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) {
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 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
928 if (is_neg != nullptr && pos_index != nullptr) {
929 // Emulate effect of Parse::adjust_map_after_if.
930 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
931 (*pos_index) = _gvn.transform(ccast);
932 }
933 return is_neg;
934 }
935
936 // Make sure that 'position' is a valid limit index, in [0..length].
937 // There are two equivalent plans for checking this:
938 // A. (offset + copyLength) unsigned<= arrayLength
939 // B. offset <= (arrayLength - copyLength)
940 // We require that all of the values above, except for the sum and
941 // difference, are already known to be non-negative.
942 // Plan A is robust in the face of overflow, if offset and copyLength
943 // are both hugely positive.
944 //
945 // Plan B is less direct and intuitive, but it does not overflow at
946 // all, since the difference of two non-negatives is always
947 // representable. Whenever Java methods must perform the equivalent
948 // check they generally use Plan B instead of Plan A.
949 // For the moment we use Plan A.
950 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
951 Node* subseq_length,
952 Node* array_length,
953 RegionNode* region) {
954 if (stopped())
955 return nullptr; // already stopped
956 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
957 if (zero_offset && subseq_length->eqv_uncast(array_length))
958 return nullptr; // common case of whole-array copy
959 Node* last = subseq_length;
960 if (!zero_offset) // last += offset
961 last = _gvn.transform(new AddINode(last, offset));
962 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
963 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
964 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
965 return is_over;
966 }
967
968 // Emit range checks for the given String.value byte array
969 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
970 if (stopped()) {
971 return; // already stopped
972 }
973 RegionNode* bailout = new RegionNode(1);
974 record_for_igvn(bailout);
975 if (char_count) {
976 // Convert char count to byte count
977 count = _gvn.transform(new LShiftINode(count, intcon(1)));
978 }
979
980 // Offset and count must not be negative
981 generate_negative_guard(offset, bailout);
982 generate_negative_guard(count, bailout);
983 // Offset + count must not exceed length of array
984 generate_limit_guard(offset, count, load_array_length(array), bailout);
985
986 if (bailout->req() > 1) {
987 PreserveJVMState pjvms(this);
988 set_control(_gvn.transform(bailout));
989 uncommon_trap(Deoptimization::Reason_intrinsic,
990 Deoptimization::Action_maybe_recompile);
991 }
992 }
993
994 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
995 bool is_immutable) {
996 ciKlass* thread_klass = env()->Thread_klass();
997 const Type* thread_type
998 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
999
1000 Node* thread = _gvn.transform(new ThreadLocalNode());
1001 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
1002 tls_output = thread;
1003
1004 Node* thread_obj_handle
1005 = (is_immutable
1006 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
1007 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
1008 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
1009 thread_obj_handle = _gvn.transform(thread_obj_handle);
1010
1011 DecoratorSet decorators = IN_NATIVE;
1012 if (is_immutable) {
1013 decorators |= C2_IMMUTABLE_MEMORY;
1014 }
1015 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
1016 }
1017
1018 //--------------------------generate_current_thread--------------------
1019 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1020 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
1021 /*is_immutable*/false);
1022 }
1023
1024 //--------------------------generate_virtual_thread--------------------
1025 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
1026 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
1027 !C->method()->changes_current_thread());
1028 }
1029
1030 //------------------------------make_string_method_node------------------------
1031 // Helper method for String intrinsic functions. This version is called with
1032 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1033 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1034 // containing the lengths of str1 and str2.
1035 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1036 Node* result = nullptr;
1037 switch (opcode) {
1038 case Op_StrIndexOf:
1039 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1040 str1_start, cnt1, str2_start, cnt2, ae);
1041 break;
1042 case Op_StrComp:
1043 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1044 str1_start, cnt1, str2_start, cnt2, ae);
1045 break;
1046 case Op_StrEquals:
1047 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1048 // Use the constant length if there is one because optimized match rule may exist.
1049 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1050 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1051 break;
1052 default:
1053 ShouldNotReachHere();
1054 return nullptr;
1055 }
1056
1057 // All these intrinsics have checks.
1058 C->set_has_split_ifs(true); // Has chance for split-if optimization
1059 clear_upper_avx();
1060
1061 return _gvn.transform(result);
1062 }
1063
1064 //------------------------------inline_string_compareTo------------------------
1065 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1066 Node* arg1 = argument(0);
1067 Node* arg2 = argument(1);
1068
1069 arg1 = must_be_not_null(arg1, true);
1070 arg2 = must_be_not_null(arg2, true);
1071
1072 // Get start addr and length of first argument
1073 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1074 Node* arg1_cnt = load_array_length(arg1);
1075
1076 // Get start addr and length of second argument
1077 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1078 Node* arg2_cnt = load_array_length(arg2);
1079
1080 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1081 set_result(result);
1082 return true;
1083 }
1084
1085 //------------------------------inline_string_equals------------------------
1086 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1087 Node* arg1 = argument(0);
1088 Node* arg2 = argument(1);
1089
1090 // paths (plus control) merge
1091 RegionNode* region = new RegionNode(3);
1092 Node* phi = new PhiNode(region, TypeInt::BOOL);
1093
1094 if (!stopped()) {
1095
1096 arg1 = must_be_not_null(arg1, true);
1097 arg2 = must_be_not_null(arg2, true);
1098
1099 // Get start addr and length of first argument
1100 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1101 Node* arg1_cnt = load_array_length(arg1);
1102
1103 // Get start addr and length of second argument
1104 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1105 Node* arg2_cnt = load_array_length(arg2);
1106
1107 // Check for arg1_cnt != arg2_cnt
1108 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1109 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1110 Node* if_ne = generate_slow_guard(bol, nullptr);
1111 if (if_ne != nullptr) {
1112 phi->init_req(2, intcon(0));
1113 region->init_req(2, if_ne);
1114 }
1115
1116 // Check for count == 0 is done by assembler code for StrEquals.
1117
1118 if (!stopped()) {
1119 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1120 phi->init_req(1, equals);
1121 region->init_req(1, control());
1122 }
1123 }
1124
1125 // post merge
1126 set_control(_gvn.transform(region));
1127 record_for_igvn(region);
1128
1129 set_result(_gvn.transform(phi));
1130 return true;
1131 }
1132
1133 //------------------------------inline_array_equals----------------------------
1134 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1135 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1136 Node* arg1 = argument(0);
1137 Node* arg2 = argument(1);
1138
1139 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1140 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1141 clear_upper_avx();
1142
1143 return true;
1144 }
1145
1146
1147 //------------------------------inline_countPositives------------------------------
1148 bool LibraryCallKit::inline_countPositives() {
1149 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1150 return false;
1151 }
1152
1153 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1154 // no receiver since it is static method
1155 Node* ba = argument(0);
1156 Node* offset = argument(1);
1157 Node* len = argument(2);
1158
1159 ba = must_be_not_null(ba, true);
1160
1161 // Range checks
1162 generate_string_range_check(ba, offset, len, false);
1163 if (stopped()) {
1164 return true;
1165 }
1166 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1167 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1168 set_result(_gvn.transform(result));
1169 clear_upper_avx();
1170 return true;
1171 }
1172
1173 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1174 Node* index = argument(0);
1175 Node* length = bt == T_INT ? argument(1) : argument(2);
1176 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1177 return false;
1178 }
1179
1180 // check that length is positive
1181 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1182 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1183
1184 {
1185 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1186 uncommon_trap(Deoptimization::Reason_intrinsic,
1187 Deoptimization::Action_make_not_entrant);
1188 }
1189
1190 if (stopped()) {
1191 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1192 return true;
1193 }
1194
1195 // length is now known positive, add a cast node to make this explicit
1196 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1197 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1198 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1199 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1200 casted_length = _gvn.transform(casted_length);
1201 replace_in_map(length, casted_length);
1202 length = casted_length;
1203
1204 // Use an unsigned comparison for the range check itself
1205 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1206 BoolTest::mask btest = BoolTest::lt;
1207 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1208 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1209 _gvn.set_type(rc, rc->Value(&_gvn));
1210 if (!rc_bool->is_Con()) {
1211 record_for_igvn(rc);
1212 }
1213 set_control(_gvn.transform(new IfTrueNode(rc)));
1214 {
1215 PreserveJVMState pjvms(this);
1216 set_control(_gvn.transform(new IfFalseNode(rc)));
1217 uncommon_trap(Deoptimization::Reason_range_check,
1218 Deoptimization::Action_make_not_entrant);
1219 }
1220
1221 if (stopped()) {
1222 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1223 return true;
1224 }
1225
1226 // index is now known to be >= 0 and < length, cast it
1227 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1228 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1229 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1230 result = _gvn.transform(result);
1231 set_result(result);
1232 replace_in_map(index, result);
1233 return true;
1234 }
1235
1236 //------------------------------inline_string_indexOf------------------------
1237 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1238 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1239 return false;
1240 }
1241 Node* src = argument(0);
1242 Node* tgt = argument(1);
1243
1244 // Make the merge point
1245 RegionNode* result_rgn = new RegionNode(4);
1246 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1247
1248 src = must_be_not_null(src, true);
1249 tgt = must_be_not_null(tgt, true);
1250
1251 // Get start addr and length of source string
1252 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1253 Node* src_count = load_array_length(src);
1254
1255 // Get start addr and length of substring
1256 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1257 Node* tgt_count = load_array_length(tgt);
1258
1259 Node* result = nullptr;
1260 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1261
1262 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1263 // Divide src size by 2 if String is UTF16 encoded
1264 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1265 }
1266 if (ae == StrIntrinsicNode::UU) {
1267 // Divide substring size by 2 if String is UTF16 encoded
1268 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1269 }
1270
1271 if (call_opt_stub) {
1272 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1273 StubRoutines::_string_indexof_array[ae],
1274 "stringIndexOf", TypePtr::BOTTOM, src_start,
1275 src_count, tgt_start, tgt_count);
1276 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1277 } else {
1278 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1279 result_rgn, result_phi, ae);
1280 }
1281 if (result != nullptr) {
1282 result_phi->init_req(3, result);
1283 result_rgn->init_req(3, control());
1284 }
1285 set_control(_gvn.transform(result_rgn));
1286 record_for_igvn(result_rgn);
1287 set_result(_gvn.transform(result_phi));
1288
1289 return true;
1290 }
1291
1292 //-----------------------------inline_string_indexOfI-----------------------
1293 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1294 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1295 return false;
1296 }
1297 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1298 return false;
1299 }
1300
1301 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1302 Node* src = argument(0); // byte[]
1303 Node* src_count = argument(1); // char count
1304 Node* tgt = argument(2); // byte[]
1305 Node* tgt_count = argument(3); // char count
1306 Node* from_index = argument(4); // char index
1307
1308 src = must_be_not_null(src, true);
1309 tgt = must_be_not_null(tgt, true);
1310
1311 // Multiply byte array index by 2 if String is UTF16 encoded
1312 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1313 src_count = _gvn.transform(new SubINode(src_count, from_index));
1314 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1315 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1316
1317 // Range checks
1318 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1319 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1320 if (stopped()) {
1321 return true;
1322 }
1323
1324 RegionNode* region = new RegionNode(5);
1325 Node* phi = new PhiNode(region, TypeInt::INT);
1326 Node* result = nullptr;
1327
1328 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1329
1330 if (call_opt_stub) {
1331 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1332 StubRoutines::_string_indexof_array[ae],
1333 "stringIndexOf", TypePtr::BOTTOM, src_start,
1334 src_count, tgt_start, tgt_count);
1335 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1336 } else {
1337 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1338 region, phi, ae);
1339 }
1340 if (result != nullptr) {
1341 // The result is index relative to from_index if substring was found, -1 otherwise.
