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_type->is_inlinetypeptr()) {
2448 value = InlineTypeNode::make_from_oop(this, value, value_type->inline_klass());
2449 }
2450 set_result(value);
2451 return true;
2452 }
2453 }
2454 }
2455 {
2456 // Re-execute the unsafe access if allocation triggers deoptimization.
2457 PreserveReexecuteState preexecs(this);
2458 jvms()->set_should_reexecute(true);
2459 vt = vt->buffer(this);
2460 }
2461 base = vt->get_oop();
2462 }
2463
2464 // 32-bit machines ignore the high half!
2465 offset = ConvL2X(offset);
2466
2467 // Save state and restore on bailout
2468 SavedState old_state(this);
2469
2470 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2471 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2472
2473 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2474 if (type != T_OBJECT) {
2475 decorators |= IN_NATIVE; // off-heap primitive access
2476 } else {
2477 return false; // off-heap oop accesses are not supported
2478 }
2479 } else {
2480 heap_base_oop = base; // on-heap or mixed access
2481 }
2482
2483 // Can base be null? Otherwise, always on-heap access.
2484 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2485
2486 if (!can_access_non_heap) {
2487 decorators |= IN_HEAP;
2488 }
2489
2490 Node* val = is_store ? argument(4) : nullptr;
2491
2492 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2493 if (adr_type == TypePtr::NULL_PTR) {
2494 return false; // off-heap access with zero address
2495 }
2496
2497 // Try to categorize the address.
2498 Compile::AliasType* alias_type = C->alias_type(adr_type);
2499 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2500
2501 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2502 alias_type->adr_type() == TypeAryPtr::RANGE) {
2503 return false; // not supported
2504 }
2505
2506 bool mismatched = false;
2507 BasicType bt = T_ILLEGAL;
2508 ciField* field = nullptr;
2509 if (adr_type->isa_instptr()) {
2510 const TypeInstPtr* instptr = adr_type->is_instptr();
2511 ciInstanceKlass* k = instptr->instance_klass();
2512 int off = instptr->offset();
2513 if (instptr->const_oop() != nullptr &&
2514 k == ciEnv::current()->Class_klass() &&
2515 instptr->offset() >= (k->size_helper() * wordSize)) {
2516 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2517 field = k->get_field_by_offset(off, true);
2518 } else {
2519 field = k->get_non_flat_field_by_offset(off);
2520 }
2521 if (field != nullptr) {
2522 bt = type2field[field->type()->basic_type()];
2523 }
2524 if (bt != alias_type->basic_type()) {
2525 // Type mismatch. Is it an access to a nested flat field?
2526 field = k->get_field_by_offset(off, false);
2527 if (field != nullptr) {
2528 bt = type2field[field->type()->basic_type()];
2529 }
2530 }
2531 assert(bt == alias_type->basic_type(), "should match");
2532 } else {
2533 bt = alias_type->basic_type();
2534 }
2535
2536 if (bt != T_ILLEGAL) {
2537 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2538 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2539 // Alias type doesn't differentiate between byte[] and boolean[]).
2540 // Use address type to get the element type.
2541 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2542 }
2543 if (is_reference_type(bt, true)) {
2544 // accessing an array field with getReference is not a mismatch
2545 bt = T_OBJECT;
2546 }
2547 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2548 // Don't intrinsify mismatched object accesses
2549 return false;
2550 }
2551 mismatched = (bt != type);
2552 } else if (alias_type->adr_type()->isa_oopptr()) {
2553 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2554 }
2555
2556 old_state.discard();
2557 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2558
2559 if (mismatched) {
2560 decorators |= C2_MISMATCHED;
2561 }
2562
2563 // First guess at the value type.
2564 const Type *value_type = Type::get_const_basic_type(type);
2565
2566 // Figure out the memory ordering.
2567 decorators |= mo_decorator_for_access_kind(kind);
2568
2569 if (!is_store) {
2570 if (type == T_OBJECT) {
2571 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2572 if (tjp != nullptr) {
2573 value_type = tjp;
2574 }
2575 }
2576 }
2577
2578 receiver = null_check(receiver);
2579 if (stopped()) {
2580 return true;
2581 }
2582 // Heap pointers get a null-check from the interpreter,
2583 // as a courtesy. However, this is not guaranteed by Unsafe,
2584 // and it is not possible to fully distinguish unintended nulls
2585 // from intended ones in this API.
2586
2587 if (!is_store) {
2588 Node* p = nullptr;
2589 // Try to constant fold a load from a constant field
2590
2591 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !field->is_flat() && !mismatched) {
2592 // final or stable field
2593 p = make_constant_from_field(field, heap_base_oop);
2594 }
2595
2596 if (p == nullptr) { // Could not constant fold the load
2597 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2598 const TypeOopPtr* ptr = value_type->make_oopptr();
2599 if (ptr != nullptr && ptr->is_inlinetypeptr()) {
2600 // Load a non-flattened inline type from memory
2601 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass());
2602 }
2603 // Normalize the value returned by getBoolean in the following cases
2604 if (type == T_BOOLEAN &&
2605 (mismatched ||
2606 heap_base_oop == top() || // - heap_base_oop is null or
2607 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2608 // and the unsafe access is made to large offset
2609 // (i.e., larger than the maximum offset necessary for any
2610 // field access)
2611 ) {
2612 IdealKit ideal = IdealKit(this);
2613 #define __ ideal.
2614 IdealVariable normalized_result(ideal);
2615 __ declarations_done();
2616 __ set(normalized_result, p);
2617 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2618 __ set(normalized_result, ideal.ConI(1));
2619 ideal.end_if();
2620 final_sync(ideal);
2621 p = __ value(normalized_result);
2622 #undef __
2623 }
2624 }
2625 if (type == T_ADDRESS) {
2626 p = gvn().transform(new CastP2XNode(nullptr, p));
2627 p = ConvX2UL(p);
2628 }
2629 // The load node has the control of the preceding MemBarCPUOrder. All
2630 // following nodes will have the control of the MemBarCPUOrder inserted at
2631 // the end of this method. So, pushing the load onto the stack at a later
2632 // point is fine.
2633 set_result(p);
2634 } else {
2635 if (bt == T_ADDRESS) {
2636 // Repackage the long as a pointer.
2637 val = ConvL2X(val);
2638 val = gvn().transform(new CastX2PNode(val));
2639 }
2640 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2641 }
2642
2643 return true;
2644 }
2645
2646 bool LibraryCallKit::inline_unsafe_flat_access(bool is_store, AccessKind kind) {
2647 #ifdef ASSERT
2648 {
2649 ResourceMark rm;
2650 // Check the signatures.
2651 ciSignature* sig = callee()->signature();
2652 assert(sig->type_at(0)->basic_type() == T_OBJECT, "base should be object, but is %s", type2name(sig->type_at(0)->basic_type()));
2653 assert(sig->type_at(1)->basic_type() == T_LONG, "offset should be long, but is %s", type2name(sig->type_at(1)->basic_type()));
2654 assert(sig->type_at(2)->basic_type() == T_INT, "layout kind should be int, but is %s", type2name(sig->type_at(3)->basic_type()));
2655 assert(sig->type_at(3)->basic_type() == T_OBJECT, "value klass should be object, but is %s", type2name(sig->type_at(4)->basic_type()));
2656 if (is_store) {
2657 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value, but returns %s", type2name(sig->return_type()->basic_type()));
2658 assert(sig->count() == 5, "flat putter should have 5 arguments, but has %d", sig->count());
2659 assert(sig->type_at(4)->basic_type() == T_OBJECT, "put value should be object, but is %s", type2name(sig->type_at(5)->basic_type()));
2660 } else {
2661 assert(sig->return_type()->basic_type() == T_OBJECT, "getter must return an object, but returns %s", type2name(sig->return_type()->basic_type()));
2662 assert(sig->count() == 4, "flat getter should have 4 arguments, but has %d", sig->count());
2663 }
2664 }
2665 #endif // ASSERT
2666
2667 assert(kind == Relaxed, "Only plain accesses for now");
2668 if (callee()->is_static()) {
2669 // caller must have the capability!
2670 return false;
2671 }
2672 C->set_has_unsafe_access(true);
2673
2674 const TypeInstPtr* value_klass_node = _gvn.type(argument(5))->isa_instptr();
2675 if (value_klass_node == nullptr || value_klass_node->const_oop() == nullptr) {
2676 // parameter valueType is not a constant
2677 return false;
2678 }
2679 ciType* mirror_type = value_klass_node->const_oop()->as_instance()->java_mirror_type();
2680 if (!mirror_type->is_inlinetype()) {
2681 // Dead code
2682 return false;
2683 }
2684 ciInlineKlass* value_klass = mirror_type->as_inline_klass();
2685
2686 const TypeInt* layout_type = _gvn.type(argument(4))->isa_int();
2687 if (layout_type == nullptr || !layout_type->is_con()) {
2688 // parameter layoutKind is not a constant
2689 return false;
2690 }
2691 assert(layout_type->get_con() >= static_cast<int>(LayoutKind::REFERENCE) &&
2692 layout_type->get_con() < static_cast<int>(LayoutKind::UNKNOWN),
2693 "invalid layoutKind %d", layout_type->get_con());
2694 LayoutKind layout = static_cast<LayoutKind>(layout_type->get_con());
2695 assert(layout == LayoutKind::REFERENCE || layout == LayoutKind::NULL_FREE_NON_ATOMIC_FLAT ||
2696 layout == LayoutKind::NULL_FREE_ATOMIC_FLAT || layout == LayoutKind::NULLABLE_ATOMIC_FLAT,
2697 "unexpected layoutKind %d", layout_type->get_con());
2698
2699 null_check(argument(0));
2700 if (stopped()) {
2701 return true;
2702 }
2703
2704 Node* base = must_be_not_null(argument(1), true);
2705 Node* offset = argument(2);
2706 const Type* base_type = _gvn.type(base);
2707
2708 Node* ptr;
2709 bool immutable_memory = false;
2710 DecoratorSet decorators = C2_UNSAFE_ACCESS | IN_HEAP | MO_UNORDERED;
2711 if (base_type->isa_instptr()) {
2712 const TypeLong* offset_type = _gvn.type(offset)->isa_long();
2713 if (offset_type == nullptr || !offset_type->is_con()) {
2714 // Offset into a non-array should be a constant
2715 decorators |= C2_MISMATCHED;
2716 } else {
2717 int offset_con = checked_cast<int>(offset_type->get_con());
2718 ciInstanceKlass* base_klass = base_type->is_instptr()->instance_klass();
2719 ciField* field = base_klass->get_non_flat_field_by_offset(offset_con);
2720 if (field == nullptr) {
2721 assert(!base_klass->is_final(), "non-existence field at offset %d of class %s", offset_con, base_klass->name()->as_utf8());
2722 decorators |= C2_MISMATCHED;
2723 } else {
2724 assert(field->type() == value_klass, "field at offset %d of %s is of type %s, but valueType is %s",
2725 offset_con, base_klass->name()->as_utf8(), field->type()->name(), value_klass->name()->as_utf8());
2726 immutable_memory = field->is_strict() && field->is_final();
2727
2728 if (base->is_InlineType()) {
2729 assert(!is_store, "Cannot store into a non-larval value object");
2730 set_result(base->as_InlineType()->field_value_by_offset(offset_con, false));
2731 return true;
2732 }
2733 }
2734 }
2735
2736 if (base->is_InlineType()) {
2737 assert(!is_store, "Cannot store into a non-larval value object");
2738 base = base->as_InlineType()->buffer(this, true);
2739 }
2740 ptr = basic_plus_adr(base, ConvL2X(offset));
2741 } else if (base_type->isa_aryptr()) {
2742 decorators |= IS_ARRAY;
2743 if (layout == LayoutKind::REFERENCE) {
2744 if (!base_type->is_aryptr()->is_not_flat()) {
2745 const TypeAryPtr* array_type = base_type->is_aryptr()->cast_to_not_flat();
2746 // TODO 8350865 This should be a CheckCastPP, can we add a test?
2747 Node* new_base = _gvn.transform(new CastPPNode(control(), base, array_type, ConstraintCastNode::DependencyType::NonFloatingNarrowing));
2748 replace_in_map(base, new_base);
2749 base = new_base;
2750 }
2751 ptr = basic_plus_adr(base, ConvL2X(offset));
2752 } else {
2753 if (UseArrayFlattening) {
2754 // Flat array must have an exact type
2755 bool is_null_free = !LayoutKindHelper::is_nullable_flat(layout);
2756 bool is_atomic = LayoutKindHelper::is_atomic_flat(layout);
2757 Node* new_base = cast_to_flat_array_exact(base, value_klass, is_null_free, is_atomic);
2758 replace_in_map(base, new_base);
2759 base = new_base;
2760 ptr = basic_plus_adr(base, ConvL2X(offset));
2761 const TypeAryPtr* ptr_type = _gvn.type(ptr)->is_aryptr();
2762 if (ptr_type->field_offset().get() != 0) {
2763 // TODO 8350865 This should be a CheckCastPP, can we add a test?
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 CheckCastPPNode(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 Node* map = access_load_at(mirror, map_addr, TypeAryPtr::INTS, val_type, T_ARRAY, IN_HEAP | MO_UNORDERED);
3187
3188 set_result(map);
3189 return true;
3190 }
3191
3192 bool LibraryCallKit::inline_onspinwait() {
3193 insert_mem_bar(Op_OnSpinWait);
3194 return true;
3195 }
3196
3197 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3198 if (!kls->is_Con()) {
3199 return true;
3200 }
3201 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
3202 if (klsptr == nullptr) {
3203 return true;
3204 }
3205 ciInstanceKlass* ik = klsptr->instance_klass();
3206 // don't need a guard for a klass that is already initialized
3207 return !ik->is_initialized();
3208 }
3209
3210 //----------------------------inline_unsafe_writeback0-------------------------
3211 // public native void Unsafe.writeback0(long address)
3212 bool LibraryCallKit::inline_unsafe_writeback0() {
3213 if (!Matcher::has_match_rule(Op_CacheWB)) {
3214 return false;
3215 }
3216 #ifndef PRODUCT
3217 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
3218 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
3219 ciSignature* sig = callee()->signature();
3220 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
3221 #endif
3222 null_check_receiver(); // null-check, then ignore
3223 Node *addr = argument(1);
3224 addr = new CastX2PNode(addr);
3225 addr = _gvn.transform(addr);
3226 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
3227 flush = _gvn.transform(flush);
3228 set_memory(flush, TypeRawPtr::BOTTOM);
3229 return true;
3230 }
3231
3232 //----------------------------inline_unsafe_writeback0-------------------------
3233 // public native void Unsafe.writeback0(long address)
3234 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
3235 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
3236 return false;
3237 }
3238 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
3239 return false;
3240 }
3241 #ifndef PRODUCT
3242 assert(Matcher::has_match_rule(Op_CacheWB),
3243 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
3244 : "found match rule for CacheWBPostSync but not CacheWB"));
3245
3246 #endif
3247 null_check_receiver(); // null-check, then ignore
3248 Node *sync;
3249 if (is_pre) {
3250 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3251 } else {
3252 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3253 }
3254 sync = _gvn.transform(sync);
3255 set_memory(sync, TypeRawPtr::BOTTOM);
3256 return true;
3257 }
3258
3259 //----------------------------inline_unsafe_allocate---------------------------
3260 // public native Object Unsafe.allocateInstance(Class<?> cls);
3261 bool LibraryCallKit::inline_unsafe_allocate() {
3262
3263 #if INCLUDE_JVMTI
3264 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3265 return false;
3266 }
3267 #endif //INCLUDE_JVMTI
3268
3269 if (callee()->is_static()) return false; // caller must have the capability!
3270
3271 null_check_receiver(); // null-check, then ignore
3272 Node* cls = null_check(argument(1));
3273 if (stopped()) return true;
3274
3275 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
3276 kls = null_check(kls);
3277 if (stopped()) return true; // argument was like int.class
3278
3279 #if INCLUDE_JVMTI
3280 // Don't try to access new allocated obj in the intrinsic.
3281 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
3282 // Deoptimize and allocate in interpreter instead.
3283 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
3284 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
3285 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
3286 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
3287 {
3288 BuildCutout unless(this, tst, PROB_MAX);
3289 uncommon_trap(Deoptimization::Reason_intrinsic,
3290 Deoptimization::Action_make_not_entrant);
3291 }
3292 if (stopped()) {
3293 return true;
3294 }
3295 #endif //INCLUDE_JVMTI
3296
3297 Node* test = nullptr;
3298 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3299 // Note: The argument might still be an illegal value like
3300 // Serializable.class or Object[].class. The runtime will handle it.
3301 // But we must make an explicit check for initialization.
3302 Node* insp = off_heap_plus_addr(kls, in_bytes(InstanceKlass::init_state_offset()));
3303 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3304 // can generate code to load it as unsigned byte.
3305 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
3306 Node* bits = intcon(InstanceKlass::fully_initialized);
3307 test = _gvn.transform(new SubINode(inst, bits));
3308 // The 'test' is non-zero if we need to take a slow path.
3309 }
3310 Node* obj = new_instance(kls, test);
3311 set_result(obj);
3312 return true;
3313 }
3314
3315 //------------------------inline_native_time_funcs--------------
3316 // inline code for System.currentTimeMillis() and System.nanoTime()
3317 // these have the same type and signature
3318 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3319 const TypeFunc* tf = OptoRuntime::void_long_Type();
3320 const TypePtr* no_memory_effects = nullptr;
3321 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3322 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3323 #ifdef ASSERT
3324 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3325 assert(value_top == top(), "second value must be top");
3326 #endif
3327 set_result(value);
3328 return true;
3329 }
3330
3331 //--------------------inline_native_vthread_start_transition--------------------
3332 // inline void startTransition(boolean is_mount);
3333 // inline void startFinalTransition();
3334 // Pseudocode of implementation:
3335 //
3336 // java_lang_Thread::set_is_in_vthread_transition(vthread, true);
3337 // carrier->set_is_in_vthread_transition(true);
3338 // OrderAccess::storeload();
3339 // int disable_requests = java_lang_Thread::vthread_transition_disable_count(vthread)
3340 // + global_vthread_transition_disable_count();
3341 // if (disable_requests > 0) {
3342 // slow path: runtime call
3343 // }
3344 bool LibraryCallKit::inline_native_vthread_start_transition(address funcAddr, const char* funcName, bool is_final_transition) {
3345 Node* vt_oop = must_be_not_null(argument(0), true); // VirtualThread this argument
3346 IdealKit ideal(this);
3347
3348 Node* thread = ideal.thread();
3349 Node* jt_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3350 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3351 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3352 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3353 insert_mem_bar(Op_MemBarStoreLoad);
3354 ideal.sync_kit(this);
3355
3356 Node* global_disable_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::global_vthread_transition_disable_count_address()));
3357 Node* global_disable = ideal.load(ideal.ctrl(), global_disable_addr, TypeInt::INT, T_INT, Compile::AliasIdxRaw, true /*require_atomic_access*/);
3358 Node* vt_disable_addr = basic_plus_adr(vt_oop, java_lang_Thread::vthread_transition_disable_count_offset());
3359 const TypePtr* vt_disable_addr_t = _gvn.type(vt_disable_addr)->is_ptr();
3360 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*/);
3361 Node* disabled = _gvn.transform(new AddINode(global_disable, vt_disable));
3362
3363 ideal.if_then(disabled, BoolTest::ne, ideal.ConI(0)); {
3364 sync_kit(ideal);
3365 Node* is_mount = is_final_transition ? ideal.ConI(0) : argument(1);
3366 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3367 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3368 ideal.sync_kit(this);
3369 }
3370 ideal.end_if();
3371
3372 final_sync(ideal);
3373 return true;
3374 }
3375
3376 bool LibraryCallKit::inline_native_vthread_end_transition(address funcAddr, const char* funcName, bool is_first_transition) {
3377 Node* vt_oop = must_be_not_null(argument(0), true); // VirtualThread this argument
3378 IdealKit ideal(this);
3379
3380 Node* _notify_jvmti_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::notify_jvmti_events_address()));
3381 Node* _notify_jvmti = ideal.load(ideal.ctrl(), _notify_jvmti_addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3382
3383 ideal.if_then(_notify_jvmti, BoolTest::eq, ideal.ConI(1)); {
3384 sync_kit(ideal);
3385 Node* is_mount = is_first_transition ? ideal.ConI(1) : argument(1);
3386 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3387 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3388 ideal.sync_kit(this);
3389 } ideal.else_(); {
3390 Node* thread = ideal.thread();
3391 Node* jt_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3392 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3393
3394 sync_kit(ideal);
3395 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3396 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3397 ideal.sync_kit(this);
3398 } ideal.end_if();
3399
3400 final_sync(ideal);
3401 return true;
3402 }
3403
3404 #if INCLUDE_JVMTI
3405
3406 // Always update the is_disable_suspend bit.
3407 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3408 if (!DoJVMTIVirtualThreadTransitions) {
3409 return true;
3410 }
3411 IdealKit ideal(this);
3412
3413 {
3414 // unconditionally update the is_disable_suspend bit in current JavaThread
3415 Node* thread = ideal.thread();
3416 Node* arg = argument(0); // argument for notification
3417 Node* addr = off_heap_plus_addr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3418 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3419
3420 sync_kit(ideal);
3421 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3422 ideal.sync_kit(this);
3423 }
3424 final_sync(ideal);
3425
3426 return true;
3427 }
3428
3429 #endif // INCLUDE_JVMTI
3430
3431 #ifdef JFR_HAVE_INTRINSICS
3432
3433 /**
3434 * if oop->klass != null
3435 * // normal class
3436 * epoch = _epoch_state ? 2 : 1
3437 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3438 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3439 * }
3440 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3441 * else
3442 * // primitive class
3443 * if oop->array_klass != null
3444 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3445 * else
3446 * id = LAST_TYPE_ID + 1 // void class path
3447 * if (!signaled)
3448 * signaled = true
3449 */
3450 bool LibraryCallKit::inline_native_classID() {
3451 Node* cls = argument(0);
3452
3453 IdealKit ideal(this);
3454 #define __ ideal.
3455 IdealVariable result(ideal); __ declarations_done();
3456 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3457 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3458 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3459
3460
3461 __ if_then(kls, BoolTest::ne, null()); {
3462 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3463 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3464
3465 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3466 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3467 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3468 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3469 mask = _gvn.transform(new OrLNode(mask, epoch));
3470 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3471
3472 float unlikely = PROB_UNLIKELY(0.999);
3473 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3474 sync_kit(ideal);
3475 make_runtime_call(RC_LEAF,
3476 OptoRuntime::class_id_load_barrier_Type(),
3477 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3478 "class id load barrier",
3479 TypePtr::BOTTOM,
3480 kls);
3481 ideal.sync_kit(this);
3482 } __ end_if();
3483
3484 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3485 } __ else_(); {
3486 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3487 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3488 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3489 __ if_then(array_kls, BoolTest::ne, null()); {
3490 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3491 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3492 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3493 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3494 } __ else_(); {
3495 // void class case
3496 ideal.set(result, longcon(LAST_TYPE_ID + 1));
3497 } __ end_if();
3498
3499 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3500 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3501 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3502 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3503 } __ end_if();
3504 } __ end_if();
3505
3506 final_sync(ideal);
3507 set_result(ideal.value(result));
3508 #undef __
3509 return true;
3510 }
3511
3512 //------------------------inline_native_jvm_commit------------------
3513 bool LibraryCallKit::inline_native_jvm_commit() {
3514 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3515
3516 // Save input memory and i_o state.
3517 Node* input_memory_state = reset_memory();
3518 set_all_memory(input_memory_state);
3519 Node* input_io_state = i_o();
3520
3521 // TLS.
3522 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3523 // Jfr java buffer.
3524 Node* java_buffer_offset = _gvn.transform(AddPNode::make_off_heap(tls_ptr, MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR))));
3525 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3526 Node* java_buffer_pos_offset = _gvn.transform(AddPNode::make_off_heap(java_buffer, MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET))));
3527
3528 // Load the current value of the notified field in the JfrThreadLocal.
3529 Node* notified_offset = off_heap_plus_addr(tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3530 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3531
3532 // Test for notification.
3533 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3534 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3535 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3536
3537 // True branch, is notified.
3538 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3539 set_control(is_notified);
3540
3541 // Reset notified state.
3542 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3543 Node* notified_reset_memory = reset_memory();
3544
3545 // 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.
3546 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3547 // Convert the machine-word to a long.
3548 Node* current_pos = ConvX2L(current_pos_X);
3549
3550 // False branch, not notified.
3551 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3552 set_control(not_notified);
3553 set_all_memory(input_memory_state);
3554
3555 // Arg is the next position as a long.
3556 Node* arg = argument(0);
3557 // Convert long to machine-word.
3558 Node* next_pos_X = ConvL2X(arg);
3559
3560 // Store the next_position to the underlying jfr java buffer.
3561 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3562
3563 Node* commit_memory = reset_memory();
3564 set_all_memory(commit_memory);
3565
3566 // 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.
3567 Node* java_buffer_flags_offset = _gvn.transform(AddPNode::make_off_heap(java_buffer, MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET))));
3568 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3569 Node* lease_constant = _gvn.intcon(4);
3570
3571 // And flags with lease constant.
3572 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3573
3574 // Branch on lease to conditionalize returning the leased java buffer.
3575 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3576 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3577 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3578
3579 // False branch, not a lease.
3580 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3581
3582 // True branch, is lease.
3583 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3584 set_control(is_lease);
3585
3586 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3587 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3588 OptoRuntime::void_void_Type(),
3589 SharedRuntime::jfr_return_lease(),
3590 "return_lease", TypePtr::BOTTOM);
3591 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3592
3593 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3594 record_for_igvn(lease_compare_rgn);
3595 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3596 record_for_igvn(lease_compare_mem);
3597 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3598 record_for_igvn(lease_compare_io);
3599 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3600 record_for_igvn(lease_result_value);
3601
3602 // Update control and phi nodes.
3603 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3604 lease_compare_rgn->init_req(_false_path, not_lease);
3605
3606 lease_compare_mem->init_req(_true_path, reset_memory());
3607 lease_compare_mem->init_req(_false_path, commit_memory);
3608
3609 lease_compare_io->init_req(_true_path, i_o());
3610 lease_compare_io->init_req(_false_path, input_io_state);
3611
3612 lease_result_value->init_req(_true_path, _gvn.longcon(0)); // if the lease was returned, return 0L.
3613 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3614
3615 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3616 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3617 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3618 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3619
3620 // Update control and phi nodes.
3621 result_rgn->init_req(_true_path, is_notified);
3622 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3623
3624 result_mem->init_req(_true_path, notified_reset_memory);
3625 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3626
3627 result_io->init_req(_true_path, input_io_state);
3628 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3629
3630 result_value->init_req(_true_path, current_pos);
3631 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3632
3633 // Set output state.
3634 set_control(_gvn.transform(result_rgn));
3635 set_all_memory(_gvn.transform(result_mem));
3636 set_i_o(_gvn.transform(result_io));
3637 set_result(result_rgn, result_value);
3638 return true;
3639 }
3640
3641 /*
3642 * The intrinsic is a model of this pseudo-code:
3643 *
3644 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3645 * jobject h_event_writer = tl->java_event_writer();
3646 * if (h_event_writer == nullptr) {
3647 * return nullptr;
3648 * }
3649 * oop threadObj = Thread::threadObj();
3650 * oop vthread = java_lang_Thread::vthread(threadObj);
3651 * traceid tid;
3652 * bool pinVirtualThread;
3653 * bool excluded;
3654 * if (vthread != threadObj) { // i.e. current thread is virtual
3655 * tid = java_lang_Thread::tid(vthread);
3656 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3657 * pinVirtualThread = VMContinuations;
3658 * excluded = vthread_epoch_raw & excluded_mask;
3659 * if (!excluded) {
3660 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3661 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3662 * if (vthread_epoch != current_epoch) {
3663 * write_checkpoint();
3664 * }
3665 * }
3666 * } else {
3667 * tid = java_lang_Thread::tid(threadObj);
3668 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3669 * pinVirtualThread = false;
3670 * excluded = thread_epoch_raw & excluded_mask;
3671 * }
3672 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3673 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3674 * if (tid_in_event_writer != tid) {
3675 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3676 * setField(event_writer, "excluded", excluded);
3677 * setField(event_writer, "threadID", tid);
3678 * }
3679 * return event_writer
3680 */
3681 bool LibraryCallKit::inline_native_getEventWriter() {
3682 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3683
3684 // Save input memory and i_o state.
