1 /*
2 * Copyright (c) 1999, 2026, 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 "asm/macroAssembler.hpp"
26 #include "ci/ciArrayKlass.hpp"
27 #include "ci/ciFlatArrayKlass.hpp"
28 #include "ci/ciInstanceKlass.hpp"
29 #include "ci/ciSymbols.hpp"
30 #include "ci/ciUtilities.inline.hpp"
31 #include "classfile/vmIntrinsics.hpp"
32 #include "compiler/compileBroker.hpp"
33 #include "compiler/compileLog.hpp"
34 #include "gc/shared/barrierSet.hpp"
35 #include "gc/shared/c2/barrierSetC2.hpp"
36 #include "jfr/support/jfrIntrinsics.hpp"
37 #include "memory/resourceArea.hpp"
38 #include "oops/accessDecorators.hpp"
39 #include "oops/klass.inline.hpp"
40 #include "oops/layoutKind.hpp"
41 #include "oops/objArrayKlass.hpp"
42 #include "opto/addnode.hpp"
43 #include "opto/arraycopynode.hpp"
44 #include "opto/c2compiler.hpp"
45 #include "opto/castnode.hpp"
46 #include "opto/cfgnode.hpp"
47 #include "opto/convertnode.hpp"
48 #include "opto/countbitsnode.hpp"
49 #include "opto/graphKit.hpp"
50 #include "opto/idealKit.hpp"
51 #include "opto/inlinetypenode.hpp"
52 #include "opto/library_call.hpp"
53 #include "opto/mathexactnode.hpp"
54 #include "opto/mulnode.hpp"
55 #include "opto/narrowptrnode.hpp"
56 #include "opto/opaquenode.hpp"
57 #include "opto/opcodes.hpp"
58 #include "opto/parse.hpp"
59 #include "opto/rootnode.hpp"
60 #include "opto/runtime.hpp"
61 #include "opto/subnode.hpp"
62 #include "opto/type.hpp"
63 #include "opto/vectornode.hpp"
64 #include "prims/jvmtiExport.hpp"
65 #include "prims/jvmtiThreadState.hpp"
66 #include "prims/unsafe.hpp"
67 #include "runtime/globals.hpp"
68 #include "runtime/jniHandles.inline.hpp"
69 #include "runtime/mountUnmountDisabler.hpp"
70 #include "runtime/objectMonitor.hpp"
71 #include "runtime/sharedRuntime.hpp"
72 #include "runtime/stubRoutines.hpp"
73 #include "utilities/globalDefinitions.hpp"
74 #include "utilities/macros.hpp"
75 #include "utilities/powerOfTwo.hpp"
76
77 //---------------------------make_vm_intrinsic----------------------------
78 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
79 vmIntrinsicID id = m->intrinsic_id();
80 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
81
82 if (!m->is_loaded()) {
83 // Do not attempt to inline unloaded methods.
84 return nullptr;
85 }
86
87 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
88 bool is_available = false;
89
90 {
91 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
92 // the compiler must transition to '_thread_in_vm' state because both
93 // methods access VM-internal data.
94 VM_ENTRY_MARK;
95 methodHandle mh(THREAD, m->get_Method());
96 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
97 if (is_available && is_virtual) {
98 is_available = vmIntrinsics::does_virtual_dispatch(id);
99 }
100 }
101
102 if (is_available) {
103 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
104 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
105 return new LibraryIntrinsic(m, is_virtual,
106 vmIntrinsics::predicates_needed(id),
107 vmIntrinsics::does_virtual_dispatch(id),
108 id);
109 } else {
110 return nullptr;
111 }
112 }
113
114 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
115 LibraryCallKit kit(jvms, this);
116 Compile* C = kit.C;
117 int nodes = C->unique();
118 #ifndef PRODUCT
119 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
120 char buf[1000];
121 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
122 tty->print_cr("Intrinsic %s", str);
123 }
124 #endif
125 ciMethod* callee = kit.callee();
126 const int bci = kit.bci();
127 #ifdef ASSERT
128 Node* ctrl = kit.control();
129 #endif
130 // Try to inline the intrinsic.
131 if (callee->check_intrinsic_candidate() &&
132 kit.try_to_inline(_last_predicate)) {
133 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
134 : "(intrinsic)";
135 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
136 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
137 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
138 if (C->log()) {
139 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
140 vmIntrinsics::name_at(intrinsic_id()),
141 (is_virtual() ? " virtual='1'" : ""),
142 C->unique() - nodes);
143 }
144 // Push the result from the inlined method onto the stack.
145 kit.push_result();
146 return kit.transfer_exceptions_into_jvms();
147 }
148
149 // The intrinsic bailed out
150 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
151 assert(jvms->map() == kit.map(), "Out of sync JVM state");
152 if (jvms->has_method()) {
153 // Not a root compile.
154 const char* msg;
155 if (callee->intrinsic_candidate()) {
156 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
157 } else {
158 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
159 : "failed to inline (intrinsic), method not annotated";
160 }
161 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
162 C->inline_printer()->record(callee, jvms, InliningResult::FAILURE, msg);
163 } else {
164 // Root compile
165 ResourceMark rm;
166 stringStream msg_stream;
167 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
168 vmIntrinsics::name_at(intrinsic_id()),
169 is_virtual() ? " (virtual)" : "", bci);
170 const char *msg = msg_stream.freeze();
171 log_debug(jit, inlining)("%s", msg);
172 if (C->print_intrinsics() || C->print_inlining()) {
173 tty->print("%s", msg);
174 }
175 }
176 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
177
178 return nullptr;
179 }
180
181 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
182 LibraryCallKit kit(jvms, this);
183 Compile* C = kit.C;
184 int nodes = C->unique();
185 _last_predicate = predicate;
186 #ifndef PRODUCT
187 assert(is_predicated() && predicate < predicates_count(), "sanity");
188 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
189 char buf[1000];
190 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
191 tty->print_cr("Predicate for intrinsic %s", str);
192 }
193 #endif
194 ciMethod* callee = kit.callee();
195 const int bci = kit.bci();
196
197 Node* slow_ctl = kit.try_to_predicate(predicate);
198 if (!kit.failing()) {
199 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
200 : "(intrinsic, predicate)";
201 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
202 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
203
204 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
205 if (C->log()) {
206 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
207 vmIntrinsics::name_at(intrinsic_id()),
208 (is_virtual() ? " virtual='1'" : ""),
209 C->unique() - nodes);
210 }
211 return slow_ctl; // Could be null if the check folds.
212 }
213
214 // The intrinsic bailed out
215 if (jvms->has_method()) {
216 // Not a root compile.
217 const char* msg = "failed to generate predicate for intrinsic";
218 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
219 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
220 } else {
221 // Root compile
222 ResourceMark rm;
223 stringStream msg_stream;
224 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
225 vmIntrinsics::name_at(intrinsic_id()),
226 is_virtual() ? " (virtual)" : "", bci);
227 const char *msg = msg_stream.freeze();
228 log_debug(jit, inlining)("%s", msg);
229 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
230 }
231 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
232 return nullptr;
233 }
234
235 bool LibraryCallKit::try_to_inline(int predicate) {
236 // Handle symbolic names for otherwise undistinguished boolean switches:
237 const bool is_store = true;
238 const bool is_compress = true;
239 const bool is_static = true;
240 const bool is_volatile = true;
241
242 if (!jvms()->has_method()) {
243 // Root JVMState has a null method.
244 assert(map()->memory()->Opcode() == Op_Parm, "");
245 // Insert the memory aliasing node
246 set_all_memory(reset_memory());
247 }
248 assert(merged_memory(), "");
249
250 switch (intrinsic_id()) {
251 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
252 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
253 case vmIntrinsics::_getClass: return inline_native_getClass();
254
255 case vmIntrinsics::_ceil:
256 case vmIntrinsics::_floor:
257 case vmIntrinsics::_rint:
258 case vmIntrinsics::_dsin:
259 case vmIntrinsics::_dcos:
260 case vmIntrinsics::_dtan:
261 case vmIntrinsics::_dsinh:
262 case vmIntrinsics::_dtanh:
263 case vmIntrinsics::_dcbrt:
264 case vmIntrinsics::_dabs:
265 case vmIntrinsics::_fabs:
266 case vmIntrinsics::_iabs:
267 case vmIntrinsics::_labs:
268 case vmIntrinsics::_datan2:
269 case vmIntrinsics::_dsqrt:
270 case vmIntrinsics::_dsqrt_strict:
271 case vmIntrinsics::_dexp:
272 case vmIntrinsics::_dlog:
273 case vmIntrinsics::_dlog10:
274 case vmIntrinsics::_dpow:
275 case vmIntrinsics::_dcopySign:
276 case vmIntrinsics::_fcopySign:
277 case vmIntrinsics::_dsignum:
278 case vmIntrinsics::_roundF:
279 case vmIntrinsics::_roundD:
280 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
281
282 case vmIntrinsics::_notify:
283 case vmIntrinsics::_notifyAll:
284 return inline_notify(intrinsic_id());
285
286 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
287 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
288 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
289 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
290 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
291 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
292 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
293 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
294 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
295 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
296 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
297 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
298 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
299 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
300
301 case vmIntrinsics::_arraycopy: return inline_arraycopy();
302
303 case vmIntrinsics::_arraySort: return inline_array_sort();
304 case vmIntrinsics::_arrayPartition: return inline_array_partition();
305
306 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
307 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
308 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
309 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
310
311 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
312 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
313 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
314 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
315 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
316 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
317 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
318 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
319
320 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
321
322 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
323
324 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
325 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
326 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
327 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
328
329 case vmIntrinsics::_compressStringC:
330 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
331 case vmIntrinsics::_inflateStringC:
332 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
333
334 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
335 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
336 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
337 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
338 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
339 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
340 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
341 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
342 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
343
344 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
345 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
346 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
347 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
348 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
349 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
350 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
351 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
352 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
353
354 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
355 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
356 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
357 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
358 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
359 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
360 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
361 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
362 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
363
364 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
365 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
366 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
367 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
368 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
369 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
370 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
371 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
372 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
373
374 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
375 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
376 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
377 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
378
379 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
380 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
381 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
382 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
383
384 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
385 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
386 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
387 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
388 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
389 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
390 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
391 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
392 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
393
394 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
395 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
396 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
397 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
398 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
399 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
400 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
401 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
402 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
403
404 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
405 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
406 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
407 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
408 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
409 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
410 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
411 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
412 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
413
414 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
415 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
416 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
417 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
418 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
419 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
420 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
421 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
422 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
423
424 case vmIntrinsics::_getFlatValue: return inline_unsafe_flat_access(!is_store, Relaxed);
425 case vmIntrinsics::_putFlatValue: return inline_unsafe_flat_access( is_store, Relaxed);
426
427 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
428 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
429 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
430 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
431 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
432
433 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
434 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
435 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
436 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
437 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
438 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
439 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
440 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
441 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
442 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
443 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
444 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
445 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
446 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
447 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
448 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
449 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
450 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
451 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
452 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
453
454 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
455 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
456 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
457 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
458 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
459 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
460 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
461 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
462 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
463 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
464 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
465 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
466 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
467 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
468 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
469
470 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
471 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
472 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
473 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
474
475 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
476 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
477 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
478 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
479 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
480
481 case vmIntrinsics::_loadFence:
482 case vmIntrinsics::_storeFence:
483 case vmIntrinsics::_storeStoreFence:
484 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
485
486 case vmIntrinsics::_arrayInstanceBaseOffset: return inline_arrayInstanceBaseOffset();
487 case vmIntrinsics::_arrayInstanceIndexScale: return inline_arrayInstanceIndexScale();
488 case vmIntrinsics::_arrayLayout: return inline_arrayLayout();
489 case vmIntrinsics::_getFieldMap: return inline_getFieldMap();
490
491 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
492
493 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
494 case vmIntrinsics::_currentThread: return inline_native_currentThread();
495 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
496
497 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
498 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
499
500 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
501 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
502
503 case vmIntrinsics::_vthreadEndFirstTransition: return inline_native_vthread_end_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_end_first_transition_Java()),
504 "endFirstTransition", true);
505 case vmIntrinsics::_vthreadStartFinalTransition: return inline_native_vthread_start_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_start_final_transition_Java()),
506 "startFinalTransition", true);
507 case vmIntrinsics::_vthreadStartTransition: return inline_native_vthread_start_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_start_transition_Java()),
508 "startTransition", false);
509 case vmIntrinsics::_vthreadEndTransition: return inline_native_vthread_end_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_end_transition_Java()),
510 "endTransition", false);
511 #if INCLUDE_JVMTI
512 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
513 #endif
514
515 #ifdef JFR_HAVE_INTRINSICS
516 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
517 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
518 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
519 #endif
520 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
521 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
522 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
523 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
524 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
525 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
526 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
527 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
528 case vmIntrinsics::_getLength: return inline_native_getLength();
529 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
530 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
531 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
532 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
533 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
534 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
535 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
536
537 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
538 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
539 case vmIntrinsics::_newNullRestrictedNonAtomicArray: return inline_newArray(/* null_free */ true, /* atomic */ false);
540 case vmIntrinsics::_newNullRestrictedAtomicArray: return inline_newArray(/* null_free */ true, /* atomic */ true);
541 case vmIntrinsics::_newNullableAtomicArray: return inline_newArray(/* null_free */ false, /* atomic */ true);
542 case vmIntrinsics::_isFlatArray: return inline_getArrayProperties(IsFlat);
543 case vmIntrinsics::_isNullRestrictedArray: return inline_getArrayProperties(IsNullRestricted);
544 case vmIntrinsics::_isAtomicArray: return inline_getArrayProperties(IsAtomic);
545
546 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
547
548 case vmIntrinsics::_isInstance:
549 case vmIntrinsics::_isHidden:
550 case vmIntrinsics::_getSuperclass: return inline_native_Class_query(intrinsic_id());
551
552 case vmIntrinsics::_floatToRawIntBits:
553 case vmIntrinsics::_floatToIntBits:
554 case vmIntrinsics::_intBitsToFloat:
555 case vmIntrinsics::_doubleToRawLongBits:
556 case vmIntrinsics::_doubleToLongBits:
557 case vmIntrinsics::_longBitsToDouble:
558 case vmIntrinsics::_floatToFloat16:
559 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
560 case vmIntrinsics::_sqrt_float16: return inline_fp16_operations(intrinsic_id(), 1);
561 case vmIntrinsics::_fma_float16: return inline_fp16_operations(intrinsic_id(), 3);
562 case vmIntrinsics::_floatIsFinite:
563 case vmIntrinsics::_floatIsInfinite:
564 case vmIntrinsics::_doubleIsFinite:
565 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
566
567 case vmIntrinsics::_numberOfLeadingZeros_i:
568 case vmIntrinsics::_numberOfLeadingZeros_l:
569 case vmIntrinsics::_numberOfTrailingZeros_i:
570 case vmIntrinsics::_numberOfTrailingZeros_l:
571 case vmIntrinsics::_bitCount_i:
572 case vmIntrinsics::_bitCount_l:
573 case vmIntrinsics::_reverse_i:
574 case vmIntrinsics::_reverse_l:
575 case vmIntrinsics::_reverseBytes_i:
576 case vmIntrinsics::_reverseBytes_l:
577 case vmIntrinsics::_reverseBytes_s:
578 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
579
580 case vmIntrinsics::_compress_i:
581 case vmIntrinsics::_compress_l:
582 case vmIntrinsics::_expand_i:
583 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
584
585 case vmIntrinsics::_compareUnsigned_i:
586 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
587
588 case vmIntrinsics::_divideUnsigned_i:
589 case vmIntrinsics::_divideUnsigned_l:
590 case vmIntrinsics::_remainderUnsigned_i:
591 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
592
593 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
594
595 case vmIntrinsics::_Reference_get0: return inline_reference_get0();
596 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
597 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
598 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
599 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
600
601 case vmIntrinsics::_Class_cast: return inline_Class_cast();
602
603 case vmIntrinsics::_aescrypt_encryptBlock:
604 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
605
606 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
607 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
608 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
609
610 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
611 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
612 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
613
614 case vmIntrinsics::_counterMode_AESCrypt:
615 return inline_counterMode_AESCrypt(intrinsic_id());
616
617 case vmIntrinsics::_galoisCounterMode_AESCrypt:
618 return inline_galoisCounterMode_AESCrypt();
619
620 case vmIntrinsics::_md5_implCompress:
621 case vmIntrinsics::_sha_implCompress:
622 case vmIntrinsics::_sha2_implCompress:
623 case vmIntrinsics::_sha5_implCompress:
624 case vmIntrinsics::_sha3_implCompress:
625 return inline_digestBase_implCompress(intrinsic_id());
626 case vmIntrinsics::_double_keccak:
627 return inline_double_keccak();
628
629 case vmIntrinsics::_digestBase_implCompressMB:
630 return inline_digestBase_implCompressMB(predicate);
631
632 case vmIntrinsics::_multiplyToLen:
633 return inline_multiplyToLen();
634
635 case vmIntrinsics::_squareToLen:
636 return inline_squareToLen();
637
638 case vmIntrinsics::_mulAdd:
639 return inline_mulAdd();
640
641 case vmIntrinsics::_montgomeryMultiply:
642 return inline_montgomeryMultiply();
643 case vmIntrinsics::_montgomerySquare:
644 return inline_montgomerySquare();
645
646 case vmIntrinsics::_bigIntegerRightShiftWorker:
647 return inline_bigIntegerShift(true);
648 case vmIntrinsics::_bigIntegerLeftShiftWorker:
649 return inline_bigIntegerShift(false);
650
651 case vmIntrinsics::_vectorizedMismatch:
652 return inline_vectorizedMismatch();
653
654 case vmIntrinsics::_ghash_processBlocks:
655 return inline_ghash_processBlocks();
656 case vmIntrinsics::_chacha20Block:
657 return inline_chacha20Block();
658 case vmIntrinsics::_kyberNtt:
659 return inline_kyberNtt();
660 case vmIntrinsics::_kyberInverseNtt:
661 return inline_kyberInverseNtt();
662 case vmIntrinsics::_kyberNttMult:
663 return inline_kyberNttMult();
664 case vmIntrinsics::_kyberAddPoly_2:
665 return inline_kyberAddPoly_2();
666 case vmIntrinsics::_kyberAddPoly_3:
667 return inline_kyberAddPoly_3();
668 case vmIntrinsics::_kyber12To16:
669 return inline_kyber12To16();
670 case vmIntrinsics::_kyberBarrettReduce:
671 return inline_kyberBarrettReduce();
672 case vmIntrinsics::_dilithiumAlmostNtt:
673 return inline_dilithiumAlmostNtt();
674 case vmIntrinsics::_dilithiumAlmostInverseNtt:
675 return inline_dilithiumAlmostInverseNtt();
676 case vmIntrinsics::_dilithiumNttMult:
677 return inline_dilithiumNttMult();
678 case vmIntrinsics::_dilithiumMontMulByConstant:
679 return inline_dilithiumMontMulByConstant();
680 case vmIntrinsics::_dilithiumDecomposePoly:
681 return inline_dilithiumDecomposePoly();
682 case vmIntrinsics::_base64_encodeBlock:
683 return inline_base64_encodeBlock();
684 case vmIntrinsics::_base64_decodeBlock:
685 return inline_base64_decodeBlock();
686 case vmIntrinsics::_poly1305_processBlocks:
687 return inline_poly1305_processBlocks();
688 case vmIntrinsics::_intpoly_montgomeryMult_P256:
689 return inline_intpoly_montgomeryMult_P256();
690 case vmIntrinsics::_intpoly_assign:
691 return inline_intpoly_assign();
692 case vmIntrinsics::_encodeISOArray:
693 case vmIntrinsics::_encodeByteISOArray:
694 return inline_encodeISOArray(false);
695 case vmIntrinsics::_encodeAsciiArray:
696 return inline_encodeISOArray(true);
697
698 case vmIntrinsics::_updateCRC32:
699 return inline_updateCRC32();
700 case vmIntrinsics::_updateBytesCRC32:
701 return inline_updateBytesCRC32();
702 case vmIntrinsics::_updateByteBufferCRC32:
703 return inline_updateByteBufferCRC32();
704
705 case vmIntrinsics::_updateBytesCRC32C:
706 return inline_updateBytesCRC32C();
707 case vmIntrinsics::_updateDirectByteBufferCRC32C:
708 return inline_updateDirectByteBufferCRC32C();
709
710 case vmIntrinsics::_updateBytesAdler32:
711 return inline_updateBytesAdler32();
712 case vmIntrinsics::_updateByteBufferAdler32:
713 return inline_updateByteBufferAdler32();
714
715 case vmIntrinsics::_profileBoolean:
716 return inline_profileBoolean();
717 case vmIntrinsics::_isCompileConstant:
718 return inline_isCompileConstant();
719
720 case vmIntrinsics::_countPositives:
721 return inline_countPositives();
722
723 case vmIntrinsics::_fmaD:
724 case vmIntrinsics::_fmaF:
725 return inline_fma(intrinsic_id());
726
727 case vmIntrinsics::_isDigit:
728 case vmIntrinsics::_isLowerCase:
729 case vmIntrinsics::_isUpperCase:
730 case vmIntrinsics::_isWhitespace:
731 return inline_character_compare(intrinsic_id());
732
733 case vmIntrinsics::_min:
734 case vmIntrinsics::_max:
735 case vmIntrinsics::_min_strict:
736 case vmIntrinsics::_max_strict:
737 case vmIntrinsics::_minL:
738 case vmIntrinsics::_maxL:
739 case vmIntrinsics::_minF:
740 case vmIntrinsics::_maxF:
741 case vmIntrinsics::_minD:
742 case vmIntrinsics::_maxD:
743 case vmIntrinsics::_minF_strict:
744 case vmIntrinsics::_maxF_strict:
745 case vmIntrinsics::_minD_strict:
746 case vmIntrinsics::_maxD_strict:
747 return inline_min_max(intrinsic_id());
748
749 case vmIntrinsics::_VectorUnaryOp:
750 return inline_vector_nary_operation(1);
751 case vmIntrinsics::_VectorBinaryOp:
752 return inline_vector_nary_operation(2);
753 case vmIntrinsics::_VectorUnaryLibOp:
754 return inline_vector_call(1);
755 case vmIntrinsics::_VectorBinaryLibOp:
756 return inline_vector_call(2);
757 case vmIntrinsics::_VectorTernaryOp:
758 return inline_vector_nary_operation(3);
759 case vmIntrinsics::_VectorFromBitsCoerced:
760 return inline_vector_frombits_coerced();
761 case vmIntrinsics::_VectorMaskOp:
762 return inline_vector_mask_operation();
763 case vmIntrinsics::_VectorLoadOp:
764 return inline_vector_mem_operation(/*is_store=*/false);
765 case vmIntrinsics::_VectorLoadMaskedOp:
766 return inline_vector_mem_masked_operation(/*is_store*/false);
767 case vmIntrinsics::_VectorStoreOp:
768 return inline_vector_mem_operation(/*is_store=*/true);
769 case vmIntrinsics::_VectorStoreMaskedOp:
770 return inline_vector_mem_masked_operation(/*is_store=*/true);
771 case vmIntrinsics::_VectorGatherOp:
772 return inline_vector_gather_scatter(/*is_scatter*/ false);
773 case vmIntrinsics::_VectorScatterOp:
774 return inline_vector_gather_scatter(/*is_scatter*/ true);
775 case vmIntrinsics::_VectorReductionCoerced:
776 return inline_vector_reduction();
777 case vmIntrinsics::_VectorTest:
778 return inline_vector_test();
779 case vmIntrinsics::_VectorBlend:
780 return inline_vector_blend();
781 case vmIntrinsics::_VectorRearrange:
782 return inline_vector_rearrange();
783 case vmIntrinsics::_VectorSelectFrom:
784 return inline_vector_select_from();
785 case vmIntrinsics::_VectorCompare:
786 return inline_vector_compare();
787 case vmIntrinsics::_VectorBroadcastInt:
788 return inline_vector_broadcast_int();
789 case vmIntrinsics::_VectorConvert:
790 return inline_vector_convert();
791 case vmIntrinsics::_VectorInsert:
792 return inline_vector_insert();
793 case vmIntrinsics::_VectorExtract:
794 return inline_vector_extract();
795 case vmIntrinsics::_VectorCompressExpand:
796 return inline_vector_compress_expand();
797 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
798 return inline_vector_select_from_two_vectors();
799 case vmIntrinsics::_IndexVector:
800 return inline_index_vector();
801 case vmIntrinsics::_IndexPartiallyInUpperRange:
802 return inline_index_partially_in_upper_range();
803
804 case vmIntrinsics::_getObjectSize:
805 return inline_getObjectSize();
806
807 case vmIntrinsics::_blackhole:
808 return inline_blackhole();
809
810 default:
811 // If you get here, it may be that someone has added a new intrinsic
812 // to the list in vmIntrinsics.hpp without implementing it here.
813 #ifndef PRODUCT
814 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
815 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
816 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
817 }
818 #endif
819 return false;
820 }
821 }
822
823 Node* LibraryCallKit::try_to_predicate(int predicate) {
824 if (!jvms()->has_method()) {
825 // Root JVMState has a null method.
826 assert(map()->memory()->Opcode() == Op_Parm, "");
827 // Insert the memory aliasing node
828 set_all_memory(reset_memory());
829 }
830 assert(merged_memory(), "");
831
832 switch (intrinsic_id()) {
833 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
834 return inline_cipherBlockChaining_AESCrypt_predicate(false);
835 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
836 return inline_cipherBlockChaining_AESCrypt_predicate(true);
837 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
838 return inline_electronicCodeBook_AESCrypt_predicate(false);
839 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
840 return inline_electronicCodeBook_AESCrypt_predicate(true);
841 case vmIntrinsics::_counterMode_AESCrypt:
842 return inline_counterMode_AESCrypt_predicate();
843 case vmIntrinsics::_digestBase_implCompressMB:
844 return inline_digestBase_implCompressMB_predicate(predicate);
845 case vmIntrinsics::_galoisCounterMode_AESCrypt:
846 return inline_galoisCounterMode_AESCrypt_predicate();
847
848 default:
849 // If you get here, it may be that someone has added a new intrinsic
850 // to the list in vmIntrinsics.hpp without implementing it here.
851 #ifndef PRODUCT
852 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
853 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
854 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
855 }
856 #endif
857 Node* slow_ctl = control();
858 set_control(top()); // No fast path intrinsic
859 return slow_ctl;
860 }
861 }
862
863 //------------------------------set_result-------------------------------
864 // Helper function for finishing intrinsics.
865 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
866 record_for_igvn(region);
867 set_control(_gvn.transform(region));
868 set_result( _gvn.transform(value));
869 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
870 }
871
872 RegionNode* LibraryCallKit::create_bailout() {
873 RegionNode* bailout = new RegionNode(1);
874 record_for_igvn(bailout);
875 return bailout;
876 }
877
878 bool LibraryCallKit::check_bailout(RegionNode* bailout) {
879 if (bailout->req() > 1) {
880 bailout = _gvn.transform(bailout)->as_Region();
881 Node* frame = _gvn.transform(new ParmNode(C->start(), TypeFunc::FramePtr));
882 Node* halt = _gvn.transform(new HaltNode(bailout, frame, "unexpected guard failure in intrinsic"));
883 C->root()->add_req(halt);
884 }
885 return stopped();
886 }
887
888 //------------------------------generate_guard---------------------------
889 // Helper function for generating guarded fast-slow graph structures.
890 // The given 'test', if true, guards a slow path. If the test fails
891 // then a fast path can be taken. (We generally hope it fails.)
892 // In all cases, GraphKit::control() is updated to the fast path.
893 // The returned value represents the control for the slow path.
894 // The return value is never 'top'; it is either a valid control
895 // or null if it is obvious that the slow path can never be taken.
896 // Also, if region and the slow control are not null, the slow edge
897 // is appended to the region.
898 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
899 if (stopped()) {
900 // Already short circuited.
901 return nullptr;
902 }
903
904 // Build an if node and its projections.
905 // If test is true we take the slow path, which we assume is uncommon.
906 if (_gvn.type(test) == TypeInt::ZERO) {
907 // The slow branch is never taken. No need to build this guard.
908 return nullptr;
909 }
910
911 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
912
913 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
914 if (if_slow == top()) {
915 // The slow branch is never taken. No need to build this guard.
916 return nullptr;
917 }
918
919 if (region != nullptr)
920 region->add_req(if_slow);
921
922 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
923 set_control(if_fast);
924
925 return if_slow;
926 }
927
928 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
929 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
930 }
931 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
932 return generate_guard(test, region, PROB_FAIR);
933 }
934
935 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
936 Node** pos_index, bool with_opaque) {
937 if (stopped())
938 return nullptr; // already stopped
939 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
940 return nullptr; // index is already adequately typed
941 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
942 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
943 if (with_opaque) {
944 bol_lt = _gvn.transform(new OpaqueConstantBoolNode(C, bol_lt, false));
945 }
946 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
947 if (is_neg != nullptr && pos_index != nullptr) {
948 // Emulate effect of Parse::adjust_map_after_if.
949 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
950 (*pos_index) = _gvn.transform(ccast);
951 }
952 return is_neg;
953 }
954
955 // Make sure that 'position' is a valid limit index, in [0..length].
956 // There are two equivalent plans for checking this:
957 // A. (offset + copyLength) unsigned<= arrayLength
958 // B. offset <= (arrayLength - copyLength)
959 // We require that all of the values above, except for the sum and
960 // difference, are already known to be non-negative.
961 // Plan A is robust in the face of overflow, if offset and copyLength
962 // are both hugely positive.
963 //
964 // Plan B is less direct and intuitive, but it does not overflow at
965 // all, since the difference of two non-negatives is always
966 // representable. Whenever Java methods must perform the equivalent
967 // check they generally use Plan B instead of Plan A.
968 // For the moment we use Plan A.
969 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
970 Node* subseq_length,
971 Node* array_length,
972 RegionNode* region,
973 bool with_opaque) {
974 if (stopped())
975 return nullptr; // already stopped
976 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
977 if (zero_offset && subseq_length->eqv_uncast(array_length))
978 return nullptr; // common case of whole-array copy
979 Node* last = subseq_length;
980 if (!zero_offset) // last += offset
981 last = _gvn.transform(new AddINode(last, offset));
982 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
983 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
984 if (with_opaque) {
985 bol_lt = _gvn.transform(new OpaqueConstantBoolNode(C, bol_lt, false));
986 }
987 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
988 return is_over;
989 }
990
991 // Emit range checks for the given String.value byte array
992 void LibraryCallKit::generate_string_range_check(Node* array,
993 Node* offset,
994 Node* count,
995 bool char_count,
996 RegionNode* region) {
997 if (stopped()) {
998 return; // already stopped
999 }
1000 if (char_count) {
1001 // Convert char count to byte count
1002 count = _gvn.transform(new LShiftINode(count, intcon(1)));
1003 }
1004 // Offset and count must not be negative
1005 generate_negative_guard(offset, region, nullptr, true);
1006 generate_negative_guard(count, region, nullptr, true);
1007 // Offset + count must not exceed length of array
1008 generate_limit_guard(offset, count, load_array_length(array), region, true);
1009 }
1010
1011 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
1012 bool is_immutable) {
1013 ciKlass* thread_klass = env()->Thread_klass();
1014 const Type* thread_type
1015 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1016
1017 Node* thread = _gvn.transform(new ThreadLocalNode());
1018 Node* p = off_heap_plus_addr(thread, in_bytes(handle_offset));
1019 tls_output = thread;
1020
1021 Node* thread_obj_handle
1022 = (is_immutable
1023 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
1024 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
1025 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
1026 thread_obj_handle = _gvn.transform(thread_obj_handle);
1027
1028 DecoratorSet decorators = IN_NATIVE;
1029 if (is_immutable) {
1030 decorators |= C2_IMMUTABLE_MEMORY;
1031 }
1032 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
1033 }
1034
1035 //--------------------------generate_current_thread--------------------
1036 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1037 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
1038 /*is_immutable*/false);
1039 }
1040
1041 //--------------------------generate_virtual_thread--------------------
1042 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
1043 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
1044 !C->method()->changes_current_thread());
1045 }
1046
1047 //------------------------------make_string_method_node------------------------
1048 // Helper method for String intrinsic functions. This version is called with
1049 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1050 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1051 // containing the lengths of str1 and str2.
1052 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1053 Node* result = nullptr;
1054 switch (opcode) {
1055 case Op_StrIndexOf:
1056 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1057 str1_start, cnt1, str2_start, cnt2, ae);
1058 break;
1059 case Op_StrComp:
1060 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1061 str1_start, cnt1, str2_start, cnt2, ae);
1062 break;
1063 case Op_StrEquals:
1064 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1065 // Use the constant length if there is one because optimized match rule may exist.
1066 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1067 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1068 break;
1069 default:
1070 ShouldNotReachHere();
1071 return nullptr;
1072 }
1073
1074 // All these intrinsics have checks.
1075 C->set_has_split_ifs(true); // Has chance for split-if optimization
1076 clear_upper_avx();
1077
1078 return _gvn.transform(result);
1079 }
1080
1081 //------------------------------inline_string_compareTo------------------------
1082 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1083 Node* arg1 = argument(0);
1084 Node* arg2 = argument(1);
1085
1086 arg1 = must_be_not_null(arg1, true);
1087 arg2 = must_be_not_null(arg2, true);
1088
1089 // Get start addr and length of first argument
1090 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1091 Node* arg1_cnt = load_array_length(arg1);
1092
1093 // Get start addr and length of second argument
1094 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1095 Node* arg2_cnt = load_array_length(arg2);
1096
1097 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1098 set_result(result);
1099 return true;
1100 }
1101
1102 //------------------------------inline_string_equals------------------------
1103 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1104 Node* arg1 = argument(0);
1105 Node* arg2 = argument(1);
1106
1107 // paths (plus control) merge
1108 RegionNode* region = new RegionNode(3);
1109 Node* phi = new PhiNode(region, TypeInt::BOOL);
1110
1111 if (!stopped()) {
1112
1113 arg1 = must_be_not_null(arg1, true);
1114 arg2 = must_be_not_null(arg2, true);
1115
1116 // Get start addr and length of first argument
1117 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1118 Node* arg1_cnt = load_array_length(arg1);
1119
1120 // Get start addr and length of second argument
1121 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1122 Node* arg2_cnt = load_array_length(arg2);
1123
1124 // Check for arg1_cnt != arg2_cnt
1125 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1126 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1127 Node* if_ne = generate_slow_guard(bol, nullptr);
1128 if (if_ne != nullptr) {
1129 phi->init_req(2, intcon(0));
1130 region->init_req(2, if_ne);
1131 }
1132
1133 // Check for count == 0 is done by assembler code for StrEquals.
1134
1135 if (!stopped()) {
1136 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1137 phi->init_req(1, equals);
1138 region->init_req(1, control());
1139 }
1140 }
1141
1142 // post merge
1143 set_control(_gvn.transform(region));
1144 record_for_igvn(region);
1145
1146 set_result(_gvn.transform(phi));
1147 return true;
1148 }
1149
1150 //------------------------------inline_array_equals----------------------------
1151 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1152 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1153 Node* arg1 = argument(0);
1154 Node* arg2 = argument(1);
1155
1156 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1157 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1158 clear_upper_avx();
1159
1160 return true;
1161 }
1162
1163
1164 //------------------------------inline_countPositives------------------------------
1165 // int java.lang.StringCoding#countPositives0(byte[] ba, int off, int len)
1166 bool LibraryCallKit::inline_countPositives() {
1167 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1168 // no receiver since it is static method
1169 Node* ba = argument(0);
1170 Node* offset = argument(1);
1171 Node* len = argument(2);
1172
1173 ba = must_be_not_null(ba, true);
1174 RegionNode* bailout = create_bailout();
1175 generate_string_range_check(ba, offset, len, false, bailout);
1176 if (check_bailout(bailout)) {
1177 return true;
1178 }
1179
1180 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1181 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1182 set_result(_gvn.transform(result));
1183 clear_upper_avx();
1184 return true;
1185 }
1186
1187 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1188 Node* index = argument(0);
1189 Node* length = bt == T_INT ? argument(1) : argument(2);
1190 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1191 return false;
1192 }
1193
1194 // check that length is positive
1195 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1196 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1197
1198 {
1199 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1200 uncommon_trap(Deoptimization::Reason_intrinsic,
1201 Deoptimization::Action_make_not_entrant);
1202 }
1203
1204 if (stopped()) {
1205 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1206 return true;
1207 }
1208
1209 // length is now known positive, add a cast node to make this explicit
1210 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1211 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1212 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1213 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1214 casted_length = _gvn.transform(casted_length);
1215 replace_in_map(length, casted_length);
1216 length = casted_length;
1217
1218 // Use an unsigned comparison for the range check itself
1219 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1220 BoolTest::mask btest = BoolTest::lt;
1221 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1222 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1223 _gvn.set_type(rc, rc->Value(&_gvn));
1224 if (!rc_bool->is_Con()) {
1225 record_for_igvn(rc);
1226 }
1227 set_control(_gvn.transform(new IfTrueNode(rc)));
1228 {
1229 PreserveJVMState pjvms(this);
1230 set_control(_gvn.transform(new IfFalseNode(rc)));
1231 uncommon_trap(Deoptimization::Reason_range_check,
1232 Deoptimization::Action_make_not_entrant);
1233 }
1234
1235 if (stopped()) {
1236 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1237 return true;
1238 }
1239
1240 // index is now known to be >= 0 and < length, cast it
1241 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1242 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1243 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1244 result = _gvn.transform(result);
1245 set_result(result);
1246 replace_in_map(index, result);
1247 return true;
1248 }
1249
1250 //------------------------------inline_string_indexOf------------------------
1251 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1252 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1253 return false;
1254 }
1255 Node* src = argument(0);
1256 Node* tgt = argument(1);
1257
1258 // Make the merge point
1259 RegionNode* result_rgn = new RegionNode(4);
1260 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1261
1262 src = must_be_not_null(src, true);
1263 tgt = must_be_not_null(tgt, true);
1264
1265 // Get start addr and length of source string
1266 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1267 Node* src_count = load_array_length(src);
1268
1269 // Get start addr and length of substring
1270 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1271 Node* tgt_count = load_array_length(tgt);
1272
1273 Node* result = nullptr;
1274 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1275
1276 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1277 // Divide src size by 2 if String is UTF16 encoded
1278 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1279 }
1280 if (ae == StrIntrinsicNode::UU) {
1281 // Divide substring size by 2 if String is UTF16 encoded
1282 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1283 }
1284
1285 if (call_opt_stub) {
1286 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1287 StubRoutines::_string_indexof_array[ae],
1288 "stringIndexOf", TypePtr::BOTTOM, src_start,
1289 src_count, tgt_start, tgt_count);
1290 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1291 } else {
1292 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1293 result_rgn, result_phi, ae);
1294 }
1295 if (result != nullptr) {
1296 result_phi->init_req(3, result);
1297 result_rgn->init_req(3, control());
1298 }
1299 set_control(_gvn.transform(result_rgn));
1300 record_for_igvn(result_rgn);
1301 set_result(_gvn.transform(result_phi));
1302
1303 return true;
1304 }
1305
1306 //-----------------------------inline_string_indexOfI-----------------------
1307 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1308 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1309 return false;
1310 }
1311
1312 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1313 Node* src = argument(0); // byte[]
1314 Node* src_count = argument(1); // char count
1315 Node* tgt = argument(2); // byte[]
1316 Node* tgt_count = argument(3); // char count
1317 Node* from_index = argument(4); // char index
1318
1319 src = must_be_not_null(src, true);
1320 tgt = must_be_not_null(tgt, true);
1321
1322 // Multiply byte array index by 2 if String is UTF16 encoded
1323 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1324 src_count = _gvn.transform(new SubINode(src_count, from_index));
1325 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1326 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1327
1328 // Range checks
1329 RegionNode* bailout = create_bailout();
1330 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL, bailout);
1331 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU, bailout);
1332 if (check_bailout(bailout)) {
1333 return true;
1334 }
1335
1336 RegionNode* region = new RegionNode(5);
1337 Node* phi = new PhiNode(region, TypeInt::INT);
1338 Node* result = nullptr;
1339
1340 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1341
1342 if (call_opt_stub) {
1343 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1344 StubRoutines::_string_indexof_array[ae],
1345 "stringIndexOf", TypePtr::BOTTOM, src_start,
1346 src_count, tgt_start, tgt_count);
1347 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1348 } else {
1349 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1350 region, phi, ae);
1351 }
1352 if (result != nullptr) {
1353 // The result is index relative to from_index if substring was found, -1 otherwise.
