1 /*
2 * Copyright (c) 1999, 2026, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "asm/macroAssembler.hpp"
26 #include "ci/ciArrayKlass.hpp"
27 #include "ci/ciFlatArrayKlass.hpp"
28 #include "ci/ciInstanceKlass.hpp"
29 #include "ci/ciSymbols.hpp"
30 #include "ci/ciUtilities.inline.hpp"
31 #include "classfile/vmIntrinsics.hpp"
32 #include "compiler/compileBroker.hpp"
33 #include "compiler/compileLog.hpp"
34 #include "gc/shared/barrierSet.hpp"
35 #include "gc/shared/c2/barrierSetC2.hpp"
36 #include "jfr/support/jfrIntrinsics.hpp"
37 #include "memory/resourceArea.hpp"
38 #include "oops/accessDecorators.hpp"
39 #include "oops/klass.inline.hpp"
40 #include "oops/layoutKind.hpp"
41 #include "oops/objArrayKlass.hpp"
42 #include "opto/addnode.hpp"
43 #include "opto/arraycopynode.hpp"
44 #include "opto/c2compiler.hpp"
45 #include "opto/castnode.hpp"
46 #include "opto/cfgnode.hpp"
47 #include "opto/convertnode.hpp"
48 #include "opto/countbitsnode.hpp"
49 #include "opto/graphKit.hpp"
50 #include "opto/idealKit.hpp"
51 #include "opto/inlinetypenode.hpp"
52 #include "opto/library_call.hpp"
53 #include "opto/mathexactnode.hpp"
54 #include "opto/mulnode.hpp"
55 #include "opto/narrowptrnode.hpp"
56 #include "opto/opaquenode.hpp"
57 #include "opto/opcodes.hpp"
58 #include "opto/parse.hpp"
59 #include "opto/rootnode.hpp"
60 #include "opto/runtime.hpp"
61 #include "opto/subnode.hpp"
62 #include "opto/type.hpp"
63 #include "opto/vectornode.hpp"
64 #include "prims/jvmtiExport.hpp"
65 #include "prims/jvmtiThreadState.hpp"
66 #include "prims/unsafe.hpp"
67 #include "runtime/globals.hpp"
68 #include "runtime/jniHandles.inline.hpp"
69 #include "runtime/mountUnmountDisabler.hpp"
70 #include "runtime/objectMonitor.hpp"
71 #include "runtime/sharedRuntime.hpp"
72 #include "runtime/stubRoutines.hpp"
73 #include "utilities/globalDefinitions.hpp"
74 #include "utilities/macros.hpp"
75 #include "utilities/powerOfTwo.hpp"
76
77 //---------------------------make_vm_intrinsic----------------------------
78 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
79 vmIntrinsicID id = m->intrinsic_id();
80 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
81
82 if (!m->is_loaded()) {
83 // Do not attempt to inline unloaded methods.
84 return nullptr;
85 }
86
87 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
88 bool is_available = false;
89
90 {
91 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
92 // the compiler must transition to '_thread_in_vm' state because both
93 // methods access VM-internal data.
94 VM_ENTRY_MARK;
95 methodHandle mh(THREAD, m->get_Method());
96 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
97 if (is_available && is_virtual) {
98 is_available = vmIntrinsics::does_virtual_dispatch(id);
99 }
100 }
101
102 if (is_available) {
103 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
104 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
105 return new LibraryIntrinsic(m, is_virtual,
106 vmIntrinsics::predicates_needed(id),
107 vmIntrinsics::does_virtual_dispatch(id),
108 id);
109 } else {
110 return nullptr;
111 }
112 }
113
114 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
115 LibraryCallKit kit(jvms, this);
116 Compile* C = kit.C;
117 int nodes = C->unique();
118 #ifndef PRODUCT
119 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
120 char buf[1000];
121 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
122 tty->print_cr("Intrinsic %s", str);
123 }
124 #endif
125 ciMethod* callee = kit.callee();
126 const int bci = kit.bci();
127 #ifdef ASSERT
128 Node* ctrl = kit.control();
129 #endif
130 // Try to inline the intrinsic.
131 if (callee->check_intrinsic_candidate() &&
132 kit.try_to_inline(_last_predicate)) {
133 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
134 : "(intrinsic)";
135 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
136 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
137 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
138 if (C->log()) {
139 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
140 vmIntrinsics::name_at(intrinsic_id()),
141 (is_virtual() ? " virtual='1'" : ""),
142 C->unique() - nodes);
143 }
144 // Push the result from the inlined method onto the stack.
145 kit.push_result();
146 return kit.transfer_exceptions_into_jvms();
147 }
148
149 // The intrinsic bailed out
150 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
151 assert(jvms->map() == kit.map(), "Out of sync JVM state");
152 if (jvms->has_method()) {
153 // Not a root compile.
154 const char* msg;
155 if (callee->intrinsic_candidate()) {
156 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
157 } else {
158 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
159 : "failed to inline (intrinsic), method not annotated";
160 }
161 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
162 C->inline_printer()->record(callee, jvms, InliningResult::FAILURE, msg);
163 } else {
164 // Root compile
165 ResourceMark rm;
166 stringStream msg_stream;
167 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
168 vmIntrinsics::name_at(intrinsic_id()),
169 is_virtual() ? " (virtual)" : "", bci);
170 const char *msg = msg_stream.freeze();
171 log_debug(jit, inlining)("%s", msg);
172 if (C->print_intrinsics() || C->print_inlining()) {
173 tty->print("%s", msg);
174 }
175 }
176 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
177
178 return nullptr;
179 }
180
181 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
182 LibraryCallKit kit(jvms, this);
183 Compile* C = kit.C;
184 int nodes = C->unique();
185 _last_predicate = predicate;
186 #ifndef PRODUCT
187 assert(is_predicated() && predicate < predicates_count(), "sanity");
188 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
189 char buf[1000];
190 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
191 tty->print_cr("Predicate for intrinsic %s", str);
192 }
193 #endif
194 ciMethod* callee = kit.callee();
195 const int bci = kit.bci();
196
197 Node* slow_ctl = kit.try_to_predicate(predicate);
198 if (!kit.failing()) {
199 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
200 : "(intrinsic, predicate)";
201 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
202 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
203
204 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
205 if (C->log()) {
206 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
207 vmIntrinsics::name_at(intrinsic_id()),
208 (is_virtual() ? " virtual='1'" : ""),
209 C->unique() - nodes);
210 }
211 return slow_ctl; // Could be null if the check folds.
212 }
213
214 // The intrinsic bailed out
215 if (jvms->has_method()) {
216 // Not a root compile.
217 const char* msg = "failed to generate predicate for intrinsic";
218 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
219 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
220 } else {
221 // Root compile
222 ResourceMark rm;
223 stringStream msg_stream;
224 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
225 vmIntrinsics::name_at(intrinsic_id()),
226 is_virtual() ? " (virtual)" : "", bci);
227 const char *msg = msg_stream.freeze();
228 log_debug(jit, inlining)("%s", msg);
229 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
230 }
231 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
232 return nullptr;
233 }
234
235 bool LibraryCallKit::try_to_inline(int predicate) {
236 // Handle symbolic names for otherwise undistinguished boolean switches:
237 const bool is_store = true;
238 const bool is_compress = true;
239 const bool is_static = true;
240 const bool is_volatile = true;
241
242 if (!jvms()->has_method()) {
243 // Root JVMState has a null method.
244 assert(map()->memory()->Opcode() == Op_Parm, "");
245 // Insert the memory aliasing node
246 set_all_memory(reset_memory());
247 }
248 assert(merged_memory(), "");
249
250 switch (intrinsic_id()) {
251 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
252 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
253 case vmIntrinsics::_getClass: return inline_native_getClass();
254
255 case vmIntrinsics::_ceil:
256 case vmIntrinsics::_floor:
257 case vmIntrinsics::_rint:
258 case vmIntrinsics::_dsin:
259 case vmIntrinsics::_dcos:
260 case vmIntrinsics::_dtan:
261 case vmIntrinsics::_dsinh:
262 case vmIntrinsics::_dtanh:
263 case vmIntrinsics::_dcbrt:
264 case vmIntrinsics::_dabs:
265 case vmIntrinsics::_fabs:
266 case vmIntrinsics::_iabs:
267 case vmIntrinsics::_labs:
268 case vmIntrinsics::_datan2:
269 case vmIntrinsics::_dsqrt:
270 case vmIntrinsics::_dsqrt_strict:
271 case vmIntrinsics::_dexp:
272 case vmIntrinsics::_dlog:
273 case vmIntrinsics::_dlog10:
274 case vmIntrinsics::_dpow:
275 case vmIntrinsics::_dcopySign:
276 case vmIntrinsics::_fcopySign:
277 case vmIntrinsics::_dsignum:
278 case vmIntrinsics::_roundF:
279 case vmIntrinsics::_roundD:
280 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
281
282 case vmIntrinsics::_notify:
283 case vmIntrinsics::_notifyAll:
284 return inline_notify(intrinsic_id());
285
286 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
287 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
288 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
289 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
290 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
291 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
292 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
293 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
294 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
295 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
296 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
297 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
298 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
299 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
300
301 case vmIntrinsics::_arraycopy: return inline_arraycopy();
302
303 case vmIntrinsics::_arraySort: return inline_array_sort();
304 case vmIntrinsics::_arrayPartition: return inline_array_partition();
305
306 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
307 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
308 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
309 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
310
311 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
312 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
313 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
314 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
315 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
316 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
317 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
318 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
319
320 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
321
322 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
323
324 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
325 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
326 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
327 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
328
329 case vmIntrinsics::_compressStringC:
330 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
331 case vmIntrinsics::_inflateStringC:
332 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
333
334 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
335 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
336 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
337 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
338 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
339 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
340 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
341 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
342 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
343
344 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
345 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
346 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
347 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
348 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
349 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
350 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
351 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
352 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
353
354 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
355 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
356 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
357 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
358 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
359 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
360 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
361 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
362 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
363
364 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
365 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
366 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
367 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
368 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
369 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
370 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
371 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
372 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
373
374 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
375 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
376 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
377 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
378
379 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
380 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
381 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
382 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
383
384 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
385 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
386 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
387 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
388 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
389 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
390 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
391 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
392 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
393
394 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
395 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
396 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
397 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
398 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
399 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
400 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
401 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
402 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
403
404 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
405 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
406 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
407 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
408 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
409 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
410 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
411 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
412 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
413
414 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
415 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
416 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
417 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
418 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
419 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
420 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
421 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
422 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
423
424 case vmIntrinsics::_getFlatValue: return inline_unsafe_flat_access(!is_store, Relaxed);
425 case vmIntrinsics::_putFlatValue: return inline_unsafe_flat_access( is_store, Relaxed);
426
427 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
428 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
429 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
430 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
431 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
432
433 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
434 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
435 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
436 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
437 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
438 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
439 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
440 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
441 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
442 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
443 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
444 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
445 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
446 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
447 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
448 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
449 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
450 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
451 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
452 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
453
454 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
455 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
456 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
457 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
458 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
459 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
460 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
461 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
462 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
463 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
464 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
465 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
466 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
467 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
468 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
469
470 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
471 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
472 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
473 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
474
475 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
476 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
477 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
478 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
479 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
480
481 case vmIntrinsics::_loadFence:
482 case vmIntrinsics::_storeFence:
483 case vmIntrinsics::_storeStoreFence:
484 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
485
486 case vmIntrinsics::_arrayInstanceBaseOffset: return inline_arrayInstanceBaseOffset();
487 case vmIntrinsics::_arrayInstanceIndexScale: return inline_arrayInstanceIndexScale();
488 case vmIntrinsics::_arrayLayout: return inline_arrayLayout();
489 case vmIntrinsics::_getFieldMap: return inline_getFieldMap();
490
491 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
492
493 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
494 case vmIntrinsics::_currentThread: return inline_native_currentThread();
495 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
496
497 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
498 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
499
500 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
501 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
502
503 case vmIntrinsics::_vthreadEndFirstTransition: return inline_native_vthread_end_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_end_first_transition_Java()),
504 "endFirstTransition", true);
505 case vmIntrinsics::_vthreadStartFinalTransition: return inline_native_vthread_start_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_start_final_transition_Java()),
506 "startFinalTransition", true);
507 case vmIntrinsics::_vthreadStartTransition: return inline_native_vthread_start_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_start_transition_Java()),
508 "startTransition", false);
509 case vmIntrinsics::_vthreadEndTransition: return inline_native_vthread_end_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_end_transition_Java()),
510 "endTransition", false);
511 #if INCLUDE_JVMTI
512 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
513 #endif
514
515 #ifdef JFR_HAVE_INTRINSICS
516 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
517 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
518 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
519 #endif
520 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
521 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
522 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
523 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
524 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
525 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
526 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
527 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
528 case vmIntrinsics::_getLength: return inline_native_getLength();
529 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
530 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
531 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
532 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
533 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
534 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
535 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
536
537 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
538 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
539 case vmIntrinsics::_newNullRestrictedNonAtomicArray: return inline_newArray(/* null_free */ true, /* atomic */ false);
540 case vmIntrinsics::_newNullRestrictedAtomicArray: return inline_newArray(/* null_free */ true, /* atomic */ true);
541 case vmIntrinsics::_newNullableAtomicArray: return inline_newArray(/* null_free */ false, /* atomic */ true);
542 case vmIntrinsics::_isFlatArray: return inline_getArrayProperties(IsFlat);
543 case vmIntrinsics::_isNullRestrictedArray: return inline_getArrayProperties(IsNullRestricted);
544 case vmIntrinsics::_isAtomicArray: return inline_getArrayProperties(IsAtomic);
545
546 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
547
548 case vmIntrinsics::_isInstance:
549 case vmIntrinsics::_isHidden:
550 case vmIntrinsics::_getSuperclass: return inline_native_Class_query(intrinsic_id());
551
552 case vmIntrinsics::_floatToRawIntBits:
553 case vmIntrinsics::_floatToIntBits:
554 case vmIntrinsics::_intBitsToFloat:
555 case vmIntrinsics::_doubleToRawLongBits:
556 case vmIntrinsics::_doubleToLongBits:
557 case vmIntrinsics::_longBitsToDouble:
558 case vmIntrinsics::_floatToFloat16:
559 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
560 case vmIntrinsics::_sqrt_float16: return inline_fp16_operations(intrinsic_id(), 1);
561 case vmIntrinsics::_fma_float16: return inline_fp16_operations(intrinsic_id(), 3);
562 case vmIntrinsics::_floatIsFinite:
563 case vmIntrinsics::_floatIsInfinite:
564 case vmIntrinsics::_doubleIsFinite:
565 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
566
567 case vmIntrinsics::_numberOfLeadingZeros_i:
568 case vmIntrinsics::_numberOfLeadingZeros_l:
569 case vmIntrinsics::_numberOfTrailingZeros_i:
570 case vmIntrinsics::_numberOfTrailingZeros_l:
571 case vmIntrinsics::_bitCount_i:
572 case vmIntrinsics::_bitCount_l:
573 case vmIntrinsics::_reverse_i:
574 case vmIntrinsics::_reverse_l:
575 case vmIntrinsics::_reverseBytes_i:
576 case vmIntrinsics::_reverseBytes_l:
577 case vmIntrinsics::_reverseBytes_s:
578 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
579
580 case vmIntrinsics::_compress_i:
581 case vmIntrinsics::_compress_l:
582 case vmIntrinsics::_expand_i:
583 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
584
585 case vmIntrinsics::_compareUnsigned_i:
586 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
587
588 case vmIntrinsics::_divideUnsigned_i:
589 case vmIntrinsics::_divideUnsigned_l:
590 case vmIntrinsics::_remainderUnsigned_i:
591 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
592
593 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
594
595 case vmIntrinsics::_Reference_get0: return inline_reference_get0();
596 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
597 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
598 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
599 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
600
601 case vmIntrinsics::_Class_cast: return inline_Class_cast();
602
603 case vmIntrinsics::_aescrypt_encryptBlock:
604 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
605
606 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
607 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
608 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
609
610 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
611 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
612 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
613
614 case vmIntrinsics::_counterMode_AESCrypt:
615 return inline_counterMode_AESCrypt(intrinsic_id());
616
617 case vmIntrinsics::_galoisCounterMode_AESCrypt:
618 return inline_galoisCounterMode_AESCrypt();
619
620 case vmIntrinsics::_md5_implCompress:
621 case vmIntrinsics::_sha_implCompress:
622 case vmIntrinsics::_sha2_implCompress:
623 case vmIntrinsics::_sha5_implCompress:
624 case vmIntrinsics::_sha3_implCompress:
625 return inline_digestBase_implCompress(intrinsic_id());
626 case vmIntrinsics::_double_keccak:
627 return inline_double_keccak();
628
629 case vmIntrinsics::_digestBase_implCompressMB:
630 return inline_digestBase_implCompressMB(predicate);
631
632 case vmIntrinsics::_multiplyToLen:
633 return inline_multiplyToLen();
634
635 case vmIntrinsics::_squareToLen:
636 return inline_squareToLen();
637
638 case vmIntrinsics::_mulAdd:
639 return inline_mulAdd();
640
641 case vmIntrinsics::_montgomeryMultiply:
642 return inline_montgomeryMultiply();
643 case vmIntrinsics::_montgomerySquare:
644 return inline_montgomerySquare();
645
646 case vmIntrinsics::_bigIntegerRightShiftWorker:
647 return inline_bigIntegerShift(true);
648 case vmIntrinsics::_bigIntegerLeftShiftWorker:
649 return inline_bigIntegerShift(false);
650
651 case vmIntrinsics::_vectorizedMismatch:
652 return inline_vectorizedMismatch();
653
654 case vmIntrinsics::_ghash_processBlocks:
655 return inline_ghash_processBlocks();
656 case vmIntrinsics::_chacha20Block:
657 return inline_chacha20Block();
658 case vmIntrinsics::_kyberNtt:
659 return inline_kyberNtt();
660 case vmIntrinsics::_kyberInverseNtt:
661 return inline_kyberInverseNtt();
662 case vmIntrinsics::_kyberNttMult:
663 return inline_kyberNttMult();
664 case vmIntrinsics::_kyberAddPoly_2:
665 return inline_kyberAddPoly_2();
666 case vmIntrinsics::_kyberAddPoly_3:
667 return inline_kyberAddPoly_3();
668 case vmIntrinsics::_kyber12To16:
669 return inline_kyber12To16();
670 case vmIntrinsics::_kyberBarrettReduce:
671 return inline_kyberBarrettReduce();
672 case vmIntrinsics::_dilithiumAlmostNtt:
673 return inline_dilithiumAlmostNtt();
674 case vmIntrinsics::_dilithiumAlmostInverseNtt:
675 return inline_dilithiumAlmostInverseNtt();
676 case vmIntrinsics::_dilithiumNttMult:
677 return inline_dilithiumNttMult();
678 case vmIntrinsics::_dilithiumMontMulByConstant:
679 return inline_dilithiumMontMulByConstant();
680 case vmIntrinsics::_dilithiumDecomposePoly:
681 return inline_dilithiumDecomposePoly();
682 case vmIntrinsics::_base64_encodeBlock:
683 return inline_base64_encodeBlock();
684 case vmIntrinsics::_base64_decodeBlock:
685 return inline_base64_decodeBlock();
686 case vmIntrinsics::_poly1305_processBlocks:
687 return inline_poly1305_processBlocks();
688 case vmIntrinsics::_intpoly_montgomeryMult_P256:
689 return inline_intpoly_montgomeryMult_P256();
690 case vmIntrinsics::_intpoly_assign:
691 return inline_intpoly_assign();
692 case vmIntrinsics::_encodeISOArray:
693 case vmIntrinsics::_encodeByteISOArray:
694 return inline_encodeISOArray(false);
695 case vmIntrinsics::_encodeAsciiArray:
696 return inline_encodeISOArray(true);
697
698 case vmIntrinsics::_updateCRC32:
699 return inline_updateCRC32();
700 case vmIntrinsics::_updateBytesCRC32:
701 return inline_updateBytesCRC32();
702 case vmIntrinsics::_updateByteBufferCRC32:
703 return inline_updateByteBufferCRC32();
704
705 case vmIntrinsics::_updateBytesCRC32C:
706 return inline_updateBytesCRC32C();
707 case vmIntrinsics::_updateDirectByteBufferCRC32C:
708 return inline_updateDirectByteBufferCRC32C();
709
710 case vmIntrinsics::_updateBytesAdler32:
711 return inline_updateBytesAdler32();
712 case vmIntrinsics::_updateByteBufferAdler32:
713 return inline_updateByteBufferAdler32();
714
715 case vmIntrinsics::_profileBoolean:
716 return inline_profileBoolean();
717 case vmIntrinsics::_isCompileConstant:
718 return inline_isCompileConstant();
719
720 case vmIntrinsics::_countPositives:
721 return inline_countPositives();
722
723 case vmIntrinsics::_fmaD:
724 case vmIntrinsics::_fmaF:
725 return inline_fma(intrinsic_id());
726
727 case vmIntrinsics::_isDigit:
728 case vmIntrinsics::_isLowerCase:
729 case vmIntrinsics::_isUpperCase:
730 case vmIntrinsics::_isWhitespace:
731 return inline_character_compare(intrinsic_id());
732
733 case vmIntrinsics::_min:
734 case vmIntrinsics::_max:
735 case vmIntrinsics::_min_strict:
736 case vmIntrinsics::_max_strict:
737 case vmIntrinsics::_minL:
738 case vmIntrinsics::_maxL:
739 case vmIntrinsics::_minF:
740 case vmIntrinsics::_maxF:
741 case vmIntrinsics::_minD:
742 case vmIntrinsics::_maxD:
743 case vmIntrinsics::_minF_strict:
744 case vmIntrinsics::_maxF_strict:
745 case vmIntrinsics::_minD_strict:
746 case vmIntrinsics::_maxD_strict:
747 return inline_min_max(intrinsic_id());
748
749 case vmIntrinsics::_VectorUnaryOp:
750 return inline_vector_nary_operation(1);
751 case vmIntrinsics::_VectorBinaryOp:
752 return inline_vector_nary_operation(2);
753 case vmIntrinsics::_VectorUnaryLibOp:
754 return inline_vector_call(1);
755 case vmIntrinsics::_VectorBinaryLibOp:
756 return inline_vector_call(2);
757 case vmIntrinsics::_VectorTernaryOp:
758 return inline_vector_nary_operation(3);
759 case vmIntrinsics::_VectorFromBitsCoerced:
760 return inline_vector_frombits_coerced();
761 case vmIntrinsics::_VectorMaskOp:
762 return inline_vector_mask_operation();
763 case vmIntrinsics::_VectorLoadOp:
764 return inline_vector_mem_operation(/*is_store=*/false);
765 case vmIntrinsics::_VectorLoadMaskedOp:
766 return inline_vector_mem_masked_operation(/*is_store*/false);
767 case vmIntrinsics::_VectorStoreOp:
768 return inline_vector_mem_operation(/*is_store=*/true);
769 case vmIntrinsics::_VectorStoreMaskedOp:
770 return inline_vector_mem_masked_operation(/*is_store=*/true);
771 case vmIntrinsics::_VectorGatherOp:
772 return inline_vector_gather_scatter(/*is_scatter*/ false);
773 case vmIntrinsics::_VectorScatterOp:
774 return inline_vector_gather_scatter(/*is_scatter*/ true);
775 case vmIntrinsics::_VectorReductionCoerced:
776 return inline_vector_reduction();
777 case vmIntrinsics::_VectorTest:
778 return inline_vector_test();
779 case vmIntrinsics::_VectorBlend:
780 return inline_vector_blend();
781 case vmIntrinsics::_VectorRearrange:
782 return inline_vector_rearrange();
783 case vmIntrinsics::_VectorSelectFrom:
784 return inline_vector_select_from();
785 case vmIntrinsics::_VectorCompare:
786 return inline_vector_compare();
787 case vmIntrinsics::_VectorBroadcastInt:
788 return inline_vector_broadcast_int();
789 case vmIntrinsics::_VectorConvert:
790 return inline_vector_convert();
791 case vmIntrinsics::_VectorInsert:
792 return inline_vector_insert();
793 case vmIntrinsics::_VectorExtract:
794 return inline_vector_extract();
795 case vmIntrinsics::_VectorCompressExpand:
796 return inline_vector_compress_expand();
797 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
798 return inline_vector_select_from_two_vectors();
799 case vmIntrinsics::_IndexVector:
800 return inline_index_vector();
801 case vmIntrinsics::_IndexPartiallyInUpperRange:
802 return inline_index_partially_in_upper_range();
803
804 case vmIntrinsics::_getObjectSize:
805 return inline_getObjectSize();
806
807 case vmIntrinsics::_blackhole:
808 return inline_blackhole();
809
810 default:
811 // If you get here, it may be that someone has added a new intrinsic
812 // to the list in vmIntrinsics.hpp without implementing it here.
813 #ifndef PRODUCT
814 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
815 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
816 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
817 }
818 #endif
819 return false;
820 }
821 }
822
823 Node* LibraryCallKit::try_to_predicate(int predicate) {
824 if (!jvms()->has_method()) {
825 // Root JVMState has a null method.
826 assert(map()->memory()->Opcode() == Op_Parm, "");
827 // Insert the memory aliasing node
828 set_all_memory(reset_memory());
829 }
830 assert(merged_memory(), "");
831
832 switch (intrinsic_id()) {
833 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
834 return inline_cipherBlockChaining_AESCrypt_predicate(false);
835 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
836 return inline_cipherBlockChaining_AESCrypt_predicate(true);
837 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
838 return inline_electronicCodeBook_AESCrypt_predicate(false);
839 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
840 return inline_electronicCodeBook_AESCrypt_predicate(true);
841 case vmIntrinsics::_counterMode_AESCrypt:
842 return inline_counterMode_AESCrypt_predicate();
843 case vmIntrinsics::_digestBase_implCompressMB:
844 return inline_digestBase_implCompressMB_predicate(predicate);
845 case vmIntrinsics::_galoisCounterMode_AESCrypt:
846 return inline_galoisCounterMode_AESCrypt_predicate();
847
848 default:
849 // If you get here, it may be that someone has added a new intrinsic
850 // to the list in vmIntrinsics.hpp without implementing it here.
851 #ifndef PRODUCT
852 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
853 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
854 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
855 }
856 #endif
857 Node* slow_ctl = control();
858 set_control(top()); // No fast path intrinsic
859 return slow_ctl;
860 }
861 }
862
863 //------------------------------set_result-------------------------------
864 // Helper function for finishing intrinsics.
865 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
866 record_for_igvn(region);
867 set_control(_gvn.transform(region));
868 set_result( _gvn.transform(value));
869 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
870 }
871
872 //------------------------------generate_guard---------------------------
873 // Helper function for generating guarded fast-slow graph structures.
874 // The given 'test', if true, guards a slow path. If the test fails
875 // then a fast path can be taken. (We generally hope it fails.)
876 // In all cases, GraphKit::control() is updated to the fast path.
877 // The returned value represents the control for the slow path.
878 // The return value is never 'top'; it is either a valid control
879 // or null if it is obvious that the slow path can never be taken.
880 // Also, if region and the slow control are not null, the slow edge
881 // is appended to the region.
882 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
883 if (stopped()) {
884 // Already short circuited.
885 return nullptr;
886 }
887
888 // Build an if node and its projections.
889 // If test is true we take the slow path, which we assume is uncommon.
890 if (_gvn.type(test) == TypeInt::ZERO) {
891 // The slow branch is never taken. No need to build this guard.
892 return nullptr;
893 }
894
895 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
896
897 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
898 if (if_slow == top()) {
899 // The slow branch is never taken. No need to build this guard.
900 return nullptr;
901 }
902
903 if (region != nullptr)
904 region->add_req(if_slow);
905
906 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
907 set_control(if_fast);
908
909 return if_slow;
910 }
911
912 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
913 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
914 }
915 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
916 return generate_guard(test, region, PROB_FAIR);
917 }
918
919 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
920 Node** pos_index, bool with_opaque) {
921 if (stopped())
922 return nullptr; // already stopped
923 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
924 return nullptr; // index is already adequately typed
925 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
926 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
927 if (with_opaque) {
928 bol_lt = _gvn.transform(new OpaqueConstantBoolNode(C, bol_lt, false));
929 }
930 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
931 if (is_neg != nullptr && pos_index != nullptr) {
932 // Emulate effect of Parse::adjust_map_after_if.
933 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
934 (*pos_index) = _gvn.transform(ccast);
935 }
936 return is_neg;
937 }
938
939 // Make sure that 'position' is a valid limit index, in [0..length].
940 // There are two equivalent plans for checking this:
941 // A. (offset + copyLength) unsigned<= arrayLength
942 // B. offset <= (arrayLength - copyLength)
943 // We require that all of the values above, except for the sum and
944 // difference, are already known to be non-negative.
945 // Plan A is robust in the face of overflow, if offset and copyLength
946 // are both hugely positive.
947 //
948 // Plan B is less direct and intuitive, but it does not overflow at
949 // all, since the difference of two non-negatives is always
950 // representable. Whenever Java methods must perform the equivalent
951 // check they generally use Plan B instead of Plan A.
952 // For the moment we use Plan A.
953 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
954 Node* subseq_length,
955 Node* array_length,
956 RegionNode* region,
957 bool with_opaque) {
958 if (stopped())
959 return nullptr; // already stopped
960 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
961 if (zero_offset && subseq_length->eqv_uncast(array_length))
962 return nullptr; // common case of whole-array copy
963 Node* last = subseq_length;
964 if (!zero_offset) // last += offset
965 last = _gvn.transform(new AddINode(last, offset));
966 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
967 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
968 if (with_opaque) {
969 bol_lt = _gvn.transform(new OpaqueConstantBoolNode(C, bol_lt, false));
970 }
971 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
972 return is_over;
973 }
974
975 // Emit range checks for the given String.value byte array
976 void LibraryCallKit::generate_string_range_check(Node* array,
977 Node* offset,
978 Node* count,
979 bool char_count,
980 bool halt_on_oob) {
981 if (stopped()) {
982 return; // already stopped
983 }
984 RegionNode* bailout = new RegionNode(1);
985 record_for_igvn(bailout);
986 if (char_count) {
987 // Convert char count to byte count
988 count = _gvn.transform(new LShiftINode(count, intcon(1)));
989 }
990
991 // Offset and count must not be negative
992 generate_negative_guard(offset, bailout, nullptr, halt_on_oob);
993 generate_negative_guard(count, bailout, nullptr, halt_on_oob);
994 // Offset + count must not exceed length of array
995 generate_limit_guard(offset, count, load_array_length(array), bailout, halt_on_oob);
996
997 if (bailout->req() > 1) {
998 if (halt_on_oob) {
999 bailout = _gvn.transform(bailout)->as_Region();
1000 Node* frame = _gvn.transform(new ParmNode(C->start(), TypeFunc::FramePtr));
1001 Node* halt = _gvn.transform(new HaltNode(bailout, frame, "unexpected guard failure in intrinsic"));
1002 C->root()->add_req(halt);
1003 } else {
1004 PreserveJVMState pjvms(this);
1005 set_control(_gvn.transform(bailout));
1006 uncommon_trap(Deoptimization::Reason_intrinsic,
1007 Deoptimization::Action_maybe_recompile);
1008 }
1009 }
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 = basic_plus_adr(top()/*!oop*/, 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 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1169 return false;
1170 }
1171
1172 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1173 // no receiver since it is static method
1174 Node* ba = argument(0);
1175 Node* offset = argument(1);
1176 Node* len = argument(2);
1177
1178 ba = must_be_not_null(ba, true);
1179 generate_string_range_check(ba, offset, len, false, true);
1180 if (stopped()) {
1181 return true;
1182 }
1183
1184 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1185 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1186 set_result(_gvn.transform(result));
1187 clear_upper_avx();
1188 return true;
1189 }
1190
1191 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1192 Node* index = argument(0);
1193 Node* length = bt == T_INT ? argument(1) : argument(2);
1194 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1195 return false;
1196 }
1197
1198 // check that length is positive
1199 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1200 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1201
1202 {
1203 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1204 uncommon_trap(Deoptimization::Reason_intrinsic,
1205 Deoptimization::Action_make_not_entrant);
1206 }
1207
1208 if (stopped()) {
1209 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1210 return true;
1211 }
1212
1213 // length is now known positive, add a cast node to make this explicit
1214 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1215 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1216 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1217 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1218 casted_length = _gvn.transform(casted_length);
1219 replace_in_map(length, casted_length);
1220 length = casted_length;
1221
1222 // Use an unsigned comparison for the range check itself
1223 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1224 BoolTest::mask btest = BoolTest::lt;
1225 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1226 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1227 _gvn.set_type(rc, rc->Value(&_gvn));
1228 if (!rc_bool->is_Con()) {
1229 record_for_igvn(rc);
1230 }
1231 set_control(_gvn.transform(new IfTrueNode(rc)));
1232 {
1233 PreserveJVMState pjvms(this);
1234 set_control(_gvn.transform(new IfFalseNode(rc)));
1235 uncommon_trap(Deoptimization::Reason_range_check,
1236 Deoptimization::Action_make_not_entrant);
1237 }
1238
1239 if (stopped()) {
1240 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1241 return true;
1242 }
1243
1244 // index is now known to be >= 0 and < length, cast it
1245 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1246 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1247 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1248 result = _gvn.transform(result);
1249 set_result(result);
1250 replace_in_map(index, result);
1251 return true;
1252 }
1253
1254 //------------------------------inline_string_indexOf------------------------
1255 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1256 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1257 return false;
1258 }
1259 Node* src = argument(0);
1260 Node* tgt = argument(1);
1261
1262 // Make the merge point
1263 RegionNode* result_rgn = new RegionNode(4);
1264 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1265
1266 src = must_be_not_null(src, true);
1267 tgt = must_be_not_null(tgt, true);
1268
1269 // Get start addr and length of source string
1270 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1271 Node* src_count = load_array_length(src);
1272
1273 // Get start addr and length of substring
1274 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1275 Node* tgt_count = load_array_length(tgt);
1276
1277 Node* result = nullptr;
1278 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1279
1280 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1281 // Divide src size by 2 if String is UTF16 encoded
1282 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1283 }
1284 if (ae == StrIntrinsicNode::UU) {
1285 // Divide substring size by 2 if String is UTF16 encoded
1286 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1287 }
1288
1289 if (call_opt_stub) {
1290 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1291 StubRoutines::_string_indexof_array[ae],
1292 "stringIndexOf", TypePtr::BOTTOM, src_start,
1293 src_count, tgt_start, tgt_count);
1294 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1295 } else {
1296 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1297 result_rgn, result_phi, ae);
1298 }
1299 if (result != nullptr) {
1300 result_phi->init_req(3, result);
1301 result_rgn->init_req(3, control());
1302 }
1303 set_control(_gvn.transform(result_rgn));
1304 record_for_igvn(result_rgn);
1305 set_result(_gvn.transform(result_phi));
1306
1307 return true;
1308 }
1309
1310 //-----------------------------inline_string_indexOfI-----------------------
1311 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1312 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1313 return false;
1314 }
1315 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1316 return false;
1317 }
1318
1319 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1320 Node* src = argument(0); // byte[]
1321 Node* src_count = argument(1); // char count
1322 Node* tgt = argument(2); // byte[]
1323 Node* tgt_count = argument(3); // char count
1324 Node* from_index = argument(4); // char index
1325
1326 src = must_be_not_null(src, true);
1327 tgt = must_be_not_null(tgt, true);
1328
1329 // Multiply byte array index by 2 if String is UTF16 encoded
1330 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1331 src_count = _gvn.transform(new SubINode(src_count, from_index));
1332 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1333 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1334
1335 // Range checks
1336 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL, true);
1337 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU, true);
1338 if (stopped()) {
1339 return true;
1340 }
1341
1342 RegionNode* region = new RegionNode(5);
1343 Node* phi = new PhiNode(region, TypeInt::INT);
1344 Node* result = nullptr;
1345
1346 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1347
1348 if (call_opt_stub) {
1349 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1350 StubRoutines::_string_indexof_array[ae],
1351 "stringIndexOf", TypePtr::BOTTOM, src_start,
1352 src_count, tgt_start, tgt_count);
1353 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1354 } else {
1355 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1356 region, phi, ae);
1357 }
1358 if (result != nullptr) {
1359 // The result is index relative to from_index if substring was found, -1 otherwise.
1360 // Generate code which will fold into cmove.
1361 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1362 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1363
1364 Node* if_lt = generate_slow_guard(bol, nullptr);
1365 if (if_lt != nullptr) {
1366 // result == -1
1367 phi->init_req(3, result);
1368 region->init_req(3, if_lt);
1369 }
1370 if (!stopped()) {
1371 result = _gvn.transform(new AddINode(result, from_index));
1372 phi->init_req(4, result);
1373 region->init_req(4, control());
1374 }
1375 }
1376
1377 set_control(_gvn.transform(region));
1378 record_for_igvn(region);
1379 set_result(_gvn.transform(phi));
1380 clear_upper_avx();
1381
1382 return true;
1383 }
1384
1385 // Create StrIndexOfNode with fast path checks
1386 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1387 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1388 // Check for substr count > string count
1389 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1390 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1391 Node* if_gt = generate_slow_guard(bol, nullptr);
1392 if (if_gt != nullptr) {
1393 phi->init_req(1, intcon(-1));
1394 region->init_req(1, if_gt);
1395 }
1396 if (!stopped()) {
1397 // Check for substr count == 0
1398 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1399 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1400 Node* if_zero = generate_slow_guard(bol, nullptr);
1401 if (if_zero != nullptr) {
1402 phi->init_req(2, intcon(0));
1403 region->init_req(2, if_zero);
1404 }
1405 }
1406 if (!stopped()) {
1407 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1408 }
1409 return nullptr;
1410 }
1411
1412 //-----------------------------inline_string_indexOfChar-----------------------
1413 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1414 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1415 return false;
1416 }
1417 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1418 return false;
1419 }
1420 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1421 Node* src = argument(0); // byte[]
1422 Node* int_ch = argument(1);
1423 Node* from_index = argument(2);
1424 Node* max = argument(3);
1425
1426 src = must_be_not_null(src, true);
1427
1428 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1429 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1430 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1431
1432 // Range checks
1433 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U, true);
1434
1435 // Check for int_ch >= 0
1436 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1437 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1438 {
1439 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1440 uncommon_trap(Deoptimization::Reason_intrinsic,
1441 Deoptimization::Action_maybe_recompile);
1442 }
1443 if (stopped()) {
1444 return true;
1445 }
1446
1447 RegionNode* region = new RegionNode(3);
1448 Node* phi = new PhiNode(region, TypeInt::INT);
1449
1450 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1451 C->set_has_split_ifs(true); // Has chance for split-if optimization
1452 _gvn.transform(result);
1453
1454 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1455 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1456
1457 Node* if_lt = generate_slow_guard(bol, nullptr);
1458 if (if_lt != nullptr) {
1459 // result == -1
1460 phi->init_req(2, result);
1461 region->init_req(2, if_lt);
1462 }
1463 if (!stopped()) {
1464 result = _gvn.transform(new AddINode(result, from_index));
1465 phi->init_req(1, result);
1466 region->init_req(1, control());
1467 }
1468 set_control(_gvn.transform(region));
1469 record_for_igvn(region);
1470 set_result(_gvn.transform(phi));
1471 clear_upper_avx();
1472
1473 return true;
1474 }
1475 //---------------------------inline_string_copy---------------------
1476 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1477 // int StringUTF16.compress0(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1478 // int StringUTF16.compress0(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1479 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1480 // void StringLatin1.inflate0(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1481 // void StringLatin1.inflate0(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1482 bool LibraryCallKit::inline_string_copy(bool compress) {
1483 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1484 return false;
1485 }
1486 int nargs = 5; // 2 oops, 3 ints
1487 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1488
1489 Node* src = argument(0);
1490 Node* src_offset = argument(1);
1491 Node* dst = argument(2);
1492 Node* dst_offset = argument(3);
1493 Node* length = argument(4);
1494
1495 // Check for allocation before we add nodes that would confuse
1496 // tightly_coupled_allocation()
1497 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1498
1499 // Figure out the size and type of the elements we will be copying.