3685 Node* input_memory_state = reset_memory();
3686 set_all_memory(input_memory_state);
3687 Node* input_io_state = i_o();
3688
3689 // The most significant bit of the u2 is used to denote thread exclusion
3690 Node* excluded_shift = _gvn.intcon(15);
3691 Node* excluded_mask = _gvn.intcon(1 << 15);
3692 // The epoch generation is the range [1-32767]
3693 Node* epoch_mask = _gvn.intcon(32767);
3694
3695 // TLS
3696 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3697
3698 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3699 Node* jobj_ptr = off_heap_plus_addr(tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3700
3701 // Load the eventwriter jobject handle.
3702 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3703
3704 // Null check the jobject handle.
3705 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3706 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3707 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3708
3709 // False path, jobj is null.
3710 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3711
3712 // True path, jobj is not null.
3713 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3714
3715 set_control(jobj_is_not_null);
3716
3717 // Load the threadObj for the CarrierThread.
3718 Node* threadObj = generate_current_thread(tls_ptr);
3719
3720 // Load the vthread.
3721 Node* vthread = generate_virtual_thread(tls_ptr);
3722
3723 // If vthread != threadObj, this is a virtual thread.
3724 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3725 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3726 IfNode* iff_vthread_not_equal_threadObj =
3727 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3728
3729 // False branch, fallback to threadObj.
3730 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3731 set_control(vthread_equal_threadObj);
3732
3733 // Load the tid field from the vthread object.
3734 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3735
3736 // Load the raw epoch value from the threadObj.
3737 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3738 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3739 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3740 TypeInt::CHAR, T_CHAR,
3741 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3742
3743 // Mask off the excluded information from the epoch.
3744 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3745
3746 // True branch, this is a virtual thread.
3747 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3748 set_control(vthread_not_equal_threadObj);
3749
3750 // Load the tid field from the vthread object.
3751 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3752
3753 // Continuation support determines if a virtual thread should be pinned.
3754 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3755 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3756
3757 // Load the raw epoch value from the vthread.
3758 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3759 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3760 TypeInt::CHAR, T_CHAR,
3761 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3762
3763 // Mask off the excluded information from the epoch.
3764 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, excluded_mask));
3765
3766 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3767 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, excluded_mask));
3768 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3769 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3770
3771 // False branch, vthread is excluded, no need to write epoch info.
3772 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3773
3774 // True branch, vthread is included, update epoch info.
3775 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3776 set_control(included);
3777
3778 // Get epoch value.
3779 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, epoch_mask));
3780
3781 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3782 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3783 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3784
3785 // Compare the epoch in the vthread to the current epoch generation.
3786 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3787 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3788 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3789
3790 // False path, epoch is equal, checkpoint information is valid.
3791 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3792
3793 // True path, epoch is not equal, write a checkpoint for the vthread.
3794 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3795
3796 set_control(epoch_is_not_equal);
3797
3798 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3799 // The call also updates the native thread local thread id and the vthread with the current epoch.
3800 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3801 OptoRuntime::jfr_write_checkpoint_Type(),
3802 SharedRuntime::jfr_write_checkpoint(),
3803 "write_checkpoint", TypePtr::BOTTOM);
3804 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3805
3806 // vthread epoch != current epoch
3807 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3808 record_for_igvn(epoch_compare_rgn);
3809 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3810 record_for_igvn(epoch_compare_mem);
3811 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3812 record_for_igvn(epoch_compare_io);
3813
3814 // Update control and phi nodes.
3815 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3816 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3817 epoch_compare_mem->init_req(_true_path, reset_memory());
3818 epoch_compare_mem->init_req(_false_path, input_memory_state);
3819 epoch_compare_io->init_req(_true_path, i_o());
3820 epoch_compare_io->init_req(_false_path, input_io_state);
3821
3822 // excluded != true
3823 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3824 record_for_igvn(exclude_compare_rgn);
3825 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3826 record_for_igvn(exclude_compare_mem);
3827 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3828 record_for_igvn(exclude_compare_io);
3829
3830 // Update control and phi nodes.
3831 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3832 exclude_compare_rgn->init_req(_false_path, excluded);
3833 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3834 exclude_compare_mem->init_req(_false_path, input_memory_state);
3835 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3836 exclude_compare_io->init_req(_false_path, input_io_state);
3837
3838 // vthread != threadObj
3839 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3840 record_for_igvn(vthread_compare_rgn);
3841 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3842 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3843 record_for_igvn(vthread_compare_io);
3844 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3845 record_for_igvn(tid);
3846 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3847 record_for_igvn(exclusion);
3848 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3849 record_for_igvn(pinVirtualThread);
3850
3851 // Update control and phi nodes.
3852 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3853 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3854 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3855 vthread_compare_mem->init_req(_false_path, input_memory_state);
3856 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3857 vthread_compare_io->init_req(_false_path, input_io_state);
3858 tid->init_req(_true_path, vthread_tid);
3859 tid->init_req(_false_path, thread_obj_tid);
3860 exclusion->init_req(_true_path, vthread_is_excluded);
3861 exclusion->init_req(_false_path, threadObj_is_excluded);
3862 pinVirtualThread->init_req(_true_path, continuation_support);
3863 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3864
3865 // Update branch state.
3866 set_control(_gvn.transform(vthread_compare_rgn));
3867 set_all_memory(_gvn.transform(vthread_compare_mem));
3868 set_i_o(_gvn.transform(vthread_compare_io));
3869
3870 // Load the event writer oop by dereferencing the jobject handle.
3871 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3872 assert(klass_EventWriter->is_loaded(), "invariant");
3873 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3874 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3875 const TypeOopPtr* const xtype = aklass->as_instance_type();
3876 Node* jobj_untagged = _gvn.transform(AddPNode::make_off_heap(jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3877 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3878
3879 // Load the current thread id from the event writer object.
3880 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3881 // Get the field offset to, conditionally, store an updated tid value later.
3882 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3883 // Get the field offset to, conditionally, store an updated exclusion value later.
3884 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3885 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3886 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3887
3888 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3889 record_for_igvn(event_writer_tid_compare_rgn);
3890 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3891 record_for_igvn(event_writer_tid_compare_mem);
3892 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3893 record_for_igvn(event_writer_tid_compare_io);
3894
3895 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3896 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3897 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3898 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3899
3900 // False path, tids are the same.
3901 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3902
3903 // True path, tid is not equal, need to update the tid in the event writer.
3904 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3905 record_for_igvn(tid_is_not_equal);
3906
3907 // Store the pin state to the event writer.
3908 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3909
3910 // Store the exclusion state to the event writer.
3911 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3912 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3913
3914 // Store the tid to the event writer.
3915 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3916
3917 // Update control and phi nodes.
3918 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3919 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3920 event_writer_tid_compare_mem->init_req(_true_path, reset_memory());
3921 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3922 event_writer_tid_compare_io->init_req(_true_path, i_o());
3923 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3924
3925 // Result of top level CFG, Memory, IO and Value.
3926 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3927 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3928 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3929 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3930
3931 // Result control.
3932 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3933 result_rgn->init_req(_false_path, jobj_is_null);
3934
3935 // Result memory.
3936 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3937 result_mem->init_req(_false_path, input_memory_state);
3938
3939 // Result IO.
3940 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3941 result_io->init_req(_false_path, input_io_state);
3942
3943 // Result value.
3944 result_value->init_req(_true_path, event_writer); // return event writer oop
3945 result_value->init_req(_false_path, null()); // return null
3946
3947 // Set output state.
3948 set_control(_gvn.transform(result_rgn));
3949 set_all_memory(_gvn.transform(result_mem));
3950 set_i_o(_gvn.transform(result_io));
3951 set_result(result_rgn, result_value);
3952 return true;
3953 }
3954
3955 /*
3956 * The intrinsic is a model of this pseudo-code:
3957 *
3958 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3959 * if (carrierThread != thread) { // is virtual thread
3960 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3961 * bool excluded = vthread_epoch_raw & excluded_mask;
3962 * AtomicAccess::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3963 * AtomicAccess::store(&tl->_contextual_thread_excluded, is_excluded);
3964 * if (!excluded) {
3965 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3966 * AtomicAccess::store(&tl->_vthread_epoch, vthread_epoch);
3967 * }
3968 * AtomicAccess::release_store(&tl->_vthread, true);
3969 * return;
3970 * }
3971 * AtomicAccess::release_store(&tl->_vthread, false);
3972 */
3973 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3974 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3975
3976 Node* input_memory_state = reset_memory();
3977 set_all_memory(input_memory_state);
3978
3979 // The most significant bit of the u2 is used to denote thread exclusion
3980 Node* excluded_mask = _gvn.intcon(1 << 15);
3981 // The epoch generation is the range [1-32767]
3982 Node* epoch_mask = _gvn.intcon(32767);
3983
3984 Node* const carrierThread = generate_current_thread(jt);
3985 // If thread != carrierThread, this is a virtual thread.
3986 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3987 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3988 IfNode* iff_thread_not_equal_carrierThread =
3989 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3990
3991 Node* vthread_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3992
3993 // False branch, is carrierThread.
3994 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3995 // Store release
3996 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3997
3998 set_all_memory(input_memory_state);
3999
4000 // True branch, is virtual thread.
4001 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
4002 set_control(thread_not_equal_carrierThread);
4003
4004 // Load the raw epoch value from the vthread.
4005 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
4006 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
4007 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
4008
4009 // Mask off the excluded information from the epoch.
4010 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, excluded_mask));
4011
4012 // Load the tid field from the thread.
4013 Node* tid = load_field_from_object(thread, "tid", "J");
4014
4015 // Store the vthread tid to the jfr thread local.
4016 Node* thread_id_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
4017 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
4018
4019 // Branch is_excluded to conditionalize updating the epoch .
4020 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, excluded_mask));
4021 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
4022 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
4023
4024 // True branch, vthread is excluded, no need to write epoch info.
4025 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
4026 set_control(excluded);
4027 Node* vthread_is_excluded = _gvn.intcon(1);
4028
4029 // False branch, vthread is included, update epoch info.
4030 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
4031 set_control(included);
4032 Node* vthread_is_included = _gvn.intcon(0);
4033
4034 // Get epoch value.
4035 Node* epoch = _gvn.transform(new AndINode(epoch_raw, epoch_mask));
4036
4037 // Store the vthread epoch to the jfr thread local.
4038 Node* vthread_epoch_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
4039 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
4040
4041 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
4042 record_for_igvn(excluded_rgn);
4043 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
4044 record_for_igvn(excluded_mem);
4045 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
4046 record_for_igvn(exclusion);
4047
4048 // Merge the excluded control and memory.
4049 excluded_rgn->init_req(_true_path, excluded);
4050 excluded_rgn->init_req(_false_path, included);
4051 excluded_mem->init_req(_true_path, tid_memory);
4052 excluded_mem->init_req(_false_path, included_memory);
4053 exclusion->init_req(_true_path, vthread_is_excluded);
4054 exclusion->init_req(_false_path, vthread_is_included);
4055
4056 // Set intermediate state.
4057 set_control(_gvn.transform(excluded_rgn));
4058 set_all_memory(excluded_mem);
4059
4060 // Store the vthread exclusion state to the jfr thread local.
4061 Node* thread_local_excluded_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
4062 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
4063
4064 // Store release
4065 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
4066
4067 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
4068 record_for_igvn(thread_compare_rgn);
4069 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
4070 record_for_igvn(thread_compare_mem);
4071 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
4072 record_for_igvn(vthread);
4073
4074 // Merge the thread_compare control and memory.
4075 thread_compare_rgn->init_req(_true_path, control());
4076 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
4077 thread_compare_mem->init_req(_true_path, vthread_true_memory);
4078 thread_compare_mem->init_req(_false_path, vthread_false_memory);
4079
4080 // Set output state.
4081 set_control(_gvn.transform(thread_compare_rgn));
4082 set_all_memory(_gvn.transform(thread_compare_mem));
4083 }
4084
4085 #endif // JFR_HAVE_INTRINSICS
4086
4087 //------------------------inline_native_currentCarrierThread------------------
4088 bool LibraryCallKit::inline_native_currentCarrierThread() {
4089 Node* junk = nullptr;
4090 set_result(generate_current_thread(junk));
4091 return true;
4092 }
4093
4094 //------------------------inline_native_currentThread------------------
4095 bool LibraryCallKit::inline_native_currentThread() {
4096 Node* junk = nullptr;
4097 set_result(generate_virtual_thread(junk));
4098 return true;
4099 }
4100
4101 //------------------------inline_native_setVthread------------------
4102 bool LibraryCallKit::inline_native_setCurrentThread() {
4103 assert(C->method()->changes_current_thread(),
4104 "method changes current Thread but is not annotated ChangesCurrentThread");
4105 Node* arr = argument(1);
4106 Node* thread = _gvn.transform(new ThreadLocalNode());
4107 Node* p = off_heap_plus_addr(thread, in_bytes(JavaThread::vthread_offset()));
4108 Node* thread_obj_handle
4109 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
4110 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
4111 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
4112
4113 // Change the _monitor_owner_id of the JavaThread
4114 Node* tid = load_field_from_object(arr, "tid", "J");
4115 Node* monitor_owner_id_offset = off_heap_plus_addr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
4116 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
4117
4118 JFR_ONLY(extend_setCurrentThread(thread, arr);)
4119 return true;
4120 }
4121
4122 const Type* LibraryCallKit::scopedValueCache_type() {
4123 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
4124 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
4125 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS, /* stable= */ false, /* flat= */ false, /* not_flat= */ true, /* not_null_free= */ true, true);
4126
4127 // Because we create the scopedValue cache lazily we have to make the
4128 // type of the result BotPTR.
4129 bool xk = etype->klass_is_exact();
4130 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, TypeAryPtr::Offset(0));
4131 return objects_type;
4132 }
4133
4134 Node* LibraryCallKit::scopedValueCache_helper() {
4135 Node* thread = _gvn.transform(new ThreadLocalNode());
4136 Node* p = off_heap_plus_addr(thread, in_bytes(JavaThread::scopedValueCache_offset()));
4137 // We cannot use immutable_memory() because we might flip onto a
4138 // different carrier thread, at which point we'll need to use that
4139 // carrier thread's cache.
4140 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
4141 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
4142 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
4143 }
4144
4145 //------------------------inline_native_scopedValueCache------------------
4146 bool LibraryCallKit::inline_native_scopedValueCache() {
4147 Node* cache_obj_handle = scopedValueCache_helper();
4148 const Type* objects_type = scopedValueCache_type();
4149 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
4150
4151 return true;
4152 }
4153
4154 //------------------------inline_native_setScopedValueCache------------------
4155 bool LibraryCallKit::inline_native_setScopedValueCache() {
4156 Node* arr = argument(0);
4157 Node* cache_obj_handle = scopedValueCache_helper();
4158 const Type* objects_type = scopedValueCache_type();
4159
4160 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
4161 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
4162
4163 return true;
4164 }
4165
4166 //------------------------inline_native_Continuation_pin and unpin-----------
4167
4168 // Shared implementation routine for both pin and unpin.
4169 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
4170 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
4171
4172 // Save input memory.
4173 Node* input_memory_state = reset_memory();
4174 set_all_memory(input_memory_state);
4175
4176 // TLS
4177 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
4178 Node* last_continuation_offset = off_heap_plus_addr(tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
4179 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
4180
4181 // Null check the last continuation object.
4182 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
4183 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
4184 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
4185
4186 // False path, last continuation is null.
4187 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
4188
4189 // True path, last continuation is not null.
4190 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
4191
4192 set_control(continuation_is_not_null);
4193
4194 // Load the pin count from the last continuation.
4195 Node* pin_count_offset = off_heap_plus_addr(last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
4196 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
4197
4198 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
4199 Node* pin_count_rhs;
4200 if (unpin) {
4201 pin_count_rhs = _gvn.intcon(0);
4202 } else {
4203 pin_count_rhs = _gvn.intcon(UINT32_MAX);
4204 }
4205 Node* pin_count_cmp = _gvn.transform(new CmpUNode(pin_count, pin_count_rhs));
4206 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
4207 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
4208
4209 // True branch, pin count over/underflow.
4210 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
4211 {
4212 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
4213 // which will throw IllegalStateException for pin count over/underflow.
4214 // No memory changed so far - we can use memory create by reset_memory()
4215 // at the beginning of this intrinsic. No need to call reset_memory() again.
4216 PreserveJVMState pjvms(this);
4217 set_control(pin_count_over_underflow);
4218 uncommon_trap(Deoptimization::Reason_intrinsic,
4219 Deoptimization::Action_none);
4220 assert(stopped(), "invariant");
4221 }
4222
4223 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
4224 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
4225 set_control(valid_pin_count);
4226
4227 Node* next_pin_count;
4228 if (unpin) {
4229 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
4230 } else {
4231 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
4232 }
4233
4234 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
4235
4236 // Result of top level CFG and Memory.
4237 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
4238 record_for_igvn(result_rgn);
4239 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
4240 record_for_igvn(result_mem);
4241
4242 result_rgn->init_req(_true_path, valid_pin_count);
4243 result_rgn->init_req(_false_path, continuation_is_null);
4244 result_mem->init_req(_true_path, reset_memory());
4245 result_mem->init_req(_false_path, input_memory_state);
4246
4247 // Set output state.
4248 set_control(_gvn.transform(result_rgn));
4249 set_all_memory(_gvn.transform(result_mem));
4250
4251 return true;
4252 }
4253
4254 //---------------------------load_mirror_from_klass----------------------------
4255 // Given a klass oop, load its java mirror (a java.lang.Class oop).
4256 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
4257 Node* p = off_heap_plus_addr(klass, in_bytes(Klass::java_mirror_offset()));
4258 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4259 // mirror = ((OopHandle)mirror)->resolve();
4260 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
4261 }
4262
4263 //-----------------------load_klass_from_mirror_common-------------------------
4264 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
4265 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
4266 // and branch to the given path on the region.
4267 // If never_see_null, take an uncommon trap on null, so we can optimistically
4268 // compile for the non-null case.
4269 // If the region is null, force never_see_null = true.
4270 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
4271 bool never_see_null,
4272 RegionNode* region,
4273 int null_path,
4274 int offset) {
4275 if (region == nullptr) never_see_null = true;
4276 Node* p = basic_plus_adr(mirror, offset);
4277 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4278 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
4279 Node* null_ctl = top();
4280 kls = null_check_oop(kls, &null_ctl, never_see_null);
4281 if (region != nullptr) {
4282 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
4283 region->init_req(null_path, null_ctl);
4284 } else {
4285 assert(null_ctl == top(), "no loose ends");
4286 }
4287 return kls;
4288 }
4289
4290 //--------------------(inline_native_Class_query helpers)---------------------
4291 // Use this for JVM_ACC_INTERFACE.
4292 // Fall through if (mods & mask) == bits, take the guard otherwise.
4293 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
4294 ByteSize offset, const Type* type, BasicType bt) {
4295 // Branch around if the given klass has the given modifier bit set.
4296 // Like generate_guard, adds a new path onto the region.
4297 Node* modp = off_heap_plus_addr(kls, in_bytes(offset));
4298 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
4299 Node* mask = intcon(modifier_mask);
4300 Node* bits = intcon(modifier_bits);
4301 Node* mbit = _gvn.transform(new AndINode(mods, mask));
4302 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
4303 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
4304 return generate_fair_guard(bol, region);
4305 }
4306
4307 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
4308 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
4309 InstanceKlass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
4310 }
4311
4312 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
4313 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
4314 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
4315 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
4316 }
4317
4318 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
4319 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
4320 }
4321
4322 //-------------------------inline_native_Class_query-------------------
4323 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
4324 const Type* return_type = TypeInt::BOOL;
4325 Node* prim_return_value = top(); // what happens if it's a primitive class?
4326 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4327 bool expect_prim = false; // most of these guys expect to work on refs
4328
4329 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
4330
4331 Node* mirror = argument(0);
4332 Node* obj = top();
4333
4334 switch (id) {
4335 case vmIntrinsics::_isInstance:
4336 // nothing is an instance of a primitive type
4337 prim_return_value = intcon(0);
4338 obj = argument(1);
4339 break;
4340 case vmIntrinsics::_isHidden:
4341 prim_return_value = intcon(0);
4342 break;
4343 case vmIntrinsics::_getSuperclass:
4344 prim_return_value = null();
4345 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
4346 break;
4347 default:
4348 fatal_unexpected_iid(id);
4349 break;
4350 }
4351
4352 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4353 if (mirror_con == nullptr) return false; // cannot happen?
4354
4355 #ifndef PRODUCT
4356 if (C->print_intrinsics() || C->print_inlining()) {
4357 ciType* k = mirror_con->java_mirror_type();
4358 if (k) {
4359 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4360 k->print_name();
4361 tty->cr();
4362 }
4363 }
4364 #endif
4365
4366 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4367 RegionNode* region = new RegionNode(PATH_LIMIT);
4368 record_for_igvn(region);
4369 PhiNode* phi = new PhiNode(region, return_type);
4370
4371 // The mirror will never be null of Reflection.getClassAccessFlags, however
4372 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4373 // if it is. See bug 4774291.
4374
4375 // For Reflection.getClassAccessFlags(), the null check occurs in
4376 // the wrong place; see inline_unsafe_access(), above, for a similar
4377 // situation.
4378 mirror = null_check(mirror);
4379 // If mirror or obj is dead, only null-path is taken.
4380 if (stopped()) return true;
4381
4382 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4383
4384 // Now load the mirror's klass metaobject, and null-check it.
4385 // Side-effects region with the control path if the klass is null.
4386 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4387 // If kls is null, we have a primitive mirror.
4388 phi->init_req(_prim_path, prim_return_value);
4389 if (stopped()) { set_result(region, phi); return true; }
4390 bool safe_for_replace = (region->in(_prim_path) == top());
4391
4392 Node* p; // handy temp
4393 Node* null_ctl;
4394
4395 // Now that we have the non-null klass, we can perform the real query.
4396 // For constant classes, the query will constant-fold in LoadNode::Value.
4397 Node* query_value = top();
4398 switch (id) {
4399 case vmIntrinsics::_isInstance:
4400 // nothing is an instance of a primitive type
4401 query_value = gen_instanceof(obj, kls, safe_for_replace);
4402 break;
4403
4404 case vmIntrinsics::_isHidden:
4405 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4406 if (generate_hidden_class_guard(kls, region) != nullptr)
4407 // A guard was added. If the guard is taken, it was an hidden class.
4408 phi->add_req(intcon(1));
4409 // If we fall through, it's a plain class.
4410 query_value = intcon(0);
4411 break;
4412
4413
4414 case vmIntrinsics::_getSuperclass:
4415 // The rules here are somewhat unfortunate, but we can still do better
4416 // with random logic than with a JNI call.
4417 // Interfaces store null or Object as _super, but must report null.
4418 // Arrays store an intermediate super as _super, but must report Object.
4419 // Other types can report the actual _super.
4420 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4421 if (generate_array_guard(kls, region) != nullptr) {
4422 // A guard was added. If the guard is taken, it was an array.
4423 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4424 }
4425 // Check for interface after array since this checks AccessFlags offset into InstanceKlass.
4426 // In other words, we are accessing subtype-specific information, so we need to determine the subtype first.
4427 if (generate_interface_guard(kls, region) != nullptr) {
4428 // A guard was added. If the guard is taken, it was an interface.
4429 phi->add_req(null());
4430 }
4431 // If we fall through, it's a plain class. Get its _super.
4432 if (!stopped()) {
4433 p = basic_plus_adr(top(), kls, in_bytes(Klass::super_offset()));
4434 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4435 null_ctl = top();
4436 kls = null_check_oop(kls, &null_ctl);
4437 if (null_ctl != top()) {
4438 // If the guard is taken, Object.superClass is null (both klass and mirror).
4439 region->add_req(null_ctl);
4440 phi ->add_req(null());
4441 }
4442 if (!stopped()) {
4443 query_value = load_mirror_from_klass(kls);
4444 }
4445 }
4446 break;
4447
4448 default:
4449 fatal_unexpected_iid(id);
4450 break;
4451 }
4452
4453 // Fall-through is the normal case of a query to a real class.
4454 phi->init_req(1, query_value);
4455 region->init_req(1, control());
4456
4457 C->set_has_split_ifs(true); // Has chance for split-if optimization
4458 set_result(region, phi);
4459 return true;
4460 }
4461
4462
4463 //-------------------------inline_Class_cast-------------------
4464 bool LibraryCallKit::inline_Class_cast() {
4465 Node* mirror = argument(0); // Class
4466 Node* obj = argument(1);
4467 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4468 if (mirror_con == nullptr) {
4469 return false; // dead path (mirror->is_top()).
4470 }
4471 if (obj == nullptr || obj->is_top()) {
4472 return false; // dead path
4473 }
4474 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4475
4476 // First, see if Class.cast() can be folded statically.
4477 // java_mirror_type() returns non-null for compile-time Class constants.
4478 ciType* tm = mirror_con->java_mirror_type();
4479 if (tm != nullptr && tm->is_klass() &&
4480 tp != nullptr) {
4481 if (!tp->is_loaded()) {
4482 // Don't use intrinsic when class is not loaded.
4483 return false;
4484 } else {
4485 const TypeKlassPtr* tklass = TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces);
4486 int static_res = C->static_subtype_check(tklass, tp->as_klass_type());
4487 if (static_res == Compile::SSC_always_true) {
4488 // isInstance() is true - fold the code.
4489 set_result(obj);
4490 return true;
4491 } else if (static_res == Compile::SSC_always_false) {
4492 // Don't use intrinsic, have to throw ClassCastException.
4493 // If the reference is null, the non-intrinsic bytecode will
4494 // be optimized appropriately.
4495 return false;
4496 }
4497 }
4498 }
4499
4500 // Bailout intrinsic and do normal inlining if exception path is frequent.
4501 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4502 return false;
4503 }
4504
4505 // Generate dynamic checks.
4506 // Class.cast() is java implementation of _checkcast bytecode.
4507 // Do checkcast (Parse::do_checkcast()) optimizations here.
4508
4509 mirror = null_check(mirror);
4510 // If mirror is dead, only null-path is taken.
4511 if (stopped()) {
4512 return true;
4513 }
4514
4515 // Not-subtype or the mirror's klass ptr is nullptr (in case it is a primitive).
4516 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
4517 RegionNode* region = new RegionNode(PATH_LIMIT);
4518 record_for_igvn(region);
4519
4520 // Now load the mirror's klass metaobject, and null-check it.
4521 // If kls is null, we have a primitive mirror and
4522 // nothing is an instance of a primitive type.
4523 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4524
4525 Node* res = top();
4526 Node* io = i_o();
4527 Node* mem = merged_memory();
4528 SafePointNode* new_cast_failure_map = nullptr;
4529
4530 if (!stopped()) {
4531
4532 Node* bad_type_ctrl = top();
4533 // Do checkcast optimizations.
4534 res = gen_checkcast(obj, kls, &bad_type_ctrl, &new_cast_failure_map);
4535 region->init_req(_bad_type_path, bad_type_ctrl);
4536 }
4537 if (region->in(_prim_path) != top() ||
4538 region->in(_bad_type_path) != top() ||
4539 region->in(_npe_path) != top()) {
4540 // Let Interpreter throw ClassCastException.
4541 PreserveJVMState pjvms(this);
4542 if (new_cast_failure_map != nullptr) {
4543 // The current map on the success path could have been modified. Use the dedicated failure path map.
4544 set_map(new_cast_failure_map);
4545 }
4546 set_control(_gvn.transform(region));
4547 // Set IO and memory because gen_checkcast may override them when buffering inline types
4548 set_i_o(io);
4549 set_all_memory(mem);
4550 uncommon_trap(Deoptimization::Reason_intrinsic,
4551 Deoptimization::Action_maybe_recompile);
4552 }
4553 if (!stopped()) {
4554 set_result(res);
4555 }
4556 return true;
4557 }
4558
4559
4560 //--------------------------inline_native_subtype_check------------------------
4561 // This intrinsic takes the JNI calls out of the heart of
4562 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4563 bool LibraryCallKit::inline_native_subtype_check() {
4564 // Pull both arguments off the stack.
4565 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4566 args[0] = argument(0);
4567 args[1] = argument(1);
4568 Node* klasses[2]; // corresponding Klasses: superk, subk
4569 klasses[0] = klasses[1] = top();
4570
4571 enum {
4572 // A full decision tree on {superc is prim, subc is prim}:
4573 _prim_0_path = 1, // {P,N} => false
4574 // {P,P} & superc!=subc => false
4575 _prim_same_path, // {P,P} & superc==subc => true
4576 _prim_1_path, // {N,P} => false
4577 _ref_subtype_path, // {N,N} & subtype check wins => true
4578 _both_ref_path, // {N,N} & subtype check loses => false
4579 PATH_LIMIT
4580 };
4581
4582 RegionNode* region = new RegionNode(PATH_LIMIT);
4583 RegionNode* prim_region = new RegionNode(2);
4584 Node* phi = new PhiNode(region, TypeInt::BOOL);
4585 record_for_igvn(region);
4586 record_for_igvn(prim_region);
4587
4588 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4589 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4590 int class_klass_offset = java_lang_Class::klass_offset();
4591
4592 // First null-check both mirrors and load each mirror's klass metaobject.
