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::_Reference_reachabilityFence: return inline_reference_reachabilityFence();
598 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
599 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
600 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
601
602 case vmIntrinsics::_Class_cast: return inline_Class_cast();
603
604 case vmIntrinsics::_aescrypt_encryptBlock:
605 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
606
607 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
608 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
609 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
610
611 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
612 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
613 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
614
615 case vmIntrinsics::_counterMode_AESCrypt:
616 return inline_counterMode_AESCrypt(intrinsic_id());
617
618 case vmIntrinsics::_galoisCounterMode_AESCrypt:
619 return inline_galoisCounterMode_AESCrypt();
620
621 case vmIntrinsics::_md5_implCompress:
622 case vmIntrinsics::_sha_implCompress:
623 case vmIntrinsics::_sha2_implCompress:
624 case vmIntrinsics::_sha5_implCompress:
625 case vmIntrinsics::_sha3_implCompress:
626 return inline_digestBase_implCompress(intrinsic_id());
627 case vmIntrinsics::_double_keccak:
628 return inline_double_keccak();
629
630 case vmIntrinsics::_digestBase_implCompressMB:
631 return inline_digestBase_implCompressMB(predicate);
632
633 case vmIntrinsics::_multiplyToLen:
634 return inline_multiplyToLen();
635
636 case vmIntrinsics::_squareToLen:
637 return inline_squareToLen();
638
639 case vmIntrinsics::_mulAdd:
640 return inline_mulAdd();
641
642 case vmIntrinsics::_montgomeryMultiply:
643 return inline_montgomeryMultiply();
644 case vmIntrinsics::_montgomerySquare:
645 return inline_montgomerySquare();
646
647 case vmIntrinsics::_bigIntegerRightShiftWorker:
648 return inline_bigIntegerShift(true);
649 case vmIntrinsics::_bigIntegerLeftShiftWorker:
650 return inline_bigIntegerShift(false);
651
652 case vmIntrinsics::_vectorizedMismatch:
653 return inline_vectorizedMismatch();
654
655 case vmIntrinsics::_ghash_processBlocks:
656 return inline_ghash_processBlocks();
657 case vmIntrinsics::_chacha20Block:
658 return inline_chacha20Block();
659 case vmIntrinsics::_kyberNtt:
660 return inline_kyberNtt();
661 case vmIntrinsics::_kyberInverseNtt:
662 return inline_kyberInverseNtt();
663 case vmIntrinsics::_kyberNttMult:
664 return inline_kyberNttMult();
665 case vmIntrinsics::_kyberAddPoly_2:
666 return inline_kyberAddPoly_2();
667 case vmIntrinsics::_kyberAddPoly_3:
668 return inline_kyberAddPoly_3();
669 case vmIntrinsics::_kyber12To16:
670 return inline_kyber12To16();
671 case vmIntrinsics::_kyberBarrettReduce:
672 return inline_kyberBarrettReduce();
673 case vmIntrinsics::_dilithiumAlmostNtt:
674 return inline_dilithiumAlmostNtt();
675 case vmIntrinsics::_dilithiumAlmostInverseNtt:
676 return inline_dilithiumAlmostInverseNtt();
677 case vmIntrinsics::_dilithiumNttMult:
678 return inline_dilithiumNttMult();
679 case vmIntrinsics::_dilithiumMontMulByConstant:
680 return inline_dilithiumMontMulByConstant();
681 case vmIntrinsics::_dilithiumDecomposePoly:
682 return inline_dilithiumDecomposePoly();
683 case vmIntrinsics::_base64_encodeBlock:
684 return inline_base64_encodeBlock();
685 case vmIntrinsics::_base64_decodeBlock:
686 return inline_base64_decodeBlock();
687 case vmIntrinsics::_poly1305_processBlocks:
688 return inline_poly1305_processBlocks();
689 case vmIntrinsics::_intpoly_montgomeryMult_P256:
690 return inline_intpoly_montgomeryMult_P256();
691 case vmIntrinsics::_intpoly_assign:
692 return inline_intpoly_assign();
693 case vmIntrinsics::_encodeISOArray:
694 case vmIntrinsics::_encodeByteISOArray:
695 return inline_encodeISOArray(false);
696 case vmIntrinsics::_encodeAsciiArray:
697 return inline_encodeISOArray(true);
698
699 case vmIntrinsics::_updateCRC32:
700 return inline_updateCRC32();
701 case vmIntrinsics::_updateBytesCRC32:
702 return inline_updateBytesCRC32();
703 case vmIntrinsics::_updateByteBufferCRC32:
704 return inline_updateByteBufferCRC32();
705
706 case vmIntrinsics::_updateBytesCRC32C:
707 return inline_updateBytesCRC32C();
708 case vmIntrinsics::_updateDirectByteBufferCRC32C:
709 return inline_updateDirectByteBufferCRC32C();
710
711 case vmIntrinsics::_updateBytesAdler32:
712 return inline_updateBytesAdler32();
713 case vmIntrinsics::_updateByteBufferAdler32:
714 return inline_updateByteBufferAdler32();
715
716 case vmIntrinsics::_profileBoolean:
717 return inline_profileBoolean();
718 case vmIntrinsics::_isCompileConstant:
719 return inline_isCompileConstant();
720
721 case vmIntrinsics::_countPositives:
722 return inline_countPositives();
723
724 case vmIntrinsics::_fmaD:
725 case vmIntrinsics::_fmaF:
726 return inline_fma(intrinsic_id());
727
728 case vmIntrinsics::_isDigit:
729 case vmIntrinsics::_isLowerCase:
730 case vmIntrinsics::_isUpperCase:
731 case vmIntrinsics::_isWhitespace:
732 return inline_character_compare(intrinsic_id());
733
734 case vmIntrinsics::_min:
735 case vmIntrinsics::_max:
736 case vmIntrinsics::_min_strict:
737 case vmIntrinsics::_max_strict:
738 case vmIntrinsics::_minL:
739 case vmIntrinsics::_maxL:
740 case vmIntrinsics::_minF:
741 case vmIntrinsics::_maxF:
742 case vmIntrinsics::_minD:
743 case vmIntrinsics::_maxD:
744 case vmIntrinsics::_minF_strict:
745 case vmIntrinsics::_maxF_strict:
746 case vmIntrinsics::_minD_strict:
747 case vmIntrinsics::_maxD_strict:
748 return inline_min_max(intrinsic_id());
749
750 case vmIntrinsics::_VectorUnaryOp:
751 return inline_vector_nary_operation(1);
752 case vmIntrinsics::_VectorBinaryOp:
753 return inline_vector_nary_operation(2);
754 case vmIntrinsics::_VectorUnaryLibOp:
755 return inline_vector_call(1);
756 case vmIntrinsics::_VectorBinaryLibOp:
757 return inline_vector_call(2);
758 case vmIntrinsics::_VectorTernaryOp:
759 return inline_vector_nary_operation(3);
760 case vmIntrinsics::_VectorFromBitsCoerced:
761 return inline_vector_frombits_coerced();
762 case vmIntrinsics::_VectorMaskOp:
763 return inline_vector_mask_operation();
764 case vmIntrinsics::_VectorLoadOp:
765 return inline_vector_mem_operation(/*is_store=*/false);
766 case vmIntrinsics::_VectorLoadMaskedOp:
767 return inline_vector_mem_masked_operation(/*is_store*/false);
768 case vmIntrinsics::_VectorStoreOp:
769 return inline_vector_mem_operation(/*is_store=*/true);
770 case vmIntrinsics::_VectorStoreMaskedOp:
771 return inline_vector_mem_masked_operation(/*is_store=*/true);
772 case vmIntrinsics::_VectorGatherOp:
773 return inline_vector_gather_scatter(/*is_scatter*/ false);
774 case vmIntrinsics::_VectorScatterOp:
775 return inline_vector_gather_scatter(/*is_scatter*/ true);
776 case vmIntrinsics::_VectorReductionCoerced:
777 return inline_vector_reduction();
778 case vmIntrinsics::_VectorTest:
779 return inline_vector_test();
780 case vmIntrinsics::_VectorBlend:
781 return inline_vector_blend();
782 case vmIntrinsics::_VectorRearrange:
783 return inline_vector_rearrange();
784 case vmIntrinsics::_VectorSelectFrom:
785 return inline_vector_select_from();
786 case vmIntrinsics::_VectorCompare:
787 return inline_vector_compare();
788 case vmIntrinsics::_VectorBroadcastInt:
789 return inline_vector_broadcast_int();
790 case vmIntrinsics::_VectorConvert:
791 return inline_vector_convert();
792 case vmIntrinsics::_VectorInsert:
793 return inline_vector_insert();
794 case vmIntrinsics::_VectorExtract:
795 return inline_vector_extract();
796 case vmIntrinsics::_VectorCompressExpand:
797 return inline_vector_compress_expand();
798 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
799 return inline_vector_select_from_two_vectors();
800 case vmIntrinsics::_IndexVector:
801 return inline_index_vector();
802 case vmIntrinsics::_IndexPartiallyInUpperRange:
803 return inline_index_partially_in_upper_range();
804
805 case vmIntrinsics::_getObjectSize:
806 return inline_getObjectSize();
807
808 case vmIntrinsics::_blackhole:
809 return inline_blackhole();
810
811 default:
812 // If you get here, it may be that someone has added a new intrinsic
813 // to the list in vmIntrinsics.hpp without implementing it here.
814 #ifndef PRODUCT
815 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
816 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
817 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
818 }
819 #endif
820 return false;
821 }
822 }
823
824 Node* LibraryCallKit::try_to_predicate(int predicate) {
825 if (!jvms()->has_method()) {
826 // Root JVMState has a null method.
827 assert(map()->memory()->Opcode() == Op_Parm, "");
828 // Insert the memory aliasing node
829 set_all_memory(reset_memory());
830 }
831 assert(merged_memory(), "");
832
833 switch (intrinsic_id()) {
834 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
835 return inline_cipherBlockChaining_AESCrypt_predicate(false);
836 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
837 return inline_cipherBlockChaining_AESCrypt_predicate(true);
838 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
839 return inline_electronicCodeBook_AESCrypt_predicate(false);
840 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
841 return inline_electronicCodeBook_AESCrypt_predicate(true);
842 case vmIntrinsics::_counterMode_AESCrypt:
843 return inline_counterMode_AESCrypt_predicate();
844 case vmIntrinsics::_digestBase_implCompressMB:
845 return inline_digestBase_implCompressMB_predicate(predicate);
846 case vmIntrinsics::_galoisCounterMode_AESCrypt:
847 return inline_galoisCounterMode_AESCrypt_predicate();
848
849 default:
850 // If you get here, it may be that someone has added a new intrinsic
851 // to the list in vmIntrinsics.hpp without implementing it here.
852 #ifndef PRODUCT
853 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
854 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
855 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
856 }
857 #endif
858 Node* slow_ctl = control();
859 set_control(top()); // No fast path intrinsic
860 return slow_ctl;
861 }
862 }
863
864 //------------------------------set_result-------------------------------
865 // Helper function for finishing intrinsics.
866 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
867 record_for_igvn(region);
868 set_control(_gvn.transform(region));
869 set_result( _gvn.transform(value));
870 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
871 }
872
873 RegionNode* LibraryCallKit::create_bailout() {
874 RegionNode* bailout = new RegionNode(1);
875 record_for_igvn(bailout);
876 return bailout;
877 }
878
879 bool LibraryCallKit::check_bailout(RegionNode* bailout) {
880 if (bailout->req() > 1) {
881 bailout = _gvn.transform(bailout)->as_Region();
882 Node* frame = _gvn.transform(new ParmNode(C->start(), TypeFunc::FramePtr));
883 Node* halt = _gvn.transform(new HaltNode(bailout, frame, "unexpected guard failure in intrinsic"));
884 C->root()->add_req(halt);
885 }
886 return stopped();
887 }
888
889 //------------------------------generate_guard---------------------------
890 // Helper function for generating guarded fast-slow graph structures.
891 // The given 'test', if true, guards a slow path. If the test fails
892 // then a fast path can be taken. (We generally hope it fails.)
893 // In all cases, GraphKit::control() is updated to the fast path.
894 // The returned value represents the control for the slow path.
895 // The return value is never 'top'; it is either a valid control
896 // or null if it is obvious that the slow path can never be taken.
897 // Also, if region and the slow control are not null, the slow edge
898 // is appended to the region.
899 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
900 if (stopped()) {
901 // Already short circuited.
902 return nullptr;
903 }
904
905 // Build an if node and its projections.
906 // If test is true we take the slow path, which we assume is uncommon.
907 if (_gvn.type(test) == TypeInt::ZERO) {
908 // The slow branch is never taken. No need to build this guard.
909 return nullptr;
910 }
911
912 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
913
914 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
915 if (if_slow == top()) {
916 // The slow branch is never taken. No need to build this guard.
917 return nullptr;
918 }
919
920 if (region != nullptr)
921 region->add_req(if_slow);
922
923 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
924 set_control(if_fast);
925
926 return if_slow;
927 }
928
929 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
930 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
931 }
932 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
933 return generate_guard(test, region, PROB_FAIR);
934 }
935
936 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
937 Node** pos_index, bool with_opaque) {
938 if (stopped())
939 return nullptr; // already stopped
940 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
941 return nullptr; // index is already adequately typed
942 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
943 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
944 if (with_opaque) {
945 bol_lt = _gvn.transform(new OpaqueConstantBoolNode(C, bol_lt, false));
946 }
947 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
948 if (is_neg != nullptr && pos_index != nullptr) {
949 // Emulate effect of Parse::adjust_map_after_if.
950 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
951 (*pos_index) = _gvn.transform(ccast);
952 }
953 return is_neg;
954 }
955
956 // Make sure that 'position' is a valid limit index, in [0..length].
957 // There are two equivalent plans for checking this:
958 // A. (offset + copyLength) unsigned<= arrayLength
959 // B. offset <= (arrayLength - copyLength)
960 // We require that all of the values above, except for the sum and
961 // difference, are already known to be non-negative.
962 // Plan A is robust in the face of overflow, if offset and copyLength
963 // are both hugely positive.
964 //
965 // Plan B is less direct and intuitive, but it does not overflow at
966 // all, since the difference of two non-negatives is always
967 // representable. Whenever Java methods must perform the equivalent
968 // check they generally use Plan B instead of Plan A.
969 // For the moment we use Plan A.
970 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
971 Node* subseq_length,
972 Node* array_length,
973 RegionNode* region,
974 bool with_opaque) {
975 if (stopped())
976 return nullptr; // already stopped
977 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
978 if (zero_offset && subseq_length->eqv_uncast(array_length))
979 return nullptr; // common case of whole-array copy
980 Node* last = subseq_length;
981 if (!zero_offset) // last += offset
982 last = _gvn.transform(new AddINode(last, offset));
983 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
984 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
985 if (with_opaque) {
986 bol_lt = _gvn.transform(new OpaqueConstantBoolNode(C, bol_lt, false));
987 }
988 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
989 return is_over;
990 }
991
992 // Emit range checks for the given String.value byte array
993 void LibraryCallKit::generate_string_range_check(Node* array,
994 Node* offset,
995 Node* count,
996 bool char_count,
997 RegionNode* region) {
998 if (stopped()) {
999 return; // already stopped
1000 }
1001 if (char_count) {
1002 // Convert char count to byte count
1003 count = _gvn.transform(new LShiftINode(count, intcon(1)));
1004 }
1005 // Offset and count must not be negative
1006 generate_negative_guard(offset, region, nullptr, true);
1007 generate_negative_guard(count, region, nullptr, true);
1008 // Offset + count must not exceed length of array
1009 generate_limit_guard(offset, count, load_array_length(array), region, true);
1010 }
1011
1012 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
1013 bool is_immutable) {
1014 ciKlass* thread_klass = env()->Thread_klass();
1015 const Type* thread_type
1016 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1017
1018 Node* thread = _gvn.transform(new ThreadLocalNode());
1019 Node* p = off_heap_plus_addr(thread, in_bytes(handle_offset));
1020 tls_output = thread;
1021
1022 Node* thread_obj_handle
1023 = (is_immutable
1024 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
1025 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
1026 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
1027 thread_obj_handle = _gvn.transform(thread_obj_handle);
1028
1029 DecoratorSet decorators = IN_NATIVE;
1030 if (is_immutable) {
1031 decorators |= C2_IMMUTABLE_MEMORY;
1032 }
1033 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
1034 }
1035
1036 //--------------------------generate_current_thread--------------------
1037 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1038 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
1039 /*is_immutable*/false);
1040 }
1041
1042 //--------------------------generate_virtual_thread--------------------
1043 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
1044 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
1045 !C->method()->changes_current_thread());
1046 }
1047
1048 //------------------------------make_string_method_node------------------------
1049 // Helper method for String intrinsic functions. This version is called with
1050 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1051 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1052 // containing the lengths of str1 and str2.
1053 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1054 Node* result = nullptr;
1055 switch (opcode) {
1056 case Op_StrIndexOf:
1057 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1058 str1_start, cnt1, str2_start, cnt2, ae);
1059 break;
1060 case Op_StrComp:
1061 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1062 str1_start, cnt1, str2_start, cnt2, ae);
1063 break;
1064 case Op_StrEquals:
1065 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1066 // Use the constant length if there is one because optimized match rule may exist.
1067 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1068 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1069 break;
1070 default:
1071 ShouldNotReachHere();
1072 return nullptr;
1073 }
1074
1075 // All these intrinsics have checks.
1076 C->set_has_split_ifs(true); // Has chance for split-if optimization
1077 clear_upper_avx();
1078
1079 return _gvn.transform(result);
1080 }
1081
1082 //------------------------------inline_string_compareTo------------------------
1083 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1084 Node* arg1 = argument(0);
1085 Node* arg2 = argument(1);
1086
1087 arg1 = must_be_not_null(arg1, true);
1088 arg2 = must_be_not_null(arg2, true);
1089
1090 // Get start addr and length of first argument
1091 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1092 Node* arg1_cnt = load_array_length(arg1);
1093
1094 // Get start addr and length of second argument
1095 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1096 Node* arg2_cnt = load_array_length(arg2);
1097
1098 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1099 set_result(result);
1100 return true;
1101 }
1102
1103 //------------------------------inline_string_equals------------------------
1104 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1105 Node* arg1 = argument(0);
1106 Node* arg2 = argument(1);
1107
1108 // paths (plus control) merge
1109 RegionNode* region = new RegionNode(3);
1110 Node* phi = new PhiNode(region, TypeInt::BOOL);
1111
1112 if (!stopped()) {
1113
1114 arg1 = must_be_not_null(arg1, true);
1115 arg2 = must_be_not_null(arg2, true);
1116
1117 // Get start addr and length of first argument
1118 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1119 Node* arg1_cnt = load_array_length(arg1);
1120
1121 // Get start addr and length of second argument
1122 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1123 Node* arg2_cnt = load_array_length(arg2);
1124
1125 // Check for arg1_cnt != arg2_cnt
1126 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1127 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1128 Node* if_ne = generate_slow_guard(bol, nullptr);
1129 if (if_ne != nullptr) {
1130 phi->init_req(2, intcon(0));
1131 region->init_req(2, if_ne);
1132 }
1133
1134 // Check for count == 0 is done by assembler code for StrEquals.
1135
1136 if (!stopped()) {
1137 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1138 phi->init_req(1, equals);
1139 region->init_req(1, control());
1140 }
1141 }
1142
1143 // post merge
1144 set_control(_gvn.transform(region));
1145 record_for_igvn(region);
1146
1147 set_result(_gvn.transform(phi));
1148 return true;
1149 }
1150
1151 //------------------------------inline_array_equals----------------------------
1152 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1153 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1154 Node* arg1 = argument(0);
1155 Node* arg2 = argument(1);
1156
1157 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1158 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1159 clear_upper_avx();
1160
1161 return true;
1162 }
1163
1164
1165 //------------------------------inline_countPositives------------------------------
1166 // int java.lang.StringCoding#countPositives0(byte[] ba, int off, int len)
1167 bool LibraryCallKit::inline_countPositives() {
1168 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1169 // no receiver since it is static method
1170 Node* ba = argument(0);
1171 Node* offset = argument(1);
1172 Node* len = argument(2);
1173
1174 ba = must_be_not_null(ba, true);
1175 RegionNode* bailout = create_bailout();
1176 generate_string_range_check(ba, offset, len, false, bailout);
1177 if (check_bailout(bailout)) {
1178 return true;
1179 }
1180
1181 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1182 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1183 set_result(_gvn.transform(result));
1184 clear_upper_avx();
1185 return true;
1186 }
1187
1188 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1189 Node* index = argument(0);
1190 Node* length = bt == T_INT ? argument(1) : argument(2);
1191 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1192 return false;
1193 }
1194
1195 // check that length is positive
1196 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1197 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1198
1199 {
1200 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1201 uncommon_trap(Deoptimization::Reason_intrinsic,
1202 Deoptimization::Action_make_not_entrant);
1203 }
1204
1205 if (stopped()) {
1206 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1207 return true;
1208 }
1209
1210 // length is now known positive, add a cast node to make this explicit
1211 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1212 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1213 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1214 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1215 casted_length = _gvn.transform(casted_length);
1216 replace_in_map(length, casted_length);
1217 length = casted_length;
1218
1219 // Use an unsigned comparison for the range check itself
1220 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1221 BoolTest::mask btest = BoolTest::lt;
1222 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1223 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1224 _gvn.set_type(rc, rc->Value(&_gvn));
1225 if (!rc_bool->is_Con()) {
1226 record_for_igvn(rc);
1227 }
1228 set_control(_gvn.transform(new IfTrueNode(rc)));
1229 {
1230 PreserveJVMState pjvms(this);
1231 set_control(_gvn.transform(new IfFalseNode(rc)));
1232 uncommon_trap(Deoptimization::Reason_range_check,
1233 Deoptimization::Action_make_not_entrant);
1234 }
1235
1236 if (stopped()) {
1237 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1238 return true;
1239 }
1240
1241 // index is now known to be >= 0 and < length, cast it
1242 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1243 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1244 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1245 result = _gvn.transform(result);
1246 set_result(result);
1247 replace_in_map(index, result);
1248 return true;
1249 }
1250
1251 //------------------------------inline_string_indexOf------------------------
1252 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1253 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1254 return false;
1255 }
1256 Node* src = argument(0);
1257 Node* tgt = argument(1);
1258
1259 // Make the merge point
1260 RegionNode* result_rgn = new RegionNode(4);
1261 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1262
1263 src = must_be_not_null(src, true);
1264 tgt = must_be_not_null(tgt, true);
1265
1266 // Get start addr and length of source string
1267 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1268 Node* src_count = load_array_length(src);
1269
1270 // Get start addr and length of substring
1271 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1272 Node* tgt_count = load_array_length(tgt);
1273
1274 Node* result = nullptr;
1275 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1276
1277 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1278 // Divide src size by 2 if String is UTF16 encoded
1279 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1280 }
1281 if (ae == StrIntrinsicNode::UU) {
1282 // Divide substring size by 2 if String is UTF16 encoded
1283 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1284 }
1285
1286 if (call_opt_stub) {
1287 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1288 StubRoutines::_string_indexof_array[ae],
1289 "stringIndexOf", TypePtr::BOTTOM, src_start,
1290 src_count, tgt_start, tgt_count);
1291 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1292 } else {
1293 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1294 result_rgn, result_phi, ae);
1295 }
1296 if (result != nullptr) {
1297 result_phi->init_req(3, result);
1298 result_rgn->init_req(3, control());
1299 }
1300 set_control(_gvn.transform(result_rgn));
1301 record_for_igvn(result_rgn);
1302 set_result(_gvn.transform(result_phi));
1303
1304 return true;
1305 }
1306
1307 //-----------------------------inline_string_indexOfI-----------------------
1308 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1309 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1310 return false;
1311 }
1312
1313 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1314 Node* src = argument(0); // byte[]
1315 Node* src_count = argument(1); // char count
1316 Node* tgt = argument(2); // byte[]
1317 Node* tgt_count = argument(3); // char count
1318 Node* from_index = argument(4); // char index
1319
1320 src = must_be_not_null(src, true);
1321 tgt = must_be_not_null(tgt, true);
1322
1323 // Multiply byte array index by 2 if String is UTF16 encoded
1324 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1325 src_count = _gvn.transform(new SubINode(src_count, from_index));
1326 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1327 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1328
1329 // Range checks
1330 RegionNode* bailout = create_bailout();
1331 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL, bailout);
1332 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU, bailout);
1333 if (check_bailout(bailout)) {
1334 return true;
1335 }
1336
1337 RegionNode* region = new RegionNode(5);
1338 Node* phi = new PhiNode(region, TypeInt::INT);
1339 Node* result = nullptr;
1340
1341 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1342
1343 if (call_opt_stub) {
1344 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1345 StubRoutines::_string_indexof_array[ae],
1346 "stringIndexOf", TypePtr::BOTTOM, src_start,
1347 src_count, tgt_start, tgt_count);
1348 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1349 } else {
1350 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1351 region, phi, ae);
1352 }
1353 if (result != nullptr) {
1354 // The result is index relative to from_index if substring was found, -1 otherwise.
1355 // Generate code which will fold into cmove.
1356 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1357 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1358
1359 Node* if_lt = generate_slow_guard(bol, nullptr);
1360 if (if_lt != nullptr) {
1361 // result == -1
1362 phi->init_req(3, result);
1363 region->init_req(3, if_lt);
1364 }
1365 if (!stopped()) {
1366 result = _gvn.transform(new AddINode(result, from_index));
1367 phi->init_req(4, result);
1368 region->init_req(4, control());
1369 }
1370 }
1371
1372 set_control(_gvn.transform(region));
1373 record_for_igvn(region);
1374 set_result(_gvn.transform(phi));
1375 clear_upper_avx();
1376
1377 return true;
1378 }
1379
1380 // Create StrIndexOfNode with fast path checks
1381 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1382 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1383 // Check for substr count > string count
1384 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1385 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1386 Node* if_gt = generate_slow_guard(bol, nullptr);
1387 if (if_gt != nullptr) {
1388 phi->init_req(1, intcon(-1));
1389 region->init_req(1, if_gt);
1390 }
1391 if (!stopped()) {
1392 // Check for substr count == 0
1393 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1394 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1395 Node* if_zero = generate_slow_guard(bol, nullptr);
1396 if (if_zero != nullptr) {
1397 phi->init_req(2, intcon(0));
1398 region->init_req(2, if_zero);
1399 }
1400 }
1401 if (!stopped()) {
1402 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1403 }
1404 return nullptr;
1405 }
1406
1407 //-----------------------------inline_string_indexOfChar-----------------------
1408 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1409 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1410 return false;
1411 }
1412 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1413 return false;
1414 }
1415 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1416 Node* src = argument(0); // byte[]
1417 Node* int_ch = argument(1);
1418 Node* from_index = argument(2);
1419 Node* max = argument(3);
1420
1421 src = must_be_not_null(src, true);
1422
1423 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1424 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1425 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1426
1427 // Range checks
1428 RegionNode* bailout = create_bailout();
1429 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U, bailout);
1430 if (check_bailout(bailout)) {
1431 return true;
1432 }
1433
1434 // Check for int_ch >= 0
1435 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1436 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1437 {
1438 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1439 uncommon_trap(Deoptimization::Reason_intrinsic,
1440 Deoptimization::Action_maybe_recompile);
1441 }
1442 if (stopped()) {
1443 return true;
1444 }
1445
1446 RegionNode* region = new RegionNode(3);
1447 Node* phi = new PhiNode(region, TypeInt::INT);
1448
1449 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1450 C->set_has_split_ifs(true); // Has chance for split-if optimization
1451 _gvn.transform(result);
1452
1453 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1454 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1455
1456 Node* if_lt = generate_slow_guard(bol, nullptr);
1457 if (if_lt != nullptr) {
1458 // result == -1
1459 phi->init_req(2, result);
1460 region->init_req(2, if_lt);
1461 }
1462 if (!stopped()) {
1463 result = _gvn.transform(new AddINode(result, from_index));
1464 phi->init_req(1, result);
1465 region->init_req(1, control());
1466 }
1467 set_control(_gvn.transform(region));
1468 record_for_igvn(region);
1469 set_result(_gvn.transform(phi));
1470 clear_upper_avx();
1471
1472 return true;
1473 }
1474 //---------------------------inline_string_copy---------------------
1475 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1476 // int StringUTF16.compress0(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1477 // int StringUTF16.compress0(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1478 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1479 // void StringLatin1.inflate0(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1480 // void StringLatin1.inflate0(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1481 bool LibraryCallKit::inline_string_copy(bool compress) {
1482 int nargs = 5; // 2 oops, 3 ints
1483 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1484
1485 Node* src = argument(0);
1486 Node* src_offset = argument(1);
1487 Node* dst = argument(2);
1488 Node* dst_offset = argument(3);
1489 Node* length = argument(4);
1490
1491 // Check for allocation before we add nodes that would confuse
1492 // tightly_coupled_allocation()
1493 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1494
1495 // Figure out the size and type of the elements we will be copying.
1496 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1497 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1498 if (src_type == nullptr || dst_type == nullptr) {
1499 return false;
1500 }
1501 BasicType src_elem = src_type->elem()->array_element_basic_type();
1502 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1503 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1504 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1505 "Unsupported array types for inline_string_copy");
1506
1507 src = must_be_not_null(src, true);
1508 dst = must_be_not_null(dst, true);
1509
1510 // Convert char[] offsets to byte[] offsets
1511 bool convert_src = (compress && src_elem == T_BYTE);
1512 bool convert_dst = (!compress && dst_elem == T_BYTE);
1513 if (convert_src) {
1514 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1515 } else if (convert_dst) {
1516 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1517 }
1518
1519 // Range checks
1520 RegionNode* bailout = create_bailout();
1521 generate_string_range_check(src, src_offset, length, convert_src, bailout);
1522 generate_string_range_check(dst, dst_offset, length, convert_dst, bailout);
1523 if (check_bailout(bailout)) {
1524 return true;
1525 }
1526
1527 Node* src_start = array_element_address(src, src_offset, src_elem);
1528 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1529 // 'src_start' points to src array + scaled offset
1530 // 'dst_start' points to dst array + scaled offset
1531 Node* count = nullptr;
1532 if (compress) {
1533 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1534 } else {
1535 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1536 }
1537
1538 if (alloc != nullptr) {
1539 if (alloc->maybe_set_complete(&_gvn)) {
1540 // "You break it, you buy it."
1541 InitializeNode* init = alloc->initialization();
1542 assert(init->is_complete(), "we just did this");
1543 init->set_complete_with_arraycopy();
1544 assert(dst->is_CheckCastPP(), "sanity");
1545 assert(dst->in(0)->in(0) == init, "dest pinned");
1546 }
1547 // Do not let stores that initialize this object be reordered with
1548 // a subsequent store that would make this object accessible by
1549 // other threads.
1550 // Record what AllocateNode this StoreStore protects so that
1551 // escape analysis can go from the MemBarStoreStoreNode to the
1552 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1553 // based on the escape status of the AllocateNode.
1554 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1555 }
1556 if (compress) {
1557 set_result(_gvn.transform(count));
1558 }
1559 clear_upper_avx();
1560
1561 return true;
1562 }
1563
1564 #ifdef _LP64
1565 #define XTOP ,top() /*additional argument*/
1566 #else //_LP64
1567 #define XTOP /*no additional argument*/
1568 #endif //_LP64
1569
1570 //------------------------inline_string_toBytesU--------------------------
1571 // public static byte[] StringUTF16.toBytes0(char[] value, int off, int len)
1572 bool LibraryCallKit::inline_string_toBytesU() {
1573 // Get the arguments.
1574 assert(callee()->signature()->size() == 3, "character array encoder requires 3 arguments");
1575 Node* value = argument(0);
1576 Node* offset = argument(1);
1577 Node* length = argument(2);
1578
1579 Node* newcopy = nullptr;
1580
1581 // Set the original stack and the reexecute bit for the interpreter to reexecute
1582 // the bytecode that invokes StringUTF16.toBytes0() if deoptimization happens.
1583 { PreserveReexecuteState preexecs(this);
1584 jvms()->set_should_reexecute(true);
1585
1586 value = must_be_not_null(value, true);
1587 RegionNode* bailout = create_bailout();
1588 generate_negative_guard(offset, bailout, nullptr, true);
1589 generate_negative_guard(length, bailout, nullptr, true);
1590 generate_limit_guard(offset, length, load_array_length(value), bailout, true);
1591 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1592 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout, true);
1593 if (check_bailout(bailout)) {
1594 return true;
1595 }
1596
1597 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1598 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1599 newcopy = new_array(klass_node, size, 0); // no arguments to push
1600 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1601 guarantee(alloc != nullptr, "created above");
1602
1603 // Calculate starting addresses.
1604 Node* src_start = array_element_address(value, offset, T_CHAR);
1605 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1606
1607 // Check if dst array address is aligned to HeapWordSize
1608 bool aligned = (arrayOopDesc::base_offset_in_bytes(T_BYTE) % HeapWordSize == 0);
1609 // If true, then check if src array address is aligned to HeapWordSize
1610 if (aligned) {
1611 const TypeInt* toffset = gvn().type(offset)->is_int();
1612 aligned = toffset->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) +
1613 toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1614 }
1615
1616 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1617 const char* copyfunc_name = "arraycopy";
1618 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1619 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1620 OptoRuntime::fast_arraycopy_Type(),
1621 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1622 src_start, dst_start, ConvI2X(length) XTOP);
1623 // Do not let reads from the cloned object float above the arraycopy.
1624 if (alloc->maybe_set_complete(&_gvn)) {
1625 // "You break it, you buy it."
1626 InitializeNode* init = alloc->initialization();
1627 assert(init->is_complete(), "we just did this");
1628 init->set_complete_with_arraycopy();
1629 assert(newcopy->is_CheckCastPP(), "sanity");
1630 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1631 }
1632 // Do not let stores that initialize this object be reordered with
1633 // a subsequent store that would make this object accessible by
1634 // other threads.
1635 // Record what AllocateNode this StoreStore protects so that
1636 // escape analysis can go from the MemBarStoreStoreNode to the
1637 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1638 // based on the escape status of the AllocateNode.
1639 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1640 } // original reexecute is set back here
1641
1642 C->set_has_split_ifs(true); // Has chance for split-if optimization
1643 if (!stopped()) {
1644 set_result(newcopy);
1645 }
1646 clear_upper_avx();
1647
1648 return true;
1649 }
1650
1651 //------------------------inline_string_getCharsU--------------------------
1652 // public void StringUTF16.getChars0(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1653 bool LibraryCallKit::inline_string_getCharsU() {
1654 assert(callee()->signature()->size() == 5, "StringUTF16.getChars0() has 5 arguments");
1655 // Get the arguments.
1656 Node* src = argument(0);
1657 Node* src_begin = argument(1);
1658 Node* src_end = argument(2); // exclusive offset (i < src_end)
1659 Node* dst = argument(3);
1660 Node* dst_begin = argument(4);
1661
1662 // Check for allocation before we add nodes that would confuse
1663 // tightly_coupled_allocation()
1664 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1665
1666 // Check if a null path was taken unconditionally.
1667 src = must_be_not_null(src, true);
1668 dst = must_be_not_null(dst, true);
1669 if (stopped()) {
1670 return true;
1671 }
1672
1673 // Get length and convert char[] offset to byte[] offset
1674 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1675 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1676
1677 // Range checks
1678 RegionNode* bailout = create_bailout();
1679 generate_string_range_check(src, src_begin, length, true, bailout);
1680 generate_string_range_check(dst, dst_begin, length, false, bailout);
1681 if (check_bailout(bailout)) {
1682 return true;
1683 }
1684
1685 // Calculate starting addresses.
1686 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1687 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1688
1689 // Check if array addresses are aligned to HeapWordSize
1690 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1691 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1692 bool aligned = tsrc->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_BYTE) + tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1693 tdst->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) + tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1694
1695 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1696 const char* copyfunc_name = "arraycopy";
1697 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1698 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1699 OptoRuntime::fast_arraycopy_Type(),
1700 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1701 src_start, dst_start, ConvI2X(length) XTOP);
1702 // Do not let reads from the cloned object float above the arraycopy.
1703 if (alloc != nullptr) {
1704 if (alloc->maybe_set_complete(&_gvn)) {
1705 // "You break it, you buy it."
1706 InitializeNode* init = alloc->initialization();
1707 assert(init->is_complete(), "we just did this");
1708 init->set_complete_with_arraycopy();
1709 assert(dst->is_CheckCastPP(), "sanity");
1710 assert(dst->in(0)->in(0) == init, "dest pinned");
1711 }
1712 // Do not let stores that initialize this object be reordered with
1713 // a subsequent store that would make this object accessible by
1714 // other threads.
1715 // Record what AllocateNode this StoreStore protects so that
1716 // escape analysis can go from the MemBarStoreStoreNode to the
1717 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1718 // based on the escape status of the AllocateNode.
1719 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1720 } else {
1721 insert_mem_bar(Op_MemBarCPUOrder);
1722 }
1723
1724 C->set_has_split_ifs(true); // Has chance for split-if optimization
1725 return true;
1726 }
1727
1728 //----------------------inline_string_char_access----------------------------
1729 // Store/Load char to/from byte[] array.