1500 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1501 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1502 if (src_type == nullptr || dst_type == nullptr) {
1503 return false;
1504 }
1505 BasicType src_elem = src_type->elem()->array_element_basic_type();
1506 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1507 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1508 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1509 "Unsupported array types for inline_string_copy");
1510
1511 src = must_be_not_null(src, true);
1512 dst = must_be_not_null(dst, true);
1513
1514 // Convert char[] offsets to byte[] offsets
1515 bool convert_src = (compress && src_elem == T_BYTE);
1516 bool convert_dst = (!compress && dst_elem == T_BYTE);
1517 if (convert_src) {
1518 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1519 } else if (convert_dst) {
1520 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1521 }
1522
1523 // Range checks
1524 generate_string_range_check(src, src_offset, length, convert_src, true);
1525 generate_string_range_check(dst, dst_offset, length, convert_dst, true);
1526 if (stopped()) {
1527 return true;
1528 }
1529
1530 Node* src_start = array_element_address(src, src_offset, src_elem);
1531 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1532 // 'src_start' points to src array + scaled offset
1533 // 'dst_start' points to dst array + scaled offset
1534 Node* count = nullptr;
1535 if (compress) {
1536 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1537 } else {
1538 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1539 }
1540
1541 if (alloc != nullptr) {
1542 if (alloc->maybe_set_complete(&_gvn)) {
1543 // "You break it, you buy it."
1544 InitializeNode* init = alloc->initialization();
1545 assert(init->is_complete(), "we just did this");
1546 init->set_complete_with_arraycopy();
1547 assert(dst->is_CheckCastPP(), "sanity");
1548 assert(dst->in(0)->in(0) == init, "dest pinned");
1549 }
1550 // Do not let stores that initialize this object be reordered with
1551 // a subsequent store that would make this object accessible by
1552 // other threads.
1553 // Record what AllocateNode this StoreStore protects so that
1554 // escape analysis can go from the MemBarStoreStoreNode to the
1555 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1556 // based on the escape status of the AllocateNode.
1557 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1558 }
1559 if (compress) {
1560 set_result(_gvn.transform(count));
1561 }
1562 clear_upper_avx();
1563
1564 return true;
1565 }
1566
1567 #ifdef _LP64
1568 #define XTOP ,top() /*additional argument*/
1569 #else //_LP64
1570 #define XTOP /*no additional argument*/
1571 #endif //_LP64
1572
1573 //------------------------inline_string_toBytesU--------------------------
1574 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1575 bool LibraryCallKit::inline_string_toBytesU() {
1576 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1577 return false;
1578 }
1579 // Get the arguments.
1580 Node* value = argument(0);
1581 Node* offset = argument(1);
1582 Node* length = argument(2);
1583
1584 Node* newcopy = nullptr;
1585
1586 // Set the original stack and the reexecute bit for the interpreter to reexecute
1587 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1588 { PreserveReexecuteState preexecs(this);
1589 jvms()->set_should_reexecute(true);
1590
1591 // Check if a null path was taken unconditionally.
1592 value = null_check(value);
1593
1594 RegionNode* bailout = new RegionNode(1);
1595 record_for_igvn(bailout);
1596
1597 // Range checks
1598 generate_negative_guard(offset, bailout);
1599 generate_negative_guard(length, bailout);
1600 generate_limit_guard(offset, length, load_array_length(value), bailout);
1601 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1602 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1603
1604 if (bailout->req() > 1) {
1605 PreserveJVMState pjvms(this);
1606 set_control(_gvn.transform(bailout));
1607 uncommon_trap(Deoptimization::Reason_intrinsic,
1608 Deoptimization::Action_maybe_recompile);
1609 }
1610 if (stopped()) {
1611 return true;
1612 }
1613
1614 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1615 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1616 newcopy = new_array(klass_node, size, 0); // no arguments to push
1617 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1618 guarantee(alloc != nullptr, "created above");
1619
1620 // Calculate starting addresses.
1621 Node* src_start = array_element_address(value, offset, T_CHAR);
1622 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1623
1624 // Check if dst array address is aligned to HeapWordSize
1625 bool aligned = (arrayOopDesc::base_offset_in_bytes(T_BYTE) % HeapWordSize == 0);
1626 // If true, then check if src array address is aligned to HeapWordSize
1627 if (aligned) {
1628 const TypeInt* toffset = gvn().type(offset)->is_int();
1629 aligned = toffset->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) +
1630 toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1631 }
1632
1633 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1634 const char* copyfunc_name = "arraycopy";
1635 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1636 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1637 OptoRuntime::fast_arraycopy_Type(),
1638 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1639 src_start, dst_start, ConvI2X(length) XTOP);
1640 // Do not let reads from the cloned object float above the arraycopy.
1641 if (alloc->maybe_set_complete(&_gvn)) {
1642 // "You break it, you buy it."
1643 InitializeNode* init = alloc->initialization();
1644 assert(init->is_complete(), "we just did this");
1645 init->set_complete_with_arraycopy();
1646 assert(newcopy->is_CheckCastPP(), "sanity");
1647 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1648 }
1649 // Do not let stores that initialize this object be reordered with
1650 // a subsequent store that would make this object accessible by
1651 // other threads.
1652 // Record what AllocateNode this StoreStore protects so that
1653 // escape analysis can go from the MemBarStoreStoreNode to the
1654 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1655 // based on the escape status of the AllocateNode.
1656 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1657 } // original reexecute is set back here
1658
1659 C->set_has_split_ifs(true); // Has chance for split-if optimization
1660 if (!stopped()) {
1661 set_result(newcopy);
1662 }
1663 clear_upper_avx();
1664
1665 return true;
1666 }
1667
1668 //------------------------inline_string_getCharsU--------------------------
1669 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1670 bool LibraryCallKit::inline_string_getCharsU() {
1671 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1672 return false;
1673 }
1674
1675 // Get the arguments.
1676 Node* src = argument(0);
1677 Node* src_begin = argument(1);
1678 Node* src_end = argument(2); // exclusive offset (i < src_end)
1679 Node* dst = argument(3);
1680 Node* dst_begin = argument(4);
1681
1682 // Check for allocation before we add nodes that would confuse
1683 // tightly_coupled_allocation()
1684 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1685
1686 // Check if a null path was taken unconditionally.
1687 src = null_check(src);
1688 dst = null_check(dst);
1689 if (stopped()) {
1690 return true;
1691 }
1692
1693 // Get length and convert char[] offset to byte[] offset
1694 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1695 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1696
1697 // Range checks
1698 generate_string_range_check(src, src_begin, length, true);
1699 generate_string_range_check(dst, dst_begin, length, false);
1700 if (stopped()) {
1701 return true;
1702 }
1703
1704 if (!stopped()) {
1705 // Calculate starting addresses.
1706 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1707 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1708
1709 // Check if array addresses are aligned to HeapWordSize
1710 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1711 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1712 bool aligned = tsrc->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_BYTE) + tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1713 tdst->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) + tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1714
1715 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1716 const char* copyfunc_name = "arraycopy";
1717 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1718 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1719 OptoRuntime::fast_arraycopy_Type(),
1720 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1721 src_start, dst_start, ConvI2X(length) XTOP);
1722 // Do not let reads from the cloned object float above the arraycopy.
1723 if (alloc != nullptr) {
1724 if (alloc->maybe_set_complete(&_gvn)) {
1725 // "You break it, you buy it."
1726 InitializeNode* init = alloc->initialization();
1727 assert(init->is_complete(), "we just did this");
1728 init->set_complete_with_arraycopy();
1729 assert(dst->is_CheckCastPP(), "sanity");
1730 assert(dst->in(0)->in(0) == init, "dest pinned");
1731 }
1732 // Do not let stores that initialize this object be reordered with
1733 // a subsequent store that would make this object accessible by
1734 // other threads.
1735 // Record what AllocateNode this StoreStore protects so that
1736 // escape analysis can go from the MemBarStoreStoreNode to the
1737 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1738 // based on the escape status of the AllocateNode.
1739 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1740 } else {
1741 insert_mem_bar(Op_MemBarCPUOrder);
1742 }
1743 }
1744
1745 C->set_has_split_ifs(true); // Has chance for split-if optimization
1746 return true;
1747 }
1748
1749 //----------------------inline_string_char_access----------------------------
1750 // Store/Load char to/from byte[] array.
1751 // static void StringUTF16.putChar(byte[] val, int index, int c)
1752 // static char StringUTF16.getChar(byte[] val, int index)
1753 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1754 Node* value = argument(0);
1755 Node* index = argument(1);
1756 Node* ch = is_store ? argument(2) : nullptr;
1757
1758 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1759 // correctly requires matched array shapes.
1760 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1761 "sanity: byte[] and char[] bases agree");
1762 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1763 "sanity: byte[] and char[] scales agree");
1764
1765 // Bail when getChar over constants is requested: constant folding would
1766 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1767 // Java method would constant fold nicely instead.
1768 if (!is_store && value->is_Con() && index->is_Con()) {
1769 return false;
1770 }
1771
1772 // Save state and restore on bailout
1773 SavedState old_state(this);
1774
1775 value = must_be_not_null(value, true);
1776
1777 Node* adr = array_element_address(value, index, T_CHAR);
1778 if (adr->is_top()) {
1779 return false;
1780 }
1781 old_state.discard();
1782 if (is_store) {
1783 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1784 } else {
1785 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);
1786 set_result(ch);
1787 }
1788 return true;
1789 }
1790
1791
1792 //------------------------------inline_math-----------------------------------
1793 // public static double Math.abs(double)
1794 // public static double Math.sqrt(double)
1795 // public static double Math.log(double)
1796 // public static double Math.log10(double)
1797 // public static double Math.round(double)
1798 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1799 Node* arg = argument(0);
1800 Node* n = nullptr;
1801 switch (id) {
1802 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1803 case vmIntrinsics::_dsqrt:
1804 case vmIntrinsics::_dsqrt_strict:
1805 n = new SqrtDNode(C, control(), arg); break;
1806 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1807 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1808 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1809 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1810 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1811 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1812 default: fatal_unexpected_iid(id); break;
1813 }
1814 set_result(_gvn.transform(n));
1815 return true;
1816 }
1817
1818 //------------------------------inline_math-----------------------------------
1819 // public static float Math.abs(float)
1820 // public static int Math.abs(int)
1821 // public static long Math.abs(long)
1822 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1823 Node* arg = argument(0);
1824 Node* n = nullptr;
1825 switch (id) {
1826 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1827 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1828 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1829 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1830 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1831 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1832 default: fatal_unexpected_iid(id); break;
1833 }
1834 set_result(_gvn.transform(n));
1835 return true;
1836 }
1837
1838 //------------------------------runtime_math-----------------------------
1839 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1840 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1841 "must be (DD)D or (D)D type");
1842
1843 // Inputs
1844 Node* a = argument(0);
1845 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1846
1847 const TypePtr* no_memory_effects = nullptr;
1848 Node* trig = make_runtime_call(RC_LEAF | RC_PURE, call_type, funcAddr, funcName,
1849 no_memory_effects,
1850 a, top(), b, b ? top() : nullptr);
1851 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1852 #ifdef ASSERT
1853 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1854 assert(value_top == top(), "second value must be top");
1855 #endif
1856
1857 set_result(value);
1858 return true;
1859 }
1860
1861 //------------------------------inline_math_pow-----------------------------
1862 bool LibraryCallKit::inline_math_pow() {
1863 Node* exp = argument(2);
1864 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1865 if (d != nullptr) {
1866 if (d->getd() == 2.0) {
1867 // Special case: pow(x, 2.0) => x * x
1868 Node* base = argument(0);
1869 set_result(_gvn.transform(new MulDNode(base, base)));
1870 return true;
1871 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1872 // Special case: pow(x, 0.5) => sqrt(x)
1873 Node* base = argument(0);
1874 Node* zero = _gvn.zerocon(T_DOUBLE);
1875
1876 RegionNode* region = new RegionNode(3);
1877 Node* phi = new PhiNode(region, Type::DOUBLE);
1878
1879 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1880 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1881 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1882 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1883 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1884
1885 Node* if_pow = generate_slow_guard(test, nullptr);
1886 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1887 phi->init_req(1, value_sqrt);
1888 region->init_req(1, control());
1889
1890 if (if_pow != nullptr) {
1891 set_control(if_pow);
1892 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1893 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1894 const TypePtr* no_memory_effects = nullptr;
1895 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1896 no_memory_effects, base, top(), exp, top());
1897 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1898 #ifdef ASSERT
1899 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1900 assert(value_top == top(), "second value must be top");
1901 #endif
1902 phi->init_req(2, value_pow);
1903 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1904 }
1905
1906 C->set_has_split_ifs(true); // Has chance for split-if optimization
1907 set_control(_gvn.transform(region));
1908 record_for_igvn(region);
1909 set_result(_gvn.transform(phi));
1910
1911 return true;
1912 }
1913 }
1914
1915 return StubRoutines::dpow() != nullptr ?
1916 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1917 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1918 }
1919
1920 //------------------------------inline_math_native-----------------------------
1921 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1922 switch (id) {
1923 case vmIntrinsics::_dsin:
1924 return StubRoutines::dsin() != nullptr ?
1925 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1926 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1927 case vmIntrinsics::_dcos:
1928 return StubRoutines::dcos() != nullptr ?
1929 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1930 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1931 case vmIntrinsics::_dtan:
1932 return StubRoutines::dtan() != nullptr ?
1933 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1934 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1935 case vmIntrinsics::_dsinh:
1936 return StubRoutines::dsinh() != nullptr ?
1937 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsinh(), "dsinh") : false;
1938 case vmIntrinsics::_dtanh:
1939 return StubRoutines::dtanh() != nullptr ?
1940 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1941 case vmIntrinsics::_dcbrt:
1942 return StubRoutines::dcbrt() != nullptr ?
1943 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcbrt(), "dcbrt") : false;
1944 case vmIntrinsics::_dexp:
1945 return StubRoutines::dexp() != nullptr ?
1946 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1947 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1948 case vmIntrinsics::_dlog:
1949 return StubRoutines::dlog() != nullptr ?
1950 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1951 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1952 case vmIntrinsics::_dlog10:
1953 return StubRoutines::dlog10() != nullptr ?
1954 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1955 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1956
1957 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1958 case vmIntrinsics::_ceil:
1959 case vmIntrinsics::_floor:
1960 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1961
1962 case vmIntrinsics::_dsqrt:
1963 case vmIntrinsics::_dsqrt_strict:
1964 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1965 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1966 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1967 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1968 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1969
1970 case vmIntrinsics::_dpow: return inline_math_pow();
1971 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1972 case vmIntrinsics::_fcopySign: return inline_math(id);
1973 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1974 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1975 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1976
1977 // These intrinsics are not yet correctly implemented
1978 case vmIntrinsics::_datan2:
1979 return false;
1980
1981 default:
1982 fatal_unexpected_iid(id);
1983 return false;
1984 }
1985 }
1986
1987 //----------------------------inline_notify-----------------------------------*
1988 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1989 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1990 address func;
1991 if (id == vmIntrinsics::_notify) {
1992 func = OptoRuntime::monitor_notify_Java();
1993 } else {
1994 func = OptoRuntime::monitor_notifyAll_Java();
1995 }
1996 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1997 make_slow_call_ex(call, env()->Throwable_klass(), false);
1998 return true;
1999 }
2000
2001
2002 //----------------------------inline_min_max-----------------------------------
2003 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
2004 Node* a = nullptr;
2005 Node* b = nullptr;
2006 Node* n = nullptr;
2007 switch (id) {
2008 case vmIntrinsics::_min:
2009 case vmIntrinsics::_max:
2010 case vmIntrinsics::_minF:
2011 case vmIntrinsics::_maxF:
2012 case vmIntrinsics::_minF_strict:
2013 case vmIntrinsics::_maxF_strict:
2014 case vmIntrinsics::_min_strict:
2015 case vmIntrinsics::_max_strict:
2016 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
2017 a = argument(0);
2018 b = argument(1);
2019 break;
2020 case vmIntrinsics::_minD:
2021 case vmIntrinsics::_maxD:
2022 case vmIntrinsics::_minD_strict:
2023 case vmIntrinsics::_maxD_strict:
2024 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
2025 a = argument(0);
2026 b = argument(2);
2027 break;
2028 case vmIntrinsics::_minL:
2029 case vmIntrinsics::_maxL:
2030 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
2031 a = argument(0);
2032 b = argument(2);
2033 break;
2034 default:
2035 fatal_unexpected_iid(id);
2036 break;
2037 }
2038
2039 switch (id) {
2040 case vmIntrinsics::_min:
2041 case vmIntrinsics::_min_strict:
2042 n = new MinINode(a, b);
2043 break;
2044 case vmIntrinsics::_max:
2045 case vmIntrinsics::_max_strict:
2046 n = new MaxINode(a, b);
2047 break;
2048 case vmIntrinsics::_minF:
2049 case vmIntrinsics::_minF_strict:
2050 n = new MinFNode(a, b);
2051 break;
2052 case vmIntrinsics::_maxF:
2053 case vmIntrinsics::_maxF_strict:
2054 n = new MaxFNode(a, b);
2055 break;
2056 case vmIntrinsics::_minD:
2057 case vmIntrinsics::_minD_strict:
2058 n = new MinDNode(a, b);
2059 break;
2060 case vmIntrinsics::_maxD:
2061 case vmIntrinsics::_maxD_strict:
2062 n = new MaxDNode(a, b);
2063 break;
2064 case vmIntrinsics::_minL:
2065 n = new MinLNode(_gvn.C, a, b);
2066 break;
2067 case vmIntrinsics::_maxL:
2068 n = new MaxLNode(_gvn.C, a, b);
2069 break;
2070 default:
2071 fatal_unexpected_iid(id);
2072 break;
2073 }
2074
2075 set_result(_gvn.transform(n));
2076 return true;
2077 }
2078
2079 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
2080 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
2081 env()->ArithmeticException_instance())) {
2082 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
2083 // so let's bail out intrinsic rather than risking deopting again.
2084 return false;
2085 }
2086
2087 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2088 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2089 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2090 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2091
2092 {
2093 PreserveJVMState pjvms(this);
2094 PreserveReexecuteState preexecs(this);
2095 jvms()->set_should_reexecute(true);
2096
2097 set_control(slow_path);
2098 set_i_o(i_o());
2099
2100 builtin_throw(Deoptimization::Reason_intrinsic,
2101 env()->ArithmeticException_instance(),
2102 /*allow_too_many_traps*/ false);
2103 }
2104
2105 set_control(fast_path);
2106 set_result(math);
2107 return true;
2108 }
2109
2110 template <typename OverflowOp>
2111 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2112 typedef typename OverflowOp::MathOp MathOp;
2113
2114 MathOp* mathOp = new MathOp(arg1, arg2);
2115 Node* operation = _gvn.transform( mathOp );
2116 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2117 return inline_math_mathExact(operation, ofcheck);
2118 }
2119
2120 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2121 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2122 }
2123
2124 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2125 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2126 }
2127
2128 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2129 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2130 }
2131
2132 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2133 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2134 }
2135
2136 bool LibraryCallKit::inline_math_negateExactI() {
2137 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2138 }
2139
2140 bool LibraryCallKit::inline_math_negateExactL() {
2141 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2142 }
2143
2144 bool LibraryCallKit::inline_math_multiplyExactI() {
2145 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2146 }
2147
2148 bool LibraryCallKit::inline_math_multiplyExactL() {
2149 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2150 }
2151
2152 bool LibraryCallKit::inline_math_multiplyHigh() {
2153 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2154 return true;
2155 }
2156
2157 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2158 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2159 return true;
2160 }
2161
2162 inline int
2163 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2164 const TypePtr* base_type = TypePtr::NULL_PTR;
2165 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2166 if (base_type == nullptr) {
2167 // Unknown type.
2168 return Type::AnyPtr;
2169 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2170 // Since this is a null+long form, we have to switch to a rawptr.
2171 base = _gvn.transform(new CastX2PNode(offset));
2172 offset = MakeConX(0);
2173 return Type::RawPtr;
2174 } else if (base_type->base() == Type::RawPtr) {
2175 return Type::RawPtr;
2176 } else if (base_type->isa_oopptr()) {
2177 // Base is never null => always a heap address.
2178 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2179 return Type::OopPtr;
2180 }
2181 // Offset is small => always a heap address.
2182 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2183 if (offset_type != nullptr &&
2184 base_type->offset() == 0 && // (should always be?)
2185 offset_type->_lo >= 0 &&
2186 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2187 return Type::OopPtr;
2188 } else if (type == T_OBJECT) {
2189 // off heap access to an oop doesn't make any sense. Has to be on
2190 // heap.
2191 return Type::OopPtr;
2192 }
2193 // Otherwise, it might either be oop+off or null+addr.
2194 return Type::AnyPtr;
2195 } else {
2196 // No information:
2197 return Type::AnyPtr;
2198 }
2199 }
2200
2201 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2202 Node* uncasted_base = base;
2203 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2204 if (kind == Type::RawPtr) {
2205 return basic_plus_adr(top(), uncasted_base, offset);
2206 } else if (kind == Type::AnyPtr) {
2207 assert(base == uncasted_base, "unexpected base change");
2208 if (can_cast) {
2209 if (!_gvn.type(base)->speculative_maybe_null() &&
2210 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2211 // According to profiling, this access is always on
2212 // heap. Casting the base to not null and thus avoiding membars
2213 // around the access should allow better optimizations
2214 Node* null_ctl = top();
2215 base = null_check_oop(base, &null_ctl, true, true, true);
2216 assert(null_ctl->is_top(), "no null control here");
2217 return basic_plus_adr(base, offset);
2218 } else if (_gvn.type(base)->speculative_always_null() &&
2219 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2220 // According to profiling, this access is always off
2221 // heap.
2222 base = null_assert(base);
2223 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2224 offset = MakeConX(0);
2225 return basic_plus_adr(top(), raw_base, offset);
2226 }
2227 }
2228 // We don't know if it's an on heap or off heap access. Fall back
2229 // to raw memory access.
2230 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2231 return basic_plus_adr(top(), raw, offset);
2232 } else {
2233 assert(base == uncasted_base, "unexpected base change");
2234 // We know it's an on heap access so base can't be null
2235 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2236 base = must_be_not_null(base, true);
2237 }
2238 return basic_plus_adr(base, offset);
2239 }
2240 }
2241
2242 //--------------------------inline_number_methods-----------------------------
2243 // inline int Integer.numberOfLeadingZeros(int)
2244 // inline int Long.numberOfLeadingZeros(long)
2245 //
2246 // inline int Integer.numberOfTrailingZeros(int)
2247 // inline int Long.numberOfTrailingZeros(long)
2248 //
2249 // inline int Integer.bitCount(int)
2250 // inline int Long.bitCount(long)
2251 //
2252 // inline char Character.reverseBytes(char)
2253 // inline short Short.reverseBytes(short)
2254 // inline int Integer.reverseBytes(int)
2255 // inline long Long.reverseBytes(long)
2256 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2257 Node* arg = argument(0);
2258 Node* n = nullptr;
2259 switch (id) {
2260 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2261 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2262 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2263 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2264 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2265 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2266 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2267 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2268 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2269 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2270 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2271 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2272 default: fatal_unexpected_iid(id); break;
2273 }
2274 set_result(_gvn.transform(n));
2275 return true;
2276 }
2277
2278 //--------------------------inline_bitshuffle_methods-----------------------------
2279 // inline int Integer.compress(int, int)
2280 // inline int Integer.expand(int, int)
2281 // inline long Long.compress(long, long)
2282 // inline long Long.expand(long, long)
2283 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2284 Node* n = nullptr;
2285 switch (id) {
2286 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2287 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2288 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2289 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2290 default: fatal_unexpected_iid(id); break;
2291 }
2292 set_result(_gvn.transform(n));
2293 return true;
2294 }
2295
2296 //--------------------------inline_number_methods-----------------------------
2297 // inline int Integer.compareUnsigned(int, int)
2298 // inline int Long.compareUnsigned(long, long)
2299 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2300 Node* arg1 = argument(0);
2301 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2302 Node* n = nullptr;
2303 switch (id) {
2304 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2305 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2306 default: fatal_unexpected_iid(id); break;
2307 }
2308 set_result(_gvn.transform(n));
2309 return true;
2310 }
2311
2312 //--------------------------inline_unsigned_divmod_methods-----------------------------
2313 // inline int Integer.divideUnsigned(int, int)
2314 // inline int Integer.remainderUnsigned(int, int)
2315 // inline long Long.divideUnsigned(long, long)
2316 // inline long Long.remainderUnsigned(long, long)
2317 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2318 Node* n = nullptr;
2319 switch (id) {
2320 case vmIntrinsics::_divideUnsigned_i: {
2321 zero_check_int(argument(1));
2322 // Compile-time detect of null-exception
2323 if (stopped()) {
2324 return true; // keep the graph constructed so far
2325 }
2326 n = new UDivINode(control(), argument(0), argument(1));
2327 break;
2328 }
2329 case vmIntrinsics::_divideUnsigned_l: {
2330 zero_check_long(argument(2));
2331 // Compile-time detect of null-exception
2332 if (stopped()) {
2333 return true; // keep the graph constructed so far
2334 }
2335 n = new UDivLNode(control(), argument(0), argument(2));
2336 break;
2337 }
2338 case vmIntrinsics::_remainderUnsigned_i: {
2339 zero_check_int(argument(1));
2340 // Compile-time detect of null-exception
2341 if (stopped()) {
2342 return true; // keep the graph constructed so far
2343 }
2344 n = new UModINode(control(), argument(0), argument(1));
2345 break;
2346 }
2347 case vmIntrinsics::_remainderUnsigned_l: {
2348 zero_check_long(argument(2));
2349 // Compile-time detect of null-exception
2350 if (stopped()) {
2351 return true; // keep the graph constructed so far
2352 }
2353 n = new UModLNode(control(), argument(0), argument(2));
2354 break;
2355 }
2356 default: fatal_unexpected_iid(id); break;
2357 }
2358 set_result(_gvn.transform(n));
2359 return true;
2360 }
2361
2362 //----------------------------inline_unsafe_access----------------------------
2363
2364 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2365 // Attempt to infer a sharper value type from the offset and base type.
2366 ciKlass* sharpened_klass = nullptr;
2367 bool null_free = false;
2368
2369 // See if it is an instance field, with an object type.
2370 if (alias_type->field() != nullptr) {
2371 if (alias_type->field()->type()->is_klass()) {
2372 sharpened_klass = alias_type->field()->type()->as_klass();
2373 null_free = alias_type->field()->is_null_free();
2374 }
2375 }
2376
2377 const TypeOopPtr* result = nullptr;
2378 // See if it is a narrow oop array.
2379 if (adr_type->isa_aryptr()) {
2380 if (adr_type->offset() >= refArrayOopDesc::base_offset_in_bytes()) {
2381 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2382 null_free = adr_type->is_aryptr()->is_null_free();
2383 if (elem_type != nullptr && elem_type->is_loaded()) {
2384 // Sharpen the value type.
2385 result = elem_type;
2386 }
2387 }
2388 }
2389
2390 // The sharpened class might be unloaded if there is no class loader
2391 // contraint in place.
2392 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2393 // Sharpen the value type.
2394 result = TypeOopPtr::make_from_klass(sharpened_klass);
2395 if (null_free) {
2396 result = result->join_speculative(TypePtr::NOTNULL)->is_oopptr();
2397 }
2398 }
2399 if (result != nullptr) {
2400 #ifndef PRODUCT
2401 if (C->print_intrinsics() || C->print_inlining()) {
2402 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2403 tty->print(" sharpened value: "); result->dump(); tty->cr();
2404 }
2405 #endif
2406 }
2407 return result;
2408 }
2409
2410 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2411 switch (kind) {
2412 case Relaxed:
2413 return MO_UNORDERED;
2414 case Opaque:
2415 return MO_RELAXED;
2416 case Acquire:
2417 return MO_ACQUIRE;
2418 case Release:
2419 return MO_RELEASE;
2420 case Volatile:
2421 return MO_SEQ_CST;
2422 default:
2423 ShouldNotReachHere();
2424 return 0;
2425 }
2426 }
2427
2428 LibraryCallKit::SavedState::SavedState(LibraryCallKit* kit) :
2429 _kit(kit),
2430 _sp(kit->sp()),
2431 _jvms(kit->jvms()),
2432 _map(kit->clone_map()),
2433 _discarded(false)
2434 {
2435 for (DUIterator_Fast imax, i = kit->control()->fast_outs(imax); i < imax; i++) {
2436 Node* out = kit->control()->fast_out(i);
2437 if (out->is_CFG()) {
2438 _ctrl_succ.push(out);
2439 }
2440 }
2441 }
2442
2443 LibraryCallKit::SavedState::~SavedState() {
2444 if (_discarded) {
2445 _kit->destruct_map_clone(_map);
2446 return;
2447 }
2448 _kit->jvms()->set_map(_map);
2449 _kit->jvms()->set_sp(_sp);
2450 _map->set_jvms(_kit->jvms());
2451 _kit->set_map(_map);
2452 _kit->set_sp(_sp);
2453 for (DUIterator_Fast imax, i = _kit->control()->fast_outs(imax); i < imax; i++) {
2454 Node* out = _kit->control()->fast_out(i);
2455 if (out->is_CFG() && out->in(0) == _kit->control() && out != _kit->map() && !_ctrl_succ.member(out)) {
2456 _kit->_gvn.hash_delete(out);
2457 out->set_req(0, _kit->C->top());
2458 _kit->C->record_for_igvn(out);
2459 --i; --imax;
2460 _kit->_gvn.hash_find_insert(out);
2461 }
2462 }
2463 }
2464
2465 void LibraryCallKit::SavedState::discard() {
2466 _discarded = true;
2467 }
2468
2469 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2470 if (callee()->is_static()) return false; // caller must have the capability!
2471 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2472 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2473 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2474 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2475
2476 if (is_reference_type(type)) {
2477 decorators |= ON_UNKNOWN_OOP_REF;
2478 }
2479
2480 if (unaligned) {
2481 decorators |= C2_UNALIGNED;
2482 }
2483
2484 #ifndef PRODUCT
2485 {
2486 ResourceMark rm;
2487 // Check the signatures.
2488 ciSignature* sig = callee()->signature();
2489 #ifdef ASSERT
2490 if (!is_store) {
2491 // Object getReference(Object base, int/long offset), etc.
2492 BasicType rtype = sig->return_type()->basic_type();
2493 assert(rtype == type, "getter must return the expected value");
2494 assert(sig->count() == 2, "oop getter has 2 arguments");
2495 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2496 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2497 } else {
2498 // void putReference(Object base, int/long offset, Object x), etc.
2499 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2500 assert(sig->count() == 3, "oop putter has 3 arguments");
2501 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2502 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2503 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2504 assert(vtype == type, "putter must accept the expected value");
2505 }
2506 #endif // ASSERT
2507 }
2508 #endif //PRODUCT
2509
2510 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2511
2512 Node* receiver = argument(0); // type: oop
2513
2514 // Build address expression.
2515 Node* heap_base_oop = top();
2516
2517 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2518 Node* base = argument(1); // type: oop
2519 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2520 Node* offset = argument(2); // type: long
2521 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2522 // to be plain byte offsets, which are also the same as those accepted
2523 // by oopDesc::field_addr.
2524 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2525 "fieldOffset must be byte-scaled");
2526
2527 if (base->is_InlineType()) {
2528 assert(!is_store, "InlineTypeNodes are non-larval value objects");
2529 InlineTypeNode* vt = base->as_InlineType();
2530 if (offset->is_Con()) {
2531 long off = find_long_con(offset, 0);
2532 ciInlineKlass* vk = vt->type()->inline_klass();
2533 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2534 return false;
2535 }
2536
2537 ciField* field = vk->get_non_flat_field_by_offset(off);
2538 if (field != nullptr) {
2539 BasicType bt = type2field[field->type()->basic_type()];
2540 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2541 bt = T_OBJECT;
2542 }
2543 if (bt == type && !field->is_flat()) {
2544 Node* value = vt->field_value_by_offset(off, false);
2545 const Type* value_type = _gvn.type(value);
2546 if (value->is_InlineType()) {
2547 value = value->as_InlineType()->adjust_scalarization_depth(this);
2548 } else if (value_type->is_inlinetypeptr()) {
2549 value = InlineTypeNode::make_from_oop(this, value, value_type->inline_klass());
2550 }
2551 set_result(value);
2552 return true;
2553 }
2554 }
2555 }
2556 {
2557 // Re-execute the unsafe access if allocation triggers deoptimization.
2558 PreserveReexecuteState preexecs(this);
2559 jvms()->set_should_reexecute(true);
2560 vt = vt->buffer(this);
2561 }
2562 base = vt->get_oop();
2563 }
2564
2565 // 32-bit machines ignore the high half!
2566 offset = ConvL2X(offset);
2567
2568 // Save state and restore on bailout
2569 SavedState old_state(this);
2570
2571 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2572 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2573
2574 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2575 if (type != T_OBJECT) {
2576 decorators |= IN_NATIVE; // off-heap primitive access
2577 } else {
2578 return false; // off-heap oop accesses are not supported
2579 }
2580 } else {
2581 heap_base_oop = base; // on-heap or mixed access
2582 }
2583
2584 // Can base be null? Otherwise, always on-heap access.
2585 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2586
2587 if (!can_access_non_heap) {
2588 decorators |= IN_HEAP;
2589 }
2590
2591 Node* val = is_store ? argument(4) : nullptr;
2592
2593 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2594 if (adr_type == TypePtr::NULL_PTR) {
2595 return false; // off-heap access with zero address
2596 }
2597
2598 // Try to categorize the address.
2599 Compile::AliasType* alias_type = C->alias_type(adr_type);
2600 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2601
2602 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2603 alias_type->adr_type() == TypeAryPtr::RANGE) {
2604 return false; // not supported
2605 }
2606
2607 bool mismatched = false;
2608 BasicType bt = T_ILLEGAL;
2609 ciField* field = nullptr;
2610 if (adr_type->isa_instptr()) {
2611 const TypeInstPtr* instptr = adr_type->is_instptr();
2612 ciInstanceKlass* k = instptr->instance_klass();
2613 int off = instptr->offset();
2614 if (instptr->const_oop() != nullptr &&
2615 k == ciEnv::current()->Class_klass() &&
2616 instptr->offset() >= (k->size_helper() * wordSize)) {
2617 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2618 field = k->get_field_by_offset(off, true);
2619 } else {
2620 field = k->get_non_flat_field_by_offset(off);
2621 }
2622 if (field != nullptr) {
2623 bt = type2field[field->type()->basic_type()];
2624 }
2625 if (bt != alias_type->basic_type()) {
2626 // Type mismatch. Is it an access to a nested flat field?
2627 field = k->get_field_by_offset(off, false);
2628 if (field != nullptr) {
2629 bt = type2field[field->type()->basic_type()];
2630 }
2631 }
2632 assert(bt == alias_type->basic_type(), "should match");
2633 } else {
2634 bt = alias_type->basic_type();
2635 }
2636
2637 if (bt != T_ILLEGAL) {
2638 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2639 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2640 // Alias type doesn't differentiate between byte[] and boolean[]).
2641 // Use address type to get the element type.
2642 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2643 }
2644 if (is_reference_type(bt, true)) {
2645 // accessing an array field with getReference is not a mismatch
2646 bt = T_OBJECT;
2647 }
2648 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2649 // Don't intrinsify mismatched object accesses
2650 return false;
2651 }
2652 mismatched = (bt != type);
2653 } else if (alias_type->adr_type()->isa_oopptr()) {
2654 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2655 }
2656
2657 old_state.discard();
2658 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2659
2660 if (mismatched) {
2661 decorators |= C2_MISMATCHED;
2662 }
2663
2664 // First guess at the value type.
2665 const Type *value_type = Type::get_const_basic_type(type);
2666
2667 // Figure out the memory ordering.
2668 decorators |= mo_decorator_for_access_kind(kind);
2669
2670 if (!is_store) {
2671 if (type == T_OBJECT) {
2672 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2673 if (tjp != nullptr) {
2674 value_type = tjp;
2675 }
2676 }
2677 }
2678
2679 receiver = null_check(receiver);
2680 if (stopped()) {
2681 return true;
2682 }
2683 // Heap pointers get a null-check from the interpreter,
2684 // as a courtesy. However, this is not guaranteed by Unsafe,
2685 // and it is not possible to fully distinguish unintended nulls
2686 // from intended ones in this API.
2687
2688 if (!is_store) {
2689 Node* p = nullptr;
2690 // Try to constant fold a load from a constant field
2691
2692 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !field->is_flat() && !mismatched) {
2693 // final or stable field
2694 p = make_constant_from_field(field, heap_base_oop);
2695 }
2696
2697 if (p == nullptr) { // Could not constant fold the load
2698 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2699 const TypeOopPtr* ptr = value_type->make_oopptr();
2700 if (ptr != nullptr && ptr->is_inlinetypeptr()) {
2701 // Load a non-flattened inline type from memory
2702 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass());
2703 }
2704 // Normalize the value returned by getBoolean in the following cases
2705 if (type == T_BOOLEAN &&
2706 (mismatched ||
2707 heap_base_oop == top() || // - heap_base_oop is null or
2708 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2709 // and the unsafe access is made to large offset
2710 // (i.e., larger than the maximum offset necessary for any
2711 // field access)
2712 ) {
2713 IdealKit ideal = IdealKit(this);
2714 #define __ ideal.