4593 int which_arg;
4594 for (which_arg = 0; which_arg <= 1; which_arg++) {
4595 Node* arg = args[which_arg];
4596 arg = null_check(arg);
4597 if (stopped()) break;
4598 args[which_arg] = arg;
4599
4600 Node* p = basic_plus_adr(arg, class_klass_offset);
4601 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4602 klasses[which_arg] = _gvn.transform(kls);
4603 }
4604
4605 // Having loaded both klasses, test each for null.
4606 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4607 for (which_arg = 0; which_arg <= 1; which_arg++) {
4608 Node* kls = klasses[which_arg];
4609 Node* null_ctl = top();
4610 kls = null_check_oop(kls, &null_ctl, never_see_null);
4611 if (which_arg == 0) {
4612 prim_region->init_req(1, null_ctl);
4613 } else {
4614 region->init_req(_prim_1_path, null_ctl);
4615 }
4616 if (stopped()) break;
4617 klasses[which_arg] = kls;
4618 }
4619
4620 if (!stopped()) {
4621 // now we have two reference types, in klasses[0..1]
4622 Node* subk = klasses[1]; // the argument to isAssignableFrom
4623 Node* superk = klasses[0]; // the receiver
4624 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4625 region->set_req(_ref_subtype_path, control());
4626 }
4627
4628 // If both operands are primitive (both klasses null), then
4629 // we must return true when they are identical primitives.
4630 // It is convenient to test this after the first null klass check.
4631 // This path is also used if superc is a value mirror.
4632 set_control(_gvn.transform(prim_region));
4633 if (!stopped()) {
4634 // Since superc is primitive, make a guard for the superc==subc case.
4635 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4636 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4637 generate_fair_guard(bol_eq, region);
4638 if (region->req() == PATH_LIMIT+1) {
4639 // A guard was added. If the added guard is taken, superc==subc.
4640 region->swap_edges(PATH_LIMIT, _prim_same_path);
4641 region->del_req(PATH_LIMIT);
4642 }
4643 region->set_req(_prim_0_path, control()); // Not equal after all.
4644 }
4645
4646 // these are the only paths that produce 'true':
4647 phi->set_req(_prim_same_path, intcon(1));
4648 phi->set_req(_ref_subtype_path, intcon(1));
4649
4650 // pull together the cases:
4651 assert(region->req() == PATH_LIMIT, "sane region");
4652 for (uint i = 1; i < region->req(); i++) {
4653 Node* ctl = region->in(i);
4654 if (ctl == nullptr || ctl == top()) {
4655 region->set_req(i, top());
4656 phi ->set_req(i, top());
4657 } else if (phi->in(i) == nullptr) {
4658 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4659 }
4660 }
4661
4662 set_control(_gvn.transform(region));
4663 set_result(_gvn.transform(phi));
4664 return true;
4665 }
4666
4667 //---------------------generate_array_guard_common------------------------
4668 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind, Node** obj) {
4669
4670 if (stopped()) {
4671 return nullptr;
4672 }
4673
4674 // Like generate_guard, adds a new path onto the region.
4675 jint layout_con = 0;
4676 Node* layout_val = get_layout_helper(kls, layout_con);
4677 if (layout_val == nullptr) {
4678 bool query = 0;
4679 switch(kind) {
4680 case RefArray: query = Klass::layout_helper_is_refArray(layout_con); break;
4681 case NonRefArray: query = !Klass::layout_helper_is_refArray(layout_con); break;
4682 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
4683 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
4684 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
4685 default:
4686 ShouldNotReachHere();
4687 }
4688 if (!query) {
4689 return nullptr; // never a branch
4690 } else { // always a branch
4691 Node* always_branch = control();
4692 if (region != nullptr)
4693 region->add_req(always_branch);
4694 set_control(top());
4695 return always_branch;
4696 }
4697 }
4698 unsigned int value = 0;
4699 BoolTest::mask btest = BoolTest::illegal;
4700 switch(kind) {
4701 case RefArray:
4702 case NonRefArray: {
4703 value = Klass::_lh_array_tag_ref_value;
4704 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4705 btest = (kind == RefArray) ? BoolTest::eq : BoolTest::ne;
4706 break;
4707 }
4708 case TypeArray: {
4709 value = Klass::_lh_array_tag_type_value;
4710 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4711 btest = BoolTest::eq;
4712 break;
4713 }
4714 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
4715 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
4716 default:
4717 ShouldNotReachHere();
4718 }
4719 // Now test the correct condition.
4720 jint nval = (jint)value;
4721 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4722 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4723 Node* ctrl = generate_fair_guard(bol, region);
4724 Node* is_array_ctrl = kind == NonArray ? control() : ctrl;
4725 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4726 // Keep track of the fact that 'obj' is an array to prevent
4727 // array specific accesses from floating above the guard.
4728 *obj = _gvn.transform(new CheckCastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4729 }
4730 return ctrl;
4731 }
4732
4733 // public static native Object[] ValueClass::newNullRestrictedAtomicArray(Class<?> componentType, int length, Object initVal);
4734 // public static native Object[] ValueClass::newNullRestrictedNonAtomicArray(Class<?> componentType, int length, Object initVal);
4735 // public static native Object[] ValueClass::newNullableAtomicArray(Class<?> componentType, int length);
4736 bool LibraryCallKit::inline_newArray(bool null_free, bool atomic) {
4737 assert(null_free || atomic, "nullable implies atomic");
4738 Node* componentType = argument(0);
4739 Node* length = argument(1);
4740 Node* init_val = null_free ? argument(2) : nullptr;
4741
4742 const TypeInstPtr* tp = _gvn.type(componentType)->isa_instptr();
4743 if (tp != nullptr) {
4744 ciInstanceKlass* ik = tp->instance_klass();
4745 if (ik == C->env()->Class_klass()) {
4746 ciType* t = tp->java_mirror_type();
4747 if (t != nullptr && t->is_inlinetype()) {
4748
4749 ciArrayKlass* array_klass = ciArrayKlass::make(t, null_free, atomic, true);
4750 assert(array_klass->is_elem_null_free() == null_free, "inconsistency");
4751
4752 // TOOD 8350865 ZGC needs card marks on initializing oop stores
4753 if (UseZGC && null_free && !array_klass->is_flat_array_klass()) {
4754 return false;
4755 }
4756
4757 if (array_klass->is_loaded() && array_klass->element_klass()->as_inline_klass()->is_initialized()) {
4758 const TypeAryKlassPtr* array_klass_type = TypeAryKlassPtr::make(array_klass, Type::trust_interfaces);
4759 if (null_free) {
4760 if (init_val->is_InlineType()) {
4761 if (array_klass_type->is_flat() && init_val->as_InlineType()->is_all_zero(&gvn(), /* flat */ true)) {
4762 // Zeroing is enough because the init value is the all-zero value
4763 init_val = nullptr;
4764 } else {
4765 init_val = init_val->as_InlineType()->buffer(this);
4766 }
4767 }
4768 if (init_val != nullptr) {
4769 #ifdef ASSERT
4770 init_val = null_check(init_val);
4771 Node* wrong_type_ctl = gen_subtype_check(init_val, makecon(TypeKlassPtr::make(array_klass->element_klass())));
4772 {
4773 PreserveJVMState pjvms(this);
4774 set_control(wrong_type_ctl);
4775 halt(control(), frameptr(), "incompatible type for initVal in newArray");
4776 stop_and_kill_map();
4777 }
4778 #endif
4779 init_val = _gvn.transform(new CheckCastPPNode(control(), init_val, TypeOopPtr::make_from_klass(array_klass->element_klass()), ConstraintCastNode::DependencyType::NonFloatingNarrowing));
4780 }
4781 }
4782 Node* obj = new_array(makecon(array_klass_type), length, 0, nullptr, false, init_val);
4783 const TypeAryPtr* arytype = gvn().type(obj)->is_aryptr();
4784 assert(arytype->is_null_free() == null_free, "inconsistency");
4785 assert(arytype->is_not_null_free() == !null_free, "inconsistency");
4786 set_result(obj);
4787 return true;
4788 }
4789 }
4790 }
4791 }
4792 return false;
4793 }
4794
4795 // public static native boolean ValueClass::isFlatArray(Object array);
4796 // public static native boolean ValueClass::isNullRestrictedArray(Object array);
4797 // public static native boolean ValueClass::isAtomicArray(Object array);
4798 bool LibraryCallKit::inline_getArrayProperties(ArrayPropertiesCheck check) {
4799 Node* array = argument(0);
4800
4801 Node* bol;
4802 switch(check) {
4803 case IsFlat:
4804 bol = flat_array_test(load_object_klass(array));
4805 break;
4806 case IsNullRestricted:
4807 bol = null_free_array_test(array);
4808 break;
4809 case IsAtomic: {
4810 // See conditions in JVM_IsAtomicArray
4811 // 1. If not flat, then atomic, or else...
4812 RegionNode* atomic_region = new RegionNode(1);
4813 RegionNode* non_atomic_region = new RegionNode(1);
4814 Node* array_klass = load_object_klass(array);
4815 Node* is_flat_bol = flat_array_test(array_klass);
4816 IfNode* iff_is_flat = create_and_xform_if(control(), is_flat_bol, PROB_FAIR, COUNT_UNKNOWN);
4817 atomic_region->add_req(_gvn.transform(new IfFalseNode(iff_is_flat)));
4818 set_control(_gvn.transform(new IfTrueNode(iff_is_flat)));
4819
4820 // 2. ...if the layout is atomic, then atomic, or else...
4821 Node* layout_kind = atomic_layout_array_test_and_get_layout_kind(array, atomic_region);
4822
4823 // 3. ...if the element type is naturally atomic and null-free OR empty and nullable, then atomic, or else...
4824 int element_klass_offset = in_bytes(ObjArrayKlass::element_klass_offset());
4825 Node* array_element_klass_addr = off_heap_plus_addr(array_klass, element_klass_offset);
4826 Node* array_element_klass = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), array_element_klass_addr, _gvn.type(array_klass)->is_klassptr()));
4827 int klass_flags_offset = in_bytes(InstanceKlass::misc_flags_offset() + InstanceKlassFlags::flags_offset());
4828 Node* array_element_klass_flags_addr = off_heap_plus_addr(array_element_klass, klass_flags_offset);
4829 Node* array_element_klass_flags = make_load(control(), array_element_klass_flags_addr, TypeInt::INT, T_INT, MemNode::unordered);
4830
4831 // Here, layout can only be non-atomic, otherwise atomic_layout_array_test_and_get_layout_kind already decides the array to be atomic.
4832 Node* is_null_free_cmp = _gvn.transform(new CmpINode(layout_kind, intcon(static_cast<jint>(LayoutKind::NULL_FREE_NON_ATOMIC_FLAT))));
4833 Node* is_null_free_bol = _gvn.transform(new BoolNode(is_null_free_cmp, BoolTest::eq));
4834 IfNode* iff_is_null_free_bol = create_and_xform_if(control(), is_null_free_bol, PROB_FAIR, COUNT_UNKNOWN);
4835 Node* is_null_free_ctl = _gvn.transform(new IfTrueNode(iff_is_null_free_bol));
4836 Node* is_nullable_ctl = _gvn.transform(new IfFalseNode(iff_is_null_free_bol));
4837
4838 Node* is_naturally_atomic_flag = _gvn.transform(new AndINode(array_element_klass_flags, intcon(InstanceKlassFlags::_misc_is_naturally_atomic)));
4839 Node* is_naturally_atomic_cmp = _gvn.transform(new CmpINode(is_naturally_atomic_flag, intcon(0)));
4840 Node* is_naturally_atomic_bol = _gvn.transform(new BoolNode(is_naturally_atomic_cmp, BoolTest::ne));
4841 IfNode* iff_is_naturally_atomic = create_and_xform_if(is_null_free_ctl, is_naturally_atomic_bol, PROB_FAIR, COUNT_UNKNOWN);
4842 Node* is_naturally_atomic_ctl = _gvn.transform(new IfTrueNode(iff_is_naturally_atomic));
4843 Node* is_not_naturally_atomic_ctl = _gvn.transform(new IfFalseNode(iff_is_naturally_atomic));
4844 atomic_region->add_req(is_naturally_atomic_ctl);
4845 non_atomic_region->add_req(is_not_naturally_atomic_ctl);
4846
4847 Node* is_empty_inline_type_flag = _gvn.transform(new AndINode(array_element_klass_flags, intcon(InstanceKlassFlags::_misc_is_empty_inline_type)));
4848 Node* is_empty_inline_type_cmp = _gvn.transform(new CmpINode(is_empty_inline_type_flag, intcon(0)));
4849 Node* is_empty_inline_type_bol = _gvn.transform(new BoolNode(is_empty_inline_type_cmp, BoolTest::ne));
4850 IfNode* iff_is_empty_inline_type = create_and_xform_if(is_nullable_ctl, is_empty_inline_type_bol, PROB_FAIR, COUNT_UNKNOWN);
4851 Node* is_empty_inline_type_ctl = _gvn.transform(new IfTrueNode(iff_is_empty_inline_type));
4852 Node* is_nonempty_inline_type_ctl = _gvn.transform(new IfFalseNode(iff_is_empty_inline_type));
4853 atomic_region->add_req(is_empty_inline_type_ctl);
4854 non_atomic_region->add_req(is_nonempty_inline_type_ctl);
4855
4856 // ...non-atomic, but we tried everything.
4857 RegionNode* decision = new RegionNode(3);
4858 decision->set_req(1, _gvn.transform(atomic_region));
4859 decision->set_req(2, _gvn.transform(non_atomic_region));
4860 PhiNode* result = PhiNode::make(decision, intcon(1), TypeInt::BOOL);
4861 result->set_req(2, intcon(0));
4862 set_control(_gvn.transform(decision));
4863 set_result(_gvn.transform(result));
4864 return true;
4865 }
4866 default:
4867 ShouldNotReachHere();
4868 }
4869
4870 Node* res = gvn().transform(new CMoveINode(bol, intcon(0), intcon(1), TypeInt::BOOL));
4871 set_result(res);
4872 return true;
4873 }
4874
4875 // Load the default refined array klass from an ObjArrayKlass. This relies on the first entry in the
4876 // '_next_refined_array_klass' linked list being the default (see ObjArrayKlass::klass_with_properties).
4877 Node* LibraryCallKit::load_default_refined_array_klass(Node* klass_node, bool type_array_guard) {
4878 RegionNode* region = new RegionNode(2);
4879 Node* phi = new PhiNode(region, TypeInstKlassPtr::OBJECT_OR_NULL);
4880
4881 if (type_array_guard) {
4882 generate_typeArray_guard(klass_node, region);
4883 if (region->req() == 3) {
4884 phi->add_req(klass_node);
4885 }
4886 }
4887 Node* adr_refined_klass = basic_plus_adr(top(), klass_node, in_bytes(ObjArrayKlass::next_refined_array_klass_offset()));
4888 Node* refined_klass = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), adr_refined_klass, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4889
4890 // Can be null if not initialized yet, just deopt
4891 Node* null_ctl = top();
4892 refined_klass = null_check_oop(refined_klass, &null_ctl, /* never_see_null= */ true);
4893
4894 region->init_req(1, control());
4895 phi->init_req(1, refined_klass);
4896
4897 set_control(_gvn.transform(region));
4898 return _gvn.transform(phi);
4899 }
4900
4901 // Load the non-refined array klass from an ObjArrayKlass.
4902 Node* LibraryCallKit::load_non_refined_array_klass(Node* klass_node) {
4903 const TypeAryKlassPtr* ary_klass_ptr = _gvn.type(klass_node)->isa_aryklassptr();
4904 if (ary_klass_ptr != nullptr && ary_klass_ptr->klass_is_exact()) {
4905 return _gvn.makecon(ary_klass_ptr->cast_to_refined_array_klass_ptr(false));
4906 }
4907
4908 RegionNode* region = new RegionNode(2);
4909 Node* phi = new PhiNode(region, TypeInstKlassPtr::OBJECT);
4910
4911 generate_typeArray_guard(klass_node, region);
4912 if (region->req() == 3) {
4913 phi->add_req(klass_node);
4914 }
4915 Node* super_adr = basic_plus_adr(top(), klass_node, in_bytes(Klass::super_offset()));
4916 Node* super_klass = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), super_adr, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT));
4917
4918 region->init_req(1, control());
4919 phi->init_req(1, super_klass);
4920
4921 set_control(_gvn.transform(region));
4922 return _gvn.transform(phi);
4923 }
4924
4925 //-----------------------inline_native_newArray--------------------------
4926 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
4927 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4928 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4929 Node* mirror;
4930 Node* count_val;
4931 if (uninitialized) {
4932 null_check_receiver();
4933 mirror = argument(1);
4934 count_val = argument(2);
4935 } else {
4936 mirror = argument(0);
4937 count_val = argument(1);
4938 }
4939
4940 mirror = null_check(mirror);
4941 // If mirror or obj is dead, only null-path is taken.
4942 if (stopped()) return true;
4943
4944 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4945 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4946 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4947 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4948 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4949
4950 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4951 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4952 result_reg, _slow_path);
4953 Node* normal_ctl = control();
4954 Node* no_array_ctl = result_reg->in(_slow_path);
4955
4956 // Generate code for the slow case. We make a call to newArray().
4957 set_control(no_array_ctl);
4958 if (!stopped()) {
4959 // Either the input type is void.class, or else the
4960 // array klass has not yet been cached. Either the
4961 // ensuing call will throw an exception, or else it
4962 // will cache the array klass for next time.
4963 PreserveJVMState pjvms(this);
4964 CallJavaNode* slow_call = nullptr;
4965 if (uninitialized) {
4966 // Generate optimized virtual call (holder class 'Unsafe' is final)
4967 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4968 } else {
4969 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4970 }
4971 Node* slow_result = set_results_for_java_call(slow_call);
4972 // this->control() comes from set_results_for_java_call
4973 result_reg->set_req(_slow_path, control());
4974 result_val->set_req(_slow_path, slow_result);
4975 result_io ->set_req(_slow_path, i_o());
4976 result_mem->set_req(_slow_path, reset_memory());
4977 }
4978
4979 set_control(normal_ctl);
4980 if (!stopped()) {
4981 // Normal case: The array type has been cached in the java.lang.Class.
4982 // The following call works fine even if the array type is polymorphic.
4983 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4984
4985 klass_node = load_default_refined_array_klass(klass_node);
4986
4987 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4988 result_reg->init_req(_normal_path, control());
4989 result_val->init_req(_normal_path, obj);
4990 result_io ->init_req(_normal_path, i_o());
4991 result_mem->init_req(_normal_path, reset_memory());
4992
4993 if (uninitialized) {
4994 // Mark the allocation so that zeroing is skipped
4995 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4996 alloc->maybe_set_complete(&_gvn);
4997 }
4998 }
4999
5000 // Return the combined state.
5001 set_i_o( _gvn.transform(result_io) );
5002 set_all_memory( _gvn.transform(result_mem));
5003
5004 C->set_has_split_ifs(true); // Has chance for split-if optimization
5005 set_result(result_reg, result_val);
5006 return true;
5007 }
5008
5009 //----------------------inline_native_getLength--------------------------
5010 // public static native int java.lang.reflect.Array.getLength(Object array);
5011 bool LibraryCallKit::inline_native_getLength() {
5012 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
5013
5014 Node* array = null_check(argument(0));
5015 // If array is dead, only null-path is taken.
5016 if (stopped()) return true;
5017
5018 // Deoptimize if it is a non-array.
5019 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
5020
5021 if (non_array != nullptr) {
5022 PreserveJVMState pjvms(this);
5023 set_control(non_array);
5024 uncommon_trap(Deoptimization::Reason_intrinsic,
5025 Deoptimization::Action_maybe_recompile);
5026 }
5027
5028 // If control is dead, only non-array-path is taken.
5029 if (stopped()) return true;
5030
5031 // The works fine even if the array type is polymorphic.
5032 // It could be a dynamic mix of int[], boolean[], Object[], etc.
5033 Node* result = load_array_length(array);
5034
5035 C->set_has_split_ifs(true); // Has chance for split-if optimization
5036 set_result(result);
5037 return true;
5038 }
5039
5040 //------------------------inline_array_copyOf----------------------------
5041 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
5042 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
5043 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
5044 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
5045
5046 // Get the arguments.
5047 Node* original = argument(0);
5048 Node* start = is_copyOfRange? argument(1): intcon(0);
5049 Node* end = is_copyOfRange? argument(2): argument(1);
5050 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
5051
5052 Node* newcopy = nullptr;
5053
5054 // Set the original stack and the reexecute bit for the interpreter to reexecute
5055 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
5056 { PreserveReexecuteState preexecs(this);
5057 jvms()->set_should_reexecute(true);
5058
5059 array_type_mirror = null_check(array_type_mirror);
5060 original = null_check(original);
5061
5062 // Check if a null path was taken unconditionally.
5063 if (stopped()) return true;
5064
5065 Node* orig_length = load_array_length(original);
5066
5067 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
5068 klass_node = null_check(klass_node);
5069
5070 RegionNode* bailout = new RegionNode(1);
5071 record_for_igvn(bailout);
5072
5073 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
5074 // Bail out if that is so.
5075 // Inline type array may have object field that would require a
5076 // write barrier. Conservatively, go to slow path.
5077 // TODO 8251971: Optimize for the case when flat src/dst are later found
5078 // to not contain oops (i.e., move this check to the macro expansion phase).
5079 // TODO 8382226: Revisit for flat abstract value class arrays
5080 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5081 const TypeAryPtr* orig_t = _gvn.type(original)->isa_aryptr();
5082 const TypeKlassPtr* tklass = _gvn.type(klass_node)->is_klassptr();
5083 bool exclude_flat = UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing) &&
5084 // Can src array be flat and contain oops?
5085 (orig_t == nullptr || (!orig_t->is_not_flat() && (!orig_t->is_flat() || orig_t->elem()->inline_klass()->contains_oops()))) &&
5086 // Can dest array be flat and contain oops?
5087 tklass->can_be_inline_array() && (!tklass->is_flat() || tklass->is_aryklassptr()->elem()->is_instklassptr()->instance_klass()->as_inline_klass()->contains_oops());
5088 Node* not_objArray = exclude_flat ? generate_non_refArray_guard(klass_node, bailout) : generate_typeArray_guard(klass_node, bailout);
5089
5090 Node* refined_klass_node = load_default_refined_array_klass(klass_node, /* type_array_guard= */ false);
5091
5092 if (not_objArray != nullptr) {
5093 // Improve the klass node's type from the new optimistic assumption:
5094 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
5095 bool not_flat = !UseArrayFlattening;
5096 bool not_null_free = !Arguments::is_valhalla_enabled();
5097 const Type* akls = TypeAryKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0), Type::trust_interfaces, not_flat, not_null_free, false, false, not_flat, true);
5098 Node* cast = new CastPPNode(control(), refined_klass_node, akls);
5099 refined_klass_node = _gvn.transform(cast);
5100 }
5101
5102 // Bail out if either start or end is negative.
5103 generate_negative_guard(start, bailout, &start);
5104 generate_negative_guard(end, bailout, &end);
5105
5106 Node* length = end;
5107 if (_gvn.type(start) != TypeInt::ZERO) {
5108 length = _gvn.transform(new SubINode(end, start));
5109 }
5110
5111 // Bail out if length is negative (i.e., if start > end).
5112 // Without this the new_array would throw
5113 // NegativeArraySizeException but IllegalArgumentException is what
5114 // should be thrown
5115 generate_negative_guard(length, bailout, &length);
5116
5117 // Handle inline type arrays
5118 // TODO 8251971 This is too strong
5119 generate_fair_guard(flat_array_test(load_object_klass(original)), bailout);
5120 generate_fair_guard(flat_array_test(refined_klass_node), bailout);
5121 generate_fair_guard(null_free_array_test(original), bailout);
5122
5123 // Bail out if start is larger than the original length
5124 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
5125 generate_negative_guard(orig_tail, bailout, &orig_tail);
5126
5127 if (bailout->req() > 1) {
5128 PreserveJVMState pjvms(this);
5129 set_control(_gvn.transform(bailout));
5130 uncommon_trap(Deoptimization::Reason_intrinsic,
5131 Deoptimization::Action_maybe_recompile);
5132 }
5133
5134 if (!stopped()) {
5135 // How many elements will we copy from the original?
5136 // The answer is MinI(orig_tail, length).
5137 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
5138
5139 // Generate a direct call to the right arraycopy function(s).
5140 // We know the copy is disjoint but we might not know if the
5141 // oop stores need checking.
5142 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
5143 // This will fail a store-check if x contains any non-nulls.
5144
5145 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
5146 // loads/stores but it is legal only if we're sure the
5147 // Arrays.copyOf would succeed. So we need all input arguments
5148 // to the copyOf to be validated, including that the copy to the
5149 // new array won't trigger an ArrayStoreException. That subtype
5150 // check can be optimized if we know something on the type of
5151 // the input array from type speculation.
5152 if (_gvn.type(klass_node)->singleton()) {
5153 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
5154 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
5155
5156 int test = C->static_subtype_check(superk, subk);
5157 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
5158 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
5159 if (t_original->speculative_type() != nullptr) {
5160 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
5161 }
5162 }
5163 }
5164
5165 bool validated = false;
5166 // Reason_class_check rather than Reason_intrinsic because we
5167 // want to intrinsify even if this traps.
5168 if (!too_many_traps(Deoptimization::Reason_class_check)) {
5169 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
5170
5171 if (not_subtype_ctrl != top()) {
5172 PreserveJVMState pjvms(this);
5173 set_control(not_subtype_ctrl);
5174 uncommon_trap(Deoptimization::Reason_class_check,
5175 Deoptimization::Action_make_not_entrant);
5176 assert(stopped(), "Should be stopped");
5177 }
5178 validated = true;
5179 }
5180
5181 if (!stopped()) {
5182 newcopy = new_array(refined_klass_node, length, 0); // no arguments to push
5183
5184 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
5185 load_object_klass(original), klass_node);
5186 if (!is_copyOfRange) {
5187 ac->set_copyof(validated);
5188 } else {
5189 ac->set_copyofrange(validated);
5190 }
5191 Node* n = _gvn.transform(ac);
5192 if (n == ac) {
5193 ac->connect_outputs(this);
5194 } else {
5195 assert(validated, "shouldn't transform if all arguments not validated");
5196 set_all_memory(n);
5197 }
5198 }
5199 }
5200 } // original reexecute is set back here
5201
5202 C->set_has_split_ifs(true); // Has chance for split-if optimization
5203 if (!stopped()) {
5204 set_result(newcopy);
5205 }
5206 return true;
5207 }
5208
5209
5210 //----------------------generate_virtual_guard---------------------------
5211 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
5212 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
5213 RegionNode* slow_region) {
5214 ciMethod* method = callee();
5215 int vtable_index = method->vtable_index();
5216 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5217 "bad index %d", vtable_index);
5218 // Get the Method* out of the appropriate vtable entry.
5219 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
5220 vtable_index*vtableEntry::size_in_bytes() +
5221 in_bytes(vtableEntry::method_offset());
5222 Node* entry_addr = off_heap_plus_addr(obj_klass, entry_offset);
5223 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
5224
5225 // Compare the target method with the expected method (e.g., Object.hashCode).
5226 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
5227
5228 Node* native_call = makecon(native_call_addr);
5229 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
5230 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
5231
5232 return generate_slow_guard(test_native, slow_region);
5233 }
5234
5235 //-----------------------generate_method_call----------------------------
5236 // Use generate_method_call to make a slow-call to the real
5237 // method if the fast path fails. An alternative would be to
5238 // use a stub like OptoRuntime::slow_arraycopy_Java.
5239 // This only works for expanding the current library call,
5240 // not another intrinsic. (E.g., don't use this for making an
5241 // arraycopy call inside of the copyOf intrinsic.)
5242 CallJavaNode*
5243 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
5244 // When compiling the intrinsic method itself, do not use this technique.
5245 guarantee(callee() != C->method(), "cannot make slow-call to self");
5246
5247 ciMethod* method = callee();
5248 // ensure the JVMS we have will be correct for this call
5249 guarantee(method_id == method->intrinsic_id(), "must match");
5250
5251 const TypeFunc* tf = TypeFunc::make(method);
5252 if (res_not_null) {
5253 assert(tf->return_type() == T_OBJECT, "");
5254 const TypeTuple* range = tf->range_cc();
5255 const Type** fields = TypeTuple::fields(range->cnt());
5256 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
5257 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
5258 tf = TypeFunc::make(tf->domain_cc(), new_range);
5259 }
5260 CallJavaNode* slow_call;
5261 if (is_static) {
5262 assert(!is_virtual, "");
5263 slow_call = new CallStaticJavaNode(C, tf,
5264 SharedRuntime::get_resolve_static_call_stub(), method);
5265 } else if (is_virtual) {
5266 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5267 int vtable_index = Method::invalid_vtable_index;
5268 if (UseInlineCaches) {
5269 // Suppress the vtable call
5270 } else {
5271 // hashCode and clone are not a miranda methods,
5272 // so the vtable index is fixed.