1354 // Generate code which will fold into cmove.
1355 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1356 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1357
1358 Node* if_lt = generate_slow_guard(bol, nullptr);
1359 if (if_lt != nullptr) {
1360 // result == -1
1361 phi->init_req(3, result);
1362 region->init_req(3, if_lt);
1363 }
1364 if (!stopped()) {
1365 result = _gvn.transform(new AddINode(result, from_index));
1366 phi->init_req(4, result);
1367 region->init_req(4, control());
1368 }
1369 }
1370
1371 set_control(_gvn.transform(region));
1372 record_for_igvn(region);
1373 set_result(_gvn.transform(phi));
1374 clear_upper_avx();
1375
1376 return true;
1377 }
1378
1379 // Create StrIndexOfNode with fast path checks
1380 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1381 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1382 // Check for substr count > string count
1383 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1384 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1385 Node* if_gt = generate_slow_guard(bol, nullptr);
1386 if (if_gt != nullptr) {
1387 phi->init_req(1, intcon(-1));
1388 region->init_req(1, if_gt);
1389 }
1390 if (!stopped()) {
1391 // Check for substr count == 0
1392 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1393 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1394 Node* if_zero = generate_slow_guard(bol, nullptr);
1395 if (if_zero != nullptr) {
1396 phi->init_req(2, intcon(0));
1397 region->init_req(2, if_zero);
1398 }
1399 }
1400 if (!stopped()) {
1401 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1402 }
1403 return nullptr;
1404 }
1405
1406 //-----------------------------inline_string_indexOfChar-----------------------
1407 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1408 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1409 return false;
1410 }
1411 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1412 return false;
1413 }
1414 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1415 Node* src = argument(0); // byte[]
1416 Node* int_ch = argument(1);
1417 Node* from_index = argument(2);
1418 Node* max = argument(3);
1419
1420 src = must_be_not_null(src, true);
1421
1422 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1423 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1424 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1425
1426 // Range checks
1427 RegionNode* bailout = create_bailout();
1428 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U, bailout);
1429 if (check_bailout(bailout)) {
1430 return true;
1431 }
1432
1433 // Check for int_ch >= 0
1434 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1435 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1436 {
1437 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1438 uncommon_trap(Deoptimization::Reason_intrinsic,
1439 Deoptimization::Action_maybe_recompile);
1440 }
1441 if (stopped()) {
1442 return true;
1443 }
1444
1445 RegionNode* region = new RegionNode(3);
1446 Node* phi = new PhiNode(region, TypeInt::INT);
1447
1448 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1449 C->set_has_split_ifs(true); // Has chance for split-if optimization
1450 _gvn.transform(result);
1451
1452 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1453 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1454
1455 Node* if_lt = generate_slow_guard(bol, nullptr);
1456 if (if_lt != nullptr) {
1457 // result == -1
1458 phi->init_req(2, result);
1459 region->init_req(2, if_lt);
1460 }
1461 if (!stopped()) {
1462 result = _gvn.transform(new AddINode(result, from_index));
1463 phi->init_req(1, result);
1464 region->init_req(1, control());
1465 }
1466 set_control(_gvn.transform(region));
1467 record_for_igvn(region);
1468 set_result(_gvn.transform(phi));
1469 clear_upper_avx();
1470
1471 return true;
1472 }
1473 //---------------------------inline_string_copy---------------------
1474 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1475 // int StringUTF16.compress0(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1476 // int StringUTF16.compress0(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1477 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1478 // void StringLatin1.inflate0(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1479 // void StringLatin1.inflate0(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1480 bool LibraryCallKit::inline_string_copy(bool compress) {
1481 int nargs = 5; // 2 oops, 3 ints
1482 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1483
1484 Node* src = argument(0);
1485 Node* src_offset = argument(1);
1486 Node* dst = argument(2);
1487 Node* dst_offset = argument(3);
1488 Node* length = argument(4);
1489
1490 // Check for allocation before we add nodes that would confuse
1491 // tightly_coupled_allocation()
1492 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1493
1494 // Figure out the size and type of the elements we will be copying.
1495 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1496 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1497 if (src_type == nullptr || dst_type == nullptr) {
1498 return false;
1499 }
1500 BasicType src_elem = src_type->elem()->array_element_basic_type();
1501 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1502 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1503 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1504 "Unsupported array types for inline_string_copy");
1505
1506 src = must_be_not_null(src, true);
1507 dst = must_be_not_null(dst, true);
1508
1509 // Convert char[] offsets to byte[] offsets
1510 bool convert_src = (compress && src_elem == T_BYTE);
1511 bool convert_dst = (!compress && dst_elem == T_BYTE);
1512 if (convert_src) {
1513 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1514 } else if (convert_dst) {
1515 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1516 }
1517
1518 // Range checks
1519 RegionNode* bailout = create_bailout();
1520 generate_string_range_check(src, src_offset, length, convert_src, bailout);
1521 generate_string_range_check(dst, dst_offset, length, convert_dst, bailout);
1522 if (check_bailout(bailout)) {
1523 return true;
1524 }
1525
1526 Node* src_start = array_element_address(src, src_offset, src_elem);
1527 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1528 // 'src_start' points to src array + scaled offset
1529 // 'dst_start' points to dst array + scaled offset
1530 Node* count = nullptr;
1531 if (compress) {
1532 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1533 } else {
1534 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1535 }
1536
1537 if (alloc != nullptr) {
1538 if (alloc->maybe_set_complete(&_gvn)) {
1539 // "You break it, you buy it."
1540 InitializeNode* init = alloc->initialization();
1541 assert(init->is_complete(), "we just did this");
1542 init->set_complete_with_arraycopy();
1543 assert(dst->is_CheckCastPP(), "sanity");
1544 assert(dst->in(0)->in(0) == init, "dest pinned");
1545 }
1546 // Do not let stores that initialize this object be reordered with
1547 // a subsequent store that would make this object accessible by
1548 // other threads.
1549 // Record what AllocateNode this StoreStore protects so that
1550 // escape analysis can go from the MemBarStoreStoreNode to the
1551 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1552 // based on the escape status of the AllocateNode.
1553 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1554 }
1555 if (compress) {
1556 set_result(_gvn.transform(count));
1557 }
1558 clear_upper_avx();
1559
1560 return true;
1561 }
1562
1563 #ifdef _LP64
1564 #define XTOP ,top() /*additional argument*/
1565 #else //_LP64
1566 #define XTOP /*no additional argument*/
1567 #endif //_LP64
1568
1569 //------------------------inline_string_toBytesU--------------------------
1570 // public static byte[] StringUTF16.toBytes0(char[] value, int off, int len)
1571 bool LibraryCallKit::inline_string_toBytesU() {
1572 // Get the arguments.
1573 assert(callee()->signature()->size() == 3, "character array encoder requires 3 arguments");
1574 Node* value = argument(0);
1575 Node* offset = argument(1);
1576 Node* length = argument(2);
1577
1578 Node* newcopy = nullptr;
1579
1580 // Set the original stack and the reexecute bit for the interpreter to reexecute
1581 // the bytecode that invokes StringUTF16.toBytes0() if deoptimization happens.
1582 { PreserveReexecuteState preexecs(this);
1583 jvms()->set_should_reexecute(true);
1584
1585 value = must_be_not_null(value, true);
1586 RegionNode* bailout = create_bailout();
1587 generate_negative_guard(offset, bailout, nullptr, true);
1588 generate_negative_guard(length, bailout, nullptr, true);
1589 generate_limit_guard(offset, length, load_array_length(value), bailout, true);
1590 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1591 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout, true);
1592 if (check_bailout(bailout)) {
1593 return true;
1594 }
1595
1596 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1597 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1598 newcopy = new_array(klass_node, size, 0); // no arguments to push
1599 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1600 guarantee(alloc != nullptr, "created above");
1601
1602 // Calculate starting addresses.
1603 Node* src_start = array_element_address(value, offset, T_CHAR);
1604 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1605
1606 // Check if dst array address is aligned to HeapWordSize
1607 bool aligned = (arrayOopDesc::base_offset_in_bytes(T_BYTE) % HeapWordSize == 0);
1608 // If true, then check if src array address is aligned to HeapWordSize
1609 if (aligned) {
1610 const TypeInt* toffset = gvn().type(offset)->is_int();
1611 aligned = toffset->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) +
1612 toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1613 }
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->maybe_set_complete(&_gvn)) {
1624 // "You break it, you buy it."
1625 InitializeNode* init = alloc->initialization();
1626 assert(init->is_complete(), "we just did this");
1627 init->set_complete_with_arraycopy();
1628 assert(newcopy->is_CheckCastPP(), "sanity");
1629 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1630 }
1631 // Do not let stores that initialize this object be reordered with
1632 // a subsequent store that would make this object accessible by
1633 // other threads.
1634 // Record what AllocateNode this StoreStore protects so that
1635 // escape analysis can go from the MemBarStoreStoreNode to the
1636 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1637 // based on the escape status of the AllocateNode.
1638 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1639 } // original reexecute is set back here
1640
1641 C->set_has_split_ifs(true); // Has chance for split-if optimization
1642 if (!stopped()) {
1643 set_result(newcopy);
1644 }
1645 clear_upper_avx();
1646
1647 return true;
1648 }
1649
1650 //------------------------inline_string_getCharsU--------------------------
1651 // public void StringUTF16.getChars0(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1652 bool LibraryCallKit::inline_string_getCharsU() {
1653 assert(callee()->signature()->size() == 5, "StringUTF16.getChars0() has 5 arguments");
1654 // Get the arguments.
1655 Node* src = argument(0);
1656 Node* src_begin = argument(1);
1657 Node* src_end = argument(2); // exclusive offset (i < src_end)
1658 Node* dst = argument(3);
1659 Node* dst_begin = argument(4);
1660
1661 // Check for allocation before we add nodes that would confuse
1662 // tightly_coupled_allocation()
1663 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1664
1665 // Check if a null path was taken unconditionally.
1666 src = must_be_not_null(src, true);
1667 dst = must_be_not_null(dst, true);
1668 if (stopped()) {
1669 return true;
1670 }
1671
1672 // Get length and convert char[] offset to byte[] offset
1673 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1674 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1675
1676 // Range checks
1677 RegionNode* bailout = create_bailout();
1678 generate_string_range_check(src, src_begin, length, true, bailout);
1679 generate_string_range_check(dst, dst_begin, length, false, bailout);
1680 if (check_bailout(bailout)) {
1681 return true;
1682 }
1683
1684 // Calculate starting addresses.
1685 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1686 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1687
1688 // Check if array addresses are aligned to HeapWordSize
1689 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1690 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1691 bool aligned = tsrc->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_BYTE) + tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1692 tdst->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) + tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1693
1694 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1695 const char* copyfunc_name = "arraycopy";
1696 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1697 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1698 OptoRuntime::fast_arraycopy_Type(),
1699 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1700 src_start, dst_start, ConvI2X(length) XTOP);
1701 // Do not let reads from the cloned object float above the arraycopy.
1702 if (alloc != nullptr) {
1703 if (alloc->maybe_set_complete(&_gvn)) {
1704 // "You break it, you buy it."
1705 InitializeNode* init = alloc->initialization();
1706 assert(init->is_complete(), "we just did this");
1707 init->set_complete_with_arraycopy();
1708 assert(dst->is_CheckCastPP(), "sanity");
1709 assert(dst->in(0)->in(0) == init, "dest pinned");
1710 }
1711 // Do not let stores that initialize this object be reordered with
1712 // a subsequent store that would make this object accessible by
1713 // other threads.
1714 // Record what AllocateNode this StoreStore protects so that
1715 // escape analysis can go from the MemBarStoreStoreNode to the
1716 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1717 // based on the escape status of the AllocateNode.
1718 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1719 } else {
1720 insert_mem_bar(Op_MemBarCPUOrder);
1721 }
1722
1723 C->set_has_split_ifs(true); // Has chance for split-if optimization
1724 return true;
1725 }
1726
1727 //----------------------inline_string_char_access----------------------------
1728 // Store/Load char to/from byte[] array.
1729 // static void StringUTF16.putChar(byte[] val, int index, int c)
1730 // static char StringUTF16.getChar(byte[] val, int index)
1731 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1732 Node* ch;
1733 if (is_store) {
1734 assert(callee()->signature()->size() == 3, "StringUTF16.putChar() has 3 arguments");
1735 ch = argument(2);
1736 } else {
1737 assert(callee()->signature()->size() == 2, "StringUTF16.getChar() has 2 arguments");
1738 ch = nullptr;
1739 }
1740 Node* value = argument(0);
1741 Node* index = argument(1);
1742
1743 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1744 // correctly requires matched array shapes.
1745 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1746 "sanity: byte[] and char[] bases agree");
1747 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1748 "sanity: byte[] and char[] scales agree");
1749
1750 // Bail when getChar over constants is requested: constant folding would
1751 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1752 // Java method would constant fold nicely instead.
1753 if (!is_store && value->is_Con() && index->is_Con()) {
1754 return false;
1755 }
1756
1757 // Save state and restore on bailout
1758 SavedState old_state(this);
1759
1760 value = must_be_not_null(value, true);
1761
1762 Node* adr = array_element_address(value, index, T_CHAR);
1763 if (adr->is_top()) {
1764 return false;
1765 }
1766 old_state.discard();
1767 if (is_store) {
1768 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1769 } else {
1770 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);
1771 set_result(ch);
1772 }
1773 return true;
1774 }
1775
1776
1777 //------------------------------inline_math-----------------------------------
1778 // public static double Math.abs(double)
1779 // public static double Math.sqrt(double)
1780 // public static double Math.log(double)
1781 // public static double Math.log10(double)
1782 // public static double Math.round(double)
1783 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1784 Node* arg = argument(0);
1785 Node* n = nullptr;
1786 switch (id) {
1787 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1788 case vmIntrinsics::_dsqrt:
1789 case vmIntrinsics::_dsqrt_strict:
1790 n = new SqrtDNode(C, control(), arg); break;
1791 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1792 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1793 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1794 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1795 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1796 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1797 default: fatal_unexpected_iid(id); break;
1798 }
1799 set_result(_gvn.transform(n));
1800 return true;
1801 }
1802
1803 //------------------------------inline_math-----------------------------------
1804 // public static float Math.abs(float)
1805 // public static int Math.abs(int)
1806 // public static long Math.abs(long)
1807 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1808 Node* arg = argument(0);
1809 Node* n = nullptr;
1810 switch (id) {
1811 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1812 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1813 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1814 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1815 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1816 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1817 default: fatal_unexpected_iid(id); break;
1818 }
1819 set_result(_gvn.transform(n));
1820 return true;
1821 }
1822
1823 //------------------------------runtime_math-----------------------------
1824 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1825 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1826 "must be (DD)D or (D)D type");
1827
1828 // Inputs
1829 Node* a = argument(0);
1830 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1831
1832 const TypePtr* no_memory_effects = nullptr;
1833 Node* trig = make_runtime_call(RC_LEAF | RC_PURE, call_type, funcAddr, funcName,
1834 no_memory_effects,
1835 a, top(), b, b ? top() : nullptr);
1836 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1837 #ifdef ASSERT
1838 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1839 assert(value_top == top(), "second value must be top");
1840 #endif
1841
1842 set_result(value);
1843 return true;
1844 }
1845
1846 //------------------------------inline_math_pow-----------------------------
1847 bool LibraryCallKit::inline_math_pow() {
1848 Node* base = argument(0);
1849 Node* exp = argument(2);
1850
1851 CallNode* pow = new PowDNode(C, base, exp);
1852 set_predefined_input_for_runtime_call(pow);
1853 pow = _gvn.transform(pow)->as_CallLeafPure();
1854 set_predefined_output_for_runtime_call(pow);
1855 Node* result = _gvn.transform(new ProjNode(pow, TypeFunc::Parms + 0));
1856 record_for_igvn(pow);
1857 set_result(result);
1858 return true;
1859 }
1860
1861 //------------------------------inline_math_native-----------------------------
1862 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1863 switch (id) {
1864 case vmIntrinsics::_dsin:
1865 return StubRoutines::dsin() != nullptr ?
1866 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1867 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1868 case vmIntrinsics::_dcos:
1869 return StubRoutines::dcos() != nullptr ?
1870 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1871 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1872 case vmIntrinsics::_dtan:
1873 return StubRoutines::dtan() != nullptr ?
1874 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1875 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1876 case vmIntrinsics::_dsinh:
1877 return StubRoutines::dsinh() != nullptr ?
1878 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsinh(), "dsinh") : false;
1879 case vmIntrinsics::_dtanh:
1880 return StubRoutines::dtanh() != nullptr ?
1881 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1882 case vmIntrinsics::_dcbrt:
1883 return StubRoutines::dcbrt() != nullptr ?
1884 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcbrt(), "dcbrt") : false;
1885 case vmIntrinsics::_dexp:
1886 return StubRoutines::dexp() != nullptr ?
1887 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1888 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1889 case vmIntrinsics::_dlog:
1890 return StubRoutines::dlog() != nullptr ?
1891 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1892 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1893 case vmIntrinsics::_dlog10:
1894 return StubRoutines::dlog10() != nullptr ?
1895 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1896 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1897
1898 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1899 case vmIntrinsics::_ceil:
1900 case vmIntrinsics::_floor:
1901 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1902
1903 case vmIntrinsics::_dsqrt:
1904 case vmIntrinsics::_dsqrt_strict:
1905 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1906 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1907 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1908 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1909 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1910
1911 case vmIntrinsics::_dpow: return inline_math_pow();
1912 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1913 case vmIntrinsics::_fcopySign: return inline_math(id);
1914 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1915 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1916 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1917
1918 // These intrinsics are not yet correctly implemented
1919 case vmIntrinsics::_datan2:
1920 return false;
1921
1922 default:
1923 fatal_unexpected_iid(id);
1924 return false;
1925 }
1926 }
1927
1928 //----------------------------inline_notify-----------------------------------*
1929 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1930 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1931 address func;
1932 if (id == vmIntrinsics::_notify) {
1933 func = OptoRuntime::monitor_notify_Java();
1934 } else {
1935 func = OptoRuntime::monitor_notifyAll_Java();
1936 }
1937 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1938 make_slow_call_ex(call, env()->Throwable_klass(), false);
1939 return true;
1940 }
1941
1942
1943 //----------------------------inline_min_max-----------------------------------
1944 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1945 Node* a = nullptr;
1946 Node* b = nullptr;
1947 Node* n = nullptr;
1948 switch (id) {
1949 case vmIntrinsics::_min:
1950 case vmIntrinsics::_max:
1951 case vmIntrinsics::_minF:
1952 case vmIntrinsics::_maxF:
1953 case vmIntrinsics::_minF_strict:
1954 case vmIntrinsics::_maxF_strict:
1955 case vmIntrinsics::_min_strict:
1956 case vmIntrinsics::_max_strict:
1957 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
1958 a = argument(0);
1959 b = argument(1);
1960 break;
1961 case vmIntrinsics::_minD:
1962 case vmIntrinsics::_maxD:
1963 case vmIntrinsics::_minD_strict:
1964 case vmIntrinsics::_maxD_strict:
1965 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
1966 a = argument(0);
1967 b = argument(2);
1968 break;
1969 case vmIntrinsics::_minL:
1970 case vmIntrinsics::_maxL:
1971 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
1972 a = argument(0);
1973 b = argument(2);
1974 break;
1975 default:
1976 fatal_unexpected_iid(id);
1977 break;
1978 }
1979
1980 switch (id) {
1981 case vmIntrinsics::_min:
1982 case vmIntrinsics::_min_strict:
1983 n = new MinINode(a, b);
1984 break;
1985 case vmIntrinsics::_max:
1986 case vmIntrinsics::_max_strict:
1987 n = new MaxINode(a, b);
1988 break;
1989 case vmIntrinsics::_minF:
1990 case vmIntrinsics::_minF_strict:
1991 n = new MinFNode(a, b);
1992 break;
1993 case vmIntrinsics::_maxF:
1994 case vmIntrinsics::_maxF_strict:
1995 n = new MaxFNode(a, b);
1996 break;
1997 case vmIntrinsics::_minD:
1998 case vmIntrinsics::_minD_strict:
1999 n = new MinDNode(a, b);
2000 break;
2001 case vmIntrinsics::_maxD:
2002 case vmIntrinsics::_maxD_strict:
2003 n = new MaxDNode(a, b);
2004 break;
2005 case vmIntrinsics::_minL:
2006 n = new MinLNode(_gvn.C, a, b);
2007 break;
2008 case vmIntrinsics::_maxL:
2009 n = new MaxLNode(_gvn.C, a, b);
2010 break;
2011 default:
2012 fatal_unexpected_iid(id);
2013 break;
2014 }
2015
2016 set_result(_gvn.transform(n));
2017 return true;
2018 }
2019
2020 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
2021 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
2022 env()->ArithmeticException_instance())) {
2023 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
2024 // so let's bail out intrinsic rather than risking deopting again.
2025 return false;
2026 }
2027
2028 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2029 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2030 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2031 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2032
2033 {
2034 PreserveJVMState pjvms(this);
2035 PreserveReexecuteState preexecs(this);
2036 jvms()->set_should_reexecute(true);
2037
2038 set_control(slow_path);
2039 set_i_o(i_o());
2040
2041 builtin_throw(Deoptimization::Reason_intrinsic,
2042 env()->ArithmeticException_instance(),
2043 /*allow_too_many_traps*/ false);
2044 }
2045
2046 set_control(fast_path);
2047 set_result(math);
2048 return true;
2049 }
2050
2051 template <typename OverflowOp>
2052 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2053 typedef typename OverflowOp::MathOp MathOp;
2054
2055 MathOp* mathOp = new MathOp(arg1, arg2);
2056 Node* operation = _gvn.transform( mathOp );
2057 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2058 return inline_math_mathExact(operation, ofcheck);
2059 }
2060
2061 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2062 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2063 }
2064
2065 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2066 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2067 }
2068
2069 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2070 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2071 }
2072
2073 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2074 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2075 }
2076
2077 bool LibraryCallKit::inline_math_negateExactI() {
2078 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2079 }
2080
2081 bool LibraryCallKit::inline_math_negateExactL() {
2082 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2083 }
2084
2085 bool LibraryCallKit::inline_math_multiplyExactI() {
2086 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2087 }
2088
2089 bool LibraryCallKit::inline_math_multiplyExactL() {
2090 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2091 }
2092
2093 bool LibraryCallKit::inline_math_multiplyHigh() {
2094 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2095 return true;
2096 }
2097
2098 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2099 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2100 return true;
2101 }
2102
2103 inline int
2104 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2105 const TypePtr* base_type = TypePtr::NULL_PTR;
2106 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2107 if (base_type == nullptr) {
2108 // Unknown type.
2109 return Type::AnyPtr;
2110 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2111 // Since this is a null+long form, we have to switch to a rawptr.
2112 base = _gvn.transform(new CastX2PNode(offset));
2113 offset = MakeConX(0);
2114 return Type::RawPtr;
2115 } else if (base_type->base() == Type::RawPtr) {
2116 return Type::RawPtr;
2117 } else if (base_type->isa_oopptr()) {
2118 // Base is never null => always a heap address.
2119 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2120 return Type::OopPtr;
2121 }
2122 // Offset is small => always a heap address.
2123 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2124 if (offset_type != nullptr &&
2125 base_type->offset() == 0 && // (should always be?)
2126 offset_type->_lo >= 0 &&
2127 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2128 return Type::OopPtr;
2129 } else if (type == T_OBJECT) {
2130 // off heap access to an oop doesn't make any sense. Has to be on
2131 // heap.
2132 return Type::OopPtr;
2133 }
2134 // Otherwise, it might either be oop+off or null+addr.
2135 return Type::AnyPtr;
2136 } else {
2137 // No information:
2138 return Type::AnyPtr;
2139 }
2140 }
2141
2142 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2143 Node* uncasted_base = base;
2144 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2145 if (kind == Type::RawPtr) {
2146 return off_heap_plus_addr(uncasted_base, offset);
2147 } else if (kind == Type::AnyPtr) {
2148 assert(base == uncasted_base, "unexpected base change");
2149 if (can_cast) {
2150 if (!_gvn.type(base)->speculative_maybe_null() &&
2151 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2152 // According to profiling, this access is always on
2153 // heap. Casting the base to not null and thus avoiding membars
2154 // around the access should allow better optimizations
2155 Node* null_ctl = top();
2156 base = null_check_oop(base, &null_ctl, true, true, true);
2157 assert(null_ctl->is_top(), "no null control here");
2158 return basic_plus_adr(base, offset);
2159 } else if (_gvn.type(base)->speculative_always_null() &&
2160 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2161 // According to profiling, this access is always off
2162 // heap.
2163 base = null_assert(base);
2164 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2165 offset = MakeConX(0);
2166 return off_heap_plus_addr(raw_base, offset);
2167 }
2168 }
2169 // We don't know if it's an on heap or off heap access. Fall back
2170 // to raw memory access.
2171 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2172 return off_heap_plus_addr(raw, offset);
2173 } else {
2174 assert(base == uncasted_base, "unexpected base change");
2175 // We know it's an on heap access so base can't be null
2176 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2177 base = must_be_not_null(base, true);
2178 }
2179 return basic_plus_adr(base, offset);
2180 }
2181 }
2182
2183 //--------------------------inline_number_methods-----------------------------
2184 // inline int Integer.numberOfLeadingZeros(int)
2185 // inline int Long.numberOfLeadingZeros(long)
2186 //
2187 // inline int Integer.numberOfTrailingZeros(int)
2188 // inline int Long.numberOfTrailingZeros(long)
2189 //
2190 // inline int Integer.bitCount(int)
2191 // inline int Long.bitCount(long)
2192 //
2193 // inline char Character.reverseBytes(char)
2194 // inline short Short.reverseBytes(short)
2195 // inline int Integer.reverseBytes(int)
2196 // inline long Long.reverseBytes(long)
2197 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2198 Node* arg = argument(0);
2199 Node* n = nullptr;
2200 switch (id) {
2201 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2202 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2203 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2204 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2205 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2206 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2207 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2208 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2209 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2210 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2211 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2212 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2213 default: fatal_unexpected_iid(id); break;
2214 }
2215 set_result(_gvn.transform(n));
2216 return true;
2217 }
2218
2219 //--------------------------inline_bitshuffle_methods-----------------------------
2220 // inline int Integer.compress(int, int)
2221 // inline int Integer.expand(int, int)
2222 // inline long Long.compress(long, long)
2223 // inline long Long.expand(long, long)
2224 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2225 Node* n = nullptr;
2226 switch (id) {
2227 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2228 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2229 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2230 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2231 default: fatal_unexpected_iid(id); break;
2232 }
2233 set_result(_gvn.transform(n));
2234 return true;
2235 }
2236
2237 //--------------------------inline_number_methods-----------------------------
2238 // inline int Integer.compareUnsigned(int, int)
2239 // inline int Long.compareUnsigned(long, long)
2240 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2241 Node* arg1 = argument(0);
2242 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2243 Node* n = nullptr;
2244 switch (id) {
2245 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2246 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2247 default: fatal_unexpected_iid(id); break;
2248 }
2249 set_result(_gvn.transform(n));
2250 return true;
2251 }
2252
2253 //--------------------------inline_unsigned_divmod_methods-----------------------------
2254 // inline int Integer.divideUnsigned(int, int)
2255 // inline int Integer.remainderUnsigned(int, int)
2256 // inline long Long.divideUnsigned(long, long)
2257 // inline long Long.remainderUnsigned(long, long)
2258 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2259 Node* n = nullptr;
2260 switch (id) {
2261 case vmIntrinsics::_divideUnsigned_i: {
2262 zero_check_int(argument(1));
2263 // Compile-time detect of null-exception
2264 if (stopped()) {
2265 return true; // keep the graph constructed so far
2266 }
2267 n = new UDivINode(control(), argument(0), argument(1));
2268 break;
2269 }
2270 case vmIntrinsics::_divideUnsigned_l: {
2271 zero_check_long(argument(2));
2272 // Compile-time detect of null-exception
2273 if (stopped()) {
2274 return true; // keep the graph constructed so far
2275 }
2276 n = new UDivLNode(control(), argument(0), argument(2));
2277 break;
2278 }
2279 case vmIntrinsics::_remainderUnsigned_i: {
2280 zero_check_int(argument(1));
2281 // Compile-time detect of null-exception
2282 if (stopped()) {
2283 return true; // keep the graph constructed so far
2284 }
2285 n = new UModINode(control(), argument(0), argument(1));
2286 break;
2287 }
2288 case vmIntrinsics::_remainderUnsigned_l: {
2289 zero_check_long(argument(2));
2290 // Compile-time detect of null-exception
2291 if (stopped()) {
2292 return true; // keep the graph constructed so far
2293 }
2294 n = new UModLNode(control(), argument(0), argument(2));
2295 break;
2296 }
2297 default: fatal_unexpected_iid(id); break;
2298 }
2299 set_result(_gvn.transform(n));
2300 return true;
2301 }
2302
2303 //----------------------------inline_unsafe_access----------------------------
2304
2305 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2306 // Attempt to infer a sharper value type from the offset and base type.
2307 ciKlass* sharpened_klass = nullptr;
2308 bool null_free = false;
2309
2310 // See if it is an instance field, with an object type.
2311 if (alias_type->field() != nullptr) {
2312 if (alias_type->field()->type()->is_klass()) {
2313 sharpened_klass = alias_type->field()->type()->as_klass();
2314 null_free = alias_type->field()->is_null_free();
2315 }
2316 }
2317
2318 const TypeOopPtr* result = nullptr;
2319 // See if it is a narrow oop array.
2320 if (adr_type->isa_aryptr()) {
2321 if (adr_type->offset() >= refArrayOopDesc::base_offset_in_bytes()) {
2322 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2323 null_free = adr_type->is_aryptr()->is_null_free();
2324 if (elem_type != nullptr && elem_type->is_loaded()) {
2325 // Sharpen the value type.
2326 result = elem_type;
2327 }
2328 }
2329 }
2330
2331 // The sharpened class might be unloaded if there is no class loader
2332 // contraint in place.
2333 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2334 // Sharpen the value type.
2335 result = TypeOopPtr::make_from_klass(sharpened_klass);
2336 if (null_free) {
2337 result = result->join_speculative(TypePtr::NOTNULL)->is_oopptr();
2338 }
2339 }
2340 if (result != nullptr) {
2341 #ifndef PRODUCT
2342 if (C->print_intrinsics() || C->print_inlining()) {
2343 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2344 tty->print(" sharpened value: "); result->dump(); tty->cr();
2345 }
2346 #endif
2347 }
2348 return result;
2349 }
2350
2351 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2352 switch (kind) {
2353 case Relaxed:
2354 return MO_UNORDERED;
2355 case Opaque:
2356 return MO_RELAXED;
2357 case Acquire:
2358 return MO_ACQUIRE;
2359 case Release:
2360 return MO_RELEASE;
2361 case Volatile:
2362 return MO_SEQ_CST;
2363 default:
2364 ShouldNotReachHere();
2365 return 0;
2366 }
2367 }
2368
2369 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2370 if (callee()->is_static()) return false; // caller must have the capability!
2371 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2372 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2373 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2374 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2375
2376 if (is_reference_type(type)) {
2377 decorators |= ON_UNKNOWN_OOP_REF;
2378 }
2379
2380 if (unaligned) {
2381 decorators |= C2_UNALIGNED;
2382 }
2383
2384 #ifndef PRODUCT
2385 {
2386 ResourceMark rm;
2387 // Check the signatures.
2388 ciSignature* sig = callee()->signature();
2389 #ifdef ASSERT
2390 if (!is_store) {
2391 // Object getReference(Object base, int/long offset), etc.
2392 BasicType rtype = sig->return_type()->basic_type();
2393 assert(rtype == type, "getter must return the expected value");
2394 assert(sig->count() == 2, "oop getter has 2 arguments");
2395 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2396 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2397 } else {
2398 // void putReference(Object base, int/long offset, Object x), etc.
2399 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2400 assert(sig->count() == 3, "oop putter has 3 arguments");
2401 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2402 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2403 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2404 assert(vtype == type, "putter must accept the expected value");
2405 }
2406 #endif // ASSERT
2407 }
2408 #endif //PRODUCT
2409
2410 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2411
2412 Node* receiver = argument(0); // type: oop
2413
2414 // Build address expression.
2415 Node* heap_base_oop = top();
2416
2417 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2418 Node* base = argument(1); // type: oop
2419 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2420 Node* offset = argument(2); // type: long
2421 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2422 // to be plain byte offsets, which are also the same as those accepted
2423 // by oopDesc::field_addr.
2424 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2425 "fieldOffset must be byte-scaled");
2426
2427 if (base->is_InlineType()) {
2428 assert(!is_store, "InlineTypeNodes are non-larval value objects");
2429 InlineTypeNode* vt = base->as_InlineType();
2430 if (offset->is_Con()) {
2431 long off = find_long_con(offset, 0);
2432 ciInlineKlass* vk = vt->type()->inline_klass();
2433 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2434 return false;
2435 }
2436
2437 ciField* field = vk->get_non_flat_field_by_offset(off);
2438 if (field != nullptr) {
2439 BasicType bt = type2field[field->type()->basic_type()];
2440 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2441 bt = T_OBJECT;
2442 }
2443 if (bt == type && !field->is_flat()) {
2444 Node* value = vt->field_value_by_offset(off, false);
2445 const Type* value_type = _gvn.type(value);
2446 if (value->is_InlineType()) {
2447 value = value->as_InlineType()->adjust_scalarization_depth(this);
2448 } else if (value_type->is_inlinetypeptr()) {
2449 value = InlineTypeNode::make_from_oop(this, value, value_type->inline_klass());
2450 }
2451 set_result(value);
2452 return true;
2453 }
2454 }
2455 }
2456 {
2457 // Re-execute the unsafe access if allocation triggers deoptimization.
2458 PreserveReexecuteState preexecs(this);
2459 jvms()->set_should_reexecute(true);
2460 vt = vt->buffer(this);
2461 }
2462 base = vt->get_oop();
2463 }
2464
2465 // 32-bit machines ignore the high half!
2466 offset = ConvL2X(offset);
2467
2468 // Save state and restore on bailout
2469 SavedState old_state(this);
2470
2471 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2472 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2473
2474 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2475 if (type != T_OBJECT) {
2476 decorators |= IN_NATIVE; // off-heap primitive access
2477 } else {
2478 return false; // off-heap oop accesses are not supported
2479 }
2480 } else {
2481 heap_base_oop = base; // on-heap or mixed access
2482 }
2483
2484 // Can base be null? Otherwise, always on-heap access.
2485 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2486
2487 if (!can_access_non_heap) {
2488 decorators |= IN_HEAP;
2489 }
2490
2491 Node* val = is_store ? argument(4) : nullptr;
2492
2493 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2494 if (adr_type == TypePtr::NULL_PTR) {
2495 return false; // off-heap access with zero address
2496 }
2497
2498 // Try to categorize the address.
2499 Compile::AliasType* alias_type = C->alias_type(adr_type);
2500 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2501
2502 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2503 alias_type->adr_type() == TypeAryPtr::RANGE) {
2504 return false; // not supported
2505 }
2506
2507 bool mismatched = false;
2508 BasicType bt = T_ILLEGAL;
2509 ciField* field = nullptr;
2510 if (adr_type->isa_instptr()) {
2511 const TypeInstPtr* instptr = adr_type->is_instptr();
2512 ciInstanceKlass* k = instptr->instance_klass();
2513 int off = instptr->offset();
2514 if (instptr->const_oop() != nullptr &&
2515 k == ciEnv::current()->Class_klass() &&
2516 instptr->offset() >= (k->size_helper() * wordSize)) {
2517 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2518 field = k->get_field_by_offset(off, true);
2519 } else {
2520 field = k->get_non_flat_field_by_offset(off);
2521 }
2522 if (field != nullptr) {
2523 bt = type2field[field->type()->basic_type()];
2524 }
2525 if (bt != alias_type->basic_type()) {
2526 // Type mismatch. Is it an access to a nested flat field?
2527 field = k->get_field_by_offset(off, false);
2528 if (field != nullptr) {
2529 bt = type2field[field->type()->basic_type()];
2530 }
2531 }
2532 assert(bt == alias_type->basic_type(), "should match");
2533 } else {
2534 bt = alias_type->basic_type();
2535 }
2536
2537 if (bt != T_ILLEGAL) {
2538 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2539 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2540 // Alias type doesn't differentiate between byte[] and boolean[]).
2541 // Use address type to get the element type.
2542 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2543 }
2544 if (is_reference_type(bt, true)) {
2545 // accessing an array field with getReference is not a mismatch
2546 bt = T_OBJECT;
2547 }
2548 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2549 // Don't intrinsify mismatched object accesses
2550 return false;
2551 }
2552 mismatched = (bt != type);
2553 } else if (alias_type->adr_type()->isa_oopptr()) {
2554 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2555 }
2556
2557 old_state.discard();
2558 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2559
2560 if (mismatched) {
2561 decorators |= C2_MISMATCHED;
2562 }
2563
2564 // First guess at the value type.
2565 const Type *value_type = Type::get_const_basic_type(type);
2566
2567 // Figure out the memory ordering.
2568 decorators |= mo_decorator_for_access_kind(kind);
2569
2570 if (!is_store) {
2571 if (type == T_OBJECT) {
2572 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2573 if (tjp != nullptr) {
2574 value_type = tjp;
2575 }
2576 }
2577 }
2578
2579 receiver = null_check(receiver);
2580 if (stopped()) {
2581 return true;
2582 }
2583 // Heap pointers get a null-check from the interpreter,
2584 // as a courtesy. However, this is not guaranteed by Unsafe,
2585 // and it is not possible to fully distinguish unintended nulls
2586 // from intended ones in this API.
2587
2588 if (!is_store) {
2589 Node* p = nullptr;
2590 // Try to constant fold a load from a constant field
2591
2592 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !field->is_flat() && !mismatched) {
2593 // final or stable field
2594 p = make_constant_from_field(field, heap_base_oop);
2595 }
2596
2597 if (p == nullptr) { // Could not constant fold the load
2598 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2599 const TypeOopPtr* ptr = value_type->make_oopptr();
2600 if (ptr != nullptr && ptr->is_inlinetypeptr()) {
2601 // Load a non-flattened inline type from memory
2602 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass());
2603 }
2604 // Normalize the value returned by getBoolean in the following cases
2605 if (type == T_BOOLEAN &&
2606 (mismatched ||
2607 heap_base_oop == top() || // - heap_base_oop is null or
2608 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2609 // and the unsafe access is made to large offset
2610 // (i.e., larger than the maximum offset necessary for any
2611 // field access)
2612 ) {
2613 IdealKit ideal = IdealKit(this);
2614 #define __ ideal.