1342 // Generate code which will fold into cmove.
1343 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1344 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1345
1346 Node* if_lt = generate_slow_guard(bol, nullptr);
1347 if (if_lt != nullptr) {
1348 // result == -1
1349 phi->init_req(3, result);
1350 region->init_req(3, if_lt);
1351 }
1352 if (!stopped()) {
1353 result = _gvn.transform(new AddINode(result, from_index));
1354 phi->init_req(4, result);
1355 region->init_req(4, control());
1356 }
1357 }
1358
1359 set_control(_gvn.transform(region));
1360 record_for_igvn(region);
1361 set_result(_gvn.transform(phi));
1362 clear_upper_avx();
1363
1364 return true;
1365 }
1366
1367 // Create StrIndexOfNode with fast path checks
1368 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1369 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1370 // Check for substr count > string count
1371 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1372 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1373 Node* if_gt = generate_slow_guard(bol, nullptr);
1374 if (if_gt != nullptr) {
1375 phi->init_req(1, intcon(-1));
1376 region->init_req(1, if_gt);
1377 }
1378 if (!stopped()) {
1379 // Check for substr count == 0
1380 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1381 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1382 Node* if_zero = generate_slow_guard(bol, nullptr);
1383 if (if_zero != nullptr) {
1384 phi->init_req(2, intcon(0));
1385 region->init_req(2, if_zero);
1386 }
1387 }
1388 if (!stopped()) {
1389 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1390 }
1391 return nullptr;
1392 }
1393
1394 //-----------------------------inline_string_indexOfChar-----------------------
1395 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1396 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1397 return false;
1398 }
1399 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1400 return false;
1401 }
1402 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1403 Node* src = argument(0); // byte[]
1404 Node* int_ch = argument(1);
1405 Node* from_index = argument(2);
1406 Node* max = argument(3);
1407
1408 src = must_be_not_null(src, true);
1409
1410 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1411 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1412 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1413
1414 // Range checks
1415 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1416
1417 // Check for int_ch >= 0
1418 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1419 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1420 {
1421 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1422 uncommon_trap(Deoptimization::Reason_intrinsic,
1423 Deoptimization::Action_maybe_recompile);
1424 }
1425 if (stopped()) {
1426 return true;
1427 }
1428
1429 RegionNode* region = new RegionNode(3);
1430 Node* phi = new PhiNode(region, TypeInt::INT);
1431
1432 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1433 C->set_has_split_ifs(true); // Has chance for split-if optimization
1434 _gvn.transform(result);
1435
1436 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1437 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1438
1439 Node* if_lt = generate_slow_guard(bol, nullptr);
1440 if (if_lt != nullptr) {
1441 // result == -1
1442 phi->init_req(2, result);
1443 region->init_req(2, if_lt);
1444 }
1445 if (!stopped()) {
1446 result = _gvn.transform(new AddINode(result, from_index));
1447 phi->init_req(1, result);
1448 region->init_req(1, control());
1449 }
1450 set_control(_gvn.transform(region));
1451 record_for_igvn(region);
1452 set_result(_gvn.transform(phi));
1453 clear_upper_avx();
1454
1455 return true;
1456 }
1457 //---------------------------inline_string_copy---------------------
1458 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1459 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1460 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1461 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1462 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1463 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1464 bool LibraryCallKit::inline_string_copy(bool compress) {
1465 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1466 return false;
1467 }
1468 int nargs = 5; // 2 oops, 3 ints
1469 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1470
1471 Node* src = argument(0);
1472 Node* src_offset = argument(1);
1473 Node* dst = argument(2);
1474 Node* dst_offset = argument(3);
1475 Node* length = argument(4);
1476
1477 // Check for allocation before we add nodes that would confuse
1478 // tightly_coupled_allocation()
1479 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1480
1481 // Figure out the size and type of the elements we will be copying.
1482 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1483 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1484 if (src_type == nullptr || dst_type == nullptr) {
1485 return false;
1486 }
1487 BasicType src_elem = src_type->elem()->array_element_basic_type();
1488 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1489 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1490 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1491 "Unsupported array types for inline_string_copy");
1492
1493 src = must_be_not_null(src, true);
1494 dst = must_be_not_null(dst, true);
1495
1496 // Convert char[] offsets to byte[] offsets
1497 bool convert_src = (compress && src_elem == T_BYTE);
1498 bool convert_dst = (!compress && dst_elem == T_BYTE);
1499 if (convert_src) {
1500 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1501 } else if (convert_dst) {
1502 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1503 }
1504
1505 // Range checks
1506 generate_string_range_check(src, src_offset, length, convert_src);
1507 generate_string_range_check(dst, dst_offset, length, convert_dst);
1508 if (stopped()) {
1509 return true;
1510 }
1511
1512 Node* src_start = array_element_address(src, src_offset, src_elem);
1513 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1514 // 'src_start' points to src array + scaled offset
1515 // 'dst_start' points to dst array + scaled offset
1516 Node* count = nullptr;
1517 if (compress) {
1518 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1519 } else {
1520 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1521 }
1522
1523 if (alloc != nullptr) {
1524 if (alloc->maybe_set_complete(&_gvn)) {
1525 // "You break it, you buy it."
1526 InitializeNode* init = alloc->initialization();
1527 assert(init->is_complete(), "we just did this");
1528 init->set_complete_with_arraycopy();
1529 assert(dst->is_CheckCastPP(), "sanity");
1530 assert(dst->in(0)->in(0) == init, "dest pinned");
1531 }
1532 // Do not let stores that initialize this object be reordered with
1533 // a subsequent store that would make this object accessible by
1534 // other threads.
1535 // Record what AllocateNode this StoreStore protects so that
1536 // escape analysis can go from the MemBarStoreStoreNode to the
1537 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1538 // based on the escape status of the AllocateNode.
1539 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1540 }
1541 if (compress) {
1542 set_result(_gvn.transform(count));
1543 }
1544 clear_upper_avx();
1545
1546 return true;
1547 }
1548
1549 #ifdef _LP64
1550 #define XTOP ,top() /*additional argument*/
1551 #else //_LP64
1552 #define XTOP /*no additional argument*/
1553 #endif //_LP64
1554
1555 //------------------------inline_string_toBytesU--------------------------
1556 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1557 bool LibraryCallKit::inline_string_toBytesU() {
1558 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1559 return false;
1560 }
1561 // Get the arguments.
1562 Node* value = argument(0);
1563 Node* offset = argument(1);
1564 Node* length = argument(2);
1565
1566 Node* newcopy = nullptr;
1567
1568 // Set the original stack and the reexecute bit for the interpreter to reexecute
1569 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1570 { PreserveReexecuteState preexecs(this);
1571 jvms()->set_should_reexecute(true);
1572
1573 // Check if a null path was taken unconditionally.
1574 value = null_check(value);
1575
1576 RegionNode* bailout = new RegionNode(1);
1577 record_for_igvn(bailout);
1578
1579 // Range checks
1580 generate_negative_guard(offset, bailout);
1581 generate_negative_guard(length, bailout);
1582 generate_limit_guard(offset, length, load_array_length(value), bailout);
1583 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1584 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1585
1586 if (bailout->req() > 1) {
1587 PreserveJVMState pjvms(this);
1588 set_control(_gvn.transform(bailout));
1589 uncommon_trap(Deoptimization::Reason_intrinsic,
1590 Deoptimization::Action_maybe_recompile);
1591 }
1592 if (stopped()) {
1593 return true;
1594 }
1595
1596 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1597 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1598 newcopy = new_array(klass_node, size, 0); // no arguments to push
1599 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1600 guarantee(alloc != nullptr, "created above");
1601
1602 // Calculate starting addresses.
1603 Node* src_start = array_element_address(value, offset, T_CHAR);
1604 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1605
1606 // Check if dst array address is aligned to HeapWordSize
1607 bool aligned = (arrayOopDesc::base_offset_in_bytes(T_BYTE) % HeapWordSize == 0);
1608 // If true, then check if src array address is aligned to HeapWordSize
1609 if (aligned) {
1610 const TypeInt* toffset = gvn().type(offset)->is_int();
1611 aligned = toffset->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) +
1612 toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1613 }
1614
1615 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1616 const char* copyfunc_name = "arraycopy";
1617 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1618 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1619 OptoRuntime::fast_arraycopy_Type(),
1620 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1621 src_start, dst_start, ConvI2X(length) XTOP);
1622 // Do not let reads from the cloned object float above the arraycopy.
1623 if (alloc->maybe_set_complete(&_gvn)) {
1624 // "You break it, you buy it."
1625 InitializeNode* init = alloc->initialization();
1626 assert(init->is_complete(), "we just did this");
1627 init->set_complete_with_arraycopy();
1628 assert(newcopy->is_CheckCastPP(), "sanity");
1629 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1630 }
1631 // Do not let stores that initialize this object be reordered with
1632 // a subsequent store that would make this object accessible by
1633 // other threads.
1634 // Record what AllocateNode this StoreStore protects so that
1635 // escape analysis can go from the MemBarStoreStoreNode to the
1636 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1637 // based on the escape status of the AllocateNode.
1638 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1639 } // original reexecute is set back here
1640
1641 C->set_has_split_ifs(true); // Has chance for split-if optimization
1642 if (!stopped()) {
1643 set_result(newcopy);
1644 }
1645 clear_upper_avx();
1646
1647 return true;
1648 }
1649
1650 //------------------------inline_string_getCharsU--------------------------
1651 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1652 bool LibraryCallKit::inline_string_getCharsU() {
1653 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1654 return false;
1655 }
1656
1657 // Get the arguments.
1658 Node* src = argument(0);
1659 Node* src_begin = argument(1);
1660 Node* src_end = argument(2); // exclusive offset (i < src_end)
1661 Node* dst = argument(3);
1662 Node* dst_begin = argument(4);
1663
1664 // Check for allocation before we add nodes that would confuse
1665 // tightly_coupled_allocation()
1666 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1667
1668 // Check if a null path was taken unconditionally.
1669 src = null_check(src);
1670 dst = null_check(dst);
1671 if (stopped()) {
1672 return true;
1673 }
1674
1675 // Get length and convert char[] offset to byte[] offset
1676 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1677 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1678
1679 // Range checks
1680 generate_string_range_check(src, src_begin, length, true);
1681 generate_string_range_check(dst, dst_begin, length, false);
1682 if (stopped()) {
1683 return true;
1684 }
1685
1686 if (!stopped()) {
1687 // Calculate starting addresses.