5273 // No need to use the linkResolver to get it.
5274 vtable_index = method->vtable_index();
5275 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5276 "bad index %d", vtable_index);
5277 }
5278 slow_call = new CallDynamicJavaNode(tf,
5279 SharedRuntime::get_resolve_virtual_call_stub(),
5280 method, vtable_index);
5281 } else { // neither virtual nor static: opt_virtual
5282 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5283 slow_call = new CallStaticJavaNode(C, tf,
5284 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
5285 slow_call->set_optimized_virtual(true);
5286 }
5287 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
5288 // To be able to issue a direct call (optimized virtual or virtual)
5289 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
5290 // about the method being invoked should be attached to the call site to
5291 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
5292 slow_call->set_override_symbolic_info(true);
5293 }
5294 set_arguments_for_java_call(slow_call);
5295 set_edges_for_java_call(slow_call);
5296 return slow_call;
5297 }
5298
5299
5300 /**
5301 * Build special case code for calls to hashCode on an object. This call may
5302 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
5303 * slightly different code.
5304 */
5305 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
5306 assert(is_static == callee()->is_static(), "correct intrinsic selection");
5307 assert(!(is_virtual && is_static), "either virtual, special, or static");
5308
5309 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
5310
5311 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5312 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
5313 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
5314 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5315 Node* obj = argument(0);
5316
5317 // Don't intrinsify hashcode on inline types for now.
5318 // The "is locked" runtime check also subsumes the inline type check (as inline types cannot be locked) and goes to the slow path.
5319 if (gvn().type(obj)->is_inlinetypeptr()) {
5320 return false;
5321 }
5322
5323 if (!is_static) {
5324 // Check for hashing null object
5325 obj = null_check_receiver();
5326 if (stopped()) return true; // unconditionally null
5327 result_reg->init_req(_null_path, top());
5328 result_val->init_req(_null_path, top());
5329 } else {
5330 // Do a null check, and return zero if null.
5331 // System.identityHashCode(null) == 0
5332 Node* null_ctl = top();
5333 obj = null_check_oop(obj, &null_ctl);
5334 result_reg->init_req(_null_path, null_ctl);
5335 result_val->init_req(_null_path, _gvn.intcon(0));
5336 }
5337
5338 // Unconditionally null? Then return right away.
5339 if (stopped()) {
5340 set_control( result_reg->in(_null_path));
5341 if (!stopped())
5342 set_result(result_val->in(_null_path));
5343 return true;
5344 }
5345
5346 // We only go to the fast case code if we pass a number of guards. The
5347 // paths which do not pass are accumulated in the slow_region.
5348 RegionNode* slow_region = new RegionNode(1);
5349 record_for_igvn(slow_region);
5350
5351 // If this is a virtual call, we generate a funny guard. We pull out
5352 // the vtable entry corresponding to hashCode() from the target object.
5353 // If the target method which we are calling happens to be the native
5354 // Object hashCode() method, we pass the guard. We do not need this
5355 // guard for non-virtual calls -- the caller is known to be the native
5356 // Object hashCode().
5357 if (is_virtual) {
5358 // After null check, get the object's klass.
5359 Node* obj_klass = load_object_klass(obj);
5360 generate_virtual_guard(obj_klass, slow_region);
5361 }
5362
5363 // Get the header out of the object, use LoadMarkNode when available
5364 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
5365 // The control of the load must be null. Otherwise, the load can move before
5366 // the null check after castPP removal.
5367 Node* no_ctrl = nullptr;
5368 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
5369
5370 if (!UseObjectMonitorTable) {
5371 // Test the header to see if it is safe to read w.r.t. locking.
5372 // We cannot use the inline type mask as this may check bits that are overriden
5373 // by an object monitor's pointer when inflating locking.
5374 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
5375 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
5376 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
5377 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
5378 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
5379
5380 generate_slow_guard(test_monitor, slow_region);
5381 }
5382
5383 // Get the hash value and check to see that it has been properly assigned.
5384 // We depend on hash_mask being at most 32 bits and avoid the use of
5385 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
5386 // vm: see markWord.hpp.
5387 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
5388 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
5389 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
5390 // This hack lets the hash bits live anywhere in the mark object now, as long
5391 // as the shift drops the relevant bits into the low 32 bits. Note that
5392 // Java spec says that HashCode is an int so there's no point in capturing
5393 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
5394 hshifted_header = ConvX2I(hshifted_header);
5395 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
5396
5397 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
5398 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
5399 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
5400
5401 generate_slow_guard(test_assigned, slow_region);
5402
5403 Node* init_mem = reset_memory();
5404 // fill in the rest of the null path:
5405 result_io ->init_req(_null_path, i_o());
5406 result_mem->init_req(_null_path, init_mem);
5407
5408 result_val->init_req(_fast_path, hash_val);
5409 result_reg->init_req(_fast_path, control());
5410 result_io ->init_req(_fast_path, i_o());
5411 result_mem->init_req(_fast_path, init_mem);
5412
5413 // Generate code for the slow case. We make a call to hashCode().
5414 set_control(_gvn.transform(slow_region));
5415 if (!stopped()) {
5416 // No need for PreserveJVMState, because we're using up the present state.
5417 set_all_memory(init_mem);
5418 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
5419 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
5420 Node* slow_result = set_results_for_java_call(slow_call);
5421 // this->control() comes from set_results_for_java_call
5422 result_reg->init_req(_slow_path, control());
5423 result_val->init_req(_slow_path, slow_result);
5424 result_io ->set_req(_slow_path, i_o());
5425 result_mem ->set_req(_slow_path, reset_memory());
5426 }
5427
5428 // Return the combined state.
5429 set_i_o( _gvn.transform(result_io) );
5430 set_all_memory( _gvn.transform(result_mem));
5431
5432 set_result(result_reg, result_val);
5433 return true;
5434 }
5435
5436 //---------------------------inline_native_getClass----------------------------
5437 // public final native Class<?> java.lang.Object.getClass();
5438 //
5439 // Build special case code for calls to getClass on an object.
5440 bool LibraryCallKit::inline_native_getClass() {
5441 Node* obj = argument(0);
5442 if (obj->is_InlineType()) {
5443 const Type* t = _gvn.type(obj);
5444 if (t->maybe_null()) {
5445 null_check(obj);
5446 }
5447 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror())));
5448 return true;
5449 }
5450 obj = null_check_receiver();
5451 if (stopped()) return true;
5452 set_result(load_mirror_from_klass(load_object_klass(obj)));
5453 return true;
5454 }
5455
5456 //-----------------inline_native_Reflection_getCallerClass---------------------
5457 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
5458 //
5459 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
5460 //
5461 // NOTE: This code must perform the same logic as JVM_GetCallerClass
5462 // in that it must skip particular security frames and checks for
5463 // caller sensitive methods.
5464 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
5465 #ifndef PRODUCT
5466 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5467 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
5468 }
5469 #endif
5470
5471 if (!jvms()->has_method()) {
5472 #ifndef PRODUCT
5473 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5474 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
5475 }
5476 #endif
5477 return false;
5478 }
5479
5480 // Walk back up the JVM state to find the caller at the required
5481 // depth.
5482 JVMState* caller_jvms = jvms();
5483
5484 // Cf. JVM_GetCallerClass
5485 // NOTE: Start the loop at depth 1 because the current JVM state does
5486 // not include the Reflection.getCallerClass() frame.
5487 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
5488 ciMethod* m = caller_jvms->method();
5489 switch (n) {
5490 case 0:
5491 fatal("current JVM state does not include the Reflection.getCallerClass frame");
5492 break;
5493 case 1:
5494 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
5495 if (!m->caller_sensitive()) {
5496 #ifndef PRODUCT
5497 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5498 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
5499 }
5500 #endif
5501 return false; // bail-out; let JVM_GetCallerClass do the work
5502 }
5503 break;
5504 default:
5505 if (!m->is_ignored_by_security_stack_walk()) {
5506 // We have reached the desired frame; return the holder class.
5507 // Acquire method holder as java.lang.Class and push as constant.
5508 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
5509 ciInstance* caller_mirror = caller_klass->java_mirror();
5510 set_result(makecon(TypeInstPtr::make(caller_mirror)));
5511
5512 #ifndef PRODUCT
5513 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5514 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());
5515 tty->print_cr(" JVM state at this point:");
5516 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5517 ciMethod* m = jvms()->of_depth(i)->method();
5518 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5519 }
5520 }
5521 #endif
5522 return true;
5523 }
5524 break;
5525 }
5526 }
5527
5528 #ifndef PRODUCT
5529 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5530 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
5531 tty->print_cr(" JVM state at this point:");
5532 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5533 ciMethod* m = jvms()->of_depth(i)->method();
5534 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5535 }
5536 }
5537 #endif
5538
5539 return false; // bail-out; let JVM_GetCallerClass do the work
5540 }
5541
5542 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
5543 Node* arg = argument(0);
5544 Node* result = nullptr;
5545
5546 switch (id) {
5547 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
5548 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
5549 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5550 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5551 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5552 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5553
5554 case vmIntrinsics::_doubleToLongBits: {
5555 // two paths (plus control) merge in a wood
5556 RegionNode *r = new RegionNode(3);
5557 Node *phi = new PhiNode(r, TypeLong::LONG);
5558
5559 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5560 // Build the boolean node
5561 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5562
5563 // Branch either way.
5564 // NaN case is less traveled, which makes all the difference.
5565 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5566 Node *opt_isnan = _gvn.transform(ifisnan);
5567 assert( opt_isnan->is_If(), "Expect an IfNode");
5568 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5569 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5570
5571 set_control(iftrue);
5572
5573 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5574 Node *slow_result = longcon(nan_bits); // return NaN
5575 phi->init_req(1, _gvn.transform( slow_result ));
5576 r->init_req(1, iftrue);
5577
5578 // Else fall through
5579 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5580 set_control(iffalse);
5581
5582 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5583 r->init_req(2, iffalse);
5584
5585 // Post merge
5586 set_control(_gvn.transform(r));
5587 record_for_igvn(r);
5588
5589 C->set_has_split_ifs(true); // Has chance for split-if optimization
5590 result = phi;
5591 assert(result->bottom_type()->isa_long(), "must be");
5592 break;
5593 }
5594
5595 case vmIntrinsics::_floatToIntBits: {
5596 // two paths (plus control) merge in a wood
5597 RegionNode *r = new RegionNode(3);
5598 Node *phi = new PhiNode(r, TypeInt::INT);
5599
5600 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5601 // Build the boolean node
5602 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5603
5604 // Branch either way.
5605 // NaN case is less traveled, which makes all the difference.
5606 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5607 Node *opt_isnan = _gvn.transform(ifisnan);
5608 assert( opt_isnan->is_If(), "Expect an IfNode");
5609 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5610 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5611
5612 set_control(iftrue);
5613
5614 static const jint nan_bits = 0x7fc00000;
5615 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5616 phi->init_req(1, _gvn.transform( slow_result ));
5617 r->init_req(1, iftrue);
5618
5619 // Else fall through
5620 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5621 set_control(iffalse);
5622
5623 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5624 r->init_req(2, iffalse);
5625
5626 // Post merge
5627 set_control(_gvn.transform(r));
5628 record_for_igvn(r);
5629
5630 C->set_has_split_ifs(true); // Has chance for split-if optimization
5631 result = phi;
5632 assert(result->bottom_type()->isa_int(), "must be");
5633 break;
5634 }
5635
5636 default:
5637 fatal_unexpected_iid(id);
5638 break;
5639 }
5640 set_result(_gvn.transform(result));
5641 return true;
5642 }
5643
5644 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5645 Node* arg = argument(0);
5646 Node* result = nullptr;
5647
5648 switch (id) {
5649 case vmIntrinsics::_floatIsInfinite:
5650 result = new IsInfiniteFNode(arg);
5651 break;
5652 case vmIntrinsics::_floatIsFinite:
5653 result = new IsFiniteFNode(arg);
5654 break;
5655 case vmIntrinsics::_doubleIsInfinite:
5656 result = new IsInfiniteDNode(arg);
5657 break;
5658 case vmIntrinsics::_doubleIsFinite:
5659 result = new IsFiniteDNode(arg);
5660 break;
5661 default:
5662 fatal_unexpected_iid(id);
5663 break;
5664 }
5665 set_result(_gvn.transform(result));
5666 return true;
5667 }
5668
5669 //----------------------inline_unsafe_copyMemory-------------------------
5670 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5671
5672 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5673 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5674 const Type* base_t = gvn.type(base);
5675
5676 bool in_native = (base_t == TypePtr::NULL_PTR);
5677 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5678 bool is_mixed = !in_heap && !in_native;
5679
5680 if (is_mixed) {
5681 return true; // mixed accesses can touch both on-heap and off-heap memory
5682 }
5683 if (in_heap) {
5684 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5685 if (!is_prim_array) {
5686 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5687 // there's not enough type information available to determine proper memory slice for it.
5688 return true;
5689 }
5690 }
5691 return false;
5692 }
5693
5694 bool LibraryCallKit::inline_unsafe_copyMemory() {
5695 if (callee()->is_static()) return false; // caller must have the capability!
5696 null_check_receiver(); // null-check receiver
5697 if (stopped()) return true;
5698
5699 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5700
5701 Node* src_base = argument(1); // type: oop
5702 Node* src_off = ConvL2X(argument(2)); // type: long
5703 Node* dst_base = argument(4); // type: oop
5704 Node* dst_off = ConvL2X(argument(5)); // type: long
5705 Node* size = ConvL2X(argument(7)); // type: long
5706
5707 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5708 "fieldOffset must be byte-scaled");
5709
5710 Node* src_addr = make_unsafe_address(src_base, src_off);
5711 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5712
5713 Node* thread = _gvn.transform(new ThreadLocalNode());
5714 Node* doing_unsafe_access_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5715 BasicType doing_unsafe_access_bt = T_BYTE;
5716 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5717
5718 // update volatile field
5719 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5720
5721 int flags = RC_LEAF | RC_NO_FP;
5722
5723 const TypePtr* dst_type = TypePtr::BOTTOM;
5724
5725 // Adjust memory effects of the runtime call based on input values.
5726 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5727 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5728 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5729
5730 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5731 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5732 flags |= RC_NARROW_MEM; // narrow in memory
5733 }
5734 }
5735
5736 // Call it. Note that the length argument is not scaled.
5737 make_runtime_call(flags,
5738 OptoRuntime::fast_arraycopy_Type(),
5739 StubRoutines::unsafe_arraycopy(),
5740 "unsafe_arraycopy",
5741 dst_type,
5742 src_addr, dst_addr, size XTOP);
5743
5744 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5745
5746 return true;
5747 }
5748
5749 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5750 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5751 bool LibraryCallKit::inline_unsafe_setMemory() {
5752 if (callee()->is_static()) return false; // caller must have the capability!
5753 null_check_receiver(); // null-check receiver
5754 if (stopped()) return true;
5755
5756 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5757
5758 Node* dst_base = argument(1); // type: oop
5759 Node* dst_off = ConvL2X(argument(2)); // type: long
5760 Node* size = ConvL2X(argument(4)); // type: long
5761 Node* byte = argument(6); // type: byte
5762
5763 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5764 "fieldOffset must be byte-scaled");
5765
5766 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5767
5768 Node* thread = _gvn.transform(new ThreadLocalNode());
5769 Node* doing_unsafe_access_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5770 BasicType doing_unsafe_access_bt = T_BYTE;
5771 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5772
5773 // update volatile field
5774 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5775
5776 int flags = RC_LEAF | RC_NO_FP;
5777
5778 const TypePtr* dst_type = TypePtr::BOTTOM;
5779
5780 // Adjust memory effects of the runtime call based on input values.
5781 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5782 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5783
5784 flags |= RC_NARROW_MEM; // narrow in memory
5785 }
5786
5787 // Call it. Note that the length argument is not scaled.
5788 make_runtime_call(flags,
5789 OptoRuntime::unsafe_setmemory_Type(),
5790 StubRoutines::unsafe_setmemory(),
5791 "unsafe_setmemory",
5792 dst_type,
5793 dst_addr, size XTOP, byte);
5794
5795 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5796
5797 return true;
5798 }
5799
5800 #undef XTOP
5801
5802 //------------------------clone_coping-----------------------------------
5803 // Helper function for inline_native_clone.
5804 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5805 assert(obj_size != nullptr, "");
5806 Node* raw_obj = alloc_obj->in(1);
5807 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5808
5809 AllocateNode* alloc = nullptr;
5810 if (ReduceBulkZeroing &&
5811 // If we are implementing an array clone without knowing its source type
5812 // (can happen when compiling the array-guarded branch of a reflective
5813 // Object.clone() invocation), initialize the array within the allocation.
5814 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5815 // to a runtime clone call that assumes fully initialized source arrays.
5816 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5817 // We will be completely responsible for initializing this object -
5818 // mark Initialize node as complete.
5819 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5820 // The object was just allocated - there should be no any stores!
5821 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5822 // Mark as complete_with_arraycopy so that on AllocateNode
5823 // expansion, we know this AllocateNode is initialized by an array
5824 // copy and a StoreStore barrier exists after the array copy.
5825 alloc->initialization()->set_complete_with_arraycopy();
5826 }
5827
5828 Node* size = _gvn.transform(obj_size);
5829 access_clone(obj, alloc_obj, size, is_array);
5830
5831 // Do not let reads from the cloned object float above the arraycopy.
5832 if (alloc != nullptr) {
5833 // Do not let stores that initialize this object be reordered with
5834 // a subsequent store that would make this object accessible by
5835 // other threads.
5836 // Record what AllocateNode this StoreStore protects so that
5837 // escape analysis can go from the MemBarStoreStoreNode to the
5838 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5839 // based on the escape status of the AllocateNode.
5840 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5841 } else {
5842 insert_mem_bar(Op_MemBarCPUOrder);
5843 }
5844 }
5845
5846 //------------------------inline_native_clone----------------------------
5847 // protected native Object java.lang.Object.clone();
5848 //
5849 // Here are the simple edge cases:
5850 // null receiver => normal trap
5851 // virtual and clone was overridden => slow path to out-of-line clone
5852 // not cloneable or finalizer => slow path to out-of-line Object.clone
5853 //
5854 // The general case has two steps, allocation and copying.
5855 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5856 //
5857 // Copying also has two cases, oop arrays and everything else.
5858 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5859 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5860 //
5861 // These steps fold up nicely if and when the cloned object's klass
5862 // can be sharply typed as an object array, a type array, or an instance.
5863 //
5864 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5865 PhiNode* result_val;
5866
5867 // Set the reexecute bit for the interpreter to reexecute
5868 // the bytecode that invokes Object.clone if deoptimization happens.
5869 { PreserveReexecuteState preexecs(this);
5870 jvms()->set_should_reexecute(true);
5871
5872 Node* obj = argument(0);
5873 obj = null_check_receiver();
5874 if (stopped()) return true;
5875
5876 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5877 if (obj_type->is_inlinetypeptr()) {
5878 // If the object to clone is an inline type, we can simply return it (i.e. a nop) since inline types have
5879 // no identity.
5880 set_result(obj);
5881 return true;
5882 }
5883
5884 // If we are going to clone an instance, we need its exact type to
5885 // know the number and types of fields to convert the clone to
5886 // loads/stores. Maybe a speculative type can help us.
5887 if (!obj_type->klass_is_exact() &&
5888 obj_type->speculative_type() != nullptr &&
5889 obj_type->speculative_type()->is_instance_klass() &&
5890 !obj_type->speculative_type()->is_inlinetype()) {
5891 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5892 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5893 !spec_ik->has_injected_fields()) {
5894 if (!obj_type->isa_instptr() ||
5895 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5896 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5897 }
5898 }
5899 }
5900
5901 // Conservatively insert a memory barrier on all memory slices.
5902 // Do not let writes into the original float below the clone.
5903 insert_mem_bar(Op_MemBarCPUOrder);
5904
5905 // paths into result_reg:
5906 enum {
5907 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5908 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5909 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5910 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5911 PATH_LIMIT
5912 };
5913 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5914 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5915 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5916 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5917 record_for_igvn(result_reg);
5918
5919 Node* obj_klass = load_object_klass(obj);
5920 // We only go to the fast case code if we pass a number of guards.
5921 // The paths which do not pass are accumulated in the slow_region.
5922 RegionNode* slow_region = new RegionNode(1);
5923 record_for_igvn(slow_region);
5924
5925 Node* array_obj = obj;
5926 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5927 if (array_ctl != nullptr) {
5928 // It's an array.
5929 PreserveJVMState pjvms(this);
5930 set_control(array_ctl);
5931
5932 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5933 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr();
5934 if (UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) &&
5935 obj_type->can_be_inline_array() &&
5936 (ary_ptr == nullptr || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) {
5937 // Flat inline type array may have object field that would require a
5938 // write barrier. Conservatively, go to slow path.
5939 generate_fair_guard(flat_array_test(obj_klass), slow_region);
5940 }
5941
5942 if (!stopped()) {
5943 Node* obj_length = load_array_length(array_obj);
5944 Node* array_size = nullptr; // Size of the array without object alignment padding.
5945 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5946
5947 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5948 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5949 // If it is an oop array, it requires very special treatment,
5950 // because gc barriers are required when accessing the array.
5951 Node* is_obja = generate_refArray_guard(obj_klass, (RegionNode*)nullptr);
5952 if (is_obja != nullptr) {
5953 PreserveJVMState pjvms2(this);
5954 set_control(is_obja);
5955 // Generate a direct call to the right arraycopy function(s).
5956 // Clones are always tightly coupled.
5957 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5958 ac->set_clone_oop_array();
5959 Node* n = _gvn.transform(ac);
5960 assert(n == ac, "cannot disappear");
5961 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5962
5963 result_reg->init_req(_objArray_path, control());
5964 result_val->init_req(_objArray_path, alloc_obj);
5965 result_i_o ->set_req(_objArray_path, i_o());
5966 result_mem ->set_req(_objArray_path, reset_memory());
5967 }
5968 }
5969 // Otherwise, there are no barriers to worry about.
5970 // (We can dispense with card marks if we know the allocation
5971 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5972 // causes the non-eden paths to take compensating steps to
5973 // simulate a fresh allocation, so that no further
5974 // card marks are required in compiled code to initialize
5975 // the object.)
5976
5977 if (!stopped()) {
5978 copy_to_clone(obj, alloc_obj, array_size, true);
5979
5980 // Present the results of the copy.
5981 result_reg->init_req(_array_path, control());
5982 result_val->init_req(_array_path, alloc_obj);
5983 result_i_o ->set_req(_array_path, i_o());
5984 result_mem ->set_req(_array_path, reset_memory());
5985 }
5986 }
5987 }
5988
5989 if (!stopped()) {
5990 // It's an instance (we did array above). Make the slow-path tests.
5991 // If this is a virtual call, we generate a funny guard. We grab
5992 // the vtable entry corresponding to clone() from the target object.
5993 // If the target method which we are calling happens to be the
5994 // Object clone() method, we pass the guard. We do not need this
5995 // guard for non-virtual calls; the caller is known to be the native
5996 // Object clone().
5997 if (is_virtual) {
5998 generate_virtual_guard(obj_klass, slow_region);
5999 }
6000
6001 // The object must be easily cloneable and must not have a finalizer.
6002 // Both of these conditions may be checked in a single test.
6003 // We could optimize the test further, but we don't care.
6004 generate_misc_flags_guard(obj_klass,
6005 // Test both conditions:
6006 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
6007 // Must be cloneable but not finalizer:
6008 KlassFlags::_misc_is_cloneable_fast,
6009 slow_region);
6010 }
6011
6012 if (!stopped()) {
6013 // It's an instance, and it passed the slow-path tests.
6014 PreserveJVMState pjvms(this);
6015 Node* obj_size = nullptr; // Total object size, including object alignment padding.
6016 // Need to deoptimize on exception from allocation since Object.clone intrinsic
6017 // is reexecuted if deoptimization occurs and there could be problems when merging
6018 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
6019 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
6020
6021 copy_to_clone(obj, alloc_obj, obj_size, false);
6022
6023 // Present the results of the slow call.
6024 result_reg->init_req(_instance_path, control());
6025 result_val->init_req(_instance_path, alloc_obj);
6026 result_i_o ->set_req(_instance_path, i_o());
6027 result_mem ->set_req(_instance_path, reset_memory());
6028 }
6029
6030 // Generate code for the slow case. We make a call to clone().
6031 set_control(_gvn.transform(slow_region));
6032 if (!stopped()) {
6033 PreserveJVMState pjvms(this);
6034 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
6035 // We need to deoptimize on exception (see comment above)
6036 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
6037 // this->control() comes from set_results_for_java_call
6038 result_reg->init_req(_slow_path, control());
6039 result_val->init_req(_slow_path, slow_result);
6040 result_i_o ->set_req(_slow_path, i_o());
6041 result_mem ->set_req(_slow_path, reset_memory());
6042 }
6043
6044 // Return the combined state.
6045 set_control( _gvn.transform(result_reg));
6046 set_i_o( _gvn.transform(result_i_o));
6047 set_all_memory( _gvn.transform(result_mem));
6048 } // original reexecute is set back here
6049
6050 set_result(_gvn.transform(result_val));
6051 return true;
6052 }
6053
6054 // If we have a tightly coupled allocation, the arraycopy may take care
6055 // of the array initialization. If one of the guards we insert between
6056 // the allocation and the arraycopy causes a deoptimization, an
6057 // uninitialized array will escape the compiled method. To prevent that
6058 // we set the JVM state for uncommon traps between the allocation and
6059 // the arraycopy to the state before the allocation so, in case of
6060 // deoptimization, we'll reexecute the allocation and the
6061 // initialization.
6062 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
6063 if (alloc != nullptr) {
6064 ciMethod* trap_method = alloc->jvms()->method();
6065 int trap_bci = alloc->jvms()->bci();
6066
6067 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6068 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
6069 // Make sure there's no store between the allocation and the
6070 // arraycopy otherwise visible side effects could be rexecuted
6071 // in case of deoptimization and cause incorrect execution.
6072 bool no_interfering_store = true;
6073 Node* mem = alloc->in(TypeFunc::Memory);
6074 if (mem->is_MergeMem()) {
6075 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
6076 Node* n = mms.memory();
6077 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
6078 assert(n->is_Store(), "what else?");
6079 no_interfering_store = false;
6080 break;
6081 }
6082 }
6083 } else {
6084 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
6085 Node* n = mms.memory();
6086 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
6087 assert(n->is_Store(), "what else?");
6088 no_interfering_store = false;
6089 break;
6090 }
6091 }
6092 }
6093
6094 if (no_interfering_store) {
6095 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6096
6097 JVMState* saved_jvms = jvms();
6098 saved_reexecute_sp = _reexecute_sp;
6099
6100 set_jvms(sfpt->jvms());
6101 _reexecute_sp = jvms()->sp();
6102
6103 return saved_jvms;
6104 }
6105 }
6106 }
6107 return nullptr;
6108 }
6109
6110 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
6111 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
6112 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
6113 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
6114 uint size = alloc->req();
6115 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
6116 old_jvms->set_map(sfpt);
6117 for (uint i = 0; i < size; i++) {
6118 sfpt->init_req(i, alloc->in(i));
6119 }
6120 int adjustment = 1;
6121 const TypeAryKlassPtr* ary_klass_ptr = alloc->in(AllocateNode::KlassNode)->bottom_type()->is_aryklassptr();
6122 if (ary_klass_ptr->is_null_free()) {
6123 // A null-free, tightly coupled array allocation can only come from LibraryCallKit::inline_newArray which
6124 // also requires the componentType and initVal on stack for re-execution.
6125 // Re-create and push the componentType.
6126 ciArrayKlass* klass = ary_klass_ptr->exact_klass()->as_array_klass();
6127 ciInstance* instance = klass->component_mirror_instance();
6128 const TypeInstPtr* t_instance = TypeInstPtr::make(instance);
6129 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), makecon(t_instance));
6130 adjustment++;
6131 }
6132 // re-push array length for deoptimization
6133 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment - 1, alloc->in(AllocateNode::ALength));
6134 if (ary_klass_ptr->is_null_free()) {
6135 // Re-create and push the initVal.