2615 IdealVariable normalized_result(ideal);
2616 __ declarations_done();
2617 __ set(normalized_result, p);
2618 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2619 __ set(normalized_result, ideal.ConI(1));
2620 ideal.end_if();
2621 final_sync(ideal);
2622 p = __ value(normalized_result);
2623 #undef __
2624 }
2625 }
2626 if (type == T_ADDRESS) {
2627 p = gvn().transform(new CastP2XNode(nullptr, p));
2628 p = ConvX2UL(p);
2629 }
2630 // The load node has the control of the preceding MemBarCPUOrder. All
2631 // following nodes will have the control of the MemBarCPUOrder inserted at
2632 // the end of this method. So, pushing the load onto the stack at a later
2633 // point is fine.
2634 set_result(p);
2635 } else {
2636 if (bt == T_ADDRESS) {
2637 // Repackage the long as a pointer.
2638 val = ConvL2X(val);
2639 val = gvn().transform(new CastX2PNode(val));
2640 }
2641 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2642 }
2643
2644 return true;
2645 }
2646
2647 bool LibraryCallKit::inline_unsafe_flat_access(bool is_store, AccessKind kind) {
2648 #ifdef ASSERT
2649 {
2650 ResourceMark rm;
2651 // Check the signatures.
2652 ciSignature* sig = callee()->signature();
2653 assert(sig->type_at(0)->basic_type() == T_OBJECT, "base should be object, but is %s", type2name(sig->type_at(0)->basic_type()));
2654 assert(sig->type_at(1)->basic_type() == T_LONG, "offset should be long, but is %s", type2name(sig->type_at(1)->basic_type()));
2655 assert(sig->type_at(2)->basic_type() == T_INT, "layout kind should be int, but is %s", type2name(sig->type_at(3)->basic_type()));
2656 assert(sig->type_at(3)->basic_type() == T_OBJECT, "value klass should be object, but is %s", type2name(sig->type_at(4)->basic_type()));
2657 if (is_store) {
2658 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value, but returns %s", type2name(sig->return_type()->basic_type()));
2659 assert(sig->count() == 5, "flat putter should have 5 arguments, but has %d", sig->count());
2660 assert(sig->type_at(4)->basic_type() == T_OBJECT, "put value should be object, but is %s", type2name(sig->type_at(5)->basic_type()));
2661 } else {
2662 assert(sig->return_type()->basic_type() == T_OBJECT, "getter must return an object, but returns %s", type2name(sig->return_type()->basic_type()));
2663 assert(sig->count() == 4, "flat getter should have 4 arguments, but has %d", sig->count());
2664 }
2665 }
2666 #endif // ASSERT
2667
2668 assert(kind == Relaxed, "Only plain accesses for now");
2669 if (callee()->is_static()) {
2670 // caller must have the capability!
2671 return false;
2672 }
2673 C->set_has_unsafe_access(true);
2674
2675 const TypeInstPtr* value_klass_node = _gvn.type(argument(5))->isa_instptr();
2676 if (value_klass_node == nullptr || value_klass_node->const_oop() == nullptr) {
2677 // parameter valueType is not a constant
2678 return false;
2679 }
2680 ciType* mirror_type = value_klass_node->const_oop()->as_instance()->java_mirror_type();
2681 if (!mirror_type->is_inlinetype()) {
2682 // Dead code
2683 return false;
2684 }
2685 ciInlineKlass* value_klass = mirror_type->as_inline_klass();
2686
2687 const TypeInt* layout_type = _gvn.type(argument(4))->isa_int();
2688 if (layout_type == nullptr || !layout_type->is_con()) {
2689 // parameter layoutKind is not a constant
2690 return false;
2691 }
2692 assert(layout_type->get_con() >= static_cast<int>(LayoutKind::REFERENCE) &&
2693 layout_type->get_con() < static_cast<int>(LayoutKind::UNKNOWN),
2694 "invalid layoutKind %d", layout_type->get_con());
2695 LayoutKind layout = static_cast<LayoutKind>(layout_type->get_con());
2696 assert(layout == LayoutKind::REFERENCE || layout == LayoutKind::NULL_FREE_NON_ATOMIC_FLAT ||
2697 layout == LayoutKind::NULL_FREE_ATOMIC_FLAT || layout == LayoutKind::NULLABLE_ATOMIC_FLAT,
2698 "unexpected layoutKind %d", layout_type->get_con());
2699
2700 null_check(argument(0));
2701 if (stopped()) {
2702 return true;
2703 }
2704
2705 Node* base = must_be_not_null(argument(1), true);
2706 Node* offset = argument(2);
2707 const Type* base_type = _gvn.type(base);
2708
2709 Node* ptr;
2710 bool immutable_memory = false;
2711 DecoratorSet decorators = C2_UNSAFE_ACCESS | IN_HEAP | MO_UNORDERED;
2712 if (base_type->isa_instptr()) {
2713 const TypeLong* offset_type = _gvn.type(offset)->isa_long();
2714 if (offset_type == nullptr || !offset_type->is_con()) {
2715 // Offset into a non-array should be a constant
2716 decorators |= C2_MISMATCHED;
2717 } else {
2718 int offset_con = checked_cast<int>(offset_type->get_con());
2719 ciInstanceKlass* base_klass = base_type->is_instptr()->instance_klass();
2720 ciField* field = base_klass->get_non_flat_field_by_offset(offset_con);
2721 if (field == nullptr) {
2722 assert(!base_klass->is_final(), "non-existence field at offset %d of class %s", offset_con, base_klass->name()->as_utf8());
2723 decorators |= C2_MISMATCHED;
2724 } else {
2725 assert(field->type() == value_klass, "field at offset %d of %s is of type %s, but valueType is %s",
2726 offset_con, base_klass->name()->as_utf8(), field->type()->name(), value_klass->name()->as_utf8());
2727 immutable_memory = field->is_strict() && field->is_final();
2728
2729 if (base->is_InlineType()) {
2730 assert(!is_store, "Cannot store into a non-larval value object");
2731 set_result(base->as_InlineType()->field_value_by_offset(offset_con, false));
2732 return true;
2733 }
2734 }
2735 }
2736
2737 if (base->is_InlineType()) {
2738 assert(!is_store, "Cannot store into a non-larval value object");
2739 base = base->as_InlineType()->buffer(this, true);
2740 }
2741 ptr = basic_plus_adr(base, ConvL2X(offset));
2742 } else if (base_type->isa_aryptr()) {
2743 decorators |= IS_ARRAY;
2744 if (layout == LayoutKind::REFERENCE) {
2745 if (!base_type->is_aryptr()->is_not_flat()) {
2746 const TypeAryPtr* array_type = base_type->is_aryptr()->cast_to_not_flat();
2747 Node* new_base = _gvn.transform(new CastPPNode(control(), base, array_type, ConstraintCastNode::DependencyType::NonFloatingNarrowing));
2748 replace_in_map(base, new_base);
2749 base = new_base;
2750 }
2751 ptr = basic_plus_adr(base, ConvL2X(offset));
2752 } else {
2753 if (UseArrayFlattening) {
2754 // Flat array must have an exact type
2755 bool is_null_free = !LayoutKindHelper::is_nullable_flat(layout);
2756 bool is_atomic = LayoutKindHelper::is_atomic_flat(layout);
2757 Node* new_base = cast_to_flat_array_exact(base, value_klass, is_null_free, is_atomic);
2758 replace_in_map(base, new_base);
2759 base = new_base;
2760 ptr = basic_plus_adr(base, ConvL2X(offset));
2761 const TypeAryPtr* ptr_type = _gvn.type(ptr)->is_aryptr();
2762 if (ptr_type->field_offset().get() != 0) {
2763 ptr = _gvn.transform(new CastPPNode(control(), ptr, ptr_type->with_field_offset(0), ConstraintCastNode::DependencyType::NonFloatingNarrowing));
2764 }
2765 } else {
2766 uncommon_trap(Deoptimization::Reason_intrinsic,
2767 Deoptimization::Action_none);
2768 return true;
2769 }
2770 }
2771 } else {
2772 decorators |= C2_MISMATCHED;
2773 ptr = basic_plus_adr(base, ConvL2X(offset));
2774 }
2775
2776 if (is_store) {
2777 Node* value = argument(6);
2778 const Type* value_type = _gvn.type(value);
2779 if (!value_type->is_inlinetypeptr()) {
2780 value_type = Type::get_const_type(value_klass)->filter_speculative(value_type);
2781 Node* new_value = _gvn.transform(new CastPPNode(control(), value, value_type, ConstraintCastNode::DependencyType::NonFloatingNarrowing));
2782 new_value = InlineTypeNode::make_from_oop(this, new_value, value_klass);
2783 replace_in_map(value, new_value);
2784 value = new_value;
2785 }
2786
2787 assert(value_type->inline_klass() == value_klass, "value is of type %s while valueType is %s", value_type->inline_klass()->name()->as_utf8(), value_klass->name()->as_utf8());
2788 if (layout == LayoutKind::REFERENCE) {
2789 const TypePtr* ptr_type = (decorators & C2_MISMATCHED) != 0 ? TypeRawPtr::BOTTOM : _gvn.type(ptr)->is_ptr();
2790 access_store_at(base, ptr, ptr_type, value, value_type, T_OBJECT, decorators);
2791 } else {
2792 bool atomic = LayoutKindHelper::is_atomic_flat(layout);
2793 bool null_free = !LayoutKindHelper::is_nullable_flat(layout);
2794 value->as_InlineType()->store_flat(this, base, ptr, atomic, immutable_memory, null_free, decorators);
2795 }
2796
2797 return true;
2798 } else {
2799 decorators |= (C2_CONTROL_DEPENDENT_LOAD | C2_UNKNOWN_CONTROL_LOAD);
2800 InlineTypeNode* result;
2801 if (layout == LayoutKind::REFERENCE) {
2802 const TypePtr* ptr_type = (decorators & C2_MISMATCHED) != 0 ? TypeRawPtr::BOTTOM : _gvn.type(ptr)->is_ptr();
2803 Node* oop = access_load_at(base, ptr, ptr_type, Type::get_const_type(value_klass), T_OBJECT, decorators);
2804 result = InlineTypeNode::make_from_oop(this, oop, value_klass);
2805 } else {
2806 bool atomic = LayoutKindHelper::is_atomic_flat(layout);
2807 bool null_free = !LayoutKindHelper::is_nullable_flat(layout);
2808 result = InlineTypeNode::make_from_flat(this, value_klass, base, ptr, atomic, immutable_memory, null_free, decorators);
2809 }
2810
2811 set_result(result);
2812 return true;
2813 }
2814 }
2815
2816 //----------------------------inline_unsafe_load_store----------------------------
2817 // This method serves a couple of different customers (depending on LoadStoreKind):
2818 //
2819 // LS_cmp_swap:
2820 //
2821 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2822 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2823 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2824 //
2825 // LS_cmp_swap_weak:
2826 //
2827 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2828 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2829 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2830 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2831 //
2832 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2833 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2834 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2835 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2836 //
2837 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2838 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2839 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2840 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2841 //
2842 // LS_cmp_exchange:
2843 //
2844 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2845 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2846 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2847 //
2848 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2849 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2850 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2851 //
2852 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2853 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2854 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2855 //
2856 // LS_get_add:
2857 //
2858 // int getAndAddInt( Object o, long offset, int delta)
2859 // long getAndAddLong(Object o, long offset, long delta)
2860 //
2861 // LS_get_set:
2862 //
2863 // int getAndSet(Object o, long offset, int newValue)
2864 // long getAndSet(Object o, long offset, long newValue)
2865 // Object getAndSet(Object o, long offset, Object newValue)
2866 //
2867 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2868 // This basic scheme here is the same as inline_unsafe_access, but
2869 // differs in enough details that combining them would make the code
2870 // overly confusing. (This is a true fact! I originally combined
2871 // them, but even I was confused by it!) As much code/comments as
2872 // possible are retained from inline_unsafe_access though to make
2873 // the correspondences clearer. - dl
2874
2875 if (callee()->is_static()) return false; // caller must have the capability!
2876
2877 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2878 decorators |= mo_decorator_for_access_kind(access_kind);
2879
2880 #ifndef PRODUCT
2881 BasicType rtype;
2882 {
2883 ResourceMark rm;
2884 // Check the signatures.
2885 ciSignature* sig = callee()->signature();
2886 rtype = sig->return_type()->basic_type();
2887 switch(kind) {
2888 case LS_get_add:
2889 case LS_get_set: {
2890 // Check the signatures.
2891 #ifdef ASSERT
2892 assert(rtype == type, "get and set must return the expected type");
2893 assert(sig->count() == 3, "get and set has 3 arguments");
2894 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2895 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2896 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2897 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2898 #endif // ASSERT
2899 break;
2900 }
2901 case LS_cmp_swap:
2902 case LS_cmp_swap_weak: {
2903 // Check the signatures.
2904 #ifdef ASSERT
2905 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2906 assert(sig->count() == 4, "CAS has 4 arguments");
2907 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2908 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2909 #endif // ASSERT
2910 break;
2911 }
2912 case LS_cmp_exchange: {
2913 // Check the signatures.
2914 #ifdef ASSERT
2915 assert(rtype == type, "CAS must return the expected type");
2916 assert(sig->count() == 4, "CAS has 4 arguments");
2917 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2918 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2919 #endif // ASSERT
2920 break;
2921 }
2922 default:
2923 ShouldNotReachHere();
2924 }
2925 }
2926 #endif //PRODUCT
2927
2928 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2929
2930 // Get arguments:
2931 Node* receiver = nullptr;
2932 Node* base = nullptr;
2933 Node* offset = nullptr;
2934 Node* oldval = nullptr;
2935 Node* newval = nullptr;
2936 switch(kind) {
2937 case LS_cmp_swap:
2938 case LS_cmp_swap_weak:
2939 case LS_cmp_exchange: {
2940 const bool two_slot_type = type2size[type] == 2;
2941 receiver = argument(0); // type: oop
2942 base = argument(1); // type: oop
2943 offset = argument(2); // type: long
2944 oldval = argument(4); // type: oop, int, or long
2945 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2946 break;
2947 }
2948 case LS_get_add:
2949 case LS_get_set: {
2950 receiver = argument(0); // type: oop
2951 base = argument(1); // type: oop
2952 offset = argument(2); // type: long
2953 oldval = nullptr;
2954 newval = argument(4); // type: oop, int, or long
2955 break;
2956 }
2957 default:
2958 ShouldNotReachHere();
2959 }
2960
2961 // Build field offset expression.
2962 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2963 // to be plain byte offsets, which are also the same as those accepted
2964 // by oopDesc::field_addr.
2965 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2966 // 32-bit machines ignore the high half of long offsets
2967 offset = ConvL2X(offset);
2968 // Save state and restore on bailout
2969 SavedState old_state(this);
2970 Node* adr = make_unsafe_address(base, offset,type, false);
2971 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2972
2973 Compile::AliasType* alias_type = C->alias_type(adr_type);
2974 BasicType bt = alias_type->basic_type();
2975 if (bt != T_ILLEGAL &&
2976 (is_reference_type(bt) != (type == T_OBJECT))) {
2977 // Don't intrinsify mismatched object accesses.
2978 return false;
2979 }
2980
2981 old_state.discard();
2982
2983 // For CAS, unlike inline_unsafe_access, there seems no point in
2984 // trying to refine types. Just use the coarse types here.
2985 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2986 const Type *value_type = Type::get_const_basic_type(type);
2987
2988 switch (kind) {
2989 case LS_get_set:
2990 case LS_cmp_exchange: {
2991 if (type == T_OBJECT) {
2992 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2993 if (tjp != nullptr) {
2994 value_type = tjp;
2995 }
2996 }
2997 break;
2998 }
2999 case LS_cmp_swap:
3000 case LS_cmp_swap_weak:
3001 case LS_get_add:
3002 break;
3003 default:
3004 ShouldNotReachHere();
3005 }
3006
3007 // Null check receiver.
3008 receiver = null_check(receiver);
3009 if (stopped()) {
3010 return true;
3011 }
3012
3013 int alias_idx = C->get_alias_index(adr_type);
3014
3015 if (is_reference_type(type)) {
3016 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
3017
3018 if (oldval != nullptr && oldval->is_InlineType()) {
3019 // Re-execute the unsafe access if allocation triggers deoptimization.
3020 PreserveReexecuteState preexecs(this);
3021 jvms()->set_should_reexecute(true);
3022 oldval = oldval->as_InlineType()->buffer(this)->get_oop();
3023 }
3024 if (newval != nullptr && newval->is_InlineType()) {
3025 // Re-execute the unsafe access if allocation triggers deoptimization.
3026 PreserveReexecuteState preexecs(this);
3027 jvms()->set_should_reexecute(true);
3028 newval = newval->as_InlineType()->buffer(this)->get_oop();
3029 }
3030
3031 // Transformation of a value which could be null pointer (CastPP #null)
3032 // could be delayed during Parse (for example, in adjust_map_after_if()).
3033 // Execute transformation here to avoid barrier generation in such case.
3034 if (_gvn.type(newval) == TypePtr::NULL_PTR)
3035 newval = _gvn.makecon(TypePtr::NULL_PTR);
3036
3037 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
3038 // Refine the value to a null constant, when it is known to be null
3039 oldval = _gvn.makecon(TypePtr::NULL_PTR);
3040 }
3041 }
3042
3043 Node* result = nullptr;
3044 switch (kind) {
3045 case LS_cmp_exchange: {
3046 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
3047 oldval, newval, value_type, type, decorators);
3048 break;
3049 }
3050 case LS_cmp_swap_weak:
3051 decorators |= C2_WEAK_CMPXCHG;
3052 case LS_cmp_swap: {
3053 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
3054 oldval, newval, value_type, type, decorators);
3055 break;
3056 }
3057 case LS_get_set: {
3058 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
3059 newval, value_type, type, decorators);
3060 break;
3061 }
3062 case LS_get_add: {
3063 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
3064 newval, value_type, type, decorators);
3065 break;
3066 }
3067 default:
3068 ShouldNotReachHere();
3069 }
3070
3071 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3072 set_result(result);
3073 return true;
3074 }
3075
3076 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3077 // Regardless of form, don't allow previous ld/st to move down,
3078 // then issue acquire, release, or volatile mem_bar.
3079 insert_mem_bar(Op_MemBarCPUOrder);
3080 switch(id) {
3081 case vmIntrinsics::_loadFence:
3082 insert_mem_bar(Op_LoadFence);
3083 return true;
3084 case vmIntrinsics::_storeFence:
3085 insert_mem_bar(Op_StoreFence);
3086 return true;
3087 case vmIntrinsics::_storeStoreFence:
3088 insert_mem_bar(Op_StoreStoreFence);
3089 return true;
3090 case vmIntrinsics::_fullFence:
3091 insert_mem_bar(Op_MemBarFull);
3092 return true;
3093 default:
3094 fatal_unexpected_iid(id);
3095 return false;
3096 }
3097 }
3098
3099 // private native int arrayInstanceBaseOffset0(Object[] array);
3100 bool LibraryCallKit::inline_arrayInstanceBaseOffset() {
3101 Node* array = argument(1);
3102 Node* klass_node = load_object_klass(array);
3103
3104 jint layout_con = Klass::_lh_neutral_value;
3105 Node* layout_val = get_layout_helper(klass_node, layout_con);
3106 int layout_is_con = (layout_val == nullptr);
3107
3108 Node* header_size = nullptr;
3109 if (layout_is_con) {
3110 int hsize = Klass::layout_helper_header_size(layout_con);
3111 header_size = intcon(hsize);
3112 } else {
3113 Node* hss = intcon(Klass::_lh_header_size_shift);
3114 Node* hsm = intcon(Klass::_lh_header_size_mask);
3115 header_size = _gvn.transform(new URShiftINode(layout_val, hss));
3116 header_size = _gvn.transform(new AndINode(header_size, hsm));
3117 }
3118 set_result(header_size);
3119 return true;
3120 }
3121
3122 // private native int arrayInstanceIndexScale0(Object[] array);
3123 bool LibraryCallKit::inline_arrayInstanceIndexScale() {
3124 Node* array = argument(1);
3125 Node* klass_node = load_object_klass(array);
3126
3127 jint layout_con = Klass::_lh_neutral_value;
3128 Node* layout_val = get_layout_helper(klass_node, layout_con);
3129 int layout_is_con = (layout_val == nullptr);
3130
3131 Node* element_size = nullptr;
3132 if (layout_is_con) {
3133 int log_element_size = Klass::layout_helper_log2_element_size(layout_con);
3134 int elem_size = 1 << log_element_size;
3135 element_size = intcon(elem_size);
3136 } else {
3137 Node* ess = intcon(Klass::_lh_log2_element_size_shift);
3138 Node* esm = intcon(Klass::_lh_log2_element_size_mask);
3139 Node* log_element_size = _gvn.transform(new URShiftINode(layout_val, ess));
3140 log_element_size = _gvn.transform(new AndINode(log_element_size, esm));
3141 element_size = _gvn.transform(new LShiftINode(intcon(1), log_element_size));
3142 }
3143 set_result(element_size);
3144 return true;
3145 }
3146
3147 // private native int arrayLayout0(Object[] array);
3148 bool LibraryCallKit::inline_arrayLayout() {
3149 RegionNode* region = new RegionNode(2);
3150 Node* phi = new PhiNode(region, TypeInt::POS);
3151
3152 Node* array = argument(1);
3153 Node* klass_node = load_object_klass(array);
3154 generate_refArray_guard(klass_node, region);
3155 if (region->req() == 3) {
3156 phi->add_req(intcon((jint)LayoutKind::REFERENCE));
3157 }
3158
3159 int layout_kind_offset = in_bytes(FlatArrayKlass::layout_kind_offset());
3160 Node* layout_kind_addr = basic_plus_adr(top(), klass_node, layout_kind_offset);
3161 Node* layout_kind = make_load(nullptr, layout_kind_addr, TypeInt::POS, T_INT, MemNode::unordered);
3162
3163 region->init_req(1, control());
3164 phi->init_req(1, layout_kind);
3165
3166 set_control(_gvn.transform(region));
3167 set_result(_gvn.transform(phi));
3168 return true;
3169 }
3170
3171 // private native int[] getFieldMap0(Class <?> c);
3172 // int offset = c._klass._acmp_maps_offset;
3173 // return (int[])c.obj_field(offset);
3174 bool LibraryCallKit::inline_getFieldMap() {
3175 Node* mirror = argument(1);
3176 Node* klass = load_klass_from_mirror(mirror, false, nullptr, 0);
3177
3178 int field_map_offset_offset = in_bytes(InstanceKlass::acmp_maps_offset_offset());
3179 Node* field_map_offset_addr = basic_plus_adr(top(), klass, field_map_offset_offset);
3180 Node* field_map_offset = make_load(nullptr, field_map_offset_addr, TypeInt::INT, T_INT, MemNode::unordered);
3181 field_map_offset = _gvn.transform(ConvI2L(field_map_offset));
3182
3183 Node* map_addr = basic_plus_adr(mirror, field_map_offset);
3184 const TypeAryPtr* val_type = TypeAryPtr::INTS->cast_to_ptr_type(TypePtr::NotNull)->with_offset(0);
3185 // TODO 8350865 Remove this
3186 val_type = val_type->cast_to_not_flat(true)->cast_to_not_null_free(true);
3187 Node* map = access_load_at(mirror, map_addr, TypeAryPtr::INTS, val_type, T_ARRAY, IN_HEAP | MO_UNORDERED);
3188
3189 set_result(map);
3190 return true;
3191 }
3192
3193 bool LibraryCallKit::inline_onspinwait() {
3194 insert_mem_bar(Op_OnSpinWait);
3195 return true;
3196 }
3197
3198 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3199 if (!kls->is_Con()) {
3200 return true;
3201 }
3202 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
3203 if (klsptr == nullptr) {
3204 return true;
3205 }
3206 ciInstanceKlass* ik = klsptr->instance_klass();
3207 // don't need a guard for a klass that is already initialized
3208 return !ik->is_initialized();
3209 }
3210
3211 //----------------------------inline_unsafe_writeback0-------------------------
3212 // public native void Unsafe.writeback0(long address)
3213 bool LibraryCallKit::inline_unsafe_writeback0() {
3214 if (!Matcher::has_match_rule(Op_CacheWB)) {
3215 return false;
3216 }
3217 #ifndef PRODUCT
3218 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
3219 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
3220 ciSignature* sig = callee()->signature();
3221 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
3222 #endif
3223 null_check_receiver(); // null-check, then ignore
3224 Node *addr = argument(1);
3225 addr = new CastX2PNode(addr);
3226 addr = _gvn.transform(addr);
3227 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
3228 flush = _gvn.transform(flush);
3229 set_memory(flush, TypeRawPtr::BOTTOM);
3230 return true;
3231 }
3232
3233 //----------------------------inline_unsafe_writeback0-------------------------
3234 // public native void Unsafe.writeback0(long address)
3235 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
3236 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
3237 return false;
3238 }
3239 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
3240 return false;
3241 }
3242 #ifndef PRODUCT
3243 assert(Matcher::has_match_rule(Op_CacheWB),
3244 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
3245 : "found match rule for CacheWBPostSync but not CacheWB"));
3246
3247 #endif
3248 null_check_receiver(); // null-check, then ignore
3249 Node *sync;
3250 if (is_pre) {
3251 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3252 } else {
3253 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3254 }
3255 sync = _gvn.transform(sync);
3256 set_memory(sync, TypeRawPtr::BOTTOM);
3257 return true;
3258 }
3259
3260 //----------------------------inline_unsafe_allocate---------------------------
3261 // public native Object Unsafe.allocateInstance(Class<?> cls);
3262 bool LibraryCallKit::inline_unsafe_allocate() {
3263
3264 #if INCLUDE_JVMTI
3265 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3266 return false;
3267 }
3268 #endif //INCLUDE_JVMTI
3269
3270 if (callee()->is_static()) return false; // caller must have the capability!
3271
3272 null_check_receiver(); // null-check, then ignore
3273 Node* cls = null_check(argument(1));
3274 if (stopped()) return true;
3275
3276 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
3277 kls = null_check(kls);
3278 if (stopped()) return true; // argument was like int.class
3279
3280 #if INCLUDE_JVMTI
3281 // Don't try to access new allocated obj in the intrinsic.
3282 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
3283 // Deoptimize and allocate in interpreter instead.
3284 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
3285 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
3286 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
3287 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
3288 {
3289 BuildCutout unless(this, tst, PROB_MAX);
3290 uncommon_trap(Deoptimization::Reason_intrinsic,
3291 Deoptimization::Action_make_not_entrant);
3292 }
3293 if (stopped()) {
3294 return true;
3295 }
3296 #endif //INCLUDE_JVMTI
3297
3298 Node* test = nullptr;
3299 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3300 // Note: The argument might still be an illegal value like
3301 // Serializable.class or Object[].class. The runtime will handle it.
3302 // But we must make an explicit check for initialization.
3303 Node* insp = off_heap_plus_addr(kls, in_bytes(InstanceKlass::init_state_offset()));
3304 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3305 // can generate code to load it as unsigned byte.
3306 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
3307 Node* bits = intcon(InstanceKlass::fully_initialized);
3308 test = _gvn.transform(new SubINode(inst, bits));
3309 // The 'test' is non-zero if we need to take a slow path.
3310 }
3311 Node* obj = new_instance(kls, test);
3312 set_result(obj);
3313 return true;
3314 }
3315
3316 //------------------------inline_native_time_funcs--------------
3317 // inline code for System.currentTimeMillis() and System.nanoTime()
3318 // these have the same type and signature
3319 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3320 const TypeFunc* tf = OptoRuntime::void_long_Type();
3321 const TypePtr* no_memory_effects = nullptr;
3322 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3323 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3324 #ifdef ASSERT
3325 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3326 assert(value_top == top(), "second value must be top");
3327 #endif
3328 set_result(value);
3329 return true;
3330 }
3331
3332 //--------------------inline_native_vthread_start_transition--------------------
3333 // inline void startTransition(boolean is_mount);
3334 // inline void startFinalTransition();
3335 // Pseudocode of implementation:
3336 //
3337 // java_lang_Thread::set_is_in_vthread_transition(vthread, true);
3338 // carrier->set_is_in_vthread_transition(true);
3339 // OrderAccess::storeload();
3340 // int disable_requests = java_lang_Thread::vthread_transition_disable_count(vthread)
3341 // + global_vthread_transition_disable_count();
3342 // if (disable_requests > 0) {
3343 // slow path: runtime call
3344 // }
3345 bool LibraryCallKit::inline_native_vthread_start_transition(address funcAddr, const char* funcName, bool is_final_transition) {
3346 Node* vt_oop = must_be_not_null(argument(0), true); // VirtualThread this argument
3347 IdealKit ideal(this);
3348
3349 Node* thread = ideal.thread();
3350 Node* jt_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3351 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3352 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3353 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3354 insert_mem_bar(Op_MemBarStoreLoad);
3355 ideal.sync_kit(this);
3356
3357 Node* global_disable_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::global_vthread_transition_disable_count_address()));
3358 Node* global_disable = ideal.load(ideal.ctrl(), global_disable_addr, TypeInt::INT, T_INT, Compile::AliasIdxRaw, true /*require_atomic_access*/);
3359 Node* vt_disable_addr = basic_plus_adr(vt_oop, java_lang_Thread::vthread_transition_disable_count_offset());
3360 const TypePtr* vt_disable_addr_t = _gvn.type(vt_disable_addr)->is_ptr();
3361 Node* vt_disable = ideal.load(ideal.ctrl(), vt_disable_addr, TypeInt::INT, T_INT, C->get_alias_index(vt_disable_addr_t), true /*require_atomic_access*/);
3362 Node* disabled = _gvn.transform(new AddINode(global_disable, vt_disable));
3363
3364 ideal.if_then(disabled, BoolTest::ne, ideal.ConI(0)); {
3365 sync_kit(ideal);
3366 Node* is_mount = is_final_transition ? ideal.ConI(0) : argument(1);
3367 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3368 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3369 ideal.sync_kit(this);
3370 }
3371 ideal.end_if();
3372
3373 final_sync(ideal);
3374 return true;
3375 }
3376
3377 bool LibraryCallKit::inline_native_vthread_end_transition(address funcAddr, const char* funcName, bool is_first_transition) {
3378 Node* vt_oop = must_be_not_null(argument(0), true); // VirtualThread this argument
3379 IdealKit ideal(this);
3380
3381 Node* _notify_jvmti_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::notify_jvmti_events_address()));
3382 Node* _notify_jvmti = ideal.load(ideal.ctrl(), _notify_jvmti_addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3383
3384 ideal.if_then(_notify_jvmti, BoolTest::eq, ideal.ConI(1)); {
3385 sync_kit(ideal);
3386 Node* is_mount = is_first_transition ? ideal.ConI(1) : argument(1);
3387 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3388 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3389 ideal.sync_kit(this);
3390 } ideal.else_(); {
3391 Node* thread = ideal.thread();
3392 Node* jt_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3393 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3394
3395 sync_kit(ideal);
3396 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3397 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3398 ideal.sync_kit(this);
3399 } ideal.end_if();
3400
3401 final_sync(ideal);
3402 return true;
3403 }
3404
3405 #if INCLUDE_JVMTI
3406
3407 // Always update the is_disable_suspend bit.
3408 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3409 if (!DoJVMTIVirtualThreadTransitions) {
3410 return true;
3411 }
3412 IdealKit ideal(this);
3413
3414 {
3415 // unconditionally update the is_disable_suspend bit in current JavaThread
3416 Node* thread = ideal.thread();
3417 Node* arg = argument(0); // argument for notification
3418 Node* addr = off_heap_plus_addr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3419 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3420
3421 sync_kit(ideal);
3422 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3423 ideal.sync_kit(this);
3424 }
3425 final_sync(ideal);
3426
3427 return true;
3428 }
3429
3430 #endif // INCLUDE_JVMTI
3431
3432 #ifdef JFR_HAVE_INTRINSICS
3433
3434 /**
3435 * if oop->klass != null
3436 * // normal class
3437 * epoch = _epoch_state ? 2 : 1
3438 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3439 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3440 * }
3441 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3442 * else
3443 * // primitive class
3444 * if oop->array_klass != null
3445 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3446 * else
3447 * id = LAST_TYPE_ID + 1 // void class path
3448 * if (!signaled)
3449 * signaled = true
3450 */
3451 bool LibraryCallKit::inline_native_classID() {
3452 Node* cls = argument(0);
3453
3454 IdealKit ideal(this);
3455 #define __ ideal.
3456 IdealVariable result(ideal); __ declarations_done();
3457 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3458 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3459 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3460
3461
3462 __ if_then(kls, BoolTest::ne, null()); {
3463 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3464 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3465
3466 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3467 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3468 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3469 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3470 mask = _gvn.transform(new OrLNode(mask, epoch));
3471 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3472
3473 float unlikely = PROB_UNLIKELY(0.999);
3474 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3475 sync_kit(ideal);
3476 make_runtime_call(RC_LEAF,
3477 OptoRuntime::class_id_load_barrier_Type(),
3478 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3479 "class id load barrier",
3480 TypePtr::BOTTOM,
3481 kls);
3482 ideal.sync_kit(this);
3483 } __ end_if();
3484
3485 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3486 } __ else_(); {
3487 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3488 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3489 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3490 __ if_then(array_kls, BoolTest::ne, null()); {
3491 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3492 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3493 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3494 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3495 } __ else_(); {
3496 // void class case
3497 ideal.set(result, longcon(LAST_TYPE_ID + 1));
3498 } __ end_if();
3499
3500 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3501 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3502 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3503 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3504 } __ end_if();
3505 } __ end_if();
3506
3507 final_sync(ideal);
3508 set_result(ideal.value(result));
3509 #undef __
3510 return true;
3511 }
3512
3513 //------------------------inline_native_jvm_commit------------------
3514 bool LibraryCallKit::inline_native_jvm_commit() {
3515 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3516
3517 // Save input memory and i_o state.
3518 Node* input_memory_state = reset_memory();
3519 set_all_memory(input_memory_state);
3520 Node* input_io_state = i_o();
3521
3522 // TLS.
3523 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3524 // Jfr java buffer.
3525 Node* java_buffer_offset = _gvn.transform(AddPNode::make_off_heap(tls_ptr, MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR))));
3526 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3527 Node* java_buffer_pos_offset = _gvn.transform(AddPNode::make_off_heap(java_buffer, MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET))));
3528
3529 // Load the current value of the notified field in the JfrThreadLocal.
3530 Node* notified_offset = off_heap_plus_addr(tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3531 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3532
3533 // Test for notification.
3534 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3535 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3536 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3537
3538 // True branch, is notified.
3539 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3540 set_control(is_notified);
3541
3542 // Reset notified state.
3543 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3544 Node* notified_reset_memory = reset_memory();
3545
3546 // 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.
3547 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3548 // Convert the machine-word to a long.
3549 Node* current_pos = ConvX2L(current_pos_X);
3550
3551 // False branch, not notified.
3552 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3553 set_control(not_notified);
3554 set_all_memory(input_memory_state);
3555
3556 // Arg is the next position as a long.
3557 Node* arg = argument(0);
3558 // Convert long to machine-word.
3559 Node* next_pos_X = ConvL2X(arg);
3560
3561 // Store the next_position to the underlying jfr java buffer.
3562 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3563
3564 Node* commit_memory = reset_memory();
3565 set_all_memory(commit_memory);
3566
3567 // 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.
3568 Node* java_buffer_flags_offset = _gvn.transform(AddPNode::make_off_heap(java_buffer, MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET))));
3569 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3570 Node* lease_constant = _gvn.intcon(4);
3571
3572 // And flags with lease constant.
3573 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3574
3575 // Branch on lease to conditionalize returning the leased java buffer.
3576 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3577 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3578 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3579
3580 // False branch, not a lease.
3581 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3582
3583 // True branch, is lease.
3584 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3585 set_control(is_lease);
3586
3587 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3588 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3589 OptoRuntime::void_void_Type(),
3590 SharedRuntime::jfr_return_lease(),
3591 "return_lease", TypePtr::BOTTOM);
3592 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3593
3594 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3595 record_for_igvn(lease_compare_rgn);
3596 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3597 record_for_igvn(lease_compare_mem);
3598 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3599 record_for_igvn(lease_compare_io);
3600 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3601 record_for_igvn(lease_result_value);
3602
3603 // Update control and phi nodes.
3604 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3605 lease_compare_rgn->init_req(_false_path, not_lease);
3606
3607 lease_compare_mem->init_req(_true_path, reset_memory());
3608 lease_compare_mem->init_req(_false_path, commit_memory);
3609
3610 lease_compare_io->init_req(_true_path, i_o());
3611 lease_compare_io->init_req(_false_path, input_io_state);
3612
3613 lease_result_value->init_req(_true_path, _gvn.longcon(0)); // if the lease was returned, return 0L.
3614 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3615
3616 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3617 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3618 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3619 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3620
3621 // Update control and phi nodes.
3622 result_rgn->init_req(_true_path, is_notified);
3623 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3624
3625 result_mem->init_req(_true_path, notified_reset_memory);
3626 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3627
3628 result_io->init_req(_true_path, input_io_state);
3629 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3630
3631 result_value->init_req(_true_path, current_pos);
3632 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3633
3634 // Set output state.
3635 set_control(_gvn.transform(result_rgn));
3636 set_all_memory(_gvn.transform(result_mem));
3637 set_i_o(_gvn.transform(result_io));
3638 set_result(result_rgn, result_value);
3639 return true;
3640 }
3641
3642 /*
3643 * The intrinsic is a model of this pseudo-code:
3644 *
3645 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3646 * jobject h_event_writer = tl->java_event_writer();
3647 * if (h_event_writer == nullptr) {
3648 * return nullptr;
3649 * }
3650 * oop threadObj = Thread::threadObj();
3651 * oop vthread = java_lang_Thread::vthread(threadObj);
3652 * traceid tid;
3653 * bool pinVirtualThread;
3654 * bool excluded;
3655 * if (vthread != threadObj) { // i.e. current thread is virtual
3656 * tid = java_lang_Thread::tid(vthread);
3657 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3658 * pinVirtualThread = VMContinuations;
3659 * excluded = vthread_epoch_raw & excluded_mask;
3660 * if (!excluded) {
3661 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3662 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3663 * if (vthread_epoch != current_epoch) {
3664 * write_checkpoint();
3665 * }
3666 * }
3667 * } else {
3668 * tid = java_lang_Thread::tid(threadObj);
3669 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3670 * pinVirtualThread = false;
3671 * excluded = thread_epoch_raw & excluded_mask;
3672 * }
3673 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3674 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3675 * if (tid_in_event_writer != tid) {
3676 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3677 * setField(event_writer, "excluded", excluded);
3678 * setField(event_writer, "threadID", tid);
3679 * }
3680 * return event_writer
3681 */
3682 bool LibraryCallKit::inline_native_getEventWriter() {
3683 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3684
3685 // Save input memory and i_o state.