1688 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1689 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1690
1691 // Check if array addresses are aligned to HeapWordSize
1692 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1693 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1694 bool aligned = tsrc->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_BYTE) + tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1695 tdst->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) + tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1696
1697 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1698 const char* copyfunc_name = "arraycopy";
1699 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1700 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1701 OptoRuntime::fast_arraycopy_Type(),
1702 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1703 src_start, dst_start, ConvI2X(length) XTOP);
1704 // Do not let reads from the cloned object float above the arraycopy.
1705 if (alloc != nullptr) {
1706 if (alloc->maybe_set_complete(&_gvn)) {
1707 // "You break it, you buy it."
1708 InitializeNode* init = alloc->initialization();
1709 assert(init->is_complete(), "we just did this");
1710 init->set_complete_with_arraycopy();
1711 assert(dst->is_CheckCastPP(), "sanity");
1712 assert(dst->in(0)->in(0) == init, "dest pinned");
1713 }
1714 // Do not let stores that initialize this object be reordered with
1715 // a subsequent store that would make this object accessible by
1716 // other threads.
1717 // Record what AllocateNode this StoreStore protects so that
1718 // escape analysis can go from the MemBarStoreStoreNode to the
1719 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1720 // based on the escape status of the AllocateNode.
1721 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1722 } else {
1723 insert_mem_bar(Op_MemBarCPUOrder);
1724 }
1725 }
1726
1727 C->set_has_split_ifs(true); // Has chance for split-if optimization
1728 return true;
1729 }
1730
1731 //----------------------inline_string_char_access----------------------------
1732 // Store/Load char to/from byte[] array.
1733 // static void StringUTF16.putChar(byte[] val, int index, int c)
1734 // static char StringUTF16.getChar(byte[] val, int index)
1735 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1736 Node* value = argument(0);
1737 Node* index = argument(1);
1738 Node* ch = is_store ? argument(2) : nullptr;
1739
1740 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1741 // correctly requires matched array shapes.
1742 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1743 "sanity: byte[] and char[] bases agree");
1744 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1745 "sanity: byte[] and char[] scales agree");
1746
1747 // Bail when getChar over constants is requested: constant folding would
1748 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1749 // Java method would constant fold nicely instead.
1750 if (!is_store && value->is_Con() && index->is_Con()) {
1751 return false;
1752 }
1753
1754 // Save state and restore on bailout
1755 SavedState old_state(this);
1756
1757 value = must_be_not_null(value, true);
1758
1759 Node* adr = array_element_address(value, index, T_CHAR);
1760 if (adr->is_top()) {
1761 return false;
1762 }
1763 old_state.discard();
1764 if (is_store) {
1765 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1766 } else {
1767 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);
1768 set_result(ch);
1769 }
1770 return true;
1771 }
1772
1773
1774 //------------------------------inline_math-----------------------------------
1775 // public static double Math.abs(double)
1776 // public static double Math.sqrt(double)
1777 // public static double Math.log(double)
1778 // public static double Math.log10(double)
1779 // public static double Math.round(double)
1780 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1781 Node* arg = argument(0);
1782 Node* n = nullptr;
1783 switch (id) {
1784 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1785 case vmIntrinsics::_dsqrt:
1786 case vmIntrinsics::_dsqrt_strict:
1787 n = new SqrtDNode(C, control(), arg); break;
1788 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1789 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1790 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1791 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1792 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1793 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1794 default: fatal_unexpected_iid(id); break;
1795 }
1796 set_result(_gvn.transform(n));
1797 return true;
1798 }
1799
1800 //------------------------------inline_math-----------------------------------
1801 // public static float Math.abs(float)
1802 // public static int Math.abs(int)
1803 // public static long Math.abs(long)
1804 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1805 Node* arg = argument(0);
1806 Node* n = nullptr;
1807 switch (id) {
1808 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1809 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1810 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1811 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1812 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1813 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1814 default: fatal_unexpected_iid(id); break;
1815 }
1816 set_result(_gvn.transform(n));
1817 return true;
1818 }
1819
1820 //------------------------------runtime_math-----------------------------
1821 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1822 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1823 "must be (DD)D or (D)D type");
1824
1825 // Inputs
1826 Node* a = argument(0);
1827 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1828
1829 const TypePtr* no_memory_effects = nullptr;
1830 Node* trig = make_runtime_call(RC_LEAF | RC_PURE, call_type, funcAddr, funcName,
1831 no_memory_effects,
1832 a, top(), b, b ? top() : nullptr);
1833 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1834 #ifdef ASSERT
1835 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1836 assert(value_top == top(), "second value must be top");
1837 #endif
1838
1839 set_result(value);
1840 return true;
1841 }
1842
1843 //------------------------------inline_math_pow-----------------------------
1844 bool LibraryCallKit::inline_math_pow() {
1845 Node* exp = argument(2);
1846 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1847 if (d != nullptr) {
1848 if (d->getd() == 2.0) {
1849 // Special case: pow(x, 2.0) => x * x
1850 Node* base = argument(0);
1851 set_result(_gvn.transform(new MulDNode(base, base)));
1852 return true;
1853 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1854 // Special case: pow(x, 0.5) => sqrt(x)
1855 Node* base = argument(0);
1856 Node* zero = _gvn.zerocon(T_DOUBLE);
1857
1858 RegionNode* region = new RegionNode(3);
1859 Node* phi = new PhiNode(region, Type::DOUBLE);
1860
1861 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1862 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1863 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1864 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1865 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1866
1867 Node* if_pow = generate_slow_guard(test, nullptr);
1868 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1869 phi->init_req(1, value_sqrt);
1870 region->init_req(1, control());
1871
1872 if (if_pow != nullptr) {
1873 set_control(if_pow);
1874 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1875 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1876 const TypePtr* no_memory_effects = nullptr;
1877 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1878 no_memory_effects, base, top(), exp, top());
1879 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1880 #ifdef ASSERT
1881 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1882 assert(value_top == top(), "second value must be top");
1883 #endif
1884 phi->init_req(2, value_pow);
1885 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1886 }
1887
1888 C->set_has_split_ifs(true); // Has chance for split-if optimization
1889 set_control(_gvn.transform(region));
1890 record_for_igvn(region);
1891 set_result(_gvn.transform(phi));
1892
1893 return true;
1894 }
1895 }
1896
1897 return StubRoutines::dpow() != nullptr ?
1898 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1899 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1900 }
1901
1902 //------------------------------inline_math_native-----------------------------
1903 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1904 switch (id) {
1905 case vmIntrinsics::_dsin:
1906 return StubRoutines::dsin() != nullptr ?
1907 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1908 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1909 case vmIntrinsics::_dcos:
1910 return StubRoutines::dcos() != nullptr ?
1911 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1912 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1913 case vmIntrinsics::_dtan:
1914 return StubRoutines::dtan() != nullptr ?
1915 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1916 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1917 case vmIntrinsics::_dsinh:
1918 return StubRoutines::dsinh() != nullptr ?
1919 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsinh(), "dsinh") : false;
1920 case vmIntrinsics::_dtanh:
1921 return StubRoutines::dtanh() != nullptr ?
1922 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1923 case vmIntrinsics::_dcbrt:
1924 return StubRoutines::dcbrt() != nullptr ?
1925 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcbrt(), "dcbrt") : false;
1926 case vmIntrinsics::_dexp:
1927 return StubRoutines::dexp() != nullptr ?
1928 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1929 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1930 case vmIntrinsics::_dlog:
1931 return StubRoutines::dlog() != nullptr ?
1932 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1933 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1934 case vmIntrinsics::_dlog10:
1935 return StubRoutines::dlog10() != nullptr ?
1936 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1937 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1938
1939 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1940 case vmIntrinsics::_ceil:
1941 case vmIntrinsics::_floor:
1942 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1943
1944 case vmIntrinsics::_dsqrt:
1945 case vmIntrinsics::_dsqrt_strict:
1946 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1947 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1948 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1949 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1950 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1951
1952 case vmIntrinsics::_dpow: return inline_math_pow();
1953 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1954 case vmIntrinsics::_fcopySign: return inline_math(id);
1955 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1956 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1957 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1958
1959 // These intrinsics are not yet correctly implemented
1960 case vmIntrinsics::_datan2:
1961 return false;
1962
1963 default:
1964 fatal_unexpected_iid(id);
1965 return false;
1966 }
1967 }
1968
1969 //----------------------------inline_notify-----------------------------------*
1970 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1971 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1972 address func;
1973 if (id == vmIntrinsics::_notify) {
1974 func = OptoRuntime::monitor_notify_Java();
1975 } else {
1976 func = OptoRuntime::monitor_notifyAll_Java();
1977 }
1978 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1979 make_slow_call_ex(call, env()->Throwable_klass(), false);
1980 return true;
1981 }
1982
1983
1984 //----------------------------inline_min_max-----------------------------------
1985 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1986 Node* a = nullptr;
1987 Node* b = nullptr;
1988 Node* n = nullptr;
1989 switch (id) {
1990 case vmIntrinsics::_min:
1991 case vmIntrinsics::_max:
1992 case vmIntrinsics::_minF:
1993 case vmIntrinsics::_maxF:
1994 case vmIntrinsics::_minF_strict:
1995 case vmIntrinsics::_maxF_strict:
1996 case vmIntrinsics::_min_strict:
1997 case vmIntrinsics::_max_strict:
1998 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
1999 a = argument(0);
2000 b = argument(1);
2001 break;
2002 case vmIntrinsics::_minD:
2003 case vmIntrinsics::_maxD:
2004 case vmIntrinsics::_minD_strict:
2005 case vmIntrinsics::_maxD_strict:
2006 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
2007 a = argument(0);
2008 b = argument(2);
2009 break;
2010 case vmIntrinsics::_minL:
2011 case vmIntrinsics::_maxL:
2012 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
2013 a = argument(0);
2014 b = argument(2);
2015 break;
2016 default:
2017 fatal_unexpected_iid(id);
2018 break;
2019 }
2020
2021 switch (id) {
2022 case vmIntrinsics::_min:
2023 case vmIntrinsics::_min_strict:
2024 n = new MinINode(a, b);
2025 break;
2026 case vmIntrinsics::_max:
2027 case vmIntrinsics::_max_strict:
2028 n = new MaxINode(a, b);
2029 break;
2030 case vmIntrinsics::_minF:
2031 case vmIntrinsics::_minF_strict:
2032 n = new MinFNode(a, b);
2033 break;
2034 case vmIntrinsics::_maxF:
2035 case vmIntrinsics::_maxF_strict:
2036 n = new MaxFNode(a, b);
2037 break;
2038 case vmIntrinsics::_minD:
2039 case vmIntrinsics::_minD_strict:
2040 n = new MinDNode(a, b);
2041 break;
2042 case vmIntrinsics::_maxD:
2043 case vmIntrinsics::_maxD_strict:
2044 n = new MaxDNode(a, b);
2045 break;
2046 case vmIntrinsics::_minL:
2047 n = new MinLNode(_gvn.C, a, b);
2048 break;
2049 case vmIntrinsics::_maxL:
2050 n = new MaxLNode(_gvn.C, a, b);
2051 break;
2052 default:
2053 fatal_unexpected_iid(id);
2054 break;
2055 }
2056
2057 set_result(_gvn.transform(n));
2058 return true;
2059 }
2060
2061 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
2062 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
2063 env()->ArithmeticException_instance())) {
2064 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
2065 // so let's bail out intrinsic rather than risking deopting again.