6136 Node* init_val = alloc->in(AllocateNode::InitValue);
6137 if (init_val == nullptr) {
6138 init_val = InlineTypeNode::make_all_zero(_gvn, ary_klass_ptr->elem()->is_instklassptr()->instance_klass()->as_inline_klass());
6139 } else if (UseCompressedOops) {
6140 init_val = _gvn.transform(new DecodeNNode(init_val, init_val->bottom_type()->make_ptr()));
6141 }
6142 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment, init_val);
6143 adjustment++;
6144 }
6145 old_jvms->set_sp(old_jvms->sp() + adjustment);
6146 old_jvms->set_monoff(old_jvms->monoff() + adjustment);
6147 old_jvms->set_scloff(old_jvms->scloff() + adjustment);
6148 old_jvms->set_endoff(old_jvms->endoff() + adjustment);
6149 old_jvms->set_should_reexecute(true);
6150
6151 sfpt->set_i_o(map()->i_o());
6152 sfpt->set_memory(map()->memory());
6153 sfpt->set_control(map()->control());
6154 return sfpt;
6155 }
6156
6157 // In case of a deoptimization, we restart execution at the
6158 // allocation, allocating a new array. We would leave an uninitialized
6159 // array in the heap that GCs wouldn't expect. Move the allocation
6160 // after the traps so we don't allocate the array if we
6161 // deoptimize. This is possible because tightly_coupled_allocation()
6162 // guarantees there's no observer of the allocated array at this point
6163 // and the control flow is simple enough.
6164 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
6165 int saved_reexecute_sp, uint new_idx) {
6166 if (saved_jvms_before_guards != nullptr && !stopped()) {
6167 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
6168
6169 assert(alloc != nullptr, "only with a tightly coupled allocation");
6170 // restore JVM state to the state at the arraycopy
6171 saved_jvms_before_guards->map()->set_control(map()->control());
6172 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
6173 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
6174 // If we've improved the types of some nodes (null check) while
6175 // emitting the guards, propagate them to the current state
6176 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
6177 set_jvms(saved_jvms_before_guards);
6178 _reexecute_sp = saved_reexecute_sp;
6179
6180 // Remove the allocation from above the guards
6181 CallProjections* callprojs = alloc->extract_projections(true);
6182 InitializeNode* init = alloc->initialization();
6183 Node* alloc_mem = alloc->in(TypeFunc::Memory);
6184 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
6185 init->replace_mem_projs_by(alloc_mem, C);
6186
6187 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
6188 // the allocation (i.e. is only valid if the allocation succeeds):
6189 // 1) replace CastIINode with AllocateArrayNode's length here
6190 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
6191 //
6192 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
6193 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
6194 Node* init_control = init->proj_out(TypeFunc::Control);
6195 Node* alloc_length = alloc->Ideal_length();
6196 #ifdef ASSERT
6197 Node* prev_cast = nullptr;
6198 #endif
6199 for (uint i = 0; i < init_control->outcnt(); i++) {
6200 Node* init_out = init_control->raw_out(i);
6201 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
6202 #ifdef ASSERT
6203 if (prev_cast == nullptr) {
6204 prev_cast = init_out;
6205 } else {
6206 if (prev_cast->cmp(*init_out) == false) {
6207 prev_cast->dump();
6208 init_out->dump();
6209 assert(false, "not equal CastIINode");
6210 }
6211 }
6212 #endif
6213 C->gvn_replace_by(init_out, alloc_length);
6214 }
6215 }
6216 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
6217
6218 // move the allocation here (after the guards)
6219 _gvn.hash_delete(alloc);
6220 alloc->set_req(TypeFunc::Control, control());
6221 alloc->set_req(TypeFunc::I_O, i_o());
6222 Node *mem = reset_memory();
6223 set_all_memory(mem);
6224 alloc->set_req(TypeFunc::Memory, mem);
6225 set_control(init->proj_out_or_null(TypeFunc::Control));
6226 set_i_o(callprojs->fallthrough_ioproj);
6227
6228 // Update memory as done in GraphKit::set_output_for_allocation()
6229 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
6230 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
6231 if (ary_type->isa_aryptr() && length_type != nullptr) {
6232 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
6233 }
6234 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
6235 int elemidx = C->get_alias_index(telemref);
6236 // Need to properly move every memory projection for the Initialize
6237 #ifdef ASSERT
6238 int mark_idx = C->get_alias_index(ary_type->add_offset(oopDesc::mark_offset_in_bytes()));
6239 int klass_idx = C->get_alias_index(ary_type->add_offset(oopDesc::klass_offset_in_bytes()));
6240 #endif
6241 auto move_proj = [&](ProjNode* proj) {
6242 int alias_idx = C->get_alias_index(proj->adr_type());
6243 assert(alias_idx == Compile::AliasIdxRaw ||
6244 alias_idx == elemidx ||
6245 alias_idx == mark_idx ||
6246 alias_idx == klass_idx, "should be raw memory or array element type");
6247 set_memory(proj, alias_idx);
6248 };
6249 init->for_each_proj(move_proj, TypeFunc::Memory);
6250
6251 Node* allocx = _gvn.transform(alloc);
6252 assert(allocx == alloc, "where has the allocation gone?");
6253 assert(dest->is_CheckCastPP(), "not an allocation result?");
6254
6255 _gvn.hash_delete(dest);
6256 dest->set_req(0, control());
6257 Node* destx = _gvn.transform(dest);
6258 assert(destx == dest, "where has the allocation result gone?");
6259
6260 array_ideal_length(alloc, ary_type, true);
6261 }
6262 }
6263
6264 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
6265 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
6266 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
6267 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
6268 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
6269 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
6270 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
6271 JVMState* saved_jvms_before_guards) {
6272 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
6273 // There is at least one unrelated uncommon trap which needs to be replaced.
6274 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6275
6276 JVMState* saved_jvms = jvms();
6277 const int saved_reexecute_sp = _reexecute_sp;
6278 set_jvms(sfpt->jvms());
6279 _reexecute_sp = jvms()->sp();
6280
6281 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
6282
6283 // Restore state
6284 set_jvms(saved_jvms);
6285 _reexecute_sp = saved_reexecute_sp;
6286 }
6287 }
6288
6289 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
6290 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
6291 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
6292 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
6293 while (if_proj->is_IfProj()) {
6294 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
6295 if (uncommon_trap != nullptr) {
6296 create_new_uncommon_trap(uncommon_trap);
6297 }
6298 assert(if_proj->in(0)->is_If(), "must be If");
6299 if_proj = if_proj->in(0)->in(0);
6300 }
6301 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
6302 "must have reached control projection of init node");
6303 }
6304
6305 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
6306 const int trap_request = uncommon_trap_call->uncommon_trap_request();
6307 assert(trap_request != 0, "no valid UCT trap request");
6308 PreserveJVMState pjvms(this);
6309 set_control(uncommon_trap_call->in(0));
6310 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
6311 Deoptimization::trap_request_action(trap_request));
6312 assert(stopped(), "Should be stopped");
6313 _gvn.hash_delete(uncommon_trap_call);
6314 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
6315 }
6316
6317 // Common checks for array sorting intrinsics arguments.
6318 // Returns `true` if checks passed.
6319 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
6320 // check address of the class
6321 if (elementType == nullptr || elementType->is_top()) {
6322 return false; // dead path
6323 }
6324 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
6325 if (elem_klass == nullptr) {
6326 return false; // dead path
6327 }
6328 // java_mirror_type() returns non-null for compile-time Class constants only
6329 ciType* elem_type = elem_klass->java_mirror_type();
6330 if (elem_type == nullptr) {
6331 return false;
6332 }
6333 bt = elem_type->basic_type();
6334 // Disable the intrinsic if the CPU does not support SIMD sort
6335 if (!Matcher::supports_simd_sort(bt)) {
6336 return false;
6337 }
6338 // check address of the array
6339 if (obj == nullptr || obj->is_top()) {
6340 return false; // dead path
6341 }
6342 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
6343 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
6344 return false; // failed input validation
6345 }
6346 return true;
6347 }
6348
6349 //------------------------------inline_array_partition-----------------------
6350 bool LibraryCallKit::inline_array_partition() {
6351 address stubAddr = StubRoutines::select_array_partition_function();
6352 if (stubAddr == nullptr) {
6353 return false; // Intrinsic's stub is not implemented on this platform
6354 }
6355 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
6356
6357 // no receiver because it is a static method
6358 Node* elementType = argument(0);
6359 Node* obj = argument(1);
6360 Node* offset = argument(2); // long
6361 Node* fromIndex = argument(4);
6362 Node* toIndex = argument(5);
6363 Node* indexPivot1 = argument(6);
6364 Node* indexPivot2 = argument(7);
6365 // PartitionOperation: argument(8) is ignored
6366
6367 Node* pivotIndices = nullptr;
6368 BasicType bt = T_ILLEGAL;
6369
6370 if (!check_array_sort_arguments(elementType, obj, bt)) {
6371 return false;
6372 }
6373 null_check(obj);
6374 // If obj is dead, only null-path is taken.
6375 if (stopped()) {
6376 return true;
6377 }
6378 // Set the original stack and the reexecute bit for the interpreter to reexecute
6379 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
6380 { PreserveReexecuteState preexecs(this);
6381 jvms()->set_should_reexecute(true);
6382
6383 Node* obj_adr = make_unsafe_address(obj, offset);
6384
6385 // create the pivotIndices array of type int and size = 2
6386 Node* size = intcon(2);
6387 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
6388 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
6389 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
6390 guarantee(alloc != nullptr, "created above");
6391 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
6392
6393 // pass the basic type enum to the stub
6394 Node* elemType = intcon(bt);
6395
6396 // Call the stub
6397 const char *stubName = "array_partition_stub";
6398 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
6399 stubAddr, stubName, TypePtr::BOTTOM,
6400 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
6401 indexPivot1, indexPivot2);
6402
6403 } // original reexecute is set back here
6404
6405 if (!stopped()) {
6406 set_result(pivotIndices);
6407 }
6408
6409 return true;
6410 }
6411
6412
6413 //------------------------------inline_array_sort-----------------------
6414 bool LibraryCallKit::inline_array_sort() {
6415 address stubAddr = StubRoutines::select_arraysort_function();
6416 if (stubAddr == nullptr) {
6417 return false; // Intrinsic's stub is not implemented on this platform
6418 }
6419 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
6420
6421 // no receiver because it is a static method
6422 Node* elementType = argument(0);
6423 Node* obj = argument(1);
6424 Node* offset = argument(2); // long
6425 Node* fromIndex = argument(4);
6426 Node* toIndex = argument(5);
6427 // SortOperation: argument(6) is ignored
6428
6429 BasicType bt = T_ILLEGAL;
6430
6431 if (!check_array_sort_arguments(elementType, obj, bt)) {
6432 return false;
6433 }
6434 null_check(obj);
6435 // If obj is dead, only null-path is taken.
6436 if (stopped()) {
6437 return true;
6438 }
6439 Node* obj_adr = make_unsafe_address(obj, offset);
6440
6441 // pass the basic type enum to the stub
6442 Node* elemType = intcon(bt);
6443
6444 // Call the stub.
6445 const char *stubName = "arraysort_stub";
6446 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
6447 stubAddr, stubName, TypePtr::BOTTOM,
6448 obj_adr, elemType, fromIndex, toIndex);
6449
6450 return true;
6451 }
6452
6453
6454 //------------------------------inline_arraycopy-----------------------
6455 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
6456 // Object dest, int destPos,
6457 // int length);
6458 bool LibraryCallKit::inline_arraycopy() {
6459 // Get the arguments.
6460 Node* src = argument(0); // type: oop
6461 Node* src_offset = argument(1); // type: int
6462 Node* dest = argument(2); // type: oop
6463 Node* dest_offset = argument(3); // type: int
6464 Node* length = argument(4); // type: int
6465
6466 uint new_idx = C->unique();
6467
6468 // Check for allocation before we add nodes that would confuse
6469 // tightly_coupled_allocation()
6470 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
6471
6472 int saved_reexecute_sp = -1;
6473 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
6474 // See arraycopy_restore_alloc_state() comment
6475 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
6476 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
6477 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
6478 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
6479
6480 // The following tests must be performed
6481 // (1) src and dest are arrays.
6482 // (2) src and dest arrays must have elements of the same BasicType
6483 // (3) src and dest must not be null.
6484 // (4) src_offset must not be negative.
6485 // (5) dest_offset must not be negative.
6486 // (6) length must not be negative.
6487 // (7) src_offset + length must not exceed length of src.
6488 // (8) dest_offset + length must not exceed length of dest.
6489 // (9) each element of an oop array must be assignable
6490
6491 // (3) src and dest must not be null.
6492 // always do this here because we need the JVM state for uncommon traps
6493 Node* null_ctl = top();
6494 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
6495 assert(null_ctl->is_top(), "no null control here");
6496 dest = null_check(dest, T_ARRAY);
6497
6498 if (!can_emit_guards) {
6499 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
6500 // guards but the arraycopy node could still take advantage of a
6501 // tightly allocated allocation. tightly_coupled_allocation() is
6502 // called again to make sure it takes the null check above into
6503 // account: the null check is mandatory and if it caused an
6504 // uncommon trap to be emitted then the allocation can't be
6505 // considered tightly coupled in this context.
6506 alloc = tightly_coupled_allocation(dest);
6507 }
6508
6509 bool validated = false;
6510
6511 const Type* src_type = _gvn.type(src);
6512 const Type* dest_type = _gvn.type(dest);
6513 const TypeAryPtr* top_src = src_type->isa_aryptr();
6514 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6515
6516 // Do we have the type of src?
6517 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6518 // Do we have the type of dest?
6519 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6520 // Is the type for src from speculation?
6521 bool src_spec = false;
6522 // Is the type for dest from speculation?
6523 bool dest_spec = false;
6524
6525 if ((!has_src || !has_dest) && can_emit_guards) {
6526 // We don't have sufficient type information, let's see if
6527 // speculative types can help. We need to have types for both src
6528 // and dest so that it pays off.
6529
6530 // Do we already have or could we have type information for src
6531 bool could_have_src = has_src;
6532 // Do we already have or could we have type information for dest
6533 bool could_have_dest = has_dest;
6534
6535 ciKlass* src_k = nullptr;
6536 if (!has_src) {
6537 src_k = src_type->speculative_type_not_null();
6538 if (src_k != nullptr && src_k->is_array_klass()) {
6539 could_have_src = true;
6540 }
6541 }
6542
6543 ciKlass* dest_k = nullptr;
6544 if (!has_dest) {
6545 dest_k = dest_type->speculative_type_not_null();
6546 if (dest_k != nullptr && dest_k->is_array_klass()) {
6547 could_have_dest = true;
6548 }
6549 }
6550
6551 if (could_have_src && could_have_dest) {
6552 // This is going to pay off so emit the required guards
6553 if (!has_src) {
6554 src = maybe_cast_profiled_obj(src, src_k, true);
6555 src_type = _gvn.type(src);
6556 top_src = src_type->isa_aryptr();
6557 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6558 src_spec = true;
6559 }
6560 if (!has_dest) {
6561 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6562 dest_type = _gvn.type(dest);
6563 top_dest = dest_type->isa_aryptr();
6564 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6565 dest_spec = true;
6566 }
6567 }
6568 }
6569
6570 if (has_src && has_dest && can_emit_guards) {
6571 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
6572 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
6573 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
6574 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
6575
6576 if (src_elem == dest_elem && top_src->is_flat() == top_dest->is_flat() && src_elem == T_OBJECT) {
6577 // If both arrays are object arrays then having the exact types
6578 // for both will remove the need for a subtype check at runtime
6579 // before the call and may make it possible to pick a faster copy
6580 // routine (without a subtype check on every element)
6581 // Do we have the exact type of src?
6582 bool could_have_src = src_spec;
6583 // Do we have the exact type of dest?
6584 bool could_have_dest = dest_spec;
6585 ciKlass* src_k = nullptr;
6586 ciKlass* dest_k = nullptr;
6587 if (!src_spec) {
6588 src_k = src_type->speculative_type_not_null();
6589 if (src_k != nullptr && src_k->is_array_klass()) {
6590 could_have_src = true;
6591 }
6592 }
6593 if (!dest_spec) {
6594 dest_k = dest_type->speculative_type_not_null();
6595 if (dest_k != nullptr && dest_k->is_array_klass()) {
6596 could_have_dest = true;
6597 }
6598 }
6599 if (could_have_src && could_have_dest) {
6600 // If we can have both exact types, emit the missing guards
6601 if (could_have_src && !src_spec) {
6602 src = maybe_cast_profiled_obj(src, src_k, true);
6603 src_type = _gvn.type(src);
6604 top_src = src_type->isa_aryptr();
6605 }
6606 if (could_have_dest && !dest_spec) {
6607 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6608 dest_type = _gvn.type(dest);
6609 top_dest = dest_type->isa_aryptr();
6610 }
6611 }
6612 }
6613 }
6614
6615 ciMethod* trap_method = method();
6616 int trap_bci = bci();
6617 if (saved_jvms_before_guards != nullptr) {
6618 trap_method = alloc->jvms()->method();
6619 trap_bci = alloc->jvms()->bci();
6620 }
6621
6622 bool negative_length_guard_generated = false;
6623
6624 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6625 can_emit_guards && !src->is_top() && !dest->is_top()) {
6626 // validate arguments: enables transformation the ArrayCopyNode
6627 validated = true;
6628
6629 RegionNode* slow_region = new RegionNode(1);
6630 record_for_igvn(slow_region);
6631
6632 // (1) src and dest are arrays.
6633 generate_non_array_guard(load_object_klass(src), slow_region, &src);
6634 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
6635
6636 // (2) src and dest arrays must have elements of the same BasicType
6637 // done at macro expansion or at Ideal transformation time
6638
6639 // (4) src_offset must not be negative.
6640 generate_negative_guard(src_offset, slow_region);
6641
6642 // (5) dest_offset must not be negative.
6643 generate_negative_guard(dest_offset, slow_region);
6644
6645 // (7) src_offset + length must not exceed length of src.
6646 generate_limit_guard(src_offset, length,
6647 load_array_length(src),
6648 slow_region);
6649
6650 // (8) dest_offset + length must not exceed length of dest.
6651 generate_limit_guard(dest_offset, length,
6652 load_array_length(dest),
6653 slow_region);
6654
6655 // (6) length must not be negative.
6656 // This is also checked in generate_arraycopy() during macro expansion, but
6657 // we also have to check it here for the case where the ArrayCopyNode will
6658 // be eliminated by Escape Analysis.
6659 if (EliminateAllocations) {
6660 generate_negative_guard(length, slow_region);
6661 negative_length_guard_generated = true;
6662 }
6663
6664 // (9) each element of an oop array must be assignable
6665 Node* dest_klass = load_object_klass(dest);
6666 Node* refined_dest_klass = dest_klass;
6667 if (src != dest) {
6668 dest_klass = load_non_refined_array_klass(refined_dest_klass);
6669 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6670 slow_region->add_req(not_subtype_ctrl);
6671 }
6672
6673 // TODO 8251971 Improve this. What about atomicity? Make sure this is always folded for type arrays.
6674 // If destination is null-restricted, source must be null-restricted as well: src_null_restricted || !dst_null_restricted
6675 Node* src_klass = load_object_klass(src);
6676 Node* adr_prop_src = basic_plus_adr(top(), src_klass, in_bytes(ArrayKlass::properties_offset()));
6677 Node* prop_src = _gvn.transform(LoadNode::make(_gvn, control(), immutable_memory(), adr_prop_src,
6678 _gvn.type(adr_prop_src)->is_ptr(), TypeInt::INT, T_INT,
6679 MemNode::unordered));
6680 Node* adr_prop_dest = basic_plus_adr(top(), refined_dest_klass, in_bytes(ArrayKlass::properties_offset()));
6681 Node* prop_dest = _gvn.transform(LoadNode::make(_gvn, control(), immutable_memory(), adr_prop_dest,
6682 _gvn.type(adr_prop_dest)->is_ptr(), TypeInt::INT, T_INT,
6683 MemNode::unordered));
6684
6685 const ArrayProperties props_null_restricted = ArrayProperties::Default().with_null_restricted();
6686 jint props_value = (jint)props_null_restricted.value();
6687
6688 prop_dest = _gvn.transform(new XorINode(prop_dest, intcon(props_value)));
6689 prop_src = _gvn.transform(new OrINode(prop_dest, prop_src));
6690 prop_src = _gvn.transform(new AndINode(prop_src, intcon(props_value)));
6691
6692 Node* chk = _gvn.transform(new CmpINode(prop_src, intcon(props_value)));
6693 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::ne));
6694 generate_fair_guard(tst, slow_region);
6695
6696 // TODO 8251971 This is too strong
6697 generate_fair_guard(flat_array_test(src), slow_region);
6698 generate_fair_guard(flat_array_test(dest), slow_region);
6699
6700 {
6701 PreserveJVMState pjvms(this);
6702 set_control(_gvn.transform(slow_region));
6703 uncommon_trap(Deoptimization::Reason_intrinsic,
6704 Deoptimization::Action_make_not_entrant);
6705 assert(stopped(), "Should be stopped");
6706 }
6707
6708 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->isa_klassptr();
6709 if (dest_klass_t == nullptr) {
6710 // refined_dest_klass may not be an array, which leads to dest_klass being top. This means we
6711 // are in a dead path.
6712 uncommon_trap(Deoptimization::Reason_intrinsic,
6713 Deoptimization::Action_make_not_entrant);
6714 return true;
6715 }
6716
6717 const Type* toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6718 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6719 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6720 }
6721
6722 if (stopped()) {
6723 return true;
6724 }
6725
6726 Node* dest_klass = load_object_klass(dest);
6727 dest_klass = load_non_refined_array_klass(dest_klass);
6728
6729 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6730 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6731 // so the compiler has a chance to eliminate them: during macro expansion,
6732 // we have to set their control (CastPP nodes are eliminated).
6733 load_object_klass(src), dest_klass,
6734 load_array_length(src), load_array_length(dest));
6735
6736 ac->set_arraycopy(validated);
6737
6738 Node* n = _gvn.transform(ac);
6739 if (n == ac) {
6740 ac->connect_outputs(this);
6741 } else {
6742 assert(validated, "shouldn't transform if all arguments not validated");
6743 set_all_memory(n);
6744 }
6745 clear_upper_avx();
6746
6747
6748 return true;
6749 }
6750
6751
6752 // Helper function which determines if an arraycopy immediately follows
6753 // an allocation, with no intervening tests or other escapes for the object.
6754 AllocateArrayNode*
6755 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6756 if (stopped()) return nullptr; // no fast path
6757 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6758
6759 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6760 if (alloc == nullptr) return nullptr;
6761
6762 Node* rawmem = memory(Compile::AliasIdxRaw);
6763 // Is the allocation's memory state untouched?
6764 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6765 // Bail out if there have been raw-memory effects since the allocation.
6766 // (Example: There might have been a call or safepoint.)
6767 return nullptr;
6768 }
6769 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6770 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6771 return nullptr;
6772 }
6773
6774 // There must be no unexpected observers of this allocation.
6775 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6776 Node* obs = ptr->fast_out(i);
6777 if (obs != this->map()) {
6778 return nullptr;
6779 }
6780 }
6781
6782 // This arraycopy must unconditionally follow the allocation of the ptr.
6783 Node* alloc_ctl = ptr->in(0);
6784 Node* ctl = control();
6785 while (ctl != alloc_ctl) {
6786 // There may be guards which feed into the slow_region.
6787 // Any other control flow means that we might not get a chance
6788 // to finish initializing the allocated object.
6789 // Various low-level checks bottom out in uncommon traps. These
6790 // are considered safe since we've already checked above that
6791 // there is no unexpected observer of this allocation.
6792 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6793 assert(ctl->in(0)->is_If(), "must be If");
6794 ctl = ctl->in(0)->in(0);
6795 } else {
6796 return nullptr;
6797 }
6798 }
6799
6800 // If we get this far, we have an allocation which immediately
6801 // precedes the arraycopy, and we can take over zeroing the new object.
6802 // The arraycopy will finish the initialization, and provide
6803 // a new control state to which we will anchor the destination pointer.
6804
6805 return alloc;
6806 }
6807
6808 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6809 if (node->is_IfProj()) {
6810 IfProjNode* other_proj = node->as_IfProj()->other_if_proj();
6811 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6812 Node* obs = other_proj->fast_out(j);
6813 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6814 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6815 return obs->as_CallStaticJava();
6816 }
6817 }
6818 }
6819 return nullptr;
6820 }
6821
6822 //-------------inline_encodeISOArray-----------------------------------
6823 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6824 // int java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6825 // int java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6826 // encode char[] to byte[] in ISO_8859_1 or ASCII
6827 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6828 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6829 // no receiver since it is static method
6830 Node *src = argument(0);
6831 Node *src_offset = argument(1);
6832 Node *dst = argument(2);
6833 Node *dst_offset = argument(3);
6834 Node *length = argument(4);
6835
6836 // Cast source & target arrays to not-null
6837 src = must_be_not_null(src, true);
6838 dst = must_be_not_null(dst, true);
6839 if (stopped()) {
6840 return true;
6841 }
6842
6843 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6844 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6845 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6846 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6847 // failed array check
6848 return false;
6849 }
6850
6851 // Figure out the size and type of the elements we will be copying.