1730 // static void StringUTF16.putChar(byte[] val, int index, int c)
1731 // static char StringUTF16.getChar(byte[] val, int index)
1732 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1733 Node* ch;
1734 if (is_store) {
1735 assert(callee()->signature()->size() == 3, "StringUTF16.putChar() has 3 arguments");
1736 ch = argument(2);
1737 } else {
1738 assert(callee()->signature()->size() == 2, "StringUTF16.getChar() has 2 arguments");
1739 ch = nullptr;
1740 }
1741 Node* value = argument(0);
1742 Node* index = argument(1);
1743
1744 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1745 // correctly requires matched array shapes.
1746 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1747 "sanity: byte[] and char[] bases agree");
1748 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1749 "sanity: byte[] and char[] scales agree");
1750
1751 // Bail when getChar over constants is requested: constant folding would
1752 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1753 // Java method would constant fold nicely instead.
1754 if (!is_store && value->is_Con() && index->is_Con()) {
1755 return false;
1756 }
1757
1758 // Save state and restore on bailout
1759 SavedState old_state(this);
1760
1761 value = must_be_not_null(value, true);
1762
1763 Node* adr = array_element_address(value, index, T_CHAR);
1764 if (adr->is_top()) {
1765 return false;
1766 }
1767 old_state.discard();
1768 if (is_store) {
1769 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1770 } else {
1771 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);
1772 set_result(ch);
1773 }
1774 return true;
1775 }
1776
1777
1778 //------------------------------inline_math-----------------------------------
1779 // public static double Math.abs(double)
1780 // public static double Math.sqrt(double)
1781 // public static double Math.log(double)
1782 // public static double Math.log10(double)
1783 // public static double Math.round(double)
1784 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1785 Node* arg = argument(0);
1786 Node* n = nullptr;
1787 switch (id) {
1788 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1789 case vmIntrinsics::_dsqrt:
1790 case vmIntrinsics::_dsqrt_strict:
1791 n = new SqrtDNode(C, control(), arg); break;
1792 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1793 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1794 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1795 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1796 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1797 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1798 default: fatal_unexpected_iid(id); break;
1799 }
1800 set_result(_gvn.transform(n));
1801 return true;
1802 }
1803
1804 //------------------------------inline_math-----------------------------------
1805 // public static float Math.abs(float)
1806 // public static int Math.abs(int)
1807 // public static long Math.abs(long)
1808 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1809 Node* arg = argument(0);
1810 Node* n = nullptr;
1811 switch (id) {
1812 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1813 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1814 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1815 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1816 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1817 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1818 default: fatal_unexpected_iid(id); break;
1819 }
1820 set_result(_gvn.transform(n));
1821 return true;
1822 }
1823
1824 //------------------------------runtime_math-----------------------------
1825 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1826 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1827 "must be (DD)D or (D)D type");
1828
1829 // Inputs
1830 Node* a = argument(0);
1831 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1832
1833 const TypePtr* no_memory_effects = nullptr;
1834 Node* trig = make_runtime_call(RC_LEAF | RC_PURE, call_type, funcAddr, funcName,
1835 no_memory_effects,
1836 a, top(), b, b ? top() : nullptr);
1837 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1838 #ifdef ASSERT
1839 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1840 assert(value_top == top(), "second value must be top");
1841 #endif
1842
1843 set_result(value);
1844 return true;
1845 }
1846
1847 //------------------------------inline_math_pow-----------------------------
1848 bool LibraryCallKit::inline_math_pow() {
1849 Node* base = argument(0);
1850 Node* exp = argument(2);
1851
1852 CallNode* pow = new PowDNode(C, base, exp);
1853 set_predefined_input_for_runtime_call(pow);
1854 pow = _gvn.transform(pow)->as_CallLeafPure();
1855 set_predefined_output_for_runtime_call(pow);
1856 Node* result = _gvn.transform(new ProjNode(pow, TypeFunc::Parms + 0));
1857 record_for_igvn(pow);
1858 set_result(result);
1859 return true;
1860 }
1861
1862 //------------------------------inline_math_native-----------------------------
1863 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1864 switch (id) {
1865 case vmIntrinsics::_dsin:
1866 return StubRoutines::dsin() != nullptr ?
1867 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1868 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1869 case vmIntrinsics::_dcos:
1870 return StubRoutines::dcos() != nullptr ?
1871 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1872 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1873 case vmIntrinsics::_dtan:
1874 return StubRoutines::dtan() != nullptr ?
1875 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1876 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1877 case vmIntrinsics::_dsinh:
1878 return StubRoutines::dsinh() != nullptr ?
1879 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsinh(), "dsinh") : false;
1880 case vmIntrinsics::_dtanh:
1881 return StubRoutines::dtanh() != nullptr ?
1882 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1883 case vmIntrinsics::_dcbrt:
1884 return StubRoutines::dcbrt() != nullptr ?
1885 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcbrt(), "dcbrt") : false;
1886 case vmIntrinsics::_dexp:
1887 return StubRoutines::dexp() != nullptr ?
1888 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1889 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1890 case vmIntrinsics::_dlog:
1891 return StubRoutines::dlog() != nullptr ?
1892 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1893 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1894 case vmIntrinsics::_dlog10:
1895 return StubRoutines::dlog10() != nullptr ?
1896 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1897 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1898
1899 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1900 case vmIntrinsics::_ceil:
1901 case vmIntrinsics::_floor:
1902 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1903
1904 case vmIntrinsics::_dsqrt:
1905 case vmIntrinsics::_dsqrt_strict:
1906 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1907 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1908 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1909 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1910 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1911
1912 case vmIntrinsics::_dpow: return inline_math_pow();
1913 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1914 case vmIntrinsics::_fcopySign: return inline_math(id);
1915 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1916 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1917 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1918
1919 // These intrinsics are not yet correctly implemented
1920 case vmIntrinsics::_datan2:
1921 return false;
1922
1923 default:
1924 fatal_unexpected_iid(id);
1925 return false;
1926 }
1927 }
1928
1929 //----------------------------inline_notify-----------------------------------*
1930 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1931 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1932 address func;
1933 if (id == vmIntrinsics::_notify) {
1934 func = OptoRuntime::monitor_notify_Java();
1935 } else {
1936 func = OptoRuntime::monitor_notifyAll_Java();
1937 }
1938 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1939 make_slow_call_ex(call, env()->Throwable_klass(), false);
1940 return true;
1941 }
1942
1943
1944 //----------------------------inline_min_max-----------------------------------
1945 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1946 Node* a = nullptr;
1947 Node* b = nullptr;
1948 Node* n = nullptr;
1949 switch (id) {
1950 case vmIntrinsics::_min:
1951 case vmIntrinsics::_max:
1952 case vmIntrinsics::_minF:
1953 case vmIntrinsics::_maxF:
1954 case vmIntrinsics::_minF_strict:
1955 case vmIntrinsics::_maxF_strict:
1956 case vmIntrinsics::_min_strict:
1957 case vmIntrinsics::_max_strict:
1958 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
1959 a = argument(0);
1960 b = argument(1);
1961 break;
1962 case vmIntrinsics::_minD:
1963 case vmIntrinsics::_maxD:
1964 case vmIntrinsics::_minD_strict:
1965 case vmIntrinsics::_maxD_strict:
1966 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
1967 a = argument(0);
1968 b = argument(2);
1969 break;
1970 case vmIntrinsics::_minL:
1971 case vmIntrinsics::_maxL:
1972 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
1973 a = argument(0);
1974 b = argument(2);
1975 break;
1976 default:
1977 fatal_unexpected_iid(id);
1978 break;
1979 }
1980
1981 switch (id) {
1982 case vmIntrinsics::_min:
1983 case vmIntrinsics::_min_strict:
1984 n = new MinINode(a, b);
1985 break;
1986 case vmIntrinsics::_max:
1987 case vmIntrinsics::_max_strict:
1988 n = new MaxINode(a, b);
1989 break;
1990 case vmIntrinsics::_minF:
1991 case vmIntrinsics::_minF_strict:
1992 n = new MinFNode(a, b);
1993 break;
1994 case vmIntrinsics::_maxF:
1995 case vmIntrinsics::_maxF_strict:
1996 n = new MaxFNode(a, b);
1997 break;
1998 case vmIntrinsics::_minD:
1999 case vmIntrinsics::_minD_strict:
2000 n = new MinDNode(a, b);
2001 break;
2002 case vmIntrinsics::_maxD:
2003 case vmIntrinsics::_maxD_strict:
2004 n = new MaxDNode(a, b);
2005 break;
2006 case vmIntrinsics::_minL:
2007 n = new MinLNode(_gvn.C, a, b);
2008 break;
2009 case vmIntrinsics::_maxL:
2010 n = new MaxLNode(_gvn.C, a, b);
2011 break;
2012 default:
2013 fatal_unexpected_iid(id);
2014 break;
2015 }
2016
2017 set_result(_gvn.transform(n));
2018 return true;
2019 }
2020
2021 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
2022 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
2023 env()->ArithmeticException_instance())) {
2024 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
2025 // so let's bail out intrinsic rather than risking deopting again.
2026 return false;
2027 }
2028
2029 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2030 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2031 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2032 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2033
2034 {
2035 PreserveJVMState pjvms(this);
2036 PreserveReexecuteState preexecs(this);
2037 jvms()->set_should_reexecute(true);
2038
2039 set_control(slow_path);
2040 set_i_o(i_o());
2041
2042 builtin_throw(Deoptimization::Reason_intrinsic,
2043 env()->ArithmeticException_instance(),
2044 /*allow_too_many_traps*/ false);
2045 }
2046
2047 set_control(fast_path);
2048 set_result(math);
2049 return true;
2050 }
2051
2052 template <typename OverflowOp>
2053 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2054 typedef typename OverflowOp::MathOp MathOp;
2055
2056 MathOp* mathOp = new MathOp(arg1, arg2);
2057 Node* operation = _gvn.transform( mathOp );
2058 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2059 return inline_math_mathExact(operation, ofcheck);
2060 }
2061
2062 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2063 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2064 }
2065
2066 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2067 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2068 }
2069
2070 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2071 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2072 }
2073
2074 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2075 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2076 }
2077
2078 bool LibraryCallKit::inline_math_negateExactI() {
2079 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2080 }
2081
2082 bool LibraryCallKit::inline_math_negateExactL() {
2083 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2084 }
2085
2086 bool LibraryCallKit::inline_math_multiplyExactI() {
2087 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2088 }
2089
2090 bool LibraryCallKit::inline_math_multiplyExactL() {
2091 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2092 }
2093
2094 bool LibraryCallKit::inline_math_multiplyHigh() {
2095 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2096 return true;
2097 }
2098
2099 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2100 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2101 return true;
2102 }
2103
2104 inline int
2105 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2106 const TypePtr* base_type = TypePtr::NULL_PTR;
2107 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2108 if (base_type == nullptr) {
2109 // Unknown type.
2110 return Type::AnyPtr;
2111 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2112 // Since this is a null+long form, we have to switch to a rawptr.
2113 base = _gvn.transform(new CastX2PNode(offset));
2114 offset = MakeConX(0);
2115 return Type::RawPtr;
2116 } else if (base_type->base() == Type::RawPtr) {
2117 return Type::RawPtr;
2118 } else if (base_type->isa_oopptr()) {
2119 // Base is never null => always a heap address.
2120 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2121 return Type::OopPtr;
2122 }
2123 // Offset is small => always a heap address.
2124 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2125 if (offset_type != nullptr &&
2126 base_type->offset() == 0 && // (should always be?)
2127 offset_type->_lo >= 0 &&
2128 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2129 return Type::OopPtr;
2130 } else if (type == T_OBJECT) {
2131 // off heap access to an oop doesn't make any sense. Has to be on
2132 // heap.
2133 return Type::OopPtr;
2134 }
2135 // Otherwise, it might either be oop+off or null+addr.
2136 return Type::AnyPtr;
2137 } else {
2138 // No information:
2139 return Type::AnyPtr;
2140 }
2141 }
2142
2143 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2144 Node* uncasted_base = base;
2145 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2146 if (kind == Type::RawPtr) {
2147 return off_heap_plus_addr(uncasted_base, offset);
2148 } else if (kind == Type::AnyPtr) {
2149 assert(base == uncasted_base, "unexpected base change");
2150 if (can_cast) {
2151 if (!_gvn.type(base)->speculative_maybe_null() &&
2152 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2153 // According to profiling, this access is always on
2154 // heap. Casting the base to not null and thus avoiding membars
2155 // around the access should allow better optimizations
2156 Node* null_ctl = top();
2157 base = null_check_oop(base, &null_ctl, true, true, true);
2158 assert(null_ctl->is_top(), "no null control here");
2159 return basic_plus_adr(base, offset);
2160 } else if (_gvn.type(base)->speculative_always_null() &&
2161 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2162 // According to profiling, this access is always off
2163 // heap.
2164 base = null_assert(base);
2165 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2166 offset = MakeConX(0);
2167 return off_heap_plus_addr(raw_base, offset);
2168 }
2169 }
2170 // We don't know if it's an on heap or off heap access. Fall back
2171 // to raw memory access.
2172 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2173 return off_heap_plus_addr(raw, offset);
2174 } else {
2175 assert(base == uncasted_base, "unexpected base change");
2176 // We know it's an on heap access so base can't be null
2177 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2178 base = must_be_not_null(base, true);
2179 }
2180 return basic_plus_adr(base, offset);
2181 }
2182 }
2183
2184 //--------------------------inline_number_methods-----------------------------
2185 // inline int Integer.numberOfLeadingZeros(int)
2186 // inline int Long.numberOfLeadingZeros(long)
2187 //
2188 // inline int Integer.numberOfTrailingZeros(int)
2189 // inline int Long.numberOfTrailingZeros(long)
2190 //
2191 // inline int Integer.bitCount(int)
2192 // inline int Long.bitCount(long)
2193 //
2194 // inline char Character.reverseBytes(char)
2195 // inline short Short.reverseBytes(short)
2196 // inline int Integer.reverseBytes(int)
2197 // inline long Long.reverseBytes(long)
2198 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2199 Node* arg = argument(0);
2200 Node* n = nullptr;
2201 switch (id) {
2202 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2203 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2204 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2205 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2206 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2207 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2208 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2209 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2210 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2211 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2212 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2213 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2214 default: fatal_unexpected_iid(id); break;
2215 }
2216 set_result(_gvn.transform(n));
2217 return true;
2218 }
2219
2220 //--------------------------inline_bitshuffle_methods-----------------------------
2221 // inline int Integer.compress(int, int)
2222 // inline int Integer.expand(int, int)
2223 // inline long Long.compress(long, long)
2224 // inline long Long.expand(long, long)
2225 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2226 Node* n = nullptr;
2227 switch (id) {
2228 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2229 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2230 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2231 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2232 default: fatal_unexpected_iid(id); break;
2233 }
2234 set_result(_gvn.transform(n));
2235 return true;
2236 }
2237
2238 //--------------------------inline_number_methods-----------------------------
2239 // inline int Integer.compareUnsigned(int, int)
2240 // inline int Long.compareUnsigned(long, long)
2241 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2242 Node* arg1 = argument(0);
2243 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2244 Node* n = nullptr;
2245 switch (id) {
2246 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2247 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2248 default: fatal_unexpected_iid(id); break;
2249 }
2250 set_result(_gvn.transform(n));
2251 return true;
2252 }
2253
2254 //--------------------------inline_unsigned_divmod_methods-----------------------------
2255 // inline int Integer.divideUnsigned(int, int)
2256 // inline int Integer.remainderUnsigned(int, int)
2257 // inline long Long.divideUnsigned(long, long)
2258 // inline long Long.remainderUnsigned(long, long)
2259 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2260 Node* n = nullptr;
2261 switch (id) {
2262 case vmIntrinsics::_divideUnsigned_i: {
2263 zero_check_int(argument(1));
2264 // Compile-time detect of null-exception
2265 if (stopped()) {
2266 return true; // keep the graph constructed so far
2267 }
2268 n = new UDivINode(control(), argument(0), argument(1));
2269 break;
2270 }
2271 case vmIntrinsics::_divideUnsigned_l: {
2272 zero_check_long(argument(2));
2273 // Compile-time detect of null-exception
2274 if (stopped()) {
2275 return true; // keep the graph constructed so far
2276 }
2277 n = new UDivLNode(control(), argument(0), argument(2));
2278 break;
2279 }
2280 case vmIntrinsics::_remainderUnsigned_i: {
2281 zero_check_int(argument(1));
2282 // Compile-time detect of null-exception
2283 if (stopped()) {
2284 return true; // keep the graph constructed so far
2285 }
2286 n = new UModINode(control(), argument(0), argument(1));
2287 break;
2288 }
2289 case vmIntrinsics::_remainderUnsigned_l: {
2290 zero_check_long(argument(2));
2291 // Compile-time detect of null-exception
2292 if (stopped()) {
2293 return true; // keep the graph constructed so far
2294 }
2295 n = new UModLNode(control(), argument(0), argument(2));
2296 break;
2297 }
2298 default: fatal_unexpected_iid(id); break;
2299 }
2300 set_result(_gvn.transform(n));
2301 return true;
2302 }
2303
2304 //----------------------------inline_unsafe_access----------------------------
2305
2306 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2307 // Attempt to infer a sharper value type from the offset and base type.
2308 ciKlass* sharpened_klass = nullptr;
2309 bool null_free = false;
2310
2311 // See if it is an instance field, with an object type.
2312 if (alias_type->field() != nullptr) {
2313 if (alias_type->field()->type()->is_klass()) {
2314 sharpened_klass = alias_type->field()->type()->as_klass();
2315 null_free = alias_type->field()->is_null_free();
2316 }
2317 }
2318
2319 const TypeOopPtr* result = nullptr;
2320 // See if it is a narrow oop array.
2321 if (adr_type->isa_aryptr()) {
2322 if (adr_type->offset() >= refArrayOopDesc::base_offset_in_bytes()) {
2323 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2324 null_free = adr_type->is_aryptr()->is_null_free();
2325 if (elem_type != nullptr && elem_type->is_loaded()) {
2326 // Sharpen the value type.
2327 result = elem_type;
2328 }
2329 }
2330 }
2331
2332 // The sharpened class might be unloaded if there is no class loader
2333 // contraint in place.
2334 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2335 // Sharpen the value type.
2336 result = TypeOopPtr::make_from_klass(sharpened_klass);
2337 if (null_free) {
2338 result = result->join_speculative(TypePtr::NOTNULL)->is_oopptr();
2339 }
2340 }
2341 if (result != nullptr) {
2342 #ifndef PRODUCT
2343 if (C->print_intrinsics() || C->print_inlining()) {
2344 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2345 tty->print(" sharpened value: "); result->dump(); tty->cr();
2346 }
2347 #endif
2348 }
2349 return result;
2350 }
2351
2352 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2353 switch (kind) {
2354 case Relaxed:
2355 return MO_UNORDERED;
2356 case Opaque:
2357 return MO_RELAXED;
2358 case Acquire:
2359 return MO_ACQUIRE;
2360 case Release:
2361 return MO_RELEASE;
2362 case Volatile:
2363 return MO_SEQ_CST;
2364 default:
2365 ShouldNotReachHere();
2366 return 0;
2367 }
2368 }
2369
2370 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2371 if (callee()->is_static()) return false; // caller must have the capability!
2372 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2373 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2374 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2375 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2376
2377 if (is_reference_type(type)) {
2378 decorators |= ON_UNKNOWN_OOP_REF;
2379 }
2380
2381 if (unaligned) {
2382 decorators |= C2_UNALIGNED;
2383 }
2384
2385 #ifndef PRODUCT
2386 {
2387 ResourceMark rm;
2388 // Check the signatures.
2389 ciSignature* sig = callee()->signature();
2390 #ifdef ASSERT
2391 if (!is_store) {
2392 // Object getReference(Object base, int/long offset), etc.
2393 BasicType rtype = sig->return_type()->basic_type();
2394 assert(rtype == type, "getter must return the expected value");
2395 assert(sig->count() == 2, "oop getter has 2 arguments");
2396 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2397 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2398 } else {
2399 // void putReference(Object base, int/long offset, Object x), etc.
2400 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2401 assert(sig->count() == 3, "oop putter has 3 arguments");
2402 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2403 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2404 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2405 assert(vtype == type, "putter must accept the expected value");
2406 }
2407 #endif // ASSERT
2408 }
2409 #endif //PRODUCT
2410
2411 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2412
2413 Node* receiver = argument(0); // type: oop
2414
2415 // Build address expression.
2416 Node* heap_base_oop = top();
2417
2418 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2419 Node* base = argument(1); // type: oop
2420 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2421 Node* offset = argument(2); // type: long
2422 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2423 // to be plain byte offsets, which are also the same as those accepted
2424 // by oopDesc::field_addr.
2425 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2426 "fieldOffset must be byte-scaled");
2427
2428 if (base->is_InlineType()) {
2429 assert(!is_store, "InlineTypeNodes are non-larval value objects");
2430 InlineTypeNode* vt = base->as_InlineType();
2431 if (offset->is_Con()) {
2432 long off = find_long_con(offset, 0);
2433 ciInlineKlass* vk = vt->type()->inline_klass();
2434 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2435 return false;
2436 }
2437
2438 ciField* field = vk->get_non_flat_field_by_offset(off);
2439 if (field != nullptr) {
2440 BasicType bt = type2field[field->type()->basic_type()];
2441 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2442 bt = T_OBJECT;
2443 }
2444 if (bt == type && !field->is_flat()) {
2445 Node* value = vt->field_value_by_offset(off, false);
2446 const Type* value_type = _gvn.type(value);
2447 if (value->is_InlineType()) {
2448 value = value->as_InlineType()->adjust_scalarization_depth(this);
2449 } else if (value_type->is_inlinetypeptr()) {
2450 value = InlineTypeNode::make_from_oop(this, value, value_type->inline_klass());
2451 }
2452 set_result(value);
2453 return true;
2454 }
2455 }
2456 }
2457 {
2458 // Re-execute the unsafe access if allocation triggers deoptimization.
2459 PreserveReexecuteState preexecs(this);
2460 jvms()->set_should_reexecute(true);
2461 vt = vt->buffer(this);
2462 }
2463 base = vt->get_oop();
2464 }
2465
2466 // 32-bit machines ignore the high half!
2467 offset = ConvL2X(offset);
2468
2469 // Save state and restore on bailout
2470 SavedState old_state(this);
2471
2472 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2473 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2474
2475 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2476 if (type != T_OBJECT) {
2477 decorators |= IN_NATIVE; // off-heap primitive access
2478 } else {
2479 return false; // off-heap oop accesses are not supported
2480 }
2481 } else {
2482 heap_base_oop = base; // on-heap or mixed access
2483 }
2484
2485 // Can base be null? Otherwise, always on-heap access.
2486 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2487
2488 if (!can_access_non_heap) {
2489 decorators |= IN_HEAP;
2490 }
2491
2492 Node* val = is_store ? argument(4) : nullptr;
2493
2494 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2495 if (adr_type == TypePtr::NULL_PTR) {
2496 return false; // off-heap access with zero address
2497 }
2498
2499 // Try to categorize the address.
2500 Compile::AliasType* alias_type = C->alias_type(adr_type);
2501 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2502
2503 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2504 alias_type->adr_type() == TypeAryPtr::RANGE) {
2505 return false; // not supported
2506 }
2507
2508 bool mismatched = false;
2509 BasicType bt = T_ILLEGAL;
2510 ciField* field = nullptr;
2511 if (adr_type->isa_instptr()) {
2512 const TypeInstPtr* instptr = adr_type->is_instptr();
2513 ciInstanceKlass* k = instptr->instance_klass();
2514 int off = instptr->offset();
2515 if (instptr->const_oop() != nullptr &&
2516 k == ciEnv::current()->Class_klass() &&
2517 instptr->offset() >= (k->size_helper() * wordSize)) {
2518 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2519 field = k->get_field_by_offset(off, true);
2520 } else {
2521 field = k->get_non_flat_field_by_offset(off);
2522 }
2523 if (field != nullptr) {
2524 bt = type2field[field->type()->basic_type()];
2525 }
2526 if (bt != alias_type->basic_type()) {
2527 // Type mismatch. Is it an access to a nested flat field?
2528 field = k->get_field_by_offset(off, false);
2529 if (field != nullptr) {
2530 bt = type2field[field->type()->basic_type()];
2531 }
2532 }
2533 assert(bt == alias_type->basic_type(), "should match");
2534 } else {
2535 bt = alias_type->basic_type();
2536 }
2537
2538 if (bt != T_ILLEGAL) {
2539 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2540 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2541 // Alias type doesn't differentiate between byte[] and boolean[]).
2542 // Use address type to get the element type.
2543 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2544 }
2545 if (is_reference_type(bt, true)) {
2546 // accessing an array field with getReference is not a mismatch
2547 bt = T_OBJECT;
2548 }
2549 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2550 // Don't intrinsify mismatched object accesses
2551 return false;
2552 }
2553 mismatched = (bt != type);
2554 } else if (alias_type->adr_type()->isa_oopptr()) {
2555 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2556 }
2557
2558 old_state.discard();
2559 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2560
2561 if (mismatched) {
2562 decorators |= C2_MISMATCHED;
2563 }
2564
2565 // First guess at the value type.
2566 const Type *value_type = Type::get_const_basic_type(type);
2567
2568 // Figure out the memory ordering.
2569 decorators |= mo_decorator_for_access_kind(kind);
2570
2571 if (!is_store) {
2572 if (type == T_OBJECT) {
2573 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2574 if (tjp != nullptr) {
2575 value_type = tjp;
2576 }
2577 }
2578 }
2579
2580 receiver = null_check(receiver);
2581 if (stopped()) {
2582 return true;
2583 }
2584 // Heap pointers get a null-check from the interpreter,
2585 // as a courtesy. However, this is not guaranteed by Unsafe,
2586 // and it is not possible to fully distinguish unintended nulls
2587 // from intended ones in this API.
2588
2589 if (!is_store) {
2590 Node* p = nullptr;
2591 // Try to constant fold a load from a constant field
2592
2593 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !field->is_flat() && !mismatched) {
2594 // final or stable field
2595 p = make_constant_from_field(field, heap_base_oop);
2596 }
2597
2598 if (p == nullptr) { // Could not constant fold the load
2599 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2600 const TypeOopPtr* ptr = value_type->make_oopptr();
2601 if (ptr != nullptr && ptr->is_inlinetypeptr()) {
2602 // Load a non-flattened inline type from memory
2603 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass());
2604 }
2605 // Normalize the value returned by getBoolean in the following cases
2606 if (type == T_BOOLEAN &&
2607 (mismatched ||
2608 heap_base_oop == top() || // - heap_base_oop is null or
2609 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2610 // and the unsafe access is made to large offset
2611 // (i.e., larger than the maximum offset necessary for any
2612 // field access)
2613 ) {
2614 IdealKit ideal = IdealKit(this);
2615 #define __ ideal.
2616 IdealVariable normalized_result(ideal);
2617 __ declarations_done();
2618 __ set(normalized_result, p);
2619 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2620 __ set(normalized_result, ideal.ConI(1));
2621 ideal.end_if();
2622 final_sync(ideal);
2623 p = __ value(normalized_result);
2624 #undef __
2625 }
2626 }
2627 if (type == T_ADDRESS) {
2628 p = gvn().transform(new CastP2XNode(nullptr, p));
2629 p = ConvX2UL(p);
2630 }
2631 // The load node has the control of the preceding MemBarCPUOrder. All
2632 // following nodes will have the control of the MemBarCPUOrder inserted at
2633 // the end of this method. So, pushing the load onto the stack at a later
2634 // point is fine.
2635 set_result(p);
2636 } else {
2637 if (bt == T_ADDRESS) {
2638 // Repackage the long as a pointer.
2639 val = ConvL2X(val);
2640 val = gvn().transform(new CastX2PNode(val));
2641 }
2642 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2643 }
2644
2645 return true;
2646 }
2647
2648 bool LibraryCallKit::inline_unsafe_flat_access(bool is_store, AccessKind kind) {
2649 #ifdef ASSERT
2650 {
2651 ResourceMark rm;
2652 // Check the signatures.
2653 ciSignature* sig = callee()->signature();
2654 assert(sig->type_at(0)->basic_type() == T_OBJECT, "base should be object, but is %s", type2name(sig->type_at(0)->basic_type()));
2655 assert(sig->type_at(1)->basic_type() == T_LONG, "offset should be long, but is %s", type2name(sig->type_at(1)->basic_type()));
2656 assert(sig->type_at(2)->basic_type() == T_INT, "layout kind should be int, but is %s", type2name(sig->type_at(3)->basic_type()));
2657 assert(sig->type_at(3)->basic_type() == T_OBJECT, "value klass should be object, but is %s", type2name(sig->type_at(4)->basic_type()));
2658 if (is_store) {
2659 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value, but returns %s", type2name(sig->return_type()->basic_type()));
2660 assert(sig->count() == 5, "flat putter should have 5 arguments, but has %d", sig->count());
2661 assert(sig->type_at(4)->basic_type() == T_OBJECT, "put value should be object, but is %s", type2name(sig->type_at(5)->basic_type()));
2662 } else {
2663 assert(sig->return_type()->basic_type() == T_OBJECT, "getter must return an object, but returns %s", type2name(sig->return_type()->basic_type()));
2664 assert(sig->count() == 4, "flat getter should have 4 arguments, but has %d", sig->count());
2665 }
2666 }
2667 #endif // ASSERT
2668
2669 assert(kind == Relaxed, "Only plain accesses for now");
2670 if (callee()->is_static()) {
2671 // caller must have the capability!
2672 return false;
2673 }
2674 C->set_has_unsafe_access(true);
2675
2676 const TypeInstPtr* value_klass_node = _gvn.type(argument(5))->isa_instptr();
2677 if (value_klass_node == nullptr || value_klass_node->const_oop() == nullptr) {
2678 // parameter valueType is not a constant
2679 return false;
2680 }
2681 ciType* mirror_type = value_klass_node->const_oop()->as_instance()->java_mirror_type();
2682 if (!mirror_type->is_inlinetype()) {
2683 // Dead code
2684 return false;
2685 }
2686 ciInlineKlass* value_klass = mirror_type->as_inline_klass();
2687
2688 const TypeInt* layout_type = _gvn.type(argument(4))->isa_int();
2689 if (layout_type == nullptr || !layout_type->is_con()) {
2690 // parameter layoutKind is not a constant
2691 return false;
2692 }
2693 assert(layout_type->get_con() >= static_cast<int>(LayoutKind::REFERENCE) &&
2694 layout_type->get_con() < static_cast<int>(LayoutKind::UNKNOWN),
2695 "invalid layoutKind %d", layout_type->get_con());
2696 LayoutKind layout = static_cast<LayoutKind>(layout_type->get_con());
2697 assert(layout == LayoutKind::REFERENCE || layout == LayoutKind::NULL_FREE_NON_ATOMIC_FLAT ||
2698 layout == LayoutKind::NULL_FREE_ATOMIC_FLAT || layout == LayoutKind::NULLABLE_ATOMIC_FLAT,
2699 "unexpected layoutKind %d", layout_type->get_con());
2700
2701 null_check(argument(0));
2702 if (stopped()) {
2703 return true;
2704 }
2705
2706 Node* base = must_be_not_null(argument(1), true);
2707 Node* offset = argument(2);
2708 const Type* base_type = _gvn.type(base);
2709
2710 Node* ptr;
2711 bool immutable_memory = false;
2712 DecoratorSet decorators = C2_UNSAFE_ACCESS | IN_HEAP | MO_UNORDERED;
2713 if (base_type->isa_instptr()) {
2714 const TypeLong* offset_type = _gvn.type(offset)->isa_long();
2715 if (offset_type == nullptr || !offset_type->is_con()) {
2716 // Offset into a non-array should be a constant
2717 decorators |= C2_MISMATCHED;
2718 } else {
2719 int offset_con = checked_cast<int>(offset_type->get_con());
2720 ciInstanceKlass* base_klass = base_type->is_instptr()->instance_klass();
2721 ciField* field = base_klass->get_non_flat_field_by_offset(offset_con);
2722 if (field == nullptr) {
2723 assert(!base_klass->is_final(), "non-existence field at offset %d of class %s", offset_con, base_klass->name()->as_utf8());
2724 decorators |= C2_MISMATCHED;
2725 } else {
2726 assert(field->type() == value_klass, "field at offset %d of %s is of type %s, but valueType is %s",
2727 offset_con, base_klass->name()->as_utf8(), field->type()->name(), value_klass->name()->as_utf8());
2728 immutable_memory = field->is_strict() && field->is_final();
2729
2730 if (base->is_InlineType()) {
2731 assert(!is_store, "Cannot store into a non-larval value object");
2732 set_result(base->as_InlineType()->field_value_by_offset(offset_con, false));
2733 return true;
2734 }
2735 }
2736 }
2737
2738 if (base->is_InlineType()) {
2739 assert(!is_store, "Cannot store into a non-larval value object");
2740 base = base->as_InlineType()->buffer(this, true);
2741 }
2742 ptr = basic_plus_adr(base, ConvL2X(offset));
2743 } else if (base_type->isa_aryptr()) {
2744 decorators |= IS_ARRAY;
2745 if (layout == LayoutKind::REFERENCE) {
2746 if (!base_type->is_aryptr()->is_not_flat()) {
2747 const TypeAryPtr* array_type = base_type->is_aryptr()->cast_to_not_flat();
2748 Node* new_base = _gvn.transform(new CastPPNode(control(), base, array_type, ConstraintCastNode::DependencyType::NonFloatingNarrowing));
2749 replace_in_map(base, new_base);
2750 base = new_base;
2751 }
2752 ptr = basic_plus_adr(base, ConvL2X(offset));
2753 } else {
2754 if (UseArrayFlattening) {
2755 // Flat array must have an exact type
2756 bool is_null_free = !LayoutKindHelper::is_nullable_flat(layout);
2757 bool is_atomic = LayoutKindHelper::is_atomic_flat(layout);
2758 Node* new_base = cast_to_flat_array_exact(base, value_klass, is_null_free, is_atomic);
2759 replace_in_map(base, new_base);
2760 base = new_base;
2761 ptr = basic_plus_adr(base, ConvL2X(offset));
2762 const TypeAryPtr* ptr_type = _gvn.type(ptr)->is_aryptr();
2763 if (ptr_type->field_offset().get() != 0) {
2764 ptr = _gvn.transform(new CastPPNode(control(), ptr, ptr_type->with_field_offset(0), ConstraintCastNode::DependencyType::NonFloatingNarrowing));
2765 }
2766 } else {
2767 uncommon_trap(Deoptimization::Reason_intrinsic,
2768 Deoptimization::Action_none);
2769 return true;
2770 }
2771 }
2772 } else {
2773 decorators |= C2_MISMATCHED;
2774 ptr = basic_plus_adr(base, ConvL2X(offset));
2775 }
2776
2777 if (is_store) {
2778 Node* value = argument(6);
2779 const Type* value_type = _gvn.type(value);
2780 if (!value_type->is_inlinetypeptr()) {
2781 value_type = Type::get_const_type(value_klass)->filter_speculative(value_type);
2782 Node* new_value = _gvn.transform(new CastPPNode(control(), value, value_type, ConstraintCastNode::DependencyType::NonFloatingNarrowing));
2783 new_value = InlineTypeNode::make_from_oop(this, new_value, value_klass);
2784 replace_in_map(value, new_value);
2785 value = new_value;
2786 }
2787
2788 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());
2789 if (layout == LayoutKind::REFERENCE) {
2790 const TypePtr* ptr_type = (decorators & C2_MISMATCHED) != 0 ? TypeRawPtr::BOTTOM : _gvn.type(ptr)->is_ptr();
2791 access_store_at(base, ptr, ptr_type, value, value_type, T_OBJECT, decorators);
2792 } else {
2793 bool atomic = LayoutKindHelper::is_atomic_flat(layout);
2794 bool null_free = !LayoutKindHelper::is_nullable_flat(layout);
2795 value->as_InlineType()->store_flat(this, base, ptr, atomic, immutable_memory, null_free, decorators);
2796 }
2797
2798 return true;
2799 } else {
2800 decorators |= (C2_CONTROL_DEPENDENT_LOAD | C2_UNKNOWN_CONTROL_LOAD);
2801 InlineTypeNode* result;
2802 if (layout == LayoutKind::REFERENCE) {
2803 const TypePtr* ptr_type = (decorators & C2_MISMATCHED) != 0 ? TypeRawPtr::BOTTOM : _gvn.type(ptr)->is_ptr();
2804 Node* oop = access_load_at(base, ptr, ptr_type, Type::get_const_type(value_klass), T_OBJECT, decorators);
2805 result = InlineTypeNode::make_from_oop(this, oop, value_klass);
2806 } else {
2807 bool atomic = LayoutKindHelper::is_atomic_flat(layout);
2808 bool null_free = !LayoutKindHelper::is_nullable_flat(layout);
2809 result = InlineTypeNode::make_from_flat(this, value_klass, base, ptr, atomic, immutable_memory, null_free, decorators);
2810 }
2811
2812 set_result(result);
2813 return true;
2814 }
2815 }
2816
2817 //----------------------------inline_unsafe_load_store----------------------------
2818 // This method serves a couple of different customers (depending on LoadStoreKind):
2819 //
2820 // LS_cmp_swap:
2821 //
2822 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2823 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2824 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2825 //
2826 // LS_cmp_swap_weak:
2827 //
2828 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2829 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2830 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2831 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2832 //
2833 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2834 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2835 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2836 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2837 //
2838 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2839 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2840 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2841 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2842 //
2843 // LS_cmp_exchange:
2844 //
2845 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2846 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2847 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2848 //
2849 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2850 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2851 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2852 //
2853 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2854 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2855 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2856 //
2857 // LS_get_add:
2858 //
2859 // int getAndAddInt( Object o, long offset, int delta)
2860 // long getAndAddLong(Object o, long offset, long delta)
2861 //
2862 // LS_get_set:
2863 //
2864 // int getAndSet(Object o, long offset, int newValue)
2865 // long getAndSet(Object o, long offset, long newValue)
2866 // Object getAndSet(Object o, long offset, Object newValue)
2867 //
2868 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2869 // This basic scheme here is the same as inline_unsafe_access, but
2870 // differs in enough details that combining them would make the code
2871 // overly confusing. (This is a true fact! I originally combined
2872 // them, but even I was confused by it!) As much code/comments as
2873 // possible are retained from inline_unsafe_access though to make
2874 // the correspondences clearer. - dl
2875
2876 if (callee()->is_static()) return false; // caller must have the capability!