2066 return false;
2067 }
2068
2069 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2070 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2071 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2072 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2073
2074 {
2075 PreserveJVMState pjvms(this);
2076 PreserveReexecuteState preexecs(this);
2077 jvms()->set_should_reexecute(true);
2078
2079 set_control(slow_path);
2080 set_i_o(i_o());
2081
2082 builtin_throw(Deoptimization::Reason_intrinsic,
2083 env()->ArithmeticException_instance(),
2084 /*allow_too_many_traps*/ false);
2085 }
2086
2087 set_control(fast_path);
2088 set_result(math);
2089 return true;
2090 }
2091
2092 template <typename OverflowOp>
2093 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2094 typedef typename OverflowOp::MathOp MathOp;
2095
2096 MathOp* mathOp = new MathOp(arg1, arg2);
2097 Node* operation = _gvn.transform( mathOp );
2098 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2099 return inline_math_mathExact(operation, ofcheck);
2100 }
2101
2102 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2103 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2104 }
2105
2106 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2107 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2108 }
2109
2110 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2111 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2112 }
2113
2114 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2115 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2116 }
2117
2118 bool LibraryCallKit::inline_math_negateExactI() {
2119 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2120 }
2121
2122 bool LibraryCallKit::inline_math_negateExactL() {
2123 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2124 }
2125
2126 bool LibraryCallKit::inline_math_multiplyExactI() {
2127 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2128 }
2129
2130 bool LibraryCallKit::inline_math_multiplyExactL() {
2131 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2132 }
2133
2134 bool LibraryCallKit::inline_math_multiplyHigh() {
2135 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2136 return true;
2137 }
2138
2139 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2140 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2141 return true;
2142 }
2143
2144 inline int
2145 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2146 const TypePtr* base_type = TypePtr::NULL_PTR;
2147 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2148 if (base_type == nullptr) {
2149 // Unknown type.
2150 return Type::AnyPtr;
2151 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2152 // Since this is a null+long form, we have to switch to a rawptr.
2153 base = _gvn.transform(new CastX2PNode(offset));
2154 offset = MakeConX(0);
2155 return Type::RawPtr;
2156 } else if (base_type->base() == Type::RawPtr) {
2157 return Type::RawPtr;
2158 } else if (base_type->isa_oopptr()) {
2159 // Base is never null => always a heap address.
2160 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2161 return Type::OopPtr;
2162 }
2163 // Offset is small => always a heap address.
2164 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2165 if (offset_type != nullptr &&
2166 base_type->offset() == 0 && // (should always be?)
2167 offset_type->_lo >= 0 &&
2168 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2169 return Type::OopPtr;
2170 } else if (type == T_OBJECT) {
2171 // off heap access to an oop doesn't make any sense. Has to be on
2172 // heap.
2173 return Type::OopPtr;
2174 }
2175 // Otherwise, it might either be oop+off or null+addr.
2176 return Type::AnyPtr;
2177 } else {
2178 // No information:
2179 return Type::AnyPtr;
2180 }
2181 }
2182
2183 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2184 Node* uncasted_base = base;
2185 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2186 if (kind == Type::RawPtr) {
2187 return basic_plus_adr(top(), uncasted_base, offset);
2188 } else if (kind == Type::AnyPtr) {
2189 assert(base == uncasted_base, "unexpected base change");
2190 if (can_cast) {
2191 if (!_gvn.type(base)->speculative_maybe_null() &&
2192 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2193 // According to profiling, this access is always on
2194 // heap. Casting the base to not null and thus avoiding membars
2195 // around the access should allow better optimizations
2196 Node* null_ctl = top();
2197 base = null_check_oop(base, &null_ctl, true, true, true);
2198 assert(null_ctl->is_top(), "no null control here");
2199 return basic_plus_adr(base, offset);
2200 } else if (_gvn.type(base)->speculative_always_null() &&
2201 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2202 // According to profiling, this access is always off
2203 // heap.
2204 base = null_assert(base);
2205 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2206 offset = MakeConX(0);
2207 return basic_plus_adr(top(), raw_base, offset);
2208 }
2209 }
2210 // We don't know if it's an on heap or off heap access. Fall back
2211 // to raw memory access.
2212 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2213 return basic_plus_adr(top(), raw, offset);
2214 } else {
2215 assert(base == uncasted_base, "unexpected base change");
2216 // We know it's an on heap access so base can't be null
2217 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2218 base = must_be_not_null(base, true);
2219 }
2220 return basic_plus_adr(base, offset);
2221 }
2222 }
2223
2224 //--------------------------inline_number_methods-----------------------------
2225 // inline int Integer.numberOfLeadingZeros(int)
2226 // inline int Long.numberOfLeadingZeros(long)
2227 //
2228 // inline int Integer.numberOfTrailingZeros(int)
2229 // inline int Long.numberOfTrailingZeros(long)
2230 //
2231 // inline int Integer.bitCount(int)
2232 // inline int Long.bitCount(long)
2233 //
2234 // inline char Character.reverseBytes(char)
2235 // inline short Short.reverseBytes(short)
2236 // inline int Integer.reverseBytes(int)
2237 // inline long Long.reverseBytes(long)
2238 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2239 Node* arg = argument(0);
2240 Node* n = nullptr;
2241 switch (id) {
2242 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2243 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2244 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2245 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2246 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2247 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2248 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2249 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2250 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2251 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2252 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2253 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2254 default: fatal_unexpected_iid(id); break;
2255 }
2256 set_result(_gvn.transform(n));
2257 return true;
2258 }
2259
2260 //--------------------------inline_bitshuffle_methods-----------------------------
2261 // inline int Integer.compress(int, int)
2262 // inline int Integer.expand(int, int)
2263 // inline long Long.compress(long, long)
2264 // inline long Long.expand(long, long)
2265 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2266 Node* n = nullptr;
2267 switch (id) {
2268 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2269 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2270 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2271 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2272 default: fatal_unexpected_iid(id); break;
2273 }
2274 set_result(_gvn.transform(n));
2275 return true;
2276 }
2277
2278 //--------------------------inline_number_methods-----------------------------
2279 // inline int Integer.compareUnsigned(int, int)
2280 // inline int Long.compareUnsigned(long, long)
2281 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2282 Node* arg1 = argument(0);
2283 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2284 Node* n = nullptr;
2285 switch (id) {
2286 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2287 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2288 default: fatal_unexpected_iid(id); break;
2289 }
2290 set_result(_gvn.transform(n));
2291 return true;
2292 }
2293
2294 //--------------------------inline_unsigned_divmod_methods-----------------------------
2295 // inline int Integer.divideUnsigned(int, int)
2296 // inline int Integer.remainderUnsigned(int, int)
2297 // inline long Long.divideUnsigned(long, long)
2298 // inline long Long.remainderUnsigned(long, long)
2299 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2300 Node* n = nullptr;
2301 switch (id) {
2302 case vmIntrinsics::_divideUnsigned_i: {
2303 zero_check_int(argument(1));
2304 // Compile-time detect of null-exception
2305 if (stopped()) {
2306 return true; // keep the graph constructed so far
2307 }
2308 n = new UDivINode(control(), argument(0), argument(1));
2309 break;
2310 }
2311 case vmIntrinsics::_divideUnsigned_l: {
2312 zero_check_long(argument(2));
2313 // Compile-time detect of null-exception
2314 if (stopped()) {
2315 return true; // keep the graph constructed so far
2316 }
2317 n = new UDivLNode(control(), argument(0), argument(2));
2318 break;
2319 }
2320 case vmIntrinsics::_remainderUnsigned_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 UModINode(control(), argument(0), argument(1));
2327 break;
2328 }
2329 case vmIntrinsics::_remainderUnsigned_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 UModLNode(control(), argument(0), argument(2));
2336 break;
2337 }
2338 default: fatal_unexpected_iid(id); break;
2339 }
2340 set_result(_gvn.transform(n));
2341 return true;
2342 }
2343
2344 //----------------------------inline_unsafe_access----------------------------
2345
2346 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2347 // Attempt to infer a sharper value type from the offset and base type.
2348 ciKlass* sharpened_klass = nullptr;
2349 bool null_free = false;
2350
2351 // See if it is an instance field, with an object type.
2352 if (alias_type->field() != nullptr) {
2353 if (alias_type->field()->type()->is_klass()) {
2354 sharpened_klass = alias_type->field()->type()->as_klass();
2355 null_free = alias_type->field()->is_null_free();
2356 }
2357 }
2358
2359 const TypeOopPtr* result = nullptr;
2360 // See if it is a narrow oop array.
2361 if (adr_type->isa_aryptr()) {
2362 if (adr_type->offset() >= refArrayOopDesc::base_offset_in_bytes()) {
2363 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2364 null_free = adr_type->is_aryptr()->is_null_free();
2365 if (elem_type != nullptr && elem_type->is_loaded()) {
2366 // Sharpen the value type.
2367 result = elem_type;
2368 }
2369 }
2370 }
2371
2372 // The sharpened class might be unloaded if there is no class loader
2373 // contraint in place.
2374 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2375 // Sharpen the value type.
2376 result = TypeOopPtr::make_from_klass(sharpened_klass);
2377 if (null_free) {
2378 result = result->join_speculative(TypePtr::NOTNULL)->is_oopptr();
2379 }
2380 }
2381 if (result != nullptr) {
2382 #ifndef PRODUCT
2383 if (C->print_intrinsics() || C->print_inlining()) {
2384 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2385 tty->print(" sharpened value: "); result->dump(); tty->cr();
2386 }
2387 #endif
2388 }
2389 return result;
2390 }
2391
2392 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2393 switch (kind) {
2394 case Relaxed:
2395 return MO_UNORDERED;
2396 case Opaque:
2397 return MO_RELAXED;
2398 case Acquire:
2399 return MO_ACQUIRE;
2400 case Release:
2401 return MO_RELEASE;
2402 case Volatile:
2403 return MO_SEQ_CST;
2404 default:
2405 ShouldNotReachHere();
2406 return 0;
2407 }
2408 }
2409
2410 LibraryCallKit::SavedState::SavedState(LibraryCallKit* kit) :
2411 _kit(kit),
2412 _sp(kit->sp()),
2413 _jvms(kit->jvms()),
2414 _map(kit->clone_map()),
2415 _discarded(false)
2416 {
2417 for (DUIterator_Fast imax, i = kit->control()->fast_outs(imax); i < imax; i++) {
2418 Node* out = kit->control()->fast_out(i);
2419 if (out->is_CFG()) {
2420 _ctrl_succ.push(out);
2421 }
2422 }
2423 }
2424
2425 LibraryCallKit::SavedState::~SavedState() {
2426 if (_discarded) {
2427 _kit->destruct_map_clone(_map);
2428 return;
2429 }
2430 _kit->jvms()->set_map(_map);
2431 _kit->jvms()->set_sp(_sp);
2432 _map->set_jvms(_kit->jvms());
2433 _kit->set_map(_map);
2434 _kit->set_sp(_sp);
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() && out->in(0) == _kit->control() && out != _kit->map() && !_ctrl_succ.member(out)) {
2438 _kit->_gvn.hash_delete(out);
2439 out->set_req(0, _kit->C->top());
2440 _kit->C->record_for_igvn(out);
2441 --i; --imax;
2442 _kit->_gvn.hash_find_insert(out);
2443 }
2444 }
2445 }
2446
2447 void LibraryCallKit::SavedState::discard() {
2448 _discarded = true;
2449 }
2450
2451 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2452 if (callee()->is_static()) return false; // caller must have the capability!