3686 Node* input_memory_state = reset_memory();
3687 set_all_memory(input_memory_state);
3688 Node* input_io_state = i_o();
3689
3690 // The most significant bit of the u2 is used to denote thread exclusion
3691 Node* excluded_shift = _gvn.intcon(15);
3692 Node* excluded_mask = _gvn.intcon(1 << 15);
3693 // The epoch generation is the range [1-32767]
3694 Node* epoch_mask = _gvn.intcon(32767);
3695
3696 // TLS
3697 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3698
3699 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3700 Node* jobj_ptr = off_heap_plus_addr(tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3701
3702 // Load the eventwriter jobject handle.
3703 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3704
3705 // Null check the jobject handle.
3706 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3707 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3708 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3709
3710 // False path, jobj is null.
3711 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3712
3713 // True path, jobj is not null.
3714 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3715
3716 set_control(jobj_is_not_null);
3717
3718 // Load the threadObj for the CarrierThread.
3719 Node* threadObj = generate_current_thread(tls_ptr);
3720
3721 // Load the vthread.
3722 Node* vthread = generate_virtual_thread(tls_ptr);
3723
3724 // If vthread != threadObj, this is a virtual thread.
3725 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3726 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3727 IfNode* iff_vthread_not_equal_threadObj =
3728 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3729
3730 // False branch, fallback to threadObj.
3731 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3732 set_control(vthread_equal_threadObj);
3733
3734 // Load the tid field from the vthread object.
3735 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3736
3737 // Load the raw epoch value from the threadObj.
3738 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3739 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3740 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3741 TypeInt::CHAR, T_CHAR,
3742 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3743
3744 // Mask off the excluded information from the epoch.
3745 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3746
3747 // True branch, this is a virtual thread.
3748 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3749 set_control(vthread_not_equal_threadObj);
3750
3751 // Load the tid field from the vthread object.
3752 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3753
3754 // Continuation support determines if a virtual thread should be pinned.
3755 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3756 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3757
3758 // Load the raw epoch value from the vthread.
3759 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3760 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3761 TypeInt::CHAR, T_CHAR,
3762 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3763
3764 // Mask off the excluded information from the epoch.
3765 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, excluded_mask));
3766
3767 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3768 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, excluded_mask));
3769 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3770 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3771
3772 // False branch, vthread is excluded, no need to write epoch info.
3773 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3774
3775 // True branch, vthread is included, update epoch info.
3776 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3777 set_control(included);
3778
3779 // Get epoch value.
3780 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, epoch_mask));
3781
3782 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3783 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3784 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3785
3786 // Compare the epoch in the vthread to the current epoch generation.
3787 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3788 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3789 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3790
3791 // False path, epoch is equal, checkpoint information is valid.
3792 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3793
3794 // True path, epoch is not equal, write a checkpoint for the vthread.
3795 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3796
3797 set_control(epoch_is_not_equal);
3798
3799 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3800 // The call also updates the native thread local thread id and the vthread with the current epoch.
3801 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3802 OptoRuntime::jfr_write_checkpoint_Type(),
3803 SharedRuntime::jfr_write_checkpoint(),
3804 "write_checkpoint", TypePtr::BOTTOM);
3805 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3806
3807 // vthread epoch != current epoch
3808 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3809 record_for_igvn(epoch_compare_rgn);
3810 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3811 record_for_igvn(epoch_compare_mem);
3812 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3813 record_for_igvn(epoch_compare_io);
3814
3815 // Update control and phi nodes.
3816 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3817 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3818 epoch_compare_mem->init_req(_true_path, reset_memory());
3819 epoch_compare_mem->init_req(_false_path, input_memory_state);
3820 epoch_compare_io->init_req(_true_path, i_o());
3821 epoch_compare_io->init_req(_false_path, input_io_state);
3822
3823 // excluded != true
3824 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3825 record_for_igvn(exclude_compare_rgn);
3826 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3827 record_for_igvn(exclude_compare_mem);
3828 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3829 record_for_igvn(exclude_compare_io);
3830
3831 // Update control and phi nodes.
3832 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3833 exclude_compare_rgn->init_req(_false_path, excluded);
3834 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3835 exclude_compare_mem->init_req(_false_path, input_memory_state);
3836 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3837 exclude_compare_io->init_req(_false_path, input_io_state);
3838
3839 // vthread != threadObj
3840 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3841 record_for_igvn(vthread_compare_rgn);
3842 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3843 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3844 record_for_igvn(vthread_compare_io);
3845 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3846 record_for_igvn(tid);
3847 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3848 record_for_igvn(exclusion);
3849 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3850 record_for_igvn(pinVirtualThread);
3851
3852 // Update control and phi nodes.
3853 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3854 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3855 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3856 vthread_compare_mem->init_req(_false_path, input_memory_state);
3857 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3858 vthread_compare_io->init_req(_false_path, input_io_state);
3859 tid->init_req(_true_path, vthread_tid);
3860 tid->init_req(_false_path, thread_obj_tid);
3861 exclusion->init_req(_true_path, vthread_is_excluded);
3862 exclusion->init_req(_false_path, threadObj_is_excluded);
3863 pinVirtualThread->init_req(_true_path, continuation_support);
3864 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3865
3866 // Update branch state.
3867 set_control(_gvn.transform(vthread_compare_rgn));
3868 set_all_memory(_gvn.transform(vthread_compare_mem));
3869 set_i_o(_gvn.transform(vthread_compare_io));
3870
3871 // Load the event writer oop by dereferencing the jobject handle.
3872 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3873 assert(klass_EventWriter->is_loaded(), "invariant");
3874 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3875 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3876 const TypeOopPtr* const xtype = aklass->as_instance_type();
3877 Node* jobj_untagged = _gvn.transform(AddPNode::make_off_heap(jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3878 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3879
3880 // Load the current thread id from the event writer object.
3881 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3882 // Get the field offset to, conditionally, store an updated tid value later.
3883 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3884 // Get the field offset to, conditionally, store an updated exclusion value later.
3885 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3886 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3887 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3888
3889 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3890 record_for_igvn(event_writer_tid_compare_rgn);
3891 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3892 record_for_igvn(event_writer_tid_compare_mem);
3893 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3894 record_for_igvn(event_writer_tid_compare_io);
3895
3896 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3897 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3898 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3899 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3900
3901 // False path, tids are the same.
3902 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3903
3904 // True path, tid is not equal, need to update the tid in the event writer.
3905 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3906 record_for_igvn(tid_is_not_equal);
3907
3908 // Store the pin state to the event writer.
3909 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3910
3911 // Store the exclusion state to the event writer.
3912 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3913 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3914
3915 // Store the tid to the event writer.
3916 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3917
3918 // Update control and phi nodes.
3919 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3920 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3921 event_writer_tid_compare_mem->init_req(_true_path, reset_memory());
3922 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3923 event_writer_tid_compare_io->init_req(_true_path, i_o());
3924 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3925
3926 // Result of top level CFG, Memory, IO and Value.
3927 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3928 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3929 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3930 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3931
3932 // Result control.
3933 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3934 result_rgn->init_req(_false_path, jobj_is_null);
3935
3936 // Result memory.
3937 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3938 result_mem->init_req(_false_path, input_memory_state);
3939
3940 // Result IO.
3941 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3942 result_io->init_req(_false_path, input_io_state);
3943
3944 // Result value.
3945 result_value->init_req(_true_path, event_writer); // return event writer oop
3946 result_value->init_req(_false_path, null()); // return null
3947
3948 // Set output state.
3949 set_control(_gvn.transform(result_rgn));
3950 set_all_memory(_gvn.transform(result_mem));
3951 set_i_o(_gvn.transform(result_io));
3952 set_result(result_rgn, result_value);
3953 return true;
3954 }
3955
3956 /*
3957 * The intrinsic is a model of this pseudo-code:
3958 *
3959 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3960 * if (carrierThread != thread) { // is virtual thread
3961 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3962 * bool excluded = vthread_epoch_raw & excluded_mask;
3963 * AtomicAccess::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3964 * AtomicAccess::store(&tl->_contextual_thread_excluded, is_excluded);
3965 * if (!excluded) {
3966 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3967 * AtomicAccess::store(&tl->_vthread_epoch, vthread_epoch);
3968 * }
3969 * AtomicAccess::release_store(&tl->_vthread, true);
3970 * return;
3971 * }
3972 * AtomicAccess::release_store(&tl->_vthread, false);
3973 */
3974 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3975 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3976
3977 Node* input_memory_state = reset_memory();
3978 set_all_memory(input_memory_state);
3979
3980 // The most significant bit of the u2 is used to denote thread exclusion
3981 Node* excluded_mask = _gvn.intcon(1 << 15);
3982 // The epoch generation is the range [1-32767]
3983 Node* epoch_mask = _gvn.intcon(32767);
3984
3985 Node* const carrierThread = generate_current_thread(jt);
3986 // If thread != carrierThread, this is a virtual thread.
3987 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3988 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3989 IfNode* iff_thread_not_equal_carrierThread =
3990 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3991
3992 Node* vthread_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3993
3994 // False branch, is carrierThread.
3995 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3996 // Store release
3997 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3998
3999 set_all_memory(input_memory_state);
4000
4001 // True branch, is virtual thread.
4002 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
4003 set_control(thread_not_equal_carrierThread);
4004
4005 // Load the raw epoch value from the vthread.
4006 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
4007 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
4008 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
4009
4010 // Mask off the excluded information from the epoch.
4011 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, excluded_mask));
4012
4013 // Load the tid field from the thread.
4014 Node* tid = load_field_from_object(thread, "tid", "J");
4015
4016 // Store the vthread tid to the jfr thread local.
4017 Node* thread_id_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
4018 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
4019
4020 // Branch is_excluded to conditionalize updating the epoch .
4021 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, excluded_mask));
4022 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
4023 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
4024
4025 // True branch, vthread is excluded, no need to write epoch info.
4026 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
4027 set_control(excluded);
4028 Node* vthread_is_excluded = _gvn.intcon(1);
4029
4030 // False branch, vthread is included, update epoch info.
4031 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
4032 set_control(included);
4033 Node* vthread_is_included = _gvn.intcon(0);
4034
4035 // Get epoch value.
4036 Node* epoch = _gvn.transform(new AndINode(epoch_raw, epoch_mask));
4037
4038 // Store the vthread epoch to the jfr thread local.
4039 Node* vthread_epoch_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
4040 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
4041
4042 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
4043 record_for_igvn(excluded_rgn);
4044 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
4045 record_for_igvn(excluded_mem);
4046 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
4047 record_for_igvn(exclusion);
4048
4049 // Merge the excluded control and memory.
4050 excluded_rgn->init_req(_true_path, excluded);
4051 excluded_rgn->init_req(_false_path, included);
4052 excluded_mem->init_req(_true_path, tid_memory);
4053 excluded_mem->init_req(_false_path, included_memory);
4054 exclusion->init_req(_true_path, vthread_is_excluded);
4055 exclusion->init_req(_false_path, vthread_is_included);
4056
4057 // Set intermediate state.
4058 set_control(_gvn.transform(excluded_rgn));
4059 set_all_memory(excluded_mem);
4060
4061 // Store the vthread exclusion state to the jfr thread local.
4062 Node* thread_local_excluded_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
4063 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
4064
4065 // Store release
4066 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
4067
4068 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
4069 record_for_igvn(thread_compare_rgn);
4070 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
4071 record_for_igvn(thread_compare_mem);
4072 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
4073 record_for_igvn(vthread);
4074
4075 // Merge the thread_compare control and memory.
4076 thread_compare_rgn->init_req(_true_path, control());
4077 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
4078 thread_compare_mem->init_req(_true_path, vthread_true_memory);
4079 thread_compare_mem->init_req(_false_path, vthread_false_memory);
4080
4081 // Set output state.
4082 set_control(_gvn.transform(thread_compare_rgn));
4083 set_all_memory(_gvn.transform(thread_compare_mem));
4084 }
4085
4086 #endif // JFR_HAVE_INTRINSICS
4087
4088 //------------------------inline_native_currentCarrierThread------------------
4089 bool LibraryCallKit::inline_native_currentCarrierThread() {
4090 Node* junk = nullptr;
4091 set_result(generate_current_thread(junk));
4092 return true;
4093 }
4094
4095 //------------------------inline_native_currentThread------------------
4096 bool LibraryCallKit::inline_native_currentThread() {
4097 Node* junk = nullptr;
4098 set_result(generate_virtual_thread(junk));
4099 return true;
4100 }
4101
4102 //------------------------inline_native_setVthread------------------
4103 bool LibraryCallKit::inline_native_setCurrentThread() {
4104 assert(C->method()->changes_current_thread(),
4105 "method changes current Thread but is not annotated ChangesCurrentThread");
4106 Node* arr = argument(1);
4107 Node* thread = _gvn.transform(new ThreadLocalNode());
4108 Node* p = off_heap_plus_addr(thread, in_bytes(JavaThread::vthread_offset()));
4109 Node* thread_obj_handle
4110 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
4111 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
4112 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
4113
4114 // Change the _monitor_owner_id of the JavaThread
4115 Node* tid = load_field_from_object(arr, "tid", "J");
4116 Node* monitor_owner_id_offset = off_heap_plus_addr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
4117 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
4118
4119 JFR_ONLY(extend_setCurrentThread(thread, arr);)
4120 return true;
4121 }
4122
4123 const Type* LibraryCallKit::scopedValueCache_type() {
4124 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
4125 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
4126 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS, /* stable= */ false, /* flat= */ false, /* not_flat= */ true, /* not_null_free= */ true, true);
4127
4128 // Because we create the scopedValue cache lazily we have to make the
4129 // type of the result BotPTR.
4130 bool xk = etype->klass_is_exact();
4131 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, TypeAryPtr::Offset(0));
4132 return objects_type;
4133 }
4134
4135 Node* LibraryCallKit::scopedValueCache_helper() {
4136 Node* thread = _gvn.transform(new ThreadLocalNode());
4137 Node* p = off_heap_plus_addr(thread, in_bytes(JavaThread::scopedValueCache_offset()));
4138 // We cannot use immutable_memory() because we might flip onto a
4139 // different carrier thread, at which point we'll need to use that
4140 // carrier thread's cache.
4141 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
4142 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
4143 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
4144 }
4145
4146 //------------------------inline_native_scopedValueCache------------------
4147 bool LibraryCallKit::inline_native_scopedValueCache() {
4148 Node* cache_obj_handle = scopedValueCache_helper();
4149 const Type* objects_type = scopedValueCache_type();
4150 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
4151
4152 return true;
4153 }
4154
4155 //------------------------inline_native_setScopedValueCache------------------
4156 bool LibraryCallKit::inline_native_setScopedValueCache() {
4157 Node* arr = argument(0);
4158 Node* cache_obj_handle = scopedValueCache_helper();
4159 const Type* objects_type = scopedValueCache_type();
4160
4161 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
4162 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
4163
4164 return true;
4165 }
4166
4167 //------------------------inline_native_Continuation_pin and unpin-----------
4168
4169 // Shared implementation routine for both pin and unpin.
4170 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
4171 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
4172
4173 // Save input memory.
4174 Node* input_memory_state = reset_memory();
4175 set_all_memory(input_memory_state);
4176
4177 // TLS
4178 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
4179 Node* last_continuation_offset = off_heap_plus_addr(tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
4180 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
4181
4182 // Null check the last continuation object.
4183 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
4184 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
4185 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
4186
4187 // False path, last continuation is null.
4188 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
4189
4190 // True path, last continuation is not null.
4191 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
4192
4193 set_control(continuation_is_not_null);
4194
4195 // Load the pin count from the last continuation.
4196 Node* pin_count_offset = off_heap_plus_addr(last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
4197 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
4198
4199 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
4200 Node* pin_count_rhs;
4201 if (unpin) {
4202 pin_count_rhs = _gvn.intcon(0);
4203 } else {
4204 pin_count_rhs = _gvn.intcon(UINT32_MAX);
4205 }
4206 Node* pin_count_cmp = _gvn.transform(new CmpUNode(pin_count, pin_count_rhs));
4207 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
4208 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
4209
4210 // True branch, pin count over/underflow.
4211 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
4212 {
4213 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
4214 // which will throw IllegalStateException for pin count over/underflow.
4215 // No memory changed so far - we can use memory create by reset_memory()
4216 // at the beginning of this intrinsic. No need to call reset_memory() again.
4217 PreserveJVMState pjvms(this);
4218 set_control(pin_count_over_underflow);
4219 uncommon_trap(Deoptimization::Reason_intrinsic,
4220 Deoptimization::Action_none);
4221 assert(stopped(), "invariant");
4222 }
4223
4224 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
4225 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
4226 set_control(valid_pin_count);
4227
4228 Node* next_pin_count;
4229 if (unpin) {
4230 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
4231 } else {
4232 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
4233 }
4234
4235 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
4236
4237 // Result of top level CFG and Memory.
4238 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
4239 record_for_igvn(result_rgn);
4240 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
4241 record_for_igvn(result_mem);
4242
4243 result_rgn->init_req(_true_path, valid_pin_count);
4244 result_rgn->init_req(_false_path, continuation_is_null);
4245 result_mem->init_req(_true_path, reset_memory());
4246 result_mem->init_req(_false_path, input_memory_state);
4247
4248 // Set output state.
4249 set_control(_gvn.transform(result_rgn));
4250 set_all_memory(_gvn.transform(result_mem));
4251
4252 return true;
4253 }
4254
4255 //---------------------------load_mirror_from_klass----------------------------
4256 // Given a klass oop, load its java mirror (a java.lang.Class oop).
4257 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
4258 Node* p = off_heap_plus_addr(klass, in_bytes(Klass::java_mirror_offset()));
4259 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4260 // mirror = ((OopHandle)mirror)->resolve();
4261 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
4262 }
4263
4264 //-----------------------load_klass_from_mirror_common-------------------------
4265 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
4266 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
4267 // and branch to the given path on the region.
4268 // If never_see_null, take an uncommon trap on null, so we can optimistically
4269 // compile for the non-null case.
4270 // If the region is null, force never_see_null = true.
4271 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
4272 bool never_see_null,
4273 RegionNode* region,
4274 int null_path,
4275 int offset) {
4276 if (region == nullptr) never_see_null = true;
4277 Node* p = basic_plus_adr(mirror, offset);
4278 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4279 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
4280 Node* null_ctl = top();
4281 kls = null_check_oop(kls, &null_ctl, never_see_null);
4282 if (region != nullptr) {
4283 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
4284 region->init_req(null_path, null_ctl);
4285 } else {
4286 assert(null_ctl == top(), "no loose ends");
4287 }
4288 return kls;
4289 }
4290
4291 //--------------------(inline_native_Class_query helpers)---------------------
4292 // Use this for JVM_ACC_INTERFACE.
4293 // Fall through if (mods & mask) == bits, take the guard otherwise.
4294 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
4295 ByteSize offset, const Type* type, BasicType bt) {
4296 // Branch around if the given klass has the given modifier bit set.
4297 // Like generate_guard, adds a new path onto the region.
4298 Node* modp = off_heap_plus_addr(kls, in_bytes(offset));
4299 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
4300 Node* mask = intcon(modifier_mask);
4301 Node* bits = intcon(modifier_bits);
4302 Node* mbit = _gvn.transform(new AndINode(mods, mask));
4303 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
4304 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
4305 return generate_fair_guard(bol, region);
4306 }
4307
4308 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
4309 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
4310 InstanceKlass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
4311 }
4312
4313 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
4314 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
4315 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
4316 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
4317 }
4318
4319 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
4320 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
4321 }
4322
4323 //-------------------------inline_native_Class_query-------------------
4324 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
4325 const Type* return_type = TypeInt::BOOL;
4326 Node* prim_return_value = top(); // what happens if it's a primitive class?
4327 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4328 bool expect_prim = false; // most of these guys expect to work on refs
4329
4330 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
4331
4332 Node* mirror = argument(0);
4333 Node* obj = top();
4334
4335 switch (id) {
4336 case vmIntrinsics::_isInstance:
4337 // nothing is an instance of a primitive type
4338 prim_return_value = intcon(0);
4339 obj = argument(1);
4340 break;
4341 case vmIntrinsics::_isHidden:
4342 prim_return_value = intcon(0);
4343 break;
4344 case vmIntrinsics::_getSuperclass:
4345 prim_return_value = null();
4346 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
4347 break;
4348 default:
4349 fatal_unexpected_iid(id);
4350 break;
4351 }
4352
4353 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4354 if (mirror_con == nullptr) return false; // cannot happen?
4355
4356 #ifndef PRODUCT
4357 if (C->print_intrinsics() || C->print_inlining()) {
4358 ciType* k = mirror_con->java_mirror_type();
4359 if (k) {
4360 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4361 k->print_name();
4362 tty->cr();
4363 }
4364 }
4365 #endif
4366
4367 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4368 RegionNode* region = new RegionNode(PATH_LIMIT);
4369 record_for_igvn(region);
4370 PhiNode* phi = new PhiNode(region, return_type);
4371
4372 // The mirror will never be null of Reflection.getClassAccessFlags, however
4373 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4374 // if it is. See bug 4774291.
4375
4376 // For Reflection.getClassAccessFlags(), the null check occurs in
4377 // the wrong place; see inline_unsafe_access(), above, for a similar
4378 // situation.
4379 mirror = null_check(mirror);
4380 // If mirror or obj is dead, only null-path is taken.
4381 if (stopped()) return true;
4382
4383 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4384
4385 // Now load the mirror's klass metaobject, and null-check it.
4386 // Side-effects region with the control path if the klass is null.
4387 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4388 // If kls is null, we have a primitive mirror.
4389 phi->init_req(_prim_path, prim_return_value);
4390 if (stopped()) { set_result(region, phi); return true; }
4391 bool safe_for_replace = (region->in(_prim_path) == top());
4392
4393 Node* p; // handy temp
4394 Node* null_ctl;
4395
4396 // Now that we have the non-null klass, we can perform the real query.
4397 // For constant classes, the query will constant-fold in LoadNode::Value.
4398 Node* query_value = top();
4399 switch (id) {
4400 case vmIntrinsics::_isInstance:
4401 // nothing is an instance of a primitive type
4402 query_value = gen_instanceof(obj, kls, safe_for_replace);
4403 break;
4404
4405 case vmIntrinsics::_isHidden:
4406 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4407 if (generate_hidden_class_guard(kls, region) != nullptr)
4408 // A guard was added. If the guard is taken, it was an hidden class.
4409 phi->add_req(intcon(1));
4410 // If we fall through, it's a plain class.
4411 query_value = intcon(0);
4412 break;
4413
4414
4415 case vmIntrinsics::_getSuperclass:
4416 // The rules here are somewhat unfortunate, but we can still do better
4417 // with random logic than with a JNI call.
4418 // Interfaces store null or Object as _super, but must report null.
4419 // Arrays store an intermediate super as _super, but must report Object.
4420 // Other types can report the actual _super.
4421 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4422 if (generate_array_guard(kls, region) != nullptr) {
4423 // A guard was added. If the guard is taken, it was an array.
4424 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4425 }
4426 // Check for interface after array since this checks AccessFlags offset into InstanceKlass.
4427 // In other words, we are accessing subtype-specific information, so we need to determine the subtype first.
4428 if (generate_interface_guard(kls, region) != nullptr) {
4429 // A guard was added. If the guard is taken, it was an interface.
4430 phi->add_req(null());
4431 }
4432 // If we fall through, it's a plain class. Get its _super.
4433 if (!stopped()) {
4434 p = basic_plus_adr(top(), kls, in_bytes(Klass::super_offset()));
4435 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4436 null_ctl = top();
4437 kls = null_check_oop(kls, &null_ctl);
4438 if (null_ctl != top()) {
4439 // If the guard is taken, Object.superClass is null (both klass and mirror).
4440 region->add_req(null_ctl);
4441 phi ->add_req(null());
4442 }
4443 if (!stopped()) {
4444 query_value = load_mirror_from_klass(kls);
4445 }
4446 }
4447 break;
4448
4449 default:
4450 fatal_unexpected_iid(id);
4451 break;
4452 }
4453
4454 // Fall-through is the normal case of a query to a real class.
4455 phi->init_req(1, query_value);
4456 region->init_req(1, control());
4457
4458 C->set_has_split_ifs(true); // Has chance for split-if optimization
4459 set_result(region, phi);
4460 return true;
4461 }
4462
4463
4464 //-------------------------inline_Class_cast-------------------
4465 bool LibraryCallKit::inline_Class_cast() {
4466 Node* mirror = argument(0); // Class
4467 Node* obj = argument(1);
4468 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4469 if (mirror_con == nullptr) {
4470 return false; // dead path (mirror->is_top()).
4471 }
4472 if (obj == nullptr || obj->is_top()) {
4473 return false; // dead path
4474 }
4475 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4476
4477 // First, see if Class.cast() can be folded statically.
4478 // java_mirror_type() returns non-null for compile-time Class constants.
4479 ciType* tm = mirror_con->java_mirror_type();
4480 if (tm != nullptr && tm->is_klass() &&
4481 tp != nullptr) {
4482 if (!tp->is_loaded()) {
4483 // Don't use intrinsic when class is not loaded.
4484 return false;
4485 } else {
4486 const TypeKlassPtr* tklass = TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces);
4487 int static_res = C->static_subtype_check(tklass, tp->as_klass_type());
4488 if (static_res == Compile::SSC_always_true) {
4489 // isInstance() is true - fold the code.
4490 set_result(obj);
4491 return true;
4492 } else if (static_res == Compile::SSC_always_false) {
4493 // Don't use intrinsic, have to throw ClassCastException.
4494 // If the reference is null, the non-intrinsic bytecode will
4495 // be optimized appropriately.
4496 return false;
4497 }
4498 }
4499 }
4500
4501 // Bailout intrinsic and do normal inlining if exception path is frequent.
4502 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4503 return false;
4504 }
4505
4506 // Generate dynamic checks.
4507 // Class.cast() is java implementation of _checkcast bytecode.
4508 // Do checkcast (Parse::do_checkcast()) optimizations here.
4509
4510 mirror = null_check(mirror);
4511 // If mirror is dead, only null-path is taken.
4512 if (stopped()) {
4513 return true;
4514 }
4515
4516 // Not-subtype or the mirror's klass ptr is nullptr (in case it is a primitive).
4517 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
4518 RegionNode* region = new RegionNode(PATH_LIMIT);
4519 record_for_igvn(region);
4520
4521 // Now load the mirror's klass metaobject, and null-check it.
4522 // If kls is null, we have a primitive mirror and
4523 // nothing is an instance of a primitive type.
4524 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4525
4526 Node* res = top();
4527 Node* io = i_o();
4528 Node* mem = merged_memory();
4529 SafePointNode* new_cast_failure_map = nullptr;
4530
4531 if (!stopped()) {
4532
4533 Node* bad_type_ctrl = top();
4534 // Do checkcast optimizations.
4535 res = gen_checkcast(obj, kls, &bad_type_ctrl, &new_cast_failure_map);
4536 region->init_req(_bad_type_path, bad_type_ctrl);
4537 }
4538 if (region->in(_prim_path) != top() ||
4539 region->in(_bad_type_path) != top() ||
4540 region->in(_npe_path) != top()) {
4541 // Let Interpreter throw ClassCastException.
4542 PreserveJVMState pjvms(this);
4543 if (new_cast_failure_map != nullptr) {
4544 // The current map on the success path could have been modified. Use the dedicated failure path map.
4545 set_map(new_cast_failure_map);
4546 }
4547 set_control(_gvn.transform(region));
4548 // Set IO and memory because gen_checkcast may override them when buffering inline types
4549 set_i_o(io);
4550 set_all_memory(mem);
4551 uncommon_trap(Deoptimization::Reason_intrinsic,
4552 Deoptimization::Action_maybe_recompile);
4553 }
4554 if (!stopped()) {
4555 set_result(res);
4556 }
4557 return true;
4558 }
4559
4560
4561 //--------------------------inline_native_subtype_check------------------------
4562 // This intrinsic takes the JNI calls out of the heart of
4563 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4564 bool LibraryCallKit::inline_native_subtype_check() {
4565 // Pull both arguments off the stack.
4566 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4567 args[0] = argument(0);
4568 args[1] = argument(1);
4569 Node* klasses[2]; // corresponding Klasses: superk, subk
4570 klasses[0] = klasses[1] = top();
4571
4572 enum {
4573 // A full decision tree on {superc is prim, subc is prim}:
4574 _prim_0_path = 1, // {P,N} => false
4575 // {P,P} & superc!=subc => false
4576 _prim_same_path, // {P,P} & superc==subc => true
4577 _prim_1_path, // {N,P} => false
4578 _ref_subtype_path, // {N,N} & subtype check wins => true
4579 _both_ref_path, // {N,N} & subtype check loses => false
4580 PATH_LIMIT
4581 };
4582
4583 RegionNode* region = new RegionNode(PATH_LIMIT);
4584 RegionNode* prim_region = new RegionNode(2);
4585 Node* phi = new PhiNode(region, TypeInt::BOOL);
4586 record_for_igvn(region);
4587 record_for_igvn(prim_region);
4588
4589 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4590 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4591 int class_klass_offset = java_lang_Class::klass_offset();
4592
4593 // First null-check both mirrors and load each mirror's klass metaobject.
4594 int which_arg;
4595 for (which_arg = 0; which_arg <= 1; which_arg++) {
4596 Node* arg = args[which_arg];
4597 arg = null_check(arg);
4598 if (stopped()) break;
4599 args[which_arg] = arg;
4600
4601 Node* p = basic_plus_adr(arg, class_klass_offset);
4602 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4603 klasses[which_arg] = _gvn.transform(kls);
4604 }
4605
4606 // Having loaded both klasses, test each for null.
4607 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4608 for (which_arg = 0; which_arg <= 1; which_arg++) {
4609 Node* kls = klasses[which_arg];
4610 Node* null_ctl = top();
4611 kls = null_check_oop(kls, &null_ctl, never_see_null);
4612 if (which_arg == 0) {
4613 prim_region->init_req(1, null_ctl);
4614 } else {
4615 region->init_req(_prim_1_path, null_ctl);
4616 }
4617 if (stopped()) break;
4618 klasses[which_arg] = kls;
4619 }
4620
4621 if (!stopped()) {
4622 // now we have two reference types, in klasses[0..1]
4623 Node* subk = klasses[1]; // the argument to isAssignableFrom
4624 Node* superk = klasses[0]; // the receiver
4625 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4626 region->set_req(_ref_subtype_path, control());
4627 }
4628
4629 // If both operands are primitive (both klasses null), then
4630 // we must return true when they are identical primitives.
4631 // It is convenient to test this after the first null klass check.
4632 // This path is also used if superc is a value mirror.
4633 set_control(_gvn.transform(prim_region));
4634 if (!stopped()) {
4635 // Since superc is primitive, make a guard for the superc==subc case.
4636 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4637 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4638 generate_fair_guard(bol_eq, region);
4639 if (region->req() == PATH_LIMIT+1) {
4640 // A guard was added. If the added guard is taken, superc==subc.
4641 region->swap_edges(PATH_LIMIT, _prim_same_path);
4642 region->del_req(PATH_LIMIT);
4643 }
4644 region->set_req(_prim_0_path, control()); // Not equal after all.
4645 }
4646
4647 // these are the only paths that produce 'true':
4648 phi->set_req(_prim_same_path, intcon(1));
4649 phi->set_req(_ref_subtype_path, intcon(1));
4650
4651 // pull together the cases:
4652 assert(region->req() == PATH_LIMIT, "sane region");
4653 for (uint i = 1; i < region->req(); i++) {
4654 Node* ctl = region->in(i);
4655 if (ctl == nullptr || ctl == top()) {
4656 region->set_req(i, top());
4657 phi ->set_req(i, top());
4658 } else if (phi->in(i) == nullptr) {
4659 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4660 }
4661 }
4662
4663 set_control(_gvn.transform(region));
4664 set_result(_gvn.transform(phi));
4665 return true;
4666 }
4667
4668 //---------------------generate_array_guard_common------------------------
4669 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind, Node** obj) {
4670
4671 if (stopped()) {
4672 return nullptr;
4673 }
4674
4675 // Like generate_guard, adds a new path onto the region.
4676 jint layout_con = 0;
4677 Node* layout_val = get_layout_helper(kls, layout_con);
4678 if (layout_val == nullptr) {
4679 bool query = 0;
4680 switch(kind) {
4681 case RefArray: query = Klass::layout_helper_is_refArray(layout_con); break;
4682 case NonRefArray: query = !Klass::layout_helper_is_refArray(layout_con); break;
4683 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
4684 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
4685 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
4686 default:
4687 ShouldNotReachHere();
4688 }
4689 if (!query) {
4690 return nullptr; // never a branch
4691 } else { // always a branch
4692 Node* always_branch = control();
4693 if (region != nullptr)
4694 region->add_req(always_branch);
4695 set_control(top());
4696 return always_branch;
4697 }
4698 }
4699 unsigned int value = 0;
4700 BoolTest::mask btest = BoolTest::illegal;
4701 switch(kind) {
4702 case RefArray:
4703 case NonRefArray: {
4704 value = Klass::_lh_array_tag_ref_value;
4705 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4706 btest = (kind == RefArray) ? BoolTest::eq : BoolTest::ne;
4707 break;
4708 }
4709 case TypeArray: {
4710 value = Klass::_lh_array_tag_type_value;
4711 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4712 btest = BoolTest::eq;
4713 break;
4714 }
4715 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
4716 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
4717 default:
4718 ShouldNotReachHere();
4719 }
4720 // Now test the correct condition.
4721 jint nval = (jint)value;
4722 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4723 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4724 Node* ctrl = generate_fair_guard(bol, region);
4725 Node* is_array_ctrl = kind == NonArray ? control() : ctrl;
4726 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4727 // Keep track of the fact that 'obj' is an array to prevent
4728 // array specific accesses from floating above the guard.
4729 *obj = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4730 }
4731 return ctrl;
4732 }
4733
4734 // public static native Object[] ValueClass::newNullRestrictedAtomicArray(Class<?> componentType, int length, Object initVal);
4735 // public static native Object[] ValueClass::newNullRestrictedNonAtomicArray(Class<?> componentType, int length, Object initVal);
4736 // public static native Object[] ValueClass::newNullableAtomicArray(Class<?> componentType, int length);
4737 bool LibraryCallKit::inline_newArray(bool null_free, bool atomic) {
4738 assert(null_free || atomic, "nullable implies atomic");
4739 Node* componentType = argument(0);
4740 Node* length = argument(1);
4741 Node* init_val = null_free ? argument(2) : nullptr;
4742
4743 const TypeInstPtr* tp = _gvn.type(componentType)->isa_instptr();
4744 if (tp != nullptr) {
4745 ciInstanceKlass* ik = tp->instance_klass();
4746 if (ik == C->env()->Class_klass()) {
4747 ciType* t = tp->java_mirror_type();
4748 if (t != nullptr && t->is_inlinetype()) {
4749
4750 ciArrayKlass* array_klass = ciArrayKlass::make(t, null_free, atomic, true);
4751 assert(array_klass->is_elem_null_free() == null_free, "inconsistency");
4752
4753 // TOOD 8350865 ZGC needs card marks on initializing oop stores
4754 if (UseZGC && null_free && !array_klass->is_flat_array_klass()) {
4755 return false;
4756 }
4757
4758 if (array_klass->is_loaded() && array_klass->element_klass()->as_inline_klass()->is_initialized()) {
4759 const TypeAryKlassPtr* array_klass_type = TypeAryKlassPtr::make(array_klass, Type::trust_interfaces);
4760 if (null_free) {
4761 if (init_val->is_InlineType()) {
4762 if (array_klass_type->is_flat() && init_val->as_InlineType()->is_all_zero(&gvn(), /* flat */ true)) {
4763 // Zeroing is enough because the init value is the all-zero value
4764 init_val = nullptr;
4765 } else {
4766 init_val = init_val->as_InlineType()->buffer(this);
4767 }
4768 }
4769 // TODO 8350865 Should we add a check of the init_val type (maybe in debug only + halt)?
4770 // If we insert a checkcast here, we can be sure that init_val is an InlineTypeNode, so
4771 // when we folded a field load from an allocation (e.g. during escape analysis), we can
4772 // remove the check init_val->is_InlineType().
4773 }
4774 Node* obj = new_array(makecon(array_klass_type), length, 0, nullptr, false, init_val);
4775 const TypeAryPtr* arytype = gvn().type(obj)->is_aryptr();
4776 assert(arytype->is_null_free() == null_free, "inconsistency");
4777 assert(arytype->is_not_null_free() == !null_free, "inconsistency");
4778 set_result(obj);
4779 return true;
4780 }
4781 }
4782 }
4783 }
4784 return false;
4785 }
4786
4787 // public static native boolean ValueClass::isFlatArray(Object array);
4788 // public static native boolean ValueClass::isNullRestrictedArray(Object array);
4789 // public static native boolean ValueClass::isAtomicArray(Object array);
4790 bool LibraryCallKit::inline_getArrayProperties(ArrayPropertiesCheck check) {
4791 Node* array = argument(0);
4792
4793 Node* bol;
4794 switch(check) {
4795 case IsFlat:
4796 // TODO 8350865 Use the object version here instead of loading the klass
4797 // The problem is that PhaseMacroExpand::expand_flatarraycheck_node can only handle some IR shapes and will fail, for example, if the bol is directly wired to a ReturnNode
4798 bol = flat_array_test(load_object_klass(array));
4799 break;
4800 case IsNullRestricted:
4801 bol = null_free_array_test(array);
4802 break;
4803 case IsAtomic:
4804 // TODO 8350865 Implement this. It's a bit more complicated, see conditions in JVM_IsAtomicArray
4805 // Enable TestIntrinsics::test87/88 once this is implemented
4806 // bol = null_free_atomic_array_test
4807 return false;
4808 default:
4809 ShouldNotReachHere();
4810 }
4811
4812 Node* res = gvn().transform(new CMoveINode(bol, intcon(0), intcon(1), TypeInt::BOOL));
4813 set_result(res);
4814 return true;
4815 }
4816
4817 // Load the default refined array klass from an ObjArrayKlass. This relies on the first entry in the
4818 // '_next_refined_array_klass' linked list being the default (see ObjArrayKlass::klass_with_properties).
4819 Node* LibraryCallKit::load_default_refined_array_klass(Node* klass_node, bool type_array_guard) {
4820 RegionNode* region = new RegionNode(2);
4821 Node* phi = new PhiNode(region, TypeInstKlassPtr::OBJECT_OR_NULL);
4822
4823 if (type_array_guard) {
4824 generate_typeArray_guard(klass_node, region);
4825 if (region->req() == 3) {
4826 phi->add_req(klass_node);
4827 }
4828 }
4829 Node* adr_refined_klass = basic_plus_adr(top(), klass_node, in_bytes(ObjArrayKlass::next_refined_array_klass_offset()));
4830 Node* refined_klass = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), adr_refined_klass, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4831
4832 // Can be null if not initialized yet, just deopt
4833 Node* null_ctl = top();
4834 refined_klass = null_check_oop(refined_klass, &null_ctl, /* never_see_null= */ true);
4835
4836 region->init_req(1, control());
4837 phi->init_req(1, refined_klass);
4838
4839 set_control(_gvn.transform(region));
4840 return _gvn.transform(phi);
4841 }
4842
4843 // Load the non-refined array klass from an ObjArrayKlass.