2877
2878 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2879 decorators |= mo_decorator_for_access_kind(access_kind);
2880
2881 #ifndef PRODUCT
2882 BasicType rtype;
2883 {
2884 ResourceMark rm;
2885 // Check the signatures.
2886 ciSignature* sig = callee()->signature();
2887 rtype = sig->return_type()->basic_type();
2888 switch(kind) {
2889 case LS_get_add:
2890 case LS_get_set: {
2891 // Check the signatures.
2892 #ifdef ASSERT
2893 assert(rtype == type, "get and set must return the expected type");
2894 assert(sig->count() == 3, "get and set has 3 arguments");
2895 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2896 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2897 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2898 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2899 #endif // ASSERT
2900 break;
2901 }
2902 case LS_cmp_swap:
2903 case LS_cmp_swap_weak: {
2904 // Check the signatures.
2905 #ifdef ASSERT
2906 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2907 assert(sig->count() == 4, "CAS has 4 arguments");
2908 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2909 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2910 #endif // ASSERT
2911 break;
2912 }
2913 case LS_cmp_exchange: {
2914 // Check the signatures.
2915 #ifdef ASSERT
2916 assert(rtype == type, "CAS must return the expected type");
2917 assert(sig->count() == 4, "CAS has 4 arguments");
2918 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2919 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2920 #endif // ASSERT
2921 break;
2922 }
2923 default:
2924 ShouldNotReachHere();
2925 }
2926 }
2927 #endif //PRODUCT
2928
2929 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2930
2931 // Get arguments:
2932 Node* receiver = nullptr;
2933 Node* base = nullptr;
2934 Node* offset = nullptr;
2935 Node* oldval = nullptr;
2936 Node* newval = nullptr;
2937 switch(kind) {
2938 case LS_cmp_swap:
2939 case LS_cmp_swap_weak:
2940 case LS_cmp_exchange: {
2941 const bool two_slot_type = type2size[type] == 2;
2942 receiver = argument(0); // type: oop
2943 base = argument(1); // type: oop
2944 offset = argument(2); // type: long
2945 oldval = argument(4); // type: oop, int, or long
2946 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2947 break;
2948 }
2949 case LS_get_add:
2950 case LS_get_set: {
2951 receiver = argument(0); // type: oop
2952 base = argument(1); // type: oop
2953 offset = argument(2); // type: long
2954 oldval = nullptr;
2955 newval = argument(4); // type: oop, int, or long
2956 break;
2957 }
2958 default:
2959 ShouldNotReachHere();
2960 }
2961
2962 // Build field offset expression.
2963 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2964 // to be plain byte offsets, which are also the same as those accepted
2965 // by oopDesc::field_addr.
2966 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2967 // 32-bit machines ignore the high half of long offsets
2968 offset = ConvL2X(offset);
2969 // Save state and restore on bailout
2970 SavedState old_state(this);
2971 Node* adr = make_unsafe_address(base, offset,type, false);
2972 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2973
2974 Compile::AliasType* alias_type = C->alias_type(adr_type);
2975 BasicType bt = alias_type->basic_type();
2976 if (bt != T_ILLEGAL &&
2977 (is_reference_type(bt) != (type == T_OBJECT))) {
2978 // Don't intrinsify mismatched object accesses.
2979 return false;
2980 }
2981
2982 old_state.discard();
2983
2984 // For CAS, unlike inline_unsafe_access, there seems no point in
2985 // trying to refine types. Just use the coarse types here.
2986 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2987 const Type *value_type = Type::get_const_basic_type(type);
2988
2989 switch (kind) {
2990 case LS_get_set:
2991 case LS_cmp_exchange: {
2992 if (type == T_OBJECT) {
2993 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2994 if (tjp != nullptr) {
2995 value_type = tjp;
2996 }
2997 }
2998 break;
2999 }
3000 case LS_cmp_swap:
3001 case LS_cmp_swap_weak:
3002 case LS_get_add:
3003 break;
3004 default:
3005 ShouldNotReachHere();
3006 }
3007
3008 // Null check receiver.
3009 receiver = null_check(receiver);
3010 if (stopped()) {
3011 return true;
3012 }
3013
3014 int alias_idx = C->get_alias_index(adr_type);
3015
3016 if (is_reference_type(type)) {
3017 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
3018
3019 if (oldval != nullptr && oldval->is_InlineType()) {
3020 // Re-execute the unsafe access if allocation triggers deoptimization.
3021 PreserveReexecuteState preexecs(this);
3022 jvms()->set_should_reexecute(true);
3023 oldval = oldval->as_InlineType()->buffer(this)->get_oop();
3024 }
3025 if (newval != nullptr && newval->is_InlineType()) {
3026 // Re-execute the unsafe access if allocation triggers deoptimization.
3027 PreserveReexecuteState preexecs(this);
3028 jvms()->set_should_reexecute(true);
3029 newval = newval->as_InlineType()->buffer(this)->get_oop();
3030 }
3031
3032 // Transformation of a value which could be null pointer (CastPP #null)
3033 // could be delayed during Parse (for example, in adjust_map_after_if()).
3034 // Execute transformation here to avoid barrier generation in such case.
3035 if (_gvn.type(newval) == TypePtr::NULL_PTR)
3036 newval = _gvn.makecon(TypePtr::NULL_PTR);
3037
3038 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
3039 // Refine the value to a null constant, when it is known to be null
3040 oldval = _gvn.makecon(TypePtr::NULL_PTR);
3041 }
3042 }
3043
3044 Node* result = nullptr;
3045 switch (kind) {
3046 case LS_cmp_exchange: {
3047 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
3048 oldval, newval, value_type, type, decorators);
3049 break;
3050 }
3051 case LS_cmp_swap_weak:
3052 decorators |= C2_WEAK_CMPXCHG;
3053 case LS_cmp_swap: {
3054 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
3055 oldval, newval, value_type, type, decorators);
3056 break;
3057 }
3058 case LS_get_set: {
3059 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
3060 newval, value_type, type, decorators);
3061 break;
3062 }
3063 case LS_get_add: {
3064 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
3065 newval, value_type, type, decorators);
3066 break;
3067 }
3068 default:
3069 ShouldNotReachHere();
3070 }
3071
3072 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3073 set_result(result);
3074 return true;
3075 }
3076
3077 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3078 // Regardless of form, don't allow previous ld/st to move down,
3079 // then issue acquire, release, or volatile mem_bar.
3080 insert_mem_bar(Op_MemBarCPUOrder);
3081 switch(id) {
3082 case vmIntrinsics::_loadFence:
3083 insert_mem_bar(Op_LoadFence);
3084 return true;
3085 case vmIntrinsics::_storeFence:
3086 insert_mem_bar(Op_StoreFence);
3087 return true;
3088 case vmIntrinsics::_storeStoreFence:
3089 insert_mem_bar(Op_StoreStoreFence);
3090 return true;
3091 case vmIntrinsics::_fullFence:
3092 insert_mem_bar(Op_MemBarFull);
3093 return true;
3094 default:
3095 fatal_unexpected_iid(id);
3096 return false;
3097 }
3098 }
3099
3100 // private native int arrayInstanceBaseOffset0(Object[] array);
3101 bool LibraryCallKit::inline_arrayInstanceBaseOffset() {
3102 Node* array = argument(1);
3103 Node* klass_node = load_object_klass(array);
3104
3105 jint layout_con = Klass::_lh_neutral_value;
3106 Node* layout_val = get_layout_helper(klass_node, layout_con);
3107 int layout_is_con = (layout_val == nullptr);
3108
3109 Node* header_size = nullptr;
3110 if (layout_is_con) {
3111 int hsize = Klass::layout_helper_header_size(layout_con);
3112 header_size = intcon(hsize);
3113 } else {
3114 Node* hss = intcon(Klass::_lh_header_size_shift);
3115 Node* hsm = intcon(Klass::_lh_header_size_mask);
3116 header_size = _gvn.transform(new URShiftINode(layout_val, hss));
3117 header_size = _gvn.transform(new AndINode(header_size, hsm));
3118 }
3119 set_result(header_size);
3120 return true;
3121 }
3122
3123 // private native int arrayInstanceIndexScale0(Object[] array);
3124 bool LibraryCallKit::inline_arrayInstanceIndexScale() {
3125 Node* array = argument(1);
3126 Node* klass_node = load_object_klass(array);
3127
3128 jint layout_con = Klass::_lh_neutral_value;
3129 Node* layout_val = get_layout_helper(klass_node, layout_con);
3130 int layout_is_con = (layout_val == nullptr);
3131
3132 Node* element_size = nullptr;
3133 if (layout_is_con) {
3134 int log_element_size = Klass::layout_helper_log2_element_size(layout_con);
3135 int elem_size = 1 << log_element_size;
3136 element_size = intcon(elem_size);
3137 } else {
3138 Node* ess = intcon(Klass::_lh_log2_element_size_shift);
3139 Node* esm = intcon(Klass::_lh_log2_element_size_mask);
3140 Node* log_element_size = _gvn.transform(new URShiftINode(layout_val, ess));
3141 log_element_size = _gvn.transform(new AndINode(log_element_size, esm));
3142 element_size = _gvn.transform(new LShiftINode(intcon(1), log_element_size));
3143 }
3144 set_result(element_size);
3145 return true;
3146 }
3147
3148 // private native int arrayLayout0(Object[] array);
3149 bool LibraryCallKit::inline_arrayLayout() {
3150 RegionNode* region = new RegionNode(2);
3151 Node* phi = new PhiNode(region, TypeInt::POS);
3152
3153 Node* array = argument(1);
3154 Node* klass_node = load_object_klass(array);
3155 generate_refArray_guard(klass_node, region);
3156 if (region->req() == 3) {
3157 phi->add_req(intcon((jint)LayoutKind::REFERENCE));
3158 }
3159
3160 int layout_kind_offset = in_bytes(FlatArrayKlass::layout_kind_offset());
3161 Node* layout_kind_addr = basic_plus_adr(top(), klass_node, layout_kind_offset);
3162 Node* layout_kind = make_load(nullptr, layout_kind_addr, TypeInt::POS, T_INT, MemNode::unordered);
3163
3164 region->init_req(1, control());
3165 phi->init_req(1, layout_kind);
3166
3167 set_control(_gvn.transform(region));
3168 set_result(_gvn.transform(phi));
3169 return true;
3170 }
3171
3172 // private native int[] getFieldMap0(Class <?> c);
3173 // int offset = c._klass._acmp_maps_offset;
3174 // return (int[])c.obj_field(offset);
3175 bool LibraryCallKit::inline_getFieldMap() {
3176 Node* mirror = argument(1);
3177 Node* klass = load_klass_from_mirror(mirror, false, nullptr, 0);
3178
3179 int field_map_offset_offset = in_bytes(InstanceKlass::acmp_maps_offset_offset());
3180 Node* field_map_offset_addr = basic_plus_adr(top(), klass, field_map_offset_offset);
3181 Node* field_map_offset = make_load(nullptr, field_map_offset_addr, TypeInt::INT, T_INT, MemNode::unordered);
3182 field_map_offset = _gvn.transform(ConvI2L(field_map_offset));
3183
3184 Node* map_addr = basic_plus_adr(mirror, field_map_offset);
3185 const TypeAryPtr* val_type = TypeAryPtr::INTS->cast_to_ptr_type(TypePtr::NotNull)->with_offset(0);
3186 // TODO 8350865 Remove this
3187 val_type = val_type->cast_to_not_flat(true)->cast_to_not_null_free(true);
3188 Node* map = access_load_at(mirror, map_addr, TypeAryPtr::INTS, val_type, T_ARRAY, IN_HEAP | MO_UNORDERED);
3189
3190 set_result(map);
3191 return true;
3192 }
3193
3194 bool LibraryCallKit::inline_onspinwait() {
3195 insert_mem_bar(Op_OnSpinWait);
3196 return true;
3197 }
3198
3199 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3200 if (!kls->is_Con()) {
3201 return true;
3202 }
3203 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
3204 if (klsptr == nullptr) {
3205 return true;
3206 }
3207 ciInstanceKlass* ik = klsptr->instance_klass();
3208 // don't need a guard for a klass that is already initialized
3209 return !ik->is_initialized();
3210 }
3211
3212 //----------------------------inline_unsafe_writeback0-------------------------
3213 // public native void Unsafe.writeback0(long address)
3214 bool LibraryCallKit::inline_unsafe_writeback0() {
3215 if (!Matcher::has_match_rule(Op_CacheWB)) {
3216 return false;
3217 }
3218 #ifndef PRODUCT
3219 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
3220 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
3221 ciSignature* sig = callee()->signature();
3222 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
3223 #endif
3224 null_check_receiver(); // null-check, then ignore
3225 Node *addr = argument(1);
3226 addr = new CastX2PNode(addr);
3227 addr = _gvn.transform(addr);
3228 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
3229 flush = _gvn.transform(flush);
3230 set_memory(flush, TypeRawPtr::BOTTOM);
3231 return true;
3232 }
3233
3234 //----------------------------inline_unsafe_writeback0-------------------------
3235 // public native void Unsafe.writeback0(long address)
3236 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
3237 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
3238 return false;
3239 }
3240 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
3241 return false;
3242 }
3243 #ifndef PRODUCT
3244 assert(Matcher::has_match_rule(Op_CacheWB),
3245 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
3246 : "found match rule for CacheWBPostSync but not CacheWB"));
3247
3248 #endif
3249 null_check_receiver(); // null-check, then ignore
3250 Node *sync;
3251 if (is_pre) {
3252 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3253 } else {
3254 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3255 }
3256 sync = _gvn.transform(sync);
3257 set_memory(sync, TypeRawPtr::BOTTOM);
3258 return true;
3259 }
3260
3261 //----------------------------inline_unsafe_allocate---------------------------
3262 // public native Object Unsafe.allocateInstance(Class<?> cls);
3263 bool LibraryCallKit::inline_unsafe_allocate() {
3264
3265 #if INCLUDE_JVMTI
3266 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3267 return false;
3268 }
3269 #endif //INCLUDE_JVMTI
3270
3271 if (callee()->is_static()) return false; // caller must have the capability!
3272
3273 null_check_receiver(); // null-check, then ignore
3274 Node* cls = null_check(argument(1));
3275 if (stopped()) return true;
3276
3277 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
3278 kls = null_check(kls);
3279 if (stopped()) return true; // argument was like int.class
3280
3281 #if INCLUDE_JVMTI
3282 // Don't try to access new allocated obj in the intrinsic.
3283 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
3284 // Deoptimize and allocate in interpreter instead.
3285 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
3286 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
3287 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
3288 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
3289 {
3290 BuildCutout unless(this, tst, PROB_MAX);
3291 uncommon_trap(Deoptimization::Reason_intrinsic,
3292 Deoptimization::Action_make_not_entrant);
3293 }
3294 if (stopped()) {
3295 return true;
3296 }
3297 #endif //INCLUDE_JVMTI
3298
3299 Node* test = nullptr;
3300 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3301 // Note: The argument might still be an illegal value like
3302 // Serializable.class or Object[].class. The runtime will handle it.
3303 // But we must make an explicit check for initialization.
3304 Node* insp = off_heap_plus_addr(kls, in_bytes(InstanceKlass::init_state_offset()));
3305 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3306 // can generate code to load it as unsigned byte.
3307 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
3308 Node* bits = intcon(InstanceKlass::fully_initialized);
3309 test = _gvn.transform(new SubINode(inst, bits));
3310 // The 'test' is non-zero if we need to take a slow path.
3311 }
3312 Node* obj = new_instance(kls, test);
3313 set_result(obj);
3314 return true;
3315 }
3316
3317 //------------------------inline_native_time_funcs--------------
3318 // inline code for System.currentTimeMillis() and System.nanoTime()
3319 // these have the same type and signature
3320 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3321 const TypeFunc* tf = OptoRuntime::void_long_Type();
3322 const TypePtr* no_memory_effects = nullptr;
3323 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3324 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3325 #ifdef ASSERT
3326 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3327 assert(value_top == top(), "second value must be top");
3328 #endif
3329 set_result(value);
3330 return true;
3331 }
3332
3333 //--------------------inline_native_vthread_start_transition--------------------
3334 // inline void startTransition(boolean is_mount);
3335 // inline void startFinalTransition();
3336 // Pseudocode of implementation:
3337 //
3338 // java_lang_Thread::set_is_in_vthread_transition(vthread, true);
3339 // carrier->set_is_in_vthread_transition(true);
3340 // OrderAccess::storeload();
3341 // int disable_requests = java_lang_Thread::vthread_transition_disable_count(vthread)
3342 // + global_vthread_transition_disable_count();
3343 // if (disable_requests > 0) {
3344 // slow path: runtime call
3345 // }
3346 bool LibraryCallKit::inline_native_vthread_start_transition(address funcAddr, const char* funcName, bool is_final_transition) {
3347 Node* vt_oop = must_be_not_null(argument(0), true); // VirtualThread this argument
3348 IdealKit ideal(this);
3349
3350 Node* thread = ideal.thread();
3351 Node* jt_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3352 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3353 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3354 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3355 insert_mem_bar(Op_MemBarStoreLoad);
3356 ideal.sync_kit(this);
3357
3358 Node* global_disable_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::global_vthread_transition_disable_count_address()));
3359 Node* global_disable = ideal.load(ideal.ctrl(), global_disable_addr, TypeInt::INT, T_INT, Compile::AliasIdxRaw, true /*require_atomic_access*/);
3360 Node* vt_disable_addr = basic_plus_adr(vt_oop, java_lang_Thread::vthread_transition_disable_count_offset());
3361 const TypePtr* vt_disable_addr_t = _gvn.type(vt_disable_addr)->is_ptr();
3362 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*/);
3363 Node* disabled = _gvn.transform(new AddINode(global_disable, vt_disable));
3364
3365 ideal.if_then(disabled, BoolTest::ne, ideal.ConI(0)); {
3366 sync_kit(ideal);
3367 Node* is_mount = is_final_transition ? ideal.ConI(0) : argument(1);
3368 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3369 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3370 ideal.sync_kit(this);
3371 }
3372 ideal.end_if();
3373
3374 final_sync(ideal);
3375 return true;
3376 }
3377
3378 bool LibraryCallKit::inline_native_vthread_end_transition(address funcAddr, const char* funcName, bool is_first_transition) {
3379 Node* vt_oop = must_be_not_null(argument(0), true); // VirtualThread this argument
3380 IdealKit ideal(this);
3381
3382 Node* _notify_jvmti_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::notify_jvmti_events_address()));
3383 Node* _notify_jvmti = ideal.load(ideal.ctrl(), _notify_jvmti_addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3384
3385 ideal.if_then(_notify_jvmti, BoolTest::eq, ideal.ConI(1)); {
3386 sync_kit(ideal);
3387 Node* is_mount = is_first_transition ? ideal.ConI(1) : argument(1);
3388 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3389 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3390 ideal.sync_kit(this);
3391 } ideal.else_(); {
3392 Node* thread = ideal.thread();
3393 Node* jt_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3394 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3395
3396 sync_kit(ideal);
3397 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3398 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3399 ideal.sync_kit(this);
3400 } ideal.end_if();
3401
3402 final_sync(ideal);
3403 return true;
3404 }
3405
3406 #if INCLUDE_JVMTI
3407
3408 // Always update the is_disable_suspend bit.
3409 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3410 if (!DoJVMTIVirtualThreadTransitions) {
3411 return true;
3412 }
3413 IdealKit ideal(this);
3414
3415 {
3416 // unconditionally update the is_disable_suspend bit in current JavaThread
3417 Node* thread = ideal.thread();
3418 Node* arg = argument(0); // argument for notification
3419 Node* addr = off_heap_plus_addr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3420 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3421
3422 sync_kit(ideal);
3423 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3424 ideal.sync_kit(this);
3425 }
3426 final_sync(ideal);
3427
3428 return true;
3429 }
3430
3431 #endif // INCLUDE_JVMTI
3432
3433 #ifdef JFR_HAVE_INTRINSICS
3434
3435 /**
3436 * if oop->klass != null
3437 * // normal class
3438 * epoch = _epoch_state ? 2 : 1
3439 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3440 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3441 * }
3442 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3443 * else
3444 * // primitive class
3445 * if oop->array_klass != null
3446 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3447 * else
3448 * id = LAST_TYPE_ID + 1 // void class path
3449 * if (!signaled)
3450 * signaled = true
3451 */
3452 bool LibraryCallKit::inline_native_classID() {
3453 Node* cls = argument(0);
3454
3455 IdealKit ideal(this);
3456 #define __ ideal.
3457 IdealVariable result(ideal); __ declarations_done();
3458 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3459 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3460 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3461
3462
3463 __ if_then(kls, BoolTest::ne, null()); {
3464 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3465 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3466
3467 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3468 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3469 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3470 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3471 mask = _gvn.transform(new OrLNode(mask, epoch));
3472 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3473
3474 float unlikely = PROB_UNLIKELY(0.999);
3475 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3476 sync_kit(ideal);
3477 make_runtime_call(RC_LEAF,
3478 OptoRuntime::class_id_load_barrier_Type(),
3479 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3480 "class id load barrier",
3481 TypePtr::BOTTOM,
3482 kls);
3483 ideal.sync_kit(this);
3484 } __ end_if();
3485
3486 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3487 } __ else_(); {
3488 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3489 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3490 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3491 __ if_then(array_kls, BoolTest::ne, null()); {
3492 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3493 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3494 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3495 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3496 } __ else_(); {
3497 // void class case
3498 ideal.set(result, longcon(LAST_TYPE_ID + 1));
3499 } __ end_if();
3500
3501 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3502 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3503 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3504 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3505 } __ end_if();
3506 } __ end_if();
3507
3508 final_sync(ideal);
3509 set_result(ideal.value(result));
3510 #undef __
3511 return true;
3512 }
3513
3514 //------------------------inline_native_jvm_commit------------------
3515 bool LibraryCallKit::inline_native_jvm_commit() {
3516 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3517
3518 // Save input memory and i_o state.
3519 Node* input_memory_state = reset_memory();
3520 set_all_memory(input_memory_state);
3521 Node* input_io_state = i_o();
3522
3523 // TLS.
3524 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3525 // Jfr java buffer.
3526 Node* java_buffer_offset = _gvn.transform(AddPNode::make_off_heap(tls_ptr, MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR))));
3527 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3528 Node* java_buffer_pos_offset = _gvn.transform(AddPNode::make_off_heap(java_buffer, MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET))));
3529
3530 // Load the current value of the notified field in the JfrThreadLocal.
3531 Node* notified_offset = off_heap_plus_addr(tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3532 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3533
3534 // Test for notification.
3535 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3536 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3537 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3538
3539 // True branch, is notified.
3540 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3541 set_control(is_notified);
3542
3543 // Reset notified state.
3544 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3545 Node* notified_reset_memory = reset_memory();
3546
3547 // 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.
3548 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3549 // Convert the machine-word to a long.
3550 Node* current_pos = ConvX2L(current_pos_X);
3551
3552 // False branch, not notified.
3553 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3554 set_control(not_notified);
3555 set_all_memory(input_memory_state);
3556
3557 // Arg is the next position as a long.
3558 Node* arg = argument(0);
3559 // Convert long to machine-word.
3560 Node* next_pos_X = ConvL2X(arg);
3561
3562 // Store the next_position to the underlying jfr java buffer.
3563 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3564
3565 Node* commit_memory = reset_memory();
3566 set_all_memory(commit_memory);
3567
3568 // 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.
3569 Node* java_buffer_flags_offset = _gvn.transform(AddPNode::make_off_heap(java_buffer, MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET))));
3570 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3571 Node* lease_constant = _gvn.intcon(4);
3572
3573 // And flags with lease constant.
3574 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3575
3576 // Branch on lease to conditionalize returning the leased java buffer.
3577 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3578 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3579 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3580
3581 // False branch, not a lease.
3582 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3583
3584 // True branch, is lease.
3585 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3586 set_control(is_lease);
3587
3588 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3589 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3590 OptoRuntime::void_void_Type(),
3591 SharedRuntime::jfr_return_lease(),
3592 "return_lease", TypePtr::BOTTOM);
3593 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3594
3595 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3596 record_for_igvn(lease_compare_rgn);
3597 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3598 record_for_igvn(lease_compare_mem);
3599 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3600 record_for_igvn(lease_compare_io);
3601 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3602 record_for_igvn(lease_result_value);
3603
3604 // Update control and phi nodes.
3605 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3606 lease_compare_rgn->init_req(_false_path, not_lease);
3607
3608 lease_compare_mem->init_req(_true_path, reset_memory());
3609 lease_compare_mem->init_req(_false_path, commit_memory);
3610
3611 lease_compare_io->init_req(_true_path, i_o());
3612 lease_compare_io->init_req(_false_path, input_io_state);
3613
3614 lease_result_value->init_req(_true_path, _gvn.longcon(0)); // if the lease was returned, return 0L.
3615 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3616
3617 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3618 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3619 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3620 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3621
3622 // Update control and phi nodes.
3623 result_rgn->init_req(_true_path, is_notified);
3624 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3625
3626 result_mem->init_req(_true_path, notified_reset_memory);
3627 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3628
3629 result_io->init_req(_true_path, input_io_state);
3630 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3631
3632 result_value->init_req(_true_path, current_pos);
3633 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3634
3635 // Set output state.
3636 set_control(_gvn.transform(result_rgn));
3637 set_all_memory(_gvn.transform(result_mem));
3638 set_i_o(_gvn.transform(result_io));
3639 set_result(result_rgn, result_value);
3640 return true;
3641 }
3642
3643 /*
3644 * The intrinsic is a model of this pseudo-code:
3645 *
3646 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3647 * jobject h_event_writer = tl->java_event_writer();
3648 * if (h_event_writer == nullptr) {
3649 * return nullptr;
3650 * }
3651 * oop threadObj = Thread::threadObj();
3652 * oop vthread = java_lang_Thread::vthread(threadObj);
3653 * traceid tid;
3654 * bool pinVirtualThread;
3655 * bool excluded;
3656 * if (vthread != threadObj) { // i.e. current thread is virtual
3657 * tid = java_lang_Thread::tid(vthread);
3658 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3659 * pinVirtualThread = VMContinuations;
3660 * excluded = vthread_epoch_raw & excluded_mask;
3661 * if (!excluded) {
3662 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3663 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3664 * if (vthread_epoch != current_epoch) {
3665 * write_checkpoint();
3666 * }
3667 * }
3668 * } else {
3669 * tid = java_lang_Thread::tid(threadObj);
3670 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3671 * pinVirtualThread = false;
3672 * excluded = thread_epoch_raw & excluded_mask;
3673 * }
3674 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3675 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3676 * if (tid_in_event_writer != tid) {
3677 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3678 * setField(event_writer, "excluded", excluded);
3679 * setField(event_writer, "threadID", tid);
3680 * }
3681 * return event_writer
3682 */
3683 bool LibraryCallKit::inline_native_getEventWriter() {
3684 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3685
3686 // Save input memory and i_o state.
3687 Node* input_memory_state = reset_memory();
3688 set_all_memory(input_memory_state);
3689 Node* input_io_state = i_o();
3690
3691 // The most significant bit of the u2 is used to denote thread exclusion
3692 Node* excluded_shift = _gvn.intcon(15);
3693 Node* excluded_mask = _gvn.intcon(1 << 15);
3694 // The epoch generation is the range [1-32767]
3695 Node* epoch_mask = _gvn.intcon(32767);
3696
3697 // TLS
3698 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3699
3700 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3701 Node* jobj_ptr = off_heap_plus_addr(tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3702
3703 // Load the eventwriter jobject handle.
3704 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3705
3706 // Null check the jobject handle.
3707 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3708 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3709 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3710
3711 // False path, jobj is null.
3712 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3713
3714 // True path, jobj is not null.
3715 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3716
3717 set_control(jobj_is_not_null);
3718
3719 // Load the threadObj for the CarrierThread.
3720 Node* threadObj = generate_current_thread(tls_ptr);
3721
3722 // Load the vthread.
3723 Node* vthread = generate_virtual_thread(tls_ptr);
3724
3725 // If vthread != threadObj, this is a virtual thread.
3726 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3727 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3728 IfNode* iff_vthread_not_equal_threadObj =
3729 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3730
3731 // False branch, fallback to threadObj.
3732 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3733 set_control(vthread_equal_threadObj);
3734
3735 // Load the tid field from the vthread object.
3736 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3737
3738 // Load the raw epoch value from the threadObj.
3739 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3740 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3741 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3742 TypeInt::CHAR, T_CHAR,
3743 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3744
3745 // Mask off the excluded information from the epoch.
3746 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3747
3748 // True branch, this is a virtual thread.
3749 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3750 set_control(vthread_not_equal_threadObj);
3751
3752 // Load the tid field from the vthread object.
3753 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3754
3755 // Continuation support determines if a virtual thread should be pinned.
3756 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3757 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3758
3759 // Load the raw epoch value from the vthread.
3760 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3761 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3762 TypeInt::CHAR, T_CHAR,
3763 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3764
3765 // Mask off the excluded information from the epoch.
3766 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, excluded_mask));
3767
3768 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3769 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, excluded_mask));
3770 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3771 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3772
3773 // False branch, vthread is excluded, no need to write epoch info.
3774 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3775
3776 // True branch, vthread is included, update epoch info.
3777 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3778 set_control(included);
3779
3780 // Get epoch value.
3781 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, epoch_mask));
3782
3783 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3784 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3785 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3786
3787 // Compare the epoch in the vthread to the current epoch generation.
3788 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3789 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3790 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3791
3792 // False path, epoch is equal, checkpoint information is valid.
3793 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3794
3795 // True path, epoch is not equal, write a checkpoint for the vthread.
3796 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3797
3798 set_control(epoch_is_not_equal);
3799
3800 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3801 // The call also updates the native thread local thread id and the vthread with the current epoch.
3802 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3803 OptoRuntime::jfr_write_checkpoint_Type(),
3804 SharedRuntime::jfr_write_checkpoint(),
3805 "write_checkpoint", TypePtr::BOTTOM);
3806 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3807
3808 // vthread epoch != current epoch
3809 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3810 record_for_igvn(epoch_compare_rgn);
3811 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3812 record_for_igvn(epoch_compare_mem);
3813 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3814 record_for_igvn(epoch_compare_io);
3815
3816 // Update control and phi nodes.
3817 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3818 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3819 epoch_compare_mem->init_req(_true_path, reset_memory());
3820 epoch_compare_mem->init_req(_false_path, input_memory_state);
3821 epoch_compare_io->init_req(_true_path, i_o());
3822 epoch_compare_io->init_req(_false_path, input_io_state);
3823
3824 // excluded != true
3825 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3826 record_for_igvn(exclude_compare_rgn);
3827 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3828 record_for_igvn(exclude_compare_mem);
3829 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3830 record_for_igvn(exclude_compare_io);
3831
3832 // Update control and phi nodes.
3833 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3834 exclude_compare_rgn->init_req(_false_path, excluded);
3835 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3836 exclude_compare_mem->init_req(_false_path, input_memory_state);
3837 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3838 exclude_compare_io->init_req(_false_path, input_io_state);
3839
3840 // vthread != threadObj
3841 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3842 record_for_igvn(vthread_compare_rgn);
3843 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3844 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3845 record_for_igvn(vthread_compare_io);
3846 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3847 record_for_igvn(tid);
3848 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3849 record_for_igvn(exclusion);
3850 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3851 record_for_igvn(pinVirtualThread);
3852
3853 // Update control and phi nodes.
3854 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3855 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3856 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3857 vthread_compare_mem->init_req(_false_path, input_memory_state);
3858 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3859 vthread_compare_io->init_req(_false_path, input_io_state);
3860 tid->init_req(_true_path, vthread_tid);
3861 tid->init_req(_false_path, thread_obj_tid);
3862 exclusion->init_req(_true_path, vthread_is_excluded);
3863 exclusion->init_req(_false_path, threadObj_is_excluded);
3864 pinVirtualThread->init_req(_true_path, continuation_support);
3865 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3866
3867 // Update branch state.
3868 set_control(_gvn.transform(vthread_compare_rgn));
3869 set_all_memory(_gvn.transform(vthread_compare_mem));
3870 set_i_o(_gvn.transform(vthread_compare_io));
3871
3872 // Load the event writer oop by dereferencing the jobject handle.
3873 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3874 assert(klass_EventWriter->is_loaded(), "invariant");
3875 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3876 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3877 const TypeOopPtr* const xtype = aklass->as_instance_type();
3878 Node* jobj_untagged = _gvn.transform(AddPNode::make_off_heap(jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3879 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3880
3881 // Load the current thread id from the event writer object.
3882 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3883 // Get the field offset to, conditionally, store an updated tid value later.
3884 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3885 // Get the field offset to, conditionally, store an updated exclusion value later.
3886 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3887 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3888 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3889
3890 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3891 record_for_igvn(event_writer_tid_compare_rgn);
3892 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3893 record_for_igvn(event_writer_tid_compare_mem);
3894 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3895 record_for_igvn(event_writer_tid_compare_io);
3896
3897 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3898 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3899 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3900 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3901
3902 // False path, tids are the same.
3903 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3904
3905 // True path, tid is not equal, need to update the tid in the event writer.
3906 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3907 record_for_igvn(tid_is_not_equal);
3908
3909 // Store the pin state to the event writer.
3910 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3911
3912 // Store the exclusion state to the event writer.
3913 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3914 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3915
3916 // Store the tid to the event writer.
3917 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3918
3919 // Update control and phi nodes.
3920 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3921 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3922 event_writer_tid_compare_mem->init_req(_true_path, reset_memory());
3923 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3924 event_writer_tid_compare_io->init_req(_true_path, i_o());
3925 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3926
3927 // Result of top level CFG, Memory, IO and Value.
3928 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3929 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3930 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3931 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3932
3933 // Result control.
3934 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3935 result_rgn->init_req(_false_path, jobj_is_null);
3936
3937 // Result memory.
3938 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3939 result_mem->init_req(_false_path, input_memory_state);
3940
3941 // Result IO.
3942 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3943 result_io->init_req(_false_path, input_io_state);
3944
3945 // Result value.
3946 result_value->init_req(_true_path, event_writer); // return event writer oop
3947 result_value->init_req(_false_path, null()); // return null
3948
3949 // Set output state.
3950 set_control(_gvn.transform(result_rgn));
3951 set_all_memory(_gvn.transform(result_mem));
3952 set_i_o(_gvn.transform(result_io));
3953 set_result(result_rgn, result_value);
3954 return true;
3955 }
3956
3957 /*
3958 * The intrinsic is a model of this pseudo-code:
3959 *
3960 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3961 * if (carrierThread != thread) { // is virtual thread
3962 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3963 * bool excluded = vthread_epoch_raw & excluded_mask;
3964 * AtomicAccess::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3965 * AtomicAccess::store(&tl->_contextual_thread_excluded, is_excluded);
3966 * if (!excluded) {
3967 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3968 * AtomicAccess::store(&tl->_vthread_epoch, vthread_epoch);
3969 * }
3970 * AtomicAccess::release_store(&tl->_vthread, true);
3971 * return;
3972 * }
3973 * AtomicAccess::release_store(&tl->_vthread, false);
3974 */
3975 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3976 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3977
3978 Node* input_memory_state = reset_memory();
3979 set_all_memory(input_memory_state);
3980
3981 // The most significant bit of the u2 is used to denote thread exclusion
3982 Node* excluded_mask = _gvn.intcon(1 << 15);
3983 // The epoch generation is the range [1-32767]
3984 Node* epoch_mask = _gvn.intcon(32767);
3985
3986 Node* const carrierThread = generate_current_thread(jt);
3987 // If thread != carrierThread, this is a virtual thread.
3988 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3989 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3990 IfNode* iff_thread_not_equal_carrierThread =
3991 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3992
3993 Node* vthread_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3994
3995 // False branch, is carrierThread.
3996 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3997 // Store release
3998 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3999
4000 set_all_memory(input_memory_state);
4001
4002 // True branch, is virtual thread.
4003 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
4004 set_control(thread_not_equal_carrierThread);
4005
4006 // Load the raw epoch value from the vthread.
4007 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
4008 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
4009 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
4010
4011 // Mask off the excluded information from the epoch.
4012 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, excluded_mask));
4013
4014 // Load the tid field from the thread.
4015 Node* tid = load_field_from_object(thread, "tid", "J");
4016
4017 // Store the vthread tid to the jfr thread local.
4018 Node* thread_id_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
4019 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
4020
4021 // Branch is_excluded to conditionalize updating the epoch .
4022 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, excluded_mask));
4023 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
4024 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
4025
4026 // True branch, vthread is excluded, no need to write epoch info.
4027 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
4028 set_control(excluded);
4029 Node* vthread_is_excluded = _gvn.intcon(1);
4030
4031 // False branch, vthread is included, update epoch info.