2715 IdealVariable normalized_result(ideal);
2716 __ declarations_done();
2717 __ set(normalized_result, p);
2718 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2719 __ set(normalized_result, ideal.ConI(1));
2720 ideal.end_if();
2721 final_sync(ideal);
2722 p = __ value(normalized_result);
2723 #undef __
2724 }
2725 }
2726 if (type == T_ADDRESS) {
2727 p = gvn().transform(new CastP2XNode(nullptr, p));
2728 p = ConvX2UL(p);
2729 }
2730 // The load node has the control of the preceding MemBarCPUOrder. All
2731 // following nodes will have the control of the MemBarCPUOrder inserted at
2732 // the end of this method. So, pushing the load onto the stack at a later
2733 // point is fine.
2734 set_result(p);
2735 } else {
2736 if (bt == T_ADDRESS) {
2737 // Repackage the long as a pointer.
2738 val = ConvL2X(val);
2739 val = gvn().transform(new CastX2PNode(val));
2740 }
2741 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2742 }
2743
2744 return true;
2745 }
2746
2747 bool LibraryCallKit::inline_unsafe_flat_access(bool is_store, AccessKind kind) {
2748 #ifdef ASSERT
2749 {
2750 ResourceMark rm;
2751 // Check the signatures.
2752 ciSignature* sig = callee()->signature();
2753 assert(sig->type_at(0)->basic_type() == T_OBJECT, "base should be object, but is %s", type2name(sig->type_at(0)->basic_type()));
2754 assert(sig->type_at(1)->basic_type() == T_LONG, "offset should be long, but is %s", type2name(sig->type_at(1)->basic_type()));
2755 assert(sig->type_at(2)->basic_type() == T_INT, "layout kind should be int, but is %s", type2name(sig->type_at(3)->basic_type()));
2756 assert(sig->type_at(3)->basic_type() == T_OBJECT, "value klass should be object, but is %s", type2name(sig->type_at(4)->basic_type()));
2757 if (is_store) {
2758 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value, but returns %s", type2name(sig->return_type()->basic_type()));
2759 assert(sig->count() == 5, "flat putter should have 5 arguments, but has %d", sig->count());
2760 assert(sig->type_at(4)->basic_type() == T_OBJECT, "put value should be object, but is %s", type2name(sig->type_at(5)->basic_type()));
2761 } else {
2762 assert(sig->return_type()->basic_type() == T_OBJECT, "getter must return an object, but returns %s", type2name(sig->return_type()->basic_type()));
2763 assert(sig->count() == 4, "flat getter should have 4 arguments, but has %d", sig->count());
2764 }
2765 }
2766 #endif // ASSERT
2767
2768 assert(kind == Relaxed, "Only plain accesses for now");
2769 if (callee()->is_static()) {
2770 // caller must have the capability!
2771 return false;
2772 }
2773 C->set_has_unsafe_access(true);
2774
2775 const TypeInstPtr* value_klass_node = _gvn.type(argument(5))->isa_instptr();
2776 if (value_klass_node == nullptr || value_klass_node->const_oop() == nullptr) {
2777 // parameter valueType is not a constant
2778 return false;
2779 }
2780 ciType* mirror_type = value_klass_node->const_oop()->as_instance()->java_mirror_type();
2781 if (!mirror_type->is_inlinetype()) {
2782 // Dead code
2783 return false;
2784 }
2785 ciInlineKlass* value_klass = mirror_type->as_inline_klass();
2786
2787 const TypeInt* layout_type = _gvn.type(argument(4))->isa_int();
2788 if (layout_type == nullptr || !layout_type->is_con()) {
2789 // parameter layoutKind is not a constant
2790 return false;
2791 }
2792 assert(layout_type->get_con() >= static_cast<int>(LayoutKind::REFERENCE) &&
2793 layout_type->get_con() < static_cast<int>(LayoutKind::UNKNOWN),
2794 "invalid layoutKind %d", layout_type->get_con());
2795 LayoutKind layout = static_cast<LayoutKind>(layout_type->get_con());
2796 assert(layout == LayoutKind::REFERENCE || layout == LayoutKind::NULL_FREE_NON_ATOMIC_FLAT ||
2797 layout == LayoutKind::NULL_FREE_ATOMIC_FLAT || layout == LayoutKind::NULLABLE_ATOMIC_FLAT,
2798 "unexpected layoutKind %d", layout_type->get_con());
2799
2800 null_check(argument(0));
2801 if (stopped()) {
2802 return true;
2803 }
2804
2805 Node* base = must_be_not_null(argument(1), true);
2806 Node* offset = argument(2);
2807 const Type* base_type = _gvn.type(base);
2808
2809 Node* ptr;
2810 bool immutable_memory = false;
2811 DecoratorSet decorators = C2_UNSAFE_ACCESS | IN_HEAP | MO_UNORDERED;
2812 if (base_type->isa_instptr()) {
2813 const TypeLong* offset_type = _gvn.type(offset)->isa_long();
2814 if (offset_type == nullptr || !offset_type->is_con()) {
2815 // Offset into a non-array should be a constant
2816 decorators |= C2_MISMATCHED;
2817 } else {
2818 int offset_con = checked_cast<int>(offset_type->get_con());
2819 ciInstanceKlass* base_klass = base_type->is_instptr()->instance_klass();
2820 ciField* field = base_klass->get_non_flat_field_by_offset(offset_con);
2821 if (field == nullptr) {
2822 assert(!base_klass->is_final(), "non-existence field at offset %d of class %s", offset_con, base_klass->name()->as_utf8());
2823 decorators |= C2_MISMATCHED;
2824 } else {
2825 assert(field->type() == value_klass, "field at offset %d of %s is of type %s, but valueType is %s",
2826 offset_con, base_klass->name()->as_utf8(), field->type()->name(), value_klass->name()->as_utf8());
2827 immutable_memory = field->is_strict() && field->is_final();
2828
2829 if (base->is_InlineType()) {
2830 assert(!is_store, "Cannot store into a non-larval value object");
2831 set_result(base->as_InlineType()->field_value_by_offset(offset_con, false));
2832 return true;
2833 }
2834 }
2835 }
2836
2837 if (base->is_InlineType()) {
2838 assert(!is_store, "Cannot store into a non-larval value object");
2839 base = base->as_InlineType()->buffer(this, true);
2840 }
2841 ptr = basic_plus_adr(base, ConvL2X(offset));
2842 } else if (base_type->isa_aryptr()) {
2843 decorators |= IS_ARRAY;
2844 if (layout == LayoutKind::REFERENCE) {
2845 if (!base_type->is_aryptr()->is_not_flat()) {
2846 const TypeAryPtr* array_type = base_type->is_aryptr()->cast_to_not_flat();
2847 Node* new_base = _gvn.transform(new CastPPNode(control(), base, array_type, ConstraintCastNode::DependencyType::NonFloatingNarrowing));
2848 replace_in_map(base, new_base);
2849 base = new_base;
2850 }
2851 ptr = basic_plus_adr(base, ConvL2X(offset));
2852 } else {
2853 if (UseArrayFlattening) {
2854 // Flat array must have an exact type
2855 bool is_null_free = !LayoutKindHelper::is_nullable_flat(layout);
2856 bool is_atomic = LayoutKindHelper::is_atomic_flat(layout);
2857 Node* new_base = cast_to_flat_array_exact(base, value_klass, is_null_free, is_atomic);
2858 replace_in_map(base, new_base);
2859 base = new_base;
2860 ptr = basic_plus_adr(base, ConvL2X(offset));
2861 const TypeAryPtr* ptr_type = _gvn.type(ptr)->is_aryptr();
2862 if (ptr_type->field_offset().get() != 0) {
2863 ptr = _gvn.transform(new CastPPNode(control(), ptr, ptr_type->with_field_offset(0), ConstraintCastNode::DependencyType::NonFloatingNarrowing));
2864 }
2865 } else {
2866 uncommon_trap(Deoptimization::Reason_intrinsic,
2867 Deoptimization::Action_none);
2868 return true;
2869 }
2870 }
2871 } else {
2872 decorators |= C2_MISMATCHED;
2873 ptr = basic_plus_adr(base, ConvL2X(offset));
2874 }
2875
2876 if (is_store) {
2877 Node* value = argument(6);
2878 const Type* value_type = _gvn.type(value);
2879 if (!value_type->is_inlinetypeptr()) {
2880 value_type = Type::get_const_type(value_klass)->filter_speculative(value_type);
2881 Node* new_value = _gvn.transform(new CastPPNode(control(), value, value_type, ConstraintCastNode::DependencyType::NonFloatingNarrowing));
2882 new_value = InlineTypeNode::make_from_oop(this, new_value, value_klass);
2883 replace_in_map(value, new_value);
2884 value = new_value;
2885 }
2886
2887 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());
2888 if (layout == LayoutKind::REFERENCE) {
2889 const TypePtr* ptr_type = (decorators & C2_MISMATCHED) != 0 ? TypeRawPtr::BOTTOM : _gvn.type(ptr)->is_ptr();
2890 access_store_at(base, ptr, ptr_type, value, value_type, T_OBJECT, decorators);
2891 } else {
2892 bool atomic = LayoutKindHelper::is_atomic_flat(layout);
2893 bool null_free = !LayoutKindHelper::is_nullable_flat(layout);
2894 value->as_InlineType()->store_flat(this, base, ptr, atomic, immutable_memory, null_free, decorators);
2895 }
2896
2897 return true;
2898 } else {
2899 decorators |= (C2_CONTROL_DEPENDENT_LOAD | C2_UNKNOWN_CONTROL_LOAD);
2900 InlineTypeNode* result;
2901 if (layout == LayoutKind::REFERENCE) {
2902 const TypePtr* ptr_type = (decorators & C2_MISMATCHED) != 0 ? TypeRawPtr::BOTTOM : _gvn.type(ptr)->is_ptr();
2903 Node* oop = access_load_at(base, ptr, ptr_type, Type::get_const_type(value_klass), T_OBJECT, decorators);
2904 result = InlineTypeNode::make_from_oop(this, oop, value_klass);
2905 } else {
2906 bool atomic = LayoutKindHelper::is_atomic_flat(layout);
2907 bool null_free = !LayoutKindHelper::is_nullable_flat(layout);
2908 result = InlineTypeNode::make_from_flat(this, value_klass, base, ptr, atomic, immutable_memory, null_free, decorators);
2909 }
2910
2911 set_result(result);
2912 return true;
2913 }
2914 }
2915
2916 //----------------------------inline_unsafe_load_store----------------------------
2917 // This method serves a couple of different customers (depending on LoadStoreKind):
2918 //
2919 // LS_cmp_swap:
2920 //
2921 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2922 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2923 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2924 //
2925 // LS_cmp_swap_weak:
2926 //
2927 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2928 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2929 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2930 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2931 //
2932 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2933 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2934 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2935 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2936 //
2937 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2938 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2939 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2940 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2941 //
2942 // LS_cmp_exchange:
2943 //
2944 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2945 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2946 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2947 //
2948 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2949 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2950 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2951 //
2952 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2953 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2954 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2955 //
2956 // LS_get_add:
2957 //
2958 // int getAndAddInt( Object o, long offset, int delta)
2959 // long getAndAddLong(Object o, long offset, long delta)
2960 //
2961 // LS_get_set:
2962 //
2963 // int getAndSet(Object o, long offset, int newValue)
2964 // long getAndSet(Object o, long offset, long newValue)
2965 // Object getAndSet(Object o, long offset, Object newValue)
2966 //
2967 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2968 // This basic scheme here is the same as inline_unsafe_access, but
2969 // differs in enough details that combining them would make the code
2970 // overly confusing. (This is a true fact! I originally combined
2971 // them, but even I was confused by it!) As much code/comments as
2972 // possible are retained from inline_unsafe_access though to make
2973 // the correspondences clearer. - dl
2974
2975 if (callee()->is_static()) return false; // caller must have the capability!
2976
2977 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2978 decorators |= mo_decorator_for_access_kind(access_kind);
2979
2980 #ifndef PRODUCT
2981 BasicType rtype;
2982 {
2983 ResourceMark rm;
2984 // Check the signatures.
2985 ciSignature* sig = callee()->signature();
2986 rtype = sig->return_type()->basic_type();
2987 switch(kind) {
2988 case LS_get_add:
2989 case LS_get_set: {
2990 // Check the signatures.
2991 #ifdef ASSERT
2992 assert(rtype == type, "get and set must return the expected type");
2993 assert(sig->count() == 3, "get and set has 3 arguments");
2994 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2995 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2996 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2997 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2998 #endif // ASSERT
2999 break;
3000 }
3001 case LS_cmp_swap:
3002 case LS_cmp_swap_weak: {
3003 // Check the signatures.
3004 #ifdef ASSERT
3005 assert(rtype == T_BOOLEAN, "CAS must return boolean");
3006 assert(sig->count() == 4, "CAS has 4 arguments");
3007 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
3008 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
3009 #endif // ASSERT
3010 break;
3011 }
3012 case LS_cmp_exchange: {
3013 // Check the signatures.
3014 #ifdef ASSERT
3015 assert(rtype == type, "CAS must return the expected type");
3016 assert(sig->count() == 4, "CAS has 4 arguments");
3017 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
3018 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
3019 #endif // ASSERT
3020 break;
3021 }
3022 default:
3023 ShouldNotReachHere();
3024 }
3025 }
3026 #endif //PRODUCT
3027
3028 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
3029
3030 // Get arguments:
3031 Node* receiver = nullptr;
3032 Node* base = nullptr;
3033 Node* offset = nullptr;
3034 Node* oldval = nullptr;
3035 Node* newval = nullptr;
3036 switch(kind) {
3037 case LS_cmp_swap:
3038 case LS_cmp_swap_weak:
3039 case LS_cmp_exchange: {
3040 const bool two_slot_type = type2size[type] == 2;
3041 receiver = argument(0); // type: oop
3042 base = argument(1); // type: oop
3043 offset = argument(2); // type: long
3044 oldval = argument(4); // type: oop, int, or long
3045 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
3046 break;
3047 }
3048 case LS_get_add:
3049 case LS_get_set: {
3050 receiver = argument(0); // type: oop
3051 base = argument(1); // type: oop
3052 offset = argument(2); // type: long
3053 oldval = nullptr;
3054 newval = argument(4); // type: oop, int, or long
3055 break;
3056 }
3057 default:
3058 ShouldNotReachHere();
3059 }
3060
3061 // Build field offset expression.
3062 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
3063 // to be plain byte offsets, which are also the same as those accepted
3064 // by oopDesc::field_addr.
3065 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
3066 // 32-bit machines ignore the high half of long offsets
3067 offset = ConvL2X(offset);
3068 // Save state and restore on bailout
3069 SavedState old_state(this);
3070 Node* adr = make_unsafe_address(base, offset,type, false);
3071 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
3072
3073 Compile::AliasType* alias_type = C->alias_type(adr_type);
3074 BasicType bt = alias_type->basic_type();
3075 if (bt != T_ILLEGAL &&
3076 (is_reference_type(bt) != (type == T_OBJECT))) {
3077 // Don't intrinsify mismatched object accesses.
3078 return false;
3079 }
3080
3081 old_state.discard();
3082
3083 // For CAS, unlike inline_unsafe_access, there seems no point in
3084 // trying to refine types. Just use the coarse types here.
3085 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
3086 const Type *value_type = Type::get_const_basic_type(type);
3087
3088 switch (kind) {
3089 case LS_get_set:
3090 case LS_cmp_exchange: {
3091 if (type == T_OBJECT) {
3092 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
3093 if (tjp != nullptr) {
3094 value_type = tjp;
3095 }
3096 }
3097 break;
3098 }
3099 case LS_cmp_swap:
3100 case LS_cmp_swap_weak:
3101 case LS_get_add:
3102 break;
3103 default:
3104 ShouldNotReachHere();
3105 }
3106
3107 // Null check receiver.
3108 receiver = null_check(receiver);
3109 if (stopped()) {
3110 return true;
3111 }
3112
3113 int alias_idx = C->get_alias_index(adr_type);
3114
3115 if (is_reference_type(type)) {
3116 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
3117
3118 if (oldval != nullptr && oldval->is_InlineType()) {
3119 // Re-execute the unsafe access if allocation triggers deoptimization.
3120 PreserveReexecuteState preexecs(this);
3121 jvms()->set_should_reexecute(true);
3122 oldval = oldval->as_InlineType()->buffer(this)->get_oop();
3123 }
3124 if (newval != nullptr && newval->is_InlineType()) {
3125 // Re-execute the unsafe access if allocation triggers deoptimization.
3126 PreserveReexecuteState preexecs(this);
3127 jvms()->set_should_reexecute(true);
3128 newval = newval->as_InlineType()->buffer(this)->get_oop();
3129 }
3130
3131 // Transformation of a value which could be null pointer (CastPP #null)
3132 // could be delayed during Parse (for example, in adjust_map_after_if()).
3133 // Execute transformation here to avoid barrier generation in such case.
3134 if (_gvn.type(newval) == TypePtr::NULL_PTR)
3135 newval = _gvn.makecon(TypePtr::NULL_PTR);
3136
3137 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
3138 // Refine the value to a null constant, when it is known to be null
3139 oldval = _gvn.makecon(TypePtr::NULL_PTR);
3140 }
3141 }
3142
3143 Node* result = nullptr;
3144 switch (kind) {
3145 case LS_cmp_exchange: {
3146 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
3147 oldval, newval, value_type, type, decorators);
3148 break;
3149 }
3150 case LS_cmp_swap_weak:
3151 decorators |= C2_WEAK_CMPXCHG;
3152 case LS_cmp_swap: {
3153 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
3154 oldval, newval, value_type, type, decorators);
3155 break;
3156 }
3157 case LS_get_set: {
3158 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
3159 newval, value_type, type, decorators);
3160 break;
3161 }
3162 case LS_get_add: {
3163 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
3164 newval, value_type, type, decorators);
3165 break;
3166 }
3167 default:
3168 ShouldNotReachHere();
3169 }
3170
3171 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3172 set_result(result);
3173 return true;
3174 }
3175
3176 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3177 // Regardless of form, don't allow previous ld/st to move down,
3178 // then issue acquire, release, or volatile mem_bar.
3179 insert_mem_bar(Op_MemBarCPUOrder);
3180 switch(id) {
3181 case vmIntrinsics::_loadFence:
3182 insert_mem_bar(Op_LoadFence);
3183 return true;
3184 case vmIntrinsics::_storeFence:
3185 insert_mem_bar(Op_StoreFence);
3186 return true;
3187 case vmIntrinsics::_storeStoreFence:
3188 insert_mem_bar(Op_StoreStoreFence);
3189 return true;
3190 case vmIntrinsics::_fullFence:
3191 insert_mem_bar(Op_MemBarVolatile);
3192 return true;
3193 default:
3194 fatal_unexpected_iid(id);
3195 return false;
3196 }
3197 }
3198
3199 // private native int arrayInstanceBaseOffset0(Object[] array);
3200 bool LibraryCallKit::inline_arrayInstanceBaseOffset() {
3201 Node* array = argument(1);
3202 Node* klass_node = load_object_klass(array);
3203
3204 jint layout_con = Klass::_lh_neutral_value;
3205 Node* layout_val = get_layout_helper(klass_node, layout_con);
3206 int layout_is_con = (layout_val == nullptr);
3207
3208 Node* header_size = nullptr;
3209 if (layout_is_con) {
3210 int hsize = Klass::layout_helper_header_size(layout_con);
3211 header_size = intcon(hsize);
3212 } else {
3213 Node* hss = intcon(Klass::_lh_header_size_shift);
3214 Node* hsm = intcon(Klass::_lh_header_size_mask);
3215 header_size = _gvn.transform(new URShiftINode(layout_val, hss));
3216 header_size = _gvn.transform(new AndINode(header_size, hsm));
3217 }
3218 set_result(header_size);
3219 return true;
3220 }
3221
3222 // private native int arrayInstanceIndexScale0(Object[] array);
3223 bool LibraryCallKit::inline_arrayInstanceIndexScale() {
3224 Node* array = argument(1);
3225 Node* klass_node = load_object_klass(array);
3226
3227 jint layout_con = Klass::_lh_neutral_value;
3228 Node* layout_val = get_layout_helper(klass_node, layout_con);
3229 int layout_is_con = (layout_val == nullptr);
3230
3231 Node* element_size = nullptr;
3232 if (layout_is_con) {
3233 int log_element_size = Klass::layout_helper_log2_element_size(layout_con);
3234 int elem_size = 1 << log_element_size;
3235 element_size = intcon(elem_size);
3236 } else {
3237 Node* ess = intcon(Klass::_lh_log2_element_size_shift);
3238 Node* esm = intcon(Klass::_lh_log2_element_size_mask);
3239 Node* log_element_size = _gvn.transform(new URShiftINode(layout_val, ess));
3240 log_element_size = _gvn.transform(new AndINode(log_element_size, esm));
3241 element_size = _gvn.transform(new LShiftINode(intcon(1), log_element_size));
3242 }
3243 set_result(element_size);
3244 return true;
3245 }
3246
3247 // private native int arrayLayout0(Object[] array);
3248 bool LibraryCallKit::inline_arrayLayout() {
3249 RegionNode* region = new RegionNode(2);
3250 Node* phi = new PhiNode(region, TypeInt::POS);
3251
3252 Node* array = argument(1);
3253 Node* klass_node = load_object_klass(array);
3254 generate_refArray_guard(klass_node, region);
3255 if (region->req() == 3) {
3256 phi->add_req(intcon((jint)LayoutKind::REFERENCE));
3257 }
3258
3259 int layout_kind_offset = in_bytes(FlatArrayKlass::layout_kind_offset());
3260 Node* layout_kind_addr = basic_plus_adr(top(), klass_node, layout_kind_offset);
3261 Node* layout_kind = make_load(nullptr, layout_kind_addr, TypeInt::POS, T_INT, MemNode::unordered);
3262
3263 region->init_req(1, control());
3264 phi->init_req(1, layout_kind);
3265
3266 set_control(_gvn.transform(region));
3267 set_result(_gvn.transform(phi));
3268 return true;
3269 }
3270
3271 // private native int[] getFieldMap0(Class <?> c);
3272 // int offset = c._klass._acmp_maps_offset;
3273 // return (int[])c.obj_field(offset);
3274 bool LibraryCallKit::inline_getFieldMap() {
3275 Node* mirror = argument(1);
3276 Node* klass = load_klass_from_mirror(mirror, false, nullptr, 0);
3277
3278 int field_map_offset_offset = in_bytes(InstanceKlass::acmp_maps_offset_offset());
3279 Node* field_map_offset_addr = basic_plus_adr(top(), klass, field_map_offset_offset);
3280 Node* field_map_offset = make_load(nullptr, field_map_offset_addr, TypeInt::INT, T_INT, MemNode::unordered);
3281 field_map_offset = _gvn.transform(ConvI2L(field_map_offset));
3282
3283 Node* map_addr = basic_plus_adr(mirror, field_map_offset);
3284 const TypeAryPtr* val_type = TypeAryPtr::INTS->cast_to_ptr_type(TypePtr::NotNull)->with_offset(0);
3285 // TODO 8350865 Remove this
3286 val_type = val_type->cast_to_not_flat(true)->cast_to_not_null_free(true);
3287 Node* map = access_load_at(mirror, map_addr, TypeAryPtr::INTS, val_type, T_ARRAY, IN_HEAP | MO_UNORDERED);
3288
3289 set_result(map);
3290 return true;
3291 }
3292
3293 bool LibraryCallKit::inline_onspinwait() {
3294 insert_mem_bar(Op_OnSpinWait);
3295 return true;
3296 }
3297
3298 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3299 if (!kls->is_Con()) {
3300 return true;
3301 }
3302 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
3303 if (klsptr == nullptr) {
3304 return true;
3305 }
3306 ciInstanceKlass* ik = klsptr->instance_klass();
3307 // don't need a guard for a klass that is already initialized
3308 return !ik->is_initialized();
3309 }
3310
3311 //----------------------------inline_unsafe_writeback0-------------------------
3312 // public native void Unsafe.writeback0(long address)
3313 bool LibraryCallKit::inline_unsafe_writeback0() {
3314 if (!Matcher::has_match_rule(Op_CacheWB)) {
3315 return false;
3316 }
3317 #ifndef PRODUCT
3318 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
3319 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
3320 ciSignature* sig = callee()->signature();
3321 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
3322 #endif
3323 null_check_receiver(); // null-check, then ignore
3324 Node *addr = argument(1);
3325 addr = new CastX2PNode(addr);
3326 addr = _gvn.transform(addr);
3327 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
3328 flush = _gvn.transform(flush);
3329 set_memory(flush, TypeRawPtr::BOTTOM);
3330 return true;
3331 }
3332
3333 //----------------------------inline_unsafe_writeback0-------------------------
3334 // public native void Unsafe.writeback0(long address)
3335 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
3336 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
3337 return false;
3338 }
3339 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
3340 return false;
3341 }
3342 #ifndef PRODUCT
3343 assert(Matcher::has_match_rule(Op_CacheWB),
3344 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
3345 : "found match rule for CacheWBPostSync but not CacheWB"));
3346
3347 #endif
3348 null_check_receiver(); // null-check, then ignore
3349 Node *sync;
3350 if (is_pre) {
3351 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3352 } else {
3353 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3354 }
3355 sync = _gvn.transform(sync);
3356 set_memory(sync, TypeRawPtr::BOTTOM);
3357 return true;
3358 }
3359
3360 //----------------------------inline_unsafe_allocate---------------------------
3361 // public native Object Unsafe.allocateInstance(Class<?> cls);
3362 bool LibraryCallKit::inline_unsafe_allocate() {
3363
3364 #if INCLUDE_JVMTI
3365 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3366 return false;
3367 }
3368 #endif //INCLUDE_JVMTI
3369
3370 if (callee()->is_static()) return false; // caller must have the capability!
3371
3372 null_check_receiver(); // null-check, then ignore
3373 Node* cls = null_check(argument(1));
3374 if (stopped()) return true;
3375
3376 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
3377 kls = null_check(kls);
3378 if (stopped()) return true; // argument was like int.class
3379
3380 #if INCLUDE_JVMTI
3381 // Don't try to access new allocated obj in the intrinsic.
3382 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
3383 // Deoptimize and allocate in interpreter instead.
3384 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
3385 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
3386 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
3387 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
3388 {
3389 BuildCutout unless(this, tst, PROB_MAX);
3390 uncommon_trap(Deoptimization::Reason_intrinsic,
3391 Deoptimization::Action_make_not_entrant);
3392 }
3393 if (stopped()) {
3394 return true;
3395 }
3396 #endif //INCLUDE_JVMTI
3397
3398 Node* test = nullptr;
3399 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3400 // Note: The argument might still be an illegal value like
3401 // Serializable.class or Object[].class. The runtime will handle it.
3402 // But we must make an explicit check for initialization.
3403 Node* insp = basic_plus_adr(top(), kls, in_bytes(InstanceKlass::init_state_offset()));
3404 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3405 // can generate code to load it as unsigned byte.
3406 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
3407 Node* bits = intcon(InstanceKlass::fully_initialized);
3408 test = _gvn.transform(new SubINode(inst, bits));
3409 // The 'test' is non-zero if we need to take a slow path.
3410 }
3411 Node* obj = new_instance(kls, test);
3412 set_result(obj);
3413 return true;
3414 }
3415
3416 //------------------------inline_native_time_funcs--------------
3417 // inline code for System.currentTimeMillis() and System.nanoTime()
3418 // these have the same type and signature
3419 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3420 const TypeFunc* tf = OptoRuntime::void_long_Type();
3421 const TypePtr* no_memory_effects = nullptr;
3422 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3423 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3424 #ifdef ASSERT
3425 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3426 assert(value_top == top(), "second value must be top");
3427 #endif
3428 set_result(value);
3429 return true;
3430 }
3431
3432 //--------------------inline_native_vthread_start_transition--------------------
3433 // inline void startTransition(boolean is_mount);
3434 // inline void startFinalTransition();
3435 // Pseudocode of implementation:
3436 //
3437 // java_lang_Thread::set_is_in_vthread_transition(vthread, true);
3438 // carrier->set_is_in_vthread_transition(true);
3439 // OrderAccess::storeload();
3440 // int disable_requests = java_lang_Thread::vthread_transition_disable_count(vthread)
3441 // + global_vthread_transition_disable_count();
3442 // if (disable_requests > 0) {
3443 // slow path: runtime call
3444 // }
3445 bool LibraryCallKit::inline_native_vthread_start_transition(address funcAddr, const char* funcName, bool is_final_transition) {
3446 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3447 IdealKit ideal(this);
3448
3449 Node* thread = ideal.thread();
3450 Node* jt_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3451 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3452 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3453 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3454 insert_mem_bar(Op_MemBarVolatile);
3455 ideal.sync_kit(this);
3456
3457 Node* global_disable_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::global_vthread_transition_disable_count_address()));
3458 Node* global_disable = ideal.load(ideal.ctrl(), global_disable_addr, TypeInt::INT, T_INT, Compile::AliasIdxRaw, true /*require_atomic_access*/);
3459 Node* vt_disable_addr = basic_plus_adr(vt_oop, java_lang_Thread::vthread_transition_disable_count_offset());
3460 Node* vt_disable = ideal.load(ideal.ctrl(), vt_disable_addr, TypeInt::INT, T_INT, Compile::AliasIdxRaw, true /*require_atomic_access*/);
3461 Node* disabled = _gvn.transform(new AddINode(global_disable, vt_disable));
3462
3463 ideal.if_then(disabled, BoolTest::ne, ideal.ConI(0)); {
3464 sync_kit(ideal);
3465 Node* is_mount = is_final_transition ? ideal.ConI(0) : _gvn.transform(argument(1));
3466 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3467 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3468 ideal.sync_kit(this);
3469 }
3470 ideal.end_if();
3471
3472 final_sync(ideal);
3473 return true;
3474 }
3475
3476 bool LibraryCallKit::inline_native_vthread_end_transition(address funcAddr, const char* funcName, bool is_first_transition) {
3477 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3478 IdealKit ideal(this);
3479
3480 Node* _notify_jvmti_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::notify_jvmti_events_address()));
3481 Node* _notify_jvmti = ideal.load(ideal.ctrl(), _notify_jvmti_addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3482
3483 ideal.if_then(_notify_jvmti, BoolTest::eq, ideal.ConI(1)); {
3484 sync_kit(ideal);
3485 Node* is_mount = is_first_transition ? ideal.ConI(1) : _gvn.transform(argument(1));
3486 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3487 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3488 ideal.sync_kit(this);
3489 } ideal.else_(); {
3490 Node* thread = ideal.thread();
3491 Node* jt_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3492 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3493
3494 sync_kit(ideal);
3495 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3496 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3497 ideal.sync_kit(this);
3498 } ideal.end_if();
3499
3500 final_sync(ideal);
3501 return true;
3502 }
3503
3504 #if INCLUDE_JVMTI
3505
3506 // Always update the is_disable_suspend bit.
3507 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3508 if (!DoJVMTIVirtualThreadTransitions) {
3509 return true;
3510 }
3511 IdealKit ideal(this);
3512
3513 {
3514 // unconditionally update the is_disable_suspend bit in current JavaThread
3515 Node* thread = ideal.thread();
3516 Node* arg = _gvn.transform(argument(0)); // argument for notification
3517 Node* addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3518 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3519
3520 sync_kit(ideal);
3521 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3522 ideal.sync_kit(this);
3523 }
3524 final_sync(ideal);
3525
3526 return true;
3527 }
3528
3529 #endif // INCLUDE_JVMTI
3530
3531 #ifdef JFR_HAVE_INTRINSICS
3532
3533 /**
3534 * if oop->klass != null
3535 * // normal class
3536 * epoch = _epoch_state ? 2 : 1
3537 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3538 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3539 * }
3540 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3541 * else
3542 * // primitive class
3543 * if oop->array_klass != null
3544 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3545 * else
3546 * id = LAST_TYPE_ID + 1 // void class path
3547 * if (!signaled)
3548 * signaled = true
3549 */
3550 bool LibraryCallKit::inline_native_classID() {
3551 Node* cls = argument(0);
3552
3553 IdealKit ideal(this);
3554 #define __ ideal.
3555 IdealVariable result(ideal); __ declarations_done();
3556 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3557 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3558 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3559
3560
3561 __ if_then(kls, BoolTest::ne, null()); {
3562 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3563 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3564
3565 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3566 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3567 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3568 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3569 mask = _gvn.transform(new OrLNode(mask, epoch));
3570 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3571
3572 float unlikely = PROB_UNLIKELY(0.999);
3573 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3574 sync_kit(ideal);
3575 make_runtime_call(RC_LEAF,
3576 OptoRuntime::class_id_load_barrier_Type(),
3577 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3578 "class id load barrier",
3579 TypePtr::BOTTOM,
3580 kls);
3581 ideal.sync_kit(this);
3582 } __ end_if();
3583
3584 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3585 } __ else_(); {
3586 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3587 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3588 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3589 __ if_then(array_kls, BoolTest::ne, null()); {
3590 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3591 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3592 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3593 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3594 } __ else_(); {
3595 // void class case
3596 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3597 } __ end_if();
3598
3599 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3600 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3601 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3602 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3603 } __ end_if();
3604 } __ end_if();
3605
3606 final_sync(ideal);
3607 set_result(ideal.value(result));
3608 #undef __
3609 return true;
3610 }
3611
3612 //------------------------inline_native_jvm_commit------------------
3613 bool LibraryCallKit::inline_native_jvm_commit() {
3614 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3615
3616 // Save input memory and i_o state.
3617 Node* input_memory_state = reset_memory();
3618 set_all_memory(input_memory_state);
3619 Node* input_io_state = i_o();
3620
3621 // TLS.
3622 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3623 // Jfr java buffer.
3624 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3625 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3626 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3627
3628 // Load the current value of the notified field in the JfrThreadLocal.
3629 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3630 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3631
3632 // Test for notification.
3633 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3634 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3635 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3636
3637 // True branch, is notified.
3638 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3639 set_control(is_notified);
3640
3641 // Reset notified state.
3642 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3643 Node* notified_reset_memory = reset_memory();
3644
3645 // 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.
3646 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3647 // Convert the machine-word to a long.
3648 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3649
3650 // False branch, not notified.
3651 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3652 set_control(not_notified);
3653 set_all_memory(input_memory_state);
3654
3655 // Arg is the next position as a long.
3656 Node* arg = argument(0);
3657 // Convert long to machine-word.
3658 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3659
3660 // Store the next_position to the underlying jfr java buffer.
3661 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3662
3663 Node* commit_memory = reset_memory();
3664 set_all_memory(commit_memory);
3665
3666 // 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.
3667 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3668 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3669 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3670
3671 // And flags with lease constant.
3672 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3673
3674 // Branch on lease to conditionalize returning the leased java buffer.
3675 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3676 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3677 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3678
3679 // False branch, not a lease.
3680 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3681
3682 // True branch, is lease.
3683 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3684 set_control(is_lease);
3685
3686 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3687 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3688 OptoRuntime::void_void_Type(),
3689 SharedRuntime::jfr_return_lease(),
3690 "return_lease", TypePtr::BOTTOM);
3691 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3692
3693 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3694 record_for_igvn(lease_compare_rgn);
3695 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3696 record_for_igvn(lease_compare_mem);
3697 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3698 record_for_igvn(lease_compare_io);
3699 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3700 record_for_igvn(lease_result_value);
3701
3702 // Update control and phi nodes.
3703 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3704 lease_compare_rgn->init_req(_false_path, not_lease);
3705
3706 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3707 lease_compare_mem->init_req(_false_path, commit_memory);
3708
3709 lease_compare_io->init_req(_true_path, i_o());
3710 lease_compare_io->init_req(_false_path, input_io_state);
3711
3712 lease_result_value->init_req(_true_path, _gvn.longcon(0)); // if the lease was returned, return 0L.
3713 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3714
3715 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3716 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3717 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3718 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3719
3720 // Update control and phi nodes.
3721 result_rgn->init_req(_true_path, is_notified);
3722 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3723
3724 result_mem->init_req(_true_path, notified_reset_memory);
3725 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3726
3727 result_io->init_req(_true_path, input_io_state);
3728 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3729
3730 result_value->init_req(_true_path, current_pos);
3731 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3732
3733 // Set output state.
3734 set_control(_gvn.transform(result_rgn));
3735 set_all_memory(_gvn.transform(result_mem));
3736 set_i_o(_gvn.transform(result_io));
3737 set_result(result_rgn, result_value);
3738 return true;
3739 }
3740
3741 /*
3742 * The intrinsic is a model of this pseudo-code:
3743 *
3744 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3745 * jobject h_event_writer = tl->java_event_writer();
3746 * if (h_event_writer == nullptr) {
3747 * return nullptr;
3748 * }
3749 * oop threadObj = Thread::threadObj();
3750 * oop vthread = java_lang_Thread::vthread(threadObj);
3751 * traceid tid;
3752 * bool pinVirtualThread;
3753 * bool excluded;
3754 * if (vthread != threadObj) { // i.e. current thread is virtual
3755 * tid = java_lang_Thread::tid(vthread);
3756 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3757 * pinVirtualThread = VMContinuations;
3758 * excluded = vthread_epoch_raw & excluded_mask;
3759 * if (!excluded) {
3760 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3761 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3762 * if (vthread_epoch != current_epoch) {
3763 * write_checkpoint();
3764 * }
3765 * }
3766 * } else {
3767 * tid = java_lang_Thread::tid(threadObj);
3768 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3769 * pinVirtualThread = false;
3770 * excluded = thread_epoch_raw & excluded_mask;
3771 * }
3772 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3773 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3774 * if (tid_in_event_writer != tid) {
3775 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3776 * setField(event_writer, "excluded", excluded);
3777 * setField(event_writer, "threadID", tid);
3778 * }
3779 * return event_writer
3780 */
3781 bool LibraryCallKit::inline_native_getEventWriter() {
3782 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3783
3784 // Save input memory and i_o state.
3785 Node* input_memory_state = reset_memory();
3786 set_all_memory(input_memory_state);
3787 Node* input_io_state = i_o();
3788
3789 // The most significant bit of the u2 is used to denote thread exclusion
3790 Node* excluded_shift = _gvn.intcon(15);
3791 Node* excluded_mask = _gvn.intcon(1 << 15);
3792 // The epoch generation is the range [1-32767]
3793 Node* epoch_mask = _gvn.intcon(32767);
3794
3795 // TLS
3796 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3797
3798 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3799 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3800
3801 // Load the eventwriter jobject handle.
3802 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3803
3804 // Null check the jobject handle.
3805 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3806 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3807 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3808
3809 // False path, jobj is null.
3810 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3811
3812 // True path, jobj is not null.
3813 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3814
3815 set_control(jobj_is_not_null);
3816
3817 // Load the threadObj for the CarrierThread.
3818 Node* threadObj = generate_current_thread(tls_ptr);
3819
3820 // Load the vthread.
3821 Node* vthread = generate_virtual_thread(tls_ptr);
3822
3823 // If vthread != threadObj, this is a virtual thread.
3824 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3825 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3826 IfNode* iff_vthread_not_equal_threadObj =
3827 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3828
3829 // False branch, fallback to threadObj.