4844 Node* LibraryCallKit::load_non_refined_array_klass(Node* klass_node) {
4845 const TypeAryKlassPtr* ary_klass_ptr = _gvn.type(klass_node)->isa_aryklassptr();
4846 if (ary_klass_ptr != nullptr && ary_klass_ptr->klass_is_exact()) {
4847 return _gvn.makecon(ary_klass_ptr->cast_to_refined_array_klass_ptr(false));
4848 }
4849
4850 RegionNode* region = new RegionNode(2);
4851 Node* phi = new PhiNode(region, TypeInstKlassPtr::OBJECT);
4852
4853 generate_typeArray_guard(klass_node, region);
4854 if (region->req() == 3) {
4855 phi->add_req(klass_node);
4856 }
4857 Node* super_adr = basic_plus_adr(top(), klass_node, in_bytes(Klass::super_offset()));
4858 Node* super_klass = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), super_adr, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT));
4859
4860 region->init_req(1, control());
4861 phi->init_req(1, super_klass);
4862
4863 set_control(_gvn.transform(region));
4864 return _gvn.transform(phi);
4865 }
4866
4867 //-----------------------inline_native_newArray--------------------------
4868 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
4869 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4870 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4871 Node* mirror;
4872 Node* count_val;
4873 if (uninitialized) {
4874 null_check_receiver();
4875 mirror = argument(1);
4876 count_val = argument(2);
4877 } else {
4878 mirror = argument(0);
4879 count_val = argument(1);
4880 }
4881
4882 mirror = null_check(mirror);
4883 // If mirror or obj is dead, only null-path is taken.
4884 if (stopped()) return true;
4885
4886 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4887 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4888 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4889 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4890 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4891
4892 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4893 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4894 result_reg, _slow_path);
4895 Node* normal_ctl = control();
4896 Node* no_array_ctl = result_reg->in(_slow_path);
4897
4898 // Generate code for the slow case. We make a call to newArray().
4899 set_control(no_array_ctl);
4900 if (!stopped()) {
4901 // Either the input type is void.class, or else the
4902 // array klass has not yet been cached. Either the
4903 // ensuing call will throw an exception, or else it
4904 // will cache the array klass for next time.
4905 PreserveJVMState pjvms(this);
4906 CallJavaNode* slow_call = nullptr;
4907 if (uninitialized) {
4908 // Generate optimized virtual call (holder class 'Unsafe' is final)
4909 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4910 } else {
4911 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4912 }
4913 Node* slow_result = set_results_for_java_call(slow_call);
4914 // this->control() comes from set_results_for_java_call
4915 result_reg->set_req(_slow_path, control());
4916 result_val->set_req(_slow_path, slow_result);
4917 result_io ->set_req(_slow_path, i_o());
4918 result_mem->set_req(_slow_path, reset_memory());
4919 }
4920
4921 set_control(normal_ctl);
4922 if (!stopped()) {
4923 // Normal case: The array type has been cached in the java.lang.Class.
4924 // The following call works fine even if the array type is polymorphic.
4925 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4926
4927 klass_node = load_default_refined_array_klass(klass_node);
4928
4929 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4930 result_reg->init_req(_normal_path, control());
4931 result_val->init_req(_normal_path, obj);
4932 result_io ->init_req(_normal_path, i_o());
4933 result_mem->init_req(_normal_path, reset_memory());
4934
4935 if (uninitialized) {
4936 // Mark the allocation so that zeroing is skipped
4937 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4938 alloc->maybe_set_complete(&_gvn);
4939 }
4940 }
4941
4942 // Return the combined state.
4943 set_i_o( _gvn.transform(result_io) );
4944 set_all_memory( _gvn.transform(result_mem));
4945
4946 C->set_has_split_ifs(true); // Has chance for split-if optimization
4947 set_result(result_reg, result_val);
4948 return true;
4949 }
4950
4951 //----------------------inline_native_getLength--------------------------
4952 // public static native int java.lang.reflect.Array.getLength(Object array);
4953 bool LibraryCallKit::inline_native_getLength() {
4954 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4955
4956 Node* array = null_check(argument(0));
4957 // If array is dead, only null-path is taken.
4958 if (stopped()) return true;
4959
4960 // Deoptimize if it is a non-array.
4961 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
4962
4963 if (non_array != nullptr) {
4964 PreserveJVMState pjvms(this);
4965 set_control(non_array);
4966 uncommon_trap(Deoptimization::Reason_intrinsic,
4967 Deoptimization::Action_maybe_recompile);
4968 }
4969
4970 // If control is dead, only non-array-path is taken.
4971 if (stopped()) return true;
4972
4973 // The works fine even if the array type is polymorphic.
4974 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4975 Node* result = load_array_length(array);
4976
4977 C->set_has_split_ifs(true); // Has chance for split-if optimization
4978 set_result(result);
4979 return true;
4980 }
4981
4982 //------------------------inline_array_copyOf----------------------------
4983 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4984 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4985 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4986 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4987
4988 // Get the arguments.
4989 Node* original = argument(0);
4990 Node* start = is_copyOfRange? argument(1): intcon(0);
4991 Node* end = is_copyOfRange? argument(2): argument(1);
4992 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4993
4994 Node* newcopy = nullptr;
4995
4996 // Set the original stack and the reexecute bit for the interpreter to reexecute
4997 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4998 { PreserveReexecuteState preexecs(this);
4999 jvms()->set_should_reexecute(true);
5000
5001 array_type_mirror = null_check(array_type_mirror);
5002 original = null_check(original);
5003
5004 // Check if a null path was taken unconditionally.
5005 if (stopped()) return true;
5006
5007 Node* orig_length = load_array_length(original);
5008
5009 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
5010 klass_node = null_check(klass_node);
5011
5012 RegionNode* bailout = new RegionNode(1);
5013 record_for_igvn(bailout);
5014
5015 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
5016 // Bail out if that is so.
5017 // Inline type array may have object field that would require a
5018 // write barrier. Conservatively, go to slow path.
5019 // TODO 8251971: Optimize for the case when flat src/dst are later found
5020 // to not contain oops (i.e., move this check to the macro expansion phase).
5021 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5022 const TypeAryPtr* orig_t = _gvn.type(original)->isa_aryptr();
5023 const TypeKlassPtr* tklass = _gvn.type(klass_node)->is_klassptr();
5024 bool exclude_flat = UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing) &&
5025 // Can src array be flat and contain oops?
5026 (orig_t == nullptr || (!orig_t->is_not_flat() && (!orig_t->is_flat() || orig_t->elem()->inline_klass()->contains_oops()))) &&
5027 // Can dest array be flat and contain oops?
5028 tklass->can_be_inline_array() && (!tklass->is_flat() || tklass->is_aryklassptr()->elem()->is_instklassptr()->instance_klass()->as_inline_klass()->contains_oops());
5029 Node* not_objArray = exclude_flat ? generate_non_refArray_guard(klass_node, bailout) : generate_typeArray_guard(klass_node, bailout);
5030
5031 Node* refined_klass_node = load_default_refined_array_klass(klass_node, /* type_array_guard= */ false);
5032
5033 if (not_objArray != nullptr) {
5034 // Improve the klass node's type from the new optimistic assumption:
5035 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
5036 bool not_flat = !UseArrayFlattening;
5037 bool not_null_free = !Arguments::is_valhalla_enabled();
5038 const Type* akls = TypeAryKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0), Type::trust_interfaces, not_flat, not_null_free, false, false, not_flat, true);
5039 Node* cast = new CastPPNode(control(), refined_klass_node, akls);
5040 refined_klass_node = _gvn.transform(cast);
5041 }
5042
5043 // Bail out if either start or end is negative.
5044 generate_negative_guard(start, bailout, &start);
5045 generate_negative_guard(end, bailout, &end);
5046
5047 Node* length = end;
5048 if (_gvn.type(start) != TypeInt::ZERO) {
5049 length = _gvn.transform(new SubINode(end, start));
5050 }
5051
5052 // Bail out if length is negative (i.e., if start > end).
5053 // Without this the new_array would throw
5054 // NegativeArraySizeException but IllegalArgumentException is what
5055 // should be thrown
5056 generate_negative_guard(length, bailout, &length);
5057
5058 // Handle inline type arrays
5059 bool can_validate = !too_many_traps(Deoptimization::Reason_class_check);
5060 if (!stopped()) {
5061 // TODO 8251971
5062 if (!orig_t->is_null_free()) {
5063 // Not statically known to be null free, add a check
5064 generate_fair_guard(null_free_array_test(original), bailout);
5065 }
5066 orig_t = _gvn.type(original)->isa_aryptr();
5067 if (orig_t != nullptr && orig_t->is_flat()) {
5068 // Src is flat, check that dest is flat as well
5069 if (exclude_flat) {
5070 // Dest can't be flat, bail out
5071 bailout->add_req(control());
5072 set_control(top());
5073 } else {
5074 generate_fair_guard(flat_array_test(refined_klass_node, /* flat = */ false), bailout);
5075 }
5076 // TODO 8350865 This is not correct anymore. Write tests and fix logic similar to arraycopy.
5077 } else if (UseArrayFlattening && (orig_t == nullptr || !orig_t->is_not_flat()) &&
5078 // If dest is flat, src must be flat as well (guaranteed by src <: dest check if validated).
5079 ((!tklass->is_flat() && tklass->can_be_inline_array()) || !can_validate)) {
5080 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
5081 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
5082 generate_fair_guard(flat_array_test(load_object_klass(original)), bailout);
5083 if (orig_t != nullptr) {
5084 orig_t = orig_t->cast_to_not_flat();
5085 original = _gvn.transform(new CheckCastPPNode(control(), original, orig_t));
5086 }
5087 }
5088 if (!can_validate) {
5089 // No validation. The subtype check emitted at macro expansion time will not go to the slow
5090 // path but call checkcast_arraycopy which can not handle flat/null-free inline type arrays.
5091 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat/null-free.
5092 generate_fair_guard(flat_array_test(refined_klass_node), bailout);
5093 generate_fair_guard(null_free_array_test(original), bailout);
5094 }
5095 }
5096
5097 // Bail out if start is larger than the original length
5098 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
5099 generate_negative_guard(orig_tail, bailout, &orig_tail);
5100
5101 if (bailout->req() > 1) {
5102 PreserveJVMState pjvms(this);
5103 set_control(_gvn.transform(bailout));
5104 uncommon_trap(Deoptimization::Reason_intrinsic,
5105 Deoptimization::Action_maybe_recompile);
5106 }
5107
5108 if (!stopped()) {
5109 // How many elements will we copy from the original?
5110 // The answer is MinI(orig_tail, length).
5111 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
5112
5113 // Generate a direct call to the right arraycopy function(s).
5114 // We know the copy is disjoint but we might not know if the
5115 // oop stores need checking.
5116 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
5117 // This will fail a store-check if x contains any non-nulls.
5118
5119 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
5120 // loads/stores but it is legal only if we're sure the
5121 // Arrays.copyOf would succeed. So we need all input arguments
5122 // to the copyOf to be validated, including that the copy to the
5123 // new array won't trigger an ArrayStoreException. That subtype
5124 // check can be optimized if we know something on the type of
5125 // the input array from type speculation.
5126 if (_gvn.type(klass_node)->singleton()) {
5127 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
5128 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
5129
5130 int test = C->static_subtype_check(superk, subk);
5131 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
5132 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
5133 if (t_original->speculative_type() != nullptr) {
5134 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
5135 }
5136 }
5137 }
5138
5139 bool validated = false;
5140 // Reason_class_check rather than Reason_intrinsic because we
5141 // want to intrinsify even if this traps.
5142 if (can_validate) {
5143 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
5144
5145 if (not_subtype_ctrl != top()) {
5146 PreserveJVMState pjvms(this);
5147 set_control(not_subtype_ctrl);
5148 uncommon_trap(Deoptimization::Reason_class_check,
5149 Deoptimization::Action_make_not_entrant);
5150 assert(stopped(), "Should be stopped");
5151 }
5152 validated = true;
5153 }
5154
5155 if (!stopped()) {
5156 newcopy = new_array(refined_klass_node, length, 0); // no arguments to push
5157
5158 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
5159 load_object_klass(original), klass_node);
5160 if (!is_copyOfRange) {
5161 ac->set_copyof(validated);
5162 } else {
5163 ac->set_copyofrange(validated);
5164 }
5165 Node* n = _gvn.transform(ac);
5166 if (n == ac) {
5167 ac->connect_outputs(this);
5168 } else {
5169 assert(validated, "shouldn't transform if all arguments not validated");
5170 set_all_memory(n);
5171 }
5172 }
5173 }
5174 } // original reexecute is set back here
5175
5176 C->set_has_split_ifs(true); // Has chance for split-if optimization
5177 if (!stopped()) {
5178 set_result(newcopy);
5179 }
5180 return true;
5181 }
5182
5183
5184 //----------------------generate_virtual_guard---------------------------
5185 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
5186 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
5187 RegionNode* slow_region) {
5188 ciMethod* method = callee();
5189 int vtable_index = method->vtable_index();
5190 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5191 "bad index %d", vtable_index);
5192 // Get the Method* out of the appropriate vtable entry.
5193 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
5194 vtable_index*vtableEntry::size_in_bytes() +
5195 in_bytes(vtableEntry::method_offset());
5196 Node* entry_addr = off_heap_plus_addr(obj_klass, entry_offset);
5197 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
5198
5199 // Compare the target method with the expected method (e.g., Object.hashCode).
5200 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
5201
5202 Node* native_call = makecon(native_call_addr);
5203 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
5204 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
5205
5206 return generate_slow_guard(test_native, slow_region);
5207 }
5208
5209 //-----------------------generate_method_call----------------------------
5210 // Use generate_method_call to make a slow-call to the real
5211 // method if the fast path fails. An alternative would be to
5212 // use a stub like OptoRuntime::slow_arraycopy_Java.
5213 // This only works for expanding the current library call,
5214 // not another intrinsic. (E.g., don't use this for making an
5215 // arraycopy call inside of the copyOf intrinsic.)
5216 CallJavaNode*
5217 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
5218 // When compiling the intrinsic method itself, do not use this technique.
5219 guarantee(callee() != C->method(), "cannot make slow-call to self");
5220
5221 ciMethod* method = callee();
5222 // ensure the JVMS we have will be correct for this call
5223 guarantee(method_id == method->intrinsic_id(), "must match");
5224
5225 const TypeFunc* tf = TypeFunc::make(method);
5226 if (res_not_null) {
5227 assert(tf->return_type() == T_OBJECT, "");
5228 const TypeTuple* range = tf->range_cc();
5229 const Type** fields = TypeTuple::fields(range->cnt());
5230 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
5231 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
5232 tf = TypeFunc::make(tf->domain_cc(), new_range);
5233 }
5234 CallJavaNode* slow_call;
5235 if (is_static) {
5236 assert(!is_virtual, "");
5237 slow_call = new CallStaticJavaNode(C, tf,
5238 SharedRuntime::get_resolve_static_call_stub(), method);
5239 } else if (is_virtual) {
5240 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5241 int vtable_index = Method::invalid_vtable_index;
5242 if (UseInlineCaches) {
5243 // Suppress the vtable call
5244 } else {
5245 // hashCode and clone are not a miranda methods,
5246 // so the vtable index is fixed.
5247 // No need to use the linkResolver to get it.
5248 vtable_index = method->vtable_index();
5249 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5250 "bad index %d", vtable_index);
5251 }
5252 slow_call = new CallDynamicJavaNode(tf,
5253 SharedRuntime::get_resolve_virtual_call_stub(),
5254 method, vtable_index);
5255 } else { // neither virtual nor static: opt_virtual
5256 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5257 slow_call = new CallStaticJavaNode(C, tf,
5258 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
5259 slow_call->set_optimized_virtual(true);
5260 }
5261 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
5262 // To be able to issue a direct call (optimized virtual or virtual)
5263 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
5264 // about the method being invoked should be attached to the call site to
5265 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
5266 slow_call->set_override_symbolic_info(true);
5267 }
5268 set_arguments_for_java_call(slow_call);
5269 set_edges_for_java_call(slow_call);
5270 return slow_call;
5271 }
5272
5273
5274 /**
5275 * Build special case code for calls to hashCode on an object. This call may
5276 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
5277 * slightly different code.
5278 */
5279 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
5280 assert(is_static == callee()->is_static(), "correct intrinsic selection");
5281 assert(!(is_virtual && is_static), "either virtual, special, or static");
5282
5283 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
5284
5285 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5286 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
5287 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
5288 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5289 Node* obj = argument(0);
5290
5291 // Don't intrinsify hashcode on inline types for now.
5292 // The "is locked" runtime check also subsumes the inline type check (as inline types cannot be locked) and goes to the slow path.
5293 if (gvn().type(obj)->is_inlinetypeptr()) {
5294 return false;
5295 }
5296
5297 if (!is_static) {
5298 // Check for hashing null object
5299 obj = null_check_receiver();
5300 if (stopped()) return true; // unconditionally null
5301 result_reg->init_req(_null_path, top());
5302 result_val->init_req(_null_path, top());
5303 } else {
5304 // Do a null check, and return zero if null.
5305 // System.identityHashCode(null) == 0
5306 Node* null_ctl = top();
5307 obj = null_check_oop(obj, &null_ctl);
5308 result_reg->init_req(_null_path, null_ctl);
5309 result_val->init_req(_null_path, _gvn.intcon(0));
5310 }
5311
5312 // Unconditionally null? Then return right away.
5313 if (stopped()) {
5314 set_control( result_reg->in(_null_path));
5315 if (!stopped())
5316 set_result(result_val->in(_null_path));
5317 return true;
5318 }
5319
5320 // We only go to the fast case code if we pass a number of guards. The
5321 // paths which do not pass are accumulated in the slow_region.
5322 RegionNode* slow_region = new RegionNode(1);
5323 record_for_igvn(slow_region);
5324
5325 // If this is a virtual call, we generate a funny guard. We pull out
5326 // the vtable entry corresponding to hashCode() from the target object.
5327 // If the target method which we are calling happens to be the native
5328 // Object hashCode() method, we pass the guard. We do not need this
5329 // guard for non-virtual calls -- the caller is known to be the native
5330 // Object hashCode().
5331 if (is_virtual) {
5332 // After null check, get the object's klass.
5333 Node* obj_klass = load_object_klass(obj);
5334 generate_virtual_guard(obj_klass, slow_region);
5335 }
5336
5337 // Get the header out of the object, use LoadMarkNode when available
5338 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
5339 // The control of the load must be null. Otherwise, the load can move before
5340 // the null check after castPP removal.
5341 Node* no_ctrl = nullptr;
5342 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
5343
5344 if (!UseObjectMonitorTable) {
5345 // Test the header to see if it is safe to read w.r.t. locking.
5346 // We cannot use the inline type mask as this may check bits that are overriden
5347 // by an object monitor's pointer when inflating locking.
5348 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
5349 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
5350 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
5351 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
5352 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
5353
5354 generate_slow_guard(test_monitor, slow_region);
5355 }
5356
5357 // Get the hash value and check to see that it has been properly assigned.
5358 // We depend on hash_mask being at most 32 bits and avoid the use of
5359 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
5360 // vm: see markWord.hpp.
5361 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
5362 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
5363 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
5364 // This hack lets the hash bits live anywhere in the mark object now, as long
5365 // as the shift drops the relevant bits into the low 32 bits. Note that
5366 // Java spec says that HashCode is an int so there's no point in capturing
5367 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
5368 hshifted_header = ConvX2I(hshifted_header);
5369 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
5370
5371 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
5372 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
5373 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
5374
5375 generate_slow_guard(test_assigned, slow_region);
5376
5377 Node* init_mem = reset_memory();
5378 // fill in the rest of the null path:
5379 result_io ->init_req(_null_path, i_o());
5380 result_mem->init_req(_null_path, init_mem);
5381
5382 result_val->init_req(_fast_path, hash_val);
5383 result_reg->init_req(_fast_path, control());
5384 result_io ->init_req(_fast_path, i_o());
5385 result_mem->init_req(_fast_path, init_mem);
5386
5387 // Generate code for the slow case. We make a call to hashCode().
5388 set_control(_gvn.transform(slow_region));
5389 if (!stopped()) {
5390 // No need for PreserveJVMState, because we're using up the present state.
5391 set_all_memory(init_mem);
5392 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
5393 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
5394 Node* slow_result = set_results_for_java_call(slow_call);
5395 // this->control() comes from set_results_for_java_call
5396 result_reg->init_req(_slow_path, control());
5397 result_val->init_req(_slow_path, slow_result);
5398 result_io ->set_req(_slow_path, i_o());
5399 result_mem ->set_req(_slow_path, reset_memory());
5400 }
5401
5402 // Return the combined state.
5403 set_i_o( _gvn.transform(result_io) );
5404 set_all_memory( _gvn.transform(result_mem));
5405
5406 set_result(result_reg, result_val);
5407 return true;
5408 }
5409
5410 //---------------------------inline_native_getClass----------------------------
5411 // public final native Class<?> java.lang.Object.getClass();
5412 //
5413 // Build special case code for calls to getClass on an object.
5414 bool LibraryCallKit::inline_native_getClass() {
5415 Node* obj = argument(0);
5416 if (obj->is_InlineType()) {
5417 const Type* t = _gvn.type(obj);
5418 if (t->maybe_null()) {
5419 null_check(obj);
5420 }
5421 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror())));
5422 return true;
5423 }
5424 obj = null_check_receiver();
5425 if (stopped()) return true;
5426 set_result(load_mirror_from_klass(load_object_klass(obj)));
5427 return true;
5428 }
5429
5430 //-----------------inline_native_Reflection_getCallerClass---------------------
5431 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
5432 //
5433 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
5434 //
5435 // NOTE: This code must perform the same logic as JVM_GetCallerClass
5436 // in that it must skip particular security frames and checks for
5437 // caller sensitive methods.
5438 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
5439 #ifndef PRODUCT
5440 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5441 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
5442 }
5443 #endif
5444
5445 if (!jvms()->has_method()) {
5446 #ifndef PRODUCT
5447 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5448 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
5449 }
5450 #endif
5451 return false;
5452 }
5453
5454 // Walk back up the JVM state to find the caller at the required
5455 // depth.
5456 JVMState* caller_jvms = jvms();
5457
5458 // Cf. JVM_GetCallerClass
5459 // NOTE: Start the loop at depth 1 because the current JVM state does
5460 // not include the Reflection.getCallerClass() frame.
5461 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
5462 ciMethod* m = caller_jvms->method();
5463 switch (n) {
5464 case 0:
5465 fatal("current JVM state does not include the Reflection.getCallerClass frame");
5466 break;
5467 case 1:
5468 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
5469 if (!m->caller_sensitive()) {
5470 #ifndef PRODUCT
5471 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5472 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
5473 }
5474 #endif
5475 return false; // bail-out; let JVM_GetCallerClass do the work
5476 }
5477 break;
5478 default:
5479 if (!m->is_ignored_by_security_stack_walk()) {
5480 // We have reached the desired frame; return the holder class.
5481 // Acquire method holder as java.lang.Class and push as constant.
5482 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
5483 ciInstance* caller_mirror = caller_klass->java_mirror();
5484 set_result(makecon(TypeInstPtr::make(caller_mirror)));
5485
5486 #ifndef PRODUCT
5487 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5488 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());
5489 tty->print_cr(" JVM state at this point:");
5490 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5491 ciMethod* m = jvms()->of_depth(i)->method();
5492 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5493 }
5494 }
5495 #endif
5496 return true;
5497 }
5498 break;
5499 }
5500 }
5501
5502 #ifndef PRODUCT
5503 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5504 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
5505 tty->print_cr(" JVM state at this point:");
5506 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5507 ciMethod* m = jvms()->of_depth(i)->method();
5508 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5509 }
5510 }
5511 #endif
5512
5513 return false; // bail-out; let JVM_GetCallerClass do the work
5514 }
5515
5516 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
5517 Node* arg = argument(0);
5518 Node* result = nullptr;
5519
5520 switch (id) {
5521 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
5522 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
5523 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5524 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5525 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5526 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5527
5528 case vmIntrinsics::_doubleToLongBits: {
5529 // two paths (plus control) merge in a wood
5530 RegionNode *r = new RegionNode(3);
5531 Node *phi = new PhiNode(r, TypeLong::LONG);
5532
5533 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5534 // Build the boolean node
5535 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5536
5537 // Branch either way.
5538 // NaN case is less traveled, which makes all the difference.
5539 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5540 Node *opt_isnan = _gvn.transform(ifisnan);
5541 assert( opt_isnan->is_If(), "Expect an IfNode");
5542 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5543 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5544
5545 set_control(iftrue);
5546
5547 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5548 Node *slow_result = longcon(nan_bits); // return NaN
5549 phi->init_req(1, _gvn.transform( slow_result ));
5550 r->init_req(1, iftrue);
5551
5552 // Else fall through
5553 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5554 set_control(iffalse);
5555
5556 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5557 r->init_req(2, iffalse);
5558
5559 // Post merge
5560 set_control(_gvn.transform(r));
5561 record_for_igvn(r);
5562
5563 C->set_has_split_ifs(true); // Has chance for split-if optimization
5564 result = phi;
5565 assert(result->bottom_type()->isa_long(), "must be");
5566 break;
5567 }
5568
5569 case vmIntrinsics::_floatToIntBits: {
5570 // two paths (plus control) merge in a wood
5571 RegionNode *r = new RegionNode(3);
5572 Node *phi = new PhiNode(r, TypeInt::INT);
5573
5574 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5575 // Build the boolean node
5576 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5577
5578 // Branch either way.
5579 // NaN case is less traveled, which makes all the difference.
5580 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5581 Node *opt_isnan = _gvn.transform(ifisnan);
5582 assert( opt_isnan->is_If(), "Expect an IfNode");
5583 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5584 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5585
5586 set_control(iftrue);
5587
5588 static const jint nan_bits = 0x7fc00000;
5589 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5590 phi->init_req(1, _gvn.transform( slow_result ));
5591 r->init_req(1, iftrue);
5592
5593 // Else fall through
5594 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5595 set_control(iffalse);
5596
5597 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5598 r->init_req(2, iffalse);
5599
5600 // Post merge
5601 set_control(_gvn.transform(r));
5602 record_for_igvn(r);
5603
5604 C->set_has_split_ifs(true); // Has chance for split-if optimization
5605 result = phi;
5606 assert(result->bottom_type()->isa_int(), "must be");
5607 break;
5608 }
5609
5610 default:
5611 fatal_unexpected_iid(id);
5612 break;
5613 }
5614 set_result(_gvn.transform(result));
5615 return true;
5616 }
5617
5618 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5619 Node* arg = argument(0);
5620 Node* result = nullptr;
5621
5622 switch (id) {
5623 case vmIntrinsics::_floatIsInfinite:
5624 result = new IsInfiniteFNode(arg);
5625 break;
5626 case vmIntrinsics::_floatIsFinite:
5627 result = new IsFiniteFNode(arg);
5628 break;
5629 case vmIntrinsics::_doubleIsInfinite:
5630 result = new IsInfiniteDNode(arg);
5631 break;
5632 case vmIntrinsics::_doubleIsFinite:
5633 result = new IsFiniteDNode(arg);
5634 break;
5635 default:
5636 fatal_unexpected_iid(id);
5637 break;
5638 }
5639 set_result(_gvn.transform(result));
5640 return true;
5641 }
5642
5643 //----------------------inline_unsafe_copyMemory-------------------------
5644 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5645
5646 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5647 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5648 const Type* base_t = gvn.type(base);
5649
5650 bool in_native = (base_t == TypePtr::NULL_PTR);
5651 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5652 bool is_mixed = !in_heap && !in_native;
5653
5654 if (is_mixed) {
5655 return true; // mixed accesses can touch both on-heap and off-heap memory
5656 }
5657 if (in_heap) {
5658 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5659 if (!is_prim_array) {
5660 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5661 // there's not enough type information available to determine proper memory slice for it.
5662 return true;
5663 }
5664 }
5665 return false;
5666 }
5667
5668 bool LibraryCallKit::inline_unsafe_copyMemory() {
5669 if (callee()->is_static()) return false; // caller must have the capability!
5670 null_check_receiver(); // null-check receiver
5671 if (stopped()) return true;
5672
5673 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5674
5675 Node* src_base = argument(1); // type: oop
5676 Node* src_off = ConvL2X(argument(2)); // type: long
5677 Node* dst_base = argument(4); // type: oop
5678 Node* dst_off = ConvL2X(argument(5)); // type: long
5679 Node* size = ConvL2X(argument(7)); // type: long
5680
5681 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5682 "fieldOffset must be byte-scaled");
5683
5684 Node* src_addr = make_unsafe_address(src_base, src_off);
5685 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5686
5687 Node* thread = _gvn.transform(new ThreadLocalNode());
5688 Node* doing_unsafe_access_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5689 BasicType doing_unsafe_access_bt = T_BYTE;
5690 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5691
5692 // update volatile field
5693 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5694
5695 int flags = RC_LEAF | RC_NO_FP;
5696
5697 const TypePtr* dst_type = TypePtr::BOTTOM;
5698
5699 // Adjust memory effects of the runtime call based on input values.
5700 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5701 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5702 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5703
5704 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5705 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5706 flags |= RC_NARROW_MEM; // narrow in memory
5707 }
5708 }
5709
5710 // Call it. Note that the length argument is not scaled.
5711 make_runtime_call(flags,
5712 OptoRuntime::fast_arraycopy_Type(),
5713 StubRoutines::unsafe_arraycopy(),
5714 "unsafe_arraycopy",
5715 dst_type,
5716 src_addr, dst_addr, size XTOP);
5717
5718 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5719
5720 return true;
5721 }
5722
5723 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5724 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5725 bool LibraryCallKit::inline_unsafe_setMemory() {
5726 if (callee()->is_static()) return false; // caller must have the capability!
5727 null_check_receiver(); // null-check receiver
5728 if (stopped()) return true;
5729
5730 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5731
5732 Node* dst_base = argument(1); // type: oop
5733 Node* dst_off = ConvL2X(argument(2)); // type: long
5734 Node* size = ConvL2X(argument(4)); // type: long
5735 Node* byte = argument(6); // type: byte
5736
5737 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5738 "fieldOffset must be byte-scaled");
5739
5740 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5741
5742 Node* thread = _gvn.transform(new ThreadLocalNode());
5743 Node* doing_unsafe_access_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5744 BasicType doing_unsafe_access_bt = T_BYTE;
5745 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5746
5747 // update volatile field
5748 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5749
5750 int flags = RC_LEAF | RC_NO_FP;
5751
5752 const TypePtr* dst_type = TypePtr::BOTTOM;
5753
5754 // Adjust memory effects of the runtime call based on input values.
5755 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5756 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5757
5758 flags |= RC_NARROW_MEM; // narrow in memory
5759 }
5760
5761 // Call it. Note that the length argument is not scaled.
5762 make_runtime_call(flags,
5763 OptoRuntime::unsafe_setmemory_Type(),
5764 StubRoutines::unsafe_setmemory(),
5765 "unsafe_setmemory",
5766 dst_type,
5767 dst_addr, size XTOP, byte);
5768
5769 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5770
5771 return true;
5772 }
5773
5774 #undef XTOP
5775
5776 //------------------------clone_coping-----------------------------------
5777 // Helper function for inline_native_clone.
5778 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5779 assert(obj_size != nullptr, "");
5780 Node* raw_obj = alloc_obj->in(1);
5781 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5782
5783 AllocateNode* alloc = nullptr;
5784 if (ReduceBulkZeroing &&
5785 // If we are implementing an array clone without knowing its source type
5786 // (can happen when compiling the array-guarded branch of a reflective
5787 // Object.clone() invocation), initialize the array within the allocation.
5788 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5789 // to a runtime clone call that assumes fully initialized source arrays.
5790 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5791 // We will be completely responsible for initializing this object -
5792 // mark Initialize node as complete.
5793 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5794 // The object was just allocated - there should be no any stores!
5795 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5796 // Mark as complete_with_arraycopy so that on AllocateNode
5797 // expansion, we know this AllocateNode is initialized by an array
5798 // copy and a StoreStore barrier exists after the array copy.
5799 alloc->initialization()->set_complete_with_arraycopy();
5800 }
5801
5802 Node* size = _gvn.transform(obj_size);
5803 access_clone(obj, alloc_obj, size, is_array);
5804
5805 // Do not let reads from the cloned object float above the arraycopy.
5806 if (alloc != nullptr) {
5807 // Do not let stores that initialize this object be reordered with
5808 // a subsequent store that would make this object accessible by
5809 // other threads.
5810 // Record what AllocateNode this StoreStore protects so that
5811 // escape analysis can go from the MemBarStoreStoreNode to the
5812 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5813 // based on the escape status of the AllocateNode.
5814 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5815 } else {
5816 insert_mem_bar(Op_MemBarCPUOrder);
5817 }
5818 }
5819
5820 //------------------------inline_native_clone----------------------------
5821 // protected native Object java.lang.Object.clone();
5822 //
5823 // Here are the simple edge cases:
5824 // null receiver => normal trap
5825 // virtual and clone was overridden => slow path to out-of-line clone
5826 // not cloneable or finalizer => slow path to out-of-line Object.clone
5827 //
5828 // The general case has two steps, allocation and copying.
5829 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5830 //
5831 // Copying also has two cases, oop arrays and everything else.
5832 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5833 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5834 //
5835 // These steps fold up nicely if and when the cloned object's klass
5836 // can be sharply typed as an object array, a type array, or an instance.
5837 //
5838 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5839 PhiNode* result_val;
5840
5841 // Set the reexecute bit for the interpreter to reexecute
5842 // the bytecode that invokes Object.clone if deoptimization happens.
5843 { PreserveReexecuteState preexecs(this);
5844 jvms()->set_should_reexecute(true);
5845
5846 Node* obj = argument(0);
5847 obj = null_check_receiver();
5848 if (stopped()) return true;
5849
5850 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5851 if (obj_type->is_inlinetypeptr()) {
5852 // If the object to clone is an inline type, we can simply return it (i.e. a nop) since inline types have
5853 // no identity.
5854 set_result(obj);
5855 return true;
5856 }
5857
5858 // If we are going to clone an instance, we need its exact type to
5859 // know the number and types of fields to convert the clone to
5860 // loads/stores. Maybe a speculative type can help us.
5861 if (!obj_type->klass_is_exact() &&
5862 obj_type->speculative_type() != nullptr &&
5863 obj_type->speculative_type()->is_instance_klass() &&
5864 !obj_type->speculative_type()->is_inlinetype()) {
5865 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5866 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5867 !spec_ik->has_injected_fields()) {
5868 if (!obj_type->isa_instptr() ||
5869 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5870 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5871 }
5872 }
5873 }
5874
5875 // Conservatively insert a memory barrier on all memory slices.
5876 // Do not let writes into the original float below the clone.
5877 insert_mem_bar(Op_MemBarCPUOrder);
5878
5879 // paths into result_reg:
5880 enum {
5881 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5882 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5883 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5884 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5885 PATH_LIMIT
5886 };
5887 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5888 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5889 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5890 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5891 record_for_igvn(result_reg);
5892
5893 Node* obj_klass = load_object_klass(obj);
5894 // We only go to the fast case code if we pass a number of guards.
5895 // The paths which do not pass are accumulated in the slow_region.
5896 RegionNode* slow_region = new RegionNode(1);
5897 record_for_igvn(slow_region);
5898
5899 Node* array_obj = obj;
5900 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5901 if (array_ctl != nullptr) {
5902 // It's an array.
5903 PreserveJVMState pjvms(this);
5904 set_control(array_ctl);
5905
5906 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5907 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr();
5908 if (UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) &&
5909 obj_type->can_be_inline_array() &&
5910 (ary_ptr == nullptr || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) {
5911 // Flat inline type array may have object field that would require a
5912 // write barrier. Conservatively, go to slow path.
5913 generate_fair_guard(flat_array_test(obj_klass), slow_region);
5914 }
5915
5916 if (!stopped()) {
5917 Node* obj_length = load_array_length(array_obj);
5918 Node* array_size = nullptr; // Size of the array without object alignment padding.
5919 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5920
5921 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5922 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5923 // If it is an oop array, it requires very special treatment,
5924 // because gc barriers are required when accessing the array.
5925 Node* is_obja = generate_refArray_guard(obj_klass, (RegionNode*)nullptr);
5926 if (is_obja != nullptr) {
5927 PreserveJVMState pjvms2(this);
5928 set_control(is_obja);
5929 // Generate a direct call to the right arraycopy function(s).
5930 // Clones are always tightly coupled.
5931 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5932 ac->set_clone_oop_array();
5933 Node* n = _gvn.transform(ac);
5934 assert(n == ac, "cannot disappear");
5935 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5936
5937 result_reg->init_req(_objArray_path, control());
5938 result_val->init_req(_objArray_path, alloc_obj);
5939 result_i_o ->set_req(_objArray_path, i_o());
5940 result_mem ->set_req(_objArray_path, reset_memory());
5941 }
5942 }
5943 // Otherwise, there are no barriers to worry about.
5944 // (We can dispense with card marks if we know the allocation
5945 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5946 // causes the non-eden paths to take compensating steps to
5947 // simulate a fresh allocation, so that no further
5948 // card marks are required in compiled code to initialize
5949 // the object.)
5950
5951 if (!stopped()) {
5952 copy_to_clone(obj, alloc_obj, array_size, true);
5953
5954 // Present the results of the copy.
5955 result_reg->init_req(_array_path, control());
5956 result_val->init_req(_array_path, alloc_obj);
5957 result_i_o ->set_req(_array_path, i_o());
5958 result_mem ->set_req(_array_path, reset_memory());
5959 }
5960 }
5961 }
5962
5963 if (!stopped()) {
5964 // It's an instance (we did array above). Make the slow-path tests.
5965 // If this is a virtual call, we generate a funny guard. We grab
5966 // the vtable entry corresponding to clone() from the target object.
5967 // If the target method which we are calling happens to be the
5968 // Object clone() method, we pass the guard. We do not need this
5969 // guard for non-virtual calls; the caller is known to be the native
5970 // Object clone().
5971 if (is_virtual) {
5972 generate_virtual_guard(obj_klass, slow_region);
5973 }
5974
5975 // The object must be easily cloneable and must not have a finalizer.
5976 // Both of these conditions may be checked in a single test.
5977 // We could optimize the test further, but we don't care.
5978 generate_misc_flags_guard(obj_klass,
5979 // Test both conditions:
5980 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5981 // Must be cloneable but not finalizer:
5982 KlassFlags::_misc_is_cloneable_fast,
5983 slow_region);
5984 }
5985
5986 if (!stopped()) {
5987 // It's an instance, and it passed the slow-path tests.
5988 PreserveJVMState pjvms(this);
5989 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5990 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5991 // is reexecuted if deoptimization occurs and there could be problems when merging
5992 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5993 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5994
5995 copy_to_clone(obj, alloc_obj, obj_size, false);
5996
5997 // Present the results of the slow call.
5998 result_reg->init_req(_instance_path, control());
5999 result_val->init_req(_instance_path, alloc_obj);
6000 result_i_o ->set_req(_instance_path, i_o());
6001 result_mem ->set_req(_instance_path, reset_memory());
6002 }
6003
6004 // Generate code for the slow case. We make a call to clone().