4032 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
4033 set_control(included);
4034 Node* vthread_is_included = _gvn.intcon(0);
4035
4036 // Get epoch value.
4037 Node* epoch = _gvn.transform(new AndINode(epoch_raw, epoch_mask));
4038
4039 // Store the vthread epoch to the jfr thread local.
4040 Node* vthread_epoch_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
4041 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
4042
4043 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
4044 record_for_igvn(excluded_rgn);
4045 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
4046 record_for_igvn(excluded_mem);
4047 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
4048 record_for_igvn(exclusion);
4049
4050 // Merge the excluded control and memory.
4051 excluded_rgn->init_req(_true_path, excluded);
4052 excluded_rgn->init_req(_false_path, included);
4053 excluded_mem->init_req(_true_path, tid_memory);
4054 excluded_mem->init_req(_false_path, included_memory);
4055 exclusion->init_req(_true_path, vthread_is_excluded);
4056 exclusion->init_req(_false_path, vthread_is_included);
4057
4058 // Set intermediate state.
4059 set_control(_gvn.transform(excluded_rgn));
4060 set_all_memory(excluded_mem);
4061
4062 // Store the vthread exclusion state to the jfr thread local.
4063 Node* thread_local_excluded_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
4064 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
4065
4066 // Store release
4067 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
4068
4069 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
4070 record_for_igvn(thread_compare_rgn);
4071 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
4072 record_for_igvn(thread_compare_mem);
4073 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
4074 record_for_igvn(vthread);
4075
4076 // Merge the thread_compare control and memory.
4077 thread_compare_rgn->init_req(_true_path, control());
4078 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
4079 thread_compare_mem->init_req(_true_path, vthread_true_memory);
4080 thread_compare_mem->init_req(_false_path, vthread_false_memory);
4081
4082 // Set output state.
4083 set_control(_gvn.transform(thread_compare_rgn));
4084 set_all_memory(_gvn.transform(thread_compare_mem));
4085 }
4086
4087 #endif // JFR_HAVE_INTRINSICS
4088
4089 //------------------------inline_native_currentCarrierThread------------------
4090 bool LibraryCallKit::inline_native_currentCarrierThread() {
4091 Node* junk = nullptr;
4092 set_result(generate_current_thread(junk));
4093 return true;
4094 }
4095
4096 //------------------------inline_native_currentThread------------------
4097 bool LibraryCallKit::inline_native_currentThread() {
4098 Node* junk = nullptr;
4099 set_result(generate_virtual_thread(junk));
4100 return true;
4101 }
4102
4103 //------------------------inline_native_setVthread------------------
4104 bool LibraryCallKit::inline_native_setCurrentThread() {
4105 assert(C->method()->changes_current_thread(),
4106 "method changes current Thread but is not annotated ChangesCurrentThread");
4107 Node* arr = argument(1);
4108 Node* thread = _gvn.transform(new ThreadLocalNode());
4109 Node* p = off_heap_plus_addr(thread, in_bytes(JavaThread::vthread_offset()));
4110 Node* thread_obj_handle
4111 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
4112 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
4113 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
4114
4115 // Change the _monitor_owner_id of the JavaThread
4116 Node* tid = load_field_from_object(arr, "tid", "J");
4117 Node* monitor_owner_id_offset = off_heap_plus_addr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
4118 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
4119
4120 JFR_ONLY(extend_setCurrentThread(thread, arr);)
4121 return true;
4122 }
4123
4124 const Type* LibraryCallKit::scopedValueCache_type() {
4125 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
4126 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
4127 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS, /* stable= */ false, /* flat= */ false, /* not_flat= */ true, /* not_null_free= */ true, true);
4128
4129 // Because we create the scopedValue cache lazily we have to make the
4130 // type of the result BotPTR.
4131 bool xk = etype->klass_is_exact();
4132 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, TypeAryPtr::Offset(0));
4133 return objects_type;
4134 }
4135
4136 Node* LibraryCallKit::scopedValueCache_helper() {
4137 Node* thread = _gvn.transform(new ThreadLocalNode());
4138 Node* p = off_heap_plus_addr(thread, in_bytes(JavaThread::scopedValueCache_offset()));
4139 // We cannot use immutable_memory() because we might flip onto a
4140 // different carrier thread, at which point we'll need to use that
4141 // carrier thread's cache.
4142 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
4143 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
4144 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
4145 }
4146
4147 //------------------------inline_native_scopedValueCache------------------
4148 bool LibraryCallKit::inline_native_scopedValueCache() {
4149 Node* cache_obj_handle = scopedValueCache_helper();
4150 const Type* objects_type = scopedValueCache_type();
4151 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
4152
4153 return true;
4154 }
4155
4156 //------------------------inline_native_setScopedValueCache------------------
4157 bool LibraryCallKit::inline_native_setScopedValueCache() {
4158 Node* arr = argument(0);
4159 Node* cache_obj_handle = scopedValueCache_helper();
4160 const Type* objects_type = scopedValueCache_type();
4161
4162 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
4163 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
4164
4165 return true;
4166 }
4167
4168 //------------------------inline_native_Continuation_pin and unpin-----------
4169
4170 // Shared implementation routine for both pin and unpin.
4171 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
4172 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
4173
4174 // Save input memory.
4175 Node* input_memory_state = reset_memory();
4176 set_all_memory(input_memory_state);
4177
4178 // TLS
4179 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
4180 Node* last_continuation_offset = off_heap_plus_addr(tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
4181 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
4182
4183 // Null check the last continuation object.
4184 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
4185 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
4186 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
4187
4188 // False path, last continuation is null.
4189 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
4190
4191 // True path, last continuation is not null.
4192 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
4193
4194 set_control(continuation_is_not_null);
4195
4196 // Load the pin count from the last continuation.
4197 Node* pin_count_offset = off_heap_plus_addr(last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
4198 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
4199
4200 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
4201 Node* pin_count_rhs;
4202 if (unpin) {
4203 pin_count_rhs = _gvn.intcon(0);
4204 } else {
4205 pin_count_rhs = _gvn.intcon(UINT32_MAX);
4206 }
4207 Node* pin_count_cmp = _gvn.transform(new CmpUNode(pin_count, pin_count_rhs));
4208 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
4209 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
4210
4211 // True branch, pin count over/underflow.
4212 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
4213 {
4214 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
4215 // which will throw IllegalStateException for pin count over/underflow.
4216 // No memory changed so far - we can use memory create by reset_memory()
4217 // at the beginning of this intrinsic. No need to call reset_memory() again.
4218 PreserveJVMState pjvms(this);
4219 set_control(pin_count_over_underflow);
4220 uncommon_trap(Deoptimization::Reason_intrinsic,
4221 Deoptimization::Action_none);
4222 assert(stopped(), "invariant");
4223 }
4224
4225 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
4226 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
4227 set_control(valid_pin_count);
4228
4229 Node* next_pin_count;
4230 if (unpin) {
4231 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
4232 } else {
4233 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
4234 }
4235
4236 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
4237
4238 // Result of top level CFG and Memory.
4239 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
4240 record_for_igvn(result_rgn);
4241 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
4242 record_for_igvn(result_mem);
4243
4244 result_rgn->init_req(_true_path, valid_pin_count);
4245 result_rgn->init_req(_false_path, continuation_is_null);
4246 result_mem->init_req(_true_path, reset_memory());
4247 result_mem->init_req(_false_path, input_memory_state);
4248
4249 // Set output state.
4250 set_control(_gvn.transform(result_rgn));
4251 set_all_memory(_gvn.transform(result_mem));
4252
4253 return true;
4254 }
4255
4256 //---------------------------load_mirror_from_klass----------------------------
4257 // Given a klass oop, load its java mirror (a java.lang.Class oop).
4258 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
4259 Node* p = off_heap_plus_addr(klass, in_bytes(Klass::java_mirror_offset()));
4260 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4261 // mirror = ((OopHandle)mirror)->resolve();
4262 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
4263 }
4264
4265 //-----------------------load_klass_from_mirror_common-------------------------
4266 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
4267 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
4268 // and branch to the given path on the region.
4269 // If never_see_null, take an uncommon trap on null, so we can optimistically
4270 // compile for the non-null case.
4271 // If the region is null, force never_see_null = true.
4272 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
4273 bool never_see_null,
4274 RegionNode* region,
4275 int null_path,
4276 int offset) {
4277 if (region == nullptr) never_see_null = true;
4278 Node* p = basic_plus_adr(mirror, offset);
4279 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4280 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
4281 Node* null_ctl = top();
4282 kls = null_check_oop(kls, &null_ctl, never_see_null);
4283 if (region != nullptr) {
4284 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
4285 region->init_req(null_path, null_ctl);
4286 } else {
4287 assert(null_ctl == top(), "no loose ends");
4288 }
4289 return kls;
4290 }
4291
4292 //--------------------(inline_native_Class_query helpers)---------------------
4293 // Use this for JVM_ACC_INTERFACE.
4294 // Fall through if (mods & mask) == bits, take the guard otherwise.
4295 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
4296 ByteSize offset, const Type* type, BasicType bt) {
4297 // Branch around if the given klass has the given modifier bit set.
4298 // Like generate_guard, adds a new path onto the region.
4299 Node* modp = off_heap_plus_addr(kls, in_bytes(offset));
4300 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
4301 Node* mask = intcon(modifier_mask);
4302 Node* bits = intcon(modifier_bits);
4303 Node* mbit = _gvn.transform(new AndINode(mods, mask));
4304 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
4305 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
4306 return generate_fair_guard(bol, region);
4307 }
4308
4309 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
4310 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
4311 InstanceKlass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
4312 }
4313
4314 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
4315 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
4316 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
4317 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
4318 }
4319
4320 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
4321 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
4322 }
4323
4324 //-------------------------inline_native_Class_query-------------------
4325 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
4326 const Type* return_type = TypeInt::BOOL;
4327 Node* prim_return_value = top(); // what happens if it's a primitive class?
4328 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4329 bool expect_prim = false; // most of these guys expect to work on refs
4330
4331 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
4332
4333 Node* mirror = argument(0);
4334 Node* obj = top();
4335
4336 switch (id) {
4337 case vmIntrinsics::_isInstance:
4338 // nothing is an instance of a primitive type
4339 prim_return_value = intcon(0);
4340 obj = argument(1);
4341 break;
4342 case vmIntrinsics::_isHidden:
4343 prim_return_value = intcon(0);
4344 break;
4345 case vmIntrinsics::_getSuperclass:
4346 prim_return_value = null();
4347 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
4348 break;
4349 default:
4350 fatal_unexpected_iid(id);
4351 break;
4352 }
4353
4354 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4355 if (mirror_con == nullptr) return false; // cannot happen?
4356
4357 #ifndef PRODUCT
4358 if (C->print_intrinsics() || C->print_inlining()) {
4359 ciType* k = mirror_con->java_mirror_type();
4360 if (k) {
4361 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4362 k->print_name();
4363 tty->cr();
4364 }
4365 }
4366 #endif
4367
4368 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4369 RegionNode* region = new RegionNode(PATH_LIMIT);
4370 record_for_igvn(region);
4371 PhiNode* phi = new PhiNode(region, return_type);
4372
4373 // The mirror will never be null of Reflection.getClassAccessFlags, however
4374 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4375 // if it is. See bug 4774291.
4376
4377 // For Reflection.getClassAccessFlags(), the null check occurs in
4378 // the wrong place; see inline_unsafe_access(), above, for a similar
4379 // situation.
4380 mirror = null_check(mirror);
4381 // If mirror or obj is dead, only null-path is taken.
4382 if (stopped()) return true;
4383
4384 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4385
4386 // Now load the mirror's klass metaobject, and null-check it.
4387 // Side-effects region with the control path if the klass is null.
4388 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4389 // If kls is null, we have a primitive mirror.
4390 phi->init_req(_prim_path, prim_return_value);
4391 if (stopped()) { set_result(region, phi); return true; }
4392 bool safe_for_replace = (region->in(_prim_path) == top());
4393
4394 Node* p; // handy temp
4395 Node* null_ctl;
4396
4397 // Now that we have the non-null klass, we can perform the real query.
4398 // For constant classes, the query will constant-fold in LoadNode::Value.
4399 Node* query_value = top();
4400 switch (id) {
4401 case vmIntrinsics::_isInstance:
4402 // nothing is an instance of a primitive type
4403 query_value = gen_instanceof(obj, kls, safe_for_replace);
4404 break;
4405
4406 case vmIntrinsics::_isHidden:
4407 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4408 if (generate_hidden_class_guard(kls, region) != nullptr)
4409 // A guard was added. If the guard is taken, it was an hidden class.
4410 phi->add_req(intcon(1));
4411 // If we fall through, it's a plain class.
4412 query_value = intcon(0);
4413 break;
4414
4415
4416 case vmIntrinsics::_getSuperclass:
4417 // The rules here are somewhat unfortunate, but we can still do better
4418 // with random logic than with a JNI call.
4419 // Interfaces store null or Object as _super, but must report null.
4420 // Arrays store an intermediate super as _super, but must report Object.
4421 // Other types can report the actual _super.
4422 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4423 if (generate_array_guard(kls, region) != nullptr) {
4424 // A guard was added. If the guard is taken, it was an array.
4425 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4426 }
4427 // Check for interface after array since this checks AccessFlags offset into InstanceKlass.
4428 // In other words, we are accessing subtype-specific information, so we need to determine the subtype first.
4429 if (generate_interface_guard(kls, region) != nullptr) {
4430 // A guard was added. If the guard is taken, it was an interface.
4431 phi->add_req(null());
4432 }
4433 // If we fall through, it's a plain class. Get its _super.
4434 if (!stopped()) {
4435 p = basic_plus_adr(top(), kls, in_bytes(Klass::super_offset()));
4436 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4437 null_ctl = top();
4438 kls = null_check_oop(kls, &null_ctl);
4439 if (null_ctl != top()) {
4440 // If the guard is taken, Object.superClass is null (both klass and mirror).
4441 region->add_req(null_ctl);
4442 phi ->add_req(null());
4443 }
4444 if (!stopped()) {
4445 query_value = load_mirror_from_klass(kls);
4446 }
4447 }
4448 break;
4449
4450 default:
4451 fatal_unexpected_iid(id);
4452 break;
4453 }
4454
4455 // Fall-through is the normal case of a query to a real class.
4456 phi->init_req(1, query_value);
4457 region->init_req(1, control());
4458
4459 C->set_has_split_ifs(true); // Has chance for split-if optimization
4460 set_result(region, phi);
4461 return true;
4462 }
4463
4464
4465 //-------------------------inline_Class_cast-------------------
4466 bool LibraryCallKit::inline_Class_cast() {
4467 Node* mirror = argument(0); // Class
4468 Node* obj = argument(1);
4469 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4470 if (mirror_con == nullptr) {
4471 return false; // dead path (mirror->is_top()).
4472 }
4473 if (obj == nullptr || obj->is_top()) {
4474 return false; // dead path
4475 }
4476 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4477
4478 // First, see if Class.cast() can be folded statically.
4479 // java_mirror_type() returns non-null for compile-time Class constants.
4480 ciType* tm = mirror_con->java_mirror_type();
4481 if (tm != nullptr && tm->is_klass() &&
4482 tp != nullptr) {
4483 if (!tp->is_loaded()) {
4484 // Don't use intrinsic when class is not loaded.
4485 return false;
4486 } else {
4487 const TypeKlassPtr* tklass = TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces);
4488 int static_res = C->static_subtype_check(tklass, tp->as_klass_type());
4489 if (static_res == Compile::SSC_always_true) {
4490 // isInstance() is true - fold the code.
4491 set_result(obj);
4492 return true;
4493 } else if (static_res == Compile::SSC_always_false) {
4494 // Don't use intrinsic, have to throw ClassCastException.
4495 // If the reference is null, the non-intrinsic bytecode will
4496 // be optimized appropriately.
4497 return false;
4498 }
4499 }
4500 }
4501
4502 // Bailout intrinsic and do normal inlining if exception path is frequent.
4503 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4504 return false;
4505 }
4506
4507 // Generate dynamic checks.
4508 // Class.cast() is java implementation of _checkcast bytecode.
4509 // Do checkcast (Parse::do_checkcast()) optimizations here.
4510
4511 mirror = null_check(mirror);
4512 // If mirror is dead, only null-path is taken.
4513 if (stopped()) {
4514 return true;
4515 }
4516
4517 // Not-subtype or the mirror's klass ptr is nullptr (in case it is a primitive).
4518 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
4519 RegionNode* region = new RegionNode(PATH_LIMIT);
4520 record_for_igvn(region);
4521
4522 // Now load the mirror's klass metaobject, and null-check it.
4523 // If kls is null, we have a primitive mirror and
4524 // nothing is an instance of a primitive type.
4525 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4526
4527 Node* res = top();
4528 Node* io = i_o();
4529 Node* mem = merged_memory();
4530 SafePointNode* new_cast_failure_map = nullptr;
4531
4532 if (!stopped()) {
4533
4534 Node* bad_type_ctrl = top();
4535 // Do checkcast optimizations.
4536 res = gen_checkcast(obj, kls, &bad_type_ctrl, &new_cast_failure_map);
4537 region->init_req(_bad_type_path, bad_type_ctrl);
4538 }
4539 if (region->in(_prim_path) != top() ||
4540 region->in(_bad_type_path) != top() ||
4541 region->in(_npe_path) != top()) {
4542 // Let Interpreter throw ClassCastException.
4543 PreserveJVMState pjvms(this);
4544 if (new_cast_failure_map != nullptr) {
4545 // The current map on the success path could have been modified. Use the dedicated failure path map.
4546 set_map(new_cast_failure_map);
4547 }
4548 set_control(_gvn.transform(region));
4549 // Set IO and memory because gen_checkcast may override them when buffering inline types
4550 set_i_o(io);
4551 set_all_memory(mem);
4552 uncommon_trap(Deoptimization::Reason_intrinsic,
4553 Deoptimization::Action_maybe_recompile);
4554 }
4555 if (!stopped()) {
4556 set_result(res);
4557 }
4558 return true;
4559 }
4560
4561
4562 //--------------------------inline_native_subtype_check------------------------
4563 // This intrinsic takes the JNI calls out of the heart of
4564 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4565 bool LibraryCallKit::inline_native_subtype_check() {
4566 // Pull both arguments off the stack.
4567 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4568 args[0] = argument(0);
4569 args[1] = argument(1);
4570 Node* klasses[2]; // corresponding Klasses: superk, subk
4571 klasses[0] = klasses[1] = top();
4572
4573 enum {
4574 // A full decision tree on {superc is prim, subc is prim}:
4575 _prim_0_path = 1, // {P,N} => false
4576 // {P,P} & superc!=subc => false
4577 _prim_same_path, // {P,P} & superc==subc => true
4578 _prim_1_path, // {N,P} => false
4579 _ref_subtype_path, // {N,N} & subtype check wins => true
4580 _both_ref_path, // {N,N} & subtype check loses => false
4581 PATH_LIMIT
4582 };
4583
4584 RegionNode* region = new RegionNode(PATH_LIMIT);
4585 RegionNode* prim_region = new RegionNode(2);
4586 Node* phi = new PhiNode(region, TypeInt::BOOL);
4587 record_for_igvn(region);
4588 record_for_igvn(prim_region);
4589
4590 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4591 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4592 int class_klass_offset = java_lang_Class::klass_offset();
4593
4594 // First null-check both mirrors and load each mirror's klass metaobject.
4595 int which_arg;
4596 for (which_arg = 0; which_arg <= 1; which_arg++) {
4597 Node* arg = args[which_arg];
4598 arg = null_check(arg);
4599 if (stopped()) break;
4600 args[which_arg] = arg;
4601
4602 Node* p = basic_plus_adr(arg, class_klass_offset);
4603 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4604 klasses[which_arg] = _gvn.transform(kls);
4605 }
4606
4607 // Having loaded both klasses, test each for null.
4608 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4609 for (which_arg = 0; which_arg <= 1; which_arg++) {
4610 Node* kls = klasses[which_arg];
4611 Node* null_ctl = top();
4612 kls = null_check_oop(kls, &null_ctl, never_see_null);
4613 if (which_arg == 0) {
4614 prim_region->init_req(1, null_ctl);
4615 } else {
4616 region->init_req(_prim_1_path, null_ctl);
4617 }
4618 if (stopped()) break;
4619 klasses[which_arg] = kls;
4620 }
4621
4622 if (!stopped()) {
4623 // now we have two reference types, in klasses[0..1]
4624 Node* subk = klasses[1]; // the argument to isAssignableFrom
4625 Node* superk = klasses[0]; // the receiver
4626 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4627 region->set_req(_ref_subtype_path, control());
4628 }
4629
4630 // If both operands are primitive (both klasses null), then
4631 // we must return true when they are identical primitives.
4632 // It is convenient to test this after the first null klass check.
4633 // This path is also used if superc is a value mirror.
4634 set_control(_gvn.transform(prim_region));
4635 if (!stopped()) {
4636 // Since superc is primitive, make a guard for the superc==subc case.
4637 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4638 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4639 generate_fair_guard(bol_eq, region);
4640 if (region->req() == PATH_LIMIT+1) {
4641 // A guard was added. If the added guard is taken, superc==subc.
4642 region->swap_edges(PATH_LIMIT, _prim_same_path);
4643 region->del_req(PATH_LIMIT);
4644 }
4645 region->set_req(_prim_0_path, control()); // Not equal after all.
4646 }
4647
4648 // these are the only paths that produce 'true':
4649 phi->set_req(_prim_same_path, intcon(1));
4650 phi->set_req(_ref_subtype_path, intcon(1));
4651
4652 // pull together the cases:
4653 assert(region->req() == PATH_LIMIT, "sane region");
4654 for (uint i = 1; i < region->req(); i++) {
4655 Node* ctl = region->in(i);
4656 if (ctl == nullptr || ctl == top()) {
4657 region->set_req(i, top());
4658 phi ->set_req(i, top());
4659 } else if (phi->in(i) == nullptr) {
4660 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4661 }
4662 }
4663
4664 set_control(_gvn.transform(region));
4665 set_result(_gvn.transform(phi));
4666 return true;
4667 }
4668
4669 //---------------------generate_array_guard_common------------------------
4670 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind, Node** obj) {
4671
4672 if (stopped()) {
4673 return nullptr;
4674 }
4675
4676 // Like generate_guard, adds a new path onto the region.
4677 jint layout_con = 0;
4678 Node* layout_val = get_layout_helper(kls, layout_con);
4679 if (layout_val == nullptr) {
4680 bool query = 0;
4681 switch(kind) {
4682 case RefArray: query = Klass::layout_helper_is_refArray(layout_con); break;
4683 case NonRefArray: query = !Klass::layout_helper_is_refArray(layout_con); break;
4684 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
4685 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
4686 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
4687 default:
4688 ShouldNotReachHere();
4689 }
4690 if (!query) {
4691 return nullptr; // never a branch
4692 } else { // always a branch
4693 Node* always_branch = control();
4694 if (region != nullptr)
4695 region->add_req(always_branch);
4696 set_control(top());
4697 return always_branch;
4698 }
4699 }
4700 unsigned int value = 0;
4701 BoolTest::mask btest = BoolTest::illegal;
4702 switch(kind) {
4703 case RefArray:
4704 case NonRefArray: {
4705 value = Klass::_lh_array_tag_ref_value;
4706 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4707 btest = (kind == RefArray) ? BoolTest::eq : BoolTest::ne;
4708 break;
4709 }
4710 case TypeArray: {
4711 value = Klass::_lh_array_tag_type_value;
4712 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4713 btest = BoolTest::eq;
4714 break;
4715 }
4716 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
4717 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
4718 default:
4719 ShouldNotReachHere();
4720 }
4721 // Now test the correct condition.
4722 jint nval = (jint)value;
4723 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4724 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4725 Node* ctrl = generate_fair_guard(bol, region);
4726 Node* is_array_ctrl = kind == NonArray ? control() : ctrl;
4727 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4728 // Keep track of the fact that 'obj' is an array to prevent
4729 // array specific accesses from floating above the guard.
4730 *obj = _gvn.transform(new CheckCastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4731 }
4732 return ctrl;
4733 }
4734
4735 // public static native Object[] ValueClass::newNullRestrictedAtomicArray(Class<?> componentType, int length, Object initVal);
4736 // public static native Object[] ValueClass::newNullRestrictedNonAtomicArray(Class<?> componentType, int length, Object initVal);
4737 // public static native Object[] ValueClass::newNullableAtomicArray(Class<?> componentType, int length);
4738 bool LibraryCallKit::inline_newArray(bool null_free, bool atomic) {
4739 assert(null_free || atomic, "nullable implies atomic");
4740 Node* componentType = argument(0);
4741 Node* length = argument(1);
4742 Node* init_val = null_free ? argument(2) : nullptr;
4743
4744 const TypeInstPtr* tp = _gvn.type(componentType)->isa_instptr();
4745 if (tp != nullptr) {
4746 ciInstanceKlass* ik = tp->instance_klass();
4747 if (ik == C->env()->Class_klass()) {
4748 ciType* t = tp->java_mirror_type();
4749 if (t != nullptr && t->is_inlinetype()) {
4750
4751 ciArrayKlass* array_klass = ciArrayKlass::make(t, null_free, atomic, true);
4752 assert(array_klass->is_elem_null_free() == null_free, "inconsistency");
4753
4754 // TOOD 8350865 ZGC needs card marks on initializing oop stores
4755 if (UseZGC && null_free && !array_klass->is_flat_array_klass()) {
4756 return false;
4757 }
4758
4759 if (array_klass->is_loaded() && array_klass->element_klass()->as_inline_klass()->is_initialized()) {
4760 const TypeAryKlassPtr* array_klass_type = TypeAryKlassPtr::make(array_klass, Type::trust_interfaces);
4761 if (null_free) {
4762 if (init_val->is_InlineType()) {
4763 if (array_klass_type->is_flat() && init_val->as_InlineType()->is_all_zero(&gvn(), /* flat */ true)) {
4764 // Zeroing is enough because the init value is the all-zero value
4765 init_val = nullptr;
4766 } else {
4767 init_val = init_val->as_InlineType()->buffer(this);
4768 }
4769 }
4770 // TODO 8350865 Should we add a check of the init_val type (maybe in debug only + halt)?
4771 // If we insert a checkcast here, we can be sure that init_val is an InlineTypeNode, so
4772 // when we folded a field load from an allocation (e.g. during escape analysis), we can
4773 // remove the check init_val->is_InlineType().
4774 }
4775 Node* obj = new_array(makecon(array_klass_type), length, 0, nullptr, false, init_val);
4776 const TypeAryPtr* arytype = gvn().type(obj)->is_aryptr();
4777 assert(arytype->is_null_free() == null_free, "inconsistency");
4778 assert(arytype->is_not_null_free() == !null_free, "inconsistency");
4779 set_result(obj);
4780 return true;
4781 }
4782 }
4783 }
4784 }
4785 return false;
4786 }
4787
4788 // public static native boolean ValueClass::isFlatArray(Object array);
4789 // public static native boolean ValueClass::isNullRestrictedArray(Object array);
4790 // public static native boolean ValueClass::isAtomicArray(Object array);
4791 bool LibraryCallKit::inline_getArrayProperties(ArrayPropertiesCheck check) {
4792 Node* array = argument(0);
4793
4794 Node* bol;
4795 switch(check) {
4796 case IsFlat:
4797 // TODO 8350865 Use the object version here instead of loading the klass
4798 // 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
4799 bol = flat_array_test(load_object_klass(array));
4800 break;
4801 case IsNullRestricted:
4802 bol = null_free_array_test(array);
4803 break;
4804 case IsAtomic:
4805 // TODO 8350865 Implement this. It's a bit more complicated, see conditions in JVM_IsAtomicArray
4806 // Enable TestIntrinsics::test87/88 once this is implemented
4807 // bol = null_free_atomic_array_test
4808 return false;
4809 default:
4810 ShouldNotReachHere();
4811 }
4812
4813 Node* res = gvn().transform(new CMoveINode(bol, intcon(0), intcon(1), TypeInt::BOOL));
4814 set_result(res);
4815 return true;
4816 }
4817
4818 // Load the default refined array klass from an ObjArrayKlass. This relies on the first entry in the
4819 // '_next_refined_array_klass' linked list being the default (see ObjArrayKlass::klass_with_properties).
4820 Node* LibraryCallKit::load_default_refined_array_klass(Node* klass_node, bool type_array_guard) {
4821 RegionNode* region = new RegionNode(2);
4822 Node* phi = new PhiNode(region, TypeInstKlassPtr::OBJECT_OR_NULL);
4823
4824 if (type_array_guard) {
4825 generate_typeArray_guard(klass_node, region);
4826 if (region->req() == 3) {
4827 phi->add_req(klass_node);
4828 }
4829 }
4830 Node* adr_refined_klass = basic_plus_adr(top(), klass_node, in_bytes(ObjArrayKlass::next_refined_array_klass_offset()));
4831 Node* refined_klass = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), adr_refined_klass, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4832
4833 // Can be null if not initialized yet, just deopt
4834 Node* null_ctl = top();
4835 refined_klass = null_check_oop(refined_klass, &null_ctl, /* never_see_null= */ true);
4836
4837 region->init_req(1, control());
4838 phi->init_req(1, refined_klass);
4839
4840 set_control(_gvn.transform(region));
4841 return _gvn.transform(phi);
4842 }
4843
4844 // Load the non-refined array klass from an ObjArrayKlass.
4845 Node* LibraryCallKit::load_non_refined_array_klass(Node* klass_node) {
4846 const TypeAryKlassPtr* ary_klass_ptr = _gvn.type(klass_node)->isa_aryklassptr();
4847 if (ary_klass_ptr != nullptr && ary_klass_ptr->klass_is_exact()) {
4848 return _gvn.makecon(ary_klass_ptr->cast_to_refined_array_klass_ptr(false));
4849 }
4850
4851 RegionNode* region = new RegionNode(2);
4852 Node* phi = new PhiNode(region, TypeInstKlassPtr::OBJECT);
4853
4854 generate_typeArray_guard(klass_node, region);
4855 if (region->req() == 3) {
4856 phi->add_req(klass_node);
4857 }
4858 Node* super_adr = basic_plus_adr(top(), klass_node, in_bytes(Klass::super_offset()));
4859 Node* super_klass = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), super_adr, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT));
4860
4861 region->init_req(1, control());
4862 phi->init_req(1, super_klass);
4863
4864 set_control(_gvn.transform(region));
4865 return _gvn.transform(phi);
4866 }
4867
4868 //-----------------------inline_native_newArray--------------------------
4869 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
4870 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4871 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4872 Node* mirror;
4873 Node* count_val;
4874 if (uninitialized) {
4875 null_check_receiver();
4876 mirror = argument(1);
4877 count_val = argument(2);
4878 } else {
4879 mirror = argument(0);
4880 count_val = argument(1);
4881 }
4882
4883 mirror = null_check(mirror);
4884 // If mirror or obj is dead, only null-path is taken.
4885 if (stopped()) return true;
4886
4887 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4888 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4889 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4890 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4891 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4892
4893 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4894 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4895 result_reg, _slow_path);
4896 Node* normal_ctl = control();
4897 Node* no_array_ctl = result_reg->in(_slow_path);
4898
4899 // Generate code for the slow case. We make a call to newArray().
4900 set_control(no_array_ctl);
4901 if (!stopped()) {
4902 // Either the input type is void.class, or else the
4903 // array klass has not yet been cached. Either the
4904 // ensuing call will throw an exception, or else it
4905 // will cache the array klass for next time.
4906 PreserveJVMState pjvms(this);
4907 CallJavaNode* slow_call = nullptr;
4908 if (uninitialized) {
4909 // Generate optimized virtual call (holder class 'Unsafe' is final)
4910 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4911 } else {
4912 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4913 }
4914 Node* slow_result = set_results_for_java_call(slow_call);
4915 // this->control() comes from set_results_for_java_call
4916 result_reg->set_req(_slow_path, control());
4917 result_val->set_req(_slow_path, slow_result);
4918 result_io ->set_req(_slow_path, i_o());
4919 result_mem->set_req(_slow_path, reset_memory());
4920 }
4921
4922 set_control(normal_ctl);
4923 if (!stopped()) {
4924 // Normal case: The array type has been cached in the java.lang.Class.
4925 // The following call works fine even if the array type is polymorphic.
4926 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4927
4928 klass_node = load_default_refined_array_klass(klass_node);
4929
4930 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4931 result_reg->init_req(_normal_path, control());
4932 result_val->init_req(_normal_path, obj);
4933 result_io ->init_req(_normal_path, i_o());
4934 result_mem->init_req(_normal_path, reset_memory());
4935
4936 if (uninitialized) {
4937 // Mark the allocation so that zeroing is skipped
4938 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4939 alloc->maybe_set_complete(&_gvn);
4940 }
4941 }
4942
4943 // Return the combined state.
4944 set_i_o( _gvn.transform(result_io) );
4945 set_all_memory( _gvn.transform(result_mem));
4946
4947 C->set_has_split_ifs(true); // Has chance for split-if optimization
4948 set_result(result_reg, result_val);
4949 return true;
4950 }
4951
4952 //----------------------inline_native_getLength--------------------------
4953 // public static native int java.lang.reflect.Array.getLength(Object array);
4954 bool LibraryCallKit::inline_native_getLength() {
4955 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4956
4957 Node* array = null_check(argument(0));
4958 // If array is dead, only null-path is taken.
4959 if (stopped()) return true;
4960
4961 // Deoptimize if it is a non-array.
4962 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
4963
4964 if (non_array != nullptr) {
4965 PreserveJVMState pjvms(this);
4966 set_control(non_array);
4967 uncommon_trap(Deoptimization::Reason_intrinsic,
4968 Deoptimization::Action_maybe_recompile);
4969 }
4970
4971 // If control is dead, only non-array-path is taken.
4972 if (stopped()) return true;
4973
4974 // The works fine even if the array type is polymorphic.
4975 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4976 Node* result = load_array_length(array);
4977
4978 C->set_has_split_ifs(true); // Has chance for split-if optimization
4979 set_result(result);
4980 return true;
4981 }
4982
4983 //------------------------inline_array_copyOf----------------------------
4984 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4985 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4986 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4987 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4988
4989 // Get the arguments.
4990 Node* original = argument(0);
4991 Node* start = is_copyOfRange? argument(1): intcon(0);
4992 Node* end = is_copyOfRange? argument(2): argument(1);
4993 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4994
4995 Node* newcopy = nullptr;
4996
4997 // Set the original stack and the reexecute bit for the interpreter to reexecute
4998 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4999 { PreserveReexecuteState preexecs(this);
5000 jvms()->set_should_reexecute(true);
5001
5002 array_type_mirror = null_check(array_type_mirror);
5003 original = null_check(original);
5004
5005 // Check if a null path was taken unconditionally.
5006 if (stopped()) return true;
5007
5008 Node* orig_length = load_array_length(original);
5009
5010 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
5011 klass_node = null_check(klass_node);
5012
5013 RegionNode* bailout = new RegionNode(1);
5014 record_for_igvn(bailout);
5015
5016 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
5017 // Bail out if that is so.
5018 // Inline type array may have object field that would require a
5019 // write barrier. Conservatively, go to slow path.
5020 // TODO 8251971: Optimize for the case when flat src/dst are later found
5021 // to not contain oops (i.e., move this check to the macro expansion phase).
5022 // TODO 8382226: Revisit for flat abstract value class arrays
5023 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5024 const TypeAryPtr* orig_t = _gvn.type(original)->isa_aryptr();
5025 const TypeKlassPtr* tklass = _gvn.type(klass_node)->is_klassptr();
5026 bool exclude_flat = UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing) &&
5027 // Can src array be flat and contain oops?
5028 (orig_t == nullptr || (!orig_t->is_not_flat() && (!orig_t->is_flat() || orig_t->elem()->inline_klass()->contains_oops()))) &&
5029 // Can dest array be flat and contain oops?
5030 tklass->can_be_inline_array() && (!tklass->is_flat() || tklass->is_aryklassptr()->elem()->is_instklassptr()->instance_klass()->as_inline_klass()->contains_oops());
5031 Node* not_objArray = exclude_flat ? generate_non_refArray_guard(klass_node, bailout) : generate_typeArray_guard(klass_node, bailout);
5032
5033 Node* refined_klass_node = load_default_refined_array_klass(klass_node, /* type_array_guard= */ false);
5034
5035 if (not_objArray != nullptr) {
5036 // Improve the klass node's type from the new optimistic assumption:
5037 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
5038 bool not_flat = !UseArrayFlattening;
5039 bool not_null_free = !Arguments::is_valhalla_enabled();
5040 const Type* akls = TypeAryKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0), Type::trust_interfaces, not_flat, not_null_free, false, false, not_flat, true);
5041 Node* cast = new CastPPNode(control(), refined_klass_node, akls);
5042 refined_klass_node = _gvn.transform(cast);
5043 }
5044
5045 // Bail out if either start or end is negative.
5046 generate_negative_guard(start, bailout, &start);
5047 generate_negative_guard(end, bailout, &end);
5048
5049 Node* length = end;
5050 if (_gvn.type(start) != TypeInt::ZERO) {
5051 length = _gvn.transform(new SubINode(end, start));
5052 }
5053
5054 // Bail out if length is negative (i.e., if start > end).