6852 BasicType src_elem = src_type->elem()->array_element_basic_type();
6853 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6854 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6855 return false;
6856 }
6857
6858 // Check source & target bounds
6859 RegionNode* bailout = create_bailout();
6860 generate_string_range_check(src, src_offset, length, src_elem == T_BYTE, bailout);
6861 generate_string_range_check(dst, dst_offset, length, false, bailout);
6862 if (check_bailout(bailout)) {
6863 return true;
6864 }
6865
6866 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6867 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6868 // 'src_start' points to src array + scaled offset
6869 // 'dst_start' points to dst array + scaled offset
6870
6871 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6872 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6873 enc = _gvn.transform(enc);
6874 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6875 set_memory(res_mem, mtype);
6876 set_result(enc);
6877 clear_upper_avx();
6878
6879 return true;
6880 }
6881
6882 //-------------inline_multiplyToLen-----------------------------------
6883 bool LibraryCallKit::inline_multiplyToLen() {
6884 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6885
6886 address stubAddr = StubRoutines::multiplyToLen();
6887 if (stubAddr == nullptr) {
6888 return false; // Intrinsic's stub is not implemented on this platform
6889 }
6890 const char* stubName = "multiplyToLen";
6891
6892 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6893
6894 // no receiver because it is a static method
6895 Node* x = argument(0);
6896 Node* xlen = argument(1);
6897 Node* y = argument(2);
6898 Node* ylen = argument(3);
6899 Node* z = argument(4);
6900
6901 x = must_be_not_null(x, true);
6902 y = must_be_not_null(y, true);
6903
6904 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6905 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6906 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6907 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6908 // failed array check
6909 return false;
6910 }
6911
6912 BasicType x_elem = x_type->elem()->array_element_basic_type();
6913 BasicType y_elem = y_type->elem()->array_element_basic_type();
6914 if (x_elem != T_INT || y_elem != T_INT) {
6915 return false;
6916 }
6917
6918 Node* x_start = array_element_address(x, intcon(0), x_elem);
6919 Node* y_start = array_element_address(y, intcon(0), y_elem);
6920 // 'x_start' points to x array + scaled xlen
6921 // 'y_start' points to y array + scaled ylen
6922
6923 Node* z_start = array_element_address(z, intcon(0), T_INT);
6924
6925 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6926 OptoRuntime::multiplyToLen_Type(),
6927 stubAddr, stubName, TypePtr::BOTTOM,
6928 x_start, xlen, y_start, ylen, z_start);
6929
6930 C->set_has_split_ifs(true); // Has chance for split-if optimization
6931 set_result(z);
6932 return true;
6933 }
6934
6935 //-------------inline_squareToLen------------------------------------
6936 bool LibraryCallKit::inline_squareToLen() {
6937 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6938
6939 address stubAddr = StubRoutines::squareToLen();
6940 if (stubAddr == nullptr) {
6941 return false; // Intrinsic's stub is not implemented on this platform
6942 }
6943 const char* stubName = "squareToLen";
6944
6945 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6946
6947 Node* x = argument(0);
6948 Node* len = argument(1);
6949 Node* z = argument(2);
6950 Node* zlen = argument(3);
6951
6952 x = must_be_not_null(x, true);
6953 z = must_be_not_null(z, true);
6954
6955 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6956 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6957 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6958 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6959 // failed array check
6960 return false;
6961 }
6962
6963 BasicType x_elem = x_type->elem()->array_element_basic_type();
6964 BasicType z_elem = z_type->elem()->array_element_basic_type();
6965 if (x_elem != T_INT || z_elem != T_INT) {
6966 return false;
6967 }
6968
6969
6970 Node* x_start = array_element_address(x, intcon(0), x_elem);
6971 Node* z_start = array_element_address(z, intcon(0), z_elem);
6972
6973 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6974 OptoRuntime::squareToLen_Type(),
6975 stubAddr, stubName, TypePtr::BOTTOM,
6976 x_start, len, z_start, zlen);
6977
6978 set_result(z);
6979 return true;
6980 }
6981
6982 //-------------inline_mulAdd------------------------------------------
6983 bool LibraryCallKit::inline_mulAdd() {
6984 assert(UseMulAddIntrinsic, "not implemented on this platform");
6985
6986 address stubAddr = StubRoutines::mulAdd();
6987 if (stubAddr == nullptr) {
6988 return false; // Intrinsic's stub is not implemented on this platform
6989 }
6990 const char* stubName = "mulAdd";
6991
6992 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6993
6994 Node* out = argument(0);
6995 Node* in = argument(1);
6996 Node* offset = argument(2);
6997 Node* len = argument(3);
6998 Node* k = argument(4);
6999
7000 in = must_be_not_null(in, true);
7001 out = must_be_not_null(out, true);
7002
7003 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
7004 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
7005 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
7006 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
7007 // failed array check
7008 return false;
7009 }
7010
7011 BasicType out_elem = out_type->elem()->array_element_basic_type();
7012 BasicType in_elem = in_type->elem()->array_element_basic_type();
7013 if (out_elem != T_INT || in_elem != T_INT) {
7014 return false;
7015 }
7016
7017 Node* outlen = load_array_length(out);
7018 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
7019 Node* out_start = array_element_address(out, intcon(0), out_elem);
7020 Node* in_start = array_element_address(in, intcon(0), in_elem);
7021
7022 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
7023 OptoRuntime::mulAdd_Type(),
7024 stubAddr, stubName, TypePtr::BOTTOM,
7025 out_start,in_start, new_offset, len, k);
7026 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7027 set_result(result);
7028 return true;
7029 }
7030
7031 //-------------inline_montgomeryMultiply-----------------------------------
7032 bool LibraryCallKit::inline_montgomeryMultiply() {
7033 address stubAddr = StubRoutines::montgomeryMultiply();
7034 if (stubAddr == nullptr) {
7035 return false; // Intrinsic's stub is not implemented on this platform
7036 }
7037
7038 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
7039 const char* stubName = "montgomery_multiply";
7040
7041 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
7042
7043 Node* a = argument(0);
7044 Node* b = argument(1);
7045 Node* n = argument(2);
7046 Node* len = argument(3);
7047 Node* inv = argument(4);
7048 Node* m = argument(6);
7049
7050 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
7051 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
7052 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
7053 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
7054 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
7055 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
7056 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
7057 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
7058 // failed array check
7059 return false;
7060 }
7061
7062 BasicType a_elem = a_type->elem()->array_element_basic_type();
7063 BasicType b_elem = b_type->elem()->array_element_basic_type();
7064 BasicType n_elem = n_type->elem()->array_element_basic_type();
7065 BasicType m_elem = m_type->elem()->array_element_basic_type();
7066 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
7067 return false;
7068 }
7069
7070 // Make the call
7071 {
7072 Node* a_start = array_element_address(a, intcon(0), a_elem);
7073 Node* b_start = array_element_address(b, intcon(0), b_elem);
7074 Node* n_start = array_element_address(n, intcon(0), n_elem);
7075 Node* m_start = array_element_address(m, intcon(0), m_elem);
7076
7077 Node* call = make_runtime_call(RC_LEAF,
7078 OptoRuntime::montgomeryMultiply_Type(),
7079 stubAddr, stubName, TypePtr::BOTTOM,
7080 a_start, b_start, n_start, len, inv, top(),
7081 m_start);
7082 set_result(m);
7083 }
7084
7085 return true;
7086 }
7087
7088 bool LibraryCallKit::inline_montgomerySquare() {
7089 address stubAddr = StubRoutines::montgomerySquare();
7090 if (stubAddr == nullptr) {
7091 return false; // Intrinsic's stub is not implemented on this platform
7092 }
7093
7094 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
7095 const char* stubName = "montgomery_square";
7096
7097 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
7098
7099 Node* a = argument(0);
7100 Node* n = argument(1);
7101 Node* len = argument(2);
7102 Node* inv = argument(3);
7103 Node* m = argument(5);
7104
7105 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
7106 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
7107 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
7108 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
7109 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
7110 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
7111 // failed array check
7112 return false;
7113 }
7114
7115 BasicType a_elem = a_type->elem()->array_element_basic_type();
7116 BasicType n_elem = n_type->elem()->array_element_basic_type();
7117 BasicType m_elem = m_type->elem()->array_element_basic_type();
7118 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
7119 return false;
7120 }
7121
7122 // Make the call
7123 {
7124 Node* a_start = array_element_address(a, intcon(0), a_elem);
7125 Node* n_start = array_element_address(n, intcon(0), n_elem);
7126 Node* m_start = array_element_address(m, intcon(0), m_elem);
7127
7128 Node* call = make_runtime_call(RC_LEAF,
7129 OptoRuntime::montgomerySquare_Type(),
7130 stubAddr, stubName, TypePtr::BOTTOM,
7131 a_start, n_start, len, inv, top(),
7132 m_start);
7133 set_result(m);
7134 }
7135
7136 return true;
7137 }
7138
7139 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
7140 address stubAddr = nullptr;
7141 const char* stubName = nullptr;
7142
7143 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
7144 if (stubAddr == nullptr) {
7145 return false; // Intrinsic's stub is not implemented on this platform
7146 }
7147
7148 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
7149
7150 assert(callee()->signature()->size() == 5, "expected 5 arguments");
7151
7152 Node* newArr = argument(0);
7153 Node* oldArr = argument(1);
7154 Node* newIdx = argument(2);
7155 Node* shiftCount = argument(3);
7156 Node* numIter = argument(4);
7157
7158 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
7159 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
7160 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
7161 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
7162 return false;
7163 }
7164
7165 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
7166 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
7167 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
7168 return false;
7169 }
7170
7171 // Make the call
7172 {
7173 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
7174 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
7175
7176 Node* call = make_runtime_call(RC_LEAF,
7177 OptoRuntime::bigIntegerShift_Type(),
7178 stubAddr,
7179 stubName,
7180 TypePtr::BOTTOM,
7181 newArr_start,
7182 oldArr_start,
7183 newIdx,
7184 shiftCount,
7185 numIter);
7186 }
7187
7188 return true;
7189 }
7190
7191 //-------------inline_vectorizedMismatch------------------------------
7192 bool LibraryCallKit::inline_vectorizedMismatch() {
7193 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
7194
7195 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
7196 Node* obja = argument(0); // Object
7197 Node* aoffset = argument(1); // long
7198 Node* objb = argument(3); // Object
7199 Node* boffset = argument(4); // long
7200 Node* length = argument(6); // int
7201 Node* scale = argument(7); // int
7202
7203 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
7204 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
7205 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
7206 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
7207 scale == top()) {
7208 return false; // failed input validation
7209 }
7210
7211 Node* obja_adr = make_unsafe_address(obja, aoffset);
7212 Node* objb_adr = make_unsafe_address(objb, boffset);
7213
7214 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
7215 //
7216 // inline_limit = ArrayOperationPartialInlineSize / element_size;
7217 // if (length <= inline_limit) {
7218 // inline_path:
7219 // vmask = VectorMaskGen length
7220 // vload1 = LoadVectorMasked obja, vmask
7221 // vload2 = LoadVectorMasked objb, vmask
7222 // result1 = VectorCmpMasked vload1, vload2, vmask
7223 // } else {
7224 // call_stub_path:
7225 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
7226 // }
7227 // exit_block:
7228 // return Phi(result1, result2);
7229 //
7230 enum { inline_path = 1, // input is small enough to process it all at once
7231 stub_path = 2, // input is too large; call into the VM
7232 PATH_LIMIT = 3
7233 };
7234
7235 Node* exit_block = new RegionNode(PATH_LIMIT);
7236 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
7237 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
7238
7239 Node* call_stub_path = control();
7240
7241 BasicType elem_bt = T_ILLEGAL;
7242
7243 const TypeInt* scale_t = _gvn.type(scale)->is_int();
7244 if (scale_t->is_con()) {
7245 switch (scale_t->get_con()) {
7246 case 0: elem_bt = T_BYTE; break;
7247 case 1: elem_bt = T_SHORT; break;
7248 case 2: elem_bt = T_INT; break;
7249 case 3: elem_bt = T_LONG; break;
7250
7251 default: elem_bt = T_ILLEGAL; break; // not supported
7252 }
7253 }
7254
7255 int inline_limit = 0;
7256 bool do_partial_inline = false;
7257
7258 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
7259 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
7260 do_partial_inline = inline_limit >= 16;
7261 }
7262
7263 if (do_partial_inline) {
7264 assert(elem_bt != T_ILLEGAL, "sanity");
7265
7266 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
7267 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
7268 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
7269
7270 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
7271 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
7272 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
7273
7274 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
7275
7276 if (!stopped()) {
7277 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
7278
7279 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
7280 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
7281 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
7282 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
7283
7284 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
7285 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
7286 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
7287 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
7288
7289 exit_block->init_req(inline_path, control());
7290 memory_phi->init_req(inline_path, map()->memory());
7291 result_phi->init_req(inline_path, result);
7292
7293 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
7294 clear_upper_avx();
7295 }
7296 }
7297 }
7298
7299 if (call_stub_path != nullptr) {
7300 set_control(call_stub_path);
7301
7302 Node* call = make_runtime_call(RC_LEAF,
7303 OptoRuntime::vectorizedMismatch_Type(),
7304 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
7305 obja_adr, objb_adr, length, scale);
7306
7307 exit_block->init_req(stub_path, control());
7308 memory_phi->init_req(stub_path, map()->memory());
7309 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
7310 }
7311
7312 exit_block = _gvn.transform(exit_block);
7313 memory_phi = _gvn.transform(memory_phi);
7314 result_phi = _gvn.transform(result_phi);
7315
7316 record_for_igvn(exit_block);
7317 record_for_igvn(memory_phi);
7318 record_for_igvn(result_phi);
7319
7320 set_control(exit_block);
7321 set_all_memory(memory_phi);
7322 set_result(result_phi);
7323
7324 return true;
7325 }
7326
7327 //------------------------------inline_vectorizedHashcode----------------------------
7328 bool LibraryCallKit::inline_vectorizedHashCode() {
7329 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
7330
7331 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
7332 Node* array = argument(0);
7333 Node* offset = argument(1);
7334 Node* length = argument(2);
7335 Node* initialValue = argument(3);
7336 Node* basic_type = argument(4);
7337
7338 if (basic_type == top()) {
7339 return false; // failed input validation
7340 }
7341
7342 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
7343 if (!basic_type_t->is_con()) {
7344 return false; // Only intrinsify if mode argument is constant
7345 }
7346
7347 array = must_be_not_null(array, true);
7348
7349 BasicType bt = (BasicType)basic_type_t->get_con();
7350
7351 // Resolve address of first element
7352 Node* array_start = array_element_address(array, offset, bt);
7353
7354 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
7355 array_start, length, initialValue, basic_type)));
7356 clear_upper_avx();
7357
7358 return true;
7359 }
7360
7361 /**
7362 * Calculate CRC32 for byte.
7363 * int java.util.zip.CRC32.update(int crc, int b)
7364 */
7365 bool LibraryCallKit::inline_updateCRC32() {
7366 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7367 assert(callee()->signature()->size() == 2, "update has 2 parameters");
7368 // no receiver since it is static method
7369 Node* crc = argument(0); // type: int
7370 Node* b = argument(1); // type: int
7371
7372 /*
7373 * int c = ~ crc;
7374 * b = timesXtoThe32[(b ^ c) & 0xFF];
7375 * b = b ^ (c >>> 8);
7376 * crc = ~b;
7377 */
7378
7379 Node* M1 = intcon(-1);
7380 crc = _gvn.transform(new XorINode(crc, M1));
7381 Node* result = _gvn.transform(new XorINode(crc, b));
7382 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
7383
7384 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
7385 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
7386 Node* adr = off_heap_plus_addr(base, ConvI2X(offset));
7387 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
7388
7389 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
7390 result = _gvn.transform(new XorINode(crc, result));
7391 result = _gvn.transform(new XorINode(result, M1));
7392 set_result(result);
7393 return true;
7394 }
7395
7396 /**
7397 * Calculate CRC32 for byte[] array.
7398 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
7399 */
7400 bool LibraryCallKit::inline_updateBytesCRC32() {
7401 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7402 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7403 // no receiver since it is static method
7404 Node* crc = argument(0); // type: int
7405 Node* src = argument(1); // type: oop
7406 Node* offset = argument(2); // type: int
7407 Node* length = argument(3); // type: int
7408
7409 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7410 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7411 // failed array check
7412 return false;
7413 }
7414
7415 // Figure out the size and type of the elements we will be copying.
7416 BasicType src_elem = src_type->elem()->array_element_basic_type();
7417 if (src_elem != T_BYTE) {
7418 return false;
7419 }
7420
7421 // 'src_start' points to src array + scaled offset
7422 src = must_be_not_null(src, true);
7423 Node* src_start = array_element_address(src, offset, src_elem);
7424
7425 // We assume that range check is done by caller.
7426 // TODO: generate range check (offset+length < src.length) in debug VM.
7427
7428 // Call the stub.
7429 address stubAddr = StubRoutines::updateBytesCRC32();
7430 const char *stubName = "updateBytesCRC32";
7431
7432 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7433 stubAddr, stubName, TypePtr::BOTTOM,
7434 crc, src_start, length);
7435 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7436 set_result(result);
7437 return true;
7438 }
7439
7440 /**
7441 * Calculate CRC32 for ByteBuffer.
7442 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
7443 */
7444 bool LibraryCallKit::inline_updateByteBufferCRC32() {
7445 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7446 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7447 // no receiver since it is static method
7448 Node* crc = argument(0); // type: int
7449 Node* src = argument(1); // type: long
7450 Node* offset = argument(3); // type: int
7451 Node* length = argument(4); // type: int
7452
7453 src = ConvL2X(src); // adjust Java long to machine word
7454 Node* base = _gvn.transform(new CastX2PNode(src));
7455 offset = ConvI2X(offset);
7456
7457 // 'src_start' points to src array + scaled offset
7458 Node* src_start = off_heap_plus_addr(base, offset);
7459
7460 // Call the stub.
7461 address stubAddr = StubRoutines::updateBytesCRC32();
7462 const char *stubName = "updateBytesCRC32";
7463
7464 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7465 stubAddr, stubName, TypePtr::BOTTOM,
7466 crc, src_start, length);
7467 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7468 set_result(result);
7469 return true;
7470 }
7471
7472 //------------------------------get_table_from_crc32c_class-----------------------
7473 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
7474 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
7475 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
7476
7477 return table;
7478 }
7479
7480 //------------------------------inline_updateBytesCRC32C-----------------------
7481 //
7482 // Calculate CRC32C for byte[] array.
7483 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
7484 //
7485 bool LibraryCallKit::inline_updateBytesCRC32C() {
7486 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7487 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7488 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7489 // no receiver since it is a static method
7490 Node* crc = argument(0); // type: int
7491 Node* src = argument(1); // type: oop
7492 Node* offset = argument(2); // type: int
7493 Node* end = argument(3); // type: int
7494
7495 Node* length = _gvn.transform(new SubINode(end, offset));
7496
7497 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7498 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7499 // failed array check
7500 return false;
7501 }
7502
7503 // Figure out the size and type of the elements we will be copying.
7504 BasicType src_elem = src_type->elem()->array_element_basic_type();
7505 if (src_elem != T_BYTE) {
7506 return false;
7507 }
7508
7509 // 'src_start' points to src array + scaled offset
7510 src = must_be_not_null(src, true);
7511 Node* src_start = array_element_address(src, offset, src_elem);
7512
7513 // static final int[] byteTable in class CRC32C
7514 Node* table = get_table_from_crc32c_class(callee()->holder());
7515 table = must_be_not_null(table, true);
7516 Node* table_start = array_element_address(table, intcon(0), T_INT);
7517
7518 // We assume that range check is done by caller.
7519 // TODO: generate range check (offset+length < src.length) in debug VM.
7520
7521 // Call the stub.
7522 address stubAddr = StubRoutines::updateBytesCRC32C();
7523 const char *stubName = "updateBytesCRC32C";
7524
7525 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7526 stubAddr, stubName, TypePtr::BOTTOM,
7527 crc, src_start, length, table_start);
7528 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7529 set_result(result);
7530 return true;
7531 }
7532
7533 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
7534 //
7535 // Calculate CRC32C for DirectByteBuffer.
7536 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
7537 //
7538 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
7539 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7540 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
7541 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7542 // no receiver since it is a static method
7543 Node* crc = argument(0); // type: int
7544 Node* src = argument(1); // type: long
7545 Node* offset = argument(3); // type: int
7546 Node* end = argument(4); // type: int
7547
7548 Node* length = _gvn.transform(new SubINode(end, offset));
7549
7550 src = ConvL2X(src); // adjust Java long to machine word
7551 Node* base = _gvn.transform(new CastX2PNode(src));
7552 offset = ConvI2X(offset);
7553
7554 // 'src_start' points to src array + scaled offset
7555 Node* src_start = off_heap_plus_addr(base, offset);
7556
7557 // static final int[] byteTable in class CRC32C
7558 Node* table = get_table_from_crc32c_class(callee()->holder());
7559 table = must_be_not_null(table, true);
7560 Node* table_start = array_element_address(table, intcon(0), T_INT);
7561
7562 // Call the stub.
7563 address stubAddr = StubRoutines::updateBytesCRC32C();
7564 const char *stubName = "updateBytesCRC32C";
7565
7566 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7567 stubAddr, stubName, TypePtr::BOTTOM,
7568 crc, src_start, length, table_start);
7569 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7570 set_result(result);
7571 return true;
7572 }
7573
7574 //------------------------------inline_updateBytesAdler32----------------------
7575 //
7576 // Calculate Adler32 checksum for byte[] array.
7577 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
7578 //
7579 bool LibraryCallKit::inline_updateBytesAdler32() {
7580 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7581 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7582 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7583 // no receiver since it is static method
7584 Node* crc = argument(0); // type: int
7585 Node* src = argument(1); // type: oop
7586 Node* offset = argument(2); // type: int
7587 Node* length = argument(3); // type: int
7588
7589 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7590 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7591 // failed array check
7592 return false;
7593 }
7594
7595 // Figure out the size and type of the elements we will be copying.
7596 BasicType src_elem = src_type->elem()->array_element_basic_type();
7597 if (src_elem != T_BYTE) {
7598 return false;
7599 }
7600
7601 // 'src_start' points to src array + scaled offset
7602 Node* src_start = array_element_address(src, offset, src_elem);
7603
7604 // We assume that range check is done by caller.
7605 // TODO: generate range check (offset+length < src.length) in debug VM.
7606
7607 // Call the stub.
7608 address stubAddr = StubRoutines::updateBytesAdler32();
7609 const char *stubName = "updateBytesAdler32";
7610
7611 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7612 stubAddr, stubName, TypePtr::BOTTOM,
7613 crc, src_start, length);
7614 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7615 set_result(result);
7616 return true;
7617 }
7618
7619 //------------------------------inline_updateByteBufferAdler32---------------
7620 //
7621 // Calculate Adler32 checksum for DirectByteBuffer.
7622 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
7623 //
7624 bool LibraryCallKit::inline_updateByteBufferAdler32() {
7625 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7626 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7627 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7628 // no receiver since it is static method
7629 Node* crc = argument(0); // type: int
7630 Node* src = argument(1); // type: long
7631 Node* offset = argument(3); // type: int
7632 Node* length = argument(4); // type: int
7633
7634 src = ConvL2X(src); // adjust Java long to machine word
7635 Node* base = _gvn.transform(new CastX2PNode(src));
7636 offset = ConvI2X(offset);
7637
7638 // 'src_start' points to src array + scaled offset
7639 Node* src_start = off_heap_plus_addr(base, offset);
7640
7641 // Call the stub.
7642 address stubAddr = StubRoutines::updateBytesAdler32();
7643 const char *stubName = "updateBytesAdler32";
7644
7645 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7646 stubAddr, stubName, TypePtr::BOTTOM,
7647 crc, src_start, length);
7648
7649 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7650 set_result(result);
7651 return true;
7652 }
7653
7654 //----------------------------inline_reference_get0----------------------------
7655 // public T java.lang.ref.Reference.get();
7656 bool LibraryCallKit::inline_reference_get0() {
7657 const int referent_offset = java_lang_ref_Reference::referent_offset();
7658
7659 // Get the argument:
7660 Node* reference_obj = null_check_receiver();
7661 if (stopped()) return true;
7662
7663 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7664 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7665 decorators, /*is_static*/ false,
7666 env()->Reference_klass());
7667 if (result == nullptr) return false;
7668
7669 // Add memory barrier to prevent commoning reads from this field
7670 // across safepoint since GC can change its value.
7671 insert_mem_bar(Op_MemBarCPUOrder);
7672
7673 set_result(result);
7674 return true;
7675 }
7676
7677 //----------------------------inline_reference_refersTo0----------------------------
7678 // bool java.lang.ref.Reference.refersTo0();
7679 // bool java.lang.ref.PhantomReference.refersTo0();
7680 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7681 // Get arguments:
7682 Node* reference_obj = null_check_receiver();
7683 Node* other_obj = argument(1);
7684 if (stopped()) return true;
7685
7686 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7687 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7688 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7689 decorators, /*is_static*/ false,
7690 env()->Reference_klass());
7691 if (referent == nullptr) return false;
7692
7693 // Add memory barrier to prevent commoning reads from this field
7694 // across safepoint since GC can change its value.
7695 insert_mem_bar(Op_MemBarCPUOrder);
7696
7697 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7698 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7699 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7700
7701 RegionNode* region = new RegionNode(3);
7702 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7703
7704 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7705 region->init_req(1, if_true);
7706 phi->init_req(1, intcon(1));
7707
7708 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7709 region->init_req(2, if_false);
7710 phi->init_req(2, intcon(0));
7711
7712 set_control(_gvn.transform(region));
7713 record_for_igvn(region);
7714 set_result(_gvn.transform(phi));
7715 return true;
7716 }
7717
7718 //----------------------------inline_reference_clear0----------------------------
7719 // void java.lang.ref.Reference.clear0();
7720 // void java.lang.ref.PhantomReference.clear0();
7721 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7722 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7723
7724 // Get arguments
7725 Node* reference_obj = null_check_receiver();
7726 if (stopped()) return true;
7727
7728 // Common access parameters
7729 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7730 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7731 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7732 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7733 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7734
7735 Node* referent = access_load_at(reference_obj,
7736 referent_field_addr,
7737 referent_field_addr_type,
7738 val_type,
7739 T_OBJECT,
7740 decorators);
7741
7742 IdealKit ideal(this);
7743 #define __ ideal.
7744 __ if_then(referent, BoolTest::ne, null());
7745 sync_kit(ideal);
7746 access_store_at(reference_obj,
7747 referent_field_addr,
7748 referent_field_addr_type,
7749 null(),
7750 val_type,
7751 T_OBJECT,
7752 decorators);
7753 __ sync_kit(this);
7754 __ end_if();
7755 final_sync(ideal);
7756 #undef __
7757
7758 return true;
7759 }
7760
7761 //-----------------------inline_reference_reachabilityFence-----------------
7762 // bool java.lang.ref.Reference.reachabilityFence();
7763 bool LibraryCallKit::inline_reference_reachabilityFence() {
7764 Node* referent = argument(0);
7765 insert_reachability_fence(referent);
7766 return true;
7767 }
7768
7769 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7770 DecoratorSet decorators, bool is_static,
7771 ciInstanceKlass* fromKls) {
7772 if (fromKls == nullptr) {
7773 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7774 assert(tinst != nullptr, "obj is null");
7775 assert(tinst->is_loaded(), "obj is not loaded");
7776 fromKls = tinst->instance_klass();
7777 }
7778 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7779 ciSymbol::make(fieldTypeString),
7780 is_static);
7781
7782 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7783 if (field == nullptr) return (Node *) nullptr;
7784
7785 if (is_static) {
7786 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7787 fromObj = makecon(tip);
7788 }
7789
7790 // Next code copied from Parse::do_get_xxx():
7791
7792 // Compute address and memory type.
7793 int offset = field->offset_in_bytes();
7794 bool is_vol = field->is_volatile();
7795 ciType* field_klass = field->type();
7796 assert(field_klass->is_loaded(), "should be loaded");
7797 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7798 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7799 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7800 "slice of address and input slice don't match");
7801 BasicType bt = field->layout_type();
7802
7803 // Build the resultant type of the load
7804 const Type *type;
7805 if (bt == T_OBJECT) {
7806 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7807 } else {
7808 type = Type::get_const_basic_type(bt);
7809 }
7810
7811 if (is_vol) {
7812 decorators |= MO_SEQ_CST;
7813 }
7814
7815 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7816 }
7817
7818 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7819 bool is_exact /* true */, bool is_static /* false */,
7820 ciInstanceKlass * fromKls /* nullptr */) {
7821 if (fromKls == nullptr) {
7822 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7823 assert(tinst != nullptr, "obj is null");
7824 assert(tinst->is_loaded(), "obj is not loaded");
7825 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7826 fromKls = tinst->instance_klass();
7827 }
7828 else {
7829 assert(is_static, "only for static field access");
7830 }
7831 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7832 ciSymbol::make(fieldTypeString),
7833 is_static);
7834
7835 assert(field != nullptr, "undefined field");
7836 assert(!field->is_volatile(), "not defined for volatile fields");
7837
7838 if (is_static) {
7839 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7840 fromObj = makecon(tip);
7841 }
7842
7843 // Next code copied from Parse::do_get_xxx():
7844
7845 // Compute address and memory type.
7846 int offset = field->offset_in_bytes();
7847 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7848
7849 return adr;
7850 }
7851
7852 //------------------------------inline_aescrypt_Block-----------------------
7853 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7854 address stubAddr = nullptr;
7855 const char *stubName;
7856 bool is_decrypt = false;
7857 assert(UseAES, "need AES instruction support");
7858
7859 switch(id) {
7860 case vmIntrinsics::_aescrypt_encryptBlock:
7861 stubAddr = StubRoutines::aescrypt_encryptBlock();
7862 stubName = "aescrypt_encryptBlock";
7863 break;
7864 case vmIntrinsics::_aescrypt_decryptBlock:
7865 stubAddr = StubRoutines::aescrypt_decryptBlock();
7866 stubName = "aescrypt_decryptBlock";
7867 is_decrypt = true;
7868 break;
7869 default:
7870 break;
7871 }
7872 if (stubAddr == nullptr) return false;
7873
7874 Node* aescrypt_object = argument(0);
7875 Node* src = argument(1);
7876 Node* src_offset = argument(2);
7877 Node* dest = argument(3);
7878 Node* dest_offset = argument(4);
7879
7880 src = must_be_not_null(src, true);
7881 dest = must_be_not_null(dest, true);
7882
7883 // (1) src and dest are arrays.
7884 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7885 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7886 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7887 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7888
7889 // for the quick and dirty code we will skip all the checks.
7890 // we are just trying to get the call to be generated.
7891 Node* src_start = src;
7892 Node* dest_start = dest;
7893 if (src_offset != nullptr || dest_offset != nullptr) {
7894 assert(src_offset != nullptr && dest_offset != nullptr, "");
7895 src_start = array_element_address(src, src_offset, T_BYTE);
7896 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7897 }
7898
7899 // now need to get the start of its expanded key array
7900 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7901 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7902 if (k_start == nullptr) return false;
7903
7904 // Call the stub.
7905 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7906 stubAddr, stubName, TypePtr::BOTTOM,
7907 src_start, dest_start, k_start);
7908
7909 return true;
7910 }
7911
7912 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7913 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7914 address stubAddr = nullptr;
7915 const char *stubName = nullptr;
7916 bool is_decrypt = false;
7917 assert(UseAES, "need AES instruction support");
7918
7919 switch(id) {
7920 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7921 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7922 stubName = "cipherBlockChaining_encryptAESCrypt";
7923 break;
7924 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7925 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7926 stubName = "cipherBlockChaining_decryptAESCrypt";
7927 is_decrypt = true;
7928 break;
7929 default:
7930 break;
7931 }
7932 if (stubAddr == nullptr) return false;
7933
7934 Node* cipherBlockChaining_object = argument(0);
7935 Node* src = argument(1);
7936 Node* src_offset = argument(2);
7937 Node* len = argument(3);
7938 Node* dest = argument(4);
7939 Node* dest_offset = argument(5);
7940
7941 src = must_be_not_null(src, false);
7942 dest = must_be_not_null(dest, false);
7943
7944 // (1) src and dest are arrays.