3830 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3831 set_control(vthread_equal_threadObj);
3832
3833 // Load the tid field from the vthread object.
3834 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3835
3836 // Load the raw epoch value from the threadObj.
3837 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3838 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3839 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3840 TypeInt::CHAR, T_CHAR,
3841 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3842
3843 // Mask off the excluded information from the epoch.
3844 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3845
3846 // True branch, this is a virtual thread.
3847 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3848 set_control(vthread_not_equal_threadObj);
3849
3850 // Load the tid field from the vthread object.
3851 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3852
3853 // Continuation support determines if a virtual thread should be pinned.
3854 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3855 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3856
3857 // Load the raw epoch value from the vthread.
3858 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3859 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3860 TypeInt::CHAR, T_CHAR,
3861 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3862
3863 // Mask off the excluded information from the epoch.
3864 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3865
3866 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3867 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3868 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3869 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3870
3871 // False branch, vthread is excluded, no need to write epoch info.
3872 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3873
3874 // True branch, vthread is included, update epoch info.
3875 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3876 set_control(included);
3877
3878 // Get epoch value.
3879 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3880
3881 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3882 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3883 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3884
3885 // Compare the epoch in the vthread to the current epoch generation.
3886 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3887 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3888 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3889
3890 // False path, epoch is equal, checkpoint information is valid.
3891 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3892
3893 // True path, epoch is not equal, write a checkpoint for the vthread.
3894 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3895
3896 set_control(epoch_is_not_equal);
3897
3898 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3899 // The call also updates the native thread local thread id and the vthread with the current epoch.
3900 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3901 OptoRuntime::jfr_write_checkpoint_Type(),
3902 SharedRuntime::jfr_write_checkpoint(),
3903 "write_checkpoint", TypePtr::BOTTOM);
3904 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3905
3906 // vthread epoch != current epoch
3907 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3908 record_for_igvn(epoch_compare_rgn);
3909 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3910 record_for_igvn(epoch_compare_mem);
3911 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3912 record_for_igvn(epoch_compare_io);
3913
3914 // Update control and phi nodes.
3915 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3916 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3917 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3918 epoch_compare_mem->init_req(_false_path, input_memory_state);
3919 epoch_compare_io->init_req(_true_path, i_o());
3920 epoch_compare_io->init_req(_false_path, input_io_state);
3921
3922 // excluded != true
3923 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3924 record_for_igvn(exclude_compare_rgn);
3925 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3926 record_for_igvn(exclude_compare_mem);
3927 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3928 record_for_igvn(exclude_compare_io);
3929
3930 // Update control and phi nodes.
3931 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3932 exclude_compare_rgn->init_req(_false_path, excluded);
3933 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3934 exclude_compare_mem->init_req(_false_path, input_memory_state);
3935 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3936 exclude_compare_io->init_req(_false_path, input_io_state);
3937
3938 // vthread != threadObj
3939 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3940 record_for_igvn(vthread_compare_rgn);
3941 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3942 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3943 record_for_igvn(vthread_compare_io);
3944 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3945 record_for_igvn(tid);
3946 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3947 record_for_igvn(exclusion);
3948 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3949 record_for_igvn(pinVirtualThread);
3950
3951 // Update control and phi nodes.
3952 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3953 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3954 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3955 vthread_compare_mem->init_req(_false_path, input_memory_state);
3956 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3957 vthread_compare_io->init_req(_false_path, input_io_state);
3958 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3959 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3960 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3961 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3962 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3963 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3964
3965 // Update branch state.
3966 set_control(_gvn.transform(vthread_compare_rgn));
3967 set_all_memory(_gvn.transform(vthread_compare_mem));
3968 set_i_o(_gvn.transform(vthread_compare_io));
3969
3970 // Load the event writer oop by dereferencing the jobject handle.
3971 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3972 assert(klass_EventWriter->is_loaded(), "invariant");
3973 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3974 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3975 const TypeOopPtr* const xtype = aklass->as_instance_type();
3976 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3977 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3978
3979 // Load the current thread id from the event writer object.
3980 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3981 // Get the field offset to, conditionally, store an updated tid value later.
3982 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3983 // Get the field offset to, conditionally, store an updated exclusion value later.
3984 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3985 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3986 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3987
3988 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3989 record_for_igvn(event_writer_tid_compare_rgn);
3990 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3991 record_for_igvn(event_writer_tid_compare_mem);
3992 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3993 record_for_igvn(event_writer_tid_compare_io);
3994
3995 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3996 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3997 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3998 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3999
4000 // False path, tids are the same.
4001 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
4002
4003 // True path, tid is not equal, need to update the tid in the event writer.
4004 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
4005 record_for_igvn(tid_is_not_equal);
4006
4007 // Store the pin state to the event writer.
4008 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
4009
4010 // Store the exclusion state to the event writer.
4011 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
4012 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
4013
4014 // Store the tid to the event writer.
4015 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
4016
4017 // Update control and phi nodes.
4018 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
4019 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
4020 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
4021 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
4022 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
4023 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
4024
4025 // Result of top level CFG, Memory, IO and Value.
4026 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
4027 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
4028 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
4029 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
4030
4031 // Result control.
4032 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
4033 result_rgn->init_req(_false_path, jobj_is_null);
4034
4035 // Result memory.
4036 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
4037 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
4038
4039 // Result IO.
4040 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
4041 result_io->init_req(_false_path, _gvn.transform(input_io_state));
4042
4043 // Result value.
4044 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
4045 result_value->init_req(_false_path, null()); // return null
4046
4047 // Set output state.
4048 set_control(_gvn.transform(result_rgn));
4049 set_all_memory(_gvn.transform(result_mem));
4050 set_i_o(_gvn.transform(result_io));
4051 set_result(result_rgn, result_value);
4052 return true;
4053 }
4054
4055 /*
4056 * The intrinsic is a model of this pseudo-code:
4057 *
4058 * JfrThreadLocal* const tl = thread->jfr_thread_local();
4059 * if (carrierThread != thread) { // is virtual thread
4060 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
4061 * bool excluded = vthread_epoch_raw & excluded_mask;
4062 * AtomicAccess::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
4063 * AtomicAccess::store(&tl->_contextual_thread_excluded, is_excluded);
4064 * if (!excluded) {
4065 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
4066 * AtomicAccess::store(&tl->_vthread_epoch, vthread_epoch);
4067 * }
4068 * AtomicAccess::release_store(&tl->_vthread, true);
4069 * return;
4070 * }
4071 * AtomicAccess::release_store(&tl->_vthread, false);
4072 */
4073 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
4074 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
4075
4076 Node* input_memory_state = reset_memory();
4077 set_all_memory(input_memory_state);
4078
4079 // The most significant bit of the u2 is used to denote thread exclusion
4080 Node* excluded_mask = _gvn.intcon(1 << 15);
4081 // The epoch generation is the range [1-32767]
4082 Node* epoch_mask = _gvn.intcon(32767);
4083
4084 Node* const carrierThread = generate_current_thread(jt);
4085 // If thread != carrierThread, this is a virtual thread.
4086 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
4087 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
4088 IfNode* iff_thread_not_equal_carrierThread =
4089 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
4090
4091 Node* vthread_offset = basic_plus_adr(top(), jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
4092
4093 // False branch, is carrierThread.
4094 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
4095 // Store release
4096 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
4097
4098 set_all_memory(input_memory_state);
4099
4100 // True branch, is virtual thread.
4101 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
4102 set_control(thread_not_equal_carrierThread);
4103
4104 // Load the raw epoch value from the vthread.
4105 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
4106 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
4107 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
4108
4109 // Mask off the excluded information from the epoch.
4110 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
4111
4112 // Load the tid field from the thread.
4113 Node* tid = load_field_from_object(thread, "tid", "J");
4114
4115 // Store the vthread tid to the jfr thread local.
4116 Node* thread_id_offset = basic_plus_adr(top(), jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
4117 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
4118
4119 // Branch is_excluded to conditionalize updating the epoch .
4120 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
4121 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
4122 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
4123
4124 // True branch, vthread is excluded, no need to write epoch info.
4125 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
4126 set_control(excluded);
4127 Node* vthread_is_excluded = _gvn.intcon(1);
4128
4129 // False branch, vthread is included, update epoch info.
4130 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
4131 set_control(included);
4132 Node* vthread_is_included = _gvn.intcon(0);
4133
4134 // Get epoch value.
4135 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
4136
4137 // Store the vthread epoch to the jfr thread local.
4138 Node* vthread_epoch_offset = basic_plus_adr(top(), jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
4139 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
4140
4141 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
4142 record_for_igvn(excluded_rgn);
4143 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
4144 record_for_igvn(excluded_mem);
4145 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
4146 record_for_igvn(exclusion);
4147
4148 // Merge the excluded control and memory.
4149 excluded_rgn->init_req(_true_path, excluded);
4150 excluded_rgn->init_req(_false_path, included);
4151 excluded_mem->init_req(_true_path, tid_memory);
4152 excluded_mem->init_req(_false_path, included_memory);
4153 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
4154 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
4155
4156 // Set intermediate state.
4157 set_control(_gvn.transform(excluded_rgn));
4158 set_all_memory(excluded_mem);
4159
4160 // Store the vthread exclusion state to the jfr thread local.
4161 Node* thread_local_excluded_offset = basic_plus_adr(top(), jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
4162 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
4163
4164 // Store release
4165 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
4166
4167 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
4168 record_for_igvn(thread_compare_rgn);
4169 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
4170 record_for_igvn(thread_compare_mem);
4171 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
4172 record_for_igvn(vthread);
4173
4174 // Merge the thread_compare control and memory.
4175 thread_compare_rgn->init_req(_true_path, control());
4176 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
4177 thread_compare_mem->init_req(_true_path, vthread_true_memory);
4178 thread_compare_mem->init_req(_false_path, vthread_false_memory);
4179
4180 // Set output state.
4181 set_control(_gvn.transform(thread_compare_rgn));
4182 set_all_memory(_gvn.transform(thread_compare_mem));
4183 }
4184
4185 #endif // JFR_HAVE_INTRINSICS
4186
4187 //------------------------inline_native_currentCarrierThread------------------
4188 bool LibraryCallKit::inline_native_currentCarrierThread() {
4189 Node* junk = nullptr;
4190 set_result(generate_current_thread(junk));
4191 return true;
4192 }
4193
4194 //------------------------inline_native_currentThread------------------
4195 bool LibraryCallKit::inline_native_currentThread() {
4196 Node* junk = nullptr;
4197 set_result(generate_virtual_thread(junk));
4198 return true;
4199 }
4200
4201 //------------------------inline_native_setVthread------------------
4202 bool LibraryCallKit::inline_native_setCurrentThread() {
4203 assert(C->method()->changes_current_thread(),
4204 "method changes current Thread but is not annotated ChangesCurrentThread");
4205 Node* arr = argument(1);
4206 Node* thread = _gvn.transform(new ThreadLocalNode());
4207 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
4208 Node* thread_obj_handle
4209 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
4210 thread_obj_handle = _gvn.transform(thread_obj_handle);
4211 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
4212 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
4213
4214 // Change the _monitor_owner_id of the JavaThread
4215 Node* tid = load_field_from_object(arr, "tid", "J");
4216 Node* monitor_owner_id_offset = basic_plus_adr(top(), thread, in_bytes(JavaThread::monitor_owner_id_offset()));
4217 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
4218
4219 JFR_ONLY(extend_setCurrentThread(thread, arr);)
4220 return true;
4221 }
4222
4223 const Type* LibraryCallKit::scopedValueCache_type() {
4224 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
4225 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
4226 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS, /* stable= */ false, /* flat= */ false, /* not_flat= */ true, /* not_null_free= */ true, true);
4227
4228 // Because we create the scopedValue cache lazily we have to make the
4229 // type of the result BotPTR.
4230 bool xk = etype->klass_is_exact();
4231 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, TypeAryPtr::Offset(0));
4232 return objects_type;
4233 }
4234
4235 Node* LibraryCallKit::scopedValueCache_helper() {
4236 Node* thread = _gvn.transform(new ThreadLocalNode());
4237 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
4238 // We cannot use immutable_memory() because we might flip onto a
4239 // different carrier thread, at which point we'll need to use that
4240 // carrier thread's cache.
4241 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
4242 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
4243 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
4244 }
4245
4246 //------------------------inline_native_scopedValueCache------------------
4247 bool LibraryCallKit::inline_native_scopedValueCache() {
4248 Node* cache_obj_handle = scopedValueCache_helper();
4249 const Type* objects_type = scopedValueCache_type();
4250 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
4251
4252 return true;
4253 }
4254
4255 //------------------------inline_native_setScopedValueCache------------------
4256 bool LibraryCallKit::inline_native_setScopedValueCache() {
4257 Node* arr = argument(0);
4258 Node* cache_obj_handle = scopedValueCache_helper();
4259 const Type* objects_type = scopedValueCache_type();
4260
4261 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
4262 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
4263
4264 return true;
4265 }
4266
4267 //------------------------inline_native_Continuation_pin and unpin-----------
4268
4269 // Shared implementation routine for both pin and unpin.
4270 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
4271 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
4272
4273 // Save input memory.
4274 Node* input_memory_state = reset_memory();
4275 set_all_memory(input_memory_state);
4276
4277 // TLS
4278 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
4279 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
4280 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
4281
4282 // Null check the last continuation object.
4283 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
4284 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
4285 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
4286
4287 // False path, last continuation is null.
4288 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
4289
4290 // True path, last continuation is not null.
4291 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
4292
4293 set_control(continuation_is_not_null);
4294
4295 // Load the pin count from the last continuation.
4296 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
4297 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
4298
4299 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
4300 Node* pin_count_rhs;
4301 if (unpin) {
4302 pin_count_rhs = _gvn.intcon(0);
4303 } else {
4304 pin_count_rhs = _gvn.intcon(UINT32_MAX);
4305 }
4306 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
4307 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
4308 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
4309
4310 // True branch, pin count over/underflow.
4311 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
4312 {
4313 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
4314 // which will throw IllegalStateException for pin count over/underflow.
4315 // No memory changed so far - we can use memory create by reset_memory()
4316 // at the beginning of this intrinsic. No need to call reset_memory() again.
4317 PreserveJVMState pjvms(this);
4318 set_control(pin_count_over_underflow);
4319 uncommon_trap(Deoptimization::Reason_intrinsic,
4320 Deoptimization::Action_none);
4321 assert(stopped(), "invariant");
4322 }
4323
4324 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
4325 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
4326 set_control(valid_pin_count);
4327
4328 Node* next_pin_count;
4329 if (unpin) {
4330 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
4331 } else {
4332 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
4333 }
4334
4335 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
4336
4337 // Result of top level CFG and Memory.
4338 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
4339 record_for_igvn(result_rgn);
4340 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
4341 record_for_igvn(result_mem);
4342
4343 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
4344 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
4345 result_mem->init_req(_true_path, _gvn.transform(reset_memory()));
4346 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
4347
4348 // Set output state.
4349 set_control(_gvn.transform(result_rgn));
4350 set_all_memory(_gvn.transform(result_mem));
4351
4352 return true;
4353 }
4354
4355 //---------------------------load_mirror_from_klass----------------------------
4356 // Given a klass oop, load its java mirror (a java.lang.Class oop).
4357 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
4358 Node* p = basic_plus_adr(top(), klass, in_bytes(Klass::java_mirror_offset()));
4359 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4360 // mirror = ((OopHandle)mirror)->resolve();
4361 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
4362 }
4363
4364 //-----------------------load_klass_from_mirror_common-------------------------
4365 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
4366 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
4367 // and branch to the given path on the region.
4368 // If never_see_null, take an uncommon trap on null, so we can optimistically
4369 // compile for the non-null case.
4370 // If the region is null, force never_see_null = true.
4371 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
4372 bool never_see_null,
4373 RegionNode* region,
4374 int null_path,
4375 int offset) {
4376 if (region == nullptr) never_see_null = true;
4377 Node* p = basic_plus_adr(mirror, offset);
4378 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4379 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
4380 Node* null_ctl = top();
4381 kls = null_check_oop(kls, &null_ctl, never_see_null);
4382 if (region != nullptr) {
4383 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
4384 region->init_req(null_path, null_ctl);
4385 } else {
4386 assert(null_ctl == top(), "no loose ends");
4387 }
4388 return kls;
4389 }
4390
4391 //--------------------(inline_native_Class_query helpers)---------------------
4392 // Use this for JVM_ACC_INTERFACE.
4393 // Fall through if (mods & mask) == bits, take the guard otherwise.
4394 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
4395 ByteSize offset, const Type* type, BasicType bt) {
4396 // Branch around if the given klass has the given modifier bit set.
4397 // Like generate_guard, adds a new path onto the region.
4398 Node* modp = basic_plus_adr(top(), kls, in_bytes(offset));
4399 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
4400 Node* mask = intcon(modifier_mask);
4401 Node* bits = intcon(modifier_bits);
4402 Node* mbit = _gvn.transform(new AndINode(mods, mask));
4403 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
4404 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
4405 return generate_fair_guard(bol, region);
4406 }
4407
4408 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
4409 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
4410 InstanceKlass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
4411 }
4412
4413 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
4414 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
4415 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
4416 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
4417 }
4418
4419 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
4420 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
4421 }
4422
4423 //-------------------------inline_native_Class_query-------------------
4424 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
4425 const Type* return_type = TypeInt::BOOL;
4426 Node* prim_return_value = top(); // what happens if it's a primitive class?
4427 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4428 bool expect_prim = false; // most of these guys expect to work on refs
4429
4430 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
4431
4432 Node* mirror = argument(0);
4433 Node* obj = top();
4434
4435 switch (id) {
4436 case vmIntrinsics::_isInstance:
4437 // nothing is an instance of a primitive type
4438 prim_return_value = intcon(0);
4439 obj = argument(1);
4440 break;
4441 case vmIntrinsics::_isHidden:
4442 prim_return_value = intcon(0);
4443 break;
4444 case vmIntrinsics::_getSuperclass:
4445 prim_return_value = null();
4446 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
4447 break;
4448 default:
4449 fatal_unexpected_iid(id);
4450 break;
4451 }
4452
4453 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4454 if (mirror_con == nullptr) return false; // cannot happen?
4455
4456 #ifndef PRODUCT
4457 if (C->print_intrinsics() || C->print_inlining()) {
4458 ciType* k = mirror_con->java_mirror_type();
4459 if (k) {
4460 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4461 k->print_name();
4462 tty->cr();
4463 }
4464 }
4465 #endif
4466
4467 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4468 RegionNode* region = new RegionNode(PATH_LIMIT);
4469 record_for_igvn(region);
4470 PhiNode* phi = new PhiNode(region, return_type);
4471
4472 // The mirror will never be null of Reflection.getClassAccessFlags, however
4473 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4474 // if it is. See bug 4774291.
4475
4476 // For Reflection.getClassAccessFlags(), the null check occurs in
4477 // the wrong place; see inline_unsafe_access(), above, for a similar
4478 // situation.
4479 mirror = null_check(mirror);
4480 // If mirror or obj is dead, only null-path is taken.
4481 if (stopped()) return true;
4482
4483 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4484
4485 // Now load the mirror's klass metaobject, and null-check it.
4486 // Side-effects region with the control path if the klass is null.
4487 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4488 // If kls is null, we have a primitive mirror.
4489 phi->init_req(_prim_path, prim_return_value);
4490 if (stopped()) { set_result(region, phi); return true; }
4491 bool safe_for_replace = (region->in(_prim_path) == top());
4492
4493 Node* p; // handy temp
4494 Node* null_ctl;
4495
4496 // Now that we have the non-null klass, we can perform the real query.
4497 // For constant classes, the query will constant-fold in LoadNode::Value.
4498 Node* query_value = top();
4499 switch (id) {
4500 case vmIntrinsics::_isInstance:
4501 // nothing is an instance of a primitive type
4502 query_value = gen_instanceof(obj, kls, safe_for_replace);
4503 break;
4504
4505 case vmIntrinsics::_isHidden:
4506 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4507 if (generate_hidden_class_guard(kls, region) != nullptr)
4508 // A guard was added. If the guard is taken, it was an hidden class.
4509 phi->add_req(intcon(1));
4510 // If we fall through, it's a plain class.
4511 query_value = intcon(0);
4512 break;
4513
4514
4515 case vmIntrinsics::_getSuperclass:
4516 // The rules here are somewhat unfortunate, but we can still do better
4517 // with random logic than with a JNI call.
4518 // Interfaces store null or Object as _super, but must report null.
4519 // Arrays store an intermediate super as _super, but must report Object.
4520 // Other types can report the actual _super.
4521 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4522 if (generate_array_guard(kls, region) != nullptr) {
4523 // A guard was added. If the guard is taken, it was an array.
4524 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4525 }
4526 // Check for interface after array since this checks AccessFlags offset into InstanceKlass.
4527 // In other words, we are accessing subtype-specific information, so we need to determine the subtype first.
4528 if (generate_interface_guard(kls, region) != nullptr) {
4529 // A guard was added. If the guard is taken, it was an interface.
4530 phi->add_req(null());
4531 }
4532 // If we fall through, it's a plain class. Get its _super.
4533 if (!stopped()) {
4534 p = basic_plus_adr(top(), kls, in_bytes(Klass::super_offset()));
4535 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4536 null_ctl = top();
4537 kls = null_check_oop(kls, &null_ctl);
4538 if (null_ctl != top()) {
4539 // If the guard is taken, Object.superClass is null (both klass and mirror).
4540 region->add_req(null_ctl);
4541 phi ->add_req(null());
4542 }
4543 if (!stopped()) {
4544 query_value = load_mirror_from_klass(kls);
4545 }
4546 }
4547 break;
4548
4549 default:
4550 fatal_unexpected_iid(id);
4551 break;
4552 }
4553
4554 // Fall-through is the normal case of a query to a real class.
4555 phi->init_req(1, query_value);
4556 region->init_req(1, control());
4557
4558 C->set_has_split_ifs(true); // Has chance for split-if optimization
4559 set_result(region, phi);
4560 return true;
4561 }
4562
4563
4564 //-------------------------inline_Class_cast-------------------
4565 bool LibraryCallKit::inline_Class_cast() {
4566 Node* mirror = argument(0); // Class
4567 Node* obj = argument(1);
4568 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4569 if (mirror_con == nullptr) {
4570 return false; // dead path (mirror->is_top()).
4571 }
4572 if (obj == nullptr || obj->is_top()) {
4573 return false; // dead path
4574 }
4575 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4576
4577 // First, see if Class.cast() can be folded statically.
4578 // java_mirror_type() returns non-null for compile-time Class constants.
4579 ciType* tm = mirror_con->java_mirror_type();
4580 if (tm != nullptr && tm->is_klass() &&
4581 tp != nullptr) {
4582 if (!tp->is_loaded()) {
4583 // Don't use intrinsic when class is not loaded.
4584 return false;
4585 } else {
4586 const TypeKlassPtr* tklass = TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces);
4587 int static_res = C->static_subtype_check(tklass, tp->as_klass_type());
4588 if (static_res == Compile::SSC_always_true) {
4589 // isInstance() is true - fold the code.
4590 set_result(obj);
4591 return true;
4592 } else if (static_res == Compile::SSC_always_false) {
4593 // Don't use intrinsic, have to throw ClassCastException.
4594 // If the reference is null, the non-intrinsic bytecode will
4595 // be optimized appropriately.
4596 return false;
4597 }
4598 }
4599 }
4600
4601 // Bailout intrinsic and do normal inlining if exception path is frequent.
4602 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4603 return false;
4604 }
4605
4606 // Generate dynamic checks.
4607 // Class.cast() is java implementation of _checkcast bytecode.
4608 // Do checkcast (Parse::do_checkcast()) optimizations here.
4609
4610 mirror = null_check(mirror);
4611 // If mirror is dead, only null-path is taken.
4612 if (stopped()) {
4613 return true;
4614 }
4615
4616 // Not-subtype or the mirror's klass ptr is nullptr (in case it is a primitive).
4617 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
4618 RegionNode* region = new RegionNode(PATH_LIMIT);
4619 record_for_igvn(region);
4620
4621 // Now load the mirror's klass metaobject, and null-check it.
4622 // If kls is null, we have a primitive mirror and
4623 // nothing is an instance of a primitive type.
4624 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4625
4626 Node* res = top();
4627 Node* io = i_o();
4628 Node* mem = merged_memory();
4629 if (!stopped()) {
4630
4631 Node* bad_type_ctrl = top();
4632 // Do checkcast optimizations.
4633 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4634 region->init_req(_bad_type_path, bad_type_ctrl);
4635 }
4636 if (region->in(_prim_path) != top() ||
4637 region->in(_bad_type_path) != top() ||
4638 region->in(_npe_path) != top()) {
4639 // Let Interpreter throw ClassCastException.
4640 PreserveJVMState pjvms(this);
4641 set_control(_gvn.transform(region));
4642 // Set IO and memory because gen_checkcast may override them when buffering inline types
4643 set_i_o(io);
4644 set_all_memory(mem);
4645 uncommon_trap(Deoptimization::Reason_intrinsic,
4646 Deoptimization::Action_maybe_recompile);
4647 }
4648 if (!stopped()) {
4649 set_result(res);
4650 }
4651 return true;
4652 }
4653
4654
4655 //--------------------------inline_native_subtype_check------------------------
4656 // This intrinsic takes the JNI calls out of the heart of
4657 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4658 bool LibraryCallKit::inline_native_subtype_check() {
4659 // Pull both arguments off the stack.
4660 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4661 args[0] = argument(0);
4662 args[1] = argument(1);
4663 Node* klasses[2]; // corresponding Klasses: superk, subk
4664 klasses[0] = klasses[1] = top();
4665
4666 enum {
4667 // A full decision tree on {superc is prim, subc is prim}:
4668 _prim_0_path = 1, // {P,N} => false
4669 // {P,P} & superc!=subc => false
4670 _prim_same_path, // {P,P} & superc==subc => true
4671 _prim_1_path, // {N,P} => false
4672 _ref_subtype_path, // {N,N} & subtype check wins => true
4673 _both_ref_path, // {N,N} & subtype check loses => false
4674 PATH_LIMIT
4675 };
4676
4677 RegionNode* region = new RegionNode(PATH_LIMIT);
4678 RegionNode* prim_region = new RegionNode(2);
4679 Node* phi = new PhiNode(region, TypeInt::BOOL);
4680 record_for_igvn(region);
4681 record_for_igvn(prim_region);
4682
4683 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4684 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4685 int class_klass_offset = java_lang_Class::klass_offset();
4686
4687 // First null-check both mirrors and load each mirror's klass metaobject.
4688 int which_arg;
4689 for (which_arg = 0; which_arg <= 1; which_arg++) {
4690 Node* arg = args[which_arg];
4691 arg = null_check(arg);
4692 if (stopped()) break;
4693 args[which_arg] = arg;
4694
4695 Node* p = basic_plus_adr(arg, class_klass_offset);
4696 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4697 klasses[which_arg] = _gvn.transform(kls);
4698 }
4699
4700 // Having loaded both klasses, test each for null.
4701 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4702 for (which_arg = 0; which_arg <= 1; which_arg++) {
4703 Node* kls = klasses[which_arg];
4704 Node* null_ctl = top();
4705 kls = null_check_oop(kls, &null_ctl, never_see_null);
4706 if (which_arg == 0) {
4707 prim_region->init_req(1, null_ctl);
4708 } else {
4709 region->init_req(_prim_1_path, null_ctl);
4710 }
4711 if (stopped()) break;
4712 klasses[which_arg] = kls;
4713 }
4714
4715 if (!stopped()) {
4716 // now we have two reference types, in klasses[0..1]
4717 Node* subk = klasses[1]; // the argument to isAssignableFrom
4718 Node* superk = klasses[0]; // the receiver
4719 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4720 region->set_req(_ref_subtype_path, control());
4721 }
4722
4723 // If both operands are primitive (both klasses null), then
4724 // we must return true when they are identical primitives.
4725 // It is convenient to test this after the first null klass check.
4726 // This path is also used if superc is a value mirror.
4727 set_control(_gvn.transform(prim_region));
4728 if (!stopped()) {
4729 // Since superc is primitive, make a guard for the superc==subc case.
4730 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4731 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4732 generate_fair_guard(bol_eq, region);
4733 if (region->req() == PATH_LIMIT+1) {
4734 // A guard was added. If the added guard is taken, superc==subc.
4735 region->swap_edges(PATH_LIMIT, _prim_same_path);
4736 region->del_req(PATH_LIMIT);
4737 }
4738 region->set_req(_prim_0_path, control()); // Not equal after all.
4739 }
4740
4741 // these are the only paths that produce 'true':
4742 phi->set_req(_prim_same_path, intcon(1));
4743 phi->set_req(_ref_subtype_path, intcon(1));
4744
4745 // pull together the cases:
4746 assert(region->req() == PATH_LIMIT, "sane region");
4747 for (uint i = 1; i < region->req(); i++) {
4748 Node* ctl = region->in(i);
4749 if (ctl == nullptr || ctl == top()) {
4750 region->set_req(i, top());
4751 phi ->set_req(i, top());
4752 } else if (phi->in(i) == nullptr) {
4753 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4754 }
4755 }
4756
4757 set_control(_gvn.transform(region));
4758 set_result(_gvn.transform(phi));
4759 return true;
4760 }
4761
4762 //---------------------generate_array_guard_common------------------------
4763 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind, Node** obj) {
4764
4765 if (stopped()) {
4766 return nullptr;
4767 }
4768
4769 // Like generate_guard, adds a new path onto the region.
4770 jint layout_con = 0;
4771 Node* layout_val = get_layout_helper(kls, layout_con);
4772 if (layout_val == nullptr) {
4773 bool query = 0;
4774 switch(kind) {
4775 case RefArray: query = Klass::layout_helper_is_refArray(layout_con); break;
4776 case NonRefArray: query = !Klass::layout_helper_is_refArray(layout_con); break;
4777 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
4778 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
4779 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
4780 default:
4781 ShouldNotReachHere();
4782 }
4783 if (!query) {
4784 return nullptr; // never a branch
4785 } else { // always a branch
4786 Node* always_branch = control();
4787 if (region != nullptr)
4788 region->add_req(always_branch);
4789 set_control(top());
4790 return always_branch;
4791 }
4792 }
4793 unsigned int value = 0;
4794 BoolTest::mask btest = BoolTest::illegal;
4795 switch(kind) {
4796 case RefArray:
4797 case NonRefArray: {
4798 value = Klass::_lh_array_tag_ref_value;
4799 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4800 btest = (kind == RefArray) ? BoolTest::eq : BoolTest::ne;
4801 break;
4802 }
4803 case TypeArray: {
4804 value = Klass::_lh_array_tag_type_value;
4805 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4806 btest = BoolTest::eq;
4807 break;
4808 }
4809 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
4810 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
4811 default:
4812 ShouldNotReachHere();
4813 }
4814 // Now test the correct condition.
4815 jint nval = (jint)value;
4816 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4817 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4818 Node* ctrl = generate_fair_guard(bol, region);
4819 Node* is_array_ctrl = kind == NonArray ? control() : ctrl;
4820 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4821 // Keep track of the fact that 'obj' is an array to prevent
4822 // array specific accesses from floating above the guard.
4823 *obj = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4824 }
4825 return ctrl;
4826 }
4827
4828 // public static native Object[] ValueClass::newNullRestrictedAtomicArray(Class<?> componentType, int length, Object initVal);
4829 // public static native Object[] ValueClass::newNullRestrictedNonAtomicArray(Class<?> componentType, int length, Object initVal);
4830 // public static native Object[] ValueClass::newNullableAtomicArray(Class<?> componentType, int length);
4831 bool LibraryCallKit::inline_newArray(bool null_free, bool atomic) {
4832 assert(null_free || atomic, "nullable implies atomic");
4833 Node* componentType = argument(0);
4834 Node* length = argument(1);
4835 Node* init_val = null_free ? argument(2) : nullptr;
4836
4837 const TypeInstPtr* tp = _gvn.type(componentType)->isa_instptr();
4838 if (tp != nullptr) {
4839 ciInstanceKlass* ik = tp->instance_klass();
4840 if (ik == C->env()->Class_klass()) {
4841 ciType* t = tp->java_mirror_type();
4842 if (t != nullptr && t->is_inlinetype()) {
4843
4844 ciArrayKlass* array_klass = ciArrayKlass::make(t, null_free, atomic, true);
4845 assert(array_klass->is_elem_null_free() == null_free, "inconsistency");
4846
4847 // TOOD 8350865 ZGC needs card marks on initializing oop stores
4848 if (UseZGC && null_free && !array_klass->is_flat_array_klass()) {
4849 return false;
4850 }
4851
4852 if (array_klass->is_loaded() && array_klass->element_klass()->as_inline_klass()->is_initialized()) {
4853 const TypeAryKlassPtr* array_klass_type = TypeAryKlassPtr::make(array_klass, Type::trust_interfaces);
4854 if (null_free) {
4855 if (init_val->is_InlineType()) {
4856 if (array_klass_type->is_flat() && init_val->as_InlineType()->is_all_zero(&gvn(), /* flat */ true)) {
4857 // Zeroing is enough because the init value is the all-zero value
4858 init_val = nullptr;
4859 } else {
4860 init_val = init_val->as_InlineType()->buffer(this);
4861 }
4862 }
4863 // TODO 8350865 Should we add a check of the init_val type (maybe in debug only + halt)?
4864 // If we insert a checkcast here, we can be sure that init_val is an InlineTypeNode, so
4865 // when we folded a field load from an allocation (e.g. during escape analysis), we can
4866 // remove the check init_val->is_InlineType().
4867 }
4868 Node* obj = new_array(makecon(array_klass_type), length, 0, nullptr, false, init_val);
4869 const TypeAryPtr* arytype = gvn().type(obj)->is_aryptr();
4870 assert(arytype->is_null_free() == null_free, "inconsistency");
4871 assert(arytype->is_not_null_free() == !null_free, "inconsistency");
4872 set_result(obj);
4873 return true;
4874 }
4875 }
4876 }
4877 }
4878 return false;
4879 }
4880
4881 // public static native boolean ValueClass::isFlatArray(Object array);
4882 // public static native boolean ValueClass::isNullRestrictedArray(Object array);
4883 // public static native boolean ValueClass::isAtomicArray(Object array);
4884 bool LibraryCallKit::inline_getArrayProperties(ArrayPropertiesCheck check) {
4885 Node* array = argument(0);
4886
4887 Node* bol;
4888 switch(check) {
4889 case IsFlat:
4890 // TODO 8350865 Use the object version here instead of loading the klass
4891 // The problem is that PhaseMacroExpand::expand_flatarraycheck_node can only handle some IR shapes and will fail, for example, if the bol is directly wired to a ReturnNode
4892 bol = flat_array_test(load_object_klass(array));
4893 break;
4894 case IsNullRestricted:
4895 bol = null_free_array_test(array);
4896 break;
4897 case IsAtomic:
4898 // TODO 8350865 Implement this. It's a bit more complicated, see conditions in JVM_IsAtomicArray
4899 // Enable TestIntrinsics::test87/88 once this is implemented
4900 // bol = null_free_atomic_array_test
4901 return false;
4902 default:
4903 ShouldNotReachHere();
4904 }
4905
4906 Node* res = gvn().transform(new CMoveINode(bol, intcon(0), intcon(1), TypeInt::BOOL));
4907 set_result(res);
4908 return true;
4909 }
4910
4911 // Load the default refined array klass from an ObjArrayKlass. This relies on the first entry in the
4912 // '_next_refined_array_klass' linked list being the default (see ObjArrayKlass::klass_with_properties).
4913 Node* LibraryCallKit::load_default_refined_array_klass(Node* klass_node, bool type_array_guard) {
4914 RegionNode* region = new RegionNode(2);
4915 Node* phi = new PhiNode(region, TypeInstKlassPtr::OBJECT_OR_NULL);
4916
4917 if (type_array_guard) {
4918 generate_typeArray_guard(klass_node, region);
4919 if (region->req() == 3) {
4920 phi->add_req(klass_node);
4921 }
4922 }
4923 Node* adr_refined_klass = basic_plus_adr(top(), klass_node, in_bytes(ObjArrayKlass::next_refined_array_klass_offset()));
4924 Node* refined_klass = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), adr_refined_klass, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4925
4926 // Can be null if not initialized yet, just deopt
4927 Node* null_ctl = top();
4928 refined_klass = null_check_oop(refined_klass, &null_ctl, /* never_see_null= */ true);
4929
4930 region->init_req(1, control());
4931 phi->init_req(1, refined_klass);
4932
4933 set_control(_gvn.transform(region));
4934 return _gvn.transform(phi);
4935 }
4936
4937 // Load the non-refined array klass from an ObjArrayKlass.
4938 Node* LibraryCallKit::load_non_refined_array_klass(Node* klass_node) {
4939 const TypeAryKlassPtr* ary_klass_ptr = _gvn.type(klass_node)->isa_aryklassptr();
4940 if (ary_klass_ptr != nullptr && ary_klass_ptr->klass_is_exact()) {
4941 return _gvn.makecon(ary_klass_ptr->cast_to_refined_array_klass_ptr(false));
4942 }
4943
4944 RegionNode* region = new RegionNode(2);
4945 Node* phi = new PhiNode(region, TypeInstKlassPtr::OBJECT);
4946
4947 generate_typeArray_guard(klass_node, region);
4948 if (region->req() == 3) {
4949 phi->add_req(klass_node);
4950 }
4951 Node* super_adr = basic_plus_adr(top(), klass_node, in_bytes(Klass::super_offset()));
4952 Node* super_klass = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), super_adr, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT));
4953
4954 region->init_req(1, control());
4955 phi->init_req(1, super_klass);
4956
4957 set_control(_gvn.transform(region));
4958 return _gvn.transform(phi);
4959 }
4960
4961 //-----------------------inline_native_newArray--------------------------
4962 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
4963 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4964 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4965 Node* mirror;
4966 Node* count_val;
4967 if (uninitialized) {
4968 null_check_receiver();
4969 mirror = argument(1);
4970 count_val = argument(2);
4971 } else {
4972 mirror = argument(0);
4973 count_val = argument(1);
4974 }
4975
4976 mirror = null_check(mirror);
4977 // If mirror or obj is dead, only null-path is taken.
4978 if (stopped()) return true;
4979
4980 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4981 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4982 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4983 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4984 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4985
4986 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4987 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4988 result_reg, _slow_path);
4989 Node* normal_ctl = control();
4990 Node* no_array_ctl = result_reg->in(_slow_path);
4991
4992 // Generate code for the slow case. We make a call to newArray().
4993 set_control(no_array_ctl);
4994 if (!stopped()) {
4995 // Either the input type is void.class, or else the
4996 // array klass has not yet been cached. Either the
4997 // ensuing call will throw an exception, or else it
4998 // will cache the array klass for next time.