5055 // Without this the new_array would throw
5056 // NegativeArraySizeException but IllegalArgumentException is what
5057 // should be thrown
5058 generate_negative_guard(length, bailout, &length);
5059
5060 // Handle inline type arrays
5061 bool can_validate = !too_many_traps(Deoptimization::Reason_class_check);
5062 if (!stopped()) {
5063 // TODO 8251971
5064 if (!orig_t->is_null_free()) {
5065 // Not statically known to be null free, add a check
5066 generate_fair_guard(null_free_array_test(original), bailout);
5067 }
5068 orig_t = _gvn.type(original)->isa_aryptr();
5069 if (orig_t != nullptr && orig_t->is_flat()) {
5070 // Src is flat, check that dest is flat as well
5071 if (exclude_flat) {
5072 // Dest can't be flat, bail out
5073 bailout->add_req(control());
5074 set_control(top());
5075 } else {
5076 generate_fair_guard(flat_array_test(refined_klass_node, /* flat = */ false), bailout);
5077 }
5078 // TODO 8350865 This is not correct anymore. Write tests and fix logic similar to arraycopy.
5079 } else if (UseArrayFlattening && (orig_t == nullptr || !orig_t->is_not_flat()) &&
5080 // If dest is flat, src must be flat as well (guaranteed by src <: dest check if validated).
5081 ((!tklass->is_flat() && tklass->can_be_inline_array()) || !can_validate)) {
5082 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
5083 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
5084 generate_fair_guard(flat_array_test(load_object_klass(original)), bailout);
5085 if (orig_t != nullptr) {
5086 orig_t = orig_t->cast_to_not_flat();
5087 original = _gvn.transform(new CheckCastPPNode(control(), original, orig_t));
5088 }
5089 }
5090 if (!can_validate) {
5091 // No validation. The subtype check emitted at macro expansion time will not go to the slow
5092 // path but call checkcast_arraycopy which can not handle flat/null-free inline type arrays.
5093 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat/null-free.
5094 generate_fair_guard(flat_array_test(refined_klass_node), bailout);
5095 generate_fair_guard(null_free_array_test(original), bailout);
5096 }
5097 }
5098
5099 // Bail out if start is larger than the original length
5100 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
5101 generate_negative_guard(orig_tail, bailout, &orig_tail);
5102
5103 if (bailout->req() > 1) {
5104 PreserveJVMState pjvms(this);
5105 set_control(_gvn.transform(bailout));
5106 uncommon_trap(Deoptimization::Reason_intrinsic,
5107 Deoptimization::Action_maybe_recompile);
5108 }
5109
5110 if (!stopped()) {
5111 // How many elements will we copy from the original?
5112 // The answer is MinI(orig_tail, length).
5113 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
5114
5115 // Generate a direct call to the right arraycopy function(s).
5116 // We know the copy is disjoint but we might not know if the
5117 // oop stores need checking.
5118 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
5119 // This will fail a store-check if x contains any non-nulls.
5120
5121 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
5122 // loads/stores but it is legal only if we're sure the
5123 // Arrays.copyOf would succeed. So we need all input arguments
5124 // to the copyOf to be validated, including that the copy to the
5125 // new array won't trigger an ArrayStoreException. That subtype
5126 // check can be optimized if we know something on the type of
5127 // the input array from type speculation.
5128 if (_gvn.type(klass_node)->singleton()) {
5129 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
5130 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
5131
5132 int test = C->static_subtype_check(superk, subk);
5133 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
5134 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
5135 if (t_original->speculative_type() != nullptr) {
5136 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
5137 }
5138 }
5139 }
5140
5141 bool validated = false;
5142 // Reason_class_check rather than Reason_intrinsic because we
5143 // want to intrinsify even if this traps.
5144 if (can_validate) {
5145 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
5146
5147 if (not_subtype_ctrl != top()) {
5148 PreserveJVMState pjvms(this);
5149 set_control(not_subtype_ctrl);
5150 uncommon_trap(Deoptimization::Reason_class_check,
5151 Deoptimization::Action_make_not_entrant);
5152 assert(stopped(), "Should be stopped");
5153 }
5154 validated = true;
5155 }
5156
5157 if (!stopped()) {
5158 newcopy = new_array(refined_klass_node, length, 0); // no arguments to push
5159
5160 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
5161 load_object_klass(original), klass_node);
5162 if (!is_copyOfRange) {
5163 ac->set_copyof(validated);
5164 } else {
5165 ac->set_copyofrange(validated);
5166 }
5167 Node* n = _gvn.transform(ac);
5168 if (n == ac) {
5169 ac->connect_outputs(this);
5170 } else {
5171 assert(validated, "shouldn't transform if all arguments not validated");
5172 set_all_memory(n);
5173 }
5174 }
5175 }
5176 } // original reexecute is set back here
5177
5178 C->set_has_split_ifs(true); // Has chance for split-if optimization
5179 if (!stopped()) {
5180 set_result(newcopy);
5181 }
5182 return true;
5183 }
5184
5185
5186 //----------------------generate_virtual_guard---------------------------
5187 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
5188 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
5189 RegionNode* slow_region) {
5190 ciMethod* method = callee();
5191 int vtable_index = method->vtable_index();
5192 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5193 "bad index %d", vtable_index);
5194 // Get the Method* out of the appropriate vtable entry.
5195 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
5196 vtable_index*vtableEntry::size_in_bytes() +
5197 in_bytes(vtableEntry::method_offset());
5198 Node* entry_addr = off_heap_plus_addr(obj_klass, entry_offset);
5199 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
5200
5201 // Compare the target method with the expected method (e.g., Object.hashCode).
5202 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
5203
5204 Node* native_call = makecon(native_call_addr);
5205 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
5206 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
5207
5208 return generate_slow_guard(test_native, slow_region);
5209 }
5210
5211 //-----------------------generate_method_call----------------------------
5212 // Use generate_method_call to make a slow-call to the real
5213 // method if the fast path fails. An alternative would be to
5214 // use a stub like OptoRuntime::slow_arraycopy_Java.
5215 // This only works for expanding the current library call,
5216 // not another intrinsic. (E.g., don't use this for making an
5217 // arraycopy call inside of the copyOf intrinsic.)
5218 CallJavaNode*
5219 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
5220 // When compiling the intrinsic method itself, do not use this technique.
5221 guarantee(callee() != C->method(), "cannot make slow-call to self");
5222
5223 ciMethod* method = callee();
5224 // ensure the JVMS we have will be correct for this call
5225 guarantee(method_id == method->intrinsic_id(), "must match");
5226
5227 const TypeFunc* tf = TypeFunc::make(method);
5228 if (res_not_null) {
5229 assert(tf->return_type() == T_OBJECT, "");
5230 const TypeTuple* range = tf->range_cc();
5231 const Type** fields = TypeTuple::fields(range->cnt());
5232 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
5233 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
5234 tf = TypeFunc::make(tf->domain_cc(), new_range);
5235 }
5236 CallJavaNode* slow_call;
5237 if (is_static) {
5238 assert(!is_virtual, "");
5239 slow_call = new CallStaticJavaNode(C, tf,
5240 SharedRuntime::get_resolve_static_call_stub(), method);
5241 } else if (is_virtual) {
5242 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5243 int vtable_index = Method::invalid_vtable_index;
5244 if (UseInlineCaches) {
5245 // Suppress the vtable call
5246 } else {
5247 // hashCode and clone are not a miranda methods,
5248 // so the vtable index is fixed.
5249 // No need to use the linkResolver to get it.
5250 vtable_index = method->vtable_index();
5251 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5252 "bad index %d", vtable_index);
5253 }
5254 slow_call = new CallDynamicJavaNode(tf,
5255 SharedRuntime::get_resolve_virtual_call_stub(),
5256 method, vtable_index);
5257 } else { // neither virtual nor static: opt_virtual
5258 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5259 slow_call = new CallStaticJavaNode(C, tf,
5260 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
5261 slow_call->set_optimized_virtual(true);
5262 }
5263 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
5264 // To be able to issue a direct call (optimized virtual or virtual)
5265 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
5266 // about the method being invoked should be attached to the call site to
5267 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
5268 slow_call->set_override_symbolic_info(true);
5269 }
5270 set_arguments_for_java_call(slow_call);
5271 set_edges_for_java_call(slow_call);
5272 return slow_call;
5273 }
5274
5275
5276 /**
5277 * Build special case code for calls to hashCode on an object. This call may
5278 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
5279 * slightly different code.
5280 */
5281 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
5282 assert(is_static == callee()->is_static(), "correct intrinsic selection");
5283 assert(!(is_virtual && is_static), "either virtual, special, or static");
5284
5285 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
5286
5287 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5288 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
5289 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
5290 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5291 Node* obj = argument(0);
5292
5293 // Don't intrinsify hashcode on inline types for now.
5294 // The "is locked" runtime check also subsumes the inline type check (as inline types cannot be locked) and goes to the slow path.
5295 if (gvn().type(obj)->is_inlinetypeptr()) {
5296 return false;
5297 }
5298
5299 if (!is_static) {
5300 // Check for hashing null object
5301 obj = null_check_receiver();
5302 if (stopped()) return true; // unconditionally null
5303 result_reg->init_req(_null_path, top());
5304 result_val->init_req(_null_path, top());
5305 } else {
5306 // Do a null check, and return zero if null.
5307 // System.identityHashCode(null) == 0
5308 Node* null_ctl = top();
5309 obj = null_check_oop(obj, &null_ctl);
5310 result_reg->init_req(_null_path, null_ctl);
5311 result_val->init_req(_null_path, _gvn.intcon(0));
5312 }
5313
5314 // Unconditionally null? Then return right away.
5315 if (stopped()) {
5316 set_control( result_reg->in(_null_path));
5317 if (!stopped())
5318 set_result(result_val->in(_null_path));
5319 return true;
5320 }
5321
5322 // We only go to the fast case code if we pass a number of guards. The
5323 // paths which do not pass are accumulated in the slow_region.
5324 RegionNode* slow_region = new RegionNode(1);
5325 record_for_igvn(slow_region);
5326
5327 // If this is a virtual call, we generate a funny guard. We pull out
5328 // the vtable entry corresponding to hashCode() from the target object.
5329 // If the target method which we are calling happens to be the native
5330 // Object hashCode() method, we pass the guard. We do not need this
5331 // guard for non-virtual calls -- the caller is known to be the native
5332 // Object hashCode().
5333 if (is_virtual) {
5334 // After null check, get the object's klass.
5335 Node* obj_klass = load_object_klass(obj);
5336 generate_virtual_guard(obj_klass, slow_region);
5337 }
5338
5339 // Get the header out of the object, use LoadMarkNode when available
5340 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
5341 // The control of the load must be null. Otherwise, the load can move before
5342 // the null check after castPP removal.
5343 Node* no_ctrl = nullptr;
5344 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
5345
5346 if (!UseObjectMonitorTable) {
5347 // Test the header to see if it is safe to read w.r.t. locking.
5348 // We cannot use the inline type mask as this may check bits that are overriden
5349 // by an object monitor's pointer when inflating locking.
5350 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
5351 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
5352 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
5353 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
5354 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
5355
5356 generate_slow_guard(test_monitor, slow_region);
5357 }
5358
5359 // Get the hash value and check to see that it has been properly assigned.
5360 // We depend on hash_mask being at most 32 bits and avoid the use of
5361 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
5362 // vm: see markWord.hpp.
5363 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
5364 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
5365 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
5366 // This hack lets the hash bits live anywhere in the mark object now, as long
5367 // as the shift drops the relevant bits into the low 32 bits. Note that
5368 // Java spec says that HashCode is an int so there's no point in capturing
5369 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
5370 hshifted_header = ConvX2I(hshifted_header);
5371 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
5372
5373 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
5374 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
5375 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
5376
5377 generate_slow_guard(test_assigned, slow_region);
5378
5379 Node* init_mem = reset_memory();
5380 // fill in the rest of the null path:
5381 result_io ->init_req(_null_path, i_o());
5382 result_mem->init_req(_null_path, init_mem);
5383
5384 result_val->init_req(_fast_path, hash_val);
5385 result_reg->init_req(_fast_path, control());
5386 result_io ->init_req(_fast_path, i_o());
5387 result_mem->init_req(_fast_path, init_mem);
5388
5389 // Generate code for the slow case. We make a call to hashCode().
5390 set_control(_gvn.transform(slow_region));
5391 if (!stopped()) {
5392 // No need for PreserveJVMState, because we're using up the present state.
5393 set_all_memory(init_mem);
5394 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
5395 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
5396 Node* slow_result = set_results_for_java_call(slow_call);
5397 // this->control() comes from set_results_for_java_call
5398 result_reg->init_req(_slow_path, control());
5399 result_val->init_req(_slow_path, slow_result);
5400 result_io ->set_req(_slow_path, i_o());
5401 result_mem ->set_req(_slow_path, reset_memory());
5402 }
5403
5404 // Return the combined state.
5405 set_i_o( _gvn.transform(result_io) );
5406 set_all_memory( _gvn.transform(result_mem));
5407
5408 set_result(result_reg, result_val);
5409 return true;
5410 }
5411
5412 //---------------------------inline_native_getClass----------------------------
5413 // public final native Class<?> java.lang.Object.getClass();
5414 //
5415 // Build special case code for calls to getClass on an object.
5416 bool LibraryCallKit::inline_native_getClass() {
5417 Node* obj = argument(0);
5418 if (obj->is_InlineType()) {
5419 const Type* t = _gvn.type(obj);
5420 if (t->maybe_null()) {
5421 null_check(obj);
5422 }
5423 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror())));
5424 return true;
5425 }
5426 obj = null_check_receiver();
5427 if (stopped()) return true;
5428 set_result(load_mirror_from_klass(load_object_klass(obj)));
5429 return true;
5430 }
5431
5432 //-----------------inline_native_Reflection_getCallerClass---------------------
5433 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
5434 //
5435 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
5436 //
5437 // NOTE: This code must perform the same logic as JVM_GetCallerClass
5438 // in that it must skip particular security frames and checks for
5439 // caller sensitive methods.
5440 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
5441 #ifndef PRODUCT
5442 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5443 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
5444 }
5445 #endif
5446
5447 if (!jvms()->has_method()) {
5448 #ifndef PRODUCT
5449 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5450 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
5451 }
5452 #endif
5453 return false;
5454 }
5455
5456 // Walk back up the JVM state to find the caller at the required
5457 // depth.
5458 JVMState* caller_jvms = jvms();
5459
5460 // Cf. JVM_GetCallerClass
5461 // NOTE: Start the loop at depth 1 because the current JVM state does
5462 // not include the Reflection.getCallerClass() frame.
5463 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
5464 ciMethod* m = caller_jvms->method();
5465 switch (n) {
5466 case 0:
5467 fatal("current JVM state does not include the Reflection.getCallerClass frame");
5468 break;
5469 case 1:
5470 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
5471 if (!m->caller_sensitive()) {
5472 #ifndef PRODUCT
5473 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5474 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
5475 }
5476 #endif
5477 return false; // bail-out; let JVM_GetCallerClass do the work
5478 }
5479 break;
5480 default:
5481 if (!m->is_ignored_by_security_stack_walk()) {
5482 // We have reached the desired frame; return the holder class.
5483 // Acquire method holder as java.lang.Class and push as constant.
5484 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
5485 ciInstance* caller_mirror = caller_klass->java_mirror();
5486 set_result(makecon(TypeInstPtr::make(caller_mirror)));
5487
5488 #ifndef PRODUCT
5489 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5490 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());
5491 tty->print_cr(" JVM state at this point:");
5492 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5493 ciMethod* m = jvms()->of_depth(i)->method();
5494 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5495 }
5496 }
5497 #endif
5498 return true;
5499 }
5500 break;
5501 }
5502 }
5503
5504 #ifndef PRODUCT
5505 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5506 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
5507 tty->print_cr(" JVM state at this point:");
5508 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5509 ciMethod* m = jvms()->of_depth(i)->method();
5510 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5511 }
5512 }
5513 #endif
5514
5515 return false; // bail-out; let JVM_GetCallerClass do the work
5516 }
5517
5518 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
5519 Node* arg = argument(0);
5520 Node* result = nullptr;
5521
5522 switch (id) {
5523 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
5524 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
5525 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5526 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5527 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5528 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5529
5530 case vmIntrinsics::_doubleToLongBits: {
5531 // two paths (plus control) merge in a wood
5532 RegionNode *r = new RegionNode(3);
5533 Node *phi = new PhiNode(r, TypeLong::LONG);
5534
5535 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5536 // Build the boolean node
5537 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5538
5539 // Branch either way.
5540 // NaN case is less traveled, which makes all the difference.
5541 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5542 Node *opt_isnan = _gvn.transform(ifisnan);
5543 assert( opt_isnan->is_If(), "Expect an IfNode");
5544 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5545 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5546
5547 set_control(iftrue);
5548
5549 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5550 Node *slow_result = longcon(nan_bits); // return NaN
5551 phi->init_req(1, _gvn.transform( slow_result ));
5552 r->init_req(1, iftrue);
5553
5554 // Else fall through
5555 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5556 set_control(iffalse);
5557
5558 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5559 r->init_req(2, iffalse);
5560
5561 // Post merge
5562 set_control(_gvn.transform(r));
5563 record_for_igvn(r);
5564
5565 C->set_has_split_ifs(true); // Has chance for split-if optimization
5566 result = phi;
5567 assert(result->bottom_type()->isa_long(), "must be");
5568 break;
5569 }
5570
5571 case vmIntrinsics::_floatToIntBits: {
5572 // two paths (plus control) merge in a wood
5573 RegionNode *r = new RegionNode(3);
5574 Node *phi = new PhiNode(r, TypeInt::INT);
5575
5576 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5577 // Build the boolean node
5578 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5579
5580 // Branch either way.
5581 // NaN case is less traveled, which makes all the difference.
5582 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5583 Node *opt_isnan = _gvn.transform(ifisnan);
5584 assert( opt_isnan->is_If(), "Expect an IfNode");
5585 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5586 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5587
5588 set_control(iftrue);
5589
5590 static const jint nan_bits = 0x7fc00000;
5591 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5592 phi->init_req(1, _gvn.transform( slow_result ));
5593 r->init_req(1, iftrue);
5594
5595 // Else fall through
5596 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5597 set_control(iffalse);
5598
5599 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5600 r->init_req(2, iffalse);
5601
5602 // Post merge
5603 set_control(_gvn.transform(r));
5604 record_for_igvn(r);
5605
5606 C->set_has_split_ifs(true); // Has chance for split-if optimization
5607 result = phi;
5608 assert(result->bottom_type()->isa_int(), "must be");
5609 break;
5610 }
5611
5612 default:
5613 fatal_unexpected_iid(id);
5614 break;
5615 }
5616 set_result(_gvn.transform(result));
5617 return true;
5618 }
5619
5620 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5621 Node* arg = argument(0);
5622 Node* result = nullptr;
5623
5624 switch (id) {
5625 case vmIntrinsics::_floatIsInfinite:
5626 result = new IsInfiniteFNode(arg);
5627 break;
5628 case vmIntrinsics::_floatIsFinite:
5629 result = new IsFiniteFNode(arg);
5630 break;
5631 case vmIntrinsics::_doubleIsInfinite:
5632 result = new IsInfiniteDNode(arg);
5633 break;
5634 case vmIntrinsics::_doubleIsFinite:
5635 result = new IsFiniteDNode(arg);
5636 break;
5637 default:
5638 fatal_unexpected_iid(id);
5639 break;
5640 }
5641 set_result(_gvn.transform(result));
5642 return true;
5643 }
5644
5645 //----------------------inline_unsafe_copyMemory-------------------------
5646 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5647
5648 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5649 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5650 const Type* base_t = gvn.type(base);
5651
5652 bool in_native = (base_t == TypePtr::NULL_PTR);
5653 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5654 bool is_mixed = !in_heap && !in_native;
5655
5656 if (is_mixed) {
5657 return true; // mixed accesses can touch both on-heap and off-heap memory
5658 }
5659 if (in_heap) {
5660 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5661 if (!is_prim_array) {
5662 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5663 // there's not enough type information available to determine proper memory slice for it.
5664 return true;
5665 }
5666 }
5667 return false;
5668 }
5669
5670 bool LibraryCallKit::inline_unsafe_copyMemory() {
5671 if (callee()->is_static()) return false; // caller must have the capability!
5672 null_check_receiver(); // null-check receiver
5673 if (stopped()) return true;
5674
5675 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5676
5677 Node* src_base = argument(1); // type: oop
5678 Node* src_off = ConvL2X(argument(2)); // type: long
5679 Node* dst_base = argument(4); // type: oop
5680 Node* dst_off = ConvL2X(argument(5)); // type: long
5681 Node* size = ConvL2X(argument(7)); // type: long
5682
5683 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5684 "fieldOffset must be byte-scaled");
5685
5686 Node* src_addr = make_unsafe_address(src_base, src_off);
5687 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5688
5689 Node* thread = _gvn.transform(new ThreadLocalNode());
5690 Node* doing_unsafe_access_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5691 BasicType doing_unsafe_access_bt = T_BYTE;
5692 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5693
5694 // update volatile field
5695 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5696
5697 int flags = RC_LEAF | RC_NO_FP;
5698
5699 const TypePtr* dst_type = TypePtr::BOTTOM;
5700
5701 // Adjust memory effects of the runtime call based on input values.
5702 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5703 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5704 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5705
5706 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5707 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5708 flags |= RC_NARROW_MEM; // narrow in memory
5709 }
5710 }
5711
5712 // Call it. Note that the length argument is not scaled.
5713 make_runtime_call(flags,
5714 OptoRuntime::fast_arraycopy_Type(),
5715 StubRoutines::unsafe_arraycopy(),
5716 "unsafe_arraycopy",
5717 dst_type,
5718 src_addr, dst_addr, size XTOP);
5719
5720 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5721
5722 return true;
5723 }
5724
5725 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5726 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5727 bool LibraryCallKit::inline_unsafe_setMemory() {
5728 if (callee()->is_static()) return false; // caller must have the capability!
5729 null_check_receiver(); // null-check receiver
5730 if (stopped()) return true;
5731
5732 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5733
5734 Node* dst_base = argument(1); // type: oop
5735 Node* dst_off = ConvL2X(argument(2)); // type: long
5736 Node* size = ConvL2X(argument(4)); // type: long
5737 Node* byte = argument(6); // type: byte
5738
5739 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5740 "fieldOffset must be byte-scaled");
5741
5742 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5743
5744 Node* thread = _gvn.transform(new ThreadLocalNode());
5745 Node* doing_unsafe_access_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5746 BasicType doing_unsafe_access_bt = T_BYTE;
5747 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5748
5749 // update volatile field
5750 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5751
5752 int flags = RC_LEAF | RC_NO_FP;
5753
5754 const TypePtr* dst_type = TypePtr::BOTTOM;
5755
5756 // Adjust memory effects of the runtime call based on input values.
5757 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5758 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5759
5760 flags |= RC_NARROW_MEM; // narrow in memory
5761 }
5762
5763 // Call it. Note that the length argument is not scaled.
5764 make_runtime_call(flags,
5765 OptoRuntime::unsafe_setmemory_Type(),
5766 StubRoutines::unsafe_setmemory(),
5767 "unsafe_setmemory",
5768 dst_type,
5769 dst_addr, size XTOP, byte);
5770
5771 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5772
5773 return true;
5774 }
5775
5776 #undef XTOP
5777
5778 //------------------------clone_coping-----------------------------------
5779 // Helper function for inline_native_clone.
5780 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5781 assert(obj_size != nullptr, "");
5782 Node* raw_obj = alloc_obj->in(1);
5783 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5784
5785 AllocateNode* alloc = nullptr;
5786 if (ReduceBulkZeroing &&
5787 // If we are implementing an array clone without knowing its source type
5788 // (can happen when compiling the array-guarded branch of a reflective
5789 // Object.clone() invocation), initialize the array within the allocation.
5790 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5791 // to a runtime clone call that assumes fully initialized source arrays.
5792 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5793 // We will be completely responsible for initializing this object -
5794 // mark Initialize node as complete.
5795 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5796 // The object was just allocated - there should be no any stores!
5797 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5798 // Mark as complete_with_arraycopy so that on AllocateNode
5799 // expansion, we know this AllocateNode is initialized by an array
5800 // copy and a StoreStore barrier exists after the array copy.
5801 alloc->initialization()->set_complete_with_arraycopy();
5802 }
5803
5804 Node* size = _gvn.transform(obj_size);
5805 access_clone(obj, alloc_obj, size, is_array);
5806
5807 // Do not let reads from the cloned object float above the arraycopy.
5808 if (alloc != nullptr) {
5809 // Do not let stores that initialize this object be reordered with
5810 // a subsequent store that would make this object accessible by
5811 // other threads.
5812 // Record what AllocateNode this StoreStore protects so that
5813 // escape analysis can go from the MemBarStoreStoreNode to the
5814 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5815 // based on the escape status of the AllocateNode.
5816 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5817 } else {
5818 insert_mem_bar(Op_MemBarCPUOrder);
5819 }
5820 }
5821
5822 //------------------------inline_native_clone----------------------------
5823 // protected native Object java.lang.Object.clone();
5824 //
5825 // Here are the simple edge cases:
5826 // null receiver => normal trap
5827 // virtual and clone was overridden => slow path to out-of-line clone
5828 // not cloneable or finalizer => slow path to out-of-line Object.clone
5829 //
5830 // The general case has two steps, allocation and copying.
5831 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5832 //
5833 // Copying also has two cases, oop arrays and everything else.
5834 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5835 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5836 //
5837 // These steps fold up nicely if and when the cloned object's klass
5838 // can be sharply typed as an object array, a type array, or an instance.
5839 //
5840 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5841 PhiNode* result_val;
5842
5843 // Set the reexecute bit for the interpreter to reexecute
5844 // the bytecode that invokes Object.clone if deoptimization happens.
5845 { PreserveReexecuteState preexecs(this);
5846 jvms()->set_should_reexecute(true);
5847
5848 Node* obj = argument(0);
5849 obj = null_check_receiver();
5850 if (stopped()) return true;
5851
5852 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5853 if (obj_type->is_inlinetypeptr()) {
5854 // If the object to clone is an inline type, we can simply return it (i.e. a nop) since inline types have
5855 // no identity.
5856 set_result(obj);
5857 return true;
5858 }
5859
5860 // If we are going to clone an instance, we need its exact type to
5861 // know the number and types of fields to convert the clone to
5862 // loads/stores. Maybe a speculative type can help us.
5863 if (!obj_type->klass_is_exact() &&
5864 obj_type->speculative_type() != nullptr &&
5865 obj_type->speculative_type()->is_instance_klass() &&
5866 !obj_type->speculative_type()->is_inlinetype()) {
5867 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5868 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5869 !spec_ik->has_injected_fields()) {
5870 if (!obj_type->isa_instptr() ||
5871 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5872 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5873 }
5874 }
5875 }
5876
5877 // Conservatively insert a memory barrier on all memory slices.
5878 // Do not let writes into the original float below the clone.
5879 insert_mem_bar(Op_MemBarCPUOrder);
5880
5881 // paths into result_reg:
5882 enum {
5883 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5884 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5885 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5886 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5887 PATH_LIMIT
5888 };
5889 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5890 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5891 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5892 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5893 record_for_igvn(result_reg);
5894
5895 Node* obj_klass = load_object_klass(obj);
5896 // We only go to the fast case code if we pass a number of guards.
5897 // The paths which do not pass are accumulated in the slow_region.
5898 RegionNode* slow_region = new RegionNode(1);
5899 record_for_igvn(slow_region);
5900
5901 Node* array_obj = obj;
5902 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5903 if (array_ctl != nullptr) {
5904 // It's an array.
5905 PreserveJVMState pjvms(this);
5906 set_control(array_ctl);
5907
5908 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5909 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr();
5910 if (UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) &&
5911 obj_type->can_be_inline_array() &&
5912 (ary_ptr == nullptr || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) {
5913 // Flat inline type array may have object field that would require a
5914 // write barrier. Conservatively, go to slow path.
5915 generate_fair_guard(flat_array_test(obj_klass), slow_region);
5916 }
5917
5918 if (!stopped()) {
5919 Node* obj_length = load_array_length(array_obj);
5920 Node* array_size = nullptr; // Size of the array without object alignment padding.
5921 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5922
5923 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5924 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5925 // If it is an oop array, it requires very special treatment,
5926 // because gc barriers are required when accessing the array.
5927 Node* is_obja = generate_refArray_guard(obj_klass, (RegionNode*)nullptr);
5928 if (is_obja != nullptr) {
5929 PreserveJVMState pjvms2(this);
5930 set_control(is_obja);
5931 // Generate a direct call to the right arraycopy function(s).
5932 // Clones are always tightly coupled.
5933 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5934 ac->set_clone_oop_array();
5935 Node* n = _gvn.transform(ac);
5936 assert(n == ac, "cannot disappear");
5937 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5938
5939 result_reg->init_req(_objArray_path, control());
5940 result_val->init_req(_objArray_path, alloc_obj);
5941 result_i_o ->set_req(_objArray_path, i_o());
5942 result_mem ->set_req(_objArray_path, reset_memory());
5943 }
5944 }
5945 // Otherwise, there are no barriers to worry about.
5946 // (We can dispense with card marks if we know the allocation
5947 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5948 // causes the non-eden paths to take compensating steps to
5949 // simulate a fresh allocation, so that no further
5950 // card marks are required in compiled code to initialize
5951 // the object.)
5952
5953 if (!stopped()) {
5954 copy_to_clone(obj, alloc_obj, array_size, true);
5955
5956 // Present the results of the copy.
5957 result_reg->init_req(_array_path, control());
5958 result_val->init_req(_array_path, alloc_obj);
5959 result_i_o ->set_req(_array_path, i_o());
5960 result_mem ->set_req(_array_path, reset_memory());
5961 }
5962 }
5963 }
5964
5965 if (!stopped()) {
5966 // It's an instance (we did array above). Make the slow-path tests.
5967 // If this is a virtual call, we generate a funny guard. We grab
5968 // the vtable entry corresponding to clone() from the target object.
5969 // If the target method which we are calling happens to be the
5970 // Object clone() method, we pass the guard. We do not need this
5971 // guard for non-virtual calls; the caller is known to be the native
5972 // Object clone().
5973 if (is_virtual) {
5974 generate_virtual_guard(obj_klass, slow_region);
5975 }
5976
5977 // The object must be easily cloneable and must not have a finalizer.
5978 // Both of these conditions may be checked in a single test.
5979 // We could optimize the test further, but we don't care.
5980 generate_misc_flags_guard(obj_klass,
5981 // Test both conditions:
5982 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5983 // Must be cloneable but not finalizer:
5984 KlassFlags::_misc_is_cloneable_fast,
5985 slow_region);
5986 }
5987
5988 if (!stopped()) {
5989 // It's an instance, and it passed the slow-path tests.
5990 PreserveJVMState pjvms(this);
5991 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5992 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5993 // is reexecuted if deoptimization occurs and there could be problems when merging
5994 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5995 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5996
5997 copy_to_clone(obj, alloc_obj, obj_size, false);
5998
5999 // Present the results of the slow call.
6000 result_reg->init_req(_instance_path, control());
6001 result_val->init_req(_instance_path, alloc_obj);
6002 result_i_o ->set_req(_instance_path, i_o());
6003 result_mem ->set_req(_instance_path, reset_memory());
6004 }
6005
6006 // Generate code for the slow case. We make a call to clone().
6007 set_control(_gvn.transform(slow_region));
6008 if (!stopped()) {
6009 PreserveJVMState pjvms(this);
6010 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
6011 // We need to deoptimize on exception (see comment above)
6012 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
6013 // this->control() comes from set_results_for_java_call
6014 result_reg->init_req(_slow_path, control());
6015 result_val->init_req(_slow_path, slow_result);
6016 result_i_o ->set_req(_slow_path, i_o());
6017 result_mem ->set_req(_slow_path, reset_memory());
6018 }
6019
6020 // Return the combined state.
6021 set_control( _gvn.transform(result_reg));
6022 set_i_o( _gvn.transform(result_i_o));
6023 set_all_memory( _gvn.transform(result_mem));
6024 } // original reexecute is set back here
6025
6026 set_result(_gvn.transform(result_val));
6027 return true;
6028 }
6029
6030 // If we have a tightly coupled allocation, the arraycopy may take care
6031 // of the array initialization. If one of the guards we insert between
6032 // the allocation and the arraycopy causes a deoptimization, an
6033 // uninitialized array will escape the compiled method. To prevent that
6034 // we set the JVM state for uncommon traps between the allocation and
6035 // the arraycopy to the state before the allocation so, in case of
6036 // deoptimization, we'll reexecute the allocation and the
6037 // initialization.
6038 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
6039 if (alloc != nullptr) {
6040 ciMethod* trap_method = alloc->jvms()->method();
6041 int trap_bci = alloc->jvms()->bci();
6042
6043 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6044 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
6045 // Make sure there's no store between the allocation and the
6046 // arraycopy otherwise visible side effects could be rexecuted
6047 // in case of deoptimization and cause incorrect execution.
6048 bool no_interfering_store = true;
6049 Node* mem = alloc->in(TypeFunc::Memory);
6050 if (mem->is_MergeMem()) {
6051 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
6052 Node* n = mms.memory();
6053 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
6054 assert(n->is_Store(), "what else?");
6055 no_interfering_store = false;
6056 break;
6057 }
6058 }
6059 } else {
6060 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
6061 Node* n = mms.memory();
6062 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
6063 assert(n->is_Store(), "what else?");
6064 no_interfering_store = false;
6065 break;
6066 }
6067 }
6068 }
6069
6070 if (no_interfering_store) {
6071 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6072
6073 JVMState* saved_jvms = jvms();
6074 saved_reexecute_sp = _reexecute_sp;
6075
6076 set_jvms(sfpt->jvms());
6077 _reexecute_sp = jvms()->sp();
6078
6079 return saved_jvms;
6080 }
6081 }
6082 }
6083 return nullptr;
6084 }
6085
6086 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
6087 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
6088 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
6089 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
6090 uint size = alloc->req();
6091 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
6092 old_jvms->set_map(sfpt);
6093 for (uint i = 0; i < size; i++) {
6094 sfpt->init_req(i, alloc->in(i));
6095 }
6096 int adjustment = 1;
6097 const TypeAryKlassPtr* ary_klass_ptr = alloc->in(AllocateNode::KlassNode)->bottom_type()->is_aryklassptr();
6098 if (ary_klass_ptr->is_null_free()) {
6099 // A null-free, tightly coupled array allocation can only come from LibraryCallKit::inline_newArray which
6100 // also requires the componentType and initVal on stack for re-execution.
6101 // Re-create and push the componentType.
6102 ciArrayKlass* klass = ary_klass_ptr->exact_klass()->as_array_klass();
6103 ciInstance* instance = klass->component_mirror_instance();
6104 const TypeInstPtr* t_instance = TypeInstPtr::make(instance);
6105 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), makecon(t_instance));
6106 adjustment++;
6107 }
6108 // re-push array length for deoptimization
6109 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment - 1, alloc->in(AllocateNode::ALength));
6110 if (ary_klass_ptr->is_null_free()) {
6111 // Re-create and push the initVal.
6112 Node* init_val = alloc->in(AllocateNode::InitValue);
6113 if (init_val == nullptr) {
6114 init_val = InlineTypeNode::make_all_zero(_gvn, ary_klass_ptr->elem()->is_instklassptr()->instance_klass()->as_inline_klass());
6115 } else if (UseCompressedOops) {
6116 init_val = _gvn.transform(new DecodeNNode(init_val, init_val->bottom_type()->make_ptr()));
6117 }
6118 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment, init_val);
6119 adjustment++;
6120 }
6121 old_jvms->set_sp(old_jvms->sp() + adjustment);
6122 old_jvms->set_monoff(old_jvms->monoff() + adjustment);
6123 old_jvms->set_scloff(old_jvms->scloff() + adjustment);
6124 old_jvms->set_endoff(old_jvms->endoff() + adjustment);
6125 old_jvms->set_should_reexecute(true);
6126
6127 sfpt->set_i_o(map()->i_o());
6128 sfpt->set_memory(map()->memory());
6129 sfpt->set_control(map()->control());
6130 return sfpt;
6131 }
6132
6133 // In case of a deoptimization, we restart execution at the
6134 // allocation, allocating a new array. We would leave an uninitialized
6135 // array in the heap that GCs wouldn't expect. Move the allocation
6136 // after the traps so we don't allocate the array if we
6137 // deoptimize. This is possible because tightly_coupled_allocation()
6138 // guarantees there's no observer of the allocated array at this point
6139 // and the control flow is simple enough.