2453 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2454 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2455 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2456 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2457
2458 if (is_reference_type(type)) {
2459 decorators |= ON_UNKNOWN_OOP_REF;
2460 }
2461
2462 if (unaligned) {
2463 decorators |= C2_UNALIGNED;
2464 }
2465
2466 #ifndef PRODUCT
2467 {
2468 ResourceMark rm;
2469 // Check the signatures.
2470 ciSignature* sig = callee()->signature();
2471 #ifdef ASSERT
2472 if (!is_store) {
2473 // Object getReference(Object base, int/long offset), etc.
2474 BasicType rtype = sig->return_type()->basic_type();
2475 assert(rtype == type, "getter must return the expected value");
2476 assert(sig->count() == 2, "oop getter has 2 arguments");
2477 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2478 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2479 } else {
2480 // void putReference(Object base, int/long offset, Object x), etc.
2481 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2482 assert(sig->count() == 3, "oop putter has 3 arguments");
2483 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2484 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2485 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2486 assert(vtype == type, "putter must accept the expected value");
2487 }
2488 #endif // ASSERT
2489 }
2490 #endif //PRODUCT
2491
2492 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2493
2494 Node* receiver = argument(0); // type: oop
2495
2496 // Build address expression.
2497 Node* heap_base_oop = top();
2498
2499 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2500 Node* base = argument(1); // type: oop
2501 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2502 Node* offset = argument(2); // type: long
2503 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2504 // to be plain byte offsets, which are also the same as those accepted
2505 // by oopDesc::field_addr.
2506 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2507 "fieldOffset must be byte-scaled");
2508
2509 if (base->is_InlineType()) {
2510 assert(!is_store, "InlineTypeNodes are non-larval value objects");
2511 InlineTypeNode* vt = base->as_InlineType();
2512 if (offset->is_Con()) {
2513 long off = find_long_con(offset, 0);
2514 ciInlineKlass* vk = vt->type()->inline_klass();
2515 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2516 return false;
2517 }
2518
2519 ciField* field = vk->get_non_flat_field_by_offset(off);
2520 if (field != nullptr) {
2521 BasicType bt = type2field[field->type()->basic_type()];
2522 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2523 bt = T_OBJECT;
2524 }
2525 if (bt == type && !field->is_flat()) {
2526 Node* value = vt->field_value_by_offset(off, false);
2527 if (value->is_InlineType()) {
2528 value = value->as_InlineType()->adjust_scalarization_depth(this);
2529 }
2530 set_result(value);
2531 return true;
2532 }
2533 }
2534 }
2535 {
2536 // Re-execute the unsafe access if allocation triggers deoptimization.
2537 PreserveReexecuteState preexecs(this);
2538 jvms()->set_should_reexecute(true);
2539 vt = vt->buffer(this);
2540 }
2541 base = vt->get_oop();
2542 }
2543
2544 // 32-bit machines ignore the high half!
2545 offset = ConvL2X(offset);
2546
2547 // Save state and restore on bailout
2548 SavedState old_state(this);
2549
2550 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2551 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2552
2553 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2554 if (type != T_OBJECT) {
2555 decorators |= IN_NATIVE; // off-heap primitive access
2556 } else {
2557 return false; // off-heap oop accesses are not supported
2558 }
2559 } else {
2560 heap_base_oop = base; // on-heap or mixed access
2561 }
2562
2563 // Can base be null? Otherwise, always on-heap access.
2564 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2565
2566 if (!can_access_non_heap) {
2567 decorators |= IN_HEAP;
2568 }
2569
2570 Node* val = is_store ? argument(4) : nullptr;
2571
2572 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2573 if (adr_type == TypePtr::NULL_PTR) {
2574 return false; // off-heap access with zero address
2575 }
2576
2577 // Try to categorize the address.
2578 Compile::AliasType* alias_type = C->alias_type(adr_type);
2579 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2580
2581 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2582 alias_type->adr_type() == TypeAryPtr::RANGE) {
2583 return false; // not supported
2584 }
2585
2586 bool mismatched = false;
2587 BasicType bt = T_ILLEGAL;
2588 ciField* field = nullptr;
2589 if (adr_type->isa_instptr()) {
2590 const TypeInstPtr* instptr = adr_type->is_instptr();
2591 ciInstanceKlass* k = instptr->instance_klass();
2592 int off = instptr->offset();
2593 if (instptr->const_oop() != nullptr &&
2594 k == ciEnv::current()->Class_klass() &&
2595 instptr->offset() >= (k->size_helper() * wordSize)) {
2596 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2597 field = k->get_field_by_offset(off, true);
2598 } else {
2599 field = k->get_non_flat_field_by_offset(off);
2600 }
2601 if (field != nullptr) {
2602 bt = type2field[field->type()->basic_type()];
2603 }
2604 if (bt != alias_type->basic_type()) {
2605 // Type mismatch. Is it an access to a nested flat field?
2606 field = k->get_field_by_offset(off, false);
2607 if (field != nullptr) {
2608 bt = type2field[field->type()->basic_type()];
2609 }
2610 }
2611 assert(bt == alias_type->basic_type(), "should match");
2612 } else {
2613 bt = alias_type->basic_type();
2614 }
2615
2616 if (bt != T_ILLEGAL) {
2617 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2618 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2619 // Alias type doesn't differentiate between byte[] and boolean[]).
2620 // Use address type to get the element type.
2621 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2622 }
2623 if (is_reference_type(bt, true)) {
2624 // accessing an array field with getReference is not a mismatch
2625 bt = T_OBJECT;
2626 }
2627 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2628 // Don't intrinsify mismatched object accesses
2629 return false;
2630 }
2631 mismatched = (bt != type);
2632 } else if (alias_type->adr_type()->isa_oopptr()) {
2633 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2634 }
2635
2636 old_state.discard();
2637 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2638
2639 if (mismatched) {
2640 decorators |= C2_MISMATCHED;
2641 }
2642
2643 // First guess at the value type.
2644 const Type *value_type = Type::get_const_basic_type(type);
2645
2646 // Figure out the memory ordering.
2647 decorators |= mo_decorator_for_access_kind(kind);
2648
2649 if (!is_store) {
2650 if (type == T_OBJECT) {
2651 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2652 if (tjp != nullptr) {
2653 value_type = tjp;
2654 }
2655 }
2656 }
2657
2658 receiver = null_check(receiver);
2659 if (stopped()) {
2660 return true;
2661 }
2662 // Heap pointers get a null-check from the interpreter,
2663 // as a courtesy. However, this is not guaranteed by Unsafe,
2664 // and it is not possible to fully distinguish unintended nulls
2665 // from intended ones in this API.
2666
2667 if (!is_store) {
2668 Node* p = nullptr;
2669 // Try to constant fold a load from a constant field
2670
2671 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !field->is_flat() && !mismatched) {
2672 // final or stable field
2673 p = make_constant_from_field(field, heap_base_oop);
2674 }
2675
2676 if (p == nullptr) { // Could not constant fold the load
2677 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2678 const TypeOopPtr* ptr = value_type->make_oopptr();
2679 if (ptr != nullptr && ptr->is_inlinetypeptr()) {
2680 // Load a non-flattened inline type from memory
2681 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass());
2682 }
2683 // Normalize the value returned by getBoolean in the following cases
2684 if (type == T_BOOLEAN &&
2685 (mismatched ||
2686 heap_base_oop == top() || // - heap_base_oop is null or
2687 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2688 // and the unsafe access is made to large offset
2689 // (i.e., larger than the maximum offset necessary for any
2690 // field access)
2691 ) {
2692 IdealKit ideal = IdealKit(this);
2693 #define __ ideal.
2694 IdealVariable normalized_result(ideal);
2695 __ declarations_done();
2696 __ set(normalized_result, p);
2697 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2698 __ set(normalized_result, ideal.ConI(1));
2699 ideal.end_if();
2700 final_sync(ideal);
2701 p = __ value(normalized_result);
2702 #undef __
2703 }
2704 }
2705 if (type == T_ADDRESS) {
2706 p = gvn().transform(new CastP2XNode(nullptr, p));
2707 p = ConvX2UL(p);
2708 }
2709 // The load node has the control of the preceding MemBarCPUOrder. All
2710 // following nodes will have the control of the MemBarCPUOrder inserted at
2711 // the end of this method. So, pushing the load onto the stack at a later
2712 // point is fine.
2713 set_result(p);
2714 } else {
2715 if (bt == T_ADDRESS) {
2716 // Repackage the long as a pointer.
2717 val = ConvL2X(val);
2718 val = gvn().transform(new CastX2PNode(val));
2719 }
2720 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2721 }
2722
2723 return true;
2724 }
2725
2726 bool LibraryCallKit::inline_unsafe_flat_access(bool is_store, AccessKind kind) {
2727 #ifdef ASSERT
2728 {
2729 ResourceMark rm;
2730 // Check the signatures.
2731 ciSignature* sig = callee()->signature();
2732 assert(sig->type_at(0)->basic_type() == T_OBJECT, "base should be object, but is %s", type2name(sig->type_at(0)->basic_type()));
2733 assert(sig->type_at(1)->basic_type() == T_LONG, "offset should be long, but is %s", type2name(sig->type_at(1)->basic_type()));
2734 assert(sig->type_at(2)->basic_type() == T_INT, "layout kind should be int, but is %s", type2name(sig->type_at(3)->basic_type()));
2735 assert(sig->type_at(3)->basic_type() == T_OBJECT, "value klass should be object, but is %s", type2name(sig->type_at(4)->basic_type()));
2736 if (is_store) {
2737 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value, but returns %s", type2name(sig->return_type()->basic_type()));
2738 assert(sig->count() == 5, "flat putter should have 5 arguments, but has %d", sig->count());
2739 assert(sig->type_at(4)->basic_type() == T_OBJECT, "put value should be object, but is %s", type2name(sig->type_at(5)->basic_type()));
2740 } else {
2741 assert(sig->return_type()->basic_type() == T_OBJECT, "getter must return an object, but returns %s", type2name(sig->return_type()->basic_type()));
2742 assert(sig->count() == 4, "flat getter should have 4 arguments, but has %d", sig->count());
2743 }
2744 }
2745 #endif // ASSERT
2746
2747 assert(kind == Relaxed, "Only plain accesses for now");
2748 if (callee()->is_static()) {
2749 // caller must have the capability!