6005 set_control(_gvn.transform(slow_region));
6006 if (!stopped()) {
6007 PreserveJVMState pjvms(this);
6008 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
6009 // We need to deoptimize on exception (see comment above)
6010 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
6011 // this->control() comes from set_results_for_java_call
6012 result_reg->init_req(_slow_path, control());
6013 result_val->init_req(_slow_path, slow_result);
6014 result_i_o ->set_req(_slow_path, i_o());
6015 result_mem ->set_req(_slow_path, reset_memory());
6016 }
6017
6018 // Return the combined state.
6019 set_control( _gvn.transform(result_reg));
6020 set_i_o( _gvn.transform(result_i_o));
6021 set_all_memory( _gvn.transform(result_mem));
6022 } // original reexecute is set back here
6023
6024 set_result(_gvn.transform(result_val));
6025 return true;
6026 }
6027
6028 // If we have a tightly coupled allocation, the arraycopy may take care
6029 // of the array initialization. If one of the guards we insert between
6030 // the allocation and the arraycopy causes a deoptimization, an
6031 // uninitialized array will escape the compiled method. To prevent that
6032 // we set the JVM state for uncommon traps between the allocation and
6033 // the arraycopy to the state before the allocation so, in case of
6034 // deoptimization, we'll reexecute the allocation and the
6035 // initialization.
6036 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
6037 if (alloc != nullptr) {
6038 ciMethod* trap_method = alloc->jvms()->method();
6039 int trap_bci = alloc->jvms()->bci();
6040
6041 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6042 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
6043 // Make sure there's no store between the allocation and the
6044 // arraycopy otherwise visible side effects could be rexecuted
6045 // in case of deoptimization and cause incorrect execution.
6046 bool no_interfering_store = true;
6047 Node* mem = alloc->in(TypeFunc::Memory);
6048 if (mem->is_MergeMem()) {
6049 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
6050 Node* n = mms.memory();
6051 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
6052 assert(n->is_Store(), "what else?");
6053 no_interfering_store = false;
6054 break;
6055 }
6056 }
6057 } else {
6058 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
6059 Node* n = mms.memory();
6060 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
6061 assert(n->is_Store(), "what else?");
6062 no_interfering_store = false;
6063 break;
6064 }
6065 }
6066 }
6067
6068 if (no_interfering_store) {
6069 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6070
6071 JVMState* saved_jvms = jvms();
6072 saved_reexecute_sp = _reexecute_sp;
6073
6074 set_jvms(sfpt->jvms());
6075 _reexecute_sp = jvms()->sp();
6076
6077 return saved_jvms;
6078 }
6079 }
6080 }
6081 return nullptr;
6082 }
6083
6084 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
6085 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
6086 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
6087 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
6088 uint size = alloc->req();
6089 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
6090 old_jvms->set_map(sfpt);
6091 for (uint i = 0; i < size; i++) {
6092 sfpt->init_req(i, alloc->in(i));
6093 }
6094 int adjustment = 1;
6095 const TypeAryKlassPtr* ary_klass_ptr = alloc->in(AllocateNode::KlassNode)->bottom_type()->is_aryklassptr();
6096 if (ary_klass_ptr->is_null_free()) {
6097 // A null-free, tightly coupled array allocation can only come from LibraryCallKit::inline_newArray which
6098 // also requires the componentType and initVal on stack for re-execution.
6099 // Re-create and push the componentType.
6100 ciArrayKlass* klass = ary_klass_ptr->exact_klass()->as_array_klass();
6101 ciInstance* instance = klass->component_mirror_instance();
6102 const TypeInstPtr* t_instance = TypeInstPtr::make(instance);
6103 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), makecon(t_instance));
6104 adjustment++;
6105 }
6106 // re-push array length for deoptimization
6107 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment - 1, alloc->in(AllocateNode::ALength));
6108 if (ary_klass_ptr->is_null_free()) {
6109 // Re-create and push the initVal.
6110 Node* init_val = alloc->in(AllocateNode::InitValue);
6111 if (init_val == nullptr) {
6112 init_val = InlineTypeNode::make_all_zero(_gvn, ary_klass_ptr->elem()->is_instklassptr()->instance_klass()->as_inline_klass());
6113 } else if (UseCompressedOops) {
6114 init_val = _gvn.transform(new DecodeNNode(init_val, init_val->bottom_type()->make_ptr()));
6115 }
6116 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment, init_val);
6117 adjustment++;
6118 }
6119 old_jvms->set_sp(old_jvms->sp() + adjustment);
6120 old_jvms->set_monoff(old_jvms->monoff() + adjustment);
6121 old_jvms->set_scloff(old_jvms->scloff() + adjustment);
6122 old_jvms->set_endoff(old_jvms->endoff() + adjustment);
6123 old_jvms->set_should_reexecute(true);
6124
6125 sfpt->set_i_o(map()->i_o());
6126 sfpt->set_memory(map()->memory());
6127 sfpt->set_control(map()->control());
6128 return sfpt;
6129 }
6130
6131 // In case of a deoptimization, we restart execution at the
6132 // allocation, allocating a new array. We would leave an uninitialized
6133 // array in the heap that GCs wouldn't expect. Move the allocation
6134 // after the traps so we don't allocate the array if we
6135 // deoptimize. This is possible because tightly_coupled_allocation()
6136 // guarantees there's no observer of the allocated array at this point
6137 // and the control flow is simple enough.
6138 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
6139 int saved_reexecute_sp, uint new_idx) {
6140 if (saved_jvms_before_guards != nullptr && !stopped()) {
6141 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
6142
6143 assert(alloc != nullptr, "only with a tightly coupled allocation");
6144 // restore JVM state to the state at the arraycopy
6145 saved_jvms_before_guards->map()->set_control(map()->control());
6146 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
6147 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
6148 // If we've improved the types of some nodes (null check) while
6149 // emitting the guards, propagate them to the current state
6150 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
6151 set_jvms(saved_jvms_before_guards);
6152 _reexecute_sp = saved_reexecute_sp;
6153
6154 // Remove the allocation from above the guards
6155 CallProjections* callprojs = alloc->extract_projections(true);
6156 InitializeNode* init = alloc->initialization();
6157 Node* alloc_mem = alloc->in(TypeFunc::Memory);
6158 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
6159 init->replace_mem_projs_by(alloc_mem, C);
6160
6161 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
6162 // the allocation (i.e. is only valid if the allocation succeeds):
6163 // 1) replace CastIINode with AllocateArrayNode's length here
6164 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
6165 //
6166 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
6167 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
6168 Node* init_control = init->proj_out(TypeFunc::Control);
6169 Node* alloc_length = alloc->Ideal_length();
6170 #ifdef ASSERT
6171 Node* prev_cast = nullptr;
6172 #endif
6173 for (uint i = 0; i < init_control->outcnt(); i++) {
6174 Node* init_out = init_control->raw_out(i);
6175 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
6176 #ifdef ASSERT
6177 if (prev_cast == nullptr) {
6178 prev_cast = init_out;
6179 } else {
6180 if (prev_cast->cmp(*init_out) == false) {
6181 prev_cast->dump();
6182 init_out->dump();
6183 assert(false, "not equal CastIINode");
6184 }
6185 }
6186 #endif
6187 C->gvn_replace_by(init_out, alloc_length);
6188 }
6189 }
6190 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
6191
6192 // move the allocation here (after the guards)
6193 _gvn.hash_delete(alloc);
6194 alloc->set_req(TypeFunc::Control, control());
6195 alloc->set_req(TypeFunc::I_O, i_o());
6196 Node *mem = reset_memory();
6197 set_all_memory(mem);
6198 alloc->set_req(TypeFunc::Memory, mem);
6199 set_control(init->proj_out_or_null(TypeFunc::Control));
6200 set_i_o(callprojs->fallthrough_ioproj);
6201
6202 // Update memory as done in GraphKit::set_output_for_allocation()
6203 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
6204 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
6205 if (ary_type->isa_aryptr() && length_type != nullptr) {
6206 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
6207 }
6208 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
6209 int elemidx = C->get_alias_index(telemref);
6210 // Need to properly move every memory projection for the Initialize
6211 #ifdef ASSERT
6212 int mark_idx = C->get_alias_index(ary_type->add_offset(oopDesc::mark_offset_in_bytes()));
6213 int klass_idx = C->get_alias_index(ary_type->add_offset(oopDesc::klass_offset_in_bytes()));
6214 #endif
6215 auto move_proj = [&](ProjNode* proj) {
6216 int alias_idx = C->get_alias_index(proj->adr_type());
6217 assert(alias_idx == Compile::AliasIdxRaw ||
6218 alias_idx == elemidx ||
6219 alias_idx == mark_idx ||
6220 alias_idx == klass_idx, "should be raw memory or array element type");
6221 set_memory(proj, alias_idx);
6222 };
6223 init->for_each_proj(move_proj, TypeFunc::Memory);
6224
6225 Node* allocx = _gvn.transform(alloc);
6226 assert(allocx == alloc, "where has the allocation gone?");
6227 assert(dest->is_CheckCastPP(), "not an allocation result?");
6228
6229 _gvn.hash_delete(dest);
6230 dest->set_req(0, control());
6231 Node* destx = _gvn.transform(dest);
6232 assert(destx == dest, "where has the allocation result gone?");
6233
6234 array_ideal_length(alloc, ary_type, true);
6235 }
6236 }
6237
6238 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
6239 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
6240 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
6241 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
6242 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
6243 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
6244 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
6245 JVMState* saved_jvms_before_guards) {
6246 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
6247 // There is at least one unrelated uncommon trap which needs to be replaced.
6248 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6249
6250 JVMState* saved_jvms = jvms();
6251 const int saved_reexecute_sp = _reexecute_sp;
6252 set_jvms(sfpt->jvms());
6253 _reexecute_sp = jvms()->sp();
6254
6255 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
6256
6257 // Restore state
6258 set_jvms(saved_jvms);
6259 _reexecute_sp = saved_reexecute_sp;
6260 }
6261 }
6262
6263 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
6264 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
6265 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
6266 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
6267 while (if_proj->is_IfProj()) {
6268 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
6269 if (uncommon_trap != nullptr) {
6270 create_new_uncommon_trap(uncommon_trap);
6271 }
6272 assert(if_proj->in(0)->is_If(), "must be If");
6273 if_proj = if_proj->in(0)->in(0);
6274 }
6275 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
6276 "must have reached control projection of init node");
6277 }
6278
6279 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
6280 const int trap_request = uncommon_trap_call->uncommon_trap_request();
6281 assert(trap_request != 0, "no valid UCT trap request");
6282 PreserveJVMState pjvms(this);
6283 set_control(uncommon_trap_call->in(0));
6284 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
6285 Deoptimization::trap_request_action(trap_request));
6286 assert(stopped(), "Should be stopped");
6287 _gvn.hash_delete(uncommon_trap_call);
6288 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
6289 }
6290
6291 // Common checks for array sorting intrinsics arguments.
6292 // Returns `true` if checks passed.
6293 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
6294 // check address of the class
6295 if (elementType == nullptr || elementType->is_top()) {
6296 return false; // dead path
6297 }
6298 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
6299 if (elem_klass == nullptr) {
6300 return false; // dead path
6301 }
6302 // java_mirror_type() returns non-null for compile-time Class constants only
6303 ciType* elem_type = elem_klass->java_mirror_type();
6304 if (elem_type == nullptr) {
6305 return false;
6306 }
6307 bt = elem_type->basic_type();
6308 // Disable the intrinsic if the CPU does not support SIMD sort
6309 if (!Matcher::supports_simd_sort(bt)) {
6310 return false;
6311 }
6312 // check address of the array
6313 if (obj == nullptr || obj->is_top()) {
6314 return false; // dead path
6315 }
6316 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
6317 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
6318 return false; // failed input validation
6319 }
6320 return true;
6321 }
6322
6323 //------------------------------inline_array_partition-----------------------
6324 bool LibraryCallKit::inline_array_partition() {
6325 address stubAddr = StubRoutines::select_array_partition_function();
6326 if (stubAddr == nullptr) {
6327 return false; // Intrinsic's stub is not implemented on this platform
6328 }
6329 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
6330
6331 // no receiver because it is a static method
6332 Node* elementType = argument(0);
6333 Node* obj = argument(1);
6334 Node* offset = argument(2); // long
6335 Node* fromIndex = argument(4);
6336 Node* toIndex = argument(5);
6337 Node* indexPivot1 = argument(6);
6338 Node* indexPivot2 = argument(7);
6339 // PartitionOperation: argument(8) is ignored
6340
6341 Node* pivotIndices = nullptr;
6342 BasicType bt = T_ILLEGAL;
6343
6344 if (!check_array_sort_arguments(elementType, obj, bt)) {
6345 return false;
6346 }
6347 null_check(obj);
6348 // If obj is dead, only null-path is taken.
6349 if (stopped()) {
6350 return true;
6351 }
6352 // Set the original stack and the reexecute bit for the interpreter to reexecute
6353 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
6354 { PreserveReexecuteState preexecs(this);
6355 jvms()->set_should_reexecute(true);
6356
6357 Node* obj_adr = make_unsafe_address(obj, offset);
6358
6359 // create the pivotIndices array of type int and size = 2
6360 Node* size = intcon(2);
6361 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
6362 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
6363 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
6364 guarantee(alloc != nullptr, "created above");
6365 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
6366
6367 // pass the basic type enum to the stub
6368 Node* elemType = intcon(bt);
6369
6370 // Call the stub
6371 const char *stubName = "array_partition_stub";
6372 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
6373 stubAddr, stubName, TypePtr::BOTTOM,
6374 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
6375 indexPivot1, indexPivot2);
6376
6377 } // original reexecute is set back here
6378
6379 if (!stopped()) {
6380 set_result(pivotIndices);
6381 }
6382
6383 return true;
6384 }
6385
6386
6387 //------------------------------inline_array_sort-----------------------
6388 bool LibraryCallKit::inline_array_sort() {
6389 address stubAddr = StubRoutines::select_arraysort_function();
6390 if (stubAddr == nullptr) {
6391 return false; // Intrinsic's stub is not implemented on this platform
6392 }
6393 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
6394
6395 // no receiver because it is a static method
6396 Node* elementType = argument(0);
6397 Node* obj = argument(1);
6398 Node* offset = argument(2); // long
6399 Node* fromIndex = argument(4);
6400 Node* toIndex = argument(5);
6401 // SortOperation: argument(6) is ignored
6402
6403 BasicType bt = T_ILLEGAL;
6404
6405 if (!check_array_sort_arguments(elementType, obj, bt)) {
6406 return false;
6407 }
6408 null_check(obj);
6409 // If obj is dead, only null-path is taken.
6410 if (stopped()) {
6411 return true;
6412 }
6413 Node* obj_adr = make_unsafe_address(obj, offset);
6414
6415 // pass the basic type enum to the stub
6416 Node* elemType = intcon(bt);
6417
6418 // Call the stub.
6419 const char *stubName = "arraysort_stub";
6420 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
6421 stubAddr, stubName, TypePtr::BOTTOM,
6422 obj_adr, elemType, fromIndex, toIndex);
6423
6424 return true;
6425 }
6426
6427
6428 //------------------------------inline_arraycopy-----------------------
6429 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
6430 // Object dest, int destPos,
6431 // int length);
6432 bool LibraryCallKit::inline_arraycopy() {
6433 // Get the arguments.
6434 Node* src = argument(0); // type: oop
6435 Node* src_offset = argument(1); // type: int
6436 Node* dest = argument(2); // type: oop
6437 Node* dest_offset = argument(3); // type: int
6438 Node* length = argument(4); // type: int
6439
6440 uint new_idx = C->unique();
6441
6442 // Check for allocation before we add nodes that would confuse
6443 // tightly_coupled_allocation()
6444 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
6445
6446 int saved_reexecute_sp = -1;
6447 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
6448 // See arraycopy_restore_alloc_state() comment
6449 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
6450 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
6451 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
6452 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
6453
6454 // The following tests must be performed
6455 // (1) src and dest are arrays.
6456 // (2) src and dest arrays must have elements of the same BasicType
6457 // (3) src and dest must not be null.
6458 // (4) src_offset must not be negative.
6459 // (5) dest_offset must not be negative.
6460 // (6) length must not be negative.
6461 // (7) src_offset + length must not exceed length of src.
6462 // (8) dest_offset + length must not exceed length of dest.
6463 // (9) each element of an oop array must be assignable
6464
6465 // (3) src and dest must not be null.
6466 // always do this here because we need the JVM state for uncommon traps
6467 Node* null_ctl = top();
6468 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
6469 assert(null_ctl->is_top(), "no null control here");
6470 dest = null_check(dest, T_ARRAY);
6471
6472 if (!can_emit_guards) {
6473 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
6474 // guards but the arraycopy node could still take advantage of a
6475 // tightly allocated allocation. tightly_coupled_allocation() is
6476 // called again to make sure it takes the null check above into
6477 // account: the null check is mandatory and if it caused an
6478 // uncommon trap to be emitted then the allocation can't be
6479 // considered tightly coupled in this context.
6480 alloc = tightly_coupled_allocation(dest);
6481 }
6482
6483 bool validated = false;
6484
6485 const Type* src_type = _gvn.type(src);
6486 const Type* dest_type = _gvn.type(dest);
6487 const TypeAryPtr* top_src = src_type->isa_aryptr();
6488 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6489
6490 // Do we have the type of src?
6491 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6492 // Do we have the type of dest?
6493 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6494 // Is the type for src from speculation?
6495 bool src_spec = false;
6496 // Is the type for dest from speculation?
6497 bool dest_spec = false;
6498
6499 if ((!has_src || !has_dest) && can_emit_guards) {
6500 // We don't have sufficient type information, let's see if
6501 // speculative types can help. We need to have types for both src
6502 // and dest so that it pays off.
6503
6504 // Do we already have or could we have type information for src
6505 bool could_have_src = has_src;
6506 // Do we already have or could we have type information for dest
6507 bool could_have_dest = has_dest;
6508
6509 ciKlass* src_k = nullptr;
6510 if (!has_src) {
6511 src_k = src_type->speculative_type_not_null();
6512 if (src_k != nullptr && src_k->is_array_klass()) {
6513 could_have_src = true;
6514 }
6515 }
6516
6517 ciKlass* dest_k = nullptr;
6518 if (!has_dest) {
6519 dest_k = dest_type->speculative_type_not_null();
6520 if (dest_k != nullptr && dest_k->is_array_klass()) {
6521 could_have_dest = true;
6522 }
6523 }
6524
6525 if (could_have_src && could_have_dest) {
6526 // This is going to pay off so emit the required guards
6527 if (!has_src) {
6528 src = maybe_cast_profiled_obj(src, src_k, true);
6529 src_type = _gvn.type(src);
6530 top_src = src_type->isa_aryptr();
6531 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6532 src_spec = true;
6533 }
6534 if (!has_dest) {
6535 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6536 dest_type = _gvn.type(dest);
6537 top_dest = dest_type->isa_aryptr();
6538 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6539 dest_spec = true;
6540 }
6541 }
6542 }
6543
6544 if (has_src && has_dest && can_emit_guards) {
6545 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
6546 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
6547 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
6548 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
6549
6550 if (src_elem == dest_elem && top_src->is_flat() == top_dest->is_flat() && src_elem == T_OBJECT) {
6551 // If both arrays are object arrays then having the exact types
6552 // for both will remove the need for a subtype check at runtime
6553 // before the call and may make it possible to pick a faster copy
6554 // routine (without a subtype check on every element)
6555 // Do we have the exact type of src?
6556 bool could_have_src = src_spec;
6557 // Do we have the exact type of dest?
6558 bool could_have_dest = dest_spec;
6559 ciKlass* src_k = nullptr;
6560 ciKlass* dest_k = nullptr;
6561 if (!src_spec) {
6562 src_k = src_type->speculative_type_not_null();
6563 if (src_k != nullptr && src_k->is_array_klass()) {
6564 could_have_src = true;
6565 }
6566 }
6567 if (!dest_spec) {
6568 dest_k = dest_type->speculative_type_not_null();
6569 if (dest_k != nullptr && dest_k->is_array_klass()) {
6570 could_have_dest = true;
6571 }
6572 }
6573 if (could_have_src && could_have_dest) {
6574 // If we can have both exact types, emit the missing guards
6575 if (could_have_src && !src_spec) {
6576 src = maybe_cast_profiled_obj(src, src_k, true);
6577 src_type = _gvn.type(src);
6578 top_src = src_type->isa_aryptr();
6579 }
6580 if (could_have_dest && !dest_spec) {
6581 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6582 dest_type = _gvn.type(dest);
6583 top_dest = dest_type->isa_aryptr();
6584 }
6585 }
6586 }
6587 }
6588
6589 ciMethod* trap_method = method();
6590 int trap_bci = bci();
6591 if (saved_jvms_before_guards != nullptr) {
6592 trap_method = alloc->jvms()->method();
6593 trap_bci = alloc->jvms()->bci();
6594 }
6595
6596 bool negative_length_guard_generated = false;
6597
6598 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6599 can_emit_guards && !src->is_top() && !dest->is_top()) {
6600 // validate arguments: enables transformation the ArrayCopyNode
6601 validated = true;
6602
6603 RegionNode* slow_region = new RegionNode(1);
6604 record_for_igvn(slow_region);
6605
6606 // (1) src and dest are arrays.
6607 generate_non_array_guard(load_object_klass(src), slow_region, &src);
6608 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
6609
6610 // (2) src and dest arrays must have elements of the same BasicType
6611 // done at macro expansion or at Ideal transformation time
6612
6613 // (4) src_offset must not be negative.
6614 generate_negative_guard(src_offset, slow_region);
6615
6616 // (5) dest_offset must not be negative.
6617 generate_negative_guard(dest_offset, slow_region);
6618
6619 // (7) src_offset + length must not exceed length of src.
6620 generate_limit_guard(src_offset, length,
6621 load_array_length(src),
6622 slow_region);
6623
6624 // (8) dest_offset + length must not exceed length of dest.
6625 generate_limit_guard(dest_offset, length,
6626 load_array_length(dest),
6627 slow_region);
6628
6629 // (6) length must not be negative.
6630 // This is also checked in generate_arraycopy() during macro expansion, but
6631 // we also have to check it here for the case where the ArrayCopyNode will
6632 // be eliminated by Escape Analysis.
6633 if (EliminateAllocations) {
6634 generate_negative_guard(length, slow_region);
6635 negative_length_guard_generated = true;
6636 }
6637
6638 // (9) each element of an oop array must be assignable
6639 Node* dest_klass = load_object_klass(dest);
6640 Node* refined_dest_klass = dest_klass;
6641 if (src != dest) {
6642 dest_klass = load_non_refined_array_klass(refined_dest_klass);
6643 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6644 slow_region->add_req(not_subtype_ctrl);
6645 }
6646
6647 // TODO 8350865 Improve this. What about atomicity? Make sure this is always folded for type arrays.
6648 // If destination is null-restricted, source must be null-restricted as well: src_null_restricted || !dst_null_restricted
6649 Node* src_klass = load_object_klass(src);
6650 Node* adr_prop_src = basic_plus_adr(top(), src_klass, in_bytes(ArrayKlass::properties_offset()));
6651 Node* prop_src = _gvn.transform(LoadNode::make(_gvn, control(), immutable_memory(), adr_prop_src,
6652 _gvn.type(adr_prop_src)->is_ptr(), TypeInt::INT, T_INT,
6653 MemNode::unordered));
6654 Node* adr_prop_dest = basic_plus_adr(top(), refined_dest_klass, in_bytes(ArrayKlass::properties_offset()));
6655 Node* prop_dest = _gvn.transform(LoadNode::make(_gvn, control(), immutable_memory(), adr_prop_dest,
6656 _gvn.type(adr_prop_dest)->is_ptr(), TypeInt::INT, T_INT,
6657 MemNode::unordered));
6658
6659 const ArrayProperties props_null_restricted = ArrayProperties::Default().with_null_restricted();
6660 jint props_value = (jint)props_null_restricted.value();
6661
6662 prop_dest = _gvn.transform(new XorINode(prop_dest, intcon(props_value)));
6663 prop_src = _gvn.transform(new OrINode(prop_dest, prop_src));
6664 prop_src = _gvn.transform(new AndINode(prop_src, intcon(props_value)));
6665
6666 Node* chk = _gvn.transform(new CmpINode(prop_src, intcon(props_value)));
6667 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::ne));
6668 generate_fair_guard(tst, slow_region);
6669
6670 // TODO 8350865 This is too strong
6671 generate_fair_guard(flat_array_test(src), slow_region);
6672 generate_fair_guard(flat_array_test(dest), slow_region);
6673
6674 {
6675 PreserveJVMState pjvms(this);
6676 set_control(_gvn.transform(slow_region));
6677 uncommon_trap(Deoptimization::Reason_intrinsic,
6678 Deoptimization::Action_make_not_entrant);
6679 assert(stopped(), "Should be stopped");
6680 }
6681
6682 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->isa_klassptr();
6683 if (dest_klass_t == nullptr) {
6684 // refined_dest_klass may not be an array, which leads to dest_klass being top. This means we
6685 // are in a dead path.
6686 uncommon_trap(Deoptimization::Reason_intrinsic,
6687 Deoptimization::Action_make_not_entrant);
6688 return true;
6689 }
6690
6691 const Type* toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6692 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6693 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6694 }
6695
6696 if (stopped()) {
6697 return true;
6698 }
6699
6700 Node* dest_klass = load_object_klass(dest);
6701 dest_klass = load_non_refined_array_klass(dest_klass);
6702
6703 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6704 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6705 // so the compiler has a chance to eliminate them: during macro expansion,
6706 // we have to set their control (CastPP nodes are eliminated).
6707 load_object_klass(src), dest_klass,
6708 load_array_length(src), load_array_length(dest));
6709
6710 ac->set_arraycopy(validated);
6711
6712 Node* n = _gvn.transform(ac);
6713 if (n == ac) {
6714 ac->connect_outputs(this);
6715 } else {
6716 assert(validated, "shouldn't transform if all arguments not validated");
6717 set_all_memory(n);
6718 }
6719 clear_upper_avx();
6720
6721
6722 return true;
6723 }
6724
6725
6726 // Helper function which determines if an arraycopy immediately follows
6727 // an allocation, with no intervening tests or other escapes for the object.
6728 AllocateArrayNode*
6729 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6730 if (stopped()) return nullptr; // no fast path
6731 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6732
6733 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6734 if (alloc == nullptr) return nullptr;
6735
6736 Node* rawmem = memory(Compile::AliasIdxRaw);
6737 // Is the allocation's memory state untouched?
6738 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6739 // Bail out if there have been raw-memory effects since the allocation.
6740 // (Example: There might have been a call or safepoint.)
6741 return nullptr;
6742 }
6743 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6744 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6745 return nullptr;
6746 }
6747
6748 // There must be no unexpected observers of this allocation.
6749 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6750 Node* obs = ptr->fast_out(i);
6751 if (obs != this->map()) {
6752 return nullptr;
6753 }
6754 }
6755
6756 // This arraycopy must unconditionally follow the allocation of the ptr.
6757 Node* alloc_ctl = ptr->in(0);
6758 Node* ctl = control();
6759 while (ctl != alloc_ctl) {
6760 // There may be guards which feed into the slow_region.
6761 // Any other control flow means that we might not get a chance
6762 // to finish initializing the allocated object.
6763 // Various low-level checks bottom out in uncommon traps. These
6764 // are considered safe since we've already checked above that
6765 // there is no unexpected observer of this allocation.
6766 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6767 assert(ctl->in(0)->is_If(), "must be If");
6768 ctl = ctl->in(0)->in(0);
6769 } else {
6770 return nullptr;
6771 }
6772 }
6773
6774 // If we get this far, we have an allocation which immediately
6775 // precedes the arraycopy, and we can take over zeroing the new object.
6776 // The arraycopy will finish the initialization, and provide
6777 // a new control state to which we will anchor the destination pointer.
6778
6779 return alloc;
6780 }
6781
6782 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6783 if (node->is_IfProj()) {
6784 IfProjNode* other_proj = node->as_IfProj()->other_if_proj();
6785 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6786 Node* obs = other_proj->fast_out(j);
6787 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6788 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6789 return obs->as_CallStaticJava();
6790 }
6791 }
6792 }
6793 return nullptr;
6794 }
6795
6796 //-------------inline_encodeISOArray-----------------------------------
6797 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6798 // int java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6799 // int java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6800 // encode char[] to byte[] in ISO_8859_1 or ASCII
6801 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6802 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6803 // no receiver since it is static method
6804 Node *src = argument(0);
6805 Node *src_offset = argument(1);
6806 Node *dst = argument(2);
6807 Node *dst_offset = argument(3);
6808 Node *length = argument(4);
6809
6810 // Cast source & target arrays to not-null
6811 src = must_be_not_null(src, true);
6812 dst = must_be_not_null(dst, true);
6813 if (stopped()) {
6814 return true;
6815 }
6816
6817 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6818 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6819 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6820 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6821 // failed array check
6822 return false;
6823 }
6824
6825 // Figure out the size and type of the elements we will be copying.
6826 BasicType src_elem = src_type->elem()->array_element_basic_type();
6827 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6828 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6829 return false;
6830 }
6831
6832 // Check source & target bounds
6833 RegionNode* bailout = create_bailout();
6834 generate_string_range_check(src, src_offset, length, src_elem == T_BYTE, bailout);
6835 generate_string_range_check(dst, dst_offset, length, false, bailout);
6836 if (check_bailout(bailout)) {
6837 return true;
6838 }
6839
6840 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6841 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6842 // 'src_start' points to src array + scaled offset
6843 // 'dst_start' points to dst array + scaled offset
6844
6845 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6846 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6847 enc = _gvn.transform(enc);
6848 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6849 set_memory(res_mem, mtype);
6850 set_result(enc);
6851 clear_upper_avx();
6852
6853 return true;
6854 }
6855
6856 //-------------inline_multiplyToLen-----------------------------------
6857 bool LibraryCallKit::inline_multiplyToLen() {
6858 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6859
6860 address stubAddr = StubRoutines::multiplyToLen();
6861 if (stubAddr == nullptr) {
6862 return false; // Intrinsic's stub is not implemented on this platform
6863 }
6864 const char* stubName = "multiplyToLen";
6865
6866 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6867
6868 // no receiver because it is a static method
6869 Node* x = argument(0);
6870 Node* xlen = argument(1);
6871 Node* y = argument(2);
6872 Node* ylen = argument(3);
6873 Node* z = argument(4);
6874
6875 x = must_be_not_null(x, true);
6876 y = must_be_not_null(y, true);
6877
6878 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6879 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6880 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6881 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6882 // failed array check
6883 return false;
6884 }
6885
6886 BasicType x_elem = x_type->elem()->array_element_basic_type();
6887 BasicType y_elem = y_type->elem()->array_element_basic_type();
6888 if (x_elem != T_INT || y_elem != T_INT) {
6889 return false;
6890 }
6891
6892 Node* x_start = array_element_address(x, intcon(0), x_elem);
6893 Node* y_start = array_element_address(y, intcon(0), y_elem);
6894 // 'x_start' points to x array + scaled xlen
6895 // 'y_start' points to y array + scaled ylen
6896
6897 Node* z_start = array_element_address(z, intcon(0), T_INT);
6898
6899 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6900 OptoRuntime::multiplyToLen_Type(),
6901 stubAddr, stubName, TypePtr::BOTTOM,
6902 x_start, xlen, y_start, ylen, z_start);
6903
6904 C->set_has_split_ifs(true); // Has chance for split-if optimization
6905 set_result(z);
6906 return true;
6907 }
6908
6909 //-------------inline_squareToLen------------------------------------
6910 bool LibraryCallKit::inline_squareToLen() {
6911 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6912
6913 address stubAddr = StubRoutines::squareToLen();
6914 if (stubAddr == nullptr) {
6915 return false; // Intrinsic's stub is not implemented on this platform
6916 }
6917 const char* stubName = "squareToLen";
6918
6919 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6920
6921 Node* x = argument(0);
6922 Node* len = argument(1);
6923 Node* z = argument(2);
6924 Node* zlen = argument(3);
6925
6926 x = must_be_not_null(x, true);
6927 z = must_be_not_null(z, true);
6928
6929 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6930 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6931 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6932 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6933 // failed array check
6934 return false;
6935 }
6936
6937 BasicType x_elem = x_type->elem()->array_element_basic_type();
6938 BasicType z_elem = z_type->elem()->array_element_basic_type();
6939 if (x_elem != T_INT || z_elem != T_INT) {
6940 return false;
6941 }
6942
6943
6944 Node* x_start = array_element_address(x, intcon(0), x_elem);
6945 Node* z_start = array_element_address(z, intcon(0), z_elem);
6946
6947 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6948 OptoRuntime::squareToLen_Type(),
6949 stubAddr, stubName, TypePtr::BOTTOM,
6950 x_start, len, z_start, zlen);
6951
6952 set_result(z);
6953 return true;
6954 }
6955
6956 //-------------inline_mulAdd------------------------------------------
6957 bool LibraryCallKit::inline_mulAdd() {
6958 assert(UseMulAddIntrinsic, "not implemented on this platform");
6959
6960 address stubAddr = StubRoutines::mulAdd();
6961 if (stubAddr == nullptr) {
6962 return false; // Intrinsic's stub is not implemented on this platform
6963 }
6964 const char* stubName = "mulAdd";
6965
6966 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6967
6968 Node* out = argument(0);
6969 Node* in = argument(1);
6970 Node* offset = argument(2);
6971 Node* len = argument(3);
6972 Node* k = argument(4);
6973
6974 in = must_be_not_null(in, true);
6975 out = must_be_not_null(out, true);
6976
6977 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6978 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6979 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6980 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6981 // failed array check
6982 return false;
6983 }
6984
6985 BasicType out_elem = out_type->elem()->array_element_basic_type();
6986 BasicType in_elem = in_type->elem()->array_element_basic_type();
6987 if (out_elem != T_INT || in_elem != T_INT) {
6988 return false;
6989 }
6990
6991 Node* outlen = load_array_length(out);
6992 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6993 Node* out_start = array_element_address(out, intcon(0), out_elem);
6994 Node* in_start = array_element_address(in, intcon(0), in_elem);
6995
6996 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6997 OptoRuntime::mulAdd_Type(),
6998 stubAddr, stubName, TypePtr::BOTTOM,
6999 out_start,in_start, new_offset, len, k);
7000 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7001 set_result(result);
7002 return true;
7003 }
7004
7005 //-------------inline_montgomeryMultiply-----------------------------------
7006 bool LibraryCallKit::inline_montgomeryMultiply() {
7007 address stubAddr = StubRoutines::montgomeryMultiply();
7008 if (stubAddr == nullptr) {
7009 return false; // Intrinsic's stub is not implemented on this platform
7010 }
7011
7012 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
7013 const char* stubName = "montgomery_multiply";
7014
7015 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
7016
7017 Node* a = argument(0);
7018 Node* b = argument(1);
7019 Node* n = argument(2);
7020 Node* len = argument(3);
7021 Node* inv = argument(4);
7022 Node* m = argument(6);
7023
7024 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
7025 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
7026 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
7027 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
7028 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
7029 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
7030 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
7031 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
7032 // failed array check
7033 return false;
7034 }
7035
7036 BasicType a_elem = a_type->elem()->array_element_basic_type();
7037 BasicType b_elem = b_type->elem()->array_element_basic_type();
7038 BasicType n_elem = n_type->elem()->array_element_basic_type();
7039 BasicType m_elem = m_type->elem()->array_element_basic_type();
7040 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
7041 return false;
7042 }
7043
7044 // Make the call
7045 {
7046 Node* a_start = array_element_address(a, intcon(0), a_elem);
7047 Node* b_start = array_element_address(b, intcon(0), b_elem);
7048 Node* n_start = array_element_address(n, intcon(0), n_elem);
7049 Node* m_start = array_element_address(m, intcon(0), m_elem);
7050
7051 Node* call = make_runtime_call(RC_LEAF,
7052 OptoRuntime::montgomeryMultiply_Type(),
7053 stubAddr, stubName, TypePtr::BOTTOM,
7054 a_start, b_start, n_start, len, inv, top(),
7055 m_start);
7056 set_result(m);
7057 }
7058
7059 return true;
7060 }
7061
7062 bool LibraryCallKit::inline_montgomerySquare() {
7063 address stubAddr = StubRoutines::montgomerySquare();
7064 if (stubAddr == nullptr) {
7065 return false; // Intrinsic's stub is not implemented on this platform
7066 }
7067
7068 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
7069 const char* stubName = "montgomery_square";
7070
7071 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
7072
7073 Node* a = argument(0);
7074 Node* n = argument(1);
7075 Node* len = argument(2);
7076 Node* inv = argument(3);
7077 Node* m = argument(5);
7078
7079 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
7080 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
7081 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
7082 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
7083 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
7084 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
7085 // failed array check
7086 return false;
7087 }
7088
7089 BasicType a_elem = a_type->elem()->array_element_basic_type();
7090 BasicType n_elem = n_type->elem()->array_element_basic_type();
7091 BasicType m_elem = m_type->elem()->array_element_basic_type();
7092 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
7093 return false;
7094 }
7095
7096 // Make the call
7097 {
7098 Node* a_start = array_element_address(a, intcon(0), a_elem);
7099 Node* n_start = array_element_address(n, intcon(0), n_elem);
7100 Node* m_start = array_element_address(m, intcon(0), m_elem);
7101
7102 Node* call = make_runtime_call(RC_LEAF,
7103 OptoRuntime::montgomerySquare_Type(),
7104 stubAddr, stubName, TypePtr::BOTTOM,
7105 a_start, n_start, len, inv, top(),
7106 m_start);
7107 set_result(m);
7108 }
7109
7110 return true;
7111 }
7112
7113 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
7114 address stubAddr = nullptr;
7115 const char* stubName = nullptr;
7116
7117 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
7118 if (stubAddr == nullptr) {
7119 return false; // Intrinsic's stub is not implemented on this platform
7120 }
7121
7122 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
7123
7124 assert(callee()->signature()->size() == 5, "expected 5 arguments");
7125
7126 Node* newArr = argument(0);
7127 Node* oldArr = argument(1);
7128 Node* newIdx = argument(2);
7129 Node* shiftCount = argument(3);
7130 Node* numIter = argument(4);
7131
7132 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
7133 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
7134 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
7135 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
7136 return false;
7137 }
7138
7139 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
7140 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
7141 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
7142 return false;
7143 }
7144
7145 // Make the call
7146 {
7147 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
7148 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
7149
7150 Node* call = make_runtime_call(RC_LEAF,
7151 OptoRuntime::bigIntegerShift_Type(),
7152 stubAddr,
7153 stubName,
7154 TypePtr::BOTTOM,
7155 newArr_start,
7156 oldArr_start,
7157 newIdx,
7158 shiftCount,
7159 numIter);
7160 }
7161
7162 return true;
7163 }
7164
7165 //-------------inline_vectorizedMismatch------------------------------
7166 bool LibraryCallKit::inline_vectorizedMismatch() {
7167 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
7168
7169 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
7170 Node* obja = argument(0); // Object
7171 Node* aoffset = argument(1); // long
7172 Node* objb = argument(3); // Object
7173 Node* boffset = argument(4); // long
7174 Node* length = argument(6); // int
7175 Node* scale = argument(7); // int
7176
7177 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
7178 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
7179 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
7180 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
7181 scale == top()) {
7182 return false; // failed input validation
7183 }
7184
7185 Node* obja_adr = make_unsafe_address(obja, aoffset);
7186 Node* objb_adr = make_unsafe_address(objb, boffset);
7187
7188 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
7189 //
7190 // inline_limit = ArrayOperationPartialInlineSize / element_size;
7191 // if (length <= inline_limit) {
7192 // inline_path:
7193 // vmask = VectorMaskGen length
7194 // vload1 = LoadVectorMasked obja, vmask
7195 // vload2 = LoadVectorMasked objb, vmask
7196 // result1 = VectorCmpMasked vload1, vload2, vmask
7197 // } else {
7198 // call_stub_path:
7199 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
7200 // }
7201 // exit_block:
7202 // return Phi(result1, result2);
7203 //
7204 enum { inline_path = 1, // input is small enough to process it all at once
7205 stub_path = 2, // input is too large; call into the VM
7206 PATH_LIMIT = 3
7207 };
7208
7209 Node* exit_block = new RegionNode(PATH_LIMIT);
7210 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
7211 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
7212
7213 Node* call_stub_path = control();
7214
7215 BasicType elem_bt = T_ILLEGAL;
7216
7217 const TypeInt* scale_t = _gvn.type(scale)->is_int();
7218 if (scale_t->is_con()) {
7219 switch (scale_t->get_con()) {
7220 case 0: elem_bt = T_BYTE; break;
7221 case 1: elem_bt = T_SHORT; break;
7222 case 2: elem_bt = T_INT; break;
7223 case 3: elem_bt = T_LONG; break;
7224
7225 default: elem_bt = T_ILLEGAL; break; // not supported
7226 }
7227 }
7228
7229 int inline_limit = 0;
7230 bool do_partial_inline = false;
7231
7232 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
7233 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
7234 do_partial_inline = inline_limit >= 16;
7235 }
7236
7237 if (do_partial_inline) {
7238 assert(elem_bt != T_ILLEGAL, "sanity");
7239
7240 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
7241 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
7242 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
7243
7244 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
7245 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
7246 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
7247
7248 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
7249
7250 if (!stopped()) {
7251 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
7252
7253 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
7254 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
7255 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
7256 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
7257
7258 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
7259 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
7260 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
7261 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
7262
7263 exit_block->init_req(inline_path, control());
7264 memory_phi->init_req(inline_path, map()->memory());
7265 result_phi->init_req(inline_path, result);
7266
7267 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
7268 clear_upper_avx();
7269 }
7270 }
7271 }
7272
7273 if (call_stub_path != nullptr) {
7274 set_control(call_stub_path);
7275
7276 Node* call = make_runtime_call(RC_LEAF,
7277 OptoRuntime::vectorizedMismatch_Type(),
7278 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
7279 obja_adr, objb_adr, length, scale);
7280
7281 exit_block->init_req(stub_path, control());
7282 memory_phi->init_req(stub_path, map()->memory());
7283 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
7284 }
7285
7286 exit_block = _gvn.transform(exit_block);
7287 memory_phi = _gvn.transform(memory_phi);
7288 result_phi = _gvn.transform(result_phi);
7289
7290 record_for_igvn(exit_block);
7291 record_for_igvn(memory_phi);
7292 record_for_igvn(result_phi);
7293
7294 set_control(exit_block);
7295 set_all_memory(memory_phi);
7296 set_result(result_phi);
7297
7298 return true;
7299 }
7300
7301 //------------------------------inline_vectorizedHashcode----------------------------
7302 bool LibraryCallKit::inline_vectorizedHashCode() {
7303 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
7304
7305 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
7306 Node* array = argument(0);
7307 Node* offset = argument(1);
7308 Node* length = argument(2);
7309 Node* initialValue = argument(3);
7310 Node* basic_type = argument(4);
7311
7312 if (basic_type == top()) {
7313 return false; // failed input validation
7314 }
7315
7316 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
7317 if (!basic_type_t->is_con()) {
7318 return false; // Only intrinsify if mode argument is constant
7319 }
7320
7321 array = must_be_not_null(array, true);
7322
7323 BasicType bt = (BasicType)basic_type_t->get_con();
7324
7325 // Resolve address of first element
7326 Node* array_start = array_element_address(array, offset, bt);
7327
7328 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
7329 array_start, length, initialValue, basic_type)));
7330 clear_upper_avx();
7331
7332 return true;
7333 }
7334
7335 /**
7336 * Calculate CRC32 for byte.