4999 PreserveJVMState pjvms(this);
5000 CallJavaNode* slow_call = nullptr;
5001 if (uninitialized) {
5002 // Generate optimized virtual call (holder class 'Unsafe' is final)
5003 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
5004 } else {
5005 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
5006 }
5007 Node* slow_result = set_results_for_java_call(slow_call);
5008 // this->control() comes from set_results_for_java_call
5009 result_reg->set_req(_slow_path, control());
5010 result_val->set_req(_slow_path, slow_result);
5011 result_io ->set_req(_slow_path, i_o());
5012 result_mem->set_req(_slow_path, reset_memory());
5013 }
5014
5015 set_control(normal_ctl);
5016 if (!stopped()) {
5017 // Normal case: The array type has been cached in the java.lang.Class.
5018 // The following call works fine even if the array type is polymorphic.
5019 // It could be a dynamic mix of int[], boolean[], Object[], etc.
5020
5021 klass_node = load_default_refined_array_klass(klass_node);
5022
5023 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
5024 result_reg->init_req(_normal_path, control());
5025 result_val->init_req(_normal_path, obj);
5026 result_io ->init_req(_normal_path, i_o());
5027 result_mem->init_req(_normal_path, reset_memory());
5028
5029 if (uninitialized) {
5030 // Mark the allocation so that zeroing is skipped
5031 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
5032 alloc->maybe_set_complete(&_gvn);
5033 }
5034 }
5035
5036 // Return the combined state.
5037 set_i_o( _gvn.transform(result_io) );
5038 set_all_memory( _gvn.transform(result_mem));
5039
5040 C->set_has_split_ifs(true); // Has chance for split-if optimization
5041 set_result(result_reg, result_val);
5042 return true;
5043 }
5044
5045 //----------------------inline_native_getLength--------------------------
5046 // public static native int java.lang.reflect.Array.getLength(Object array);
5047 bool LibraryCallKit::inline_native_getLength() {
5048 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
5049
5050 Node* array = null_check(argument(0));
5051 // If array is dead, only null-path is taken.
5052 if (stopped()) return true;
5053
5054 // Deoptimize if it is a non-array.
5055 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
5056
5057 if (non_array != nullptr) {
5058 PreserveJVMState pjvms(this);
5059 set_control(non_array);
5060 uncommon_trap(Deoptimization::Reason_intrinsic,
5061 Deoptimization::Action_maybe_recompile);
5062 }
5063
5064 // If control is dead, only non-array-path is taken.
5065 if (stopped()) return true;
5066
5067 // The works fine even if the array type is polymorphic.
5068 // It could be a dynamic mix of int[], boolean[], Object[], etc.
5069 Node* result = load_array_length(array);
5070
5071 C->set_has_split_ifs(true); // Has chance for split-if optimization
5072 set_result(result);
5073 return true;
5074 }
5075
5076 //------------------------inline_array_copyOf----------------------------
5077 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
5078 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
5079 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
5080 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
5081
5082 // Get the arguments.
5083 Node* original = argument(0);
5084 Node* start = is_copyOfRange? argument(1): intcon(0);
5085 Node* end = is_copyOfRange? argument(2): argument(1);
5086 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
5087
5088 Node* newcopy = nullptr;
5089
5090 // Set the original stack and the reexecute bit for the interpreter to reexecute
5091 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
5092 { PreserveReexecuteState preexecs(this);
5093 jvms()->set_should_reexecute(true);
5094
5095 array_type_mirror = null_check(array_type_mirror);
5096 original = null_check(original);
5097
5098 // Check if a null path was taken unconditionally.
5099 if (stopped()) return true;
5100
5101 Node* orig_length = load_array_length(original);
5102
5103 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
5104 klass_node = null_check(klass_node);
5105
5106 RegionNode* bailout = new RegionNode(1);
5107 record_for_igvn(bailout);
5108
5109 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
5110 // Bail out if that is so.
5111 // Inline type array may have object field that would require a
5112 // write barrier. Conservatively, go to slow path.
5113 // TODO 8251971: Optimize for the case when flat src/dst are later found
5114 // to not contain oops (i.e., move this check to the macro expansion phase).
5115 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5116 const TypeAryPtr* orig_t = _gvn.type(original)->isa_aryptr();
5117 const TypeKlassPtr* tklass = _gvn.type(klass_node)->is_klassptr();
5118 bool exclude_flat = UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing) &&
5119 // Can src array be flat and contain oops?
5120 (orig_t == nullptr || (!orig_t->is_not_flat() && (!orig_t->is_flat() || orig_t->elem()->inline_klass()->contains_oops()))) &&
5121 // Can dest array be flat and contain oops?
5122 tklass->can_be_inline_array() && (!tklass->is_flat() || tklass->is_aryklassptr()->elem()->is_instklassptr()->instance_klass()->as_inline_klass()->contains_oops());
5123 Node* not_objArray = exclude_flat ? generate_non_refArray_guard(klass_node, bailout) : generate_typeArray_guard(klass_node, bailout);
5124
5125 Node* refined_klass_node = load_default_refined_array_klass(klass_node, /* type_array_guard= */ false);
5126
5127 if (not_objArray != nullptr) {
5128 // Improve the klass node's type from the new optimistic assumption:
5129 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
5130 bool not_flat = !UseArrayFlattening;
5131 bool not_null_free = !Arguments::is_valhalla_enabled();
5132 const Type* akls = TypeAryKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0), Type::trust_interfaces, not_flat, not_null_free, false, false, not_flat, true);
5133 Node* cast = new CastPPNode(control(), refined_klass_node, akls);
5134 refined_klass_node = _gvn.transform(cast);
5135 }
5136
5137 // Bail out if either start or end is negative.
5138 generate_negative_guard(start, bailout, &start);
5139 generate_negative_guard(end, bailout, &end);
5140
5141 Node* length = end;
5142 if (_gvn.type(start) != TypeInt::ZERO) {
5143 length = _gvn.transform(new SubINode(end, start));
5144 }
5145
5146 // Bail out if length is negative (i.e., if start > end).
5147 // Without this the new_array would throw
5148 // NegativeArraySizeException but IllegalArgumentException is what
5149 // should be thrown
5150 generate_negative_guard(length, bailout, &length);
5151
5152 // Handle inline type arrays
5153 bool can_validate = !too_many_traps(Deoptimization::Reason_class_check);
5154 if (!stopped()) {
5155 // TODO 8251971
5156 if (!orig_t->is_null_free()) {
5157 // Not statically known to be null free, add a check
5158 generate_fair_guard(null_free_array_test(original), bailout);
5159 }
5160 orig_t = _gvn.type(original)->isa_aryptr();
5161 if (orig_t != nullptr && orig_t->is_flat()) {
5162 // Src is flat, check that dest is flat as well
5163 if (exclude_flat) {
5164 // Dest can't be flat, bail out
5165 bailout->add_req(control());
5166 set_control(top());
5167 } else {
5168 generate_fair_guard(flat_array_test(refined_klass_node, /* flat = */ false), bailout);
5169 }
5170 // TODO 8350865 This is not correct anymore. Write tests and fix logic similar to arraycopy.
5171 } else if (UseArrayFlattening && (orig_t == nullptr || !orig_t->is_not_flat()) &&
5172 // If dest is flat, src must be flat as well (guaranteed by src <: dest check if validated).
5173 ((!tklass->is_flat() && tklass->can_be_inline_array()) || !can_validate)) {
5174 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
5175 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
5176 generate_fair_guard(flat_array_test(load_object_klass(original)), bailout);
5177 if (orig_t != nullptr) {
5178 orig_t = orig_t->cast_to_not_flat();
5179 original = _gvn.transform(new CheckCastPPNode(control(), original, orig_t));
5180 }
5181 }
5182 if (!can_validate) {
5183 // No validation. The subtype check emitted at macro expansion time will not go to the slow
5184 // path but call checkcast_arraycopy which can not handle flat/null-free inline type arrays.
5185 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat/null-free.
5186 generate_fair_guard(flat_array_test(refined_klass_node), bailout);
5187 generate_fair_guard(null_free_array_test(original), bailout);
5188 }
5189 }
5190
5191 // Bail out if start is larger than the original length
5192 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
5193 generate_negative_guard(orig_tail, bailout, &orig_tail);
5194
5195 if (bailout->req() > 1) {
5196 PreserveJVMState pjvms(this);
5197 set_control(_gvn.transform(bailout));
5198 uncommon_trap(Deoptimization::Reason_intrinsic,
5199 Deoptimization::Action_maybe_recompile);
5200 }
5201
5202 if (!stopped()) {
5203 // How many elements will we copy from the original?
5204 // The answer is MinI(orig_tail, length).
5205 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
5206
5207 // Generate a direct call to the right arraycopy function(s).
5208 // We know the copy is disjoint but we might not know if the
5209 // oop stores need checking.
5210 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
5211 // This will fail a store-check if x contains any non-nulls.
5212
5213 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
5214 // loads/stores but it is legal only if we're sure the
5215 // Arrays.copyOf would succeed. So we need all input arguments
5216 // to the copyOf to be validated, including that the copy to the
5217 // new array won't trigger an ArrayStoreException. That subtype
5218 // check can be optimized if we know something on the type of
5219 // the input array from type speculation.
5220 if (_gvn.type(klass_node)->singleton()) {
5221 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
5222 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
5223
5224 int test = C->static_subtype_check(superk, subk);
5225 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
5226 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
5227 if (t_original->speculative_type() != nullptr) {
5228 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
5229 }
5230 }
5231 }
5232
5233 bool validated = false;
5234 // Reason_class_check rather than Reason_intrinsic because we
5235 // want to intrinsify even if this traps.
5236 if (can_validate) {
5237 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
5238
5239 if (not_subtype_ctrl != top()) {
5240 PreserveJVMState pjvms(this);
5241 set_control(not_subtype_ctrl);
5242 uncommon_trap(Deoptimization::Reason_class_check,
5243 Deoptimization::Action_make_not_entrant);
5244 assert(stopped(), "Should be stopped");
5245 }
5246 validated = true;
5247 }
5248
5249 if (!stopped()) {
5250 newcopy = new_array(refined_klass_node, length, 0); // no arguments to push
5251
5252 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
5253 load_object_klass(original), klass_node);
5254 if (!is_copyOfRange) {
5255 ac->set_copyof(validated);
5256 } else {
5257 ac->set_copyofrange(validated);
5258 }
5259 Node* n = _gvn.transform(ac);
5260 if (n == ac) {
5261 ac->connect_outputs(this);
5262 } else {
5263 assert(validated, "shouldn't transform if all arguments not validated");
5264 set_all_memory(n);
5265 }
5266 }
5267 }
5268 } // original reexecute is set back here
5269
5270 C->set_has_split_ifs(true); // Has chance for split-if optimization
5271 if (!stopped()) {
5272 set_result(newcopy);
5273 }
5274 return true;
5275 }
5276
5277
5278 //----------------------generate_virtual_guard---------------------------
5279 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
5280 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
5281 RegionNode* slow_region) {
5282 ciMethod* method = callee();
5283 int vtable_index = method->vtable_index();
5284 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5285 "bad index %d", vtable_index);
5286 // Get the Method* out of the appropriate vtable entry.
5287 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
5288 vtable_index*vtableEntry::size_in_bytes() +
5289 in_bytes(vtableEntry::method_offset());
5290 Node* entry_addr = basic_plus_adr(top(), obj_klass, entry_offset);
5291 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
5292
5293 // Compare the target method with the expected method (e.g., Object.hashCode).
5294 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
5295
5296 Node* native_call = makecon(native_call_addr);
5297 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
5298 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
5299
5300 return generate_slow_guard(test_native, slow_region);
5301 }
5302
5303 //-----------------------generate_method_call----------------------------
5304 // Use generate_method_call to make a slow-call to the real
5305 // method if the fast path fails. An alternative would be to
5306 // use a stub like OptoRuntime::slow_arraycopy_Java.
5307 // This only works for expanding the current library call,
5308 // not another intrinsic. (E.g., don't use this for making an
5309 // arraycopy call inside of the copyOf intrinsic.)
5310 CallJavaNode*
5311 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
5312 // When compiling the intrinsic method itself, do not use this technique.
5313 guarantee(callee() != C->method(), "cannot make slow-call to self");
5314
5315 ciMethod* method = callee();
5316 // ensure the JVMS we have will be correct for this call
5317 guarantee(method_id == method->intrinsic_id(), "must match");
5318
5319 const TypeFunc* tf = TypeFunc::make(method);
5320 if (res_not_null) {
5321 assert(tf->return_type() == T_OBJECT, "");
5322 const TypeTuple* range = tf->range_cc();
5323 const Type** fields = TypeTuple::fields(range->cnt());
5324 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
5325 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
5326 tf = TypeFunc::make(tf->domain_cc(), new_range);
5327 }
5328 CallJavaNode* slow_call;
5329 if (is_static) {
5330 assert(!is_virtual, "");
5331 slow_call = new CallStaticJavaNode(C, tf,
5332 SharedRuntime::get_resolve_static_call_stub(), method);
5333 } else if (is_virtual) {
5334 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5335 int vtable_index = Method::invalid_vtable_index;
5336 if (UseInlineCaches) {
5337 // Suppress the vtable call
5338 } else {
5339 // hashCode and clone are not a miranda methods,
5340 // so the vtable index is fixed.
5341 // No need to use the linkResolver to get it.
5342 vtable_index = method->vtable_index();
5343 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5344 "bad index %d", vtable_index);
5345 }
5346 slow_call = new CallDynamicJavaNode(tf,
5347 SharedRuntime::get_resolve_virtual_call_stub(),
5348 method, vtable_index);
5349 } else { // neither virtual nor static: opt_virtual
5350 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5351 slow_call = new CallStaticJavaNode(C, tf,
5352 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
5353 slow_call->set_optimized_virtual(true);
5354 }
5355 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
5356 // To be able to issue a direct call (optimized virtual or virtual)
5357 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
5358 // about the method being invoked should be attached to the call site to
5359 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
5360 slow_call->set_override_symbolic_info(true);
5361 }
5362 set_arguments_for_java_call(slow_call);
5363 set_edges_for_java_call(slow_call);
5364 return slow_call;
5365 }
5366
5367
5368 /**
5369 * Build special case code for calls to hashCode on an object. This call may
5370 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
5371 * slightly different code.
5372 */
5373 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
5374 assert(is_static == callee()->is_static(), "correct intrinsic selection");
5375 assert(!(is_virtual && is_static), "either virtual, special, or static");
5376
5377 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
5378
5379 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5380 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
5381 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
5382 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5383 Node* obj = argument(0);
5384
5385 // Don't intrinsify hashcode on inline types for now.
5386 // The "is locked" runtime check also subsumes the inline type check (as inline types cannot be locked) and goes to the slow path.
5387 if (gvn().type(obj)->is_inlinetypeptr()) {
5388 return false;
5389 }
5390
5391 if (!is_static) {
5392 // Check for hashing null object
5393 obj = null_check_receiver();
5394 if (stopped()) return true; // unconditionally null
5395 result_reg->init_req(_null_path, top());
5396 result_val->init_req(_null_path, top());
5397 } else {
5398 // Do a null check, and return zero if null.
5399 // System.identityHashCode(null) == 0
5400 Node* null_ctl = top();
5401 obj = null_check_oop(obj, &null_ctl);
5402 result_reg->init_req(_null_path, null_ctl);
5403 result_val->init_req(_null_path, _gvn.intcon(0));
5404 }
5405
5406 // Unconditionally null? Then return right away.
5407 if (stopped()) {
5408 set_control( result_reg->in(_null_path));
5409 if (!stopped())
5410 set_result(result_val->in(_null_path));
5411 return true;
5412 }
5413
5414 // We only go to the fast case code if we pass a number of guards. The
5415 // paths which do not pass are accumulated in the slow_region.
5416 RegionNode* slow_region = new RegionNode(1);
5417 record_for_igvn(slow_region);
5418
5419 // If this is a virtual call, we generate a funny guard. We pull out
5420 // the vtable entry corresponding to hashCode() from the target object.
5421 // If the target method which we are calling happens to be the native
5422 // Object hashCode() method, we pass the guard. We do not need this
5423 // guard for non-virtual calls -- the caller is known to be the native
5424 // Object hashCode().
5425 if (is_virtual) {
5426 // After null check, get the object's klass.
5427 Node* obj_klass = load_object_klass(obj);
5428 generate_virtual_guard(obj_klass, slow_region);
5429 }
5430
5431 // Get the header out of the object, use LoadMarkNode when available
5432 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
5433 // The control of the load must be null. Otherwise, the load can move before
5434 // the null check after castPP removal.
5435 Node* no_ctrl = nullptr;
5436 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
5437
5438 if (!UseObjectMonitorTable) {
5439 // Test the header to see if it is safe to read w.r.t. locking.
5440 // We cannot use the inline type mask as this may check bits that are overriden
5441 // by an object monitor's pointer when inflating locking.
5442 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
5443 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
5444 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
5445 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
5446 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
5447
5448 generate_slow_guard(test_monitor, slow_region);
5449 }
5450
5451 // Get the hash value and check to see that it has been properly assigned.
5452 // We depend on hash_mask being at most 32 bits and avoid the use of
5453 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
5454 // vm: see markWord.hpp.
5455 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
5456 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
5457 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
5458 // This hack lets the hash bits live anywhere in the mark object now, as long
5459 // as the shift drops the relevant bits into the low 32 bits. Note that
5460 // Java spec says that HashCode is an int so there's no point in capturing
5461 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
5462 hshifted_header = ConvX2I(hshifted_header);
5463 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
5464
5465 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
5466 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
5467 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
5468
5469 generate_slow_guard(test_assigned, slow_region);
5470
5471 Node* init_mem = reset_memory();
5472 // fill in the rest of the null path:
5473 result_io ->init_req(_null_path, i_o());
5474 result_mem->init_req(_null_path, init_mem);
5475
5476 result_val->init_req(_fast_path, hash_val);
5477 result_reg->init_req(_fast_path, control());
5478 result_io ->init_req(_fast_path, i_o());
5479 result_mem->init_req(_fast_path, init_mem);
5480
5481 // Generate code for the slow case. We make a call to hashCode().
5482 set_control(_gvn.transform(slow_region));
5483 if (!stopped()) {
5484 // No need for PreserveJVMState, because we're using up the present state.
5485 set_all_memory(init_mem);
5486 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
5487 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
5488 Node* slow_result = set_results_for_java_call(slow_call);
5489 // this->control() comes from set_results_for_java_call
5490 result_reg->init_req(_slow_path, control());
5491 result_val->init_req(_slow_path, slow_result);
5492 result_io ->set_req(_slow_path, i_o());
5493 result_mem ->set_req(_slow_path, reset_memory());
5494 }
5495
5496 // Return the combined state.
5497 set_i_o( _gvn.transform(result_io) );
5498 set_all_memory( _gvn.transform(result_mem));
5499
5500 set_result(result_reg, result_val);
5501 return true;
5502 }
5503
5504 //---------------------------inline_native_getClass----------------------------
5505 // public final native Class<?> java.lang.Object.getClass();
5506 //
5507 // Build special case code for calls to getClass on an object.
5508 bool LibraryCallKit::inline_native_getClass() {
5509 Node* obj = argument(0);
5510 if (obj->is_InlineType()) {
5511 const Type* t = _gvn.type(obj);
5512 if (t->maybe_null()) {
5513 null_check(obj);
5514 }
5515 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror())));
5516 return true;
5517 }
5518 obj = null_check_receiver();
5519 if (stopped()) return true;
5520 set_result(load_mirror_from_klass(load_object_klass(obj)));
5521 return true;
5522 }
5523
5524 //-----------------inline_native_Reflection_getCallerClass---------------------
5525 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
5526 //
5527 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
5528 //
5529 // NOTE: This code must perform the same logic as JVM_GetCallerClass
5530 // in that it must skip particular security frames and checks for
5531 // caller sensitive methods.
5532 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
5533 #ifndef PRODUCT
5534 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5535 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
5536 }
5537 #endif
5538
5539 if (!jvms()->has_method()) {
5540 #ifndef PRODUCT
5541 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5542 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
5543 }
5544 #endif
5545 return false;
5546 }
5547
5548 // Walk back up the JVM state to find the caller at the required
5549 // depth.
5550 JVMState* caller_jvms = jvms();
5551
5552 // Cf. JVM_GetCallerClass
5553 // NOTE: Start the loop at depth 1 because the current JVM state does
5554 // not include the Reflection.getCallerClass() frame.
5555 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
5556 ciMethod* m = caller_jvms->method();
5557 switch (n) {
5558 case 0:
5559 fatal("current JVM state does not include the Reflection.getCallerClass frame");
5560 break;
5561 case 1:
5562 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
5563 if (!m->caller_sensitive()) {
5564 #ifndef PRODUCT
5565 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5566 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
5567 }
5568 #endif
5569 return false; // bail-out; let JVM_GetCallerClass do the work
5570 }
5571 break;
5572 default:
5573 if (!m->is_ignored_by_security_stack_walk()) {
5574 // We have reached the desired frame; return the holder class.
5575 // Acquire method holder as java.lang.Class and push as constant.
5576 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
5577 ciInstance* caller_mirror = caller_klass->java_mirror();
5578 set_result(makecon(TypeInstPtr::make(caller_mirror)));
5579
5580 #ifndef PRODUCT
5581 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5582 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());
5583 tty->print_cr(" JVM state at this point:");
5584 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5585 ciMethod* m = jvms()->of_depth(i)->method();
5586 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5587 }
5588 }
5589 #endif
5590 return true;
5591 }
5592 break;
5593 }
5594 }
5595
5596 #ifndef PRODUCT
5597 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5598 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
5599 tty->print_cr(" JVM state at this point:");
5600 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5601 ciMethod* m = jvms()->of_depth(i)->method();
5602 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5603 }
5604 }
5605 #endif
5606
5607 return false; // bail-out; let JVM_GetCallerClass do the work
5608 }
5609
5610 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
5611 Node* arg = argument(0);
5612 Node* result = nullptr;
5613
5614 switch (id) {
5615 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
5616 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
5617 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5618 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5619 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5620 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5621
5622 case vmIntrinsics::_doubleToLongBits: {
5623 // two paths (plus control) merge in a wood
5624 RegionNode *r = new RegionNode(3);
5625 Node *phi = new PhiNode(r, TypeLong::LONG);
5626
5627 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5628 // Build the boolean node
5629 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5630
5631 // Branch either way.
5632 // NaN case is less traveled, which makes all the difference.
5633 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5634 Node *opt_isnan = _gvn.transform(ifisnan);
5635 assert( opt_isnan->is_If(), "Expect an IfNode");
5636 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5637 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5638
5639 set_control(iftrue);
5640
5641 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5642 Node *slow_result = longcon(nan_bits); // return NaN
5643 phi->init_req(1, _gvn.transform( slow_result ));
5644 r->init_req(1, iftrue);
5645
5646 // Else fall through
5647 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5648 set_control(iffalse);
5649
5650 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5651 r->init_req(2, iffalse);
5652
5653 // Post merge
5654 set_control(_gvn.transform(r));
5655 record_for_igvn(r);
5656
5657 C->set_has_split_ifs(true); // Has chance for split-if optimization
5658 result = phi;
5659 assert(result->bottom_type()->isa_long(), "must be");
5660 break;
5661 }
5662
5663 case vmIntrinsics::_floatToIntBits: {
5664 // two paths (plus control) merge in a wood
5665 RegionNode *r = new RegionNode(3);
5666 Node *phi = new PhiNode(r, TypeInt::INT);
5667
5668 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5669 // Build the boolean node
5670 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5671
5672 // Branch either way.
5673 // NaN case is less traveled, which makes all the difference.
5674 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5675 Node *opt_isnan = _gvn.transform(ifisnan);
5676 assert( opt_isnan->is_If(), "Expect an IfNode");
5677 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5678 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5679
5680 set_control(iftrue);
5681
5682 static const jint nan_bits = 0x7fc00000;
5683 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5684 phi->init_req(1, _gvn.transform( slow_result ));
5685 r->init_req(1, iftrue);
5686
5687 // Else fall through
5688 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5689 set_control(iffalse);
5690
5691 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5692 r->init_req(2, iffalse);
5693
5694 // Post merge
5695 set_control(_gvn.transform(r));
5696 record_for_igvn(r);
5697
5698 C->set_has_split_ifs(true); // Has chance for split-if optimization
5699 result = phi;
5700 assert(result->bottom_type()->isa_int(), "must be");
5701 break;
5702 }
5703
5704 default:
5705 fatal_unexpected_iid(id);
5706 break;
5707 }
5708 set_result(_gvn.transform(result));
5709 return true;
5710 }
5711
5712 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5713 Node* arg = argument(0);
5714 Node* result = nullptr;
5715
5716 switch (id) {
5717 case vmIntrinsics::_floatIsInfinite:
5718 result = new IsInfiniteFNode(arg);
5719 break;
5720 case vmIntrinsics::_floatIsFinite:
5721 result = new IsFiniteFNode(arg);
5722 break;
5723 case vmIntrinsics::_doubleIsInfinite:
5724 result = new IsInfiniteDNode(arg);
5725 break;
5726 case vmIntrinsics::_doubleIsFinite:
5727 result = new IsFiniteDNode(arg);
5728 break;
5729 default:
5730 fatal_unexpected_iid(id);
5731 break;
5732 }
5733 set_result(_gvn.transform(result));
5734 return true;
5735 }
5736
5737 //----------------------inline_unsafe_copyMemory-------------------------
5738 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5739
5740 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5741 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5742 const Type* base_t = gvn.type(base);
5743
5744 bool in_native = (base_t == TypePtr::NULL_PTR);
5745 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5746 bool is_mixed = !in_heap && !in_native;
5747
5748 if (is_mixed) {
5749 return true; // mixed accesses can touch both on-heap and off-heap memory
5750 }
5751 if (in_heap) {
5752 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5753 if (!is_prim_array) {
5754 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5755 // there's not enough type information available to determine proper memory slice for it.
5756 return true;
5757 }
5758 }
5759 return false;
5760 }
5761
5762 bool LibraryCallKit::inline_unsafe_copyMemory() {
5763 if (callee()->is_static()) return false; // caller must have the capability!
5764 null_check_receiver(); // null-check receiver
5765 if (stopped()) return true;
5766
5767 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5768
5769 Node* src_base = argument(1); // type: oop
5770 Node* src_off = ConvL2X(argument(2)); // type: long
5771 Node* dst_base = argument(4); // type: oop
5772 Node* dst_off = ConvL2X(argument(5)); // type: long
5773 Node* size = ConvL2X(argument(7)); // type: long
5774
5775 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5776 "fieldOffset must be byte-scaled");
5777
5778 Node* src_addr = make_unsafe_address(src_base, src_off);
5779 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5780
5781 Node* thread = _gvn.transform(new ThreadLocalNode());
5782 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5783 BasicType doing_unsafe_access_bt = T_BYTE;
5784 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5785
5786 // update volatile field
5787 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5788
5789 int flags = RC_LEAF | RC_NO_FP;
5790
5791 const TypePtr* dst_type = TypePtr::BOTTOM;
5792
5793 // Adjust memory effects of the runtime call based on input values.
5794 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5795 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5796 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5797
5798 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5799 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5800 flags |= RC_NARROW_MEM; // narrow in memory
5801 }
5802 }
5803
5804 // Call it. Note that the length argument is not scaled.
5805 make_runtime_call(flags,
5806 OptoRuntime::fast_arraycopy_Type(),
5807 StubRoutines::unsafe_arraycopy(),
5808 "unsafe_arraycopy",
5809 dst_type,
5810 src_addr, dst_addr, size XTOP);
5811
5812 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5813
5814 return true;
5815 }
5816
5817 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5818 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5819 bool LibraryCallKit::inline_unsafe_setMemory() {
5820 if (callee()->is_static()) return false; // caller must have the capability!
5821 null_check_receiver(); // null-check receiver
5822 if (stopped()) return true;
5823
5824 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5825
5826 Node* dst_base = argument(1); // type: oop
5827 Node* dst_off = ConvL2X(argument(2)); // type: long
5828 Node* size = ConvL2X(argument(4)); // type: long
5829 Node* byte = argument(6); // type: byte
5830
5831 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5832 "fieldOffset must be byte-scaled");
5833
5834 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5835
5836 Node* thread = _gvn.transform(new ThreadLocalNode());
5837 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5838 BasicType doing_unsafe_access_bt = T_BYTE;
5839 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5840
5841 // update volatile field
5842 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5843
5844 int flags = RC_LEAF | RC_NO_FP;
5845
5846 const TypePtr* dst_type = TypePtr::BOTTOM;
5847
5848 // Adjust memory effects of the runtime call based on input values.
5849 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5850 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5851
5852 flags |= RC_NARROW_MEM; // narrow in memory
5853 }
5854
5855 // Call it. Note that the length argument is not scaled.
5856 make_runtime_call(flags,
5857 OptoRuntime::unsafe_setmemory_Type(),
5858 StubRoutines::unsafe_setmemory(),
5859 "unsafe_setmemory",
5860 dst_type,
5861 dst_addr, size XTOP, byte);
5862
5863 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5864
5865 return true;
5866 }
5867
5868 #undef XTOP
5869
5870 //------------------------clone_coping-----------------------------------
5871 // Helper function for inline_native_clone.
5872 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5873 assert(obj_size != nullptr, "");
5874 Node* raw_obj = alloc_obj->in(1);
5875 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5876
5877 AllocateNode* alloc = nullptr;
5878 if (ReduceBulkZeroing &&
5879 // If we are implementing an array clone without knowing its source type
5880 // (can happen when compiling the array-guarded branch of a reflective
5881 // Object.clone() invocation), initialize the array within the allocation.
5882 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5883 // to a runtime clone call that assumes fully initialized source arrays.
5884 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5885 // We will be completely responsible for initializing this object -
5886 // mark Initialize node as complete.
5887 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5888 // The object was just allocated - there should be no any stores!
5889 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5890 // Mark as complete_with_arraycopy so that on AllocateNode
5891 // expansion, we know this AllocateNode is initialized by an array
5892 // copy and a StoreStore barrier exists after the array copy.
5893 alloc->initialization()->set_complete_with_arraycopy();
5894 }
5895
5896 Node* size = _gvn.transform(obj_size);
5897 access_clone(obj, alloc_obj, size, is_array);
5898
5899 // Do not let reads from the cloned object float above the arraycopy.
5900 if (alloc != nullptr) {
5901 // Do not let stores that initialize this object be reordered with
5902 // a subsequent store that would make this object accessible by
5903 // other threads.
5904 // Record what AllocateNode this StoreStore protects so that
5905 // escape analysis can go from the MemBarStoreStoreNode to the
5906 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5907 // based on the escape status of the AllocateNode.
5908 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5909 } else {
5910 insert_mem_bar(Op_MemBarCPUOrder);
5911 }
5912 }
5913
5914 //------------------------inline_native_clone----------------------------
5915 // protected native Object java.lang.Object.clone();
5916 //
5917 // Here are the simple edge cases:
5918 // null receiver => normal trap
5919 // virtual and clone was overridden => slow path to out-of-line clone
5920 // not cloneable or finalizer => slow path to out-of-line Object.clone
5921 //
5922 // The general case has two steps, allocation and copying.
5923 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5924 //
5925 // Copying also has two cases, oop arrays and everything else.
5926 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5927 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5928 //
5929 // These steps fold up nicely if and when the cloned object's klass
5930 // can be sharply typed as an object array, a type array, or an instance.
5931 //
5932 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5933 PhiNode* result_val;
5934
5935 // Set the reexecute bit for the interpreter to reexecute
5936 // the bytecode that invokes Object.clone if deoptimization happens.
5937 { PreserveReexecuteState preexecs(this);
5938 jvms()->set_should_reexecute(true);
5939
5940 Node* obj = argument(0);
5941 obj = null_check_receiver();
5942 if (stopped()) return true;
5943
5944 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5945 if (obj_type->is_inlinetypeptr()) {
5946 // If the object to clone is an inline type, we can simply return it (i.e. a nop) since inline types have
5947 // no identity.
5948 set_result(obj);
5949 return true;
5950 }
5951
5952 // If we are going to clone an instance, we need its exact type to
5953 // know the number and types of fields to convert the clone to
5954 // loads/stores. Maybe a speculative type can help us.
5955 if (!obj_type->klass_is_exact() &&
5956 obj_type->speculative_type() != nullptr &&
5957 obj_type->speculative_type()->is_instance_klass() &&
5958 !obj_type->speculative_type()->is_inlinetype()) {
5959 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5960 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5961 !spec_ik->has_injected_fields()) {
5962 if (!obj_type->isa_instptr() ||
5963 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5964 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5965 }
5966 }
5967 }
5968
5969 // Conservatively insert a memory barrier on all memory slices.
5970 // Do not let writes into the original float below the clone.
5971 insert_mem_bar(Op_MemBarCPUOrder);
5972
5973 // paths into result_reg:
5974 enum {
5975 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5976 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5977 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5978 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5979 PATH_LIMIT
5980 };
5981 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5982 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5983 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5984 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5985 record_for_igvn(result_reg);
5986
5987 Node* obj_klass = load_object_klass(obj);
5988 // We only go to the fast case code if we pass a number of guards.
5989 // The paths which do not pass are accumulated in the slow_region.
5990 RegionNode* slow_region = new RegionNode(1);
5991 record_for_igvn(slow_region);
5992
5993 Node* array_obj = obj;
5994 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5995 if (array_ctl != nullptr) {
5996 // It's an array.
5997 PreserveJVMState pjvms(this);
5998 set_control(array_ctl);
5999
6000 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
6001 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr();
6002 if (UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) &&
6003 obj_type->can_be_inline_array() &&
6004 (ary_ptr == nullptr || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) {
6005 // Flat inline type array may have object field that would require a
6006 // write barrier. Conservatively, go to slow path.
6007 generate_fair_guard(flat_array_test(obj_klass), slow_region);
6008 }
6009
6010 if (!stopped()) {
6011 Node* obj_length = load_array_length(array_obj);
6012 Node* array_size = nullptr; // Size of the array without object alignment padding.
6013 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
6014
6015 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
6016 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
6017 // If it is an oop array, it requires very special treatment,
6018 // because gc barriers are required when accessing the array.
6019 Node* is_obja = generate_refArray_guard(obj_klass, (RegionNode*)nullptr);
6020 if (is_obja != nullptr) {
6021 PreserveJVMState pjvms2(this);
6022 set_control(is_obja);
6023 // Generate a direct call to the right arraycopy function(s).
6024 // Clones are always tightly coupled.
6025 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
6026 ac->set_clone_oop_array();
6027 Node* n = _gvn.transform(ac);
6028 assert(n == ac, "cannot disappear");
6029 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
6030
6031 result_reg->init_req(_objArray_path, control());
6032 result_val->init_req(_objArray_path, alloc_obj);
6033 result_i_o ->set_req(_objArray_path, i_o());
6034 result_mem ->set_req(_objArray_path, reset_memory());
6035 }
6036 }
6037 // Otherwise, there are no barriers to worry about.
6038 // (We can dispense with card marks if we know the allocation
6039 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
6040 // causes the non-eden paths to take compensating steps to
6041 // simulate a fresh allocation, so that no further
6042 // card marks are required in compiled code to initialize
6043 // the object.)
6044
6045 if (!stopped()) {
6046 copy_to_clone(obj, alloc_obj, array_size, true);
6047
6048 // Present the results of the copy.
6049 result_reg->init_req(_array_path, control());
6050 result_val->init_req(_array_path, alloc_obj);
6051 result_i_o ->set_req(_array_path, i_o());
6052 result_mem ->set_req(_array_path, reset_memory());
6053 }
6054 }
6055 }
6056
6057 if (!stopped()) {
6058 // It's an instance (we did array above). Make the slow-path tests.
6059 // If this is a virtual call, we generate a funny guard. We grab
6060 // the vtable entry corresponding to clone() from the target object.
6061 // If the target method which we are calling happens to be the
6062 // Object clone() method, we pass the guard. We do not need this
6063 // guard for non-virtual calls; the caller is known to be the native
6064 // Object clone().
6065 if (is_virtual) {
6066 generate_virtual_guard(obj_klass, slow_region);
6067 }
6068
6069 // The object must be easily cloneable and must not have a finalizer.
6070 // Both of these conditions may be checked in a single test.
6071 // We could optimize the test further, but we don't care.
6072 generate_misc_flags_guard(obj_klass,
6073 // Test both conditions:
6074 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
6075 // Must be cloneable but not finalizer:
6076 KlassFlags::_misc_is_cloneable_fast,
6077 slow_region);
6078 }
6079
6080 if (!stopped()) {
6081 // It's an instance, and it passed the slow-path tests.
6082 PreserveJVMState pjvms(this);
6083 Node* obj_size = nullptr; // Total object size, including object alignment padding.
6084 // Need to deoptimize on exception from allocation since Object.clone intrinsic
6085 // is reexecuted if deoptimization occurs and there could be problems when merging
6086 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
6087 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
6088
6089 copy_to_clone(obj, alloc_obj, obj_size, false);
6090
6091 // Present the results of the slow call.
6092 result_reg->init_req(_instance_path, control());
6093 result_val->init_req(_instance_path, alloc_obj);
6094 result_i_o ->set_req(_instance_path, i_o());
6095 result_mem ->set_req(_instance_path, reset_memory());
6096 }
6097
6098 // Generate code for the slow case. We make a call to clone().
6099 set_control(_gvn.transform(slow_region));
6100 if (!stopped()) {
6101 PreserveJVMState pjvms(this);
6102 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
6103 // We need to deoptimize on exception (see comment above)
6104 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
6105 // this->control() comes from set_results_for_java_call
6106 result_reg->init_req(_slow_path, control());
6107 result_val->init_req(_slow_path, slow_result);
6108 result_i_o ->set_req(_slow_path, i_o());
6109 result_mem ->set_req(_slow_path, reset_memory());
6110 }
6111
6112 // Return the combined state.
6113 set_control( _gvn.transform(result_reg));
6114 set_i_o( _gvn.transform(result_i_o));
6115 set_all_memory( _gvn.transform(result_mem));
6116 } // original reexecute is set back here
6117
6118 set_result(_gvn.transform(result_val));
6119 return true;
6120 }
6121
6122 // If we have a tightly coupled allocation, the arraycopy may take care
6123 // of the array initialization. If one of the guards we insert between
6124 // the allocation and the arraycopy causes a deoptimization, an
6125 // uninitialized array will escape the compiled method. To prevent that
6126 // we set the JVM state for uncommon traps between the allocation and
6127 // the arraycopy to the state before the allocation so, in case of
6128 // deoptimization, we'll reexecute the allocation and the
6129 // initialization.
6130 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
6131 if (alloc != nullptr) {
6132 ciMethod* trap_method = alloc->jvms()->method();
6133 int trap_bci = alloc->jvms()->bci();
6134
6135 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6136 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
6137 // Make sure there's no store between the allocation and the
6138 // arraycopy otherwise visible side effects could be rexecuted
6139 // in case of deoptimization and cause incorrect execution.