2750 return false;
2751 }
2752 C->set_has_unsafe_access(true);
2753
2754 const TypeInstPtr* value_klass_node = _gvn.type(argument(5))->isa_instptr();
2755 if (value_klass_node == nullptr || value_klass_node->const_oop() == nullptr) {
2756 // parameter valueType is not a constant
2757 return false;
2758 }
2759 ciType* mirror_type = value_klass_node->const_oop()->as_instance()->java_mirror_type();
2760 if (!mirror_type->is_inlinetype()) {
2761 // Dead code
2762 return false;
2763 }
2764 ciInlineKlass* value_klass = mirror_type->as_inline_klass();
2765
2766 const TypeInt* layout_type = _gvn.type(argument(4))->isa_int();
2767 if (layout_type == nullptr || !layout_type->is_con()) {
2768 // parameter layoutKind is not a constant
2769 return false;
2770 }
2771 assert(layout_type->get_con() >= static_cast<int>(LayoutKind::REFERENCE) &&
2772 layout_type->get_con() <= static_cast<int>(LayoutKind::UNKNOWN),
2773 "invalid layoutKind %d", layout_type->get_con());
2774 LayoutKind layout = static_cast<LayoutKind>(layout_type->get_con());
2775 assert(layout == LayoutKind::REFERENCE || layout == LayoutKind::NULL_FREE_NON_ATOMIC_FLAT ||
2776 layout == LayoutKind::NULL_FREE_ATOMIC_FLAT || layout == LayoutKind::NULLABLE_ATOMIC_FLAT,
2777 "unexpected layoutKind %d", layout_type->get_con());
2778
2779 null_check(argument(0));
2780 if (stopped()) {
2781 return true;
2782 }
2783
2784 Node* base = must_be_not_null(argument(1), true);
2785 Node* offset = argument(2);
2786 const Type* base_type = _gvn.type(base);
2787
2788 Node* ptr;
2789 bool immutable_memory = false;
2790 DecoratorSet decorators = C2_UNSAFE_ACCESS | IN_HEAP | MO_UNORDERED;
2791 if (base_type->isa_instptr()) {
2792 const TypeLong* offset_type = _gvn.type(offset)->isa_long();
2793 if (offset_type == nullptr || !offset_type->is_con()) {
2794 // Offset into a non-array should be a constant
2795 decorators |= C2_MISMATCHED;
2796 } else {
2797 int offset_con = checked_cast<int>(offset_type->get_con());
2798 ciInstanceKlass* base_klass = base_type->is_instptr()->instance_klass();
2799 ciField* field = base_klass->get_non_flat_field_by_offset(offset_con);
2800 if (field == nullptr) {
2801 assert(!base_klass->is_final(), "non-existence field at offset %d of class %s", offset_con, base_klass->name()->as_utf8());
2802 decorators |= C2_MISMATCHED;
2803 } else {
2804 assert(field->type() == value_klass, "field at offset %d of %s is of type %s, but valueType is %s",
2805 offset_con, base_klass->name()->as_utf8(), field->type()->name(), value_klass->name()->as_utf8());
2806 immutable_memory = field->is_strict() && field->is_final();
2807
2808 if (base->is_InlineType()) {
2809 assert(!is_store, "Cannot store into a non-larval value object");
2810 set_result(base->as_InlineType()->field_value_by_offset(offset_con, false));
2811 return true;
2812 }
2813 }
2814 }
2815
2816 if (base->is_InlineType()) {
2817 assert(!is_store, "Cannot store into a non-larval value object");
2818 base = base->as_InlineType()->buffer(this, true);
2819 }
2820 ptr = basic_plus_adr(base, ConvL2X(offset));
2821 } else if (base_type->isa_aryptr()) {
2822 decorators |= IS_ARRAY;
2823 if (layout == LayoutKind::REFERENCE) {
2824 if (!base_type->is_aryptr()->is_not_flat()) {
2825 const TypeAryPtr* array_type = base_type->is_aryptr()->cast_to_not_flat();
2826 Node* new_base = _gvn.transform(new CastPPNode(control(), base, array_type, ConstraintCastNode::DependencyType::NonFloatingNarrowing));
2827 replace_in_map(base, new_base);
2828 base = new_base;
2829 }
2830 ptr = basic_plus_adr(base, ConvL2X(offset));
2831 } else {
2832 if (UseArrayFlattening) {
2833 // Flat array must have an exact type
2834 bool is_null_free = !LayoutKindHelper::is_nullable_flat(layout);
2835 bool is_atomic = LayoutKindHelper::is_atomic_flat(layout);
2836 Node* new_base = cast_to_flat_array_exact(base, value_klass, is_null_free, is_atomic);
2837 replace_in_map(base, new_base);
2838 base = new_base;
2839 ptr = basic_plus_adr(base, ConvL2X(offset));
2840 const TypeAryPtr* ptr_type = _gvn.type(ptr)->is_aryptr();
2841 if (ptr_type->field_offset().get() != 0) {
2842 ptr = _gvn.transform(new CastPPNode(control(), ptr, ptr_type->with_field_offset(0), ConstraintCastNode::DependencyType::NonFloatingNarrowing));
2843 }
2844 } else {
2845 uncommon_trap(Deoptimization::Reason_intrinsic,
2846 Deoptimization::Action_none);
2847 return true;
2848 }
2849 }
2850 } else {
2851 decorators |= C2_MISMATCHED;
2852 ptr = basic_plus_adr(base, ConvL2X(offset));
2853 }
2854
2855 if (is_store) {
2856 Node* value = argument(6);
2857 const Type* value_type = _gvn.type(value);
2858 if (!value_type->is_inlinetypeptr()) {
2859 value_type = Type::get_const_type(value_klass)->filter_speculative(value_type);
2860 Node* new_value = _gvn.transform(new CastPPNode(control(), value, value_type, ConstraintCastNode::DependencyType::NonFloatingNarrowing));
2861 new_value = InlineTypeNode::make_from_oop(this, new_value, value_klass);
2862 replace_in_map(value, new_value);
2863 value = new_value;
2864 }
2865
2866 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());
2867 if (layout == LayoutKind::REFERENCE) {
2868 const TypePtr* ptr_type = (decorators & C2_MISMATCHED) != 0 ? TypeRawPtr::BOTTOM : _gvn.type(ptr)->is_ptr();
2869 access_store_at(base, ptr, ptr_type, value, value_type, T_OBJECT, decorators);
2870 } else {
2871 bool atomic = LayoutKindHelper::is_atomic_flat(layout);
2872 bool null_free = !LayoutKindHelper::is_nullable_flat(layout);
2873 value->as_InlineType()->store_flat(this, base, ptr, atomic, immutable_memory, null_free, decorators);
2874 }
2875
2876 return true;
2877 } else {
2878 decorators |= (C2_CONTROL_DEPENDENT_LOAD | C2_UNKNOWN_CONTROL_LOAD);
2879 InlineTypeNode* result;
2880 if (layout == LayoutKind::REFERENCE) {
2881 const TypePtr* ptr_type = (decorators & C2_MISMATCHED) != 0 ? TypeRawPtr::BOTTOM : _gvn.type(ptr)->is_ptr();
2882 Node* oop = access_load_at(base, ptr, ptr_type, Type::get_const_type(value_klass), T_OBJECT, decorators);
2883 result = InlineTypeNode::make_from_oop(this, oop, value_klass);
2884 } else {
2885 bool atomic = LayoutKindHelper::is_atomic_flat(layout);
2886 bool null_free = !LayoutKindHelper::is_nullable_flat(layout);
2887 result = InlineTypeNode::make_from_flat(this, value_klass, base, ptr, atomic, immutable_memory, null_free, decorators);
2888 }
2889
2890 set_result(result);
2891 return true;
2892 }
2893 }
2894
2895 //----------------------------inline_unsafe_load_store----------------------------
2896 // This method serves a couple of different customers (depending on LoadStoreKind):
2897 //
2898 // LS_cmp_swap:
2899 //
2900 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2901 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2902 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2903 //
2904 // LS_cmp_swap_weak:
2905 //
2906 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2907 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2908 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2909 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2910 //
2911 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2912 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2913 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2914 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2915 //
2916 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2917 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2918 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2919 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2920 //
2921 // LS_cmp_exchange:
2922 //
2923 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2924 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2925 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2926 //
2927 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2928 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2929 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2930 //
2931 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2932 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2933 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2934 //
2935 // LS_get_add:
2936 //
2937 // int getAndAddInt( Object o, long offset, int delta)
2938 // long getAndAddLong(Object o, long offset, long delta)
2939 //
2940 // LS_get_set:
2941 //
2942 // int getAndSet(Object o, long offset, int newValue)
2943 // long getAndSet(Object o, long offset, long newValue)
2944 // Object getAndSet(Object o, long offset, Object newValue)
2945 //
2946 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2947 // This basic scheme here is the same as inline_unsafe_access, but
2948 // differs in enough details that combining them would make the code
2949 // overly confusing. (This is a true fact! I originally combined
2950 // them, but even I was confused by it!) As much code/comments as
2951 // possible are retained from inline_unsafe_access though to make
2952 // the correspondences clearer. - dl
2953
2954 if (callee()->is_static()) return false; // caller must have the capability!
2955
2956 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2957 decorators |= mo_decorator_for_access_kind(access_kind);
2958
2959 #ifndef PRODUCT
2960 BasicType rtype;
2961 {
2962 ResourceMark rm;
2963 // Check the signatures.
2964 ciSignature* sig = callee()->signature();
2965 rtype = sig->return_type()->basic_type();
2966 switch(kind) {
2967 case LS_get_add:
2968 case LS_get_set: {
2969 // Check the signatures.
2970 #ifdef ASSERT
2971 assert(rtype == type, "get and set must return the expected type");
2972 assert(sig->count() == 3, "get and set has 3 arguments");
2973 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2974 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2975 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2976 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2977 #endif // ASSERT
2978 break;
2979 }
2980 case LS_cmp_swap:
2981 case LS_cmp_swap_weak: {
2982 // Check the signatures.
2983 #ifdef ASSERT
2984 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2985 assert(sig->count() == 4, "CAS has 4 arguments");
2986 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2987 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2988 #endif // ASSERT
2989 break;
2990 }
2991 case LS_cmp_exchange: {
2992 // Check the signatures.