7337 * int java.util.zip.CRC32.update(int crc, int b)
7338 */
7339 bool LibraryCallKit::inline_updateCRC32() {
7340 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7341 assert(callee()->signature()->size() == 2, "update has 2 parameters");
7342 // no receiver since it is static method
7343 Node* crc = argument(0); // type: int
7344 Node* b = argument(1); // type: int
7345
7346 /*
7347 * int c = ~ crc;
7348 * b = timesXtoThe32[(b ^ c) & 0xFF];
7349 * b = b ^ (c >>> 8);
7350 * crc = ~b;
7351 */
7352
7353 Node* M1 = intcon(-1);
7354 crc = _gvn.transform(new XorINode(crc, M1));
7355 Node* result = _gvn.transform(new XorINode(crc, b));
7356 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
7357
7358 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
7359 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
7360 Node* adr = off_heap_plus_addr(base, ConvI2X(offset));
7361 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
7362
7363 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
7364 result = _gvn.transform(new XorINode(crc, result));
7365 result = _gvn.transform(new XorINode(result, M1));
7366 set_result(result);
7367 return true;
7368 }
7369
7370 /**
7371 * Calculate CRC32 for byte[] array.
7372 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
7373 */
7374 bool LibraryCallKit::inline_updateBytesCRC32() {
7375 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7376 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7377 // no receiver since it is static method
7378 Node* crc = argument(0); // type: int
7379 Node* src = argument(1); // type: oop
7380 Node* offset = argument(2); // type: int
7381 Node* length = argument(3); // type: int
7382
7383 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7384 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7385 // failed array check
7386 return false;
7387 }
7388
7389 // Figure out the size and type of the elements we will be copying.
7390 BasicType src_elem = src_type->elem()->array_element_basic_type();
7391 if (src_elem != T_BYTE) {
7392 return false;
7393 }
7394
7395 // 'src_start' points to src array + scaled offset
7396 src = must_be_not_null(src, true);
7397 Node* src_start = array_element_address(src, offset, src_elem);
7398
7399 // We assume that range check is done by caller.
7400 // TODO: generate range check (offset+length < src.length) in debug VM.
7401
7402 // Call the stub.
7403 address stubAddr = StubRoutines::updateBytesCRC32();
7404 const char *stubName = "updateBytesCRC32";
7405
7406 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7407 stubAddr, stubName, TypePtr::BOTTOM,
7408 crc, src_start, length);
7409 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7410 set_result(result);
7411 return true;
7412 }
7413
7414 /**
7415 * Calculate CRC32 for ByteBuffer.
7416 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
7417 */
7418 bool LibraryCallKit::inline_updateByteBufferCRC32() {
7419 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7420 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7421 // no receiver since it is static method
7422 Node* crc = argument(0); // type: int
7423 Node* src = argument(1); // type: long
7424 Node* offset = argument(3); // type: int
7425 Node* length = argument(4); // type: int
7426
7427 src = ConvL2X(src); // adjust Java long to machine word
7428 Node* base = _gvn.transform(new CastX2PNode(src));
7429 offset = ConvI2X(offset);
7430
7431 // 'src_start' points to src array + scaled offset
7432 Node* src_start = off_heap_plus_addr(base, offset);
7433
7434 // Call the stub.
7435 address stubAddr = StubRoutines::updateBytesCRC32();
7436 const char *stubName = "updateBytesCRC32";
7437
7438 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7439 stubAddr, stubName, TypePtr::BOTTOM,
7440 crc, src_start, length);
7441 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7442 set_result(result);
7443 return true;
7444 }
7445
7446 //------------------------------get_table_from_crc32c_class-----------------------
7447 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
7448 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
7449 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
7450
7451 return table;
7452 }
7453
7454 //------------------------------inline_updateBytesCRC32C-----------------------
7455 //
7456 // Calculate CRC32C for byte[] array.
7457 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
7458 //
7459 bool LibraryCallKit::inline_updateBytesCRC32C() {
7460 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7461 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7462 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7463 // no receiver since it is a static method
7464 Node* crc = argument(0); // type: int
7465 Node* src = argument(1); // type: oop
7466 Node* offset = argument(2); // type: int
7467 Node* end = argument(3); // type: int
7468
7469 Node* length = _gvn.transform(new SubINode(end, offset));
7470
7471 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7472 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7473 // failed array check
7474 return false;
7475 }
7476
7477 // Figure out the size and type of the elements we will be copying.
7478 BasicType src_elem = src_type->elem()->array_element_basic_type();
7479 if (src_elem != T_BYTE) {
7480 return false;
7481 }
7482
7483 // 'src_start' points to src array + scaled offset
7484 src = must_be_not_null(src, true);
7485 Node* src_start = array_element_address(src, offset, src_elem);
7486
7487 // static final int[] byteTable in class CRC32C
7488 Node* table = get_table_from_crc32c_class(callee()->holder());
7489 table = must_be_not_null(table, true);
7490 Node* table_start = array_element_address(table, intcon(0), T_INT);
7491
7492 // We assume that range check is done by caller.
7493 // TODO: generate range check (offset+length < src.length) in debug VM.
7494
7495 // Call the stub.
7496 address stubAddr = StubRoutines::updateBytesCRC32C();
7497 const char *stubName = "updateBytesCRC32C";
7498
7499 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7500 stubAddr, stubName, TypePtr::BOTTOM,
7501 crc, src_start, length, table_start);
7502 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7503 set_result(result);
7504 return true;
7505 }
7506
7507 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
7508 //
7509 // Calculate CRC32C for DirectByteBuffer.
7510 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
7511 //
7512 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
7513 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7514 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
7515 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7516 // no receiver since it is a static method
7517 Node* crc = argument(0); // type: int
7518 Node* src = argument(1); // type: long
7519 Node* offset = argument(3); // type: int
7520 Node* end = argument(4); // type: int
7521
7522 Node* length = _gvn.transform(new SubINode(end, offset));
7523
7524 src = ConvL2X(src); // adjust Java long to machine word
7525 Node* base = _gvn.transform(new CastX2PNode(src));
7526 offset = ConvI2X(offset);
7527
7528 // 'src_start' points to src array + scaled offset
7529 Node* src_start = off_heap_plus_addr(base, offset);
7530
7531 // static final int[] byteTable in class CRC32C
7532 Node* table = get_table_from_crc32c_class(callee()->holder());
7533 table = must_be_not_null(table, true);
7534 Node* table_start = array_element_address(table, intcon(0), T_INT);
7535
7536 // Call the stub.
7537 address stubAddr = StubRoutines::updateBytesCRC32C();
7538 const char *stubName = "updateBytesCRC32C";
7539
7540 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7541 stubAddr, stubName, TypePtr::BOTTOM,
7542 crc, src_start, length, table_start);
7543 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7544 set_result(result);
7545 return true;
7546 }
7547
7548 //------------------------------inline_updateBytesAdler32----------------------
7549 //
7550 // Calculate Adler32 checksum for byte[] array.
7551 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
7552 //
7553 bool LibraryCallKit::inline_updateBytesAdler32() {
7554 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7555 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7556 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7557 // no receiver since it is static method
7558 Node* crc = argument(0); // type: int
7559 Node* src = argument(1); // type: oop
7560 Node* offset = argument(2); // type: int
7561 Node* length = argument(3); // type: int
7562
7563 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7564 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7565 // failed array check
7566 return false;
7567 }
7568
7569 // Figure out the size and type of the elements we will be copying.
7570 BasicType src_elem = src_type->elem()->array_element_basic_type();
7571 if (src_elem != T_BYTE) {
7572 return false;
7573 }
7574
7575 // 'src_start' points to src array + scaled offset
7576 Node* src_start = array_element_address(src, offset, src_elem);
7577
7578 // We assume that range check is done by caller.
7579 // TODO: generate range check (offset+length < src.length) in debug VM.
7580
7581 // Call the stub.
7582 address stubAddr = StubRoutines::updateBytesAdler32();
7583 const char *stubName = "updateBytesAdler32";
7584
7585 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7586 stubAddr, stubName, TypePtr::BOTTOM,
7587 crc, src_start, length);
7588 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7589 set_result(result);
7590 return true;
7591 }
7592
7593 //------------------------------inline_updateByteBufferAdler32---------------
7594 //
7595 // Calculate Adler32 checksum for DirectByteBuffer.
7596 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
7597 //
7598 bool LibraryCallKit::inline_updateByteBufferAdler32() {
7599 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7600 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7601 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7602 // no receiver since it is static method
7603 Node* crc = argument(0); // type: int
7604 Node* src = argument(1); // type: long
7605 Node* offset = argument(3); // type: int
7606 Node* length = argument(4); // type: int
7607
7608 src = ConvL2X(src); // adjust Java long to machine word
7609 Node* base = _gvn.transform(new CastX2PNode(src));
7610 offset = ConvI2X(offset);
7611
7612 // 'src_start' points to src array + scaled offset
7613 Node* src_start = off_heap_plus_addr(base, offset);
7614
7615 // Call the stub.
7616 address stubAddr = StubRoutines::updateBytesAdler32();
7617 const char *stubName = "updateBytesAdler32";
7618
7619 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7620 stubAddr, stubName, TypePtr::BOTTOM,
7621 crc, src_start, length);
7622
7623 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7624 set_result(result);
7625 return true;
7626 }
7627
7628 //----------------------------inline_reference_get0----------------------------
7629 // public T java.lang.ref.Reference.get();
7630 bool LibraryCallKit::inline_reference_get0() {
7631 const int referent_offset = java_lang_ref_Reference::referent_offset();
7632
7633 // Get the argument:
7634 Node* reference_obj = null_check_receiver();
7635 if (stopped()) return true;
7636
7637 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7638 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7639 decorators, /*is_static*/ false, nullptr);
7640 if (result == nullptr) return false;
7641
7642 // Add memory barrier to prevent commoning reads from this field
7643 // across safepoint since GC can change its value.
7644 insert_mem_bar(Op_MemBarCPUOrder);
7645
7646 set_result(result);
7647 return true;
7648 }
7649
7650 //----------------------------inline_reference_refersTo0----------------------------
7651 // bool java.lang.ref.Reference.refersTo0();
7652 // bool java.lang.ref.PhantomReference.refersTo0();
7653 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7654 // Get arguments:
7655 Node* reference_obj = null_check_receiver();
7656 Node* other_obj = argument(1);
7657 if (stopped()) return true;
7658
7659 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7660 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7661 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7662 decorators, /*is_static*/ false, nullptr);
7663 if (referent == nullptr) return false;
7664
7665 // Add memory barrier to prevent commoning reads from this field
7666 // across safepoint since GC can change its value.
7667 insert_mem_bar(Op_MemBarCPUOrder);
7668
7669 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7670 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7671 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7672
7673 RegionNode* region = new RegionNode(3);
7674 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7675
7676 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7677 region->init_req(1, if_true);
7678 phi->init_req(1, intcon(1));
7679
7680 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7681 region->init_req(2, if_false);
7682 phi->init_req(2, intcon(0));
7683
7684 set_control(_gvn.transform(region));
7685 record_for_igvn(region);
7686 set_result(_gvn.transform(phi));
7687 return true;
7688 }
7689
7690 //----------------------------inline_reference_clear0----------------------------
7691 // void java.lang.ref.Reference.clear0();
7692 // void java.lang.ref.PhantomReference.clear0();
7693 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7694 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7695
7696 // Get arguments
7697 Node* reference_obj = null_check_receiver();
7698 if (stopped()) return true;
7699
7700 // Common access parameters
7701 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7702 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7703 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7704 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7705 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7706
7707 Node* referent = access_load_at(reference_obj,
7708 referent_field_addr,
7709 referent_field_addr_type,
7710 val_type,
7711 T_OBJECT,
7712 decorators);
7713
7714 IdealKit ideal(this);
7715 #define __ ideal.
7716 __ if_then(referent, BoolTest::ne, null());
7717 sync_kit(ideal);
7718 access_store_at(reference_obj,
7719 referent_field_addr,
7720 referent_field_addr_type,
7721 null(),
7722 val_type,
7723 T_OBJECT,
7724 decorators);
7725 __ sync_kit(this);
7726 __ end_if();
7727 final_sync(ideal);
7728 #undef __
7729
7730 return true;
7731 }
7732
7733 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7734 DecoratorSet decorators, bool is_static,
7735 ciInstanceKlass* fromKls) {
7736 if (fromKls == nullptr) {
7737 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7738 assert(tinst != nullptr, "obj is null");
7739 assert(tinst->is_loaded(), "obj is not loaded");
7740 fromKls = tinst->instance_klass();
7741 } else {
7742 assert(is_static, "only for static field access");
7743 }
7744 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7745 ciSymbol::make(fieldTypeString),
7746 is_static);
7747
7748 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7749 if (field == nullptr) return (Node *) nullptr;
7750
7751 if (is_static) {
7752 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7753 fromObj = makecon(tip);
7754 }
7755
7756 // Next code copied from Parse::do_get_xxx():
7757
7758 // Compute address and memory type.
7759 int offset = field->offset_in_bytes();
7760 bool is_vol = field->is_volatile();
7761 ciType* field_klass = field->type();
7762 assert(field_klass->is_loaded(), "should be loaded");
7763 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7764 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7765 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7766 "slice of address and input slice don't match");
7767 BasicType bt = field->layout_type();
7768
7769 // Build the resultant type of the load
7770 const Type *type;
7771 if (bt == T_OBJECT) {
7772 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7773 } else {
7774 type = Type::get_const_basic_type(bt);
7775 }
7776
7777 if (is_vol) {
7778 decorators |= MO_SEQ_CST;
7779 }
7780
7781 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7782 }
7783
7784 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7785 bool is_exact /* true */, bool is_static /* false */,
7786 ciInstanceKlass * fromKls /* nullptr */) {
7787 if (fromKls == nullptr) {
7788 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7789 assert(tinst != nullptr, "obj is null");
7790 assert(tinst->is_loaded(), "obj is not loaded");
7791 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7792 fromKls = tinst->instance_klass();
7793 }
7794 else {
7795 assert(is_static, "only for static field access");
7796 }
7797 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7798 ciSymbol::make(fieldTypeString),
7799 is_static);
7800
7801 assert(field != nullptr, "undefined field");
7802 assert(!field->is_volatile(), "not defined for volatile fields");
7803
7804 if (is_static) {
7805 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7806 fromObj = makecon(tip);
7807 }
7808
7809 // Next code copied from Parse::do_get_xxx():
7810
7811 // Compute address and memory type.
7812 int offset = field->offset_in_bytes();
7813 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7814
7815 return adr;
7816 }
7817
7818 //------------------------------inline_aescrypt_Block-----------------------
7819 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7820 address stubAddr = nullptr;
7821 const char *stubName;
7822 bool is_decrypt = false;
7823 assert(UseAES, "need AES instruction support");
7824
7825 switch(id) {
7826 case vmIntrinsics::_aescrypt_encryptBlock:
7827 stubAddr = StubRoutines::aescrypt_encryptBlock();
7828 stubName = "aescrypt_encryptBlock";
7829 break;
7830 case vmIntrinsics::_aescrypt_decryptBlock:
7831 stubAddr = StubRoutines::aescrypt_decryptBlock();
7832 stubName = "aescrypt_decryptBlock";
7833 is_decrypt = true;
7834 break;
7835 default:
7836 break;
7837 }
7838 if (stubAddr == nullptr) return false;
7839
7840 Node* aescrypt_object = argument(0);
7841 Node* src = argument(1);
7842 Node* src_offset = argument(2);
7843 Node* dest = argument(3);
7844 Node* dest_offset = argument(4);
7845
7846 src = must_be_not_null(src, true);
7847 dest = must_be_not_null(dest, true);
7848
7849 // (1) src and dest are arrays.
7850 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7851 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7852 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7853 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7854
7855 // for the quick and dirty code we will skip all the checks.
7856 // we are just trying to get the call to be generated.
7857 Node* src_start = src;
7858 Node* dest_start = dest;
7859 if (src_offset != nullptr || dest_offset != nullptr) {
7860 assert(src_offset != nullptr && dest_offset != nullptr, "");
7861 src_start = array_element_address(src, src_offset, T_BYTE);
7862 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7863 }
7864
7865 // now need to get the start of its expanded key array
7866 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7867 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7868 if (k_start == nullptr) return false;
7869
7870 // Call the stub.
7871 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7872 stubAddr, stubName, TypePtr::BOTTOM,
7873 src_start, dest_start, k_start);
7874
7875 return true;
7876 }
7877
7878 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7879 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7880 address stubAddr = nullptr;
7881 const char *stubName = nullptr;
7882 bool is_decrypt = false;
7883 assert(UseAES, "need AES instruction support");
7884
7885 switch(id) {
7886 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7887 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7888 stubName = "cipherBlockChaining_encryptAESCrypt";
7889 break;
7890 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7891 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7892 stubName = "cipherBlockChaining_decryptAESCrypt";
7893 is_decrypt = true;
7894 break;
7895 default:
7896 break;
7897 }
7898 if (stubAddr == nullptr) return false;
7899
7900 Node* cipherBlockChaining_object = argument(0);
7901 Node* src = argument(1);
7902 Node* src_offset = argument(2);
7903 Node* len = argument(3);
7904 Node* dest = argument(4);
7905 Node* dest_offset = argument(5);
7906
7907 src = must_be_not_null(src, false);
7908 dest = must_be_not_null(dest, false);
7909
7910 // (1) src and dest are arrays.
7911 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7912 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7913 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7914 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7915
7916 // checks are the responsibility of the caller
7917 Node* src_start = src;
7918 Node* dest_start = dest;
7919 if (src_offset != nullptr || dest_offset != nullptr) {
7920 assert(src_offset != nullptr && dest_offset != nullptr, "");
7921 src_start = array_element_address(src, src_offset, T_BYTE);
7922 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7923 }
7924
7925 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7926 // (because of the predicated logic executed earlier).
7927 // so we cast it here safely.
7928 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7929
7930 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7931 if (embeddedCipherObj == nullptr) return false;
7932
7933 // cast it to what we know it will be at runtime
7934 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7935 assert(tinst != nullptr, "CBC obj is null");
7936 assert(tinst->is_loaded(), "CBC obj is not loaded");
7937 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7938 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7939
7940 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7941 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7942 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7943 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7944 aescrypt_object = _gvn.transform(aescrypt_object);
7945
7946 // we need to get the start of the aescrypt_object's expanded key array
7947 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7948 if (k_start == nullptr) return false;
7949
7950 // similarly, get the start address of the r vector
7951 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7952 if (objRvec == nullptr) return false;
7953 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7954
7955 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7956 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7957 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7958 stubAddr, stubName, TypePtr::BOTTOM,
7959 src_start, dest_start, k_start, r_start, len);
7960
7961 // return cipher length (int)
7962 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7963 set_result(retvalue);
7964 return true;
7965 }
7966
7967 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7968 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7969 address stubAddr = nullptr;
7970 const char *stubName = nullptr;
7971 bool is_decrypt = false;
7972 assert(UseAES, "need AES instruction support");
7973
7974 switch (id) {
7975 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7976 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7977 stubName = "electronicCodeBook_encryptAESCrypt";
7978 break;
7979 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7980 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7981 stubName = "electronicCodeBook_decryptAESCrypt";
7982 is_decrypt = true;
7983 break;
7984 default:
7985 break;
7986 }
7987
7988 if (stubAddr == nullptr) return false;
7989
7990 Node* electronicCodeBook_object = argument(0);
7991 Node* src = argument(1);
7992 Node* src_offset = argument(2);
7993 Node* len = argument(3);
7994 Node* dest = argument(4);
7995 Node* dest_offset = argument(5);
7996
7997 // (1) src and dest are arrays.
7998 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7999 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8000 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8001 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8002
8003 // checks are the responsibility of the caller
8004 Node* src_start = src;
8005 Node* dest_start = dest;
8006 if (src_offset != nullptr || dest_offset != nullptr) {
8007 assert(src_offset != nullptr && dest_offset != nullptr, "");
8008 src_start = array_element_address(src, src_offset, T_BYTE);
8009 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8010 }
8011
8012 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8013 // (because of the predicated logic executed earlier).
8014 // so we cast it here safely.
8015 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8016
8017 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8018 if (embeddedCipherObj == nullptr) return false;
8019
8020 // cast it to what we know it will be at runtime
8021 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
8022 assert(tinst != nullptr, "ECB obj is null");
8023 assert(tinst->is_loaded(), "ECB obj is not loaded");
8024 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8025 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8026
8027 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8028 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8029 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8030 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8031 aescrypt_object = _gvn.transform(aescrypt_object);
8032
8033 // we need to get the start of the aescrypt_object's expanded key array
8034 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
8035 if (k_start == nullptr) return false;
8036
8037 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8038 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
8039 OptoRuntime::electronicCodeBook_aescrypt_Type(),
8040 stubAddr, stubName, TypePtr::BOTTOM,
8041 src_start, dest_start, k_start, len);
8042
8043 // return cipher length (int)
8044 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
8045 set_result(retvalue);
8046 return true;
8047 }
8048
8049 //------------------------------inline_counterMode_AESCrypt-----------------------
8050 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
8051 assert(UseAES, "need AES instruction support");
8052 if (!UseAESCTRIntrinsics) return false;
8053
8054 address stubAddr = nullptr;
8055 const char *stubName = nullptr;
8056 if (id == vmIntrinsics::_counterMode_AESCrypt) {
8057 stubAddr = StubRoutines::counterMode_AESCrypt();
8058 stubName = "counterMode_AESCrypt";
8059 }
8060 if (stubAddr == nullptr) return false;
8061
8062 Node* counterMode_object = argument(0);
8063 Node* src = argument(1);
8064 Node* src_offset = argument(2);
8065 Node* len = argument(3);
8066 Node* dest = argument(4);
8067 Node* dest_offset = argument(5);
8068
8069 // (1) src and dest are arrays.
8070 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8071 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8072 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8073 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8074
8075 // checks are the responsibility of the caller
8076 Node* src_start = src;
8077 Node* dest_start = dest;
8078 if (src_offset != nullptr || dest_offset != nullptr) {
8079 assert(src_offset != nullptr && dest_offset != nullptr, "");
8080 src_start = array_element_address(src, src_offset, T_BYTE);
8081 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8082 }
8083
8084 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8085 // (because of the predicated logic executed earlier).
8086 // so we cast it here safely.
8087 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8088 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8089 if (embeddedCipherObj == nullptr) return false;
8090 // cast it to what we know it will be at runtime
8091 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
8092 assert(tinst != nullptr, "CTR obj is null");
8093 assert(tinst->is_loaded(), "CTR obj is not loaded");
8094 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8095 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8096 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8097 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8098 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8099 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8100 aescrypt_object = _gvn.transform(aescrypt_object);
8101 // we need to get the start of the aescrypt_object's expanded key array
8102 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
8103 if (k_start == nullptr) return false;
8104 // similarly, get the start address of the r vector
8105 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
8106 if (obj_counter == nullptr) return false;
8107 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
8108
8109 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
8110 if (saved_encCounter == nullptr) return false;
8111 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
8112 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
8113
8114 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8115 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8116 OptoRuntime::counterMode_aescrypt_Type(),
8117 stubAddr, stubName, TypePtr::BOTTOM,
8118 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
8119
8120 // return cipher length (int)
8121 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
8122 set_result(retvalue);
8123 return true;
8124 }
8125
8126 //------------------------------get_key_start_from_aescrypt_object-----------------------
8127 Node* LibraryCallKit::get_key_start_from_aescrypt_object(Node* aescrypt_object, bool is_decrypt) {
8128 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
8129 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
8130 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
8131 // The following platform specific stubs of encryption and decryption use the same round keys.
8132 #if defined(PPC64) || defined(S390) || defined(RISCV64)
8133 bool use_decryption_key = false;
8134 #else
8135 bool use_decryption_key = is_decrypt;
8136 #endif
8137 Node* objAESCryptKey = load_field_from_object(aescrypt_object, use_decryption_key ? "sessionKd" : "sessionKe", "[I");
8138 assert(objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AES_Crypt");
8139 if (objAESCryptKey == nullptr) return (Node *) nullptr;
8140
8141 // now have the array, need to get the start address of the selected key array
8142 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
8143 return k_start;
8144 }
8145
8146 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
8147 // Return node representing slow path of predicate check.
8148 // the pseudo code we want to emulate with this predicate is:
8149 // for encryption:
8150 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8151 // for decryption:
8152 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8153 // note cipher==plain is more conservative than the original java code but that's OK
8154 //
8155 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
8156 // The receiver was checked for null already.
8157 Node* objCBC = argument(0);
8158
8159 Node* src = argument(1);
8160 Node* dest = argument(4);
8161
8162 // Load embeddedCipher field of CipherBlockChaining object.
8163 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8164
8165 // get AESCrypt klass for instanceOf check
8166 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8167 // will have same classloader as CipherBlockChaining object
8168 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
8169 assert(tinst != nullptr, "CBCobj is null");
8170 assert(tinst->is_loaded(), "CBCobj is not loaded");
8171
8172 // we want to do an instanceof comparison against the AESCrypt class
8173 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8174 if (!klass_AESCrypt->is_loaded()) {
8175 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8176 Node* ctrl = control();
8177 set_control(top()); // no regular fast path
8178 return ctrl;
8179 }
8180
8181 src = must_be_not_null(src, true);
8182 dest = must_be_not_null(dest, true);
8183
8184 // Resolve oops to stable for CmpP below.
8185 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8186
8187 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8188 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8189 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8190
8191 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8192
8193 // for encryption, we are done
8194 if (!decrypting)
8195 return instof_false; // even if it is null
8196
8197 // for decryption, we need to add a further check to avoid
8198 // taking the intrinsic path when cipher and plain are the same
8199 // see the original java code for why.
8200 RegionNode* region = new RegionNode(3);
8201 region->init_req(1, instof_false);
8202
8203 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8204 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8205 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8206 region->init_req(2, src_dest_conjoint);
8207
8208 record_for_igvn(region);
8209 return _gvn.transform(region);
8210 }
8211
8212 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
8213 // Return node representing slow path of predicate check.
8214 // the pseudo code we want to emulate with this predicate is:
8215 // for encryption:
8216 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8217 // for decryption:
8218 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8219 // note cipher==plain is more conservative than the original java code but that's OK
8220 //
8221 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
8222 // The receiver was checked for null already.
8223 Node* objECB = argument(0);
8224
8225 // Load embeddedCipher field of ElectronicCodeBook object.
8226 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8227
8228 // get AESCrypt klass for instanceOf check
8229 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8230 // will have same classloader as ElectronicCodeBook object
8231 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
8232 assert(tinst != nullptr, "ECBobj is null");
8233 assert(tinst->is_loaded(), "ECBobj is not loaded");
8234
8235 // we want to do an instanceof comparison against the AESCrypt class
8236 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8237 if (!klass_AESCrypt->is_loaded()) {
8238 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8239 Node* ctrl = control();
8240 set_control(top()); // no regular fast path
8241 return ctrl;
8242 }
8243 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8244
8245 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8246 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8247 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8248
8249 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8250
8251 // for encryption, we are done
8252 if (!decrypting)
8253 return instof_false; // even if it is null
8254
8255 // for decryption, we need to add a further check to avoid
8256 // taking the intrinsic path when cipher and plain are the same
8257 // see the original java code for why.
8258 RegionNode* region = new RegionNode(3);
8259 region->init_req(1, instof_false);
8260 Node* src = argument(1);
8261 Node* dest = argument(4);
8262 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8263 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8264 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8265 region->init_req(2, src_dest_conjoint);
8266
8267 record_for_igvn(region);
8268 return _gvn.transform(region);
8269 }
8270
8271 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
8272 // Return node representing slow path of predicate check.
8273 // the pseudo code we want to emulate with this predicate is:
8274 // for encryption:
8275 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8276 // for decryption:
8277 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8278 // note cipher==plain is more conservative than the original java code but that's OK
8279 //
8280
8281 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
8282 // The receiver was checked for null already.
8283 Node* objCTR = argument(0);
8284
8285 // Load embeddedCipher field of CipherBlockChaining object.