6140 bool no_interfering_store = true;
6141 Node* mem = alloc->in(TypeFunc::Memory);
6142 if (mem->is_MergeMem()) {
6143 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
6144 Node* n = mms.memory();
6145 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
6146 assert(n->is_Store(), "what else?");
6147 no_interfering_store = false;
6148 break;
6149 }
6150 }
6151 } else {
6152 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
6153 Node* n = mms.memory();
6154 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
6155 assert(n->is_Store(), "what else?");
6156 no_interfering_store = false;
6157 break;
6158 }
6159 }
6160 }
6161
6162 if (no_interfering_store) {
6163 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6164
6165 JVMState* saved_jvms = jvms();
6166 saved_reexecute_sp = _reexecute_sp;
6167
6168 set_jvms(sfpt->jvms());
6169 _reexecute_sp = jvms()->sp();
6170
6171 return saved_jvms;
6172 }
6173 }
6174 }
6175 return nullptr;
6176 }
6177
6178 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
6179 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
6180 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
6181 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
6182 uint size = alloc->req();
6183 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
6184 old_jvms->set_map(sfpt);
6185 for (uint i = 0; i < size; i++) {
6186 sfpt->init_req(i, alloc->in(i));
6187 }
6188 int adjustment = 1;
6189 const TypeAryKlassPtr* ary_klass_ptr = alloc->in(AllocateNode::KlassNode)->bottom_type()->is_aryklassptr();
6190 if (ary_klass_ptr->is_null_free()) {
6191 // A null-free, tightly coupled array allocation can only come from LibraryCallKit::inline_newArray which
6192 // also requires the componentType and initVal on stack for re-execution.
6193 // Re-create and push the componentType.
6194 ciArrayKlass* klass = ary_klass_ptr->exact_klass()->as_array_klass();
6195 ciInstance* instance = klass->component_mirror_instance();
6196 const TypeInstPtr* t_instance = TypeInstPtr::make(instance);
6197 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), makecon(t_instance));
6198 adjustment++;
6199 }
6200 // re-push array length for deoptimization
6201 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment - 1, alloc->in(AllocateNode::ALength));
6202 if (ary_klass_ptr->is_null_free()) {
6203 // Re-create and push the initVal.
6204 Node* init_val = alloc->in(AllocateNode::InitValue);
6205 if (init_val == nullptr) {
6206 init_val = InlineTypeNode::make_all_zero(_gvn, ary_klass_ptr->elem()->is_instklassptr()->instance_klass()->as_inline_klass());
6207 } else if (UseCompressedOops) {
6208 init_val = _gvn.transform(new DecodeNNode(init_val, init_val->bottom_type()->make_ptr()));
6209 }
6210 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment, init_val);
6211 adjustment++;
6212 }
6213 old_jvms->set_sp(old_jvms->sp() + adjustment);
6214 old_jvms->set_monoff(old_jvms->monoff() + adjustment);
6215 old_jvms->set_scloff(old_jvms->scloff() + adjustment);
6216 old_jvms->set_endoff(old_jvms->endoff() + adjustment);
6217 old_jvms->set_should_reexecute(true);
6218
6219 sfpt->set_i_o(map()->i_o());
6220 sfpt->set_memory(map()->memory());
6221 sfpt->set_control(map()->control());
6222 return sfpt;
6223 }
6224
6225 // In case of a deoptimization, we restart execution at the
6226 // allocation, allocating a new array. We would leave an uninitialized
6227 // array in the heap that GCs wouldn't expect. Move the allocation
6228 // after the traps so we don't allocate the array if we
6229 // deoptimize. This is possible because tightly_coupled_allocation()
6230 // guarantees there's no observer of the allocated array at this point
6231 // and the control flow is simple enough.
6232 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
6233 int saved_reexecute_sp, uint new_idx) {
6234 if (saved_jvms_before_guards != nullptr && !stopped()) {
6235 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
6236
6237 assert(alloc != nullptr, "only with a tightly coupled allocation");
6238 // restore JVM state to the state at the arraycopy
6239 saved_jvms_before_guards->map()->set_control(map()->control());
6240 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
6241 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
6242 // If we've improved the types of some nodes (null check) while
6243 // emitting the guards, propagate them to the current state
6244 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
6245 set_jvms(saved_jvms_before_guards);
6246 _reexecute_sp = saved_reexecute_sp;
6247
6248 // Remove the allocation from above the guards
6249 CallProjections* callprojs = alloc->extract_projections(true);
6250 InitializeNode* init = alloc->initialization();
6251 Node* alloc_mem = alloc->in(TypeFunc::Memory);
6252 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
6253 init->replace_mem_projs_by(alloc_mem, C);
6254
6255 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
6256 // the allocation (i.e. is only valid if the allocation succeeds):
6257 // 1) replace CastIINode with AllocateArrayNode's length here
6258 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
6259 //
6260 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
6261 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
6262 Node* init_control = init->proj_out(TypeFunc::Control);
6263 Node* alloc_length = alloc->Ideal_length();
6264 #ifdef ASSERT
6265 Node* prev_cast = nullptr;
6266 #endif
6267 for (uint i = 0; i < init_control->outcnt(); i++) {
6268 Node* init_out = init_control->raw_out(i);
6269 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
6270 #ifdef ASSERT
6271 if (prev_cast == nullptr) {
6272 prev_cast = init_out;
6273 } else {
6274 if (prev_cast->cmp(*init_out) == false) {
6275 prev_cast->dump();
6276 init_out->dump();
6277 assert(false, "not equal CastIINode");
6278 }
6279 }
6280 #endif
6281 C->gvn_replace_by(init_out, alloc_length);
6282 }
6283 }
6284 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
6285
6286 // move the allocation here (after the guards)
6287 _gvn.hash_delete(alloc);
6288 alloc->set_req(TypeFunc::Control, control());
6289 alloc->set_req(TypeFunc::I_O, i_o());
6290 Node *mem = reset_memory();
6291 set_all_memory(mem);
6292 alloc->set_req(TypeFunc::Memory, mem);
6293 set_control(init->proj_out_or_null(TypeFunc::Control));
6294 set_i_o(callprojs->fallthrough_ioproj);
6295
6296 // Update memory as done in GraphKit::set_output_for_allocation()
6297 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
6298 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
6299 if (ary_type->isa_aryptr() && length_type != nullptr) {
6300 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
6301 }
6302 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
6303 int elemidx = C->get_alias_index(telemref);
6304 // Need to properly move every memory projection for the Initialize
6305 #ifdef ASSERT
6306 int mark_idx = C->get_alias_index(ary_type->add_offset(oopDesc::mark_offset_in_bytes()));
6307 int klass_idx = C->get_alias_index(ary_type->add_offset(oopDesc::klass_offset_in_bytes()));
6308 #endif
6309 auto move_proj = [&](ProjNode* proj) {
6310 int alias_idx = C->get_alias_index(proj->adr_type());
6311 assert(alias_idx == Compile::AliasIdxRaw ||
6312 alias_idx == elemidx ||
6313 alias_idx == mark_idx ||
6314 alias_idx == klass_idx, "should be raw memory or array element type");
6315 set_memory(proj, alias_idx);
6316 };
6317 init->for_each_proj(move_proj, TypeFunc::Memory);
6318
6319 Node* allocx = _gvn.transform(alloc);
6320 assert(allocx == alloc, "where has the allocation gone?");
6321 assert(dest->is_CheckCastPP(), "not an allocation result?");
6322
6323 _gvn.hash_delete(dest);
6324 dest->set_req(0, control());
6325 Node* destx = _gvn.transform(dest);
6326 assert(destx == dest, "where has the allocation result gone?");
6327
6328 array_ideal_length(alloc, ary_type, true);
6329 }
6330 }
6331
6332 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
6333 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
6334 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
6335 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
6336 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
6337 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
6338 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
6339 JVMState* saved_jvms_before_guards) {
6340 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
6341 // There is at least one unrelated uncommon trap which needs to be replaced.
6342 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6343
6344 JVMState* saved_jvms = jvms();
6345 const int saved_reexecute_sp = _reexecute_sp;
6346 set_jvms(sfpt->jvms());
6347 _reexecute_sp = jvms()->sp();
6348
6349 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
6350
6351 // Restore state
6352 set_jvms(saved_jvms);
6353 _reexecute_sp = saved_reexecute_sp;
6354 }
6355 }
6356
6357 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
6358 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
6359 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
6360 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
6361 while (if_proj->is_IfProj()) {
6362 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
6363 if (uncommon_trap != nullptr) {
6364 create_new_uncommon_trap(uncommon_trap);
6365 }
6366 assert(if_proj->in(0)->is_If(), "must be If");
6367 if_proj = if_proj->in(0)->in(0);
6368 }
6369 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
6370 "must have reached control projection of init node");
6371 }
6372
6373 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
6374 const int trap_request = uncommon_trap_call->uncommon_trap_request();
6375 assert(trap_request != 0, "no valid UCT trap request");
6376 PreserveJVMState pjvms(this);
6377 set_control(uncommon_trap_call->in(0));
6378 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
6379 Deoptimization::trap_request_action(trap_request));
6380 assert(stopped(), "Should be stopped");
6381 _gvn.hash_delete(uncommon_trap_call);
6382 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
6383 }
6384
6385 // Common checks for array sorting intrinsics arguments.
6386 // Returns `true` if checks passed.
6387 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
6388 // check address of the class
6389 if (elementType == nullptr || elementType->is_top()) {
6390 return false; // dead path
6391 }
6392 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
6393 if (elem_klass == nullptr) {
6394 return false; // dead path
6395 }
6396 // java_mirror_type() returns non-null for compile-time Class constants only
6397 ciType* elem_type = elem_klass->java_mirror_type();
6398 if (elem_type == nullptr) {
6399 return false;
6400 }
6401 bt = elem_type->basic_type();
6402 // Disable the intrinsic if the CPU does not support SIMD sort
6403 if (!Matcher::supports_simd_sort(bt)) {
6404 return false;
6405 }
6406 // check address of the array
6407 if (obj == nullptr || obj->is_top()) {
6408 return false; // dead path
6409 }
6410 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
6411 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
6412 return false; // failed input validation
6413 }
6414 return true;
6415 }
6416
6417 //------------------------------inline_array_partition-----------------------
6418 bool LibraryCallKit::inline_array_partition() {
6419 address stubAddr = StubRoutines::select_array_partition_function();
6420 if (stubAddr == nullptr) {
6421 return false; // Intrinsic's stub is not implemented on this platform
6422 }
6423 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
6424
6425 // no receiver because it is a static method
6426 Node* elementType = argument(0);
6427 Node* obj = argument(1);
6428 Node* offset = argument(2); // long
6429 Node* fromIndex = argument(4);
6430 Node* toIndex = argument(5);
6431 Node* indexPivot1 = argument(6);
6432 Node* indexPivot2 = argument(7);
6433 // PartitionOperation: argument(8) is ignored
6434
6435 Node* pivotIndices = nullptr;
6436 BasicType bt = T_ILLEGAL;
6437
6438 if (!check_array_sort_arguments(elementType, obj, bt)) {
6439 return false;
6440 }
6441 null_check(obj);
6442 // If obj is dead, only null-path is taken.
6443 if (stopped()) {
6444 return true;
6445 }
6446 // Set the original stack and the reexecute bit for the interpreter to reexecute
6447 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
6448 { PreserveReexecuteState preexecs(this);
6449 jvms()->set_should_reexecute(true);
6450
6451 Node* obj_adr = make_unsafe_address(obj, offset);
6452
6453 // create the pivotIndices array of type int and size = 2
6454 Node* size = intcon(2);
6455 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
6456 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
6457 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
6458 guarantee(alloc != nullptr, "created above");
6459 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
6460
6461 // pass the basic type enum to the stub
6462 Node* elemType = intcon(bt);
6463
6464 // Call the stub
6465 const char *stubName = "array_partition_stub";
6466 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
6467 stubAddr, stubName, TypePtr::BOTTOM,
6468 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
6469 indexPivot1, indexPivot2);
6470
6471 } // original reexecute is set back here
6472
6473 if (!stopped()) {
6474 set_result(pivotIndices);
6475 }
6476
6477 return true;
6478 }
6479
6480
6481 //------------------------------inline_array_sort-----------------------
6482 bool LibraryCallKit::inline_array_sort() {
6483 address stubAddr = StubRoutines::select_arraysort_function();
6484 if (stubAddr == nullptr) {
6485 return false; // Intrinsic's stub is not implemented on this platform
6486 }
6487 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
6488
6489 // no receiver because it is a static method
6490 Node* elementType = argument(0);
6491 Node* obj = argument(1);
6492 Node* offset = argument(2); // long
6493 Node* fromIndex = argument(4);
6494 Node* toIndex = argument(5);
6495 // SortOperation: argument(6) is ignored
6496
6497 BasicType bt = T_ILLEGAL;
6498
6499 if (!check_array_sort_arguments(elementType, obj, bt)) {
6500 return false;
6501 }
6502 null_check(obj);
6503 // If obj is dead, only null-path is taken.
6504 if (stopped()) {
6505 return true;
6506 }
6507 Node* obj_adr = make_unsafe_address(obj, offset);
6508
6509 // pass the basic type enum to the stub
6510 Node* elemType = intcon(bt);
6511
6512 // Call the stub.
6513 const char *stubName = "arraysort_stub";
6514 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
6515 stubAddr, stubName, TypePtr::BOTTOM,
6516 obj_adr, elemType, fromIndex, toIndex);
6517
6518 return true;
6519 }
6520
6521
6522 //------------------------------inline_arraycopy-----------------------
6523 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
6524 // Object dest, int destPos,
6525 // int length);
6526 bool LibraryCallKit::inline_arraycopy() {
6527 // Get the arguments.
6528 Node* src = argument(0); // type: oop
6529 Node* src_offset = argument(1); // type: int
6530 Node* dest = argument(2); // type: oop
6531 Node* dest_offset = argument(3); // type: int
6532 Node* length = argument(4); // type: int
6533
6534 uint new_idx = C->unique();
6535
6536 // Check for allocation before we add nodes that would confuse
6537 // tightly_coupled_allocation()
6538 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
6539
6540 int saved_reexecute_sp = -1;
6541 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
6542 // See arraycopy_restore_alloc_state() comment
6543 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
6544 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
6545 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
6546 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
6547
6548 // The following tests must be performed
6549 // (1) src and dest are arrays.
6550 // (2) src and dest arrays must have elements of the same BasicType
6551 // (3) src and dest must not be null.
6552 // (4) src_offset must not be negative.
6553 // (5) dest_offset must not be negative.
6554 // (6) length must not be negative.
6555 // (7) src_offset + length must not exceed length of src.
6556 // (8) dest_offset + length must not exceed length of dest.
6557 // (9) each element of an oop array must be assignable
6558
6559 // (3) src and dest must not be null.
6560 // always do this here because we need the JVM state for uncommon traps
6561 Node* null_ctl = top();
6562 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
6563 assert(null_ctl->is_top(), "no null control here");
6564 dest = null_check(dest, T_ARRAY);
6565
6566 if (!can_emit_guards) {
6567 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
6568 // guards but the arraycopy node could still take advantage of a
6569 // tightly allocated allocation. tightly_coupled_allocation() is
6570 // called again to make sure it takes the null check above into
6571 // account: the null check is mandatory and if it caused an
6572 // uncommon trap to be emitted then the allocation can't be
6573 // considered tightly coupled in this context.
6574 alloc = tightly_coupled_allocation(dest);
6575 }
6576
6577 bool validated = false;
6578
6579 const Type* src_type = _gvn.type(src);
6580 const Type* dest_type = _gvn.type(dest);
6581 const TypeAryPtr* top_src = src_type->isa_aryptr();
6582 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6583
6584 // Do we have the type of src?
6585 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6586 // Do we have the type of dest?
6587 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6588 // Is the type for src from speculation?
6589 bool src_spec = false;
6590 // Is the type for dest from speculation?
6591 bool dest_spec = false;
6592
6593 if ((!has_src || !has_dest) && can_emit_guards) {
6594 // We don't have sufficient type information, let's see if
6595 // speculative types can help. We need to have types for both src
6596 // and dest so that it pays off.
6597
6598 // Do we already have or could we have type information for src
6599 bool could_have_src = has_src;
6600 // Do we already have or could we have type information for dest
6601 bool could_have_dest = has_dest;
6602
6603 ciKlass* src_k = nullptr;
6604 if (!has_src) {
6605 src_k = src_type->speculative_type_not_null();
6606 if (src_k != nullptr && src_k->is_array_klass()) {
6607 could_have_src = true;
6608 }
6609 }
6610
6611 ciKlass* dest_k = nullptr;
6612 if (!has_dest) {
6613 dest_k = dest_type->speculative_type_not_null();
6614 if (dest_k != nullptr && dest_k->is_array_klass()) {
6615 could_have_dest = true;
6616 }
6617 }
6618
6619 if (could_have_src && could_have_dest) {
6620 // This is going to pay off so emit the required guards
6621 if (!has_src) {
6622 src = maybe_cast_profiled_obj(src, src_k, true);
6623 src_type = _gvn.type(src);
6624 top_src = src_type->isa_aryptr();
6625 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6626 src_spec = true;
6627 }
6628 if (!has_dest) {
6629 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6630 dest_type = _gvn.type(dest);
6631 top_dest = dest_type->isa_aryptr();
6632 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6633 dest_spec = true;
6634 }
6635 }
6636 }
6637
6638 if (has_src && has_dest && can_emit_guards) {
6639 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
6640 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
6641 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
6642 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
6643
6644 if (src_elem == dest_elem && top_src->is_flat() == top_dest->is_flat() && src_elem == T_OBJECT) {
6645 // If both arrays are object arrays then having the exact types
6646 // for both will remove the need for a subtype check at runtime
6647 // before the call and may make it possible to pick a faster copy
6648 // routine (without a subtype check on every element)
6649 // Do we have the exact type of src?
6650 bool could_have_src = src_spec;
6651 // Do we have the exact type of dest?
6652 bool could_have_dest = dest_spec;
6653 ciKlass* src_k = nullptr;
6654 ciKlass* dest_k = nullptr;
6655 if (!src_spec) {
6656 src_k = src_type->speculative_type_not_null();
6657 if (src_k != nullptr && src_k->is_array_klass()) {
6658 could_have_src = true;
6659 }
6660 }
6661 if (!dest_spec) {
6662 dest_k = dest_type->speculative_type_not_null();
6663 if (dest_k != nullptr && dest_k->is_array_klass()) {
6664 could_have_dest = true;
6665 }
6666 }
6667 if (could_have_src && could_have_dest) {
6668 // If we can have both exact types, emit the missing guards
6669 if (could_have_src && !src_spec) {
6670 src = maybe_cast_profiled_obj(src, src_k, true);
6671 src_type = _gvn.type(src);
6672 top_src = src_type->isa_aryptr();
6673 }
6674 if (could_have_dest && !dest_spec) {
6675 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6676 dest_type = _gvn.type(dest);
6677 top_dest = dest_type->isa_aryptr();
6678 }
6679 }
6680 }
6681 }
6682
6683 ciMethod* trap_method = method();
6684 int trap_bci = bci();
6685 if (saved_jvms_before_guards != nullptr) {
6686 trap_method = alloc->jvms()->method();
6687 trap_bci = alloc->jvms()->bci();
6688 }
6689
6690 bool negative_length_guard_generated = false;
6691
6692 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6693 can_emit_guards && !src->is_top() && !dest->is_top()) {
6694 // validate arguments: enables transformation the ArrayCopyNode
6695 validated = true;
6696
6697 RegionNode* slow_region = new RegionNode(1);
6698 record_for_igvn(slow_region);
6699
6700 // (1) src and dest are arrays.
6701 generate_non_array_guard(load_object_klass(src), slow_region, &src);
6702 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
6703
6704 // (2) src and dest arrays must have elements of the same BasicType
6705 // done at macro expansion or at Ideal transformation time
6706
6707 // (4) src_offset must not be negative.
6708 generate_negative_guard(src_offset, slow_region);
6709
6710 // (5) dest_offset must not be negative.
6711 generate_negative_guard(dest_offset, slow_region);
6712
6713 // (7) src_offset + length must not exceed length of src.
6714 generate_limit_guard(src_offset, length,
6715 load_array_length(src),
6716 slow_region);
6717
6718 // (8) dest_offset + length must not exceed length of dest.
6719 generate_limit_guard(dest_offset, length,
6720 load_array_length(dest),
6721 slow_region);
6722
6723 // (6) length must not be negative.
6724 // This is also checked in generate_arraycopy() during macro expansion, but
6725 // we also have to check it here for the case where the ArrayCopyNode will
6726 // be eliminated by Escape Analysis.
6727 if (EliminateAllocations) {
6728 generate_negative_guard(length, slow_region);
6729 negative_length_guard_generated = true;
6730 }
6731
6732 // (9) each element of an oop array must be assignable
6733 Node* dest_klass = load_object_klass(dest);
6734 Node* refined_dest_klass = dest_klass;
6735 if (src != dest) {
6736 dest_klass = load_non_refined_array_klass(refined_dest_klass);
6737 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6738 slow_region->add_req(not_subtype_ctrl);
6739 }
6740
6741 // TODO 8350865 Improve this. What about atomicity? Make sure this is always folded for type arrays.
6742 // If destination is null-restricted, source must be null-restricted as well: src_null_restricted || !dst_null_restricted
6743 Node* src_klass = load_object_klass(src);
6744 Node* adr_prop_src = basic_plus_adr(top(), src_klass, in_bytes(ArrayKlass::properties_offset()));
6745 Node* prop_src = _gvn.transform(LoadNode::make(_gvn, control(), immutable_memory(), adr_prop_src, TypeRawPtr::BOTTOM, TypeInt::INT, T_INT, MemNode::unordered));
6746 Node* adr_prop_dest = basic_plus_adr(top(), refined_dest_klass, in_bytes(ArrayKlass::properties_offset()));
6747 Node* prop_dest = _gvn.transform(LoadNode::make(_gvn, control(), immutable_memory(), adr_prop_dest, TypeRawPtr::BOTTOM, TypeInt::INT, T_INT, MemNode::unordered));
6748
6749 const ArrayProperties props_null_restricted = ArrayProperties::Default().with_null_restricted();
6750 jint props_value = (jint)props_null_restricted.value();
6751
6752 prop_dest = _gvn.transform(new XorINode(prop_dest, intcon(props_value)));
6753 prop_src = _gvn.transform(new OrINode(prop_dest, prop_src));
6754 prop_src = _gvn.transform(new AndINode(prop_src, intcon(props_value)));
6755
6756 Node* chk = _gvn.transform(new CmpINode(prop_src, intcon(props_value)));
6757 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::ne));
6758 generate_fair_guard(tst, slow_region);
6759
6760 // TODO 8350865 This is too strong
6761 generate_fair_guard(flat_array_test(src), slow_region);
6762 generate_fair_guard(flat_array_test(dest), slow_region);
6763
6764 {
6765 PreserveJVMState pjvms(this);
6766 set_control(_gvn.transform(slow_region));
6767 uncommon_trap(Deoptimization::Reason_intrinsic,
6768 Deoptimization::Action_make_not_entrant);
6769 assert(stopped(), "Should be stopped");
6770 }
6771
6772 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->isa_klassptr();
6773 if (dest_klass_t == nullptr) {
6774 // refined_dest_klass may not be an array, which leads to dest_klass being top. This means we
6775 // are in a dead path.
6776 uncommon_trap(Deoptimization::Reason_intrinsic,
6777 Deoptimization::Action_make_not_entrant);
6778 return true;
6779 }
6780
6781 const Type* toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6782 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6783 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6784 }
6785
6786 if (stopped()) {
6787 return true;
6788 }
6789
6790 Node* dest_klass = load_object_klass(dest);
6791 dest_klass = load_non_refined_array_klass(dest_klass);
6792
6793 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6794 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6795 // so the compiler has a chance to eliminate them: during macro expansion,
6796 // we have to set their control (CastPP nodes are eliminated).
6797 load_object_klass(src), dest_klass,
6798 load_array_length(src), load_array_length(dest));
6799
6800 ac->set_arraycopy(validated);
6801
6802 Node* n = _gvn.transform(ac);
6803 if (n == ac) {
6804 ac->connect_outputs(this);
6805 } else {
6806 assert(validated, "shouldn't transform if all arguments not validated");
6807 set_all_memory(n);
6808 }
6809 clear_upper_avx();
6810
6811
6812 return true;
6813 }
6814
6815
6816 // Helper function which determines if an arraycopy immediately follows
6817 // an allocation, with no intervening tests or other escapes for the object.
6818 AllocateArrayNode*
6819 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6820 if (stopped()) return nullptr; // no fast path
6821 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6822
6823 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6824 if (alloc == nullptr) return nullptr;
6825
6826 Node* rawmem = memory(Compile::AliasIdxRaw);
6827 // Is the allocation's memory state untouched?
6828 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6829 // Bail out if there have been raw-memory effects since the allocation.
6830 // (Example: There might have been a call or safepoint.)
6831 return nullptr;
6832 }
6833 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6834 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6835 return nullptr;
6836 }
6837
6838 // There must be no unexpected observers of this allocation.
6839 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6840 Node* obs = ptr->fast_out(i);
6841 if (obs != this->map()) {
6842 return nullptr;
6843 }
6844 }
6845
6846 // This arraycopy must unconditionally follow the allocation of the ptr.
6847 Node* alloc_ctl = ptr->in(0);
6848 Node* ctl = control();
6849 while (ctl != alloc_ctl) {
6850 // There may be guards which feed into the slow_region.
6851 // Any other control flow means that we might not get a chance
6852 // to finish initializing the allocated object.
6853 // Various low-level checks bottom out in uncommon traps. These
6854 // are considered safe since we've already checked above that
6855 // there is no unexpected observer of this allocation.
6856 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6857 assert(ctl->in(0)->is_If(), "must be If");
6858 ctl = ctl->in(0)->in(0);
6859 } else {
6860 return nullptr;
6861 }
6862 }
6863
6864 // If we get this far, we have an allocation which immediately
6865 // precedes the arraycopy, and we can take over zeroing the new object.
6866 // The arraycopy will finish the initialization, and provide
6867 // a new control state to which we will anchor the destination pointer.
6868
6869 return alloc;
6870 }
6871
6872 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6873 if (node->is_IfProj()) {
6874 IfProjNode* other_proj = node->as_IfProj()->other_if_proj();
6875 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6876 Node* obs = other_proj->fast_out(j);
6877 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6878 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6879 return obs->as_CallStaticJava();
6880 }
6881 }
6882 }
6883 return nullptr;
6884 }
6885
6886 //-------------inline_encodeISOArray-----------------------------------
6887 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6888 // int java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6889 // int java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6890 // encode char[] to byte[] in ISO_8859_1 or ASCII
6891 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6892 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6893 // no receiver since it is static method
6894 Node *src = argument(0);
6895 Node *src_offset = argument(1);
6896 Node *dst = argument(2);
6897 Node *dst_offset = argument(3);
6898 Node *length = argument(4);
6899
6900 // Cast source & target arrays to not-null
6901 src = must_be_not_null(src, true);
6902 dst = must_be_not_null(dst, true);
6903 if (stopped()) {
6904 return true;
6905 }
6906
6907 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6908 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6909 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6910 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6911 // failed array check
6912 return false;
6913 }
6914
6915 // Figure out the size and type of the elements we will be copying.
6916 BasicType src_elem = src_type->elem()->array_element_basic_type();
6917 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6918 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6919 return false;
6920 }
6921
6922 // Check source & target bounds
6923 generate_string_range_check(src, src_offset, length, src_elem == T_BYTE, true);
6924 generate_string_range_check(dst, dst_offset, length, false, true);
6925 if (stopped()) {
6926 return true;
6927 }
6928
6929 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6930 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6931 // 'src_start' points to src array + scaled offset
6932 // 'dst_start' points to dst array + scaled offset
6933
6934 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6935 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6936 enc = _gvn.transform(enc);
6937 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6938 set_memory(res_mem, mtype);
6939 set_result(enc);
6940 clear_upper_avx();
6941
6942 return true;
6943 }
6944
6945 //-------------inline_multiplyToLen-----------------------------------
6946 bool LibraryCallKit::inline_multiplyToLen() {
6947 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6948
6949 address stubAddr = StubRoutines::multiplyToLen();
6950 if (stubAddr == nullptr) {
6951 return false; // Intrinsic's stub is not implemented on this platform
6952 }
6953 const char* stubName = "multiplyToLen";
6954
6955 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6956
6957 // no receiver because it is a static method
6958 Node* x = argument(0);
6959 Node* xlen = argument(1);
6960 Node* y = argument(2);
6961 Node* ylen = argument(3);
6962 Node* z = argument(4);
6963
6964 x = must_be_not_null(x, true);
6965 y = must_be_not_null(y, true);
6966
6967 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6968 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6969 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6970 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6971 // failed array check
6972 return false;
6973 }
6974
6975 BasicType x_elem = x_type->elem()->array_element_basic_type();
6976 BasicType y_elem = y_type->elem()->array_element_basic_type();
6977 if (x_elem != T_INT || y_elem != T_INT) {
6978 return false;
6979 }
6980
6981 Node* x_start = array_element_address(x, intcon(0), x_elem);
6982 Node* y_start = array_element_address(y, intcon(0), y_elem);
6983 // 'x_start' points to x array + scaled xlen
6984 // 'y_start' points to y array + scaled ylen
6985
6986 Node* z_start = array_element_address(z, intcon(0), T_INT);
6987
6988 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6989 OptoRuntime::multiplyToLen_Type(),
6990 stubAddr, stubName, TypePtr::BOTTOM,
6991 x_start, xlen, y_start, ylen, z_start);
6992
6993 C->set_has_split_ifs(true); // Has chance for split-if optimization
6994 set_result(z);
6995 return true;
6996 }
6997
6998 //-------------inline_squareToLen------------------------------------
6999 bool LibraryCallKit::inline_squareToLen() {
7000 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
7001
7002 address stubAddr = StubRoutines::squareToLen();
7003 if (stubAddr == nullptr) {
7004 return false; // Intrinsic's stub is not implemented on this platform
7005 }
7006 const char* stubName = "squareToLen";
7007
7008 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
7009
7010 Node* x = argument(0);
7011 Node* len = argument(1);
7012 Node* z = argument(2);
7013 Node* zlen = argument(3);
7014
7015 x = must_be_not_null(x, true);
7016 z = must_be_not_null(z, true);
7017
7018 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
7019 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
7020 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
7021 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
7022 // failed array check
7023 return false;
7024 }
7025
7026 BasicType x_elem = x_type->elem()->array_element_basic_type();
7027 BasicType z_elem = z_type->elem()->array_element_basic_type();
7028 if (x_elem != T_INT || z_elem != T_INT) {
7029 return false;
7030 }
7031
7032
7033 Node* x_start = array_element_address(x, intcon(0), x_elem);
7034 Node* z_start = array_element_address(z, intcon(0), z_elem);
7035
7036 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
7037 OptoRuntime::squareToLen_Type(),
7038 stubAddr, stubName, TypePtr::BOTTOM,
7039 x_start, len, z_start, zlen);
7040
7041 set_result(z);
7042 return true;
7043 }
7044
7045 //-------------inline_mulAdd------------------------------------------
7046 bool LibraryCallKit::inline_mulAdd() {
7047 assert(UseMulAddIntrinsic, "not implemented on this platform");
7048
7049 address stubAddr = StubRoutines::mulAdd();
7050 if (stubAddr == nullptr) {
7051 return false; // Intrinsic's stub is not implemented on this platform
7052 }
7053 const char* stubName = "mulAdd";
7054
7055 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
7056
7057 Node* out = argument(0);
7058 Node* in = argument(1);
7059 Node* offset = argument(2);
7060 Node* len = argument(3);
7061 Node* k = argument(4);
7062
7063 in = must_be_not_null(in, true);
7064 out = must_be_not_null(out, true);
7065
7066 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
7067 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
7068 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
7069 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
7070 // failed array check
7071 return false;
7072 }
7073
7074 BasicType out_elem = out_type->elem()->array_element_basic_type();
7075 BasicType in_elem = in_type->elem()->array_element_basic_type();
7076 if (out_elem != T_INT || in_elem != T_INT) {
7077 return false;
7078 }
7079
7080 Node* outlen = load_array_length(out);
7081 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
7082 Node* out_start = array_element_address(out, intcon(0), out_elem);
7083 Node* in_start = array_element_address(in, intcon(0), in_elem);
7084
7085 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
7086 OptoRuntime::mulAdd_Type(),
7087 stubAddr, stubName, TypePtr::BOTTOM,
7088 out_start,in_start, new_offset, len, k);
7089 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7090 set_result(result);
7091 return true;
7092 }
7093
7094 //-------------inline_montgomeryMultiply-----------------------------------
7095 bool LibraryCallKit::inline_montgomeryMultiply() {
7096 address stubAddr = StubRoutines::montgomeryMultiply();
7097 if (stubAddr == nullptr) {
7098 return false; // Intrinsic's stub is not implemented on this platform
7099 }
7100
7101 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
7102 const char* stubName = "montgomery_multiply";
7103
7104 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
7105
7106 Node* a = argument(0);
7107 Node* b = argument(1);
7108 Node* n = argument(2);
7109 Node* len = argument(3);
7110 Node* inv = argument(4);
7111 Node* m = argument(6);
7112
7113 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
7114 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
7115 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
7116 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
7117 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
7118 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
7119 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
7120 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
7121 // failed array check
7122 return false;
7123 }
7124
7125 BasicType a_elem = a_type->elem()->array_element_basic_type();
7126 BasicType b_elem = b_type->elem()->array_element_basic_type();
7127 BasicType n_elem = n_type->elem()->array_element_basic_type();
7128 BasicType m_elem = m_type->elem()->array_element_basic_type();
7129 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
7130 return false;
7131 }
7132
7133 // Make the call
7134 {
7135 Node* a_start = array_element_address(a, intcon(0), a_elem);
7136 Node* b_start = array_element_address(b, intcon(0), b_elem);
7137 Node* n_start = array_element_address(n, intcon(0), n_elem);
7138 Node* m_start = array_element_address(m, intcon(0), m_elem);
7139
7140 Node* call = make_runtime_call(RC_LEAF,
7141 OptoRuntime::montgomeryMultiply_Type(),
7142 stubAddr, stubName, TypePtr::BOTTOM,
7143 a_start, b_start, n_start, len, inv, top(),
7144 m_start);
7145 set_result(m);
7146 }
7147
7148 return true;
7149 }
7150
7151 bool LibraryCallKit::inline_montgomerySquare() {
7152 address stubAddr = StubRoutines::montgomerySquare();
7153 if (stubAddr == nullptr) {
7154 return false; // Intrinsic's stub is not implemented on this platform
7155 }
7156
7157 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
7158 const char* stubName = "montgomery_square";
7159
7160 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
7161
7162 Node* a = argument(0);
7163 Node* n = argument(1);
7164 Node* len = argument(2);
7165 Node* inv = argument(3);
7166 Node* m = argument(5);
7167
7168 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
7169 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
7170 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
7171 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
7172 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
7173 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
7174 // failed array check
7175 return false;
7176 }
7177
7178 BasicType a_elem = a_type->elem()->array_element_basic_type();
7179 BasicType n_elem = n_type->elem()->array_element_basic_type();
7180 BasicType m_elem = m_type->elem()->array_element_basic_type();
7181 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
7182 return false;
7183 }
7184
7185 // Make the call
7186 {
7187 Node* a_start = array_element_address(a, intcon(0), a_elem);
7188 Node* n_start = array_element_address(n, intcon(0), n_elem);
7189 Node* m_start = array_element_address(m, intcon(0), m_elem);
7190
7191 Node* call = make_runtime_call(RC_LEAF,
7192 OptoRuntime::montgomerySquare_Type(),
7193 stubAddr, stubName, TypePtr::BOTTOM,
7194 a_start, n_start, len, inv, top(),
7195 m_start);
7196 set_result(m);
7197 }
7198
7199 return true;
7200 }
7201
7202 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
7203 address stubAddr = nullptr;
7204 const char* stubName = nullptr;
7205
7206 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
7207 if (stubAddr == nullptr) {
7208 return false; // Intrinsic's stub is not implemented on this platform
7209 }
7210
7211 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
7212
7213 assert(callee()->signature()->size() == 5, "expected 5 arguments");
7214
7215 Node* newArr = argument(0);
7216 Node* oldArr = argument(1);
7217 Node* newIdx = argument(2);
7218 Node* shiftCount = argument(3);
7219 Node* numIter = argument(4);
7220
7221 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
7222 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
7223 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
7224 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
7225 return false;
7226 }
7227
7228 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
7229 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
7230 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
7231 return false;
7232 }
7233
7234 // Make the call
7235 {
7236 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
7237 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
7238
7239 Node* call = make_runtime_call(RC_LEAF,
7240 OptoRuntime::bigIntegerShift_Type(),
7241 stubAddr,
7242 stubName,
7243 TypePtr::BOTTOM,
7244 newArr_start,
7245 oldArr_start,
7246 newIdx,
7247 shiftCount,
7248 numIter);
7249 }
7250
7251 return true;
7252 }
7253
7254 //-------------inline_vectorizedMismatch------------------------------
7255 bool LibraryCallKit::inline_vectorizedMismatch() {
7256 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
7257
7258 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
7259 Node* obja = argument(0); // Object
7260 Node* aoffset = argument(1); // long
7261 Node* objb = argument(3); // Object
7262 Node* boffset = argument(4); // long
7263 Node* length = argument(6); // int
7264 Node* scale = argument(7); // int
7265
7266 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
7267 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
7268 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
7269 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
7270 scale == top()) {
7271 return false; // failed input validation
7272 }
7273
7274 Node* obja_adr = make_unsafe_address(obja, aoffset);
7275 Node* objb_adr = make_unsafe_address(objb, boffset);
7276
7277 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
7278 //
7279 // inline_limit = ArrayOperationPartialInlineSize / element_size;
7280 // if (length <= inline_limit) {
7281 // inline_path:
7282 // vmask = VectorMaskGen length
7283 // vload1 = LoadVectorMasked obja, vmask
7284 // vload2 = LoadVectorMasked objb, vmask
7285 // result1 = VectorCmpMasked vload1, vload2, vmask
7286 // } else {
7287 // call_stub_path:
7288 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
7289 // }
7290 // exit_block:
7291 // return Phi(result1, result2);
7292 //
7293 enum { inline_path = 1, // input is small enough to process it all at once
7294 stub_path = 2, // input is too large; call into the VM
7295 PATH_LIMIT = 3
7296 };
7297
7298 Node* exit_block = new RegionNode(PATH_LIMIT);
7299 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
7300 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
7301
7302 Node* call_stub_path = control();
7303
7304 BasicType elem_bt = T_ILLEGAL;
7305
7306 const TypeInt* scale_t = _gvn.type(scale)->is_int();
7307 if (scale_t->is_con()) {
7308 switch (scale_t->get_con()) {
7309 case 0: elem_bt = T_BYTE; break;
7310 case 1: elem_bt = T_SHORT; break;
7311 case 2: elem_bt = T_INT; break;
7312 case 3: elem_bt = T_LONG; break;
7313
7314 default: elem_bt = T_ILLEGAL; break; // not supported
7315 }
7316 }
7317
7318 int inline_limit = 0;
7319 bool do_partial_inline = false;
7320
7321 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
7322 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
7323 do_partial_inline = inline_limit >= 16;
7324 }
7325
7326 if (do_partial_inline) {
7327 assert(elem_bt != T_ILLEGAL, "sanity");
7328
7329 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
7330 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
7331 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
7332
7333 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
7334 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
7335 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
7336
7337 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
7338
7339 if (!stopped()) {
7340 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
7341
7342 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
7343 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
7344 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
7345 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
7346
7347 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
7348 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
7349 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
7350 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
7351
7352 exit_block->init_req(inline_path, control());
7353 memory_phi->init_req(inline_path, map()->memory());
7354 result_phi->init_req(inline_path, result);
7355
7356 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
7357 clear_upper_avx();
7358 }
7359 }
7360 }
7361
7362 if (call_stub_path != nullptr) {
7363 set_control(call_stub_path);
7364
7365 Node* call = make_runtime_call(RC_LEAF,
7366 OptoRuntime::vectorizedMismatch_Type(),
7367 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
7368 obja_adr, objb_adr, length, scale);
7369
7370 exit_block->init_req(stub_path, control());
7371 memory_phi->init_req(stub_path, map()->memory());
7372 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
7373 }
7374
7375 exit_block = _gvn.transform(exit_block);
7376 memory_phi = _gvn.transform(memory_phi);
7377 result_phi = _gvn.transform(result_phi);
7378
7379 record_for_igvn(exit_block);
7380 record_for_igvn(memory_phi);
7381 record_for_igvn(result_phi);
7382
7383 set_control(exit_block);
7384 set_all_memory(memory_phi);
7385 set_result(result_phi);
7386
7387 return true;
7388 }
7389
7390 //------------------------------inline_vectorizedHashcode----------------------------
7391 bool LibraryCallKit::inline_vectorizedHashCode() {
7392 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
7393
7394 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
7395 Node* array = argument(0);
7396 Node* offset = argument(1);
7397 Node* length = argument(2);
7398 Node* initialValue = argument(3);
7399 Node* basic_type = argument(4);
7400
7401 if (basic_type == top()) {
7402 return false; // failed input validation
7403 }
7404
7405 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
7406 if (!basic_type_t->is_con()) {
7407 return false; // Only intrinsify if mode argument is constant
7408 }
7409
7410 array = must_be_not_null(array, true);
7411
7412 BasicType bt = (BasicType)basic_type_t->get_con();
7413
7414 // Resolve address of first element
7415 Node* array_start = array_element_address(array, offset, bt);
7416
7417 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
7418 array_start, length, initialValue, basic_type)));
7419 clear_upper_avx();
7420
7421 return true;
7422 }
7423
7424 /**
7425 * Calculate CRC32 for byte.