2993 #ifdef ASSERT
2994 assert(rtype == type, "CAS must return the expected type");
2995 assert(sig->count() == 4, "CAS has 4 arguments");
2996 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2997 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2998 #endif // ASSERT
2999 break;
3000 }
3001 default:
3002 ShouldNotReachHere();
3003 }
3004 }
3005 #endif //PRODUCT
3006
3007 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
3008
3009 // Get arguments:
3010 Node* receiver = nullptr;
3011 Node* base = nullptr;
3012 Node* offset = nullptr;
3013 Node* oldval = nullptr;
3014 Node* newval = nullptr;
3015 switch(kind) {
3016 case LS_cmp_swap:
3017 case LS_cmp_swap_weak:
3018 case LS_cmp_exchange: {
3019 const bool two_slot_type = type2size[type] == 2;
3020 receiver = argument(0); // type: oop
3021 base = argument(1); // type: oop
3022 offset = argument(2); // type: long
3023 oldval = argument(4); // type: oop, int, or long
3024 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
3025 break;
3026 }
3027 case LS_get_add:
3028 case LS_get_set: {
3029 receiver = argument(0); // type: oop
3030 base = argument(1); // type: oop
3031 offset = argument(2); // type: long
3032 oldval = nullptr;
3033 newval = argument(4); // type: oop, int, or long
3034 break;
3035 }
3036 default:
3037 ShouldNotReachHere();
3038 }
3039
3040 // Build field offset expression.
3041 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
3042 // to be plain byte offsets, which are also the same as those accepted
3043 // by oopDesc::field_addr.
3044 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
3045 // 32-bit machines ignore the high half of long offsets
3046 offset = ConvL2X(offset);
3047 // Save state and restore on bailout
3048 SavedState old_state(this);
3049 Node* adr = make_unsafe_address(base, offset,type, false);
3050 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
3051
3052 Compile::AliasType* alias_type = C->alias_type(adr_type);
3053 BasicType bt = alias_type->basic_type();
3054 if (bt != T_ILLEGAL &&
3055 (is_reference_type(bt) != (type == T_OBJECT))) {
3056 // Don't intrinsify mismatched object accesses.
3057 return false;
3058 }
3059
3060 old_state.discard();
3061
3062 // For CAS, unlike inline_unsafe_access, there seems no point in
3063 // trying to refine types. Just use the coarse types here.
3064 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
3065 const Type *value_type = Type::get_const_basic_type(type);
3066
3067 switch (kind) {
3068 case LS_get_set:
3069 case LS_cmp_exchange: {
3070 if (type == T_OBJECT) {
3071 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
3072 if (tjp != nullptr) {
3073 value_type = tjp;
3074 }
3075 }
3076 break;
3077 }
3078 case LS_cmp_swap:
3079 case LS_cmp_swap_weak:
3080 case LS_get_add:
3081 break;
3082 default:
3083 ShouldNotReachHere();
3084 }
3085
3086 // Null check receiver.
3087 receiver = null_check(receiver);
3088 if (stopped()) {
3089 return true;
3090 }
3091
3092 int alias_idx = C->get_alias_index(adr_type);
3093
3094 if (is_reference_type(type)) {
3095 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
3096
3097 if (oldval != nullptr && oldval->is_InlineType()) {
3098 // Re-execute the unsafe access if allocation triggers deoptimization.
3099 PreserveReexecuteState preexecs(this);
3100 jvms()->set_should_reexecute(true);
3101 oldval = oldval->as_InlineType()->buffer(this)->get_oop();
3102 }
3103 if (newval != nullptr && newval->is_InlineType()) {
3104 // Re-execute the unsafe access if allocation triggers deoptimization.
3105 PreserveReexecuteState preexecs(this);
3106 jvms()->set_should_reexecute(true);
3107 newval = newval->as_InlineType()->buffer(this)->get_oop();
3108 }
3109
3110 // Transformation of a value which could be null pointer (CastPP #null)
3111 // could be delayed during Parse (for example, in adjust_map_after_if()).
3112 // Execute transformation here to avoid barrier generation in such case.
3113 if (_gvn.type(newval) == TypePtr::NULL_PTR)
3114 newval = _gvn.makecon(TypePtr::NULL_PTR);
3115
3116 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
3117 // Refine the value to a null constant, when it is known to be null
3118 oldval = _gvn.makecon(TypePtr::NULL_PTR);
3119 }
3120 }
3121
3122 Node* result = nullptr;
3123 switch (kind) {
3124 case LS_cmp_exchange: {
3125 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
3126 oldval, newval, value_type, type, decorators);
3127 break;
3128 }
3129 case LS_cmp_swap_weak:
3130 decorators |= C2_WEAK_CMPXCHG;
3131 case LS_cmp_swap: {
3132 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
3133 oldval, newval, value_type, type, decorators);
3134 break;
3135 }
3136 case LS_get_set: {
3137 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
3138 newval, value_type, type, decorators);
3139 break;
3140 }
3141 case LS_get_add: {
3142 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
3143 newval, value_type, type, decorators);
3144 break;
3145 }
3146 default:
3147 ShouldNotReachHere();
3148 }
3149
3150 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3151 set_result(result);
3152 return true;
3153 }
3154
3155 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3156 // Regardless of form, don't allow previous ld/st to move down,
3157 // then issue acquire, release, or volatile mem_bar.
3158 insert_mem_bar(Op_MemBarCPUOrder);
3159 switch(id) {
3160 case vmIntrinsics::_loadFence:
3161 insert_mem_bar(Op_LoadFence);
3162 return true;
3163 case vmIntrinsics::_storeFence:
3164 insert_mem_bar(Op_StoreFence);
3165 return true;
3166 case vmIntrinsics::_storeStoreFence:
3167 insert_mem_bar(Op_StoreStoreFence);
3168 return true;
3169 case vmIntrinsics::_fullFence:
3170 insert_mem_bar(Op_MemBarVolatile);
3171 return true;
3172 default:
3173 fatal_unexpected_iid(id);
3174 return false;
3175 }
3176 }
3177
3178 // private native int arrayInstanceBaseOffset0(Object[] array);
3179 bool LibraryCallKit::inline_arrayInstanceBaseOffset() {
3180 Node* array = argument(1);
3181 Node* klass_node = load_object_klass(array);
3182
3183 jint layout_con = Klass::_lh_neutral_value;
3184 Node* layout_val = get_layout_helper(klass_node, layout_con);
3185 int layout_is_con = (layout_val == nullptr);
3186
3187 Node* header_size = nullptr;
3188 if (layout_is_con) {
3189 int hsize = Klass::layout_helper_header_size(layout_con);
3190 header_size = intcon(hsize);
3191 } else {
3192 Node* hss = intcon(Klass::_lh_header_size_shift);
3193 Node* hsm = intcon(Klass::_lh_header_size_mask);
3194 header_size = _gvn.transform(new URShiftINode(layout_val, hss));
3195 header_size = _gvn.transform(new AndINode(header_size, hsm));
3196 }
3197 set_result(header_size);
3198 return true;
3199 }
3200
3201 // private native int arrayInstanceIndexScale0(Object[] array);
3202 bool LibraryCallKit::inline_arrayInstanceIndexScale() {
3203 Node* array = argument(1);
3204 Node* klass_node = load_object_klass(array);
3205
3206 jint layout_con = Klass::_lh_neutral_value;
3207 Node* layout_val = get_layout_helper(klass_node, layout_con);
3208 int layout_is_con = (layout_val == nullptr);
3209
3210 Node* element_size = nullptr;
3211 if (layout_is_con) {
3212 int log_element_size = Klass::layout_helper_log2_element_size(layout_con);
3213 int elem_size = 1 << log_element_size;
3214 element_size = intcon(elem_size);
3215 } else {
3216 Node* ess = intcon(Klass::_lh_log2_element_size_shift);
3217 Node* esm = intcon(Klass::_lh_log2_element_size_mask);
3218 Node* log_element_size = _gvn.transform(new URShiftINode(layout_val, ess));
3219 log_element_size = _gvn.transform(new AndINode(log_element_size, esm));
3220 element_size = _gvn.transform(new LShiftINode(intcon(1), log_element_size));
3221 }
3222 set_result(element_size);
3223 return true;
3224 }
3225
3226 // private native int arrayLayout0(Object[] array);
3227 bool LibraryCallKit::inline_arrayLayout() {
3228 RegionNode* region = new RegionNode(2);
3229 Node* phi = new PhiNode(region, TypeInt::POS);
3230
3231 Node* array = argument(1);
3232 Node* klass_node = load_object_klass(array);
3233 generate_refArray_guard(klass_node, region);
3234 if (region->req() == 3) {
3235 phi->add_req(intcon((jint)LayoutKind::REFERENCE));
3236 }
3237
3238 int layout_kind_offset = in_bytes(FlatArrayKlass::layout_kind_offset());
3239 Node* layout_kind_addr = basic_plus_adr(klass_node, layout_kind_offset);
3240 Node* layout_kind = make_load(nullptr, layout_kind_addr, TypeInt::POS, T_INT, MemNode::unordered);
3241
3242 region->init_req(1, control());
3243 phi->init_req(1, layout_kind);
3244
3245 set_control(_gvn.transform(region));
3246 set_result(_gvn.transform(phi));
3247 return true;
3248 }
3249
3250 // private native int[] getFieldMap0(Class <?> c);
3251 // int offset = c._klass._acmp_maps_offset;
3252 // return (int[])c.obj_field(offset);
3253 bool LibraryCallKit::inline_getFieldMap() {
3254 Node* mirror = argument(1);
3255 Node* klass = load_klass_from_mirror(mirror, false, nullptr, 0);
3256
3257 int field_map_offset_offset = in_bytes(InstanceKlass::acmp_maps_offset_offset());
3258 Node* field_map_offset_addr = basic_plus_adr(klass, field_map_offset_offset);
3259 Node* field_map_offset = make_load(nullptr, field_map_offset_addr, TypeInt::INT, T_INT, MemNode::unordered);
3260 field_map_offset = _gvn.transform(ConvI2L(field_map_offset));
3261
3262 Node* map_addr = basic_plus_adr(mirror, field_map_offset);
3263 const TypeAryPtr* val_type = TypeAryPtr::INTS->cast_to_ptr_type(TypePtr::NotNull)->with_offset(0);
3264 // TODO 8350865 Remove this
3265 val_type = val_type->cast_to_not_flat(true)->cast_to_not_null_free(true);
3266 Node* map = access_load_at(mirror, map_addr, TypeAryPtr::INTS, val_type, T_ARRAY, IN_HEAP | MO_UNORDERED);
3267
3268 set_result(map);
3269 return true;
3270 }
3271
3272 bool LibraryCallKit::inline_onspinwait() {
3273 insert_mem_bar(Op_OnSpinWait);
3274 return true;
3275 }
3276
3277 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3278 if (!kls->is_Con()) {
3279 return true;
3280 }
3281 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
3282 if (klsptr == nullptr) {
3283 return true;
3284 }
3285 ciInstanceKlass* ik = klsptr->instance_klass();
3286 // don't need a guard for a klass that is already initialized
3287 return !ik->is_initialized();
3288 }
3289
3290 //----------------------------inline_unsafe_writeback0-------------------------
3291 // public native void Unsafe.writeback0(long address)
3292 bool LibraryCallKit::inline_unsafe_writeback0() {
3293 if (!Matcher::has_match_rule(Op_CacheWB)) {
3294 return false;
3295 }
3296 #ifndef PRODUCT
3297 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
3298 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
3299 ciSignature* sig = callee()->signature();
3300 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
3301 #endif
3302 null_check_receiver(); // null-check, then ignore
3303 Node *addr = argument(1);
3304 addr = new CastX2PNode(addr);
3305 addr = _gvn.transform(addr);
3306 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
3307 flush = _gvn.transform(flush);
3308 set_memory(flush, TypeRawPtr::BOTTOM);
3309 return true;
3310 }
3311
3312 //----------------------------inline_unsafe_writeback0-------------------------
3313 // public native void Unsafe.writeback0(long address)
3314 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
3315 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
3316 return false;
3317 }
3318 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
3319 return false;
3320 }
3321 #ifndef PRODUCT
3322 assert(Matcher::has_match_rule(Op_CacheWB),
3323 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
3324 : "found match rule for CacheWBPostSync but not CacheWB"));
3325
3326 #endif
3327 null_check_receiver(); // null-check, then ignore
3328 Node *sync;
3329 if (is_pre) {
3330 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3331 } else {
3332 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3333 }
3334 sync = _gvn.transform(sync);
3335 set_memory(sync, TypeRawPtr::BOTTOM);
3336 return true;
3337 }
3338
3339 //----------------------------inline_unsafe_allocate---------------------------
3340 // public native Object Unsafe.allocateInstance(Class<?> cls);
3341 bool LibraryCallKit::inline_unsafe_allocate() {
3342
3343 #if INCLUDE_JVMTI
3344 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3345 return false;
3346 }
3347 #endif //INCLUDE_JVMTI
3348
3349 if (callee()->is_static()) return false; // caller must have the capability!