7945 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7946 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7947 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7948 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7949
7950 // checks are the responsibility of the caller
7951 Node* src_start = src;
7952 Node* dest_start = dest;
7953 if (src_offset != nullptr || dest_offset != nullptr) {
7954 assert(src_offset != nullptr && dest_offset != nullptr, "");
7955 src_start = array_element_address(src, src_offset, T_BYTE);
7956 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7957 }
7958
7959 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7960 // (because of the predicated logic executed earlier).
7961 // so we cast it here safely.
7962 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7963
7964 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7965 if (embeddedCipherObj == nullptr) return false;
7966
7967 // cast it to what we know it will be at runtime
7968 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7969 assert(tinst != nullptr, "CBC obj is null");
7970 assert(tinst->is_loaded(), "CBC obj is not loaded");
7971 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7972 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7973
7974 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7975 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7976 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7977 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7978 aescrypt_object = _gvn.transform(aescrypt_object);
7979
7980 // we need to get the start of the aescrypt_object's expanded key array
7981 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7982 if (k_start == nullptr) return false;
7983
7984 // similarly, get the start address of the r vector
7985 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7986 if (objRvec == nullptr) return false;
7987 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7988
7989 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7990 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7991 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7992 stubAddr, stubName, TypePtr::BOTTOM,
7993 src_start, dest_start, k_start, r_start, len);
7994
7995 // return cipher length (int)
7996 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7997 set_result(retvalue);
7998 return true;
7999 }
8000
8001 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
8002 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
8003 address stubAddr = nullptr;
8004 const char *stubName = nullptr;
8005 bool is_decrypt = false;
8006 assert(UseAES, "need AES instruction support");
8007
8008 switch (id) {
8009 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
8010 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
8011 stubName = "electronicCodeBook_encryptAESCrypt";
8012 break;
8013 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
8014 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
8015 stubName = "electronicCodeBook_decryptAESCrypt";
8016 is_decrypt = true;
8017 break;
8018 default:
8019 break;
8020 }
8021
8022 if (stubAddr == nullptr) return false;
8023
8024 Node* electronicCodeBook_object = argument(0);
8025 Node* src = argument(1);
8026 Node* src_offset = argument(2);
8027 Node* len = argument(3);
8028 Node* dest = argument(4);
8029 Node* dest_offset = argument(5);
8030
8031 // (1) src and dest are arrays.
8032 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8033 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8034 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8035 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8036
8037 // checks are the responsibility of the caller
8038 Node* src_start = src;
8039 Node* dest_start = dest;
8040 if (src_offset != nullptr || dest_offset != nullptr) {
8041 assert(src_offset != nullptr && dest_offset != nullptr, "");
8042 src_start = array_element_address(src, src_offset, T_BYTE);
8043 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8044 }
8045
8046 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8047 // (because of the predicated logic executed earlier).
8048 // so we cast it here safely.
8049 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8050
8051 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8052 if (embeddedCipherObj == nullptr) return false;
8053
8054 // cast it to what we know it will be at runtime
8055 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
8056 assert(tinst != nullptr, "ECB obj is null");
8057 assert(tinst->is_loaded(), "ECB obj is not loaded");
8058 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8059 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8060
8061 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8062 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8063 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8064 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8065 aescrypt_object = _gvn.transform(aescrypt_object);
8066
8067 // we need to get the start of the aescrypt_object's expanded key array
8068 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
8069 if (k_start == nullptr) return false;
8070
8071 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8072 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
8073 OptoRuntime::electronicCodeBook_aescrypt_Type(),
8074 stubAddr, stubName, TypePtr::BOTTOM,
8075 src_start, dest_start, k_start, len);
8076
8077 // return cipher length (int)
8078 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
8079 set_result(retvalue);
8080 return true;
8081 }
8082
8083 //------------------------------inline_counterMode_AESCrypt-----------------------
8084 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
8085 assert(UseAES, "need AES instruction support");
8086 if (!UseAESCTRIntrinsics) return false;
8087
8088 address stubAddr = nullptr;
8089 const char *stubName = nullptr;
8090 if (id == vmIntrinsics::_counterMode_AESCrypt) {
8091 stubAddr = StubRoutines::counterMode_AESCrypt();
8092 stubName = "counterMode_AESCrypt";
8093 }
8094 if (stubAddr == nullptr) return false;
8095
8096 Node* counterMode_object = argument(0);
8097 Node* src = argument(1);
8098 Node* src_offset = argument(2);
8099 Node* len = argument(3);
8100 Node* dest = argument(4);
8101 Node* dest_offset = argument(5);
8102
8103 // (1) src and dest are arrays.
8104 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8105 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8106 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8107 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8108
8109 // checks are the responsibility of the caller
8110 Node* src_start = src;
8111 Node* dest_start = dest;
8112 if (src_offset != nullptr || dest_offset != nullptr) {
8113 assert(src_offset != nullptr && dest_offset != nullptr, "");
8114 src_start = array_element_address(src, src_offset, T_BYTE);
8115 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8116 }
8117
8118 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8119 // (because of the predicated logic executed earlier).
8120 // so we cast it here safely.
8121 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8122 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8123 if (embeddedCipherObj == nullptr) return false;
8124 // cast it to what we know it will be at runtime
8125 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
8126 assert(tinst != nullptr, "CTR obj is null");
8127 assert(tinst->is_loaded(), "CTR obj is not loaded");
8128 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8129 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8130 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8131 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8132 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8133 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8134 aescrypt_object = _gvn.transform(aescrypt_object);
8135 // we need to get the start of the aescrypt_object's expanded key array
8136 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
8137 if (k_start == nullptr) return false;
8138 // similarly, get the start address of the r vector
8139 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
8140 if (obj_counter == nullptr) return false;
8141 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
8142
8143 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
8144 if (saved_encCounter == nullptr) return false;
8145 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
8146 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
8147
8148 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8149 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8150 OptoRuntime::counterMode_aescrypt_Type(),
8151 stubAddr, stubName, TypePtr::BOTTOM,
8152 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
8153
8154 // return cipher length (int)
8155 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
8156 set_result(retvalue);
8157 return true;
8158 }
8159
8160 //------------------------------get_key_start_from_aescrypt_object-----------------------
8161 Node* LibraryCallKit::get_key_start_from_aescrypt_object(Node* aescrypt_object, bool is_decrypt) {
8162 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
8163 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
8164 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
8165 // The following platform specific stubs of encryption and decryption use the same round keys.
8166 #if defined(PPC64) || defined(S390) || defined(RISCV64)
8167 bool use_decryption_key = false;
8168 #else
8169 bool use_decryption_key = is_decrypt;
8170 #endif
8171 Node* objAESCryptKey = load_field_from_object(aescrypt_object, use_decryption_key ? "sessionKd" : "sessionKe", "[I");
8172 assert(objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AES_Crypt");
8173 if (objAESCryptKey == nullptr) return (Node *) nullptr;
8174
8175 // now have the array, need to get the start address of the selected key array
8176 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
8177 return k_start;
8178 }
8179
8180 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
8181 // Return node representing slow path of predicate check.
8182 // the pseudo code we want to emulate with this predicate is:
8183 // for encryption:
8184 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8185 // for decryption:
8186 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8187 // note cipher==plain is more conservative than the original java code but that's OK
8188 //
8189 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
8190 // The receiver was checked for null already.
8191 Node* objCBC = argument(0);
8192
8193 Node* src = argument(1);
8194 Node* dest = argument(4);
8195
8196 // Load embeddedCipher field of CipherBlockChaining object.
8197 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8198
8199 // get AESCrypt klass for instanceOf check
8200 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8201 // will have same classloader as CipherBlockChaining object
8202 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
8203 assert(tinst != nullptr, "CBCobj is null");
8204 assert(tinst->is_loaded(), "CBCobj is not loaded");
8205
8206 // we want to do an instanceof comparison against the AESCrypt class
8207 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8208 if (!klass_AESCrypt->is_loaded()) {
8209 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8210 Node* ctrl = control();
8211 set_control(top()); // no regular fast path
8212 return ctrl;
8213 }
8214
8215 src = must_be_not_null(src, true);
8216 dest = must_be_not_null(dest, true);
8217
8218 // Resolve oops to stable for CmpP below.
8219 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8220
8221 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8222 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8223 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8224
8225 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8226
8227 // for encryption, we are done
8228 if (!decrypting)
8229 return instof_false; // even if it is null
8230
8231 // for decryption, we need to add a further check to avoid
8232 // taking the intrinsic path when cipher and plain are the same
8233 // see the original java code for why.
8234 RegionNode* region = new RegionNode(3);
8235 region->init_req(1, instof_false);
8236
8237 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8238 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8239 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8240 region->init_req(2, src_dest_conjoint);
8241
8242 record_for_igvn(region);
8243 return _gvn.transform(region);
8244 }
8245
8246 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
8247 // Return node representing slow path of predicate check.
8248 // the pseudo code we want to emulate with this predicate is:
8249 // for encryption:
8250 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8251 // for decryption:
8252 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8253 // note cipher==plain is more conservative than the original java code but that's OK
8254 //
8255 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
8256 // The receiver was checked for null already.
8257 Node* objECB = argument(0);
8258
8259 // Load embeddedCipher field of ElectronicCodeBook object.
8260 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8261
8262 // get AESCrypt klass for instanceOf check
8263 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8264 // will have same classloader as ElectronicCodeBook object
8265 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
8266 assert(tinst != nullptr, "ECBobj is null");
8267 assert(tinst->is_loaded(), "ECBobj is not loaded");
8268
8269 // we want to do an instanceof comparison against the AESCrypt class
8270 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8271 if (!klass_AESCrypt->is_loaded()) {
8272 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8273 Node* ctrl = control();
8274 set_control(top()); // no regular fast path
8275 return ctrl;
8276 }
8277 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8278
8279 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8280 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8281 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8282
8283 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8284
8285 // for encryption, we are done
8286 if (!decrypting)
8287 return instof_false; // even if it is null
8288
8289 // for decryption, we need to add a further check to avoid
8290 // taking the intrinsic path when cipher and plain are the same
8291 // see the original java code for why.
8292 RegionNode* region = new RegionNode(3);
8293 region->init_req(1, instof_false);
8294 Node* src = argument(1);
8295 Node* dest = argument(4);
8296 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8297 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8298 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8299 region->init_req(2, src_dest_conjoint);
8300
8301 record_for_igvn(region);
8302 return _gvn.transform(region);
8303 }
8304
8305 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
8306 // Return node representing slow path of predicate check.
8307 // the pseudo code we want to emulate with this predicate is:
8308 // for encryption:
8309 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8310 // for decryption:
8311 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8312 // note cipher==plain is more conservative than the original java code but that's OK
8313 //
8314
8315 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
8316 // The receiver was checked for null already.
8317 Node* objCTR = argument(0);
8318
8319 // Load embeddedCipher field of CipherBlockChaining object.
8320 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8321
8322 // get AESCrypt klass for instanceOf check
8323 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8324 // will have same classloader as CipherBlockChaining object
8325 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
8326 assert(tinst != nullptr, "CTRobj is null");
8327 assert(tinst->is_loaded(), "CTRobj is not loaded");
8328
8329 // we want to do an instanceof comparison against the AESCrypt class
8330 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8331 if (!klass_AESCrypt->is_loaded()) {
8332 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8333 Node* ctrl = control();
8334 set_control(top()); // no regular fast path
8335 return ctrl;
8336 }
8337
8338 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8339 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8340 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8341 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8342 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8343
8344 return instof_false; // even if it is null
8345 }
8346
8347 //------------------------------inline_ghash_processBlocks
8348 bool LibraryCallKit::inline_ghash_processBlocks() {
8349 address stubAddr;
8350 const char *stubName;
8351 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
8352
8353 stubAddr = StubRoutines::ghash_processBlocks();
8354 stubName = "ghash_processBlocks";
8355
8356 Node* data = argument(0);
8357 Node* offset = argument(1);
8358 Node* len = argument(2);
8359 Node* state = argument(3);
8360 Node* subkeyH = argument(4);
8361
8362 state = must_be_not_null(state, true);
8363 subkeyH = must_be_not_null(subkeyH, true);
8364 data = must_be_not_null(data, true);
8365
8366 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8367 assert(state_start, "state is null");
8368 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
8369 assert(subkeyH_start, "subkeyH is null");
8370 Node* data_start = array_element_address(data, offset, T_BYTE);
8371 assert(data_start, "data is null");
8372
8373 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
8374 OptoRuntime::ghash_processBlocks_Type(),
8375 stubAddr, stubName, TypePtr::BOTTOM,
8376 state_start, subkeyH_start, data_start, len);
8377 return true;
8378 }
8379
8380 //------------------------------inline_chacha20Block
8381 bool LibraryCallKit::inline_chacha20Block() {
8382 address stubAddr;
8383 const char *stubName;
8384 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
8385
8386 stubAddr = StubRoutines::chacha20Block();
8387 stubName = "chacha20Block";
8388
8389 Node* state = argument(0);
8390 Node* result = argument(1);
8391
8392 state = must_be_not_null(state, true);
8393 result = must_be_not_null(result, true);
8394
8395 Node* state_start = array_element_address(state, intcon(0), T_INT);
8396 assert(state_start, "state is null");
8397 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
8398 assert(result_start, "result is null");
8399
8400 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
8401 OptoRuntime::chacha20Block_Type(),
8402 stubAddr, stubName, TypePtr::BOTTOM,
8403 state_start, result_start);
8404 // return key stream length (int)
8405 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
8406 set_result(retvalue);
8407 return true;
8408 }
8409
8410 //------------------------------inline_kyberNtt
8411 bool LibraryCallKit::inline_kyberNtt() {
8412 address stubAddr;
8413 const char *stubName;
8414 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8415 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
8416
8417 stubAddr = StubRoutines::kyberNtt();
8418 stubName = "kyberNtt";
8419 if (!stubAddr) return false;
8420
8421 Node* coeffs = argument(0);
8422 Node* ntt_zetas = argument(1);
8423
8424 coeffs = must_be_not_null(coeffs, true);
8425 ntt_zetas = must_be_not_null(ntt_zetas, true);
8426
8427 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8428 assert(coeffs_start, "coeffs is null");
8429 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
8430 assert(ntt_zetas_start, "ntt_zetas is null");
8431 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8432 OptoRuntime::kyberNtt_Type(),
8433 stubAddr, stubName, TypePtr::BOTTOM,
8434 coeffs_start, ntt_zetas_start);
8435 // return an int
8436 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
8437 set_result(retvalue);
8438 return true;
8439 }
8440
8441 //------------------------------inline_kyberInverseNtt
8442 bool LibraryCallKit::inline_kyberInverseNtt() {
8443 address stubAddr;
8444 const char *stubName;
8445 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8446 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
8447
8448 stubAddr = StubRoutines::kyberInverseNtt();
8449 stubName = "kyberInverseNtt";
8450 if (!stubAddr) return false;
8451
8452 Node* coeffs = argument(0);
8453 Node* zetas = argument(1);
8454
8455 coeffs = must_be_not_null(coeffs, true);
8456 zetas = must_be_not_null(zetas, true);
8457
8458 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8459 assert(coeffs_start, "coeffs is null");
8460 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8461 assert(zetas_start, "inverseNtt_zetas is null");
8462 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8463 OptoRuntime::kyberInverseNtt_Type(),
8464 stubAddr, stubName, TypePtr::BOTTOM,
8465 coeffs_start, zetas_start);
8466
8467 // return an int
8468 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
8469 set_result(retvalue);
8470 return true;
8471 }
8472
8473 //------------------------------inline_kyberNttMult
8474 bool LibraryCallKit::inline_kyberNttMult() {
8475 address stubAddr;
8476 const char *stubName;
8477 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8478 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
8479
8480 stubAddr = StubRoutines::kyberNttMult();
8481 stubName = "kyberNttMult";
8482 if (!stubAddr) return false;
8483
8484 Node* result = argument(0);
8485 Node* ntta = argument(1);
8486 Node* nttb = argument(2);
8487 Node* zetas = argument(3);
8488
8489 result = must_be_not_null(result, true);
8490 ntta = must_be_not_null(ntta, true);
8491 nttb = must_be_not_null(nttb, true);
8492 zetas = must_be_not_null(zetas, true);
8493
8494 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8495 assert(result_start, "result is null");
8496 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
8497 assert(ntta_start, "ntta is null");
8498 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
8499 assert(nttb_start, "nttb is null");
8500 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8501 assert(zetas_start, "nttMult_zetas is null");
8502 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8503 OptoRuntime::kyberNttMult_Type(),
8504 stubAddr, stubName, TypePtr::BOTTOM,
8505 result_start, ntta_start, nttb_start,
8506 zetas_start);
8507
8508 // return an int
8509 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
8510 set_result(retvalue);
8511
8512 return true;
8513 }
8514
8515 //------------------------------inline_kyberAddPoly_2
8516 bool LibraryCallKit::inline_kyberAddPoly_2() {
8517 address stubAddr;
8518 const char *stubName;
8519 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8520 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
8521
8522 stubAddr = StubRoutines::kyberAddPoly_2();
8523 stubName = "kyberAddPoly_2";
8524 if (!stubAddr) return false;
8525
8526 Node* result = argument(0);
8527 Node* a = argument(1);
8528 Node* b = argument(2);
8529
8530 result = must_be_not_null(result, true);
8531 a = must_be_not_null(a, true);
8532 b = must_be_not_null(b, true);
8533
8534 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8535 assert(result_start, "result is null");
8536 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8537 assert(a_start, "a is null");
8538 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8539 assert(b_start, "b is null");
8540 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
8541 OptoRuntime::kyberAddPoly_2_Type(),
8542 stubAddr, stubName, TypePtr::BOTTOM,
8543 result_start, a_start, b_start);
8544 // return an int
8545 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
8546 set_result(retvalue);
8547 return true;
8548 }
8549
8550 //------------------------------inline_kyberAddPoly_3
8551 bool LibraryCallKit::inline_kyberAddPoly_3() {
8552 address stubAddr;
8553 const char *stubName;
8554 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8555 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
8556
8557 stubAddr = StubRoutines::kyberAddPoly_3();
8558 stubName = "kyberAddPoly_3";
8559 if (!stubAddr) return false;
8560
8561 Node* result = argument(0);
8562 Node* a = argument(1);
8563 Node* b = argument(2);
8564 Node* c = argument(3);
8565
8566 result = must_be_not_null(result, true);
8567 a = must_be_not_null(a, true);
8568 b = must_be_not_null(b, true);
8569 c = must_be_not_null(c, true);
8570
8571 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8572 assert(result_start, "result is null");
8573 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8574 assert(a_start, "a is null");
8575 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8576 assert(b_start, "b is null");
8577 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
8578 assert(c_start, "c is null");
8579 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
8580 OptoRuntime::kyberAddPoly_3_Type(),
8581 stubAddr, stubName, TypePtr::BOTTOM,
8582 result_start, a_start, b_start, c_start);
8583 // return an int
8584 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
8585 set_result(retvalue);
8586 return true;
8587 }
8588
8589 //------------------------------inline_kyber12To16
8590 bool LibraryCallKit::inline_kyber12To16() {
8591 address stubAddr;
8592 const char *stubName;
8593 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8594 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
8595
8596 stubAddr = StubRoutines::kyber12To16();
8597 stubName = "kyber12To16";
8598 if (!stubAddr) return false;
8599
8600 Node* condensed = argument(0);
8601 Node* condensedOffs = argument(1);
8602 Node* parsed = argument(2);
8603 Node* parsedLength = argument(3);
8604
8605 condensed = must_be_not_null(condensed, true);
8606 parsed = must_be_not_null(parsed, true);
8607
8608 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
8609 assert(condensed_start, "condensed is null");
8610 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
8611 assert(parsed_start, "parsed is null");
8612 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
8613 OptoRuntime::kyber12To16_Type(),
8614 stubAddr, stubName, TypePtr::BOTTOM,
8615 condensed_start, condensedOffs, parsed_start, parsedLength);
8616 // return an int
8617 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
8618 set_result(retvalue);
8619 return true;
8620
8621 }
8622
8623 //------------------------------inline_kyberBarrettReduce
8624 bool LibraryCallKit::inline_kyberBarrettReduce() {
8625 address stubAddr;
8626 const char *stubName;
8627 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8628 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
8629
8630 stubAddr = StubRoutines::kyberBarrettReduce();
8631 stubName = "kyberBarrettReduce";
8632 if (!stubAddr) return false;
8633
8634 Node* coeffs = argument(0);
8635
8636 coeffs = must_be_not_null(coeffs, true);
8637
8638 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8639 assert(coeffs_start, "coeffs is null");
8640 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
8641 OptoRuntime::kyberBarrettReduce_Type(),
8642 stubAddr, stubName, TypePtr::BOTTOM,
8643 coeffs_start);
8644 // return an int
8645 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
8646 set_result(retvalue);
8647 return true;
8648 }
8649
8650 //------------------------------inline_dilithiumAlmostNtt
8651 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
8652 address stubAddr;
8653 const char *stubName;
8654 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8655 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
8656
8657 stubAddr = StubRoutines::dilithiumAlmostNtt();
8658 stubName = "dilithiumAlmostNtt";
8659 if (!stubAddr) return false;
8660
8661 Node* coeffs = argument(0);
8662 Node* ntt_zetas = argument(1);
8663
8664 coeffs = must_be_not_null(coeffs, true);
8665 ntt_zetas = must_be_not_null(ntt_zetas, true);
8666
8667 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8668 assert(coeffs_start, "coeffs is null");
8669 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
8670 assert(ntt_zetas_start, "ntt_zetas is null");
8671 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8672 OptoRuntime::dilithiumAlmostNtt_Type(),
8673 stubAddr, stubName, TypePtr::BOTTOM,
8674 coeffs_start, ntt_zetas_start);
8675 // return an int
8676 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
8677 set_result(retvalue);
8678 return true;
8679 }
8680
8681 //------------------------------inline_dilithiumAlmostInverseNtt
8682 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
8683 address stubAddr;
8684 const char *stubName;
8685 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8686 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
8687
8688 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
8689 stubName = "dilithiumAlmostInverseNtt";
8690 if (!stubAddr) return false;
8691
8692 Node* coeffs = argument(0);
8693 Node* zetas = argument(1);
8694
8695 coeffs = must_be_not_null(coeffs, true);
8696 zetas = must_be_not_null(zetas, true);
8697
8698 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8699 assert(coeffs_start, "coeffs is null");
8700 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
8701 assert(zetas_start, "inverseNtt_zetas is null");
8702 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8703 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
8704 stubAddr, stubName, TypePtr::BOTTOM,
8705 coeffs_start, zetas_start);
8706 // return an int
8707 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
8708 set_result(retvalue);
8709 return true;
8710 }
8711
8712 //------------------------------inline_dilithiumNttMult
8713 bool LibraryCallKit::inline_dilithiumNttMult() {
8714 address stubAddr;
8715 const char *stubName;
8716 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8717 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
8718
8719 stubAddr = StubRoutines::dilithiumNttMult();
8720 stubName = "dilithiumNttMult";
8721 if (!stubAddr) return false;
8722
8723 Node* result = argument(0);
8724 Node* ntta = argument(1);
8725 Node* nttb = argument(2);
8726 Node* zetas = argument(3);
8727
8728 result = must_be_not_null(result, true);
8729 ntta = must_be_not_null(ntta, true);
8730 nttb = must_be_not_null(nttb, true);
8731 zetas = must_be_not_null(zetas, true);
8732
8733 Node* result_start = array_element_address(result, intcon(0), T_INT);
8734 assert(result_start, "result is null");
8735 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
8736 assert(ntta_start, "ntta is null");
8737 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
8738 assert(nttb_start, "nttb is null");
8739 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8740 OptoRuntime::dilithiumNttMult_Type(),
8741 stubAddr, stubName, TypePtr::BOTTOM,
8742 result_start, ntta_start, nttb_start);
8743
8744 // return an int
8745 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8746 set_result(retvalue);
8747
8748 return true;
8749 }
8750
8751 //------------------------------inline_dilithiumMontMulByConstant
8752 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8753 address stubAddr;
8754 const char *stubName;
8755 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8756 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8757
8758 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8759 stubName = "dilithiumMontMulByConstant";
8760 if (!stubAddr) return false;
8761
8762 Node* coeffs = argument(0);
8763 Node* constant = argument(1);
8764
8765 coeffs = must_be_not_null(coeffs, true);
8766
8767 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8768 assert(coeffs_start, "coeffs is null");
8769 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8770 OptoRuntime::dilithiumMontMulByConstant_Type(),
8771 stubAddr, stubName, TypePtr::BOTTOM,
8772 coeffs_start, constant);
8773
8774 // return an int
8775 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8776 set_result(retvalue);
8777 return true;
8778 }
8779
8780
8781 //------------------------------inline_dilithiumDecomposePoly
8782 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8783 address stubAddr;
8784 const char *stubName;
8785 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8786 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8787
8788 stubAddr = StubRoutines::dilithiumDecomposePoly();
8789 stubName = "dilithiumDecomposePoly";
8790 if (!stubAddr) return false;
8791
8792 Node* input = argument(0);
8793 Node* lowPart = argument(1);
8794 Node* highPart = argument(2);
8795 Node* twoGamma2 = argument(3);
8796 Node* multiplier = argument(4);
8797
8798 input = must_be_not_null(input, true);
8799 lowPart = must_be_not_null(lowPart, true);
8800 highPart = must_be_not_null(highPart, true);
8801
8802 Node* input_start = array_element_address(input, intcon(0), T_INT);
8803 assert(input_start, "input is null");
8804 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8805 assert(lowPart_start, "lowPart is null");
8806 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8807 assert(highPart_start, "highPart is null");
8808
8809 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8810 OptoRuntime::dilithiumDecomposePoly_Type(),
8811 stubAddr, stubName, TypePtr::BOTTOM,
8812 input_start, lowPart_start, highPart_start,
8813 twoGamma2, multiplier);
8814
8815 // return an int
8816 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8817 set_result(retvalue);
8818 return true;
8819 }
8820
8821 bool LibraryCallKit::inline_base64_encodeBlock() {
8822 address stubAddr;
8823 const char *stubName;
8824 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8825 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8826 stubAddr = StubRoutines::base64_encodeBlock();
8827 stubName = "encodeBlock";
8828
8829 if (!stubAddr) return false;
8830 Node* base64obj = argument(0);
8831 Node* src = argument(1);
8832 Node* offset = argument(2);
8833 Node* len = argument(3);
8834 Node* dest = argument(4);
8835 Node* dp = argument(5);
8836 Node* isURL = argument(6);
8837
8838 src = must_be_not_null(src, true);
8839 dest = must_be_not_null(dest, true);
8840
8841 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8842 assert(src_start, "source array is null");
8843 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8844 assert(dest_start, "destination array is null");
8845
8846 Node* base64 = make_runtime_call(RC_LEAF,
8847 OptoRuntime::base64_encodeBlock_Type(),
8848 stubAddr, stubName, TypePtr::BOTTOM,
8849 src_start, offset, len, dest_start, dp, isURL);
8850 return true;
8851 }
8852
8853 bool LibraryCallKit::inline_base64_decodeBlock() {
8854 address stubAddr;
8855 const char *stubName;
8856 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8857 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8858 stubAddr = StubRoutines::base64_decodeBlock();
8859 stubName = "decodeBlock";
8860
8861 if (!stubAddr) return false;
8862 Node* base64obj = argument(0);
8863 Node* src = argument(1);
8864 Node* src_offset = argument(2);
8865 Node* len = argument(3);
8866 Node* dest = argument(4);
8867 Node* dest_offset = argument(5);
8868 Node* isURL = argument(6);
8869 Node* isMIME = argument(7);
8870
8871 src = must_be_not_null(src, true);
8872 dest = must_be_not_null(dest, true);
8873
8874 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8875 assert(src_start, "source array is null");
8876 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8877 assert(dest_start, "destination array is null");
8878
8879 Node* call = make_runtime_call(RC_LEAF,
8880 OptoRuntime::base64_decodeBlock_Type(),
8881 stubAddr, stubName, TypePtr::BOTTOM,
8882 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8883 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8884 set_result(result);
8885 return true;
8886 }
8887
8888 bool LibraryCallKit::inline_poly1305_processBlocks() {
8889 address stubAddr;
8890 const char *stubName;
8891 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8892 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8893 stubAddr = StubRoutines::poly1305_processBlocks();
8894 stubName = "poly1305_processBlocks";
8895
8896 if (!stubAddr) return false;
8897 null_check_receiver(); // null-check receiver
8898 if (stopped()) return true;
8899
8900 Node* input = argument(1);
8901 Node* input_offset = argument(2);
8902 Node* len = argument(3);
8903 Node* alimbs = argument(4);
8904 Node* rlimbs = argument(5);
8905
8906 input = must_be_not_null(input, true);
8907 alimbs = must_be_not_null(alimbs, true);
8908 rlimbs = must_be_not_null(rlimbs, true);
8909
8910 Node* input_start = array_element_address(input, input_offset, T_BYTE);
8911 assert(input_start, "input array is null");
8912 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
8913 assert(acc_start, "acc array is null");
8914 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
8915 assert(r_start, "r array is null");
8916
8917 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8918 OptoRuntime::poly1305_processBlocks_Type(),
8919 stubAddr, stubName, TypePtr::BOTTOM,
8920 input_start, len, acc_start, r_start);
8921 return true;
8922 }
8923
8924 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
8925 address stubAddr;
8926 const char *stubName;
8927 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8928 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
8929 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
8930 stubName = "intpoly_montgomeryMult_P256";
8931
8932 if (!stubAddr) return false;
8933 null_check_receiver(); // null-check receiver
8934 if (stopped()) return true;
8935
8936 Node* a = argument(1);
8937 Node* b = argument(2);
8938 Node* r = argument(3);
8939
8940 a = must_be_not_null(a, true);
8941 b = must_be_not_null(b, true);
8942 r = must_be_not_null(r, true);
8943
8944 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8945 assert(a_start, "a array is null");
8946 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8947 assert(b_start, "b array is null");
8948 Node* r_start = array_element_address(r, intcon(0), T_LONG);
8949 assert(r_start, "r array is null");
8950
8951 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8952 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
8953 stubAddr, stubName, TypePtr::BOTTOM,
8954 a_start, b_start, r_start);
8955 return true;
8956 }
8957
8958 bool LibraryCallKit::inline_intpoly_assign() {
8959 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8960 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
8961 const char *stubName = "intpoly_assign";
8962 address stubAddr = StubRoutines::intpoly_assign();
8963 if (!stubAddr) return false;
8964
8965 Node* set = argument(0);
8966 Node* a = argument(1);
8967 Node* b = argument(2);
8968 Node* arr_length = load_array_length(a);
8969
8970 a = must_be_not_null(a, true);
8971 b = must_be_not_null(b, true);
8972
8973 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8974 assert(a_start, "a array is null");
8975 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8976 assert(b_start, "b array is null");
8977
8978 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8979 OptoRuntime::intpoly_assign_Type(),
8980 stubAddr, stubName, TypePtr::BOTTOM,
8981 set, a_start, b_start, arr_length);
8982 return true;
8983 }
8984
8985 //------------------------------inline_digestBase_implCompress-----------------------
8986 //
8987 // Calculate MD5 for single-block byte[] array.