7426 * int java.util.zip.CRC32.update(int crc, int b)
7427 */
7428 bool LibraryCallKit::inline_updateCRC32() {
7429 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7430 assert(callee()->signature()->size() == 2, "update has 2 parameters");
7431 // no receiver since it is static method
7432 Node* crc = argument(0); // type: int
7433 Node* b = argument(1); // type: int
7434
7435 /*
7436 * int c = ~ crc;
7437 * b = timesXtoThe32[(b ^ c) & 0xFF];
7438 * b = b ^ (c >>> 8);
7439 * crc = ~b;
7440 */
7441
7442 Node* M1 = intcon(-1);
7443 crc = _gvn.transform(new XorINode(crc, M1));
7444 Node* result = _gvn.transform(new XorINode(crc, b));
7445 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
7446
7447 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
7448 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
7449 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
7450 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
7451
7452 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
7453 result = _gvn.transform(new XorINode(crc, result));
7454 result = _gvn.transform(new XorINode(result, M1));
7455 set_result(result);
7456 return true;
7457 }
7458
7459 /**
7460 * Calculate CRC32 for byte[] array.
7461 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
7462 */
7463 bool LibraryCallKit::inline_updateBytesCRC32() {
7464 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7465 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7466 // no receiver since it is static method
7467 Node* crc = argument(0); // type: int
7468 Node* src = argument(1); // type: oop
7469 Node* offset = argument(2); // type: int
7470 Node* length = argument(3); // type: int
7471
7472 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7473 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7474 // failed array check
7475 return false;
7476 }
7477
7478 // Figure out the size and type of the elements we will be copying.
7479 BasicType src_elem = src_type->elem()->array_element_basic_type();
7480 if (src_elem != T_BYTE) {
7481 return false;
7482 }
7483
7484 // 'src_start' points to src array + scaled offset
7485 src = must_be_not_null(src, true);
7486 Node* src_start = array_element_address(src, offset, src_elem);
7487
7488 // We assume that range check is done by caller.
7489 // TODO: generate range check (offset+length < src.length) in debug VM.
7490
7491 // Call the stub.
7492 address stubAddr = StubRoutines::updateBytesCRC32();
7493 const char *stubName = "updateBytesCRC32";
7494
7495 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7496 stubAddr, stubName, TypePtr::BOTTOM,
7497 crc, src_start, length);
7498 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7499 set_result(result);
7500 return true;
7501 }
7502
7503 /**
7504 * Calculate CRC32 for ByteBuffer.
7505 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
7506 */
7507 bool LibraryCallKit::inline_updateByteBufferCRC32() {
7508 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7509 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7510 // no receiver since it is static method
7511 Node* crc = argument(0); // type: int
7512 Node* src = argument(1); // type: long
7513 Node* offset = argument(3); // type: int
7514 Node* length = argument(4); // type: int
7515
7516 src = ConvL2X(src); // adjust Java long to machine word
7517 Node* base = _gvn.transform(new CastX2PNode(src));
7518 offset = ConvI2X(offset);
7519
7520 // 'src_start' points to src array + scaled offset
7521 Node* src_start = basic_plus_adr(top(), base, offset);
7522
7523 // Call the stub.
7524 address stubAddr = StubRoutines::updateBytesCRC32();
7525 const char *stubName = "updateBytesCRC32";
7526
7527 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7528 stubAddr, stubName, TypePtr::BOTTOM,
7529 crc, src_start, length);
7530 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7531 set_result(result);
7532 return true;
7533 }
7534
7535 //------------------------------get_table_from_crc32c_class-----------------------
7536 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
7537 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
7538 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
7539
7540 return table;
7541 }
7542
7543 //------------------------------inline_updateBytesCRC32C-----------------------
7544 //
7545 // Calculate CRC32C for byte[] array.
7546 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
7547 //
7548 bool LibraryCallKit::inline_updateBytesCRC32C() {
7549 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7550 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7551 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7552 // no receiver since it is a static method
7553 Node* crc = argument(0); // type: int
7554 Node* src = argument(1); // type: oop
7555 Node* offset = argument(2); // type: int
7556 Node* end = argument(3); // type: int
7557
7558 Node* length = _gvn.transform(new SubINode(end, offset));
7559
7560 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7561 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7562 // failed array check
7563 return false;
7564 }
7565
7566 // Figure out the size and type of the elements we will be copying.
7567 BasicType src_elem = src_type->elem()->array_element_basic_type();
7568 if (src_elem != T_BYTE) {
7569 return false;
7570 }
7571
7572 // 'src_start' points to src array + scaled offset
7573 src = must_be_not_null(src, true);
7574 Node* src_start = array_element_address(src, offset, src_elem);
7575
7576 // static final int[] byteTable in class CRC32C
7577 Node* table = get_table_from_crc32c_class(callee()->holder());
7578 table = must_be_not_null(table, true);
7579 Node* table_start = array_element_address(table, intcon(0), T_INT);
7580
7581 // We assume that range check is done by caller.
7582 // TODO: generate range check (offset+length < src.length) in debug VM.
7583
7584 // Call the stub.
7585 address stubAddr = StubRoutines::updateBytesCRC32C();
7586 const char *stubName = "updateBytesCRC32C";
7587
7588 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7589 stubAddr, stubName, TypePtr::BOTTOM,
7590 crc, src_start, length, table_start);
7591 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7592 set_result(result);
7593 return true;
7594 }
7595
7596 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
7597 //
7598 // Calculate CRC32C for DirectByteBuffer.
7599 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
7600 //
7601 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
7602 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7603 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
7604 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7605 // no receiver since it is a static method
7606 Node* crc = argument(0); // type: int
7607 Node* src = argument(1); // type: long
7608 Node* offset = argument(3); // type: int
7609 Node* end = argument(4); // type: int
7610
7611 Node* length = _gvn.transform(new SubINode(end, offset));
7612
7613 src = ConvL2X(src); // adjust Java long to machine word
7614 Node* base = _gvn.transform(new CastX2PNode(src));
7615 offset = ConvI2X(offset);
7616
7617 // 'src_start' points to src array + scaled offset
7618 Node* src_start = basic_plus_adr(top(), base, offset);
7619
7620 // static final int[] byteTable in class CRC32C
7621 Node* table = get_table_from_crc32c_class(callee()->holder());
7622 table = must_be_not_null(table, true);
7623 Node* table_start = array_element_address(table, intcon(0), T_INT);
7624
7625 // Call the stub.
7626 address stubAddr = StubRoutines::updateBytesCRC32C();
7627 const char *stubName = "updateBytesCRC32C";
7628
7629 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7630 stubAddr, stubName, TypePtr::BOTTOM,
7631 crc, src_start, length, table_start);
7632 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7633 set_result(result);
7634 return true;
7635 }
7636
7637 //------------------------------inline_updateBytesAdler32----------------------
7638 //
7639 // Calculate Adler32 checksum for byte[] array.
7640 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
7641 //
7642 bool LibraryCallKit::inline_updateBytesAdler32() {
7643 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7644 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7645 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7646 // no receiver since it is static method
7647 Node* crc = argument(0); // type: int
7648 Node* src = argument(1); // type: oop
7649 Node* offset = argument(2); // type: int
7650 Node* length = argument(3); // type: int
7651
7652 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7653 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7654 // failed array check
7655 return false;
7656 }
7657
7658 // Figure out the size and type of the elements we will be copying.
7659 BasicType src_elem = src_type->elem()->array_element_basic_type();
7660 if (src_elem != T_BYTE) {
7661 return false;
7662 }
7663
7664 // 'src_start' points to src array + scaled offset
7665 Node* src_start = array_element_address(src, offset, src_elem);
7666
7667 // We assume that range check is done by caller.
7668 // TODO: generate range check (offset+length < src.length) in debug VM.
7669
7670 // Call the stub.
7671 address stubAddr = StubRoutines::updateBytesAdler32();
7672 const char *stubName = "updateBytesAdler32";
7673
7674 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7675 stubAddr, stubName, TypePtr::BOTTOM,
7676 crc, src_start, length);
7677 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7678 set_result(result);
7679 return true;
7680 }
7681
7682 //------------------------------inline_updateByteBufferAdler32---------------
7683 //
7684 // Calculate Adler32 checksum for DirectByteBuffer.
7685 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
7686 //
7687 bool LibraryCallKit::inline_updateByteBufferAdler32() {
7688 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7689 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7690 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7691 // no receiver since it is static method
7692 Node* crc = argument(0); // type: int
7693 Node* src = argument(1); // type: long
7694 Node* offset = argument(3); // type: int
7695 Node* length = argument(4); // type: int
7696
7697 src = ConvL2X(src); // adjust Java long to machine word
7698 Node* base = _gvn.transform(new CastX2PNode(src));
7699 offset = ConvI2X(offset);
7700
7701 // 'src_start' points to src array + scaled offset
7702 Node* src_start = basic_plus_adr(top(), base, offset);
7703
7704 // Call the stub.
7705 address stubAddr = StubRoutines::updateBytesAdler32();
7706 const char *stubName = "updateBytesAdler32";
7707
7708 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7709 stubAddr, stubName, TypePtr::BOTTOM,
7710 crc, src_start, length);
7711
7712 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7713 set_result(result);
7714 return true;
7715 }
7716
7717 //----------------------------inline_reference_get0----------------------------
7718 // public T java.lang.ref.Reference.get();
7719 bool LibraryCallKit::inline_reference_get0() {
7720 const int referent_offset = java_lang_ref_Reference::referent_offset();
7721
7722 // Get the argument:
7723 Node* reference_obj = null_check_receiver();
7724 if (stopped()) return true;
7725
7726 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7727 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7728 decorators, /*is_static*/ false, nullptr);
7729 if (result == nullptr) return false;
7730
7731 // Add memory barrier to prevent commoning reads from this field
7732 // across safepoint since GC can change its value.
7733 insert_mem_bar(Op_MemBarCPUOrder);
7734
7735 set_result(result);
7736 return true;
7737 }
7738
7739 //----------------------------inline_reference_refersTo0----------------------------
7740 // bool java.lang.ref.Reference.refersTo0();
7741 // bool java.lang.ref.PhantomReference.refersTo0();
7742 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7743 // Get arguments:
7744 Node* reference_obj = null_check_receiver();
7745 Node* other_obj = argument(1);
7746 if (stopped()) return true;
7747
7748 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7749 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7750 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7751 decorators, /*is_static*/ false, nullptr);
7752 if (referent == nullptr) return false;
7753
7754 // Add memory barrier to prevent commoning reads from this field
7755 // across safepoint since GC can change its value.
7756 insert_mem_bar(Op_MemBarCPUOrder);
7757
7758 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7759 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7760 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7761
7762 RegionNode* region = new RegionNode(3);
7763 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7764
7765 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7766 region->init_req(1, if_true);
7767 phi->init_req(1, intcon(1));
7768
7769 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7770 region->init_req(2, if_false);
7771 phi->init_req(2, intcon(0));
7772
7773 set_control(_gvn.transform(region));
7774 record_for_igvn(region);
7775 set_result(_gvn.transform(phi));
7776 return true;
7777 }
7778
7779 //----------------------------inline_reference_clear0----------------------------
7780 // void java.lang.ref.Reference.clear0();
7781 // void java.lang.ref.PhantomReference.clear0();
7782 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7783 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7784
7785 // Get arguments
7786 Node* reference_obj = null_check_receiver();
7787 if (stopped()) return true;
7788
7789 // Common access parameters
7790 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7791 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7792 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7793 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7794 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7795
7796 Node* referent = access_load_at(reference_obj,
7797 referent_field_addr,
7798 referent_field_addr_type,
7799 val_type,
7800 T_OBJECT,
7801 decorators);
7802
7803 IdealKit ideal(this);
7804 #define __ ideal.
7805 __ if_then(referent, BoolTest::ne, null());
7806 sync_kit(ideal);
7807 access_store_at(reference_obj,
7808 referent_field_addr,
7809 referent_field_addr_type,
7810 null(),
7811 val_type,
7812 T_OBJECT,
7813 decorators);
7814 __ sync_kit(this);
7815 __ end_if();
7816 final_sync(ideal);
7817 #undef __
7818
7819 return true;
7820 }
7821
7822 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7823 DecoratorSet decorators, bool is_static,
7824 ciInstanceKlass* fromKls) {
7825 if (fromKls == nullptr) {
7826 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7827 assert(tinst != nullptr, "obj is null");
7828 assert(tinst->is_loaded(), "obj is not loaded");
7829 fromKls = tinst->instance_klass();
7830 } else {
7831 assert(is_static, "only for static field access");
7832 }
7833 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7834 ciSymbol::make(fieldTypeString),
7835 is_static);
7836
7837 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7838 if (field == nullptr) return (Node *) nullptr;
7839
7840 if (is_static) {
7841 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7842 fromObj = makecon(tip);
7843 }
7844
7845 // Next code copied from Parse::do_get_xxx():
7846
7847 // Compute address and memory type.
7848 int offset = field->offset_in_bytes();
7849 bool is_vol = field->is_volatile();
7850 ciType* field_klass = field->type();
7851 assert(field_klass->is_loaded(), "should be loaded");
7852 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7853 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7854 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7855 "slice of address and input slice don't match");
7856 BasicType bt = field->layout_type();
7857
7858 // Build the resultant type of the load
7859 const Type *type;
7860 if (bt == T_OBJECT) {
7861 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7862 } else {
7863 type = Type::get_const_basic_type(bt);
7864 }
7865
7866 if (is_vol) {
7867 decorators |= MO_SEQ_CST;
7868 }
7869
7870 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7871 }
7872
7873 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7874 bool is_exact /* true */, bool is_static /* false */,
7875 ciInstanceKlass * fromKls /* nullptr */) {
7876 if (fromKls == nullptr) {
7877 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7878 assert(tinst != nullptr, "obj is null");
7879 assert(tinst->is_loaded(), "obj is not loaded");
7880 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7881 fromKls = tinst->instance_klass();
7882 }
7883 else {
7884 assert(is_static, "only for static field access");
7885 }
7886 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7887 ciSymbol::make(fieldTypeString),
7888 is_static);
7889
7890 assert(field != nullptr, "undefined field");
7891 assert(!field->is_volatile(), "not defined for volatile fields");
7892
7893 if (is_static) {
7894 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7895 fromObj = makecon(tip);
7896 }
7897
7898 // Next code copied from Parse::do_get_xxx():
7899
7900 // Compute address and memory type.
7901 int offset = field->offset_in_bytes();
7902 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7903
7904 return adr;
7905 }
7906
7907 //------------------------------inline_aescrypt_Block-----------------------
7908 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7909 address stubAddr = nullptr;
7910 const char *stubName;
7911 bool is_decrypt = false;
7912 assert(UseAES, "need AES instruction support");
7913
7914 switch(id) {
7915 case vmIntrinsics::_aescrypt_encryptBlock:
7916 stubAddr = StubRoutines::aescrypt_encryptBlock();
7917 stubName = "aescrypt_encryptBlock";
7918 break;
7919 case vmIntrinsics::_aescrypt_decryptBlock:
7920 stubAddr = StubRoutines::aescrypt_decryptBlock();
7921 stubName = "aescrypt_decryptBlock";
7922 is_decrypt = true;
7923 break;
7924 default:
7925 break;
7926 }
7927 if (stubAddr == nullptr) return false;
7928
7929 Node* aescrypt_object = argument(0);
7930 Node* src = argument(1);
7931 Node* src_offset = argument(2);
7932 Node* dest = argument(3);
7933 Node* dest_offset = argument(4);
7934
7935 src = must_be_not_null(src, true);
7936 dest = must_be_not_null(dest, true);
7937
7938 // (1) src and dest are arrays.
7939 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7940 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7941 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7942 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7943
7944 // for the quick and dirty code we will skip all the checks.
7945 // we are just trying to get the call to be generated.
7946 Node* src_start = src;
7947 Node* dest_start = dest;
7948 if (src_offset != nullptr || dest_offset != nullptr) {
7949 assert(src_offset != nullptr && dest_offset != nullptr, "");
7950 src_start = array_element_address(src, src_offset, T_BYTE);
7951 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7952 }
7953
7954 // now need to get the start of its expanded key array
7955 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7956 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7957 if (k_start == nullptr) return false;
7958
7959 // Call the stub.
7960 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7961 stubAddr, stubName, TypePtr::BOTTOM,
7962 src_start, dest_start, k_start);
7963
7964 return true;
7965 }
7966
7967 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7968 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7969 address stubAddr = nullptr;
7970 const char *stubName = nullptr;
7971 bool is_decrypt = false;
7972 assert(UseAES, "need AES instruction support");
7973
7974 switch(id) {
7975 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7976 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7977 stubName = "cipherBlockChaining_encryptAESCrypt";
7978 break;
7979 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7980 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7981 stubName = "cipherBlockChaining_decryptAESCrypt";
7982 is_decrypt = true;
7983 break;
7984 default:
7985 break;
7986 }
7987 if (stubAddr == nullptr) return false;
7988
7989 Node* cipherBlockChaining_object = argument(0);
7990 Node* src = argument(1);
7991 Node* src_offset = argument(2);
7992 Node* len = argument(3);
7993 Node* dest = argument(4);
7994 Node* dest_offset = argument(5);
7995
7996 src = must_be_not_null(src, false);
7997 dest = must_be_not_null(dest, false);
7998
7999 // (1) src and dest are arrays.
8000 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8001 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8002 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8003 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8004
8005 // checks are the responsibility of the caller
8006 Node* src_start = src;
8007 Node* dest_start = dest;
8008 if (src_offset != nullptr || dest_offset != nullptr) {
8009 assert(src_offset != nullptr && dest_offset != nullptr, "");
8010 src_start = array_element_address(src, src_offset, T_BYTE);
8011 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8012 }
8013
8014 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8015 // (because of the predicated logic executed earlier).
8016 // so we cast it here safely.
8017 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8018
8019 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8020 if (embeddedCipherObj == nullptr) return false;
8021
8022 // cast it to what we know it will be at runtime
8023 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
8024 assert(tinst != nullptr, "CBC obj is null");
8025 assert(tinst->is_loaded(), "CBC obj is not loaded");
8026 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8027 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8028
8029 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8030 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8031 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8032 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8033 aescrypt_object = _gvn.transform(aescrypt_object);
8034
8035 // we need to get the start of the aescrypt_object's expanded key array
8036 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
8037 if (k_start == nullptr) return false;
8038
8039 // similarly, get the start address of the r vector
8040 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
8041 if (objRvec == nullptr) return false;
8042 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
8043
8044 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8045 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8046 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
8047 stubAddr, stubName, TypePtr::BOTTOM,
8048 src_start, dest_start, k_start, r_start, len);
8049
8050 // return cipher length (int)
8051 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
8052 set_result(retvalue);
8053 return true;
8054 }
8055
8056 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
8057 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
8058 address stubAddr = nullptr;
8059 const char *stubName = nullptr;
8060 bool is_decrypt = false;
8061 assert(UseAES, "need AES instruction support");
8062
8063 switch (id) {
8064 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
8065 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
8066 stubName = "electronicCodeBook_encryptAESCrypt";
8067 break;
8068 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
8069 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
8070 stubName = "electronicCodeBook_decryptAESCrypt";
8071 is_decrypt = true;
8072 break;
8073 default:
8074 break;
8075 }
8076
8077 if (stubAddr == nullptr) return false;
8078
8079 Node* electronicCodeBook_object = argument(0);
8080 Node* src = argument(1);
8081 Node* src_offset = argument(2);
8082 Node* len = argument(3);
8083 Node* dest = argument(4);
8084 Node* dest_offset = argument(5);
8085
8086 // (1) src and dest are arrays.
8087 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8088 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8089 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8090 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8091
8092 // checks are the responsibility of the caller
8093 Node* src_start = src;
8094 Node* dest_start = dest;
8095 if (src_offset != nullptr || dest_offset != nullptr) {
8096 assert(src_offset != nullptr && dest_offset != nullptr, "");
8097 src_start = array_element_address(src, src_offset, T_BYTE);
8098 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8099 }
8100
8101 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8102 // (because of the predicated logic executed earlier).
8103 // so we cast it here safely.
8104 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8105
8106 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8107 if (embeddedCipherObj == nullptr) return false;
8108
8109 // cast it to what we know it will be at runtime
8110 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
8111 assert(tinst != nullptr, "ECB obj is null");
8112 assert(tinst->is_loaded(), "ECB obj is not loaded");
8113 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8114 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8115
8116 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8117 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8118 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8119 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8120 aescrypt_object = _gvn.transform(aescrypt_object);
8121
8122 // we need to get the start of the aescrypt_object's expanded key array
8123 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
8124 if (k_start == nullptr) return false;
8125
8126 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8127 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
8128 OptoRuntime::electronicCodeBook_aescrypt_Type(),
8129 stubAddr, stubName, TypePtr::BOTTOM,
8130 src_start, dest_start, k_start, len);
8131
8132 // return cipher length (int)
8133 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
8134 set_result(retvalue);
8135 return true;
8136 }
8137
8138 //------------------------------inline_counterMode_AESCrypt-----------------------
8139 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
8140 assert(UseAES, "need AES instruction support");
8141 if (!UseAESCTRIntrinsics) return false;
8142
8143 address stubAddr = nullptr;
8144 const char *stubName = nullptr;
8145 if (id == vmIntrinsics::_counterMode_AESCrypt) {
8146 stubAddr = StubRoutines::counterMode_AESCrypt();
8147 stubName = "counterMode_AESCrypt";
8148 }
8149 if (stubAddr == nullptr) return false;
8150
8151 Node* counterMode_object = argument(0);
8152 Node* src = argument(1);
8153 Node* src_offset = argument(2);
8154 Node* len = argument(3);
8155 Node* dest = argument(4);
8156 Node* dest_offset = argument(5);
8157
8158 // (1) src and dest are arrays.
8159 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8160 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8161 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8162 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8163
8164 // checks are the responsibility of the caller
8165 Node* src_start = src;
8166 Node* dest_start = dest;
8167 if (src_offset != nullptr || dest_offset != nullptr) {
8168 assert(src_offset != nullptr && dest_offset != nullptr, "");
8169 src_start = array_element_address(src, src_offset, T_BYTE);
8170 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8171 }
8172
8173 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8174 // (because of the predicated logic executed earlier).
8175 // so we cast it here safely.
8176 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8177 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8178 if (embeddedCipherObj == nullptr) return false;
8179 // cast it to what we know it will be at runtime
8180 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
8181 assert(tinst != nullptr, "CTR obj is null");
8182 assert(tinst->is_loaded(), "CTR obj is not loaded");
8183 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8184 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8185 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8186 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8187 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8188 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8189 aescrypt_object = _gvn.transform(aescrypt_object);
8190 // we need to get the start of the aescrypt_object's expanded key array
8191 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
8192 if (k_start == nullptr) return false;
8193 // similarly, get the start address of the r vector
8194 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
8195 if (obj_counter == nullptr) return false;
8196 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
8197
8198 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
8199 if (saved_encCounter == nullptr) return false;
8200 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
8201 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
8202
8203 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8204 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8205 OptoRuntime::counterMode_aescrypt_Type(),
8206 stubAddr, stubName, TypePtr::BOTTOM,
8207 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
8208
8209 // return cipher length (int)
8210 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
8211 set_result(retvalue);
8212 return true;
8213 }
8214
8215 //------------------------------get_key_start_from_aescrypt_object-----------------------
8216 Node* LibraryCallKit::get_key_start_from_aescrypt_object(Node* aescrypt_object, bool is_decrypt) {
8217 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
8218 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
8219 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
8220 // The following platform specific stubs of encryption and decryption use the same round keys.
8221 #if defined(PPC64) || defined(S390) || defined(RISCV64)
8222 bool use_decryption_key = false;
8223 #else
8224 bool use_decryption_key = is_decrypt;
8225 #endif
8226 Node* objAESCryptKey = load_field_from_object(aescrypt_object, use_decryption_key ? "sessionKd" : "sessionKe", "[I");
8227 assert(objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AES_Crypt");
8228 if (objAESCryptKey == nullptr) return (Node *) nullptr;
8229
8230 // now have the array, need to get the start address of the selected key array
8231 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
8232 return k_start;
8233 }
8234
8235 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
8236 // Return node representing slow path of predicate check.
8237 // the pseudo code we want to emulate with this predicate is:
8238 // for encryption:
8239 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8240 // for decryption:
8241 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8242 // note cipher==plain is more conservative than the original java code but that's OK
8243 //
8244 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
8245 // The receiver was checked for null already.
8246 Node* objCBC = argument(0);
8247
8248 Node* src = argument(1);
8249 Node* dest = argument(4);
8250
8251 // Load embeddedCipher field of CipherBlockChaining object.
8252 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8253
8254 // get AESCrypt klass for instanceOf check
8255 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8256 // will have same classloader as CipherBlockChaining object
8257 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
8258 assert(tinst != nullptr, "CBCobj is null");
8259 assert(tinst->is_loaded(), "CBCobj is not loaded");
8260
8261 // we want to do an instanceof comparison against the AESCrypt class
8262 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8263 if (!klass_AESCrypt->is_loaded()) {
8264 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8265 Node* ctrl = control();
8266 set_control(top()); // no regular fast path
8267 return ctrl;
8268 }
8269
8270 src = must_be_not_null(src, true);
8271 dest = must_be_not_null(dest, true);
8272
8273 // Resolve oops to stable for CmpP below.
8274 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8275
8276 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8277 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8278 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8279
8280 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8281
8282 // for encryption, we are done
8283 if (!decrypting)
8284 return instof_false; // even if it is null
8285
8286 // for decryption, we need to add a further check to avoid
8287 // taking the intrinsic path when cipher and plain are the same
8288 // see the original java code for why.
8289 RegionNode* region = new RegionNode(3);
8290 region->init_req(1, instof_false);
8291
8292 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8293 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8294 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8295 region->init_req(2, src_dest_conjoint);
8296
8297 record_for_igvn(region);
8298 return _gvn.transform(region);
8299 }
8300
8301 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
8302 // Return node representing slow path of predicate check.
8303 // the pseudo code we want to emulate with this predicate is:
8304 // for encryption:
8305 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8306 // for decryption:
8307 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8308 // note cipher==plain is more conservative than the original java code but that's OK
8309 //
8310 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
8311 // The receiver was checked for null already.
8312 Node* objECB = argument(0);
8313
8314 // Load embeddedCipher field of ElectronicCodeBook object.
8315 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8316
8317 // get AESCrypt klass for instanceOf check
8318 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8319 // will have same classloader as ElectronicCodeBook object
8320 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
8321 assert(tinst != nullptr, "ECBobj is null");
8322 assert(tinst->is_loaded(), "ECBobj is not loaded");
8323
8324 // we want to do an instanceof comparison against the AESCrypt class
8325 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8326 if (!klass_AESCrypt->is_loaded()) {
8327 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8328 Node* ctrl = control();
8329 set_control(top()); // no regular fast path
8330 return ctrl;
8331 }
8332 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8333
8334 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8335 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8336 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8337
8338 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8339
8340 // for encryption, we are done
8341 if (!decrypting)
8342 return instof_false; // even if it is null
8343
8344 // for decryption, we need to add a further check to avoid
8345 // taking the intrinsic path when cipher and plain are the same
8346 // see the original java code for why.
8347 RegionNode* region = new RegionNode(3);
8348 region->init_req(1, instof_false);
8349 Node* src = argument(1);
8350 Node* dest = argument(4);
8351 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8352 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8353 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8354 region->init_req(2, src_dest_conjoint);
8355
8356 record_for_igvn(region);
8357 return _gvn.transform(region);
8358 }
8359
8360 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
8361 // Return node representing slow path of predicate check.
8362 // the pseudo code we want to emulate with this predicate is:
8363 // for encryption:
8364 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8365 // for decryption:
8366 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8367 // note cipher==plain is more conservative than the original java code but that's OK
8368 //
8369
8370 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
8371 // The receiver was checked for null already.
8372 Node* objCTR = argument(0);
8373
8374 // Load embeddedCipher field of CipherBlockChaining object.