6140 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
6141 int saved_reexecute_sp, uint new_idx) {
6142 if (saved_jvms_before_guards != nullptr && !stopped()) {
6143 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
6144
6145 assert(alloc != nullptr, "only with a tightly coupled allocation");
6146 // restore JVM state to the state at the arraycopy
6147 saved_jvms_before_guards->map()->set_control(map()->control());
6148 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
6149 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
6150 // If we've improved the types of some nodes (null check) while
6151 // emitting the guards, propagate them to the current state
6152 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
6153 set_jvms(saved_jvms_before_guards);
6154 _reexecute_sp = saved_reexecute_sp;
6155
6156 // Remove the allocation from above the guards
6157 CallProjections* callprojs = alloc->extract_projections(true);
6158 InitializeNode* init = alloc->initialization();
6159 Node* alloc_mem = alloc->in(TypeFunc::Memory);
6160 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
6161 init->replace_mem_projs_by(alloc_mem, C);
6162
6163 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
6164 // the allocation (i.e. is only valid if the allocation succeeds):
6165 // 1) replace CastIINode with AllocateArrayNode's length here
6166 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
6167 //
6168 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
6169 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
6170 Node* init_control = init->proj_out(TypeFunc::Control);
6171 Node* alloc_length = alloc->Ideal_length();
6172 #ifdef ASSERT
6173 Node* prev_cast = nullptr;
6174 #endif
6175 for (uint i = 0; i < init_control->outcnt(); i++) {
6176 Node* init_out = init_control->raw_out(i);
6177 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
6178 #ifdef ASSERT
6179 if (prev_cast == nullptr) {
6180 prev_cast = init_out;
6181 } else {
6182 if (prev_cast->cmp(*init_out) == false) {
6183 prev_cast->dump();
6184 init_out->dump();
6185 assert(false, "not equal CastIINode");
6186 }
6187 }
6188 #endif
6189 C->gvn_replace_by(init_out, alloc_length);
6190 }
6191 }
6192 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
6193
6194 // move the allocation here (after the guards)
6195 _gvn.hash_delete(alloc);
6196 alloc->set_req(TypeFunc::Control, control());
6197 alloc->set_req(TypeFunc::I_O, i_o());
6198 Node *mem = reset_memory();
6199 set_all_memory(mem);
6200 alloc->set_req(TypeFunc::Memory, mem);
6201 set_control(init->proj_out_or_null(TypeFunc::Control));
6202 set_i_o(callprojs->fallthrough_ioproj);
6203
6204 // Update memory as done in GraphKit::set_output_for_allocation()
6205 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
6206 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
6207 if (ary_type->isa_aryptr() && length_type != nullptr) {
6208 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
6209 }
6210 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
6211 int elemidx = C->get_alias_index(telemref);
6212 // Need to properly move every memory projection for the Initialize
6213 #ifdef ASSERT
6214 int mark_idx = C->get_alias_index(ary_type->add_offset(oopDesc::mark_offset_in_bytes()));
6215 int klass_idx = C->get_alias_index(ary_type->add_offset(oopDesc::klass_offset_in_bytes()));
6216 #endif
6217 auto move_proj = [&](ProjNode* proj) {
6218 int alias_idx = C->get_alias_index(proj->adr_type());
6219 assert(alias_idx == Compile::AliasIdxRaw ||
6220 alias_idx == elemidx ||
6221 alias_idx == mark_idx ||
6222 alias_idx == klass_idx, "should be raw memory or array element type");
6223 set_memory(proj, alias_idx);
6224 };
6225 init->for_each_proj(move_proj, TypeFunc::Memory);
6226
6227 Node* allocx = _gvn.transform(alloc);
6228 assert(allocx == alloc, "where has the allocation gone?");
6229 assert(dest->is_CheckCastPP(), "not an allocation result?");
6230
6231 _gvn.hash_delete(dest);
6232 dest->set_req(0, control());
6233 Node* destx = _gvn.transform(dest);
6234 assert(destx == dest, "where has the allocation result gone?");
6235
6236 array_ideal_length(alloc, ary_type, true);
6237 }
6238 }
6239
6240 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
6241 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
6242 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
6243 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
6244 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
6245 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
6246 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
6247 JVMState* saved_jvms_before_guards) {
6248 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
6249 // There is at least one unrelated uncommon trap which needs to be replaced.
6250 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6251
6252 JVMState* saved_jvms = jvms();
6253 const int saved_reexecute_sp = _reexecute_sp;
6254 set_jvms(sfpt->jvms());
6255 _reexecute_sp = jvms()->sp();
6256
6257 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
6258
6259 // Restore state
6260 set_jvms(saved_jvms);
6261 _reexecute_sp = saved_reexecute_sp;
6262 }
6263 }
6264
6265 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
6266 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
6267 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
6268 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
6269 while (if_proj->is_IfProj()) {
6270 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
6271 if (uncommon_trap != nullptr) {
6272 create_new_uncommon_trap(uncommon_trap);
6273 }
6274 assert(if_proj->in(0)->is_If(), "must be If");
6275 if_proj = if_proj->in(0)->in(0);
6276 }
6277 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
6278 "must have reached control projection of init node");
6279 }
6280
6281 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
6282 const int trap_request = uncommon_trap_call->uncommon_trap_request();
6283 assert(trap_request != 0, "no valid UCT trap request");
6284 PreserveJVMState pjvms(this);
6285 set_control(uncommon_trap_call->in(0));
6286 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
6287 Deoptimization::trap_request_action(trap_request));
6288 assert(stopped(), "Should be stopped");
6289 _gvn.hash_delete(uncommon_trap_call);
6290 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
6291 }
6292
6293 // Common checks for array sorting intrinsics arguments.
6294 // Returns `true` if checks passed.
6295 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
6296 // check address of the class
6297 if (elementType == nullptr || elementType->is_top()) {
6298 return false; // dead path
6299 }
6300 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
6301 if (elem_klass == nullptr) {
6302 return false; // dead path
6303 }
6304 // java_mirror_type() returns non-null for compile-time Class constants only
6305 ciType* elem_type = elem_klass->java_mirror_type();
6306 if (elem_type == nullptr) {
6307 return false;
6308 }
6309 bt = elem_type->basic_type();
6310 // Disable the intrinsic if the CPU does not support SIMD sort
6311 if (!Matcher::supports_simd_sort(bt)) {
6312 return false;
6313 }
6314 // check address of the array
6315 if (obj == nullptr || obj->is_top()) {
6316 return false; // dead path
6317 }
6318 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
6319 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
6320 return false; // failed input validation
6321 }
6322 return true;
6323 }
6324
6325 //------------------------------inline_array_partition-----------------------
6326 bool LibraryCallKit::inline_array_partition() {
6327 address stubAddr = StubRoutines::select_array_partition_function();
6328 if (stubAddr == nullptr) {
6329 return false; // Intrinsic's stub is not implemented on this platform
6330 }
6331 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
6332
6333 // no receiver because it is a static method
6334 Node* elementType = argument(0);
6335 Node* obj = argument(1);
6336 Node* offset = argument(2); // long
6337 Node* fromIndex = argument(4);
6338 Node* toIndex = argument(5);
6339 Node* indexPivot1 = argument(6);
6340 Node* indexPivot2 = argument(7);
6341 // PartitionOperation: argument(8) is ignored
6342
6343 Node* pivotIndices = nullptr;
6344 BasicType bt = T_ILLEGAL;
6345
6346 if (!check_array_sort_arguments(elementType, obj, bt)) {
6347 return false;
6348 }
6349 null_check(obj);
6350 // If obj is dead, only null-path is taken.
6351 if (stopped()) {
6352 return true;
6353 }
6354 // Set the original stack and the reexecute bit for the interpreter to reexecute
6355 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
6356 { PreserveReexecuteState preexecs(this);
6357 jvms()->set_should_reexecute(true);
6358
6359 Node* obj_adr = make_unsafe_address(obj, offset);
6360
6361 // create the pivotIndices array of type int and size = 2
6362 Node* size = intcon(2);
6363 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
6364 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
6365 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
6366 guarantee(alloc != nullptr, "created above");
6367 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
6368
6369 // pass the basic type enum to the stub
6370 Node* elemType = intcon(bt);
6371
6372 // Call the stub
6373 const char *stubName = "array_partition_stub";
6374 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
6375 stubAddr, stubName, TypePtr::BOTTOM,
6376 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
6377 indexPivot1, indexPivot2);
6378
6379 } // original reexecute is set back here
6380
6381 if (!stopped()) {
6382 set_result(pivotIndices);
6383 }
6384
6385 return true;
6386 }
6387
6388
6389 //------------------------------inline_array_sort-----------------------
6390 bool LibraryCallKit::inline_array_sort() {
6391 address stubAddr = StubRoutines::select_arraysort_function();
6392 if (stubAddr == nullptr) {
6393 return false; // Intrinsic's stub is not implemented on this platform
6394 }
6395 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
6396
6397 // no receiver because it is a static method
6398 Node* elementType = argument(0);
6399 Node* obj = argument(1);
6400 Node* offset = argument(2); // long
6401 Node* fromIndex = argument(4);
6402 Node* toIndex = argument(5);
6403 // SortOperation: argument(6) is ignored
6404
6405 BasicType bt = T_ILLEGAL;
6406
6407 if (!check_array_sort_arguments(elementType, obj, bt)) {
6408 return false;
6409 }
6410 null_check(obj);
6411 // If obj is dead, only null-path is taken.
6412 if (stopped()) {
6413 return true;
6414 }
6415 Node* obj_adr = make_unsafe_address(obj, offset);
6416
6417 // pass the basic type enum to the stub
6418 Node* elemType = intcon(bt);
6419
6420 // Call the stub.
6421 const char *stubName = "arraysort_stub";
6422 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
6423 stubAddr, stubName, TypePtr::BOTTOM,
6424 obj_adr, elemType, fromIndex, toIndex);
6425
6426 return true;
6427 }
6428
6429
6430 //------------------------------inline_arraycopy-----------------------
6431 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
6432 // Object dest, int destPos,
6433 // int length);
6434 bool LibraryCallKit::inline_arraycopy() {
6435 // Get the arguments.
6436 Node* src = argument(0); // type: oop
6437 Node* src_offset = argument(1); // type: int
6438 Node* dest = argument(2); // type: oop
6439 Node* dest_offset = argument(3); // type: int
6440 Node* length = argument(4); // type: int
6441
6442 uint new_idx = C->unique();
6443
6444 // Check for allocation before we add nodes that would confuse
6445 // tightly_coupled_allocation()
6446 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
6447
6448 int saved_reexecute_sp = -1;
6449 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
6450 // See arraycopy_restore_alloc_state() comment
6451 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
6452 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
6453 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
6454 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
6455
6456 // The following tests must be performed
6457 // (1) src and dest are arrays.
6458 // (2) src and dest arrays must have elements of the same BasicType
6459 // (3) src and dest must not be null.
6460 // (4) src_offset must not be negative.
6461 // (5) dest_offset must not be negative.
6462 // (6) length must not be negative.
6463 // (7) src_offset + length must not exceed length of src.
6464 // (8) dest_offset + length must not exceed length of dest.
6465 // (9) each element of an oop array must be assignable
6466
6467 // (3) src and dest must not be null.
6468 // always do this here because we need the JVM state for uncommon traps
6469 Node* null_ctl = top();
6470 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
6471 assert(null_ctl->is_top(), "no null control here");
6472 dest = null_check(dest, T_ARRAY);
6473
6474 if (!can_emit_guards) {
6475 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
6476 // guards but the arraycopy node could still take advantage of a
6477 // tightly allocated allocation. tightly_coupled_allocation() is
6478 // called again to make sure it takes the null check above into
6479 // account: the null check is mandatory and if it caused an
6480 // uncommon trap to be emitted then the allocation can't be
6481 // considered tightly coupled in this context.
6482 alloc = tightly_coupled_allocation(dest);
6483 }
6484
6485 bool validated = false;
6486
6487 const Type* src_type = _gvn.type(src);
6488 const Type* dest_type = _gvn.type(dest);
6489 const TypeAryPtr* top_src = src_type->isa_aryptr();
6490 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6491
6492 // Do we have the type of src?
6493 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6494 // Do we have the type of dest?
6495 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6496 // Is the type for src from speculation?
6497 bool src_spec = false;
6498 // Is the type for dest from speculation?
6499 bool dest_spec = false;
6500
6501 if ((!has_src || !has_dest) && can_emit_guards) {
6502 // We don't have sufficient type information, let's see if
6503 // speculative types can help. We need to have types for both src
6504 // and dest so that it pays off.
6505
6506 // Do we already have or could we have type information for src
6507 bool could_have_src = has_src;
6508 // Do we already have or could we have type information for dest
6509 bool could_have_dest = has_dest;
6510
6511 ciKlass* src_k = nullptr;
6512 if (!has_src) {
6513 src_k = src_type->speculative_type_not_null();
6514 if (src_k != nullptr && src_k->is_array_klass()) {
6515 could_have_src = true;
6516 }
6517 }
6518
6519 ciKlass* dest_k = nullptr;
6520 if (!has_dest) {
6521 dest_k = dest_type->speculative_type_not_null();
6522 if (dest_k != nullptr && dest_k->is_array_klass()) {
6523 could_have_dest = true;
6524 }
6525 }
6526
6527 if (could_have_src && could_have_dest) {
6528 // This is going to pay off so emit the required guards
6529 if (!has_src) {
6530 src = maybe_cast_profiled_obj(src, src_k, true);
6531 src_type = _gvn.type(src);
6532 top_src = src_type->isa_aryptr();
6533 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6534 src_spec = true;
6535 }
6536 if (!has_dest) {
6537 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6538 dest_type = _gvn.type(dest);
6539 top_dest = dest_type->isa_aryptr();
6540 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6541 dest_spec = true;
6542 }
6543 }
6544 }
6545
6546 if (has_src && has_dest && can_emit_guards) {
6547 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
6548 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
6549 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
6550 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
6551
6552 if (src_elem == dest_elem && top_src->is_flat() == top_dest->is_flat() && src_elem == T_OBJECT) {
6553 // If both arrays are object arrays then having the exact types
6554 // for both will remove the need for a subtype check at runtime
6555 // before the call and may make it possible to pick a faster copy
6556 // routine (without a subtype check on every element)
6557 // Do we have the exact type of src?
6558 bool could_have_src = src_spec;
6559 // Do we have the exact type of dest?
6560 bool could_have_dest = dest_spec;
6561 ciKlass* src_k = nullptr;
6562 ciKlass* dest_k = nullptr;
6563 if (!src_spec) {
6564 src_k = src_type->speculative_type_not_null();
6565 if (src_k != nullptr && src_k->is_array_klass()) {
6566 could_have_src = true;
6567 }
6568 }
6569 if (!dest_spec) {
6570 dest_k = dest_type->speculative_type_not_null();
6571 if (dest_k != nullptr && dest_k->is_array_klass()) {
6572 could_have_dest = true;
6573 }
6574 }
6575 if (could_have_src && could_have_dest) {
6576 // If we can have both exact types, emit the missing guards
6577 if (could_have_src && !src_spec) {
6578 src = maybe_cast_profiled_obj(src, src_k, true);
6579 src_type = _gvn.type(src);
6580 top_src = src_type->isa_aryptr();
6581 }
6582 if (could_have_dest && !dest_spec) {
6583 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6584 dest_type = _gvn.type(dest);
6585 top_dest = dest_type->isa_aryptr();
6586 }
6587 }
6588 }
6589 }
6590
6591 ciMethod* trap_method = method();
6592 int trap_bci = bci();
6593 if (saved_jvms_before_guards != nullptr) {
6594 trap_method = alloc->jvms()->method();
6595 trap_bci = alloc->jvms()->bci();
6596 }
6597
6598 bool negative_length_guard_generated = false;
6599
6600 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6601 can_emit_guards && !src->is_top() && !dest->is_top()) {
6602 // validate arguments: enables transformation the ArrayCopyNode
6603 validated = true;
6604
6605 RegionNode* slow_region = new RegionNode(1);
6606 record_for_igvn(slow_region);
6607
6608 // (1) src and dest are arrays.
6609 generate_non_array_guard(load_object_klass(src), slow_region, &src);
6610 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
6611
6612 // (2) src and dest arrays must have elements of the same BasicType
6613 // done at macro expansion or at Ideal transformation time
6614
6615 // (4) src_offset must not be negative.
6616 generate_negative_guard(src_offset, slow_region);
6617
6618 // (5) dest_offset must not be negative.
6619 generate_negative_guard(dest_offset, slow_region);
6620
6621 // (7) src_offset + length must not exceed length of src.
6622 generate_limit_guard(src_offset, length,
6623 load_array_length(src),
6624 slow_region);
6625
6626 // (8) dest_offset + length must not exceed length of dest.
6627 generate_limit_guard(dest_offset, length,
6628 load_array_length(dest),
6629 slow_region);
6630
6631 // (6) length must not be negative.
6632 // This is also checked in generate_arraycopy() during macro expansion, but
6633 // we also have to check it here for the case where the ArrayCopyNode will
6634 // be eliminated by Escape Analysis.
6635 if (EliminateAllocations) {
6636 generate_negative_guard(length, slow_region);
6637 negative_length_guard_generated = true;
6638 }
6639
6640 // (9) each element of an oop array must be assignable
6641 Node* dest_klass = load_object_klass(dest);
6642 Node* refined_dest_klass = dest_klass;
6643 if (src != dest) {
6644 dest_klass = load_non_refined_array_klass(refined_dest_klass);
6645 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6646 slow_region->add_req(not_subtype_ctrl);
6647 }
6648
6649 // TODO 8350865 Improve this. What about atomicity? Make sure this is always folded for type arrays.
6650 // If destination is null-restricted, source must be null-restricted as well: src_null_restricted || !dst_null_restricted
6651 Node* src_klass = load_object_klass(src);
6652 Node* adr_prop_src = basic_plus_adr(top(), src_klass, in_bytes(ArrayKlass::properties_offset()));
6653 Node* prop_src = _gvn.transform(LoadNode::make(_gvn, control(), immutable_memory(), adr_prop_src,
6654 _gvn.type(adr_prop_src)->is_ptr(), TypeInt::INT, T_INT,
6655 MemNode::unordered));
6656 Node* adr_prop_dest = basic_plus_adr(top(), refined_dest_klass, in_bytes(ArrayKlass::properties_offset()));
6657 Node* prop_dest = _gvn.transform(LoadNode::make(_gvn, control(), immutable_memory(), adr_prop_dest,
6658 _gvn.type(adr_prop_dest)->is_ptr(), TypeInt::INT, T_INT,
6659 MemNode::unordered));
6660
6661 const ArrayProperties props_null_restricted = ArrayProperties::Default().with_null_restricted();
6662 jint props_value = (jint)props_null_restricted.value();
6663
6664 prop_dest = _gvn.transform(new XorINode(prop_dest, intcon(props_value)));
6665 prop_src = _gvn.transform(new OrINode(prop_dest, prop_src));
6666 prop_src = _gvn.transform(new AndINode(prop_src, intcon(props_value)));
6667
6668 Node* chk = _gvn.transform(new CmpINode(prop_src, intcon(props_value)));
6669 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::ne));
6670 generate_fair_guard(tst, slow_region);
6671
6672 // TODO 8350865 This is too strong
6673 generate_fair_guard(flat_array_test(src), slow_region);
6674 generate_fair_guard(flat_array_test(dest), slow_region);
6675
6676 {
6677 PreserveJVMState pjvms(this);
6678 set_control(_gvn.transform(slow_region));
6679 uncommon_trap(Deoptimization::Reason_intrinsic,
6680 Deoptimization::Action_make_not_entrant);
6681 assert(stopped(), "Should be stopped");
6682 }
6683
6684 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->isa_klassptr();
6685 if (dest_klass_t == nullptr) {
6686 // refined_dest_klass may not be an array, which leads to dest_klass being top. This means we
6687 // are in a dead path.
6688 uncommon_trap(Deoptimization::Reason_intrinsic,
6689 Deoptimization::Action_make_not_entrant);
6690 return true;
6691 }
6692
6693 const Type* toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6694 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6695 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6696 }
6697
6698 if (stopped()) {
6699 return true;
6700 }
6701
6702 Node* dest_klass = load_object_klass(dest);
6703 dest_klass = load_non_refined_array_klass(dest_klass);
6704
6705 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6706 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6707 // so the compiler has a chance to eliminate them: during macro expansion,
6708 // we have to set their control (CastPP nodes are eliminated).
6709 load_object_klass(src), dest_klass,
6710 load_array_length(src), load_array_length(dest));
6711
6712 ac->set_arraycopy(validated);
6713
6714 Node* n = _gvn.transform(ac);
6715 if (n == ac) {
6716 ac->connect_outputs(this);
6717 } else {
6718 assert(validated, "shouldn't transform if all arguments not validated");
6719 set_all_memory(n);
6720 }
6721 clear_upper_avx();
6722
6723
6724 return true;
6725 }
6726
6727
6728 // Helper function which determines if an arraycopy immediately follows
6729 // an allocation, with no intervening tests or other escapes for the object.
6730 AllocateArrayNode*
6731 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6732 if (stopped()) return nullptr; // no fast path
6733 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6734
6735 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6736 if (alloc == nullptr) return nullptr;
6737
6738 Node* rawmem = memory(Compile::AliasIdxRaw);
6739 // Is the allocation's memory state untouched?
6740 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6741 // Bail out if there have been raw-memory effects since the allocation.
6742 // (Example: There might have been a call or safepoint.)
6743 return nullptr;
6744 }
6745 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6746 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6747 return nullptr;
6748 }
6749
6750 // There must be no unexpected observers of this allocation.
6751 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6752 Node* obs = ptr->fast_out(i);
6753 if (obs != this->map()) {
6754 return nullptr;
6755 }
6756 }
6757
6758 // This arraycopy must unconditionally follow the allocation of the ptr.
6759 Node* alloc_ctl = ptr->in(0);
6760 Node* ctl = control();
6761 while (ctl != alloc_ctl) {
6762 // There may be guards which feed into the slow_region.
6763 // Any other control flow means that we might not get a chance
6764 // to finish initializing the allocated object.
6765 // Various low-level checks bottom out in uncommon traps. These
6766 // are considered safe since we've already checked above that
6767 // there is no unexpected observer of this allocation.
6768 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6769 assert(ctl->in(0)->is_If(), "must be If");
6770 ctl = ctl->in(0)->in(0);
6771 } else {
6772 return nullptr;
6773 }
6774 }
6775
6776 // If we get this far, we have an allocation which immediately
6777 // precedes the arraycopy, and we can take over zeroing the new object.
6778 // The arraycopy will finish the initialization, and provide
6779 // a new control state to which we will anchor the destination pointer.
6780
6781 return alloc;
6782 }
6783
6784 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6785 if (node->is_IfProj()) {
6786 IfProjNode* other_proj = node->as_IfProj()->other_if_proj();
6787 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6788 Node* obs = other_proj->fast_out(j);
6789 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6790 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6791 return obs->as_CallStaticJava();
6792 }
6793 }
6794 }
6795 return nullptr;
6796 }
6797
6798 //-------------inline_encodeISOArray-----------------------------------
6799 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6800 // int java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6801 // int java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6802 // encode char[] to byte[] in ISO_8859_1 or ASCII
6803 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6804 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6805 // no receiver since it is static method
6806 Node *src = argument(0);
6807 Node *src_offset = argument(1);
6808 Node *dst = argument(2);
6809 Node *dst_offset = argument(3);
6810 Node *length = argument(4);
6811
6812 // Cast source & target arrays to not-null
6813 src = must_be_not_null(src, true);
6814 dst = must_be_not_null(dst, true);
6815 if (stopped()) {
6816 return true;
6817 }
6818
6819 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6820 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6821 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6822 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6823 // failed array check
6824 return false;
6825 }
6826
6827 // Figure out the size and type of the elements we will be copying.
6828 BasicType src_elem = src_type->elem()->array_element_basic_type();
6829 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6830 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6831 return false;
6832 }
6833
6834 // Check source & target bounds
6835 RegionNode* bailout = create_bailout();
6836 generate_string_range_check(src, src_offset, length, src_elem == T_BYTE, bailout);
6837 generate_string_range_check(dst, dst_offset, length, false, bailout);
6838 if (check_bailout(bailout)) {
6839 return true;
6840 }
6841
6842 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6843 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6844 // 'src_start' points to src array + scaled offset
6845 // 'dst_start' points to dst array + scaled offset
6846
6847 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6848 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6849 enc = _gvn.transform(enc);
6850 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6851 set_memory(res_mem, mtype);
6852 set_result(enc);
6853 clear_upper_avx();
6854
6855 return true;
6856 }
6857
6858 //-------------inline_multiplyToLen-----------------------------------
6859 bool LibraryCallKit::inline_multiplyToLen() {
6860 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6861
6862 address stubAddr = StubRoutines::multiplyToLen();
6863 if (stubAddr == nullptr) {
6864 return false; // Intrinsic's stub is not implemented on this platform
6865 }
6866 const char* stubName = "multiplyToLen";
6867
6868 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6869
6870 // no receiver because it is a static method
6871 Node* x = argument(0);
6872 Node* xlen = argument(1);
6873 Node* y = argument(2);
6874 Node* ylen = argument(3);
6875 Node* z = argument(4);
6876
6877 x = must_be_not_null(x, true);
6878 y = must_be_not_null(y, true);
6879
6880 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6881 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6882 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6883 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6884 // failed array check
6885 return false;
6886 }
6887
6888 BasicType x_elem = x_type->elem()->array_element_basic_type();
6889 BasicType y_elem = y_type->elem()->array_element_basic_type();
6890 if (x_elem != T_INT || y_elem != T_INT) {
6891 return false;
6892 }
6893
6894 Node* x_start = array_element_address(x, intcon(0), x_elem);
6895 Node* y_start = array_element_address(y, intcon(0), y_elem);
6896 // 'x_start' points to x array + scaled xlen
6897 // 'y_start' points to y array + scaled ylen
6898
6899 Node* z_start = array_element_address(z, intcon(0), T_INT);
6900
6901 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6902 OptoRuntime::multiplyToLen_Type(),
6903 stubAddr, stubName, TypePtr::BOTTOM,
6904 x_start, xlen, y_start, ylen, z_start);
6905
6906 C->set_has_split_ifs(true); // Has chance for split-if optimization
6907 set_result(z);
6908 return true;
6909 }
6910
6911 //-------------inline_squareToLen------------------------------------
6912 bool LibraryCallKit::inline_squareToLen() {
6913 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6914
6915 address stubAddr = StubRoutines::squareToLen();
6916 if (stubAddr == nullptr) {
6917 return false; // Intrinsic's stub is not implemented on this platform
6918 }
6919 const char* stubName = "squareToLen";
6920
6921 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6922
6923 Node* x = argument(0);
6924 Node* len = argument(1);
6925 Node* z = argument(2);
6926 Node* zlen = argument(3);
6927
6928 x = must_be_not_null(x, true);
6929 z = must_be_not_null(z, true);
6930
6931 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6932 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6933 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6934 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6935 // failed array check
6936 return false;
6937 }
6938
6939 BasicType x_elem = x_type->elem()->array_element_basic_type();
6940 BasicType z_elem = z_type->elem()->array_element_basic_type();
6941 if (x_elem != T_INT || z_elem != T_INT) {
6942 return false;
6943 }
6944
6945
6946 Node* x_start = array_element_address(x, intcon(0), x_elem);
6947 Node* z_start = array_element_address(z, intcon(0), z_elem);
6948
6949 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6950 OptoRuntime::squareToLen_Type(),
6951 stubAddr, stubName, TypePtr::BOTTOM,
6952 x_start, len, z_start, zlen);
6953
6954 set_result(z);
6955 return true;
6956 }
6957
6958 //-------------inline_mulAdd------------------------------------------
6959 bool LibraryCallKit::inline_mulAdd() {
6960 assert(UseMulAddIntrinsic, "not implemented on this platform");
6961
6962 address stubAddr = StubRoutines::mulAdd();
6963 if (stubAddr == nullptr) {
6964 return false; // Intrinsic's stub is not implemented on this platform
6965 }
6966 const char* stubName = "mulAdd";
6967
6968 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6969
6970 Node* out = argument(0);
6971 Node* in = argument(1);
6972 Node* offset = argument(2);
6973 Node* len = argument(3);
6974 Node* k = argument(4);
6975
6976 in = must_be_not_null(in, true);
6977 out = must_be_not_null(out, true);
6978
6979 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6980 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6981 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6982 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6983 // failed array check
6984 return false;
6985 }
6986
6987 BasicType out_elem = out_type->elem()->array_element_basic_type();
6988 BasicType in_elem = in_type->elem()->array_element_basic_type();
6989 if (out_elem != T_INT || in_elem != T_INT) {
6990 return false;
6991 }
6992
6993 Node* outlen = load_array_length(out);
6994 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6995 Node* out_start = array_element_address(out, intcon(0), out_elem);
6996 Node* in_start = array_element_address(in, intcon(0), in_elem);
6997
6998 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6999 OptoRuntime::mulAdd_Type(),
7000 stubAddr, stubName, TypePtr::BOTTOM,
7001 out_start,in_start, new_offset, len, k);
7002 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7003 set_result(result);
7004 return true;
7005 }
7006
7007 //-------------inline_montgomeryMultiply-----------------------------------
7008 bool LibraryCallKit::inline_montgomeryMultiply() {
7009 address stubAddr = StubRoutines::montgomeryMultiply();
7010 if (stubAddr == nullptr) {
7011 return false; // Intrinsic's stub is not implemented on this platform
7012 }
7013
7014 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
7015 const char* stubName = "montgomery_multiply";
7016
7017 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
7018
7019 Node* a = argument(0);
7020 Node* b = argument(1);
7021 Node* n = argument(2);
7022 Node* len = argument(3);
7023 Node* inv = argument(4);
7024 Node* m = argument(6);
7025
7026 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
7027 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
7028 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
7029 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
7030 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
7031 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
7032 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
7033 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
7034 // failed array check
7035 return false;
7036 }
7037
7038 BasicType a_elem = a_type->elem()->array_element_basic_type();
7039 BasicType b_elem = b_type->elem()->array_element_basic_type();
7040 BasicType n_elem = n_type->elem()->array_element_basic_type();
7041 BasicType m_elem = m_type->elem()->array_element_basic_type();
7042 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
7043 return false;
7044 }
7045
7046 // Make the call
7047 {
7048 Node* a_start = array_element_address(a, intcon(0), a_elem);
7049 Node* b_start = array_element_address(b, intcon(0), b_elem);
7050 Node* n_start = array_element_address(n, intcon(0), n_elem);
7051 Node* m_start = array_element_address(m, intcon(0), m_elem);
7052
7053 Node* call = make_runtime_call(RC_LEAF,
7054 OptoRuntime::montgomeryMultiply_Type(),
7055 stubAddr, stubName, TypePtr::BOTTOM,
7056 a_start, b_start, n_start, len, inv, top(),
7057 m_start);
7058 set_result(m);
7059 }
7060
7061 return true;
7062 }
7063
7064 bool LibraryCallKit::inline_montgomerySquare() {
7065 address stubAddr = StubRoutines::montgomerySquare();
7066 if (stubAddr == nullptr) {
7067 return false; // Intrinsic's stub is not implemented on this platform
7068 }
7069
7070 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
7071 const char* stubName = "montgomery_square";
7072
7073 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
7074
7075 Node* a = argument(0);
7076 Node* n = argument(1);
7077 Node* len = argument(2);
7078 Node* inv = argument(3);
7079 Node* m = argument(5);
7080
7081 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
7082 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
7083 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
7084 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
7085 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
7086 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
7087 // failed array check
7088 return false;
7089 }
7090
7091 BasicType a_elem = a_type->elem()->array_element_basic_type();
7092 BasicType n_elem = n_type->elem()->array_element_basic_type();
7093 BasicType m_elem = m_type->elem()->array_element_basic_type();
7094 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
7095 return false;
7096 }
7097
7098 // Make the call
7099 {
7100 Node* a_start = array_element_address(a, intcon(0), a_elem);
7101 Node* n_start = array_element_address(n, intcon(0), n_elem);
7102 Node* m_start = array_element_address(m, intcon(0), m_elem);
7103
7104 Node* call = make_runtime_call(RC_LEAF,
7105 OptoRuntime::montgomerySquare_Type(),
7106 stubAddr, stubName, TypePtr::BOTTOM,
7107 a_start, n_start, len, inv, top(),
7108 m_start);
7109 set_result(m);
7110 }
7111
7112 return true;
7113 }
7114
7115 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
7116 address stubAddr = nullptr;
7117 const char* stubName = nullptr;
7118
7119 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
7120 if (stubAddr == nullptr) {
7121 return false; // Intrinsic's stub is not implemented on this platform
7122 }
7123
7124 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
7125
7126 assert(callee()->signature()->size() == 5, "expected 5 arguments");
7127
7128 Node* newArr = argument(0);
7129 Node* oldArr = argument(1);
7130 Node* newIdx = argument(2);
7131 Node* shiftCount = argument(3);
7132 Node* numIter = argument(4);
7133
7134 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
7135 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
7136 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
7137 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
7138 return false;
7139 }
7140
7141 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
7142 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
7143 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
7144 return false;
7145 }
7146
7147 // Make the call
7148 {
7149 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
7150 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
7151
7152 Node* call = make_runtime_call(RC_LEAF,
7153 OptoRuntime::bigIntegerShift_Type(),
7154 stubAddr,
7155 stubName,
7156 TypePtr::BOTTOM,
7157 newArr_start,
7158 oldArr_start,
7159 newIdx,
7160 shiftCount,
7161 numIter);
7162 }
7163
7164 return true;
7165 }
7166
7167 //-------------inline_vectorizedMismatch------------------------------
7168 bool LibraryCallKit::inline_vectorizedMismatch() {
7169 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
7170
7171 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
7172 Node* obja = argument(0); // Object
7173 Node* aoffset = argument(1); // long
7174 Node* objb = argument(3); // Object
7175 Node* boffset = argument(4); // long
7176 Node* length = argument(6); // int
7177 Node* scale = argument(7); // int
7178
7179 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
7180 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
7181 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
7182 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
7183 scale == top()) {
7184 return false; // failed input validation
7185 }
7186
7187 Node* obja_adr = make_unsafe_address(obja, aoffset);
7188 Node* objb_adr = make_unsafe_address(objb, boffset);
7189
7190 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
7191 //
7192 // inline_limit = ArrayOperationPartialInlineSize / element_size;
7193 // if (length <= inline_limit) {
7194 // inline_path:
7195 // vmask = VectorMaskGen length
7196 // vload1 = LoadVectorMasked obja, vmask
7197 // vload2 = LoadVectorMasked objb, vmask
7198 // result1 = VectorCmpMasked vload1, vload2, vmask
7199 // } else {
7200 // call_stub_path:
7201 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
7202 // }
7203 // exit_block:
7204 // return Phi(result1, result2);
7205 //
7206 enum { inline_path = 1, // input is small enough to process it all at once
7207 stub_path = 2, // input is too large; call into the VM
7208 PATH_LIMIT = 3
7209 };
7210
7211 Node* exit_block = new RegionNode(PATH_LIMIT);
7212 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
7213 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
7214
7215 Node* call_stub_path = control();
7216
7217 BasicType elem_bt = T_ILLEGAL;
7218
7219 const TypeInt* scale_t = _gvn.type(scale)->is_int();
7220 if (scale_t->is_con()) {
7221 switch (scale_t->get_con()) {
7222 case 0: elem_bt = T_BYTE; break;
7223 case 1: elem_bt = T_SHORT; break;
7224 case 2: elem_bt = T_INT; break;
7225 case 3: elem_bt = T_LONG; break;
7226
7227 default: elem_bt = T_ILLEGAL; break; // not supported
7228 }
7229 }
7230
7231 int inline_limit = 0;
7232 bool do_partial_inline = false;
7233
7234 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
7235 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
7236 do_partial_inline = inline_limit >= 16;
7237 }
7238
7239 if (do_partial_inline) {
7240 assert(elem_bt != T_ILLEGAL, "sanity");
7241
7242 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
7243 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
7244 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
7245
7246 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
7247 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
7248 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
7249
7250 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
7251
7252 if (!stopped()) {
7253 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
7254
7255 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
7256 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
7257 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
7258 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
7259
7260 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
7261 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
7262 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
7263 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
7264
7265 exit_block->init_req(inline_path, control());
7266 memory_phi->init_req(inline_path, map()->memory());
7267 result_phi->init_req(inline_path, result);
7268
7269 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
7270 clear_upper_avx();
7271 }
7272 }
7273 }
7274
7275 if (call_stub_path != nullptr) {
7276 set_control(call_stub_path);
7277
7278 Node* call = make_runtime_call(RC_LEAF,
7279 OptoRuntime::vectorizedMismatch_Type(),
7280 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
7281 obja_adr, objb_adr, length, scale);
7282
7283 exit_block->init_req(stub_path, control());
7284 memory_phi->init_req(stub_path, map()->memory());
7285 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
7286 }
7287
7288 exit_block = _gvn.transform(exit_block);
7289 memory_phi = _gvn.transform(memory_phi);
7290 result_phi = _gvn.transform(result_phi);
7291
7292 record_for_igvn(exit_block);
7293 record_for_igvn(memory_phi);
7294 record_for_igvn(result_phi);
7295
7296 set_control(exit_block);
7297 set_all_memory(memory_phi);
7298 set_result(result_phi);
7299
7300 return true;
7301 }
7302
7303 //------------------------------inline_vectorizedHashcode----------------------------
7304 bool LibraryCallKit::inline_vectorizedHashCode() {
7305 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
7306
7307 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
7308 Node* array = argument(0);
7309 Node* offset = argument(1);
7310 Node* length = argument(2);
7311 Node* initialValue = argument(3);
7312 Node* basic_type = argument(4);
7313
7314 if (basic_type == top()) {
7315 return false; // failed input validation
7316 }
7317
7318 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
7319 if (!basic_type_t->is_con()) {
7320 return false; // Only intrinsify if mode argument is constant
7321 }
7322
7323 array = must_be_not_null(array, true);
7324
7325 BasicType bt = (BasicType)basic_type_t->get_con();
7326
7327 // Resolve address of first element
7328 Node* array_start = array_element_address(array, offset, bt);
7329
7330 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
7331 array_start, length, initialValue, basic_type)));
7332 clear_upper_avx();
7333
7334 return true;
7335 }
7336
7337 /**
7338 * Calculate CRC32 for byte.