8286 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8287
8288 // get AESCrypt klass for instanceOf check
8289 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8290 // will have same classloader as CipherBlockChaining object
8291 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
8292 assert(tinst != nullptr, "CTRobj is null");
8293 assert(tinst->is_loaded(), "CTRobj is not loaded");
8294
8295 // we want to do an instanceof comparison against the AESCrypt class
8296 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8297 if (!klass_AESCrypt->is_loaded()) {
8298 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8299 Node* ctrl = control();
8300 set_control(top()); // no regular fast path
8301 return ctrl;
8302 }
8303
8304 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8305 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8306 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8307 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8308 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8309
8310 return instof_false; // even if it is null
8311 }
8312
8313 //------------------------------inline_ghash_processBlocks
8314 bool LibraryCallKit::inline_ghash_processBlocks() {
8315 address stubAddr;
8316 const char *stubName;
8317 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
8318
8319 stubAddr = StubRoutines::ghash_processBlocks();
8320 stubName = "ghash_processBlocks";
8321
8322 Node* data = argument(0);
8323 Node* offset = argument(1);
8324 Node* len = argument(2);
8325 Node* state = argument(3);
8326 Node* subkeyH = argument(4);
8327
8328 state = must_be_not_null(state, true);
8329 subkeyH = must_be_not_null(subkeyH, true);
8330 data = must_be_not_null(data, true);
8331
8332 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8333 assert(state_start, "state is null");
8334 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
8335 assert(subkeyH_start, "subkeyH is null");
8336 Node* data_start = array_element_address(data, offset, T_BYTE);
8337 assert(data_start, "data is null");
8338
8339 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
8340 OptoRuntime::ghash_processBlocks_Type(),
8341 stubAddr, stubName, TypePtr::BOTTOM,
8342 state_start, subkeyH_start, data_start, len);
8343 return true;
8344 }
8345
8346 //------------------------------inline_chacha20Block
8347 bool LibraryCallKit::inline_chacha20Block() {
8348 address stubAddr;
8349 const char *stubName;
8350 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
8351
8352 stubAddr = StubRoutines::chacha20Block();
8353 stubName = "chacha20Block";
8354
8355 Node* state = argument(0);
8356 Node* result = argument(1);
8357
8358 state = must_be_not_null(state, true);
8359 result = must_be_not_null(result, true);
8360
8361 Node* state_start = array_element_address(state, intcon(0), T_INT);
8362 assert(state_start, "state is null");
8363 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
8364 assert(result_start, "result is null");
8365
8366 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
8367 OptoRuntime::chacha20Block_Type(),
8368 stubAddr, stubName, TypePtr::BOTTOM,
8369 state_start, result_start);
8370 // return key stream length (int)
8371 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
8372 set_result(retvalue);
8373 return true;
8374 }
8375
8376 //------------------------------inline_kyberNtt
8377 bool LibraryCallKit::inline_kyberNtt() {
8378 address stubAddr;
8379 const char *stubName;
8380 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8381 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
8382
8383 stubAddr = StubRoutines::kyberNtt();
8384 stubName = "kyberNtt";
8385 if (!stubAddr) return false;
8386
8387 Node* coeffs = argument(0);
8388 Node* ntt_zetas = argument(1);
8389
8390 coeffs = must_be_not_null(coeffs, true);
8391 ntt_zetas = must_be_not_null(ntt_zetas, true);
8392
8393 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8394 assert(coeffs_start, "coeffs is null");
8395 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
8396 assert(ntt_zetas_start, "ntt_zetas is null");
8397 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8398 OptoRuntime::kyberNtt_Type(),
8399 stubAddr, stubName, TypePtr::BOTTOM,
8400 coeffs_start, ntt_zetas_start);
8401 // return an int
8402 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
8403 set_result(retvalue);
8404 return true;
8405 }
8406
8407 //------------------------------inline_kyberInverseNtt
8408 bool LibraryCallKit::inline_kyberInverseNtt() {
8409 address stubAddr;
8410 const char *stubName;
8411 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8412 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
8413
8414 stubAddr = StubRoutines::kyberInverseNtt();
8415 stubName = "kyberInverseNtt";
8416 if (!stubAddr) return false;
8417
8418 Node* coeffs = argument(0);
8419 Node* zetas = argument(1);
8420
8421 coeffs = must_be_not_null(coeffs, true);
8422 zetas = must_be_not_null(zetas, true);
8423
8424 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8425 assert(coeffs_start, "coeffs is null");
8426 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8427 assert(zetas_start, "inverseNtt_zetas is null");
8428 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8429 OptoRuntime::kyberInverseNtt_Type(),
8430 stubAddr, stubName, TypePtr::BOTTOM,
8431 coeffs_start, zetas_start);
8432
8433 // return an int
8434 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
8435 set_result(retvalue);
8436 return true;
8437 }
8438
8439 //------------------------------inline_kyberNttMult
8440 bool LibraryCallKit::inline_kyberNttMult() {
8441 address stubAddr;
8442 const char *stubName;
8443 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8444 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
8445
8446 stubAddr = StubRoutines::kyberNttMult();
8447 stubName = "kyberNttMult";
8448 if (!stubAddr) return false;
8449
8450 Node* result = argument(0);
8451 Node* ntta = argument(1);
8452 Node* nttb = argument(2);
8453 Node* zetas = argument(3);
8454
8455 result = must_be_not_null(result, true);
8456 ntta = must_be_not_null(ntta, true);
8457 nttb = must_be_not_null(nttb, true);
8458 zetas = must_be_not_null(zetas, true);
8459
8460 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8461 assert(result_start, "result is null");
8462 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
8463 assert(ntta_start, "ntta is null");
8464 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
8465 assert(nttb_start, "nttb is null");
8466 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8467 assert(zetas_start, "nttMult_zetas is null");
8468 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8469 OptoRuntime::kyberNttMult_Type(),
8470 stubAddr, stubName, TypePtr::BOTTOM,
8471 result_start, ntta_start, nttb_start,
8472 zetas_start);
8473
8474 // return an int
8475 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
8476 set_result(retvalue);
8477
8478 return true;
8479 }
8480
8481 //------------------------------inline_kyberAddPoly_2
8482 bool LibraryCallKit::inline_kyberAddPoly_2() {
8483 address stubAddr;
8484 const char *stubName;
8485 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8486 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
8487
8488 stubAddr = StubRoutines::kyberAddPoly_2();
8489 stubName = "kyberAddPoly_2";
8490 if (!stubAddr) return false;
8491
8492 Node* result = argument(0);
8493 Node* a = argument(1);
8494 Node* b = argument(2);
8495
8496 result = must_be_not_null(result, true);
8497 a = must_be_not_null(a, true);
8498 b = must_be_not_null(b, true);
8499
8500 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8501 assert(result_start, "result is null");
8502 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8503 assert(a_start, "a is null");
8504 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8505 assert(b_start, "b is null");
8506 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
8507 OptoRuntime::kyberAddPoly_2_Type(),
8508 stubAddr, stubName, TypePtr::BOTTOM,
8509 result_start, a_start, b_start);
8510 // return an int
8511 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
8512 set_result(retvalue);
8513 return true;
8514 }
8515
8516 //------------------------------inline_kyberAddPoly_3
8517 bool LibraryCallKit::inline_kyberAddPoly_3() {
8518 address stubAddr;
8519 const char *stubName;
8520 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8521 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
8522
8523 stubAddr = StubRoutines::kyberAddPoly_3();
8524 stubName = "kyberAddPoly_3";
8525 if (!stubAddr) return false;
8526
8527 Node* result = argument(0);
8528 Node* a = argument(1);
8529 Node* b = argument(2);
8530 Node* c = argument(3);
8531
8532 result = must_be_not_null(result, true);
8533 a = must_be_not_null(a, true);
8534 b = must_be_not_null(b, true);
8535 c = must_be_not_null(c, true);
8536
8537 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8538 assert(result_start, "result is null");
8539 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8540 assert(a_start, "a is null");
8541 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8542 assert(b_start, "b is null");
8543 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
8544 assert(c_start, "c is null");
8545 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
8546 OptoRuntime::kyberAddPoly_3_Type(),
8547 stubAddr, stubName, TypePtr::BOTTOM,
8548 result_start, a_start, b_start, c_start);
8549 // return an int
8550 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
8551 set_result(retvalue);
8552 return true;
8553 }
8554
8555 //------------------------------inline_kyber12To16
8556 bool LibraryCallKit::inline_kyber12To16() {
8557 address stubAddr;
8558 const char *stubName;
8559 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8560 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
8561
8562 stubAddr = StubRoutines::kyber12To16();
8563 stubName = "kyber12To16";
8564 if (!stubAddr) return false;
8565
8566 Node* condensed = argument(0);
8567 Node* condensedOffs = argument(1);
8568 Node* parsed = argument(2);
8569 Node* parsedLength = argument(3);
8570
8571 condensed = must_be_not_null(condensed, true);
8572 parsed = must_be_not_null(parsed, true);
8573
8574 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
8575 assert(condensed_start, "condensed is null");
8576 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
8577 assert(parsed_start, "parsed is null");
8578 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
8579 OptoRuntime::kyber12To16_Type(),
8580 stubAddr, stubName, TypePtr::BOTTOM,
8581 condensed_start, condensedOffs, parsed_start, parsedLength);
8582 // return an int
8583 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
8584 set_result(retvalue);
8585 return true;
8586
8587 }
8588
8589 //------------------------------inline_kyberBarrettReduce
8590 bool LibraryCallKit::inline_kyberBarrettReduce() {
8591 address stubAddr;
8592 const char *stubName;
8593 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8594 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
8595
8596 stubAddr = StubRoutines::kyberBarrettReduce();
8597 stubName = "kyberBarrettReduce";
8598 if (!stubAddr) return false;
8599
8600 Node* coeffs = argument(0);
8601
8602 coeffs = must_be_not_null(coeffs, true);
8603
8604 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8605 assert(coeffs_start, "coeffs is null");
8606 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
8607 OptoRuntime::kyberBarrettReduce_Type(),
8608 stubAddr, stubName, TypePtr::BOTTOM,
8609 coeffs_start);
8610 // return an int
8611 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
8612 set_result(retvalue);
8613 return true;
8614 }
8615
8616 //------------------------------inline_dilithiumAlmostNtt
8617 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
8618 address stubAddr;
8619 const char *stubName;
8620 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8621 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
8622
8623 stubAddr = StubRoutines::dilithiumAlmostNtt();
8624 stubName = "dilithiumAlmostNtt";
8625 if (!stubAddr) return false;
8626
8627 Node* coeffs = argument(0);
8628 Node* ntt_zetas = argument(1);
8629
8630 coeffs = must_be_not_null(coeffs, true);
8631 ntt_zetas = must_be_not_null(ntt_zetas, true);
8632
8633 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8634 assert(coeffs_start, "coeffs is null");
8635 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
8636 assert(ntt_zetas_start, "ntt_zetas is null");
8637 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8638 OptoRuntime::dilithiumAlmostNtt_Type(),
8639 stubAddr, stubName, TypePtr::BOTTOM,
8640 coeffs_start, ntt_zetas_start);
8641 // return an int
8642 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
8643 set_result(retvalue);
8644 return true;
8645 }
8646
8647 //------------------------------inline_dilithiumAlmostInverseNtt
8648 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
8649 address stubAddr;
8650 const char *stubName;
8651 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8652 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
8653
8654 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
8655 stubName = "dilithiumAlmostInverseNtt";
8656 if (!stubAddr) return false;
8657
8658 Node* coeffs = argument(0);
8659 Node* zetas = argument(1);
8660
8661 coeffs = must_be_not_null(coeffs, true);
8662 zetas = must_be_not_null(zetas, true);
8663
8664 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8665 assert(coeffs_start, "coeffs is null");
8666 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
8667 assert(zetas_start, "inverseNtt_zetas is null");
8668 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8669 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
8670 stubAddr, stubName, TypePtr::BOTTOM,
8671 coeffs_start, zetas_start);
8672 // return an int
8673 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
8674 set_result(retvalue);
8675 return true;
8676 }
8677
8678 //------------------------------inline_dilithiumNttMult
8679 bool LibraryCallKit::inline_dilithiumNttMult() {
8680 address stubAddr;
8681 const char *stubName;
8682 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8683 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
8684
8685 stubAddr = StubRoutines::dilithiumNttMult();
8686 stubName = "dilithiumNttMult";
8687 if (!stubAddr) return false;
8688
8689 Node* result = argument(0);
8690 Node* ntta = argument(1);
8691 Node* nttb = argument(2);
8692 Node* zetas = argument(3);
8693
8694 result = must_be_not_null(result, true);
8695 ntta = must_be_not_null(ntta, true);
8696 nttb = must_be_not_null(nttb, true);
8697 zetas = must_be_not_null(zetas, true);
8698
8699 Node* result_start = array_element_address(result, intcon(0), T_INT);
8700 assert(result_start, "result is null");
8701 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
8702 assert(ntta_start, "ntta is null");
8703 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
8704 assert(nttb_start, "nttb is null");
8705 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8706 OptoRuntime::dilithiumNttMult_Type(),
8707 stubAddr, stubName, TypePtr::BOTTOM,
8708 result_start, ntta_start, nttb_start);
8709
8710 // return an int
8711 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8712 set_result(retvalue);
8713
8714 return true;
8715 }
8716
8717 //------------------------------inline_dilithiumMontMulByConstant
8718 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8719 address stubAddr;
8720 const char *stubName;
8721 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8722 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8723
8724 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8725 stubName = "dilithiumMontMulByConstant";
8726 if (!stubAddr) return false;
8727
8728 Node* coeffs = argument(0);
8729 Node* constant = argument(1);
8730
8731 coeffs = must_be_not_null(coeffs, true);
8732
8733 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8734 assert(coeffs_start, "coeffs is null");
8735 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8736 OptoRuntime::dilithiumMontMulByConstant_Type(),
8737 stubAddr, stubName, TypePtr::BOTTOM,
8738 coeffs_start, constant);
8739
8740 // return an int
8741 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8742 set_result(retvalue);
8743 return true;
8744 }
8745
8746
8747 //------------------------------inline_dilithiumDecomposePoly
8748 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8749 address stubAddr;
8750 const char *stubName;
8751 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8752 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8753
8754 stubAddr = StubRoutines::dilithiumDecomposePoly();
8755 stubName = "dilithiumDecomposePoly";
8756 if (!stubAddr) return false;
8757
8758 Node* input = argument(0);
8759 Node* lowPart = argument(1);
8760 Node* highPart = argument(2);
8761 Node* twoGamma2 = argument(3);
8762 Node* multiplier = argument(4);
8763
8764 input = must_be_not_null(input, true);
8765 lowPart = must_be_not_null(lowPart, true);
8766 highPart = must_be_not_null(highPart, true);
8767
8768 Node* input_start = array_element_address(input, intcon(0), T_INT);
8769 assert(input_start, "input is null");
8770 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8771 assert(lowPart_start, "lowPart is null");
8772 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8773 assert(highPart_start, "highPart is null");
8774
8775 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8776 OptoRuntime::dilithiumDecomposePoly_Type(),
8777 stubAddr, stubName, TypePtr::BOTTOM,
8778 input_start, lowPart_start, highPart_start,
8779 twoGamma2, multiplier);
8780
8781 // return an int
8782 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8783 set_result(retvalue);
8784 return true;
8785 }
8786
8787 bool LibraryCallKit::inline_base64_encodeBlock() {
8788 address stubAddr;
8789 const char *stubName;
8790 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8791 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8792 stubAddr = StubRoutines::base64_encodeBlock();
8793 stubName = "encodeBlock";
8794
8795 if (!stubAddr) return false;
8796 Node* base64obj = argument(0);
8797 Node* src = argument(1);
8798 Node* offset = argument(2);
8799 Node* len = argument(3);
8800 Node* dest = argument(4);
8801 Node* dp = argument(5);
8802 Node* isURL = argument(6);
8803
8804 src = must_be_not_null(src, true);
8805 dest = must_be_not_null(dest, true);
8806
8807 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8808 assert(src_start, "source array is null");
8809 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8810 assert(dest_start, "destination array is null");
8811
8812 Node* base64 = make_runtime_call(RC_LEAF,
8813 OptoRuntime::base64_encodeBlock_Type(),
8814 stubAddr, stubName, TypePtr::BOTTOM,
8815 src_start, offset, len, dest_start, dp, isURL);
8816 return true;
8817 }
8818
8819 bool LibraryCallKit::inline_base64_decodeBlock() {
8820 address stubAddr;
8821 const char *stubName;
8822 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8823 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8824 stubAddr = StubRoutines::base64_decodeBlock();
8825 stubName = "decodeBlock";
8826
8827 if (!stubAddr) return false;
8828 Node* base64obj = argument(0);
8829 Node* src = argument(1);
8830 Node* src_offset = argument(2);
8831 Node* len = argument(3);
8832 Node* dest = argument(4);
8833 Node* dest_offset = argument(5);
8834 Node* isURL = argument(6);
8835 Node* isMIME = argument(7);
8836
8837 src = must_be_not_null(src, true);
8838 dest = must_be_not_null(dest, true);
8839
8840 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8841 assert(src_start, "source array is null");
8842 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8843 assert(dest_start, "destination array is null");
8844
8845 Node* call = make_runtime_call(RC_LEAF,
8846 OptoRuntime::base64_decodeBlock_Type(),
8847 stubAddr, stubName, TypePtr::BOTTOM,
8848 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8849 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8850 set_result(result);
8851 return true;
8852 }
8853
8854 bool LibraryCallKit::inline_poly1305_processBlocks() {
8855 address stubAddr;
8856 const char *stubName;
8857 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8858 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8859 stubAddr = StubRoutines::poly1305_processBlocks();
8860 stubName = "poly1305_processBlocks";
8861
8862 if (!stubAddr) return false;
8863 null_check_receiver(); // null-check receiver
8864 if (stopped()) return true;
8865
8866 Node* input = argument(1);
8867 Node* input_offset = argument(2);
8868 Node* len = argument(3);
8869 Node* alimbs = argument(4);
8870 Node* rlimbs = argument(5);
8871
8872 input = must_be_not_null(input, true);
8873 alimbs = must_be_not_null(alimbs, true);
8874 rlimbs = must_be_not_null(rlimbs, true);
8875
8876 Node* input_start = array_element_address(input, input_offset, T_BYTE);
8877 assert(input_start, "input array is null");
8878 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
8879 assert(acc_start, "acc array is null");
8880 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
8881 assert(r_start, "r array is null");
8882
8883 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8884 OptoRuntime::poly1305_processBlocks_Type(),
8885 stubAddr, stubName, TypePtr::BOTTOM,
8886 input_start, len, acc_start, r_start);
8887 return true;
8888 }
8889
8890 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
8891 address stubAddr;
8892 const char *stubName;
8893 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8894 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
8895 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
8896 stubName = "intpoly_montgomeryMult_P256";
8897
8898 if (!stubAddr) return false;
8899 null_check_receiver(); // null-check receiver
8900 if (stopped()) return true;
8901
8902 Node* a = argument(1);
8903 Node* b = argument(2);
8904 Node* r = argument(3);
8905
8906 a = must_be_not_null(a, true);
8907 b = must_be_not_null(b, true);
8908 r = must_be_not_null(r, true);
8909
8910 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8911 assert(a_start, "a array is null");
8912 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8913 assert(b_start, "b array is null");
8914 Node* r_start = array_element_address(r, intcon(0), T_LONG);
8915 assert(r_start, "r array is null");
8916
8917 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8918 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
8919 stubAddr, stubName, TypePtr::BOTTOM,
8920 a_start, b_start, r_start);
8921 return true;
8922 }
8923
8924 bool LibraryCallKit::inline_intpoly_assign() {
8925 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8926 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
8927 const char *stubName = "intpoly_assign";
8928 address stubAddr = StubRoutines::intpoly_assign();
8929 if (!stubAddr) return false;
8930
8931 Node* set = argument(0);
8932 Node* a = argument(1);
8933 Node* b = argument(2);
8934 Node* arr_length = load_array_length(a);
8935
8936 a = must_be_not_null(a, true);
8937 b = must_be_not_null(b, true);
8938
8939 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8940 assert(a_start, "a array is null");
8941 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8942 assert(b_start, "b array is null");
8943
8944 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8945 OptoRuntime::intpoly_assign_Type(),
8946 stubAddr, stubName, TypePtr::BOTTOM,
8947 set, a_start, b_start, arr_length);
8948 return true;
8949 }
8950
8951 //------------------------------inline_digestBase_implCompress-----------------------
8952 //
8953 // Calculate MD5 for single-block byte[] array.
8954 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
8955 //
8956 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
8957 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
8958 //
8959 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
8960 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
8961 //
8962 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
8963 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
8964 //
8965 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
8966 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
8967 //
8968 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
8969 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
8970
8971 Node* digestBase_obj = argument(0);
8972 Node* src = argument(1); // type oop
8973 Node* ofs = argument(2); // type int
8974
8975 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8976 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8977 // failed array check
8978 return false;
8979 }
8980 // Figure out the size and type of the elements we will be copying.
8981 BasicType src_elem = src_type->elem()->array_element_basic_type();
8982 if (src_elem != T_BYTE) {
8983 return false;
8984 }
8985 // 'src_start' points to src array + offset
8986 src = must_be_not_null(src, true);
8987 Node* src_start = array_element_address(src, ofs, src_elem);
8988 Node* state = nullptr;
8989 Node* block_size = nullptr;
8990 address stubAddr;
8991 const char *stubName;
8992
8993 switch(id) {
8994 case vmIntrinsics::_md5_implCompress:
8995 assert(UseMD5Intrinsics, "need MD5 instruction support");
8996 state = get_state_from_digest_object(digestBase_obj, T_INT);
8997 stubAddr = StubRoutines::md5_implCompress();
8998 stubName = "md5_implCompress";
8999 break;
9000 case vmIntrinsics::_sha_implCompress:
9001 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
9002 state = get_state_from_digest_object(digestBase_obj, T_INT);
9003 stubAddr = StubRoutines::sha1_implCompress();
9004 stubName = "sha1_implCompress";
9005 break;
9006 case vmIntrinsics::_sha2_implCompress:
9007 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
9008 state = get_state_from_digest_object(digestBase_obj, T_INT);
9009 stubAddr = StubRoutines::sha256_implCompress();
9010 stubName = "sha256_implCompress";
9011 break;
9012 case vmIntrinsics::_sha5_implCompress:
9013 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
9014 state = get_state_from_digest_object(digestBase_obj, T_LONG);
9015 stubAddr = StubRoutines::sha512_implCompress();
9016 stubName = "sha512_implCompress";
9017 break;
9018 case vmIntrinsics::_sha3_implCompress:
9019 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
9020 state = get_state_from_digest_object(digestBase_obj, T_LONG);
9021 stubAddr = StubRoutines::sha3_implCompress();
9022 stubName = "sha3_implCompress";
9023 block_size = get_block_size_from_digest_object(digestBase_obj);
9024 if (block_size == nullptr) return false;
9025 break;
9026 default:
9027 fatal_unexpected_iid(id);
9028 return false;
9029 }
9030 if (state == nullptr) return false;
9031
9032 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
9033 if (stubAddr == nullptr) return false;
9034
9035 // Call the stub.
9036 Node* call;
9037 if (block_size == nullptr) {
9038 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
9039 stubAddr, stubName, TypePtr::BOTTOM,
9040 src_start, state);
9041 } else {
9042 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
9043 stubAddr, stubName, TypePtr::BOTTOM,
9044 src_start, state, block_size);
9045 }
9046
9047 return true;
9048 }
9049
9050 //------------------------------inline_double_keccak
9051 bool LibraryCallKit::inline_double_keccak() {
9052 address stubAddr;
9053 const char *stubName;
9054 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
9055 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
9056
9057 stubAddr = StubRoutines::double_keccak();
9058 stubName = "double_keccak";
9059 if (!stubAddr) return false;
9060
9061 Node* status0 = argument(0);
9062 Node* status1 = argument(1);
9063
9064 status0 = must_be_not_null(status0, true);
9065 status1 = must_be_not_null(status1, true);
9066
9067 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
9068 assert(status0_start, "status0 is null");
9069 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
9070 assert(status1_start, "status1 is null");
9071 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
9072 OptoRuntime::double_keccak_Type(),
9073 stubAddr, stubName, TypePtr::BOTTOM,
9074 status0_start, status1_start);
9075 // return an int
9076 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
9077 set_result(retvalue);
9078 return true;
9079 }
9080
9081
9082 //------------------------------inline_digestBase_implCompressMB-----------------------
9083 //
9084 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
9085 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
9086 //
9087 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
9088 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9089 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9090 assert((uint)predicate < 5, "sanity");
9091 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
9092
9093 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
9094 Node* src = argument(1); // byte[] array
9095 Node* ofs = argument(2); // type int
9096 Node* limit = argument(3); // type int
9097
9098 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
9099 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
9100 // failed array check
9101 return false;
9102 }
9103 // Figure out the size and type of the elements we will be copying.
9104 BasicType src_elem = src_type->elem()->array_element_basic_type();
9105 if (src_elem != T_BYTE) {
9106 return false;
9107 }
9108 // 'src_start' points to src array + offset
9109 src = must_be_not_null(src, false);
9110 Node* src_start = array_element_address(src, ofs, src_elem);
9111
9112 const char* klass_digestBase_name = nullptr;
9113 const char* stub_name = nullptr;
9114 address stub_addr = nullptr;
9115 BasicType elem_type = T_INT;
9116
9117 switch (predicate) {
9118 case 0:
9119 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
9120 klass_digestBase_name = "sun/security/provider/MD5";
9121 stub_name = "md5_implCompressMB";
9122 stub_addr = StubRoutines::md5_implCompressMB();
9123 }
9124 break;
9125 case 1:
9126 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
9127 klass_digestBase_name = "sun/security/provider/SHA";
9128 stub_name = "sha1_implCompressMB";
9129 stub_addr = StubRoutines::sha1_implCompressMB();
9130 }
9131 break;
9132 case 2:
9133 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
9134 klass_digestBase_name = "sun/security/provider/SHA2";
9135 stub_name = "sha256_implCompressMB";
9136 stub_addr = StubRoutines::sha256_implCompressMB();
9137 }
9138 break;
9139 case 3:
9140 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
9141 klass_digestBase_name = "sun/security/provider/SHA5";
9142 stub_name = "sha512_implCompressMB";
9143 stub_addr = StubRoutines::sha512_implCompressMB();
9144 elem_type = T_LONG;
9145 }
9146 break;
9147 case 4:
9148 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
9149 klass_digestBase_name = "sun/security/provider/SHA3";
9150 stub_name = "sha3_implCompressMB";
9151 stub_addr = StubRoutines::sha3_implCompressMB();
9152 elem_type = T_LONG;
9153 }
9154 break;
9155 default:
9156 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
9157 }
9158 if (klass_digestBase_name != nullptr) {
9159 assert(stub_addr != nullptr, "Stub is generated");
9160 if (stub_addr == nullptr) return false;
9161
9162 // get DigestBase klass to lookup for SHA klass
9163 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
9164 assert(tinst != nullptr, "digestBase_obj is not instance???");
9165 assert(tinst->is_loaded(), "DigestBase is not loaded");
9166
9167 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
9168 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
9169 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
9170 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
9171 }
9172 return false;
9173 }
9174
9175 //------------------------------inline_digestBase_implCompressMB-----------------------
9176 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
9177 BasicType elem_type, address stubAddr, const char *stubName,
9178 Node* src_start, Node* ofs, Node* limit) {
9179 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
9180 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
9181 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
9182 digest_obj = _gvn.transform(digest_obj);
9183
9184 Node* state = get_state_from_digest_object(digest_obj, elem_type);
9185 if (state == nullptr) return false;
9186
9187 Node* block_size = nullptr;
9188 if (strcmp("sha3_implCompressMB", stubName) == 0) {
9189 block_size = get_block_size_from_digest_object(digest_obj);
9190 if (block_size == nullptr) return false;
9191 }
9192
9193 // Call the stub.
9194 Node* call;
9195 if (block_size == nullptr) {
9196 call = make_runtime_call(RC_LEAF|RC_NO_FP,
9197 OptoRuntime::digestBase_implCompressMB_Type(false),
9198 stubAddr, stubName, TypePtr::BOTTOM,
9199 src_start, state, ofs, limit);
9200 } else {
9201 call = make_runtime_call(RC_LEAF|RC_NO_FP,
9202 OptoRuntime::digestBase_implCompressMB_Type(true),
9203 stubAddr, stubName, TypePtr::BOTTOM,
9204 src_start, state, block_size, ofs, limit);
9205 }
9206
9207 // return ofs (int)
9208 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
9209 set_result(result);
9210
9211 return true;
9212 }
9213
9214 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
9215 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
9216 assert(UseAES, "need AES instruction support");
9217 address stubAddr = nullptr;
9218 const char *stubName = nullptr;
9219 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
9220 stubName = "galoisCounterMode_AESCrypt";
9221
9222 if (stubAddr == nullptr) return false;
9223
9224 Node* in = argument(0);
9225 Node* inOfs = argument(1);
9226 Node* len = argument(2);
9227 Node* ct = argument(3);
9228 Node* ctOfs = argument(4);
9229 Node* out = argument(5);
9230 Node* outOfs = argument(6);
9231 Node* gctr_object = argument(7);
9232 Node* ghash_object = argument(8);
9233
9234 // (1) in, ct and out are arrays.
9235 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
9236 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
9237 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
9238 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
9239 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
9240 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
9241
9242 // checks are the responsibility of the caller
9243 Node* in_start = in;
9244 Node* ct_start = ct;
9245 Node* out_start = out;
9246 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
9247 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
9248 in_start = array_element_address(in, inOfs, T_BYTE);
9249 ct_start = array_element_address(ct, ctOfs, T_BYTE);
9250 out_start = array_element_address(out, outOfs, T_BYTE);
9251 }
9252
9253 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
9254 // (because of the predicated logic executed earlier).
9255 // so we cast it here safely.
9256 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
9257 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9258 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
9259 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
9260 Node* state = load_field_from_object(ghash_object, "state", "[J");
9261
9262 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
9263 return false;
9264 }
9265 // cast it to what we know it will be at runtime
9266 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
9267 assert(tinst != nullptr, "GCTR obj is null");
9268 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9269 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
9270 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
9271 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9272 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
9273 const TypeOopPtr* xtype = aklass->as_instance_type();
9274 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
9275 aescrypt_object = _gvn.transform(aescrypt_object);
9276 // we need to get the start of the aescrypt_object's expanded key array
9277 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
9278 if (k_start == nullptr) return false;
9279 // similarly, get the start address of the r vector
9280 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
9281 Node* state_start = array_element_address(state, intcon(0), T_LONG);
9282 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
9283
9284
9285 // Call the stub, passing params
9286 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
9287 OptoRuntime::galoisCounterMode_aescrypt_Type(),
9288 stubAddr, stubName, TypePtr::BOTTOM,
9289 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
9290
9291 // return cipher length (int)
9292 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
9293 set_result(retvalue);
9294
9295 return true;
9296 }
9297
9298 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
9299 // Return node representing slow path of predicate check.
9300 // the pseudo code we want to emulate with this predicate is:
9301 // for encryption:
9302 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
9303 // for decryption:
9304 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
9305 // note cipher==plain is more conservative than the original java code but that's OK
9306 //
9307
9308 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
9309 // The receiver was checked for null already.
9310 Node* objGCTR = argument(7);
9311 // Load embeddedCipher field of GCTR object.
9312 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9313 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
9314
9315 // get AESCrypt klass for instanceOf check
9316 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
9317 // will have same classloader as CipherBlockChaining object
9318 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
9319 assert(tinst != nullptr, "GCTR obj is null");
9320 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9321
9322 // we want to do an instanceof comparison against the AESCrypt class
9323 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
9324 if (!klass_AESCrypt->is_loaded()) {
9325 // if AESCrypt is not even loaded, we never take the intrinsic fast path
9326 Node* ctrl = control();
9327 set_control(top()); // no regular fast path
9328 return ctrl;
9329 }
9330
9331 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9332 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
9333 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9334 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9335 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9336
9337 return instof_false; // even if it is null
9338 }
9339
9340 //------------------------------get_state_from_digest_object-----------------------
9341 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
9342 const char* state_type;
9343 switch (elem_type) {
9344 case T_BYTE: state_type = "[B"; break;
9345 case T_INT: state_type = "[I"; break;
9346 case T_LONG: state_type = "[J"; break;
9347 default: ShouldNotReachHere();
9348 }
9349 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
9350 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
9351 if (digest_state == nullptr) return (Node *) nullptr;
9352
9353 // now have the array, need to get the start address of the state array
9354 Node* state = array_element_address(digest_state, intcon(0), elem_type);
9355 return state;
9356 }
9357
9358 //------------------------------get_block_size_from_sha3_object----------------------------------
9359 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
9360 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
9361 assert (block_size != nullptr, "sanity");
9362 return block_size;
9363 }
9364
9365 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
9366 // Return node representing slow path of predicate check.
9367 // the pseudo code we want to emulate with this predicate is:
9368 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
9369 //
9370 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
9371 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9372 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9373 assert((uint)predicate < 5, "sanity");
9374
9375 // The receiver was checked for null already.
9376 Node* digestBaseObj = argument(0);
9377
9378 // get DigestBase klass for instanceOf check
9379 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
9380 assert(tinst != nullptr, "digestBaseObj is null");
9381 assert(tinst->is_loaded(), "DigestBase is not loaded");
9382
9383 const char* klass_name = nullptr;
9384 switch (predicate) {
9385 case 0:
9386 if (UseMD5Intrinsics) {
9387 // we want to do an instanceof comparison against the MD5 class
9388 klass_name = "sun/security/provider/MD5";
9389 }
9390 break;
9391 case 1:
9392 if (UseSHA1Intrinsics) {
9393 // we want to do an instanceof comparison against the SHA class
9394 klass_name = "sun/security/provider/SHA";
9395 }
9396 break;
9397 case 2:
9398 if (UseSHA256Intrinsics) {
9399 // we want to do an instanceof comparison against the SHA2 class
9400 klass_name = "sun/security/provider/SHA2";
9401 }
9402 break;
9403 case 3:
9404 if (UseSHA512Intrinsics) {
9405 // we want to do an instanceof comparison against the SHA5 class
9406 klass_name = "sun/security/provider/SHA5";
9407 }
9408 break;
9409 case 4:
9410 if (UseSHA3Intrinsics) {
9411 // we want to do an instanceof comparison against the SHA3 class
9412 klass_name = "sun/security/provider/SHA3";
9413 }
9414 break;
9415 default:
9416 fatal("unknown SHA intrinsic predicate: %d", predicate);
9417 }
9418
9419 ciKlass* klass = nullptr;
9420 if (klass_name != nullptr) {
9421 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
9422 }
9423 if ((klass == nullptr) || !klass->is_loaded()) {
9424 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
9425 Node* ctrl = control();
9426 set_control(top()); // no intrinsic path
9427 return ctrl;
9428 }
9429 ciInstanceKlass* instklass = klass->as_instance_klass();
9430
9431 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
9432 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9433 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9434 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9435
9436 return instof_false; // even if it is null
9437 }
9438
9439 //-------------inline_fma-----------------------------------
9440 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
9441 Node *a = nullptr;
9442 Node *b = nullptr;
9443 Node *c = nullptr;
9444 Node* result = nullptr;
9445 switch (id) {
9446 case vmIntrinsics::_fmaD:
9447 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
9448 // no receiver since it is static method
9449 a = argument(0);
9450 b = argument(2);
9451 c = argument(4);
9452 result = _gvn.transform(new FmaDNode(a, b, c));
9453 break;
9454 case vmIntrinsics::_fmaF:
9455 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
9456 a = argument(0);
9457 b = argument(1);
9458 c = argument(2);
9459 result = _gvn.transform(new FmaFNode(a, b, c));
9460 break;
9461 default:
9462 fatal_unexpected_iid(id); break;
9463 }
9464 set_result(result);
9465 return true;
9466 }
9467
9468 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
9469 // argument(0) is receiver
9470 Node* codePoint = argument(1);
9471 Node* n = nullptr;
9472
9473 switch (id) {
9474 case vmIntrinsics::_isDigit :
9475 n = new DigitNode(control(), codePoint);
9476 break;
9477 case vmIntrinsics::_isLowerCase :
9478 n = new LowerCaseNode(control(), codePoint);
9479 break;
9480 case vmIntrinsics::_isUpperCase :
9481 n = new UpperCaseNode(control(), codePoint);
9482 break;
9483 case vmIntrinsics::_isWhitespace :
9484 n = new WhitespaceNode(control(), codePoint);
9485 break;
9486 default:
9487 fatal_unexpected_iid(id);
9488 }
9489
9490 set_result(_gvn.transform(n));
9491 return true;
9492 }
9493
9494 bool LibraryCallKit::inline_profileBoolean() {
9495 Node* counts = argument(1);
9496 const TypeAryPtr* ary = nullptr;
9497 ciArray* aobj = nullptr;
9498 if (counts->is_Con()
9499 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
9500 && (aobj = ary->const_oop()->as_array()) != nullptr
9501 && (aobj->length() == 2)) {
9502 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
9503 jint false_cnt = aobj->element_value(0).as_int();
9504 jint true_cnt = aobj->element_value(1).as_int();
9505
9506 if (C->log() != nullptr) {
9507 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
9508 false_cnt, true_cnt);
9509 }
9510
9511 if (false_cnt + true_cnt == 0) {
9512 // According to profile, never executed.
9513 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9514 Deoptimization::Action_reinterpret);
9515 return true;
9516 }
9517
9518 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
9519 // is a number of each value occurrences.
9520 Node* result = argument(0);
9521 if (false_cnt == 0 || true_cnt == 0) {
9522 // According to profile, one value has been never seen.
9523 int expected_val = (false_cnt == 0) ? 1 : 0;
9524
9525 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
9526 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
9527
9528 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
9529 Node* fast_path = _gvn.transform(new IfTrueNode(check));
9530 Node* slow_path = _gvn.transform(new IfFalseNode(check));
9531
9532 { // Slow path: uncommon trap for never seen value and then reexecute
9533 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
9534 // the value has been seen at least once.
9535 PreserveJVMState pjvms(this);
9536 PreserveReexecuteState preexecs(this);
9537 jvms()->set_should_reexecute(true);
9538
9539 set_control(slow_path);
9540 set_i_o(i_o());
9541
9542 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9543 Deoptimization::Action_reinterpret);
9544 }
9545 // The guard for never seen value enables sharpening of the result and
9546 // returning a constant. It allows to eliminate branches on the same value
9547 // later on.
9548 set_control(fast_path);
9549 result = intcon(expected_val);
9550 }
9551 // Stop profiling.
9552 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
9553 // By replacing method body with profile data (represented as ProfileBooleanNode
9554 // on IR level) we effectively disable profiling.
9555 // It enables full speed execution once optimized code is generated.
9556 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
9557 C->record_for_igvn(profile);
9558 set_result(profile);
9559 return true;
9560 } else {
9561 // Continue profiling.
9562 // Profile data isn't available at the moment. So, execute method's bytecode version.
9563 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
9564 // is compiled and counters aren't available since corresponding MethodHandle
9565 // isn't a compile-time constant.
9566 return false;
9567 }
9568 }
9569
9570 bool LibraryCallKit::inline_isCompileConstant() {
9571 Node* n = argument(0);
9572 set_result(n->is_Con() ? intcon(1) : intcon(0));
9573 return true;
9574 }
9575
9576 //------------------------------- inline_getObjectSize --------------------------------------
9577 //
9578 // Calculate the runtime size of the object/array.
9579 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
9580 //
9581 bool LibraryCallKit::inline_getObjectSize() {
9582 Node* obj = argument(3);
9583 Node* klass_node = load_object_klass(obj);
9584
9585 jint layout_con = Klass::_lh_neutral_value;
9586 Node* layout_val = get_layout_helper(klass_node, layout_con);
9587 int layout_is_con = (layout_val == nullptr);
9588
9589 if (layout_is_con) {
9590 // Layout helper is constant, can figure out things at compile time.
9591
9592 if (Klass::layout_helper_is_instance(layout_con)) {
9593 // Instance case: layout_con contains the size itself.
9594 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
9595 set_result(size);
9596 } else {
9597 // Array case: size is round(header + element_size*arraylength).
9598 // Since arraylength is different for every array instance, we have to
9599 // compute the whole thing at runtime.
9600
9601 Node* arr_length = load_array_length(obj);
9602
9603 int round_mask = MinObjAlignmentInBytes - 1;
9604 int hsize = Klass::layout_helper_header_size(layout_con);
9605 int eshift = Klass::layout_helper_log2_element_size(layout_con);
9606
9607 if ((round_mask & ~right_n_bits(eshift)) == 0) {
9608 round_mask = 0; // strength-reduce it if it goes away completely
9609 }
9610 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
9611 Node* header_size = intcon(hsize + round_mask);
9612
9613 Node* lengthx = ConvI2X(arr_length);
9614 Node* headerx = ConvI2X(header_size);
9615
9616 Node* abody = lengthx;
9617 if (eshift != 0) {
9618 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
9619 }
9620 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
9621 if (round_mask != 0) {
9622 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
9623 }
9624 size = ConvX2L(size);
9625 set_result(size);
9626 }
9627 } else {
9628 // Layout helper is not constant, need to test for array-ness at runtime.
9629
9630 enum { _instance_path = 1, _array_path, PATH_LIMIT };
9631 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
9632 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
9633 record_for_igvn(result_reg);
9634
9635 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
9636 if (array_ctl != nullptr) {
9637 // Array case: size is round(header + element_size*arraylength).
9638 // Since arraylength is different for every array instance, we have to
9639 // compute the whole thing at runtime.
9640
9641 PreserveJVMState pjvms(this);
9642 set_control(array_ctl);
9643 Node* arr_length = load_array_length(obj);
9644
9645 int round_mask = MinObjAlignmentInBytes - 1;
9646 Node* mask = intcon(round_mask);
9647
9648 Node* hss = intcon(Klass::_lh_header_size_shift);
9649 Node* hsm = intcon(Klass::_lh_header_size_mask);
9650 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
9651 header_size = _gvn.transform(new AndINode(header_size, hsm));
9652 header_size = _gvn.transform(new AddINode(header_size, mask));
9653
9654 // There is no need to mask or shift this value.
9655 // The semantics of LShiftINode include an implicit mask to 0x1F.
9656 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
9657 Node* elem_shift = layout_val;
9658
9659 Node* lengthx = ConvI2X(arr_length);
9660 Node* headerx = ConvI2X(header_size);
9661
9662 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
9663 Node* size = _gvn.transform(new AddXNode(headerx, abody));
9664 if (round_mask != 0) {
9665 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
9666 }
9667 size = ConvX2L(size);
9668
9669 result_reg->init_req(_array_path, control());
9670 result_val->init_req(_array_path, size);
9671 }
9672
9673 if (!stopped()) {
9674 // Instance case: the layout helper gives us instance size almost directly,
9675 // but we need to mask out the _lh_instance_slow_path_bit.
9676 Node* size = ConvI2X(layout_val);
9677 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
9678 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
9679 size = _gvn.transform(new AndXNode(size, mask));
9680 size = ConvX2L(size);
9681
9682 result_reg->init_req(_instance_path, control());
9683 result_val->init_req(_instance_path, size);
9684 }
9685
9686 set_result(result_reg, result_val);
9687 }
9688
9689 return true;
9690 }
9691
9692 //------------------------------- inline_blackhole --------------------------------------
9693 //
9694 // Make sure all arguments to this node are alive.
9695 // This matches methods that were requested to be blackholed through compile commands.
9696 //
9697 bool LibraryCallKit::inline_blackhole() {
9698 assert(callee()->is_static(), "Should have been checked before: only static methods here");
9699 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
9700 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
9701
9702 // Blackhole node pinches only the control, not memory. This allows
9703 // the blackhole to be pinned in the loop that computes blackholed
9704 // values, but have no other side effects, like breaking the optimizations
9705 // across the blackhole.
9706
9707 Node* bh = _gvn.transform(new BlackholeNode(control()));
9708 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
9709
9710 // Bind call arguments as blackhole arguments to keep them alive
9711 uint nargs = callee()->arg_size();
9712 for (uint i = 0; i < nargs; i++) {
9713 bh->add_req(argument(i));
9714 }
9715
9716 return true;
9717 }
9718
9719 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9720 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9721 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9722 return nullptr; // box klass is not Float16
9723 }
9724
9725 // Null check; get notnull casted pointer
9726 Node* null_ctl = top();
9727 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9728 // If not_null_box is dead, only null-path is taken
9729 if (stopped()) {
9730 set_control(null_ctl);
9731 return nullptr;
9732 }
9733 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9734 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9735 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9736 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9737 }
9738
9739 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9740 PreserveReexecuteState preexecs(this);
9741 jvms()->set_should_reexecute(true);
9742
9743 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9744 Node* klass_node = makecon(klass_type);
9745 Node* box = new_instance(klass_node);
9746
9747 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9748 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9749
9750 Node* field_store = _gvn.transform(access_store_at(box,
9751 value_field,
9752 value_adr_type,
9753 value,
9754 TypeInt::SHORT,
9755 T_SHORT,
9756 IN_HEAP));
9757 set_memory(field_store, value_adr_type);
9758 return box;
9759 }
9760
9761 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9762 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9763 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9764 return false;
9765 }
9766
9767 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9768 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9769 return false;
9770 }
9771
9772 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9773 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9774 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9775 ciSymbols::short_signature(),
9776 false);
9777 assert(field != nullptr, "");
9778
9779 // Transformed nodes
9780 Node* fld1 = nullptr;
9781 Node* fld2 = nullptr;
9782 Node* fld3 = nullptr;
9783 switch(num_args) {
9784 case 3:
9785 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9786 if (fld3 == nullptr) {
9787 return false;
9788 }
9789 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9790 // fall-through
9791 case 2:
9792 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9793 if (fld2 == nullptr) {
9794 return false;
9795 }
9796 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9797 // fall-through
9798 case 1:
9799 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9800 if (fld1 == nullptr) {
9801 return false;
9802 }
9803 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9804 break;
9805 default: fatal("Unsupported number of arguments %d", num_args);
9806 }
9807
9808 Node* result = nullptr;
9809 switch (id) {
9810 // Unary operations
9811 case vmIntrinsics::_sqrt_float16:
9812 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9813 break;
9814 // Ternary operations
9815 case vmIntrinsics::_fma_float16:
9816 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9817 break;
9818 default:
9819 fatal_unexpected_iid(id);
9820 break;
9821 }
9822 result = _gvn.transform(new ReinterpretHF2SNode(result));
9823 set_result(box_fp16_value(float16_box_type, field, result));
9824 return true;
9825 }
9826