8988 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
8989 //
8990 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
8991 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
8992 //
8993 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
8994 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
8995 //
8996 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
8997 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
8998 //
8999 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
9000 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
9001 //
9002 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
9003 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
9004
9005 Node* digestBase_obj = argument(0);
9006 Node* src = argument(1); // type oop
9007 Node* ofs = argument(2); // type int
9008
9009 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
9010 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
9011 // failed array check
9012 return false;
9013 }
9014 // Figure out the size and type of the elements we will be copying.
9015 BasicType src_elem = src_type->elem()->array_element_basic_type();
9016 if (src_elem != T_BYTE) {
9017 return false;
9018 }
9019 // 'src_start' points to src array + offset
9020 src = must_be_not_null(src, true);
9021 Node* src_start = array_element_address(src, ofs, src_elem);
9022 Node* state = nullptr;
9023 Node* block_size = nullptr;
9024 address stubAddr;
9025 const char *stubName;
9026
9027 switch(id) {
9028 case vmIntrinsics::_md5_implCompress:
9029 assert(UseMD5Intrinsics, "need MD5 instruction support");
9030 state = get_state_from_digest_object(digestBase_obj, T_INT);
9031 stubAddr = StubRoutines::md5_implCompress();
9032 stubName = "md5_implCompress";
9033 break;
9034 case vmIntrinsics::_sha_implCompress:
9035 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
9036 state = get_state_from_digest_object(digestBase_obj, T_INT);
9037 stubAddr = StubRoutines::sha1_implCompress();
9038 stubName = "sha1_implCompress";
9039 break;
9040 case vmIntrinsics::_sha2_implCompress:
9041 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
9042 state = get_state_from_digest_object(digestBase_obj, T_INT);
9043 stubAddr = StubRoutines::sha256_implCompress();
9044 stubName = "sha256_implCompress";
9045 break;
9046 case vmIntrinsics::_sha5_implCompress:
9047 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
9048 state = get_state_from_digest_object(digestBase_obj, T_LONG);
9049 stubAddr = StubRoutines::sha512_implCompress();
9050 stubName = "sha512_implCompress";
9051 break;
9052 case vmIntrinsics::_sha3_implCompress:
9053 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
9054 state = get_state_from_digest_object(digestBase_obj, T_LONG);
9055 stubAddr = StubRoutines::sha3_implCompress();
9056 stubName = "sha3_implCompress";
9057 block_size = get_block_size_from_digest_object(digestBase_obj);
9058 if (block_size == nullptr) return false;
9059 break;
9060 default:
9061 fatal_unexpected_iid(id);
9062 return false;
9063 }
9064 if (state == nullptr) return false;
9065
9066 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
9067 if (stubAddr == nullptr) return false;
9068
9069 // Call the stub.
9070 Node* call;
9071 if (block_size == nullptr) {
9072 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
9073 stubAddr, stubName, TypePtr::BOTTOM,
9074 src_start, state);
9075 } else {
9076 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
9077 stubAddr, stubName, TypePtr::BOTTOM,
9078 src_start, state, block_size);
9079 }
9080
9081 return true;
9082 }
9083
9084 //------------------------------inline_double_keccak
9085 bool LibraryCallKit::inline_double_keccak() {
9086 address stubAddr;
9087 const char *stubName;
9088 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
9089 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
9090
9091 stubAddr = StubRoutines::double_keccak();
9092 stubName = "double_keccak";
9093 if (!stubAddr) return false;
9094
9095 Node* status0 = argument(0);
9096 Node* status1 = argument(1);
9097
9098 status0 = must_be_not_null(status0, true);
9099 status1 = must_be_not_null(status1, true);
9100
9101 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
9102 assert(status0_start, "status0 is null");
9103 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
9104 assert(status1_start, "status1 is null");
9105 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
9106 OptoRuntime::double_keccak_Type(),
9107 stubAddr, stubName, TypePtr::BOTTOM,
9108 status0_start, status1_start);
9109 // return an int
9110 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
9111 set_result(retvalue);
9112 return true;
9113 }
9114
9115
9116 //------------------------------inline_digestBase_implCompressMB-----------------------
9117 //
9118 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
9119 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
9120 //
9121 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
9122 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9123 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9124 assert((uint)predicate < 5, "sanity");
9125 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
9126
9127 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
9128 Node* src = argument(1); // byte[] array
9129 Node* ofs = argument(2); // type int
9130 Node* limit = argument(3); // type int
9131
9132 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
9133 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
9134 // failed array check
9135 return false;
9136 }
9137 // Figure out the size and type of the elements we will be copying.
9138 BasicType src_elem = src_type->elem()->array_element_basic_type();
9139 if (src_elem != T_BYTE) {
9140 return false;
9141 }
9142 // 'src_start' points to src array + offset
9143 src = must_be_not_null(src, false);
9144 Node* src_start = array_element_address(src, ofs, src_elem);
9145
9146 const char* klass_digestBase_name = nullptr;
9147 const char* stub_name = nullptr;
9148 address stub_addr = nullptr;
9149 BasicType elem_type = T_INT;
9150
9151 switch (predicate) {
9152 case 0:
9153 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
9154 klass_digestBase_name = "sun/security/provider/MD5";
9155 stub_name = "md5_implCompressMB";
9156 stub_addr = StubRoutines::md5_implCompressMB();
9157 }
9158 break;
9159 case 1:
9160 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
9161 klass_digestBase_name = "sun/security/provider/SHA";
9162 stub_name = "sha1_implCompressMB";
9163 stub_addr = StubRoutines::sha1_implCompressMB();
9164 }
9165 break;
9166 case 2:
9167 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
9168 klass_digestBase_name = "sun/security/provider/SHA2";
9169 stub_name = "sha256_implCompressMB";
9170 stub_addr = StubRoutines::sha256_implCompressMB();
9171 }
9172 break;
9173 case 3:
9174 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
9175 klass_digestBase_name = "sun/security/provider/SHA5";
9176 stub_name = "sha512_implCompressMB";
9177 stub_addr = StubRoutines::sha512_implCompressMB();
9178 elem_type = T_LONG;
9179 }
9180 break;
9181 case 4:
9182 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
9183 klass_digestBase_name = "sun/security/provider/SHA3";
9184 stub_name = "sha3_implCompressMB";
9185 stub_addr = StubRoutines::sha3_implCompressMB();
9186 elem_type = T_LONG;
9187 }
9188 break;
9189 default:
9190 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
9191 }
9192 if (klass_digestBase_name != nullptr) {
9193 assert(stub_addr != nullptr, "Stub is generated");
9194 if (stub_addr == nullptr) return false;
9195
9196 // get DigestBase klass to lookup for SHA klass
9197 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
9198 assert(tinst != nullptr, "digestBase_obj is not instance???");
9199 assert(tinst->is_loaded(), "DigestBase is not loaded");
9200
9201 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
9202 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
9203 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
9204 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
9205 }
9206 return false;
9207 }
9208
9209 //------------------------------inline_digestBase_implCompressMB-----------------------
9210 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
9211 BasicType elem_type, address stubAddr, const char *stubName,
9212 Node* src_start, Node* ofs, Node* limit) {
9213 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
9214 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
9215 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
9216 digest_obj = _gvn.transform(digest_obj);
9217
9218 Node* state = get_state_from_digest_object(digest_obj, elem_type);
9219 if (state == nullptr) return false;
9220
9221 Node* block_size = nullptr;
9222 if (strcmp("sha3_implCompressMB", stubName) == 0) {
9223 block_size = get_block_size_from_digest_object(digest_obj);
9224 if (block_size == nullptr) return false;
9225 }
9226
9227 // Call the stub.
9228 Node* call;
9229 if (block_size == nullptr) {
9230 call = make_runtime_call(RC_LEAF|RC_NO_FP,
9231 OptoRuntime::digestBase_implCompressMB_Type(false),
9232 stubAddr, stubName, TypePtr::BOTTOM,
9233 src_start, state, ofs, limit);
9234 } else {
9235 call = make_runtime_call(RC_LEAF|RC_NO_FP,
9236 OptoRuntime::digestBase_implCompressMB_Type(true),
9237 stubAddr, stubName, TypePtr::BOTTOM,
9238 src_start, state, block_size, ofs, limit);
9239 }
9240
9241 // return ofs (int)
9242 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
9243 set_result(result);
9244
9245 return true;
9246 }
9247
9248 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
9249 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
9250 assert(UseAES, "need AES instruction support");
9251 address stubAddr = nullptr;
9252 const char *stubName = nullptr;
9253 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
9254 stubName = "galoisCounterMode_AESCrypt";
9255
9256 if (stubAddr == nullptr) return false;
9257
9258 Node* in = argument(0);
9259 Node* inOfs = argument(1);
9260 Node* len = argument(2);
9261 Node* ct = argument(3);
9262 Node* ctOfs = argument(4);
9263 Node* out = argument(5);
9264 Node* outOfs = argument(6);
9265 Node* gctr_object = argument(7);
9266 Node* ghash_object = argument(8);
9267
9268 // (1) in, ct and out are arrays.
9269 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
9270 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
9271 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
9272 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
9273 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
9274 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
9275
9276 // checks are the responsibility of the caller
9277 Node* in_start = in;
9278 Node* ct_start = ct;
9279 Node* out_start = out;
9280 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
9281 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
9282 in_start = array_element_address(in, inOfs, T_BYTE);
9283 ct_start = array_element_address(ct, ctOfs, T_BYTE);
9284 out_start = array_element_address(out, outOfs, T_BYTE);
9285 }
9286
9287 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
9288 // (because of the predicated logic executed earlier).
9289 // so we cast it here safely.
9290 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
9291 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9292 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
9293 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
9294 Node* state = load_field_from_object(ghash_object, "state", "[J");
9295
9296 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
9297 return false;
9298 }
9299 // cast it to what we know it will be at runtime
9300 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
9301 assert(tinst != nullptr, "GCTR obj is null");
9302 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9303 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
9304 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
9305 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9306 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
9307 const TypeOopPtr* xtype = aklass->as_instance_type();
9308 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
9309 aescrypt_object = _gvn.transform(aescrypt_object);
9310 // we need to get the start of the aescrypt_object's expanded key array
9311 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
9312 if (k_start == nullptr) return false;
9313 // similarly, get the start address of the r vector
9314 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
9315 Node* state_start = array_element_address(state, intcon(0), T_LONG);
9316 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
9317
9318
9319 // Call the stub, passing params
9320 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
9321 OptoRuntime::galoisCounterMode_aescrypt_Type(),
9322 stubAddr, stubName, TypePtr::BOTTOM,
9323 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
9324
9325 // return cipher length (int)
9326 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
9327 set_result(retvalue);
9328
9329 return true;
9330 }
9331
9332 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
9333 // Return node representing slow path of predicate check.
9334 // the pseudo code we want to emulate with this predicate is:
9335 // for encryption:
9336 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
9337 // for decryption:
9338 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
9339 // note cipher==plain is more conservative than the original java code but that's OK
9340 //
9341
9342 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
9343 // The receiver was checked for null already.
9344 Node* objGCTR = argument(7);
9345 // Load embeddedCipher field of GCTR object.
9346 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9347 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
9348
9349 // get AESCrypt klass for instanceOf check
9350 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
9351 // will have same classloader as CipherBlockChaining object
9352 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
9353 assert(tinst != nullptr, "GCTR obj is null");
9354 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9355
9356 // we want to do an instanceof comparison against the AESCrypt class
9357 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
9358 if (!klass_AESCrypt->is_loaded()) {
9359 // if AESCrypt is not even loaded, we never take the intrinsic fast path
9360 Node* ctrl = control();
9361 set_control(top()); // no regular fast path
9362 return ctrl;
9363 }
9364
9365 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9366 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
9367 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9368 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9369 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9370
9371 return instof_false; // even if it is null
9372 }
9373
9374 //------------------------------get_state_from_digest_object-----------------------
9375 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
9376 const char* state_type;
9377 switch (elem_type) {
9378 case T_BYTE: state_type = "[B"; break;
9379 case T_INT: state_type = "[I"; break;
9380 case T_LONG: state_type = "[J"; break;
9381 default: ShouldNotReachHere();
9382 }
9383 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
9384 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
9385 if (digest_state == nullptr) return (Node *) nullptr;
9386
9387 // now have the array, need to get the start address of the state array
9388 Node* state = array_element_address(digest_state, intcon(0), elem_type);
9389 return state;
9390 }
9391
9392 //------------------------------get_block_size_from_sha3_object----------------------------------
9393 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
9394 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
9395 assert (block_size != nullptr, "sanity");
9396 return block_size;
9397 }
9398
9399 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
9400 // Return node representing slow path of predicate check.
9401 // the pseudo code we want to emulate with this predicate is:
9402 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
9403 //
9404 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
9405 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9406 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9407 assert((uint)predicate < 5, "sanity");
9408
9409 // The receiver was checked for null already.
9410 Node* digestBaseObj = argument(0);
9411
9412 // get DigestBase klass for instanceOf check
9413 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
9414 assert(tinst != nullptr, "digestBaseObj is null");
9415 assert(tinst->is_loaded(), "DigestBase is not loaded");
9416
9417 const char* klass_name = nullptr;
9418 switch (predicate) {
9419 case 0:
9420 if (UseMD5Intrinsics) {
9421 // we want to do an instanceof comparison against the MD5 class
9422 klass_name = "sun/security/provider/MD5";
9423 }
9424 break;
9425 case 1:
9426 if (UseSHA1Intrinsics) {
9427 // we want to do an instanceof comparison against the SHA class
9428 klass_name = "sun/security/provider/SHA";
9429 }
9430 break;
9431 case 2:
9432 if (UseSHA256Intrinsics) {
9433 // we want to do an instanceof comparison against the SHA2 class
9434 klass_name = "sun/security/provider/SHA2";
9435 }
9436 break;
9437 case 3:
9438 if (UseSHA512Intrinsics) {
9439 // we want to do an instanceof comparison against the SHA5 class
9440 klass_name = "sun/security/provider/SHA5";
9441 }
9442 break;
9443 case 4:
9444 if (UseSHA3Intrinsics) {
9445 // we want to do an instanceof comparison against the SHA3 class
9446 klass_name = "sun/security/provider/SHA3";
9447 }
9448 break;
9449 default:
9450 fatal("unknown SHA intrinsic predicate: %d", predicate);
9451 }
9452
9453 ciKlass* klass = nullptr;
9454 if (klass_name != nullptr) {
9455 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
9456 }
9457 if ((klass == nullptr) || !klass->is_loaded()) {
9458 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
9459 Node* ctrl = control();
9460 set_control(top()); // no intrinsic path
9461 return ctrl;
9462 }
9463 ciInstanceKlass* instklass = klass->as_instance_klass();
9464
9465 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
9466 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9467 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9468 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9469
9470 return instof_false; // even if it is null
9471 }
9472
9473 //-------------inline_fma-----------------------------------
9474 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
9475 Node *a = nullptr;
9476 Node *b = nullptr;
9477 Node *c = nullptr;
9478 Node* result = nullptr;
9479 switch (id) {
9480 case vmIntrinsics::_fmaD:
9481 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
9482 // no receiver since it is static method
9483 a = argument(0);
9484 b = argument(2);
9485 c = argument(4);
9486 result = _gvn.transform(new FmaDNode(a, b, c));
9487 break;
9488 case vmIntrinsics::_fmaF:
9489 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
9490 a = argument(0);
9491 b = argument(1);
9492 c = argument(2);
9493 result = _gvn.transform(new FmaFNode(a, b, c));
9494 break;
9495 default:
9496 fatal_unexpected_iid(id); break;
9497 }
9498 set_result(result);
9499 return true;
9500 }
9501
9502 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
9503 // argument(0) is receiver
9504 Node* codePoint = argument(1);
9505 Node* n = nullptr;
9506
9507 switch (id) {
9508 case vmIntrinsics::_isDigit :
9509 n = new DigitNode(control(), codePoint);
9510 break;
9511 case vmIntrinsics::_isLowerCase :
9512 n = new LowerCaseNode(control(), codePoint);
9513 break;
9514 case vmIntrinsics::_isUpperCase :
9515 n = new UpperCaseNode(control(), codePoint);
9516 break;
9517 case vmIntrinsics::_isWhitespace :
9518 n = new WhitespaceNode(control(), codePoint);
9519 break;
9520 default:
9521 fatal_unexpected_iid(id);
9522 }
9523
9524 set_result(_gvn.transform(n));
9525 return true;
9526 }
9527
9528 bool LibraryCallKit::inline_profileBoolean() {
9529 Node* counts = argument(1);
9530 const TypeAryPtr* ary = nullptr;
9531 ciArray* aobj = nullptr;
9532 if (counts->is_Con()
9533 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
9534 && (aobj = ary->const_oop()->as_array()) != nullptr
9535 && (aobj->length() == 2)) {
9536 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
9537 jint false_cnt = aobj->element_value(0).as_int();
9538 jint true_cnt = aobj->element_value(1).as_int();
9539
9540 if (C->log() != nullptr) {
9541 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
9542 false_cnt, true_cnt);
9543 }
9544
9545 if (false_cnt + true_cnt == 0) {
9546 // According to profile, never executed.
9547 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9548 Deoptimization::Action_reinterpret);
9549 return true;
9550 }
9551
9552 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
9553 // is a number of each value occurrences.
9554 Node* result = argument(0);
9555 if (false_cnt == 0 || true_cnt == 0) {
9556 // According to profile, one value has been never seen.
9557 int expected_val = (false_cnt == 0) ? 1 : 0;
9558
9559 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
9560 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
9561
9562 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
9563 Node* fast_path = _gvn.transform(new IfTrueNode(check));
9564 Node* slow_path = _gvn.transform(new IfFalseNode(check));
9565
9566 { // Slow path: uncommon trap for never seen value and then reexecute
9567 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
9568 // the value has been seen at least once.
9569 PreserveJVMState pjvms(this);
9570 PreserveReexecuteState preexecs(this);
9571 jvms()->set_should_reexecute(true);
9572
9573 set_control(slow_path);
9574 set_i_o(i_o());
9575
9576 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9577 Deoptimization::Action_reinterpret);
9578 }
9579 // The guard for never seen value enables sharpening of the result and
9580 // returning a constant. It allows to eliminate branches on the same value
9581 // later on.
9582 set_control(fast_path);
9583 result = intcon(expected_val);
9584 }
9585 // Stop profiling.
9586 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
9587 // By replacing method body with profile data (represented as ProfileBooleanNode
9588 // on IR level) we effectively disable profiling.
9589 // It enables full speed execution once optimized code is generated.
9590 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
9591 C->record_for_igvn(profile);
9592 set_result(profile);
9593 return true;
9594 } else {
9595 // Continue profiling.
9596 // Profile data isn't available at the moment. So, execute method's bytecode version.
9597 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
9598 // is compiled and counters aren't available since corresponding MethodHandle
9599 // isn't a compile-time constant.
9600 return false;
9601 }
9602 }
9603
9604 bool LibraryCallKit::inline_isCompileConstant() {
9605 Node* n = argument(0);
9606 set_result(n->is_Con() ? intcon(1) : intcon(0));
9607 return true;
9608 }
9609
9610 //------------------------------- inline_getObjectSize --------------------------------------
9611 //
9612 // Calculate the runtime size of the object/array.
9613 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
9614 //
9615 bool LibraryCallKit::inline_getObjectSize() {
9616 Node* obj = argument(3);
9617 Node* klass_node = load_object_klass(obj);
9618
9619 jint layout_con = Klass::_lh_neutral_value;
9620 Node* layout_val = get_layout_helper(klass_node, layout_con);
9621 int layout_is_con = (layout_val == nullptr);
9622
9623 if (layout_is_con) {
9624 // Layout helper is constant, can figure out things at compile time.
9625
9626 if (Klass::layout_helper_is_instance(layout_con)) {
9627 // Instance case: layout_con contains the size itself.
9628 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
9629 set_result(size);
9630 } else {
9631 // Array case: size is round(header + element_size*arraylength).
9632 // Since arraylength is different for every array instance, we have to
9633 // compute the whole thing at runtime.
9634
9635 Node* arr_length = load_array_length(obj);
9636
9637 int round_mask = MinObjAlignmentInBytes - 1;
9638 int hsize = Klass::layout_helper_header_size(layout_con);
9639 int eshift = Klass::layout_helper_log2_element_size(layout_con);
9640
9641 if ((round_mask & ~right_n_bits(eshift)) == 0) {
9642 round_mask = 0; // strength-reduce it if it goes away completely
9643 }
9644 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
9645 Node* header_size = intcon(hsize + round_mask);
9646
9647 Node* lengthx = ConvI2X(arr_length);
9648 Node* headerx = ConvI2X(header_size);
9649
9650 Node* abody = lengthx;
9651 if (eshift != 0) {
9652 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
9653 }
9654 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
9655 if (round_mask != 0) {
9656 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
9657 }
9658 size = ConvX2L(size);
9659 set_result(size);
9660 }
9661 } else {
9662 // Layout helper is not constant, need to test for array-ness at runtime.
9663
9664 enum { _instance_path = 1, _array_path, PATH_LIMIT };
9665 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
9666 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
9667 record_for_igvn(result_reg);
9668
9669 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
9670 if (array_ctl != nullptr) {
9671 // Array case: size is round(header + element_size*arraylength).
9672 // Since arraylength is different for every array instance, we have to
9673 // compute the whole thing at runtime.
9674
9675 PreserveJVMState pjvms(this);
9676 set_control(array_ctl);
9677 Node* arr_length = load_array_length(obj);
9678
9679 int round_mask = MinObjAlignmentInBytes - 1;
9680 Node* mask = intcon(round_mask);
9681
9682 Node* hss = intcon(Klass::_lh_header_size_shift);
9683 Node* hsm = intcon(Klass::_lh_header_size_mask);
9684 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
9685 header_size = _gvn.transform(new AndINode(header_size, hsm));
9686 header_size = _gvn.transform(new AddINode(header_size, mask));
9687
9688 // There is no need to mask or shift this value.
9689 // The semantics of LShiftINode include an implicit mask to 0x1F.
9690 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
9691 Node* elem_shift = layout_val;
9692
9693 Node* lengthx = ConvI2X(arr_length);
9694 Node* headerx = ConvI2X(header_size);
9695
9696 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
9697 Node* size = _gvn.transform(new AddXNode(headerx, abody));
9698 if (round_mask != 0) {
9699 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
9700 }
9701 size = ConvX2L(size);
9702
9703 result_reg->init_req(_array_path, control());
9704 result_val->init_req(_array_path, size);
9705 }
9706
9707 if (!stopped()) {
9708 // Instance case: the layout helper gives us instance size almost directly,
9709 // but we need to mask out the _lh_instance_slow_path_bit.
9710 Node* size = ConvI2X(layout_val);
9711 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
9712 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
9713 size = _gvn.transform(new AndXNode(size, mask));
9714 size = ConvX2L(size);
9715
9716 result_reg->init_req(_instance_path, control());
9717 result_val->init_req(_instance_path, size);
9718 }
9719
9720 set_result(result_reg, result_val);
9721 }
9722
9723 return true;
9724 }
9725
9726 //------------------------------- inline_blackhole --------------------------------------
9727 //
9728 // Make sure all arguments to this node are alive.
9729 // This matches methods that were requested to be blackholed through compile commands.
9730 //
9731 bool LibraryCallKit::inline_blackhole() {
9732 assert(callee()->is_static(), "Should have been checked before: only static methods here");
9733 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
9734 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
9735
9736 // Blackhole node pinches only the control, not memory. This allows
9737 // the blackhole to be pinned in the loop that computes blackholed
9738 // values, but have no other side effects, like breaking the optimizations
9739 // across the blackhole.
9740
9741 Node* bh = _gvn.transform(new BlackholeNode(control()));
9742 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
9743
9744 // Bind call arguments as blackhole arguments to keep them alive
9745 uint nargs = callee()->arg_size();
9746 for (uint i = 0; i < nargs; i++) {
9747 bh->add_req(argument(i));
9748 }
9749
9750 return true;
9751 }
9752
9753 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9754 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9755 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9756 return nullptr; // box klass is not Float16
9757 }
9758
9759 // Null check; get notnull casted pointer
9760 Node* null_ctl = top();
9761 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9762 // If not_null_box is dead, only null-path is taken
9763 if (stopped()) {
9764 set_control(null_ctl);
9765 return nullptr;
9766 }
9767 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9768 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9769 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9770 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9771 }
9772
9773 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9774 PreserveReexecuteState preexecs(this);
9775 jvms()->set_should_reexecute(true);
9776
9777 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9778 Node* klass_node = makecon(klass_type);
9779 Node* box = new_instance(klass_node);
9780
9781 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9782 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9783
9784 Node* field_store = _gvn.transform(access_store_at(box,
9785 value_field,
9786 value_adr_type,
9787 value,
9788 TypeInt::SHORT,
9789 T_SHORT,
9790 IN_HEAP));
9791 set_memory(field_store, value_adr_type);
9792 return box;
9793 }
9794
9795 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9796 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9797 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9798 return false;
9799 }
9800
9801 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9802 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9803 return false;
9804 }
9805
9806 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9807 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9808 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9809 ciSymbols::short_signature(),
9810 false);
9811 assert(field != nullptr, "");
9812
9813 // Transformed nodes
9814 Node* fld1 = nullptr;
9815 Node* fld2 = nullptr;
9816 Node* fld3 = nullptr;
9817 switch(num_args) {
9818 case 3:
9819 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9820 if (fld3 == nullptr) {
9821 return false;
9822 }
9823 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9824 // fall-through
9825 case 2:
9826 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9827 if (fld2 == nullptr) {
9828 return false;
9829 }
9830 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9831 // fall-through
9832 case 1:
9833 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9834 if (fld1 == nullptr) {
9835 return false;
9836 }
9837 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9838 break;
9839 default: fatal("Unsupported number of arguments %d", num_args);
9840 }
9841
9842 Node* result = nullptr;
9843 switch (id) {
9844 // Unary operations
9845 case vmIntrinsics::_sqrt_float16:
9846 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9847 break;
9848 // Ternary operations
9849 case vmIntrinsics::_fma_float16:
9850 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9851 break;
9852 default:
9853 fatal_unexpected_iid(id);
9854 break;
9855 }
9856 result = _gvn.transform(new ReinterpretHF2SNode(result));
9857 set_result(box_fp16_value(float16_box_type, field, result));
9858 return true;
9859 }
9860