7339 * int java.util.zip.CRC32.update(int crc, int b)
7340 */
7341 bool LibraryCallKit::inline_updateCRC32() {
7342 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7343 assert(callee()->signature()->size() == 2, "update has 2 parameters");
7344 // no receiver since it is static method
7345 Node* crc = argument(0); // type: int
7346 Node* b = argument(1); // type: int
7347
7348 /*
7349 * int c = ~ crc;
7350 * b = timesXtoThe32[(b ^ c) & 0xFF];
7351 * b = b ^ (c >>> 8);
7352 * crc = ~b;
7353 */
7354
7355 Node* M1 = intcon(-1);
7356 crc = _gvn.transform(new XorINode(crc, M1));
7357 Node* result = _gvn.transform(new XorINode(crc, b));
7358 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
7359
7360 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
7361 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
7362 Node* adr = off_heap_plus_addr(base, ConvI2X(offset));
7363 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
7364
7365 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
7366 result = _gvn.transform(new XorINode(crc, result));
7367 result = _gvn.transform(new XorINode(result, M1));
7368 set_result(result);
7369 return true;
7370 }
7371
7372 /**
7373 * Calculate CRC32 for byte[] array.
7374 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
7375 */
7376 bool LibraryCallKit::inline_updateBytesCRC32() {
7377 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7378 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7379 // no receiver since it is static method
7380 Node* crc = argument(0); // type: int
7381 Node* src = argument(1); // type: oop
7382 Node* offset = argument(2); // type: int
7383 Node* length = argument(3); // type: int
7384
7385 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7386 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7387 // failed array check
7388 return false;
7389 }
7390
7391 // Figure out the size and type of the elements we will be copying.
7392 BasicType src_elem = src_type->elem()->array_element_basic_type();
7393 if (src_elem != T_BYTE) {
7394 return false;
7395 }
7396
7397 // 'src_start' points to src array + scaled offset
7398 src = must_be_not_null(src, true);
7399 Node* src_start = array_element_address(src, offset, src_elem);
7400
7401 // We assume that range check is done by caller.
7402 // TODO: generate range check (offset+length < src.length) in debug VM.
7403
7404 // Call the stub.
7405 address stubAddr = StubRoutines::updateBytesCRC32();
7406 const char *stubName = "updateBytesCRC32";
7407
7408 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7409 stubAddr, stubName, TypePtr::BOTTOM,
7410 crc, src_start, length);
7411 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7412 set_result(result);
7413 return true;
7414 }
7415
7416 /**
7417 * Calculate CRC32 for ByteBuffer.
7418 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
7419 */
7420 bool LibraryCallKit::inline_updateByteBufferCRC32() {
7421 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7422 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7423 // no receiver since it is static method
7424 Node* crc = argument(0); // type: int
7425 Node* src = argument(1); // type: long
7426 Node* offset = argument(3); // type: int
7427 Node* length = argument(4); // type: int
7428
7429 src = ConvL2X(src); // adjust Java long to machine word
7430 Node* base = _gvn.transform(new CastX2PNode(src));
7431 offset = ConvI2X(offset);
7432
7433 // 'src_start' points to src array + scaled offset
7434 Node* src_start = off_heap_plus_addr(base, offset);
7435
7436 // Call the stub.
7437 address stubAddr = StubRoutines::updateBytesCRC32();
7438 const char *stubName = "updateBytesCRC32";
7439
7440 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7441 stubAddr, stubName, TypePtr::BOTTOM,
7442 crc, src_start, length);
7443 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7444 set_result(result);
7445 return true;
7446 }
7447
7448 //------------------------------get_table_from_crc32c_class-----------------------
7449 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
7450 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
7451 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
7452
7453 return table;
7454 }
7455
7456 //------------------------------inline_updateBytesCRC32C-----------------------
7457 //
7458 // Calculate CRC32C for byte[] array.
7459 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
7460 //
7461 bool LibraryCallKit::inline_updateBytesCRC32C() {
7462 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7463 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7464 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7465 // no receiver since it is a static method
7466 Node* crc = argument(0); // type: int
7467 Node* src = argument(1); // type: oop
7468 Node* offset = argument(2); // type: int
7469 Node* end = argument(3); // type: int
7470
7471 Node* length = _gvn.transform(new SubINode(end, offset));
7472
7473 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7474 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7475 // failed array check
7476 return false;
7477 }
7478
7479 // Figure out the size and type of the elements we will be copying.
7480 BasicType src_elem = src_type->elem()->array_element_basic_type();
7481 if (src_elem != T_BYTE) {
7482 return false;
7483 }
7484
7485 // 'src_start' points to src array + scaled offset
7486 src = must_be_not_null(src, true);
7487 Node* src_start = array_element_address(src, offset, src_elem);
7488
7489 // static final int[] byteTable in class CRC32C
7490 Node* table = get_table_from_crc32c_class(callee()->holder());
7491 table = must_be_not_null(table, true);
7492 Node* table_start = array_element_address(table, intcon(0), T_INT);
7493
7494 // We assume that range check is done by caller.
7495 // TODO: generate range check (offset+length < src.length) in debug VM.
7496
7497 // Call the stub.
7498 address stubAddr = StubRoutines::updateBytesCRC32C();
7499 const char *stubName = "updateBytesCRC32C";
7500
7501 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7502 stubAddr, stubName, TypePtr::BOTTOM,
7503 crc, src_start, length, table_start);
7504 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7505 set_result(result);
7506 return true;
7507 }
7508
7509 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
7510 //
7511 // Calculate CRC32C for DirectByteBuffer.
7512 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
7513 //
7514 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
7515 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7516 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
7517 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7518 // no receiver since it is a static method
7519 Node* crc = argument(0); // type: int
7520 Node* src = argument(1); // type: long
7521 Node* offset = argument(3); // type: int
7522 Node* end = argument(4); // type: int
7523
7524 Node* length = _gvn.transform(new SubINode(end, offset));
7525
7526 src = ConvL2X(src); // adjust Java long to machine word
7527 Node* base = _gvn.transform(new CastX2PNode(src));
7528 offset = ConvI2X(offset);
7529
7530 // 'src_start' points to src array + scaled offset
7531 Node* src_start = off_heap_plus_addr(base, offset);
7532
7533 // static final int[] byteTable in class CRC32C
7534 Node* table = get_table_from_crc32c_class(callee()->holder());
7535 table = must_be_not_null(table, true);
7536 Node* table_start = array_element_address(table, intcon(0), T_INT);
7537
7538 // Call the stub.
7539 address stubAddr = StubRoutines::updateBytesCRC32C();
7540 const char *stubName = "updateBytesCRC32C";
7541
7542 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7543 stubAddr, stubName, TypePtr::BOTTOM,
7544 crc, src_start, length, table_start);
7545 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7546 set_result(result);
7547 return true;
7548 }
7549
7550 //------------------------------inline_updateBytesAdler32----------------------
7551 //
7552 // Calculate Adler32 checksum for byte[] array.
7553 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
7554 //
7555 bool LibraryCallKit::inline_updateBytesAdler32() {
7556 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7557 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7558 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7559 // no receiver since it is static method
7560 Node* crc = argument(0); // type: int
7561 Node* src = argument(1); // type: oop
7562 Node* offset = argument(2); // type: int
7563 Node* length = argument(3); // type: int
7564
7565 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7566 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7567 // failed array check
7568 return false;
7569 }
7570
7571 // Figure out the size and type of the elements we will be copying.
7572 BasicType src_elem = src_type->elem()->array_element_basic_type();
7573 if (src_elem != T_BYTE) {
7574 return false;
7575 }
7576
7577 // 'src_start' points to src array + scaled offset
7578 Node* src_start = array_element_address(src, offset, src_elem);
7579
7580 // We assume that range check is done by caller.
7581 // TODO: generate range check (offset+length < src.length) in debug VM.
7582
7583 // Call the stub.
7584 address stubAddr = StubRoutines::updateBytesAdler32();
7585 const char *stubName = "updateBytesAdler32";
7586
7587 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7588 stubAddr, stubName, TypePtr::BOTTOM,
7589 crc, src_start, length);
7590 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7591 set_result(result);
7592 return true;
7593 }
7594
7595 //------------------------------inline_updateByteBufferAdler32---------------
7596 //
7597 // Calculate Adler32 checksum for DirectByteBuffer.
7598 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
7599 //
7600 bool LibraryCallKit::inline_updateByteBufferAdler32() {
7601 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7602 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7603 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7604 // no receiver since it is static method
7605 Node* crc = argument(0); // type: int
7606 Node* src = argument(1); // type: long
7607 Node* offset = argument(3); // type: int
7608 Node* length = argument(4); // type: int
7609
7610 src = ConvL2X(src); // adjust Java long to machine word
7611 Node* base = _gvn.transform(new CastX2PNode(src));
7612 offset = ConvI2X(offset);
7613
7614 // 'src_start' points to src array + scaled offset
7615 Node* src_start = off_heap_plus_addr(base, offset);
7616
7617 // Call the stub.
7618 address stubAddr = StubRoutines::updateBytesAdler32();
7619 const char *stubName = "updateBytesAdler32";
7620
7621 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7622 stubAddr, stubName, TypePtr::BOTTOM,
7623 crc, src_start, length);
7624
7625 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7626 set_result(result);
7627 return true;
7628 }
7629
7630 //----------------------------inline_reference_get0----------------------------
7631 // public T java.lang.ref.Reference.get();
7632 bool LibraryCallKit::inline_reference_get0() {
7633 const int referent_offset = java_lang_ref_Reference::referent_offset();
7634
7635 // Get the argument:
7636 Node* reference_obj = null_check_receiver();
7637 if (stopped()) return true;
7638
7639 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7640 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7641 decorators, /*is_static*/ false, nullptr);
7642 if (result == nullptr) return false;
7643
7644 // Add memory barrier to prevent commoning reads from this field
7645 // across safepoint since GC can change its value.
7646 insert_mem_bar(Op_MemBarCPUOrder);
7647
7648 set_result(result);
7649 return true;
7650 }
7651
7652 //----------------------------inline_reference_refersTo0----------------------------
7653 // bool java.lang.ref.Reference.refersTo0();
7654 // bool java.lang.ref.PhantomReference.refersTo0();
7655 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7656 // Get arguments:
7657 Node* reference_obj = null_check_receiver();
7658 Node* other_obj = argument(1);
7659 if (stopped()) return true;
7660
7661 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7662 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7663 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7664 decorators, /*is_static*/ false, nullptr);
7665 if (referent == nullptr) return false;
7666
7667 // Add memory barrier to prevent commoning reads from this field
7668 // across safepoint since GC can change its value.
7669 insert_mem_bar(Op_MemBarCPUOrder);
7670
7671 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7672 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7673 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7674
7675 RegionNode* region = new RegionNode(3);
7676 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7677
7678 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7679 region->init_req(1, if_true);
7680 phi->init_req(1, intcon(1));
7681
7682 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7683 region->init_req(2, if_false);
7684 phi->init_req(2, intcon(0));
7685
7686 set_control(_gvn.transform(region));
7687 record_for_igvn(region);
7688 set_result(_gvn.transform(phi));
7689 return true;
7690 }
7691
7692 //----------------------------inline_reference_clear0----------------------------
7693 // void java.lang.ref.Reference.clear0();
7694 // void java.lang.ref.PhantomReference.clear0();
7695 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7696 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7697
7698 // Get arguments
7699 Node* reference_obj = null_check_receiver();
7700 if (stopped()) return true;
7701
7702 // Common access parameters
7703 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7704 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7705 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7706 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7707 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7708
7709 Node* referent = access_load_at(reference_obj,
7710 referent_field_addr,
7711 referent_field_addr_type,
7712 val_type,
7713 T_OBJECT,
7714 decorators);
7715
7716 IdealKit ideal(this);
7717 #define __ ideal.
7718 __ if_then(referent, BoolTest::ne, null());
7719 sync_kit(ideal);
7720 access_store_at(reference_obj,
7721 referent_field_addr,
7722 referent_field_addr_type,
7723 null(),
7724 val_type,
7725 T_OBJECT,
7726 decorators);
7727 __ sync_kit(this);
7728 __ end_if();
7729 final_sync(ideal);
7730 #undef __
7731
7732 return true;
7733 }
7734
7735 //-----------------------inline_reference_reachabilityFence-----------------
7736 // bool java.lang.ref.Reference.reachabilityFence();
7737 bool LibraryCallKit::inline_reference_reachabilityFence() {
7738 Node* referent = argument(0);
7739 insert_reachability_fence(referent);
7740 return true;
7741 }
7742
7743 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7744 DecoratorSet decorators, bool is_static,
7745 ciInstanceKlass* fromKls) {
7746 if (fromKls == nullptr) {
7747 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7748 assert(tinst != nullptr, "obj is null");
7749 assert(tinst->is_loaded(), "obj is not loaded");
7750 fromKls = tinst->instance_klass();
7751 } else {
7752 assert(is_static, "only for static field access");
7753 }
7754 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7755 ciSymbol::make(fieldTypeString),
7756 is_static);
7757
7758 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7759 if (field == nullptr) return (Node *) nullptr;
7760
7761 if (is_static) {
7762 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7763 fromObj = makecon(tip);
7764 }
7765
7766 // Next code copied from Parse::do_get_xxx():
7767
7768 // Compute address and memory type.
7769 int offset = field->offset_in_bytes();
7770 bool is_vol = field->is_volatile();
7771 ciType* field_klass = field->type();
7772 assert(field_klass->is_loaded(), "should be loaded");
7773 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7774 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7775 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7776 "slice of address and input slice don't match");
7777 BasicType bt = field->layout_type();
7778
7779 // Build the resultant type of the load
7780 const Type *type;
7781 if (bt == T_OBJECT) {
7782 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7783 } else {
7784 type = Type::get_const_basic_type(bt);
7785 }
7786
7787 if (is_vol) {
7788 decorators |= MO_SEQ_CST;
7789 }
7790
7791 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7792 }
7793
7794 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7795 bool is_exact /* true */, bool is_static /* false */,
7796 ciInstanceKlass * fromKls /* nullptr */) {
7797 if (fromKls == nullptr) {
7798 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7799 assert(tinst != nullptr, "obj is null");
7800 assert(tinst->is_loaded(), "obj is not loaded");
7801 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7802 fromKls = tinst->instance_klass();
7803 }
7804 else {
7805 assert(is_static, "only for static field access");
7806 }
7807 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7808 ciSymbol::make(fieldTypeString),
7809 is_static);
7810
7811 assert(field != nullptr, "undefined field");
7812 assert(!field->is_volatile(), "not defined for volatile fields");
7813
7814 if (is_static) {
7815 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7816 fromObj = makecon(tip);
7817 }
7818
7819 // Next code copied from Parse::do_get_xxx():
7820
7821 // Compute address and memory type.
7822 int offset = field->offset_in_bytes();
7823 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7824
7825 return adr;
7826 }
7827
7828 //------------------------------inline_aescrypt_Block-----------------------
7829 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7830 address stubAddr = nullptr;
7831 const char *stubName;
7832 bool is_decrypt = false;
7833 assert(UseAES, "need AES instruction support");
7834
7835 switch(id) {
7836 case vmIntrinsics::_aescrypt_encryptBlock:
7837 stubAddr = StubRoutines::aescrypt_encryptBlock();
7838 stubName = "aescrypt_encryptBlock";
7839 break;
7840 case vmIntrinsics::_aescrypt_decryptBlock:
7841 stubAddr = StubRoutines::aescrypt_decryptBlock();
7842 stubName = "aescrypt_decryptBlock";
7843 is_decrypt = true;
7844 break;
7845 default:
7846 break;
7847 }
7848 if (stubAddr == nullptr) return false;
7849
7850 Node* aescrypt_object = argument(0);
7851 Node* src = argument(1);
7852 Node* src_offset = argument(2);
7853 Node* dest = argument(3);
7854 Node* dest_offset = argument(4);
7855
7856 src = must_be_not_null(src, true);
7857 dest = must_be_not_null(dest, true);
7858
7859 // (1) src and dest are arrays.
7860 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7861 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7862 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7863 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7864
7865 // for the quick and dirty code we will skip all the checks.
7866 // we are just trying to get the call to be generated.
7867 Node* src_start = src;
7868 Node* dest_start = dest;
7869 if (src_offset != nullptr || dest_offset != nullptr) {
7870 assert(src_offset != nullptr && dest_offset != nullptr, "");
7871 src_start = array_element_address(src, src_offset, T_BYTE);
7872 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7873 }
7874
7875 // now need to get the start of its expanded key array
7876 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7877 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7878 if (k_start == nullptr) return false;
7879
7880 // Call the stub.
7881 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7882 stubAddr, stubName, TypePtr::BOTTOM,
7883 src_start, dest_start, k_start);
7884
7885 return true;
7886 }
7887
7888 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7889 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7890 address stubAddr = nullptr;
7891 const char *stubName = nullptr;
7892 bool is_decrypt = false;
7893 assert(UseAES, "need AES instruction support");
7894
7895 switch(id) {
7896 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7897 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7898 stubName = "cipherBlockChaining_encryptAESCrypt";
7899 break;
7900 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7901 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7902 stubName = "cipherBlockChaining_decryptAESCrypt";
7903 is_decrypt = true;
7904 break;
7905 default:
7906 break;
7907 }
7908 if (stubAddr == nullptr) return false;
7909
7910 Node* cipherBlockChaining_object = argument(0);
7911 Node* src = argument(1);
7912 Node* src_offset = argument(2);
7913 Node* len = argument(3);
7914 Node* dest = argument(4);
7915 Node* dest_offset = argument(5);
7916
7917 src = must_be_not_null(src, false);
7918 dest = must_be_not_null(dest, false);
7919
7920 // (1) src and dest are arrays.
7921 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7922 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7923 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7924 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7925
7926 // checks are the responsibility of the caller
7927 Node* src_start = src;
7928 Node* dest_start = dest;
7929 if (src_offset != nullptr || dest_offset != nullptr) {
7930 assert(src_offset != nullptr && dest_offset != nullptr, "");
7931 src_start = array_element_address(src, src_offset, T_BYTE);
7932 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7933 }
7934
7935 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7936 // (because of the predicated logic executed earlier).
7937 // so we cast it here safely.
7938 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7939
7940 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7941 if (embeddedCipherObj == nullptr) return false;
7942
7943 // cast it to what we know it will be at runtime
7944 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7945 assert(tinst != nullptr, "CBC obj is null");
7946 assert(tinst->is_loaded(), "CBC obj is not loaded");
7947 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7948 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7949
7950 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7951 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7952 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7953 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7954 aescrypt_object = _gvn.transform(aescrypt_object);
7955
7956 // we need to get the start of the aescrypt_object's expanded key array
7957 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7958 if (k_start == nullptr) return false;
7959
7960 // similarly, get the start address of the r vector
7961 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7962 if (objRvec == nullptr) return false;
7963 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7964
7965 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7966 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7967 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7968 stubAddr, stubName, TypePtr::BOTTOM,
7969 src_start, dest_start, k_start, r_start, len);
7970
7971 // return cipher length (int)
7972 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7973 set_result(retvalue);
7974 return true;
7975 }
7976
7977 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7978 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7979 address stubAddr = nullptr;
7980 const char *stubName = nullptr;
7981 bool is_decrypt = false;
7982 assert(UseAES, "need AES instruction support");
7983
7984 switch (id) {
7985 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7986 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7987 stubName = "electronicCodeBook_encryptAESCrypt";
7988 break;
7989 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7990 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7991 stubName = "electronicCodeBook_decryptAESCrypt";
7992 is_decrypt = true;
7993 break;
7994 default:
7995 break;
7996 }
7997
7998 if (stubAddr == nullptr) return false;
7999
8000 Node* electronicCodeBook_object = argument(0);
8001 Node* src = argument(1);
8002 Node* src_offset = argument(2);
8003 Node* len = argument(3);
8004 Node* dest = argument(4);
8005 Node* dest_offset = argument(5);
8006
8007 // (1) src and dest are arrays.
8008 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8009 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8010 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8011 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8012
8013 // checks are the responsibility of the caller
8014 Node* src_start = src;
8015 Node* dest_start = dest;
8016 if (src_offset != nullptr || dest_offset != nullptr) {
8017 assert(src_offset != nullptr && dest_offset != nullptr, "");
8018 src_start = array_element_address(src, src_offset, T_BYTE);
8019 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8020 }
8021
8022 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8023 // (because of the predicated logic executed earlier).
8024 // so we cast it here safely.
8025 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8026
8027 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8028 if (embeddedCipherObj == nullptr) return false;
8029
8030 // cast it to what we know it will be at runtime
8031 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
8032 assert(tinst != nullptr, "ECB obj is null");
8033 assert(tinst->is_loaded(), "ECB obj is not loaded");
8034 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8035 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8036
8037 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8038 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8039 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8040 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8041 aescrypt_object = _gvn.transform(aescrypt_object);
8042
8043 // we need to get the start of the aescrypt_object's expanded key array
8044 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
8045 if (k_start == nullptr) return false;
8046
8047 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8048 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
8049 OptoRuntime::electronicCodeBook_aescrypt_Type(),
8050 stubAddr, stubName, TypePtr::BOTTOM,
8051 src_start, dest_start, k_start, len);
8052
8053 // return cipher length (int)
8054 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
8055 set_result(retvalue);
8056 return true;
8057 }
8058
8059 //------------------------------inline_counterMode_AESCrypt-----------------------
8060 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
8061 assert(UseAES, "need AES instruction support");
8062 if (!UseAESCTRIntrinsics) return false;
8063
8064 address stubAddr = nullptr;
8065 const char *stubName = nullptr;
8066 if (id == vmIntrinsics::_counterMode_AESCrypt) {
8067 stubAddr = StubRoutines::counterMode_AESCrypt();
8068 stubName = "counterMode_AESCrypt";
8069 }
8070 if (stubAddr == nullptr) return false;
8071
8072 Node* counterMode_object = argument(0);
8073 Node* src = argument(1);
8074 Node* src_offset = argument(2);
8075 Node* len = argument(3);
8076 Node* dest = argument(4);
8077 Node* dest_offset = argument(5);
8078
8079 // (1) src and dest are arrays.
8080 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8081 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8082 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8083 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8084
8085 // checks are the responsibility of the caller
8086 Node* src_start = src;
8087 Node* dest_start = dest;
8088 if (src_offset != nullptr || dest_offset != nullptr) {
8089 assert(src_offset != nullptr && dest_offset != nullptr, "");
8090 src_start = array_element_address(src, src_offset, T_BYTE);
8091 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8092 }
8093
8094 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8095 // (because of the predicated logic executed earlier).
8096 // so we cast it here safely.
8097 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8098 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8099 if (embeddedCipherObj == nullptr) return false;
8100 // cast it to what we know it will be at runtime
8101 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
8102 assert(tinst != nullptr, "CTR obj is null");
8103 assert(tinst->is_loaded(), "CTR obj is not loaded");
8104 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8105 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8106 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8107 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8108 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8109 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8110 aescrypt_object = _gvn.transform(aescrypt_object);
8111 // we need to get the start of the aescrypt_object's expanded key array
8112 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
8113 if (k_start == nullptr) return false;
8114 // similarly, get the start address of the r vector
8115 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
8116 if (obj_counter == nullptr) return false;
8117 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
8118
8119 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
8120 if (saved_encCounter == nullptr) return false;
8121 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
8122 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
8123
8124 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8125 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8126 OptoRuntime::counterMode_aescrypt_Type(),
8127 stubAddr, stubName, TypePtr::BOTTOM,
8128 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
8129
8130 // return cipher length (int)
8131 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
8132 set_result(retvalue);
8133 return true;
8134 }
8135
8136 //------------------------------get_key_start_from_aescrypt_object-----------------------
8137 Node* LibraryCallKit::get_key_start_from_aescrypt_object(Node* aescrypt_object, bool is_decrypt) {
8138 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
8139 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
8140 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
8141 // The following platform specific stubs of encryption and decryption use the same round keys.
8142 #if defined(PPC64) || defined(S390) || defined(RISCV64)
8143 bool use_decryption_key = false;
8144 #else
8145 bool use_decryption_key = is_decrypt;
8146 #endif
8147 Node* objAESCryptKey = load_field_from_object(aescrypt_object, use_decryption_key ? "sessionKd" : "sessionKe", "[I");
8148 assert(objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AES_Crypt");
8149 if (objAESCryptKey == nullptr) return (Node *) nullptr;
8150
8151 // now have the array, need to get the start address of the selected key array
8152 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
8153 return k_start;
8154 }
8155
8156 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
8157 // Return node representing slow path of predicate check.
8158 // the pseudo code we want to emulate with this predicate is:
8159 // for encryption:
8160 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8161 // for decryption:
8162 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8163 // note cipher==plain is more conservative than the original java code but that's OK
8164 //
8165 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
8166 // The receiver was checked for null already.
8167 Node* objCBC = argument(0);
8168
8169 Node* src = argument(1);
8170 Node* dest = argument(4);
8171
8172 // Load embeddedCipher field of CipherBlockChaining object.
8173 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8174
8175 // get AESCrypt klass for instanceOf check
8176 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8177 // will have same classloader as CipherBlockChaining object
8178 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
8179 assert(tinst != nullptr, "CBCobj is null");
8180 assert(tinst->is_loaded(), "CBCobj is not loaded");
8181
8182 // we want to do an instanceof comparison against the AESCrypt class
8183 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8184 if (!klass_AESCrypt->is_loaded()) {
8185 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8186 Node* ctrl = control();
8187 set_control(top()); // no regular fast path
8188 return ctrl;
8189 }
8190
8191 src = must_be_not_null(src, true);
8192 dest = must_be_not_null(dest, true);
8193
8194 // Resolve oops to stable for CmpP below.
8195 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8196
8197 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8198 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8199 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8200
8201 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8202
8203 // for encryption, we are done
8204 if (!decrypting)
8205 return instof_false; // even if it is null
8206
8207 // for decryption, we need to add a further check to avoid
8208 // taking the intrinsic path when cipher and plain are the same
8209 // see the original java code for why.
8210 RegionNode* region = new RegionNode(3);
8211 region->init_req(1, instof_false);
8212
8213 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8214 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8215 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8216 region->init_req(2, src_dest_conjoint);
8217
8218 record_for_igvn(region);
8219 return _gvn.transform(region);
8220 }
8221
8222 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
8223 // Return node representing slow path of predicate check.
8224 // the pseudo code we want to emulate with this predicate is:
8225 // for encryption:
8226 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8227 // for decryption:
8228 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8229 // note cipher==plain is more conservative than the original java code but that's OK
8230 //
8231 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
8232 // The receiver was checked for null already.
8233 Node* objECB = argument(0);
8234
8235 // Load embeddedCipher field of ElectronicCodeBook object.
8236 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8237
8238 // get AESCrypt klass for instanceOf check
8239 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8240 // will have same classloader as ElectronicCodeBook object
8241 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
8242 assert(tinst != nullptr, "ECBobj is null");
8243 assert(tinst->is_loaded(), "ECBobj is not loaded");
8244
8245 // we want to do an instanceof comparison against the AESCrypt class
8246 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8247 if (!klass_AESCrypt->is_loaded()) {
8248 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8249 Node* ctrl = control();
8250 set_control(top()); // no regular fast path
8251 return ctrl;
8252 }
8253 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8254
8255 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8256 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8257 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8258
8259 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8260
8261 // for encryption, we are done
8262 if (!decrypting)
8263 return instof_false; // even if it is null
8264
8265 // for decryption, we need to add a further check to avoid
8266 // taking the intrinsic path when cipher and plain are the same
8267 // see the original java code for why.
8268 RegionNode* region = new RegionNode(3);
8269 region->init_req(1, instof_false);
8270 Node* src = argument(1);
8271 Node* dest = argument(4);
8272 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8273 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8274 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8275 region->init_req(2, src_dest_conjoint);
8276
8277 record_for_igvn(region);
8278 return _gvn.transform(region);
8279 }
8280
8281 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
8282 // Return node representing slow path of predicate check.
8283 // the pseudo code we want to emulate with this predicate is:
8284 // for encryption:
8285 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8286 // for decryption:
8287 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8288 // note cipher==plain is more conservative than the original java code but that's OK
8289 //
8290
8291 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
8292 // The receiver was checked for null already.
8293 Node* objCTR = argument(0);
8294
8295 // Load embeddedCipher field of CipherBlockChaining object.