8375 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8376
8377 // get AESCrypt klass for instanceOf check
8378 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8379 // will have same classloader as CipherBlockChaining object
8380 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
8381 assert(tinst != nullptr, "CTRobj is null");
8382 assert(tinst->is_loaded(), "CTRobj is not loaded");
8383
8384 // we want to do an instanceof comparison against the AESCrypt class
8385 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8386 if (!klass_AESCrypt->is_loaded()) {
8387 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8388 Node* ctrl = control();
8389 set_control(top()); // no regular fast path
8390 return ctrl;
8391 }
8392
8393 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8394 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8395 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8396 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8397 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8398
8399 return instof_false; // even if it is null
8400 }
8401
8402 //------------------------------inline_ghash_processBlocks
8403 bool LibraryCallKit::inline_ghash_processBlocks() {
8404 address stubAddr;
8405 const char *stubName;
8406 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
8407
8408 stubAddr = StubRoutines::ghash_processBlocks();
8409 stubName = "ghash_processBlocks";
8410
8411 Node* data = argument(0);
8412 Node* offset = argument(1);
8413 Node* len = argument(2);
8414 Node* state = argument(3);
8415 Node* subkeyH = argument(4);
8416
8417 state = must_be_not_null(state, true);
8418 subkeyH = must_be_not_null(subkeyH, true);
8419 data = must_be_not_null(data, true);
8420
8421 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8422 assert(state_start, "state is null");
8423 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
8424 assert(subkeyH_start, "subkeyH is null");
8425 Node* data_start = array_element_address(data, offset, T_BYTE);
8426 assert(data_start, "data is null");
8427
8428 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
8429 OptoRuntime::ghash_processBlocks_Type(),
8430 stubAddr, stubName, TypePtr::BOTTOM,
8431 state_start, subkeyH_start, data_start, len);
8432 return true;
8433 }
8434
8435 //------------------------------inline_chacha20Block
8436 bool LibraryCallKit::inline_chacha20Block() {
8437 address stubAddr;
8438 const char *stubName;
8439 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
8440
8441 stubAddr = StubRoutines::chacha20Block();
8442 stubName = "chacha20Block";
8443
8444 Node* state = argument(0);
8445 Node* result = argument(1);
8446
8447 state = must_be_not_null(state, true);
8448 result = must_be_not_null(result, true);
8449
8450 Node* state_start = array_element_address(state, intcon(0), T_INT);
8451 assert(state_start, "state is null");
8452 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
8453 assert(result_start, "result is null");
8454
8455 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
8456 OptoRuntime::chacha20Block_Type(),
8457 stubAddr, stubName, TypePtr::BOTTOM,
8458 state_start, result_start);
8459 // return key stream length (int)
8460 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
8461 set_result(retvalue);
8462 return true;
8463 }
8464
8465 //------------------------------inline_kyberNtt
8466 bool LibraryCallKit::inline_kyberNtt() {
8467 address stubAddr;
8468 const char *stubName;
8469 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8470 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
8471
8472 stubAddr = StubRoutines::kyberNtt();
8473 stubName = "kyberNtt";
8474 if (!stubAddr) return false;
8475
8476 Node* coeffs = argument(0);
8477 Node* ntt_zetas = argument(1);
8478
8479 coeffs = must_be_not_null(coeffs, true);
8480 ntt_zetas = must_be_not_null(ntt_zetas, true);
8481
8482 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8483 assert(coeffs_start, "coeffs is null");
8484 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
8485 assert(ntt_zetas_start, "ntt_zetas is null");
8486 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8487 OptoRuntime::kyberNtt_Type(),
8488 stubAddr, stubName, TypePtr::BOTTOM,
8489 coeffs_start, ntt_zetas_start);
8490 // return an int
8491 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
8492 set_result(retvalue);
8493 return true;
8494 }
8495
8496 //------------------------------inline_kyberInverseNtt
8497 bool LibraryCallKit::inline_kyberInverseNtt() {
8498 address stubAddr;
8499 const char *stubName;
8500 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8501 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
8502
8503 stubAddr = StubRoutines::kyberInverseNtt();
8504 stubName = "kyberInverseNtt";
8505 if (!stubAddr) return false;
8506
8507 Node* coeffs = argument(0);
8508 Node* zetas = argument(1);
8509
8510 coeffs = must_be_not_null(coeffs, true);
8511 zetas = must_be_not_null(zetas, true);
8512
8513 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8514 assert(coeffs_start, "coeffs is null");
8515 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8516 assert(zetas_start, "inverseNtt_zetas is null");
8517 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8518 OptoRuntime::kyberInverseNtt_Type(),
8519 stubAddr, stubName, TypePtr::BOTTOM,
8520 coeffs_start, zetas_start);
8521
8522 // return an int
8523 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
8524 set_result(retvalue);
8525 return true;
8526 }
8527
8528 //------------------------------inline_kyberNttMult
8529 bool LibraryCallKit::inline_kyberNttMult() {
8530 address stubAddr;
8531 const char *stubName;
8532 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8533 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
8534
8535 stubAddr = StubRoutines::kyberNttMult();
8536 stubName = "kyberNttMult";
8537 if (!stubAddr) return false;
8538
8539 Node* result = argument(0);
8540 Node* ntta = argument(1);
8541 Node* nttb = argument(2);
8542 Node* zetas = argument(3);
8543
8544 result = must_be_not_null(result, true);
8545 ntta = must_be_not_null(ntta, true);
8546 nttb = must_be_not_null(nttb, true);
8547 zetas = must_be_not_null(zetas, true);
8548
8549 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8550 assert(result_start, "result is null");
8551 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
8552 assert(ntta_start, "ntta is null");
8553 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
8554 assert(nttb_start, "nttb is null");
8555 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8556 assert(zetas_start, "nttMult_zetas is null");
8557 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8558 OptoRuntime::kyberNttMult_Type(),
8559 stubAddr, stubName, TypePtr::BOTTOM,
8560 result_start, ntta_start, nttb_start,
8561 zetas_start);
8562
8563 // return an int
8564 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
8565 set_result(retvalue);
8566
8567 return true;
8568 }
8569
8570 //------------------------------inline_kyberAddPoly_2
8571 bool LibraryCallKit::inline_kyberAddPoly_2() {
8572 address stubAddr;
8573 const char *stubName;
8574 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8575 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
8576
8577 stubAddr = StubRoutines::kyberAddPoly_2();
8578 stubName = "kyberAddPoly_2";
8579 if (!stubAddr) return false;
8580
8581 Node* result = argument(0);
8582 Node* a = argument(1);
8583 Node* b = argument(2);
8584
8585 result = must_be_not_null(result, true);
8586 a = must_be_not_null(a, true);
8587 b = must_be_not_null(b, true);
8588
8589 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8590 assert(result_start, "result is null");
8591 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8592 assert(a_start, "a is null");
8593 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8594 assert(b_start, "b is null");
8595 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
8596 OptoRuntime::kyberAddPoly_2_Type(),
8597 stubAddr, stubName, TypePtr::BOTTOM,
8598 result_start, a_start, b_start);
8599 // return an int
8600 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
8601 set_result(retvalue);
8602 return true;
8603 }
8604
8605 //------------------------------inline_kyberAddPoly_3
8606 bool LibraryCallKit::inline_kyberAddPoly_3() {
8607 address stubAddr;
8608 const char *stubName;
8609 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8610 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
8611
8612 stubAddr = StubRoutines::kyberAddPoly_3();
8613 stubName = "kyberAddPoly_3";
8614 if (!stubAddr) return false;
8615
8616 Node* result = argument(0);
8617 Node* a = argument(1);
8618 Node* b = argument(2);
8619 Node* c = argument(3);
8620
8621 result = must_be_not_null(result, true);
8622 a = must_be_not_null(a, true);
8623 b = must_be_not_null(b, true);
8624 c = must_be_not_null(c, true);
8625
8626 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8627 assert(result_start, "result is null");
8628 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8629 assert(a_start, "a is null");
8630 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8631 assert(b_start, "b is null");
8632 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
8633 assert(c_start, "c is null");
8634 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
8635 OptoRuntime::kyberAddPoly_3_Type(),
8636 stubAddr, stubName, TypePtr::BOTTOM,
8637 result_start, a_start, b_start, c_start);
8638 // return an int
8639 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
8640 set_result(retvalue);
8641 return true;
8642 }
8643
8644 //------------------------------inline_kyber12To16
8645 bool LibraryCallKit::inline_kyber12To16() {
8646 address stubAddr;
8647 const char *stubName;
8648 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8649 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
8650
8651 stubAddr = StubRoutines::kyber12To16();
8652 stubName = "kyber12To16";
8653 if (!stubAddr) return false;
8654
8655 Node* condensed = argument(0);
8656 Node* condensedOffs = argument(1);
8657 Node* parsed = argument(2);
8658 Node* parsedLength = argument(3);
8659
8660 condensed = must_be_not_null(condensed, true);
8661 parsed = must_be_not_null(parsed, true);
8662
8663 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
8664 assert(condensed_start, "condensed is null");
8665 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
8666 assert(parsed_start, "parsed is null");
8667 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
8668 OptoRuntime::kyber12To16_Type(),
8669 stubAddr, stubName, TypePtr::BOTTOM,
8670 condensed_start, condensedOffs, parsed_start, parsedLength);
8671 // return an int
8672 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
8673 set_result(retvalue);
8674 return true;
8675
8676 }
8677
8678 //------------------------------inline_kyberBarrettReduce
8679 bool LibraryCallKit::inline_kyberBarrettReduce() {
8680 address stubAddr;
8681 const char *stubName;
8682 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8683 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
8684
8685 stubAddr = StubRoutines::kyberBarrettReduce();
8686 stubName = "kyberBarrettReduce";
8687 if (!stubAddr) return false;
8688
8689 Node* coeffs = argument(0);
8690
8691 coeffs = must_be_not_null(coeffs, true);
8692
8693 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8694 assert(coeffs_start, "coeffs is null");
8695 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
8696 OptoRuntime::kyberBarrettReduce_Type(),
8697 stubAddr, stubName, TypePtr::BOTTOM,
8698 coeffs_start);
8699 // return an int
8700 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
8701 set_result(retvalue);
8702 return true;
8703 }
8704
8705 //------------------------------inline_dilithiumAlmostNtt
8706 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
8707 address stubAddr;
8708 const char *stubName;
8709 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8710 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
8711
8712 stubAddr = StubRoutines::dilithiumAlmostNtt();
8713 stubName = "dilithiumAlmostNtt";
8714 if (!stubAddr) return false;
8715
8716 Node* coeffs = argument(0);
8717 Node* ntt_zetas = argument(1);
8718
8719 coeffs = must_be_not_null(coeffs, true);
8720 ntt_zetas = must_be_not_null(ntt_zetas, true);
8721
8722 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8723 assert(coeffs_start, "coeffs is null");
8724 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
8725 assert(ntt_zetas_start, "ntt_zetas is null");
8726 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8727 OptoRuntime::dilithiumAlmostNtt_Type(),
8728 stubAddr, stubName, TypePtr::BOTTOM,
8729 coeffs_start, ntt_zetas_start);
8730 // return an int
8731 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
8732 set_result(retvalue);
8733 return true;
8734 }
8735
8736 //------------------------------inline_dilithiumAlmostInverseNtt
8737 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
8738 address stubAddr;
8739 const char *stubName;
8740 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8741 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
8742
8743 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
8744 stubName = "dilithiumAlmostInverseNtt";
8745 if (!stubAddr) return false;
8746
8747 Node* coeffs = argument(0);
8748 Node* zetas = argument(1);
8749
8750 coeffs = must_be_not_null(coeffs, true);
8751 zetas = must_be_not_null(zetas, true);
8752
8753 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8754 assert(coeffs_start, "coeffs is null");
8755 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
8756 assert(zetas_start, "inverseNtt_zetas is null");
8757 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8758 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
8759 stubAddr, stubName, TypePtr::BOTTOM,
8760 coeffs_start, zetas_start);
8761 // return an int
8762 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
8763 set_result(retvalue);
8764 return true;
8765 }
8766
8767 //------------------------------inline_dilithiumNttMult
8768 bool LibraryCallKit::inline_dilithiumNttMult() {
8769 address stubAddr;
8770 const char *stubName;
8771 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8772 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
8773
8774 stubAddr = StubRoutines::dilithiumNttMult();
8775 stubName = "dilithiumNttMult";
8776 if (!stubAddr) return false;
8777
8778 Node* result = argument(0);
8779 Node* ntta = argument(1);
8780 Node* nttb = argument(2);
8781 Node* zetas = argument(3);
8782
8783 result = must_be_not_null(result, true);
8784 ntta = must_be_not_null(ntta, true);
8785 nttb = must_be_not_null(nttb, true);
8786 zetas = must_be_not_null(zetas, true);
8787
8788 Node* result_start = array_element_address(result, intcon(0), T_INT);
8789 assert(result_start, "result is null");
8790 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
8791 assert(ntta_start, "ntta is null");
8792 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
8793 assert(nttb_start, "nttb is null");
8794 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8795 OptoRuntime::dilithiumNttMult_Type(),
8796 stubAddr, stubName, TypePtr::BOTTOM,
8797 result_start, ntta_start, nttb_start);
8798
8799 // return an int
8800 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8801 set_result(retvalue);
8802
8803 return true;
8804 }
8805
8806 //------------------------------inline_dilithiumMontMulByConstant
8807 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8808 address stubAddr;
8809 const char *stubName;
8810 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8811 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8812
8813 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8814 stubName = "dilithiumMontMulByConstant";
8815 if (!stubAddr) return false;
8816
8817 Node* coeffs = argument(0);
8818 Node* constant = argument(1);
8819
8820 coeffs = must_be_not_null(coeffs, true);
8821
8822 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8823 assert(coeffs_start, "coeffs is null");
8824 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8825 OptoRuntime::dilithiumMontMulByConstant_Type(),
8826 stubAddr, stubName, TypePtr::BOTTOM,
8827 coeffs_start, constant);
8828
8829 // return an int
8830 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8831 set_result(retvalue);
8832 return true;
8833 }
8834
8835
8836 //------------------------------inline_dilithiumDecomposePoly
8837 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8838 address stubAddr;
8839 const char *stubName;
8840 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8841 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8842
8843 stubAddr = StubRoutines::dilithiumDecomposePoly();
8844 stubName = "dilithiumDecomposePoly";
8845 if (!stubAddr) return false;
8846
8847 Node* input = argument(0);
8848 Node* lowPart = argument(1);
8849 Node* highPart = argument(2);
8850 Node* twoGamma2 = argument(3);
8851 Node* multiplier = argument(4);
8852
8853 input = must_be_not_null(input, true);
8854 lowPart = must_be_not_null(lowPart, true);
8855 highPart = must_be_not_null(highPart, true);
8856
8857 Node* input_start = array_element_address(input, intcon(0), T_INT);
8858 assert(input_start, "input is null");
8859 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8860 assert(lowPart_start, "lowPart is null");
8861 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8862 assert(highPart_start, "highPart is null");
8863
8864 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8865 OptoRuntime::dilithiumDecomposePoly_Type(),
8866 stubAddr, stubName, TypePtr::BOTTOM,
8867 input_start, lowPart_start, highPart_start,
8868 twoGamma2, multiplier);
8869
8870 // return an int
8871 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8872 set_result(retvalue);
8873 return true;
8874 }
8875
8876 bool LibraryCallKit::inline_base64_encodeBlock() {
8877 address stubAddr;
8878 const char *stubName;
8879 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8880 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8881 stubAddr = StubRoutines::base64_encodeBlock();
8882 stubName = "encodeBlock";
8883
8884 if (!stubAddr) return false;
8885 Node* base64obj = argument(0);
8886 Node* src = argument(1);
8887 Node* offset = argument(2);
8888 Node* len = argument(3);
8889 Node* dest = argument(4);
8890 Node* dp = argument(5);
8891 Node* isURL = argument(6);
8892
8893 src = must_be_not_null(src, true);
8894 dest = must_be_not_null(dest, true);
8895
8896 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8897 assert(src_start, "source array is null");
8898 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8899 assert(dest_start, "destination array is null");
8900
8901 Node* base64 = make_runtime_call(RC_LEAF,
8902 OptoRuntime::base64_encodeBlock_Type(),
8903 stubAddr, stubName, TypePtr::BOTTOM,
8904 src_start, offset, len, dest_start, dp, isURL);
8905 return true;
8906 }
8907
8908 bool LibraryCallKit::inline_base64_decodeBlock() {
8909 address stubAddr;
8910 const char *stubName;
8911 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8912 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8913 stubAddr = StubRoutines::base64_decodeBlock();
8914 stubName = "decodeBlock";
8915
8916 if (!stubAddr) return false;
8917 Node* base64obj = argument(0);
8918 Node* src = argument(1);
8919 Node* src_offset = argument(2);
8920 Node* len = argument(3);
8921 Node* dest = argument(4);
8922 Node* dest_offset = argument(5);
8923 Node* isURL = argument(6);
8924 Node* isMIME = argument(7);
8925
8926 src = must_be_not_null(src, true);
8927 dest = must_be_not_null(dest, true);
8928
8929 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8930 assert(src_start, "source array is null");
8931 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8932 assert(dest_start, "destination array is null");
8933
8934 Node* call = make_runtime_call(RC_LEAF,
8935 OptoRuntime::base64_decodeBlock_Type(),
8936 stubAddr, stubName, TypePtr::BOTTOM,
8937 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8938 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8939 set_result(result);
8940 return true;
8941 }
8942
8943 bool LibraryCallKit::inline_poly1305_processBlocks() {
8944 address stubAddr;
8945 const char *stubName;
8946 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8947 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8948 stubAddr = StubRoutines::poly1305_processBlocks();
8949 stubName = "poly1305_processBlocks";
8950
8951 if (!stubAddr) return false;
8952 null_check_receiver(); // null-check receiver
8953 if (stopped()) return true;
8954
8955 Node* input = argument(1);
8956 Node* input_offset = argument(2);
8957 Node* len = argument(3);
8958 Node* alimbs = argument(4);
8959 Node* rlimbs = argument(5);
8960
8961 input = must_be_not_null(input, true);
8962 alimbs = must_be_not_null(alimbs, true);
8963 rlimbs = must_be_not_null(rlimbs, true);
8964
8965 Node* input_start = array_element_address(input, input_offset, T_BYTE);
8966 assert(input_start, "input array is null");
8967 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
8968 assert(acc_start, "acc array is null");
8969 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
8970 assert(r_start, "r array is null");
8971
8972 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8973 OptoRuntime::poly1305_processBlocks_Type(),
8974 stubAddr, stubName, TypePtr::BOTTOM,
8975 input_start, len, acc_start, r_start);
8976 return true;
8977 }
8978
8979 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
8980 address stubAddr;
8981 const char *stubName;
8982 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8983 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
8984 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
8985 stubName = "intpoly_montgomeryMult_P256";
8986
8987 if (!stubAddr) return false;
8988 null_check_receiver(); // null-check receiver
8989 if (stopped()) return true;
8990
8991 Node* a = argument(1);
8992 Node* b = argument(2);
8993 Node* r = argument(3);
8994
8995 a = must_be_not_null(a, true);
8996 b = must_be_not_null(b, true);
8997 r = must_be_not_null(r, true);
8998
8999 Node* a_start = array_element_address(a, intcon(0), T_LONG);
9000 assert(a_start, "a array is null");
9001 Node* b_start = array_element_address(b, intcon(0), T_LONG);
9002 assert(b_start, "b array is null");
9003 Node* r_start = array_element_address(r, intcon(0), T_LONG);
9004 assert(r_start, "r array is null");
9005
9006 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
9007 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
9008 stubAddr, stubName, TypePtr::BOTTOM,
9009 a_start, b_start, r_start);
9010 return true;
9011 }
9012
9013 bool LibraryCallKit::inline_intpoly_assign() {
9014 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
9015 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
9016 const char *stubName = "intpoly_assign";
9017 address stubAddr = StubRoutines::intpoly_assign();
9018 if (!stubAddr) return false;
9019
9020 Node* set = argument(0);
9021 Node* a = argument(1);
9022 Node* b = argument(2);
9023 Node* arr_length = load_array_length(a);
9024
9025 a = must_be_not_null(a, true);
9026 b = must_be_not_null(b, true);
9027
9028 Node* a_start = array_element_address(a, intcon(0), T_LONG);
9029 assert(a_start, "a array is null");
9030 Node* b_start = array_element_address(b, intcon(0), T_LONG);
9031 assert(b_start, "b array is null");
9032
9033 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
9034 OptoRuntime::intpoly_assign_Type(),
9035 stubAddr, stubName, TypePtr::BOTTOM,
9036 set, a_start, b_start, arr_length);
9037 return true;
9038 }
9039
9040 //------------------------------inline_digestBase_implCompress-----------------------
9041 //
9042 // Calculate MD5 for single-block byte[] array.
9043 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
9044 //
9045 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
9046 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
9047 //
9048 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
9049 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
9050 //
9051 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
9052 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
9053 //
9054 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
9055 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
9056 //
9057 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
9058 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
9059
9060 Node* digestBase_obj = argument(0);
9061 Node* src = argument(1); // type oop
9062 Node* ofs = argument(2); // type int
9063
9064 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
9065 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
9066 // failed array check
9067 return false;
9068 }
9069 // Figure out the size and type of the elements we will be copying.
9070 BasicType src_elem = src_type->elem()->array_element_basic_type();
9071 if (src_elem != T_BYTE) {
9072 return false;
9073 }
9074 // 'src_start' points to src array + offset
9075 src = must_be_not_null(src, true);
9076 Node* src_start = array_element_address(src, ofs, src_elem);
9077 Node* state = nullptr;
9078 Node* block_size = nullptr;
9079 address stubAddr;
9080 const char *stubName;
9081
9082 switch(id) {
9083 case vmIntrinsics::_md5_implCompress:
9084 assert(UseMD5Intrinsics, "need MD5 instruction support");
9085 state = get_state_from_digest_object(digestBase_obj, T_INT);
9086 stubAddr = StubRoutines::md5_implCompress();
9087 stubName = "md5_implCompress";
9088 break;
9089 case vmIntrinsics::_sha_implCompress:
9090 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
9091 state = get_state_from_digest_object(digestBase_obj, T_INT);
9092 stubAddr = StubRoutines::sha1_implCompress();
9093 stubName = "sha1_implCompress";
9094 break;
9095 case vmIntrinsics::_sha2_implCompress:
9096 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
9097 state = get_state_from_digest_object(digestBase_obj, T_INT);
9098 stubAddr = StubRoutines::sha256_implCompress();
9099 stubName = "sha256_implCompress";
9100 break;
9101 case vmIntrinsics::_sha5_implCompress:
9102 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
9103 state = get_state_from_digest_object(digestBase_obj, T_LONG);
9104 stubAddr = StubRoutines::sha512_implCompress();
9105 stubName = "sha512_implCompress";
9106 break;
9107 case vmIntrinsics::_sha3_implCompress:
9108 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
9109 state = get_state_from_digest_object(digestBase_obj, T_LONG);
9110 stubAddr = StubRoutines::sha3_implCompress();
9111 stubName = "sha3_implCompress";
9112 block_size = get_block_size_from_digest_object(digestBase_obj);
9113 if (block_size == nullptr) return false;
9114 break;
9115 default:
9116 fatal_unexpected_iid(id);
9117 return false;
9118 }
9119 if (state == nullptr) return false;
9120
9121 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
9122 if (stubAddr == nullptr) return false;
9123
9124 // Call the stub.
9125 Node* call;
9126 if (block_size == nullptr) {
9127 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
9128 stubAddr, stubName, TypePtr::BOTTOM,
9129 src_start, state);
9130 } else {
9131 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
9132 stubAddr, stubName, TypePtr::BOTTOM,
9133 src_start, state, block_size);
9134 }
9135
9136 return true;
9137 }
9138
9139 //------------------------------inline_double_keccak
9140 bool LibraryCallKit::inline_double_keccak() {
9141 address stubAddr;
9142 const char *stubName;
9143 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
9144 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
9145
9146 stubAddr = StubRoutines::double_keccak();
9147 stubName = "double_keccak";
9148 if (!stubAddr) return false;
9149
9150 Node* status0 = argument(0);
9151 Node* status1 = argument(1);
9152
9153 status0 = must_be_not_null(status0, true);
9154 status1 = must_be_not_null(status1, true);
9155
9156 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
9157 assert(status0_start, "status0 is null");
9158 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
9159 assert(status1_start, "status1 is null");
9160 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
9161 OptoRuntime::double_keccak_Type(),
9162 stubAddr, stubName, TypePtr::BOTTOM,
9163 status0_start, status1_start);
9164 // return an int
9165 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
9166 set_result(retvalue);
9167 return true;
9168 }
9169
9170
9171 //------------------------------inline_digestBase_implCompressMB-----------------------
9172 //
9173 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
9174 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
9175 //
9176 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
9177 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9178 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9179 assert((uint)predicate < 5, "sanity");
9180 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
9181
9182 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
9183 Node* src = argument(1); // byte[] array
9184 Node* ofs = argument(2); // type int
9185 Node* limit = argument(3); // type int
9186
9187 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
9188 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
9189 // failed array check
9190 return false;
9191 }
9192 // Figure out the size and type of the elements we will be copying.
9193 BasicType src_elem = src_type->elem()->array_element_basic_type();
9194 if (src_elem != T_BYTE) {
9195 return false;
9196 }
9197 // 'src_start' points to src array + offset
9198 src = must_be_not_null(src, false);
9199 Node* src_start = array_element_address(src, ofs, src_elem);
9200
9201 const char* klass_digestBase_name = nullptr;
9202 const char* stub_name = nullptr;
9203 address stub_addr = nullptr;
9204 BasicType elem_type = T_INT;
9205
9206 switch (predicate) {
9207 case 0:
9208 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
9209 klass_digestBase_name = "sun/security/provider/MD5";
9210 stub_name = "md5_implCompressMB";
9211 stub_addr = StubRoutines::md5_implCompressMB();
9212 }
9213 break;
9214 case 1:
9215 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
9216 klass_digestBase_name = "sun/security/provider/SHA";
9217 stub_name = "sha1_implCompressMB";
9218 stub_addr = StubRoutines::sha1_implCompressMB();
9219 }
9220 break;
9221 case 2:
9222 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
9223 klass_digestBase_name = "sun/security/provider/SHA2";
9224 stub_name = "sha256_implCompressMB";
9225 stub_addr = StubRoutines::sha256_implCompressMB();
9226 }
9227 break;
9228 case 3:
9229 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
9230 klass_digestBase_name = "sun/security/provider/SHA5";
9231 stub_name = "sha512_implCompressMB";
9232 stub_addr = StubRoutines::sha512_implCompressMB();
9233 elem_type = T_LONG;
9234 }
9235 break;
9236 case 4:
9237 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
9238 klass_digestBase_name = "sun/security/provider/SHA3";
9239 stub_name = "sha3_implCompressMB";
9240 stub_addr = StubRoutines::sha3_implCompressMB();
9241 elem_type = T_LONG;
9242 }
9243 break;
9244 default:
9245 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
9246 }
9247 if (klass_digestBase_name != nullptr) {
9248 assert(stub_addr != nullptr, "Stub is generated");
9249 if (stub_addr == nullptr) return false;
9250
9251 // get DigestBase klass to lookup for SHA klass
9252 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
9253 assert(tinst != nullptr, "digestBase_obj is not instance???");
9254 assert(tinst->is_loaded(), "DigestBase is not loaded");
9255
9256 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
9257 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
9258 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
9259 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
9260 }
9261 return false;
9262 }
9263
9264 //------------------------------inline_digestBase_implCompressMB-----------------------
9265 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
9266 BasicType elem_type, address stubAddr, const char *stubName,
9267 Node* src_start, Node* ofs, Node* limit) {
9268 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
9269 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
9270 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
9271 digest_obj = _gvn.transform(digest_obj);
9272
9273 Node* state = get_state_from_digest_object(digest_obj, elem_type);
9274 if (state == nullptr) return false;
9275
9276 Node* block_size = nullptr;
9277 if (strcmp("sha3_implCompressMB", stubName) == 0) {
9278 block_size = get_block_size_from_digest_object(digest_obj);
9279 if (block_size == nullptr) return false;
9280 }
9281
9282 // Call the stub.
9283 Node* call;
9284 if (block_size == nullptr) {
9285 call = make_runtime_call(RC_LEAF|RC_NO_FP,
9286 OptoRuntime::digestBase_implCompressMB_Type(false),
9287 stubAddr, stubName, TypePtr::BOTTOM,
9288 src_start, state, ofs, limit);
9289 } else {
9290 call = make_runtime_call(RC_LEAF|RC_NO_FP,
9291 OptoRuntime::digestBase_implCompressMB_Type(true),
9292 stubAddr, stubName, TypePtr::BOTTOM,
9293 src_start, state, block_size, ofs, limit);
9294 }
9295
9296 // return ofs (int)
9297 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
9298 set_result(result);
9299
9300 return true;
9301 }
9302
9303 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
9304 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
9305 assert(UseAES, "need AES instruction support");
9306 address stubAddr = nullptr;
9307 const char *stubName = nullptr;
9308 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
9309 stubName = "galoisCounterMode_AESCrypt";
9310
9311 if (stubAddr == nullptr) return false;
9312
9313 Node* in = argument(0);
9314 Node* inOfs = argument(1);
9315 Node* len = argument(2);
9316 Node* ct = argument(3);
9317 Node* ctOfs = argument(4);
9318 Node* out = argument(5);
9319 Node* outOfs = argument(6);
9320 Node* gctr_object = argument(7);
9321 Node* ghash_object = argument(8);
9322
9323 // (1) in, ct and out are arrays.
9324 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
9325 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
9326 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
9327 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
9328 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
9329 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
9330
9331 // checks are the responsibility of the caller
9332 Node* in_start = in;
9333 Node* ct_start = ct;
9334 Node* out_start = out;
9335 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
9336 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
9337 in_start = array_element_address(in, inOfs, T_BYTE);
9338 ct_start = array_element_address(ct, ctOfs, T_BYTE);
9339 out_start = array_element_address(out, outOfs, T_BYTE);
9340 }
9341
9342 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
9343 // (because of the predicated logic executed earlier).
9344 // so we cast it here safely.
9345 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
9346 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9347 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
9348 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
9349 Node* state = load_field_from_object(ghash_object, "state", "[J");
9350
9351 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
9352 return false;
9353 }
9354 // cast it to what we know it will be at runtime
9355 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
9356 assert(tinst != nullptr, "GCTR obj is null");
9357 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9358 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
9359 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
9360 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9361 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
9362 const TypeOopPtr* xtype = aklass->as_instance_type();
9363 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
9364 aescrypt_object = _gvn.transform(aescrypt_object);
9365 // we need to get the start of the aescrypt_object's expanded key array
9366 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
9367 if (k_start == nullptr) return false;
9368 // similarly, get the start address of the r vector
9369 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
9370 Node* state_start = array_element_address(state, intcon(0), T_LONG);
9371 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
9372
9373
9374 // Call the stub, passing params
9375 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
9376 OptoRuntime::galoisCounterMode_aescrypt_Type(),
9377 stubAddr, stubName, TypePtr::BOTTOM,
9378 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
9379
9380 // return cipher length (int)
9381 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
9382 set_result(retvalue);
9383
9384 return true;
9385 }
9386
9387 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
9388 // Return node representing slow path of predicate check.
9389 // the pseudo code we want to emulate with this predicate is:
9390 // for encryption:
9391 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
9392 // for decryption:
9393 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
9394 // note cipher==plain is more conservative than the original java code but that's OK
9395 //
9396
9397 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
9398 // The receiver was checked for null already.
9399 Node* objGCTR = argument(7);
9400 // Load embeddedCipher field of GCTR object.
9401 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9402 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
9403
9404 // get AESCrypt klass for instanceOf check
9405 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
9406 // will have same classloader as CipherBlockChaining object
9407 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
9408 assert(tinst != nullptr, "GCTR obj is null");
9409 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9410
9411 // we want to do an instanceof comparison against the AESCrypt class
9412 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
9413 if (!klass_AESCrypt->is_loaded()) {
9414 // if AESCrypt is not even loaded, we never take the intrinsic fast path
9415 Node* ctrl = control();
9416 set_control(top()); // no regular fast path
9417 return ctrl;
9418 }
9419
9420 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9421 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
9422 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9423 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9424 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9425
9426 return instof_false; // even if it is null
9427 }
9428
9429 //------------------------------get_state_from_digest_object-----------------------
9430 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
9431 const char* state_type;
9432 switch (elem_type) {
9433 case T_BYTE: state_type = "[B"; break;
9434 case T_INT: state_type = "[I"; break;
9435 case T_LONG: state_type = "[J"; break;
9436 default: ShouldNotReachHere();
9437 }
9438 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
9439 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
9440 if (digest_state == nullptr) return (Node *) nullptr;
9441
9442 // now have the array, need to get the start address of the state array
9443 Node* state = array_element_address(digest_state, intcon(0), elem_type);
9444 return state;
9445 }
9446
9447 //------------------------------get_block_size_from_sha3_object----------------------------------
9448 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
9449 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
9450 assert (block_size != nullptr, "sanity");
9451 return block_size;
9452 }
9453
9454 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
9455 // Return node representing slow path of predicate check.
9456 // the pseudo code we want to emulate with this predicate is:
9457 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
9458 //
9459 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
9460 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9461 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9462 assert((uint)predicate < 5, "sanity");
9463
9464 // The receiver was checked for null already.
9465 Node* digestBaseObj = argument(0);
9466
9467 // get DigestBase klass for instanceOf check
9468 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
9469 assert(tinst != nullptr, "digestBaseObj is null");
9470 assert(tinst->is_loaded(), "DigestBase is not loaded");
9471
9472 const char* klass_name = nullptr;
9473 switch (predicate) {
9474 case 0:
9475 if (UseMD5Intrinsics) {
9476 // we want to do an instanceof comparison against the MD5 class
9477 klass_name = "sun/security/provider/MD5";
9478 }
9479 break;
9480 case 1:
9481 if (UseSHA1Intrinsics) {
9482 // we want to do an instanceof comparison against the SHA class
9483 klass_name = "sun/security/provider/SHA";
9484 }
9485 break;
9486 case 2:
9487 if (UseSHA256Intrinsics) {
9488 // we want to do an instanceof comparison against the SHA2 class
9489 klass_name = "sun/security/provider/SHA2";
9490 }
9491 break;
9492 case 3:
9493 if (UseSHA512Intrinsics) {
9494 // we want to do an instanceof comparison against the SHA5 class
9495 klass_name = "sun/security/provider/SHA5";
9496 }
9497 break;
9498 case 4:
9499 if (UseSHA3Intrinsics) {
9500 // we want to do an instanceof comparison against the SHA3 class
9501 klass_name = "sun/security/provider/SHA3";
9502 }
9503 break;
9504 default:
9505 fatal("unknown SHA intrinsic predicate: %d", predicate);
9506 }
9507
9508 ciKlass* klass = nullptr;
9509 if (klass_name != nullptr) {
9510 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
9511 }
9512 if ((klass == nullptr) || !klass->is_loaded()) {
9513 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
9514 Node* ctrl = control();
9515 set_control(top()); // no intrinsic path
9516 return ctrl;
9517 }
9518 ciInstanceKlass* instklass = klass->as_instance_klass();
9519
9520 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
9521 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9522 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9523 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9524
9525 return instof_false; // even if it is null
9526 }
9527
9528 //-------------inline_fma-----------------------------------
9529 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
9530 Node *a = nullptr;
9531 Node *b = nullptr;
9532 Node *c = nullptr;
9533 Node* result = nullptr;
9534 switch (id) {
9535 case vmIntrinsics::_fmaD:
9536 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
9537 // no receiver since it is static method
9538 a = argument(0);
9539 b = argument(2);
9540 c = argument(4);
9541 result = _gvn.transform(new FmaDNode(a, b, c));
9542 break;
9543 case vmIntrinsics::_fmaF:
9544 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
9545 a = argument(0);
9546 b = argument(1);
9547 c = argument(2);
9548 result = _gvn.transform(new FmaFNode(a, b, c));
9549 break;
9550 default:
9551 fatal_unexpected_iid(id); break;
9552 }
9553 set_result(result);
9554 return true;
9555 }
9556
9557 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
9558 // argument(0) is receiver
9559 Node* codePoint = argument(1);
9560 Node* n = nullptr;
9561
9562 switch (id) {
9563 case vmIntrinsics::_isDigit :
9564 n = new DigitNode(control(), codePoint);
9565 break;
9566 case vmIntrinsics::_isLowerCase :
9567 n = new LowerCaseNode(control(), codePoint);
9568 break;
9569 case vmIntrinsics::_isUpperCase :
9570 n = new UpperCaseNode(control(), codePoint);
9571 break;
9572 case vmIntrinsics::_isWhitespace :
9573 n = new WhitespaceNode(control(), codePoint);
9574 break;
9575 default:
9576 fatal_unexpected_iid(id);
9577 }
9578
9579 set_result(_gvn.transform(n));
9580 return true;
9581 }
9582
9583 bool LibraryCallKit::inline_profileBoolean() {
9584 Node* counts = argument(1);
9585 const TypeAryPtr* ary = nullptr;
9586 ciArray* aobj = nullptr;
9587 if (counts->is_Con()
9588 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
9589 && (aobj = ary->const_oop()->as_array()) != nullptr
9590 && (aobj->length() == 2)) {
9591 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
9592 jint false_cnt = aobj->element_value(0).as_int();
9593 jint true_cnt = aobj->element_value(1).as_int();
9594
9595 if (C->log() != nullptr) {
9596 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
9597 false_cnt, true_cnt);
9598 }
9599
9600 if (false_cnt + true_cnt == 0) {
9601 // According to profile, never executed.
9602 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9603 Deoptimization::Action_reinterpret);
9604 return true;
9605 }
9606
9607 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
9608 // is a number of each value occurrences.
9609 Node* result = argument(0);
9610 if (false_cnt == 0 || true_cnt == 0) {
9611 // According to profile, one value has been never seen.
9612 int expected_val = (false_cnt == 0) ? 1 : 0;
9613
9614 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
9615 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
9616
9617 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
9618 Node* fast_path = _gvn.transform(new IfTrueNode(check));
9619 Node* slow_path = _gvn.transform(new IfFalseNode(check));
9620
9621 { // Slow path: uncommon trap for never seen value and then reexecute
9622 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
9623 // the value has been seen at least once.
9624 PreserveJVMState pjvms(this);
9625 PreserveReexecuteState preexecs(this);
9626 jvms()->set_should_reexecute(true);
9627
9628 set_control(slow_path);
9629 set_i_o(i_o());
9630
9631 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9632 Deoptimization::Action_reinterpret);
9633 }
9634 // The guard for never seen value enables sharpening of the result and
9635 // returning a constant. It allows to eliminate branches on the same value
9636 // later on.
9637 set_control(fast_path);
9638 result = intcon(expected_val);
9639 }
9640 // Stop profiling.
9641 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
9642 // By replacing method body with profile data (represented as ProfileBooleanNode
9643 // on IR level) we effectively disable profiling.
9644 // It enables full speed execution once optimized code is generated.
9645 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
9646 C->record_for_igvn(profile);
9647 set_result(profile);
9648 return true;
9649 } else {
9650 // Continue profiling.
9651 // Profile data isn't available at the moment. So, execute method's bytecode version.
9652 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
9653 // is compiled and counters aren't available since corresponding MethodHandle
9654 // isn't a compile-time constant.
9655 return false;
9656 }
9657 }
9658
9659 bool LibraryCallKit::inline_isCompileConstant() {
9660 Node* n = argument(0);
9661 set_result(n->is_Con() ? intcon(1) : intcon(0));
9662 return true;
9663 }
9664
9665 //------------------------------- inline_getObjectSize --------------------------------------
9666 //
9667 // Calculate the runtime size of the object/array.
9668 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
9669 //
9670 bool LibraryCallKit::inline_getObjectSize() {
9671 Node* obj = argument(3);
9672 Node* klass_node = load_object_klass(obj);
9673
9674 jint layout_con = Klass::_lh_neutral_value;
9675 Node* layout_val = get_layout_helper(klass_node, layout_con);
9676 int layout_is_con = (layout_val == nullptr);
9677
9678 if (layout_is_con) {
9679 // Layout helper is constant, can figure out things at compile time.
9680
9681 if (Klass::layout_helper_is_instance(layout_con)) {
9682 // Instance case: layout_con contains the size itself.
9683 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
9684 set_result(size);
9685 } else {
9686 // Array case: size is round(header + element_size*arraylength).
9687 // Since arraylength is different for every array instance, we have to
9688 // compute the whole thing at runtime.
9689
9690 Node* arr_length = load_array_length(obj);
9691
9692 int round_mask = MinObjAlignmentInBytes - 1;
9693 int hsize = Klass::layout_helper_header_size(layout_con);
9694 int eshift = Klass::layout_helper_log2_element_size(layout_con);
9695
9696 if ((round_mask & ~right_n_bits(eshift)) == 0) {
9697 round_mask = 0; // strength-reduce it if it goes away completely
9698 }
9699 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
9700 Node* header_size = intcon(hsize + round_mask);
9701
9702 Node* lengthx = ConvI2X(arr_length);
9703 Node* headerx = ConvI2X(header_size);
9704
9705 Node* abody = lengthx;
9706 if (eshift != 0) {
9707 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
9708 }
9709 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
9710 if (round_mask != 0) {
9711 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
9712 }
9713 size = ConvX2L(size);
9714 set_result(size);
9715 }
9716 } else {
9717 // Layout helper is not constant, need to test for array-ness at runtime.
9718
9719 enum { _instance_path = 1, _array_path, PATH_LIMIT };
9720 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
9721 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
9722 record_for_igvn(result_reg);
9723
9724 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
9725 if (array_ctl != nullptr) {
9726 // Array case: size is round(header + element_size*arraylength).
9727 // Since arraylength is different for every array instance, we have to
9728 // compute the whole thing at runtime.
9729
9730 PreserveJVMState pjvms(this);
9731 set_control(array_ctl);
9732 Node* arr_length = load_array_length(obj);
9733
9734 int round_mask = MinObjAlignmentInBytes - 1;
9735 Node* mask = intcon(round_mask);
9736
9737 Node* hss = intcon(Klass::_lh_header_size_shift);
9738 Node* hsm = intcon(Klass::_lh_header_size_mask);
9739 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
9740 header_size = _gvn.transform(new AndINode(header_size, hsm));
9741 header_size = _gvn.transform(new AddINode(header_size, mask));
9742
9743 // There is no need to mask or shift this value.
9744 // The semantics of LShiftINode include an implicit mask to 0x1F.
9745 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
9746 Node* elem_shift = layout_val;
9747
9748 Node* lengthx = ConvI2X(arr_length);
9749 Node* headerx = ConvI2X(header_size);
9750
9751 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
9752 Node* size = _gvn.transform(new AddXNode(headerx, abody));
9753 if (round_mask != 0) {
9754 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
9755 }
9756 size = ConvX2L(size);
9757
9758 result_reg->init_req(_array_path, control());
9759 result_val->init_req(_array_path, size);
9760 }
9761
9762 if (!stopped()) {
9763 // Instance case: the layout helper gives us instance size almost directly,
9764 // but we need to mask out the _lh_instance_slow_path_bit.
9765 Node* size = ConvI2X(layout_val);
9766 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
9767 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
9768 size = _gvn.transform(new AndXNode(size, mask));
9769 size = ConvX2L(size);
9770
9771 result_reg->init_req(_instance_path, control());
9772 result_val->init_req(_instance_path, size);
9773 }
9774
9775 set_result(result_reg, result_val);
9776 }
9777
9778 return true;
9779 }
9780
9781 //------------------------------- inline_blackhole --------------------------------------
9782 //
9783 // Make sure all arguments to this node are alive.
9784 // This matches methods that were requested to be blackholed through compile commands.
9785 //
9786 bool LibraryCallKit::inline_blackhole() {
9787 assert(callee()->is_static(), "Should have been checked before: only static methods here");
9788 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
9789 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
9790
9791 // Blackhole node pinches only the control, not memory. This allows
9792 // the blackhole to be pinned in the loop that computes blackholed
9793 // values, but have no other side effects, like breaking the optimizations
9794 // across the blackhole.
9795
9796 Node* bh = _gvn.transform(new BlackholeNode(control()));
9797 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
9798
9799 // Bind call arguments as blackhole arguments to keep them alive
9800 uint nargs = callee()->arg_size();
9801 for (uint i = 0; i < nargs; i++) {
9802 bh->add_req(argument(i));
9803 }
9804
9805 return true;
9806 }
9807
9808 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9809 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9810 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9811 return nullptr; // box klass is not Float16
9812 }
9813
9814 // Null check; get notnull casted pointer
9815 Node* null_ctl = top();
9816 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9817 // If not_null_box is dead, only null-path is taken
9818 if (stopped()) {
9819 set_control(null_ctl);
9820 return nullptr;
9821 }
9822 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9823 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9824 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9825 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9826 }
9827
9828 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9829 PreserveReexecuteState preexecs(this);
9830 jvms()->set_should_reexecute(true);
9831
9832 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9833 Node* klass_node = makecon(klass_type);
9834 Node* box = new_instance(klass_node);
9835
9836 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9837 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9838
9839 Node* field_store = _gvn.transform(access_store_at(box,
9840 value_field,
9841 value_adr_type,
9842 value,
9843 TypeInt::SHORT,
9844 T_SHORT,
9845 IN_HEAP));
9846 set_memory(field_store, value_adr_type);
9847 return box;
9848 }
9849
9850 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9851 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9852 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9853 return false;
9854 }
9855
9856 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9857 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9858 return false;
9859 }
9860
9861 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9862 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9863 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9864 ciSymbols::short_signature(),
9865 false);
9866 assert(field != nullptr, "");
9867
9868 // Transformed nodes
9869 Node* fld1 = nullptr;
9870 Node* fld2 = nullptr;
9871 Node* fld3 = nullptr;
9872 switch(num_args) {
9873 case 3:
9874 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9875 if (fld3 == nullptr) {
9876 return false;
9877 }
9878 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9879 // fall-through
9880 case 2:
9881 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9882 if (fld2 == nullptr) {
9883 return false;
9884 }
9885 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9886 // fall-through
9887 case 1:
9888 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9889 if (fld1 == nullptr) {
9890 return false;
9891 }
9892 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9893 break;
9894 default: fatal("Unsupported number of arguments %d", num_args);
9895 }
9896
9897 Node* result = nullptr;
9898 switch (id) {
9899 // Unary operations
9900 case vmIntrinsics::_sqrt_float16:
9901 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9902 break;
9903 // Ternary operations
9904 case vmIntrinsics::_fma_float16:
9905 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9906 break;
9907 default:
9908 fatal_unexpected_iid(id);
9909 break;
9910 }
9911 result = _gvn.transform(new ReinterpretHF2SNode(result));
9912 set_result(box_fp16_value(float16_box_type, field, result));
9913 return true;
9914 }
9915