8296 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8297
8298 // get AESCrypt klass for instanceOf check
8299 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8300 // will have same classloader as CipherBlockChaining object
8301 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
8302 assert(tinst != nullptr, "CTRobj is null");
8303 assert(tinst->is_loaded(), "CTRobj is not loaded");
8304
8305 // we want to do an instanceof comparison against the AESCrypt class
8306 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8307 if (!klass_AESCrypt->is_loaded()) {
8308 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8309 Node* ctrl = control();
8310 set_control(top()); // no regular fast path
8311 return ctrl;
8312 }
8313
8314 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8315 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8316 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8317 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8318 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8319
8320 return instof_false; // even if it is null
8321 }
8322
8323 //------------------------------inline_ghash_processBlocks
8324 bool LibraryCallKit::inline_ghash_processBlocks() {
8325 address stubAddr;
8326 const char *stubName;
8327 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
8328
8329 stubAddr = StubRoutines::ghash_processBlocks();
8330 stubName = "ghash_processBlocks";
8331
8332 Node* data = argument(0);
8333 Node* offset = argument(1);
8334 Node* len = argument(2);
8335 Node* state = argument(3);
8336 Node* subkeyH = argument(4);
8337
8338 state = must_be_not_null(state, true);
8339 subkeyH = must_be_not_null(subkeyH, true);
8340 data = must_be_not_null(data, true);
8341
8342 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8343 assert(state_start, "state is null");
8344 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
8345 assert(subkeyH_start, "subkeyH is null");
8346 Node* data_start = array_element_address(data, offset, T_BYTE);
8347 assert(data_start, "data is null");
8348
8349 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
8350 OptoRuntime::ghash_processBlocks_Type(),
8351 stubAddr, stubName, TypePtr::BOTTOM,
8352 state_start, subkeyH_start, data_start, len);
8353 return true;
8354 }
8355
8356 //------------------------------inline_chacha20Block
8357 bool LibraryCallKit::inline_chacha20Block() {
8358 address stubAddr;
8359 const char *stubName;
8360 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
8361
8362 stubAddr = StubRoutines::chacha20Block();
8363 stubName = "chacha20Block";
8364
8365 Node* state = argument(0);
8366 Node* result = argument(1);
8367
8368 state = must_be_not_null(state, true);
8369 result = must_be_not_null(result, true);
8370
8371 Node* state_start = array_element_address(state, intcon(0), T_INT);
8372 assert(state_start, "state is null");
8373 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
8374 assert(result_start, "result is null");
8375
8376 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
8377 OptoRuntime::chacha20Block_Type(),
8378 stubAddr, stubName, TypePtr::BOTTOM,
8379 state_start, result_start);
8380 // return key stream length (int)
8381 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
8382 set_result(retvalue);
8383 return true;
8384 }
8385
8386 //------------------------------inline_kyberNtt
8387 bool LibraryCallKit::inline_kyberNtt() {
8388 address stubAddr;
8389 const char *stubName;
8390 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8391 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
8392
8393 stubAddr = StubRoutines::kyberNtt();
8394 stubName = "kyberNtt";
8395 if (!stubAddr) return false;
8396
8397 Node* coeffs = argument(0);
8398 Node* ntt_zetas = argument(1);
8399
8400 coeffs = must_be_not_null(coeffs, true);
8401 ntt_zetas = must_be_not_null(ntt_zetas, true);
8402
8403 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8404 assert(coeffs_start, "coeffs is null");
8405 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
8406 assert(ntt_zetas_start, "ntt_zetas is null");
8407 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8408 OptoRuntime::kyberNtt_Type(),
8409 stubAddr, stubName, TypePtr::BOTTOM,
8410 coeffs_start, ntt_zetas_start);
8411 // return an int
8412 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
8413 set_result(retvalue);
8414 return true;
8415 }
8416
8417 //------------------------------inline_kyberInverseNtt
8418 bool LibraryCallKit::inline_kyberInverseNtt() {
8419 address stubAddr;
8420 const char *stubName;
8421 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8422 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
8423
8424 stubAddr = StubRoutines::kyberInverseNtt();
8425 stubName = "kyberInverseNtt";
8426 if (!stubAddr) return false;
8427
8428 Node* coeffs = argument(0);
8429 Node* zetas = argument(1);
8430
8431 coeffs = must_be_not_null(coeffs, true);
8432 zetas = must_be_not_null(zetas, true);
8433
8434 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8435 assert(coeffs_start, "coeffs is null");
8436 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8437 assert(zetas_start, "inverseNtt_zetas is null");
8438 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8439 OptoRuntime::kyberInverseNtt_Type(),
8440 stubAddr, stubName, TypePtr::BOTTOM,
8441 coeffs_start, zetas_start);
8442
8443 // return an int
8444 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
8445 set_result(retvalue);
8446 return true;
8447 }
8448
8449 //------------------------------inline_kyberNttMult
8450 bool LibraryCallKit::inline_kyberNttMult() {
8451 address stubAddr;
8452 const char *stubName;
8453 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8454 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
8455
8456 stubAddr = StubRoutines::kyberNttMult();
8457 stubName = "kyberNttMult";
8458 if (!stubAddr) return false;
8459
8460 Node* result = argument(0);
8461 Node* ntta = argument(1);
8462 Node* nttb = argument(2);
8463 Node* zetas = argument(3);
8464
8465 result = must_be_not_null(result, true);
8466 ntta = must_be_not_null(ntta, true);
8467 nttb = must_be_not_null(nttb, true);
8468 zetas = must_be_not_null(zetas, true);
8469
8470 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8471 assert(result_start, "result is null");
8472 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
8473 assert(ntta_start, "ntta is null");
8474 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
8475 assert(nttb_start, "nttb is null");
8476 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8477 assert(zetas_start, "nttMult_zetas is null");
8478 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8479 OptoRuntime::kyberNttMult_Type(),
8480 stubAddr, stubName, TypePtr::BOTTOM,
8481 result_start, ntta_start, nttb_start,
8482 zetas_start);
8483
8484 // return an int
8485 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
8486 set_result(retvalue);
8487
8488 return true;
8489 }
8490
8491 //------------------------------inline_kyberAddPoly_2
8492 bool LibraryCallKit::inline_kyberAddPoly_2() {
8493 address stubAddr;
8494 const char *stubName;
8495 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8496 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
8497
8498 stubAddr = StubRoutines::kyberAddPoly_2();
8499 stubName = "kyberAddPoly_2";
8500 if (!stubAddr) return false;
8501
8502 Node* result = argument(0);
8503 Node* a = argument(1);
8504 Node* b = argument(2);
8505
8506 result = must_be_not_null(result, true);
8507 a = must_be_not_null(a, true);
8508 b = must_be_not_null(b, true);
8509
8510 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8511 assert(result_start, "result is null");
8512 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8513 assert(a_start, "a is null");
8514 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8515 assert(b_start, "b is null");
8516 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
8517 OptoRuntime::kyberAddPoly_2_Type(),
8518 stubAddr, stubName, TypePtr::BOTTOM,
8519 result_start, a_start, b_start);
8520 // return an int
8521 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
8522 set_result(retvalue);
8523 return true;
8524 }
8525
8526 //------------------------------inline_kyberAddPoly_3
8527 bool LibraryCallKit::inline_kyberAddPoly_3() {
8528 address stubAddr;
8529 const char *stubName;
8530 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8531 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
8532
8533 stubAddr = StubRoutines::kyberAddPoly_3();
8534 stubName = "kyberAddPoly_3";
8535 if (!stubAddr) return false;
8536
8537 Node* result = argument(0);
8538 Node* a = argument(1);
8539 Node* b = argument(2);
8540 Node* c = argument(3);
8541
8542 result = must_be_not_null(result, true);
8543 a = must_be_not_null(a, true);
8544 b = must_be_not_null(b, true);
8545 c = must_be_not_null(c, true);
8546
8547 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8548 assert(result_start, "result is null");
8549 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8550 assert(a_start, "a is null");
8551 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8552 assert(b_start, "b is null");
8553 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
8554 assert(c_start, "c is null");
8555 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
8556 OptoRuntime::kyberAddPoly_3_Type(),
8557 stubAddr, stubName, TypePtr::BOTTOM,
8558 result_start, a_start, b_start, c_start);
8559 // return an int
8560 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
8561 set_result(retvalue);
8562 return true;
8563 }
8564
8565 //------------------------------inline_kyber12To16
8566 bool LibraryCallKit::inline_kyber12To16() {
8567 address stubAddr;
8568 const char *stubName;
8569 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8570 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
8571
8572 stubAddr = StubRoutines::kyber12To16();
8573 stubName = "kyber12To16";
8574 if (!stubAddr) return false;
8575
8576 Node* condensed = argument(0);
8577 Node* condensedOffs = argument(1);
8578 Node* parsed = argument(2);
8579 Node* parsedLength = argument(3);
8580
8581 condensed = must_be_not_null(condensed, true);
8582 parsed = must_be_not_null(parsed, true);
8583
8584 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
8585 assert(condensed_start, "condensed is null");
8586 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
8587 assert(parsed_start, "parsed is null");
8588 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
8589 OptoRuntime::kyber12To16_Type(),
8590 stubAddr, stubName, TypePtr::BOTTOM,
8591 condensed_start, condensedOffs, parsed_start, parsedLength);
8592 // return an int
8593 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
8594 set_result(retvalue);
8595 return true;
8596
8597 }
8598
8599 //------------------------------inline_kyberBarrettReduce
8600 bool LibraryCallKit::inline_kyberBarrettReduce() {
8601 address stubAddr;
8602 const char *stubName;
8603 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8604 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
8605
8606 stubAddr = StubRoutines::kyberBarrettReduce();
8607 stubName = "kyberBarrettReduce";
8608 if (!stubAddr) return false;
8609
8610 Node* coeffs = argument(0);
8611
8612 coeffs = must_be_not_null(coeffs, true);
8613
8614 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8615 assert(coeffs_start, "coeffs is null");
8616 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
8617 OptoRuntime::kyberBarrettReduce_Type(),
8618 stubAddr, stubName, TypePtr::BOTTOM,
8619 coeffs_start);
8620 // return an int
8621 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
8622 set_result(retvalue);
8623 return true;
8624 }
8625
8626 //------------------------------inline_dilithiumAlmostNtt
8627 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
8628 address stubAddr;
8629 const char *stubName;
8630 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8631 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
8632
8633 stubAddr = StubRoutines::dilithiumAlmostNtt();
8634 stubName = "dilithiumAlmostNtt";
8635 if (!stubAddr) return false;
8636
8637 Node* coeffs = argument(0);
8638 Node* ntt_zetas = argument(1);
8639
8640 coeffs = must_be_not_null(coeffs, true);
8641 ntt_zetas = must_be_not_null(ntt_zetas, true);
8642
8643 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8644 assert(coeffs_start, "coeffs is null");
8645 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
8646 assert(ntt_zetas_start, "ntt_zetas is null");
8647 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8648 OptoRuntime::dilithiumAlmostNtt_Type(),
8649 stubAddr, stubName, TypePtr::BOTTOM,
8650 coeffs_start, ntt_zetas_start);
8651 // return an int
8652 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
8653 set_result(retvalue);
8654 return true;
8655 }
8656
8657 //------------------------------inline_dilithiumAlmostInverseNtt
8658 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
8659 address stubAddr;
8660 const char *stubName;
8661 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8662 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
8663
8664 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
8665 stubName = "dilithiumAlmostInverseNtt";
8666 if (!stubAddr) return false;
8667
8668 Node* coeffs = argument(0);
8669 Node* zetas = argument(1);
8670
8671 coeffs = must_be_not_null(coeffs, true);
8672 zetas = must_be_not_null(zetas, true);
8673
8674 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8675 assert(coeffs_start, "coeffs is null");
8676 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
8677 assert(zetas_start, "inverseNtt_zetas is null");
8678 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8679 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
8680 stubAddr, stubName, TypePtr::BOTTOM,
8681 coeffs_start, zetas_start);
8682 // return an int
8683 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
8684 set_result(retvalue);
8685 return true;
8686 }
8687
8688 //------------------------------inline_dilithiumNttMult
8689 bool LibraryCallKit::inline_dilithiumNttMult() {
8690 address stubAddr;
8691 const char *stubName;
8692 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8693 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
8694
8695 stubAddr = StubRoutines::dilithiumNttMult();
8696 stubName = "dilithiumNttMult";
8697 if (!stubAddr) return false;
8698
8699 Node* result = argument(0);
8700 Node* ntta = argument(1);
8701 Node* nttb = argument(2);
8702 Node* zetas = argument(3);
8703
8704 result = must_be_not_null(result, true);
8705 ntta = must_be_not_null(ntta, true);
8706 nttb = must_be_not_null(nttb, true);
8707 zetas = must_be_not_null(zetas, true);
8708
8709 Node* result_start = array_element_address(result, intcon(0), T_INT);
8710 assert(result_start, "result is null");
8711 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
8712 assert(ntta_start, "ntta is null");
8713 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
8714 assert(nttb_start, "nttb is null");
8715 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8716 OptoRuntime::dilithiumNttMult_Type(),
8717 stubAddr, stubName, TypePtr::BOTTOM,
8718 result_start, ntta_start, nttb_start);
8719
8720 // return an int
8721 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8722 set_result(retvalue);
8723
8724 return true;
8725 }
8726
8727 //------------------------------inline_dilithiumMontMulByConstant
8728 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8729 address stubAddr;
8730 const char *stubName;
8731 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8732 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8733
8734 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8735 stubName = "dilithiumMontMulByConstant";
8736 if (!stubAddr) return false;
8737
8738 Node* coeffs = argument(0);
8739 Node* constant = argument(1);
8740
8741 coeffs = must_be_not_null(coeffs, true);
8742
8743 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8744 assert(coeffs_start, "coeffs is null");
8745 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8746 OptoRuntime::dilithiumMontMulByConstant_Type(),
8747 stubAddr, stubName, TypePtr::BOTTOM,
8748 coeffs_start, constant);
8749
8750 // return an int
8751 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8752 set_result(retvalue);
8753 return true;
8754 }
8755
8756
8757 //------------------------------inline_dilithiumDecomposePoly
8758 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8759 address stubAddr;
8760 const char *stubName;
8761 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8762 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8763
8764 stubAddr = StubRoutines::dilithiumDecomposePoly();
8765 stubName = "dilithiumDecomposePoly";
8766 if (!stubAddr) return false;
8767
8768 Node* input = argument(0);
8769 Node* lowPart = argument(1);
8770 Node* highPart = argument(2);
8771 Node* twoGamma2 = argument(3);
8772 Node* multiplier = argument(4);
8773
8774 input = must_be_not_null(input, true);
8775 lowPart = must_be_not_null(lowPart, true);
8776 highPart = must_be_not_null(highPart, true);
8777
8778 Node* input_start = array_element_address(input, intcon(0), T_INT);
8779 assert(input_start, "input is null");
8780 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8781 assert(lowPart_start, "lowPart is null");
8782 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8783 assert(highPart_start, "highPart is null");
8784
8785 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8786 OptoRuntime::dilithiumDecomposePoly_Type(),
8787 stubAddr, stubName, TypePtr::BOTTOM,
8788 input_start, lowPart_start, highPart_start,
8789 twoGamma2, multiplier);
8790
8791 // return an int
8792 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8793 set_result(retvalue);
8794 return true;
8795 }
8796
8797 bool LibraryCallKit::inline_base64_encodeBlock() {
8798 address stubAddr;
8799 const char *stubName;
8800 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8801 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8802 stubAddr = StubRoutines::base64_encodeBlock();
8803 stubName = "encodeBlock";
8804
8805 if (!stubAddr) return false;
8806 Node* base64obj = argument(0);
8807 Node* src = argument(1);
8808 Node* offset = argument(2);
8809 Node* len = argument(3);
8810 Node* dest = argument(4);
8811 Node* dp = argument(5);
8812 Node* isURL = argument(6);
8813
8814 src = must_be_not_null(src, true);
8815 dest = must_be_not_null(dest, true);
8816
8817 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8818 assert(src_start, "source array is null");
8819 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8820 assert(dest_start, "destination array is null");
8821
8822 Node* base64 = make_runtime_call(RC_LEAF,
8823 OptoRuntime::base64_encodeBlock_Type(),
8824 stubAddr, stubName, TypePtr::BOTTOM,
8825 src_start, offset, len, dest_start, dp, isURL);
8826 return true;
8827 }
8828
8829 bool LibraryCallKit::inline_base64_decodeBlock() {
8830 address stubAddr;
8831 const char *stubName;
8832 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8833 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8834 stubAddr = StubRoutines::base64_decodeBlock();
8835 stubName = "decodeBlock";
8836
8837 if (!stubAddr) return false;
8838 Node* base64obj = argument(0);
8839 Node* src = argument(1);
8840 Node* src_offset = argument(2);
8841 Node* len = argument(3);
8842 Node* dest = argument(4);
8843 Node* dest_offset = argument(5);
8844 Node* isURL = argument(6);
8845 Node* isMIME = argument(7);
8846
8847 src = must_be_not_null(src, true);
8848 dest = must_be_not_null(dest, true);
8849
8850 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8851 assert(src_start, "source array is null");
8852 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8853 assert(dest_start, "destination array is null");
8854
8855 Node* call = make_runtime_call(RC_LEAF,
8856 OptoRuntime::base64_decodeBlock_Type(),
8857 stubAddr, stubName, TypePtr::BOTTOM,
8858 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8859 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8860 set_result(result);
8861 return true;
8862 }
8863
8864 bool LibraryCallKit::inline_poly1305_processBlocks() {
8865 address stubAddr;
8866 const char *stubName;
8867 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8868 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8869 stubAddr = StubRoutines::poly1305_processBlocks();
8870 stubName = "poly1305_processBlocks";
8871
8872 if (!stubAddr) return false;
8873 null_check_receiver(); // null-check receiver
8874 if (stopped()) return true;
8875
8876 Node* input = argument(1);
8877 Node* input_offset = argument(2);
8878 Node* len = argument(3);
8879 Node* alimbs = argument(4);
8880 Node* rlimbs = argument(5);
8881
8882 input = must_be_not_null(input, true);
8883 alimbs = must_be_not_null(alimbs, true);
8884 rlimbs = must_be_not_null(rlimbs, true);
8885
8886 Node* input_start = array_element_address(input, input_offset, T_BYTE);
8887 assert(input_start, "input array is null");
8888 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
8889 assert(acc_start, "acc array is null");
8890 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
8891 assert(r_start, "r array is null");
8892
8893 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8894 OptoRuntime::poly1305_processBlocks_Type(),
8895 stubAddr, stubName, TypePtr::BOTTOM,
8896 input_start, len, acc_start, r_start);
8897 return true;
8898 }
8899
8900 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
8901 address stubAddr;
8902 const char *stubName;
8903 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8904 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
8905 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
8906 stubName = "intpoly_montgomeryMult_P256";
8907
8908 if (!stubAddr) return false;
8909 null_check_receiver(); // null-check receiver
8910 if (stopped()) return true;
8911
8912 Node* a = argument(1);
8913 Node* b = argument(2);
8914 Node* r = argument(3);
8915
8916 a = must_be_not_null(a, true);
8917 b = must_be_not_null(b, true);
8918 r = must_be_not_null(r, true);
8919
8920 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8921 assert(a_start, "a array is null");
8922 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8923 assert(b_start, "b array is null");
8924 Node* r_start = array_element_address(r, intcon(0), T_LONG);
8925 assert(r_start, "r array is null");
8926
8927 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8928 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
8929 stubAddr, stubName, TypePtr::BOTTOM,
8930 a_start, b_start, r_start);
8931 return true;
8932 }
8933
8934 bool LibraryCallKit::inline_intpoly_assign() {
8935 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8936 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
8937 const char *stubName = "intpoly_assign";
8938 address stubAddr = StubRoutines::intpoly_assign();
8939 if (!stubAddr) return false;
8940
8941 Node* set = argument(0);
8942 Node* a = argument(1);
8943 Node* b = argument(2);
8944 Node* arr_length = load_array_length(a);
8945
8946 a = must_be_not_null(a, true);
8947 b = must_be_not_null(b, true);
8948
8949 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8950 assert(a_start, "a array is null");
8951 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8952 assert(b_start, "b array is null");
8953
8954 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8955 OptoRuntime::intpoly_assign_Type(),
8956 stubAddr, stubName, TypePtr::BOTTOM,
8957 set, a_start, b_start, arr_length);
8958 return true;
8959 }
8960
8961 //------------------------------inline_digestBase_implCompress-----------------------
8962 //
8963 // Calculate MD5 for single-block byte[] array.
8964 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
8965 //
8966 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
8967 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
8968 //
8969 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
8970 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
8971 //
8972 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
8973 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
8974 //
8975 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
8976 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
8977 //
8978 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
8979 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
8980
8981 Node* digestBase_obj = argument(0);
8982 Node* src = argument(1); // type oop
8983 Node* ofs = argument(2); // type int
8984
8985 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8986 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8987 // failed array check
8988 return false;
8989 }
8990 // Figure out the size and type of the elements we will be copying.
8991 BasicType src_elem = src_type->elem()->array_element_basic_type();
8992 if (src_elem != T_BYTE) {
8993 return false;
8994 }
8995 // 'src_start' points to src array + offset
8996 src = must_be_not_null(src, true);
8997 Node* src_start = array_element_address(src, ofs, src_elem);
8998 Node* state = nullptr;
8999 Node* block_size = nullptr;
9000 address stubAddr;
9001 const char *stubName;
9002
9003 switch(id) {
9004 case vmIntrinsics::_md5_implCompress:
9005 assert(UseMD5Intrinsics, "need MD5 instruction support");
9006 state = get_state_from_digest_object(digestBase_obj, T_INT);
9007 stubAddr = StubRoutines::md5_implCompress();
9008 stubName = "md5_implCompress";
9009 break;
9010 case vmIntrinsics::_sha_implCompress:
9011 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
9012 state = get_state_from_digest_object(digestBase_obj, T_INT);
9013 stubAddr = StubRoutines::sha1_implCompress();
9014 stubName = "sha1_implCompress";
9015 break;
9016 case vmIntrinsics::_sha2_implCompress:
9017 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
9018 state = get_state_from_digest_object(digestBase_obj, T_INT);
9019 stubAddr = StubRoutines::sha256_implCompress();
9020 stubName = "sha256_implCompress";
9021 break;
9022 case vmIntrinsics::_sha5_implCompress:
9023 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
9024 state = get_state_from_digest_object(digestBase_obj, T_LONG);
9025 stubAddr = StubRoutines::sha512_implCompress();
9026 stubName = "sha512_implCompress";
9027 break;
9028 case vmIntrinsics::_sha3_implCompress:
9029 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
9030 state = get_state_from_digest_object(digestBase_obj, T_LONG);
9031 stubAddr = StubRoutines::sha3_implCompress();
9032 stubName = "sha3_implCompress";
9033 block_size = get_block_size_from_digest_object(digestBase_obj);
9034 if (block_size == nullptr) return false;
9035 break;
9036 default:
9037 fatal_unexpected_iid(id);
9038 return false;
9039 }
9040 if (state == nullptr) return false;
9041
9042 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
9043 if (stubAddr == nullptr) return false;
9044
9045 // Call the stub.
9046 Node* call;
9047 if (block_size == nullptr) {
9048 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
9049 stubAddr, stubName, TypePtr::BOTTOM,
9050 src_start, state);
9051 } else {
9052 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
9053 stubAddr, stubName, TypePtr::BOTTOM,
9054 src_start, state, block_size);
9055 }
9056
9057 return true;
9058 }
9059
9060 //------------------------------inline_double_keccak
9061 bool LibraryCallKit::inline_double_keccak() {
9062 address stubAddr;
9063 const char *stubName;
9064 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
9065 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
9066
9067 stubAddr = StubRoutines::double_keccak();
9068 stubName = "double_keccak";
9069 if (!stubAddr) return false;
9070
9071 Node* status0 = argument(0);
9072 Node* status1 = argument(1);
9073
9074 status0 = must_be_not_null(status0, true);
9075 status1 = must_be_not_null(status1, true);
9076
9077 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
9078 assert(status0_start, "status0 is null");
9079 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
9080 assert(status1_start, "status1 is null");
9081 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
9082 OptoRuntime::double_keccak_Type(),
9083 stubAddr, stubName, TypePtr::BOTTOM,
9084 status0_start, status1_start);
9085 // return an int
9086 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
9087 set_result(retvalue);
9088 return true;
9089 }
9090
9091
9092 //------------------------------inline_digestBase_implCompressMB-----------------------
9093 //
9094 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
9095 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
9096 //
9097 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
9098 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9099 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9100 assert((uint)predicate < 5, "sanity");
9101 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
9102
9103 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
9104 Node* src = argument(1); // byte[] array
9105 Node* ofs = argument(2); // type int
9106 Node* limit = argument(3); // type int
9107
9108 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
9109 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
9110 // failed array check
9111 return false;
9112 }
9113 // Figure out the size and type of the elements we will be copying.
9114 BasicType src_elem = src_type->elem()->array_element_basic_type();
9115 if (src_elem != T_BYTE) {
9116 return false;
9117 }
9118 // 'src_start' points to src array + offset
9119 src = must_be_not_null(src, false);
9120 Node* src_start = array_element_address(src, ofs, src_elem);
9121
9122 const char* klass_digestBase_name = nullptr;
9123 const char* stub_name = nullptr;
9124 address stub_addr = nullptr;
9125 BasicType elem_type = T_INT;
9126
9127 switch (predicate) {
9128 case 0:
9129 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
9130 klass_digestBase_name = "sun/security/provider/MD5";
9131 stub_name = "md5_implCompressMB";
9132 stub_addr = StubRoutines::md5_implCompressMB();
9133 }
9134 break;
9135 case 1:
9136 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
9137 klass_digestBase_name = "sun/security/provider/SHA";
9138 stub_name = "sha1_implCompressMB";
9139 stub_addr = StubRoutines::sha1_implCompressMB();
9140 }
9141 break;
9142 case 2:
9143 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
9144 klass_digestBase_name = "sun/security/provider/SHA2";
9145 stub_name = "sha256_implCompressMB";
9146 stub_addr = StubRoutines::sha256_implCompressMB();
9147 }
9148 break;
9149 case 3:
9150 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
9151 klass_digestBase_name = "sun/security/provider/SHA5";
9152 stub_name = "sha512_implCompressMB";
9153 stub_addr = StubRoutines::sha512_implCompressMB();
9154 elem_type = T_LONG;
9155 }
9156 break;
9157 case 4:
9158 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
9159 klass_digestBase_name = "sun/security/provider/SHA3";
9160 stub_name = "sha3_implCompressMB";
9161 stub_addr = StubRoutines::sha3_implCompressMB();
9162 elem_type = T_LONG;
9163 }
9164 break;
9165 default:
9166 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
9167 }
9168 if (klass_digestBase_name != nullptr) {
9169 assert(stub_addr != nullptr, "Stub is generated");
9170 if (stub_addr == nullptr) return false;
9171
9172 // get DigestBase klass to lookup for SHA klass
9173 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
9174 assert(tinst != nullptr, "digestBase_obj is not instance???");
9175 assert(tinst->is_loaded(), "DigestBase is not loaded");
9176
9177 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
9178 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
9179 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
9180 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
9181 }
9182 return false;
9183 }
9184
9185 //------------------------------inline_digestBase_implCompressMB-----------------------
9186 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
9187 BasicType elem_type, address stubAddr, const char *stubName,
9188 Node* src_start, Node* ofs, Node* limit) {
9189 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
9190 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
9191 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
9192 digest_obj = _gvn.transform(digest_obj);
9193
9194 Node* state = get_state_from_digest_object(digest_obj, elem_type);
9195 if (state == nullptr) return false;
9196
9197 Node* block_size = nullptr;
9198 if (strcmp("sha3_implCompressMB", stubName) == 0) {
9199 block_size = get_block_size_from_digest_object(digest_obj);
9200 if (block_size == nullptr) return false;
9201 }
9202
9203 // Call the stub.
9204 Node* call;
9205 if (block_size == nullptr) {
9206 call = make_runtime_call(RC_LEAF|RC_NO_FP,
9207 OptoRuntime::digestBase_implCompressMB_Type(false),
9208 stubAddr, stubName, TypePtr::BOTTOM,
9209 src_start, state, ofs, limit);
9210 } else {
9211 call = make_runtime_call(RC_LEAF|RC_NO_FP,
9212 OptoRuntime::digestBase_implCompressMB_Type(true),
9213 stubAddr, stubName, TypePtr::BOTTOM,
9214 src_start, state, block_size, ofs, limit);
9215 }
9216
9217 // return ofs (int)
9218 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
9219 set_result(result);
9220
9221 return true;
9222 }
9223
9224 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
9225 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
9226 assert(UseAES, "need AES instruction support");
9227 address stubAddr = nullptr;
9228 const char *stubName = nullptr;
9229 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
9230 stubName = "galoisCounterMode_AESCrypt";
9231
9232 if (stubAddr == nullptr) return false;
9233
9234 Node* in = argument(0);
9235 Node* inOfs = argument(1);
9236 Node* len = argument(2);
9237 Node* ct = argument(3);
9238 Node* ctOfs = argument(4);
9239 Node* out = argument(5);
9240 Node* outOfs = argument(6);
9241 Node* gctr_object = argument(7);
9242 Node* ghash_object = argument(8);
9243
9244 // (1) in, ct and out are arrays.
9245 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
9246 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
9247 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
9248 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
9249 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
9250 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
9251
9252 // checks are the responsibility of the caller
9253 Node* in_start = in;
9254 Node* ct_start = ct;
9255 Node* out_start = out;
9256 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
9257 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
9258 in_start = array_element_address(in, inOfs, T_BYTE);
9259 ct_start = array_element_address(ct, ctOfs, T_BYTE);
9260 out_start = array_element_address(out, outOfs, T_BYTE);
9261 }
9262
9263 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
9264 // (because of the predicated logic executed earlier).
9265 // so we cast it here safely.
9266 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
9267 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9268 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
9269 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
9270 Node* state = load_field_from_object(ghash_object, "state", "[J");
9271
9272 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
9273 return false;
9274 }
9275 // cast it to what we know it will be at runtime
9276 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
9277 assert(tinst != nullptr, "GCTR obj is null");
9278 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9279 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
9280 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
9281 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9282 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
9283 const TypeOopPtr* xtype = aklass->as_instance_type();
9284 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
9285 aescrypt_object = _gvn.transform(aescrypt_object);
9286 // we need to get the start of the aescrypt_object's expanded key array
9287 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
9288 if (k_start == nullptr) return false;
9289 // similarly, get the start address of the r vector
9290 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
9291 Node* state_start = array_element_address(state, intcon(0), T_LONG);
9292 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
9293
9294
9295 // Call the stub, passing params
9296 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
9297 OptoRuntime::galoisCounterMode_aescrypt_Type(),
9298 stubAddr, stubName, TypePtr::BOTTOM,
9299 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
9300
9301 // return cipher length (int)
9302 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
9303 set_result(retvalue);
9304
9305 return true;
9306 }
9307
9308 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
9309 // Return node representing slow path of predicate check.
9310 // the pseudo code we want to emulate with this predicate is:
9311 // for encryption:
9312 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
9313 // for decryption:
9314 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
9315 // note cipher==plain is more conservative than the original java code but that's OK
9316 //
9317
9318 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
9319 // The receiver was checked for null already.
9320 Node* objGCTR = argument(7);
9321 // Load embeddedCipher field of GCTR object.
9322 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9323 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
9324
9325 // get AESCrypt klass for instanceOf check
9326 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
9327 // will have same classloader as CipherBlockChaining object
9328 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
9329 assert(tinst != nullptr, "GCTR obj is null");
9330 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9331
9332 // we want to do an instanceof comparison against the AESCrypt class
9333 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
9334 if (!klass_AESCrypt->is_loaded()) {
9335 // if AESCrypt is not even loaded, we never take the intrinsic fast path
9336 Node* ctrl = control();
9337 set_control(top()); // no regular fast path
9338 return ctrl;
9339 }
9340
9341 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9342 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
9343 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9344 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9345 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9346
9347 return instof_false; // even if it is null
9348 }
9349
9350 //------------------------------get_state_from_digest_object-----------------------
9351 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
9352 const char* state_type;
9353 switch (elem_type) {
9354 case T_BYTE: state_type = "[B"; break;
9355 case T_INT: state_type = "[I"; break;
9356 case T_LONG: state_type = "[J"; break;
9357 default: ShouldNotReachHere();
9358 }
9359 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
9360 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
9361 if (digest_state == nullptr) return (Node *) nullptr;
9362
9363 // now have the array, need to get the start address of the state array
9364 Node* state = array_element_address(digest_state, intcon(0), elem_type);
9365 return state;
9366 }
9367
9368 //------------------------------get_block_size_from_sha3_object----------------------------------
9369 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
9370 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
9371 assert (block_size != nullptr, "sanity");
9372 return block_size;
9373 }
9374
9375 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
9376 // Return node representing slow path of predicate check.
9377 // the pseudo code we want to emulate with this predicate is:
9378 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
9379 //
9380 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
9381 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9382 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9383 assert((uint)predicate < 5, "sanity");
9384
9385 // The receiver was checked for null already.
9386 Node* digestBaseObj = argument(0);
9387
9388 // get DigestBase klass for instanceOf check
9389 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
9390 assert(tinst != nullptr, "digestBaseObj is null");
9391 assert(tinst->is_loaded(), "DigestBase is not loaded");
9392
9393 const char* klass_name = nullptr;
9394 switch (predicate) {
9395 case 0:
9396 if (UseMD5Intrinsics) {
9397 // we want to do an instanceof comparison against the MD5 class
9398 klass_name = "sun/security/provider/MD5";
9399 }
9400 break;
9401 case 1:
9402 if (UseSHA1Intrinsics) {
9403 // we want to do an instanceof comparison against the SHA class
9404 klass_name = "sun/security/provider/SHA";
9405 }
9406 break;
9407 case 2:
9408 if (UseSHA256Intrinsics) {
9409 // we want to do an instanceof comparison against the SHA2 class
9410 klass_name = "sun/security/provider/SHA2";
9411 }
9412 break;
9413 case 3:
9414 if (UseSHA512Intrinsics) {
9415 // we want to do an instanceof comparison against the SHA5 class
9416 klass_name = "sun/security/provider/SHA5";
9417 }
9418 break;
9419 case 4:
9420 if (UseSHA3Intrinsics) {
9421 // we want to do an instanceof comparison against the SHA3 class
9422 klass_name = "sun/security/provider/SHA3";
9423 }
9424 break;
9425 default:
9426 fatal("unknown SHA intrinsic predicate: %d", predicate);
9427 }
9428
9429 ciKlass* klass = nullptr;
9430 if (klass_name != nullptr) {
9431 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
9432 }
9433 if ((klass == nullptr) || !klass->is_loaded()) {
9434 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
9435 Node* ctrl = control();
9436 set_control(top()); // no intrinsic path
9437 return ctrl;
9438 }
9439 ciInstanceKlass* instklass = klass->as_instance_klass();
9440
9441 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
9442 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9443 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9444 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9445
9446 return instof_false; // even if it is null
9447 }
9448
9449 //-------------inline_fma-----------------------------------
9450 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
9451 Node *a = nullptr;
9452 Node *b = nullptr;
9453 Node *c = nullptr;
9454 Node* result = nullptr;
9455 switch (id) {
9456 case vmIntrinsics::_fmaD:
9457 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
9458 // no receiver since it is static method
9459 a = argument(0);
9460 b = argument(2);
9461 c = argument(4);
9462 result = _gvn.transform(new FmaDNode(a, b, c));
9463 break;
9464 case vmIntrinsics::_fmaF:
9465 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
9466 a = argument(0);
9467 b = argument(1);
9468 c = argument(2);
9469 result = _gvn.transform(new FmaFNode(a, b, c));
9470 break;
9471 default:
9472 fatal_unexpected_iid(id); break;
9473 }
9474 set_result(result);
9475 return true;
9476 }
9477
9478 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
9479 // argument(0) is receiver
9480 Node* codePoint = argument(1);
9481 Node* n = nullptr;
9482
9483 switch (id) {
9484 case vmIntrinsics::_isDigit :
9485 n = new DigitNode(control(), codePoint);
9486 break;
9487 case vmIntrinsics::_isLowerCase :
9488 n = new LowerCaseNode(control(), codePoint);
9489 break;
9490 case vmIntrinsics::_isUpperCase :
9491 n = new UpperCaseNode(control(), codePoint);
9492 break;
9493 case vmIntrinsics::_isWhitespace :
9494 n = new WhitespaceNode(control(), codePoint);
9495 break;
9496 default:
9497 fatal_unexpected_iid(id);
9498 }
9499
9500 set_result(_gvn.transform(n));
9501 return true;
9502 }
9503
9504 bool LibraryCallKit::inline_profileBoolean() {
9505 Node* counts = argument(1);
9506 const TypeAryPtr* ary = nullptr;
9507 ciArray* aobj = nullptr;
9508 if (counts->is_Con()
9509 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
9510 && (aobj = ary->const_oop()->as_array()) != nullptr
9511 && (aobj->length() == 2)) {
9512 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
9513 jint false_cnt = aobj->element_value(0).as_int();
9514 jint true_cnt = aobj->element_value(1).as_int();
9515
9516 if (C->log() != nullptr) {
9517 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
9518 false_cnt, true_cnt);
9519 }
9520
9521 if (false_cnt + true_cnt == 0) {
9522 // According to profile, never executed.
9523 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9524 Deoptimization::Action_reinterpret);
9525 return true;
9526 }
9527
9528 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
9529 // is a number of each value occurrences.
9530 Node* result = argument(0);
9531 if (false_cnt == 0 || true_cnt == 0) {
9532 // According to profile, one value has been never seen.
9533 int expected_val = (false_cnt == 0) ? 1 : 0;
9534
9535 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
9536 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
9537
9538 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
9539 Node* fast_path = _gvn.transform(new IfTrueNode(check));
9540 Node* slow_path = _gvn.transform(new IfFalseNode(check));
9541
9542 { // Slow path: uncommon trap for never seen value and then reexecute
9543 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
9544 // the value has been seen at least once.
9545 PreserveJVMState pjvms(this);
9546 PreserveReexecuteState preexecs(this);
9547 jvms()->set_should_reexecute(true);
9548
9549 set_control(slow_path);
9550 set_i_o(i_o());
9551
9552 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9553 Deoptimization::Action_reinterpret);
9554 }
9555 // The guard for never seen value enables sharpening of the result and
9556 // returning a constant. It allows to eliminate branches on the same value
9557 // later on.
9558 set_control(fast_path);
9559 result = intcon(expected_val);
9560 }
9561 // Stop profiling.
9562 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
9563 // By replacing method body with profile data (represented as ProfileBooleanNode
9564 // on IR level) we effectively disable profiling.
9565 // It enables full speed execution once optimized code is generated.
9566 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
9567 C->record_for_igvn(profile);
9568 set_result(profile);
9569 return true;
9570 } else {
9571 // Continue profiling.
9572 // Profile data isn't available at the moment. So, execute method's bytecode version.
9573 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
9574 // is compiled and counters aren't available since corresponding MethodHandle
9575 // isn't a compile-time constant.
9576 return false;
9577 }
9578 }
9579
9580 bool LibraryCallKit::inline_isCompileConstant() {
9581 Node* n = argument(0);
9582 set_result(n->is_Con() ? intcon(1) : intcon(0));
9583 return true;
9584 }
9585
9586 //------------------------------- inline_getObjectSize --------------------------------------
9587 //
9588 // Calculate the runtime size of the object/array.
9589 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
9590 //
9591 bool LibraryCallKit::inline_getObjectSize() {
9592 Node* obj = argument(3);
9593 Node* klass_node = load_object_klass(obj);
9594
9595 jint layout_con = Klass::_lh_neutral_value;
9596 Node* layout_val = get_layout_helper(klass_node, layout_con);
9597 int layout_is_con = (layout_val == nullptr);
9598
9599 if (layout_is_con) {
9600 // Layout helper is constant, can figure out things at compile time.
9601
9602 if (Klass::layout_helper_is_instance(layout_con)) {
9603 // Instance case: layout_con contains the size itself.
9604 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
9605 set_result(size);
9606 } else {
9607 // Array case: size is round(header + element_size*arraylength).
9608 // Since arraylength is different for every array instance, we have to
9609 // compute the whole thing at runtime.
9610
9611 Node* arr_length = load_array_length(obj);
9612
9613 int round_mask = MinObjAlignmentInBytes - 1;
9614 int hsize = Klass::layout_helper_header_size(layout_con);
9615 int eshift = Klass::layout_helper_log2_element_size(layout_con);
9616
9617 if ((round_mask & ~right_n_bits(eshift)) == 0) {
9618 round_mask = 0; // strength-reduce it if it goes away completely
9619 }
9620 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
9621 Node* header_size = intcon(hsize + round_mask);
9622
9623 Node* lengthx = ConvI2X(arr_length);
9624 Node* headerx = ConvI2X(header_size);
9625
9626 Node* abody = lengthx;
9627 if (eshift != 0) {
9628 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
9629 }
9630 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
9631 if (round_mask != 0) {
9632 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
9633 }
9634 size = ConvX2L(size);
9635 set_result(size);
9636 }
9637 } else {
9638 // Layout helper is not constant, need to test for array-ness at runtime.
9639
9640 enum { _instance_path = 1, _array_path, PATH_LIMIT };
9641 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
9642 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
9643 record_for_igvn(result_reg);
9644
9645 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
9646 if (array_ctl != nullptr) {
9647 // Array case: size is round(header + element_size*arraylength).
9648 // Since arraylength is different for every array instance, we have to
9649 // compute the whole thing at runtime.
9650
9651 PreserveJVMState pjvms(this);
9652 set_control(array_ctl);
9653 Node* arr_length = load_array_length(obj);
9654
9655 int round_mask = MinObjAlignmentInBytes - 1;
9656 Node* mask = intcon(round_mask);
9657
9658 Node* hss = intcon(Klass::_lh_header_size_shift);
9659 Node* hsm = intcon(Klass::_lh_header_size_mask);
9660 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
9661 header_size = _gvn.transform(new AndINode(header_size, hsm));
9662 header_size = _gvn.transform(new AddINode(header_size, mask));
9663
9664 // There is no need to mask or shift this value.
9665 // The semantics of LShiftINode include an implicit mask to 0x1F.
9666 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
9667 Node* elem_shift = layout_val;
9668
9669 Node* lengthx = ConvI2X(arr_length);
9670 Node* headerx = ConvI2X(header_size);
9671
9672 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
9673 Node* size = _gvn.transform(new AddXNode(headerx, abody));
9674 if (round_mask != 0) {
9675 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
9676 }
9677 size = ConvX2L(size);
9678
9679 result_reg->init_req(_array_path, control());
9680 result_val->init_req(_array_path, size);
9681 }
9682
9683 if (!stopped()) {
9684 // Instance case: the layout helper gives us instance size almost directly,
9685 // but we need to mask out the _lh_instance_slow_path_bit.
9686 Node* size = ConvI2X(layout_val);
9687 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
9688 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
9689 size = _gvn.transform(new AndXNode(size, mask));
9690 size = ConvX2L(size);
9691
9692 result_reg->init_req(_instance_path, control());
9693 result_val->init_req(_instance_path, size);
9694 }
9695
9696 set_result(result_reg, result_val);
9697 }
9698
9699 return true;
9700 }
9701
9702 //------------------------------- inline_blackhole --------------------------------------
9703 //
9704 // Make sure all arguments to this node are alive.
9705 // This matches methods that were requested to be blackholed through compile commands.
9706 //
9707 bool LibraryCallKit::inline_blackhole() {
9708 assert(callee()->is_static(), "Should have been checked before: only static methods here");
9709 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
9710 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
9711
9712 // Blackhole node pinches only the control, not memory. This allows
9713 // the blackhole to be pinned in the loop that computes blackholed
9714 // values, but have no other side effects, like breaking the optimizations
9715 // across the blackhole.
9716
9717 Node* bh = _gvn.transform(new BlackholeNode(control()));
9718 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
9719
9720 // Bind call arguments as blackhole arguments to keep them alive
9721 uint nargs = callee()->arg_size();
9722 for (uint i = 0; i < nargs; i++) {
9723 bh->add_req(argument(i));
9724 }
9725
9726 return true;
9727 }
9728
9729 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9730 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9731 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9732 return nullptr; // box klass is not Float16
9733 }
9734
9735 // Null check; get notnull casted pointer
9736 Node* null_ctl = top();
9737 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9738 // If not_null_box is dead, only null-path is taken
9739 if (stopped()) {
9740 set_control(null_ctl);
9741 return nullptr;
9742 }
9743 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9744 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9745 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9746 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9747 }
9748
9749 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9750 PreserveReexecuteState preexecs(this);
9751 jvms()->set_should_reexecute(true);
9752
9753 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9754 Node* klass_node = makecon(klass_type);
9755 Node* box = new_instance(klass_node);
9756
9757 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9758 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9759
9760 Node* field_store = _gvn.transform(access_store_at(box,
9761 value_field,
9762 value_adr_type,
9763 value,
9764 TypeInt::SHORT,
9765 T_SHORT,
9766 IN_HEAP));
9767 set_memory(field_store, value_adr_type);
9768 return box;
9769 }
9770
9771 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9772 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9773 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9774 return false;
9775 }
9776
9777 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9778 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9779 return false;
9780 }
9781
9782 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9783 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9784 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9785 ciSymbols::short_signature(),
9786 false);
9787 assert(field != nullptr, "");
9788
9789 // Transformed nodes
9790 Node* fld1 = nullptr;
9791 Node* fld2 = nullptr;
9792 Node* fld3 = nullptr;
9793 switch(num_args) {
9794 case 3:
9795 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9796 if (fld3 == nullptr) {
9797 return false;
9798 }
9799 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9800 // fall-through
9801 case 2:
9802 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9803 if (fld2 == nullptr) {
9804 return false;
9805 }
9806 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9807 // fall-through
9808 case 1:
9809 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9810 if (fld1 == nullptr) {
9811 return false;
9812 }
9813 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9814 break;
9815 default: fatal("Unsupported number of arguments %d", num_args);
9816 }
9817
9818 Node* result = nullptr;
9819 switch (id) {
9820 // Unary operations
9821 case vmIntrinsics::_sqrt_float16:
9822 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9823 break;
9824 // Ternary operations
9825 case vmIntrinsics::_fma_float16:
9826 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9827 break;
9828 default:
9829 fatal_unexpected_iid(id);
9830 break;
9831 }
9832 result = _gvn.transform(new ReinterpretHF2SNode(result));
9833 set_result(box_fp16_value(float16_box_type, field, result));
9834 return true;
9835 }
9836