3350
3351 null_check_receiver(); // null-check, then ignore
3352 Node* cls = null_check(argument(1));
3353 if (stopped()) return true;
3354
3355 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
3356 kls = null_check(kls);
3357 if (stopped()) return true; // argument was like int.class
3358
3359 #if INCLUDE_JVMTI
3360 // Don't try to access new allocated obj in the intrinsic.
3361 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
3362 // Deoptimize and allocate in interpreter instead.
3363 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
3364 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
3365 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
3366 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
3367 {
3368 BuildCutout unless(this, tst, PROB_MAX);
3369 uncommon_trap(Deoptimization::Reason_intrinsic,
3370 Deoptimization::Action_make_not_entrant);
3371 }
3372 if (stopped()) {
3373 return true;
3374 }
3375 #endif //INCLUDE_JVMTI
3376
3377 Node* test = nullptr;
3378 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3379 // Note: The argument might still be an illegal value like
3380 // Serializable.class or Object[].class. The runtime will handle it.
3381 // But we must make an explicit check for initialization.
3382 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3383 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3384 // can generate code to load it as unsigned byte.
3385 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
3386 Node* bits = intcon(InstanceKlass::fully_initialized);
3387 test = _gvn.transform(new SubINode(inst, bits));
3388 // The 'test' is non-zero if we need to take a slow path.
3389 }
3390 Node* obj = nullptr;
3391 const TypeInstKlassPtr* tkls = _gvn.type(kls)->isa_instklassptr();
3392 if (tkls != nullptr && tkls->instance_klass()->is_inlinetype()) {
3393 obj = InlineTypeNode::make_all_zero(_gvn, tkls->instance_klass()->as_inline_klass())->buffer(this);
3394 } else {
3395 obj = new_instance(kls, test);
3396 }
3397 set_result(obj);
3398 return true;
3399 }
3400
3401 //------------------------inline_native_time_funcs--------------
3402 // inline code for System.currentTimeMillis() and System.nanoTime()
3403 // these have the same type and signature
3404 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3405 const TypeFunc* tf = OptoRuntime::void_long_Type();
3406 const TypePtr* no_memory_effects = nullptr;
3407 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3408 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3409 #ifdef ASSERT
3410 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3411 assert(value_top == top(), "second value must be top");
3412 #endif
3413 set_result(value);
3414 return true;
3415 }
3416
3417 //--------------------inline_native_vthread_start_transition--------------------
3418 // inline void startTransition(boolean is_mount);
3419 // inline void startFinalTransition();
3420 // Pseudocode of implementation:
3421 //
3422 // java_lang_Thread::set_is_in_vthread_transition(vthread, true);
3423 // carrier->set_is_in_vthread_transition(true);
3424 // OrderAccess::storeload();
3425 // int disable_requests = java_lang_Thread::vthread_transition_disable_count(vthread)
3426 // + global_vthread_transition_disable_count();
3427 // if (disable_requests > 0) {
3428 // slow path: runtime call
3429 // }
3430 bool LibraryCallKit::inline_native_vthread_start_transition(address funcAddr, const char* funcName, bool is_final_transition) {
3431 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3432 IdealKit ideal(this);
3433
3434 Node* thread = ideal.thread();
3435 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3436 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3437 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3438 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3439 insert_mem_bar(Op_MemBarVolatile);
3440 ideal.sync_kit(this);
3441
3442 Node* global_disable_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::global_vthread_transition_disable_count_address()));
3443 Node* global_disable = ideal.load(ideal.ctrl(), global_disable_addr, TypeInt::INT, T_INT, Compile::AliasIdxRaw, true /*require_atomic_access*/);
3444 Node* vt_disable_addr = basic_plus_adr(vt_oop, java_lang_Thread::vthread_transition_disable_count_offset());
3445 Node* vt_disable = ideal.load(ideal.ctrl(), vt_disable_addr, TypeInt::INT, T_INT, Compile::AliasIdxRaw, true /*require_atomic_access*/);
3446 Node* disabled = _gvn.transform(new AddINode(global_disable, vt_disable));
3447
3448 ideal.if_then(disabled, BoolTest::ne, ideal.ConI(0)); {
3449 sync_kit(ideal);
3450 Node* is_mount = is_final_transition ? ideal.ConI(0) : _gvn.transform(argument(1));
3451 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3452 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3453 ideal.sync_kit(this);
3454 }
3455 ideal.end_if();
3456
3457 final_sync(ideal);
3458 return true;
3459 }
3460
3461 bool LibraryCallKit::inline_native_vthread_end_transition(address funcAddr, const char* funcName, bool is_first_transition) {
3462 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3463 IdealKit ideal(this);
3464
3465 Node* _notify_jvmti_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::notify_jvmti_events_address()));
3466 Node* _notify_jvmti = ideal.load(ideal.ctrl(), _notify_jvmti_addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3467
3468 ideal.if_then(_notify_jvmti, BoolTest::eq, ideal.ConI(1)); {
3469 sync_kit(ideal);
3470 Node* is_mount = is_first_transition ? ideal.ConI(1) : _gvn.transform(argument(1));
3471 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3472 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3473 ideal.sync_kit(this);
3474 } ideal.else_(); {
3475 Node* thread = ideal.thread();
3476 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3477 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3478
3479 sync_kit(ideal);
3480 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3481 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3482 ideal.sync_kit(this);
3483 } ideal.end_if();
3484
3485 final_sync(ideal);
3486 return true;
3487 }
3488
3489 #if INCLUDE_JVMTI
3490
3491 // Always update the is_disable_suspend bit.
3492 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3493 if (!DoJVMTIVirtualThreadTransitions) {
3494 return true;
3495 }
3496 IdealKit ideal(this);
3497
3498 {
3499 // unconditionally update the is_disable_suspend bit in current JavaThread
3500 Node* thread = ideal.thread();
3501 Node* arg = _gvn.transform(argument(0)); // argument for notification
3502 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3503 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3504
3505 sync_kit(ideal);
3506 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3507 ideal.sync_kit(this);
3508 }
3509 final_sync(ideal);
3510
3511 return true;
3512 }
3513
3514 #endif // INCLUDE_JVMTI
3515
3516 #ifdef JFR_HAVE_INTRINSICS
3517
3518 /**
3519 * if oop->klass != null
3520 * // normal class
3521 * epoch = _epoch_state ? 2 : 1
3522 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3523 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3524 * }
3525 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3526 * else
3527 * // primitive class
3528 * if oop->array_klass != null
3529 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3530 * else
3531 * id = LAST_TYPE_ID + 1 // void class path
3532 * if (!signaled)
3533 * signaled = true
3534 */
3535 bool LibraryCallKit::inline_native_classID() {
3536 Node* cls = argument(0);
3537
3538 IdealKit ideal(this);
3539 #define __ ideal.
3540 IdealVariable result(ideal); __ declarations_done();
3541 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3542 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3543 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3544
3545
3546 __ if_then(kls, BoolTest::ne, null()); {
3547 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3548 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3549
3550 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3551 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3552 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3553 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3554 mask = _gvn.transform(new OrLNode(mask, epoch));
3555 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3556
3557 float unlikely = PROB_UNLIKELY(0.999);
3558 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3559 sync_kit(ideal);
3560 make_runtime_call(RC_LEAF,
3561 OptoRuntime::class_id_load_barrier_Type(),
3562 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3563 "class id load barrier",
3564 TypePtr::BOTTOM,
3565 kls);
3566 ideal.sync_kit(this);
3567 } __ end_if();
3568
3569 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3570 } __ else_(); {
3571 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3572 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3573 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3574 __ if_then(array_kls, BoolTest::ne, null()); {
3575 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3576 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3577 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3578 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3579 } __ else_(); {
3580 // void class case
3581 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3582 } __ end_if();
3583
3584 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3585 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3586 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3587 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3588 } __ end_if();
3589 } __ end_if();
3590
3591 final_sync(ideal);
3592 set_result(ideal.value(result));
3593 #undef __
3594 return true;
3595 }
3596
3597 //------------------------inline_native_jvm_commit------------------
3598 bool LibraryCallKit::inline_native_jvm_commit() {
3599 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3600
3601 // Save input memory and i_o state.
3602 Node* input_memory_state = reset_memory();
3603 set_all_memory(input_memory_state);
3604 Node* input_io_state = i_o();
3605
3606 // TLS.
3607 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3608 // Jfr java buffer.
3609 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3610 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3611 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3612
3613 // Load the current value of the notified field in the JfrThreadLocal.
3614 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3615 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3616
3617 // Test for notification.
3618 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3619 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3620 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3621
3622 // True branch, is notified.
3623 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3624 set_control(is_notified);
3625
3626 // Reset notified state.
3627 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3628 Node* notified_reset_memory = reset_memory();
3629
3630 // 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.
3631 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3632 // Convert the machine-word to a long.
3633 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3634
3635 // False branch, not notified.
3636 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3637 set_control(not_notified);
3638 set_all_memory(input_memory_state);
3639
3640 // Arg is the next position as a long.
3641 Node* arg = argument(0);
3642 // Convert long to machine-word.
3643 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3644
3645 // Store the next_position to the underlying jfr java buffer.
3646 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3647
3648 Node* commit_memory = reset_memory();
3649 set_all_memory(commit_memory);
3650
3651 // 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.
3652 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3653 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3654 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3655
3656 // And flags with lease constant.
3657 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3658
3659 // Branch on lease to conditionalize returning the leased java buffer.
3660 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3661 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3662 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3663
3664 // False branch, not a lease.
3665 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3666
3667 // True branch, is lease.
3668 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3669 set_control(is_lease);
3670
3671 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3672 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3673 OptoRuntime::void_void_Type(),
3674 SharedRuntime::jfr_return_lease(),
3675 "return_lease", TypePtr::BOTTOM);
3676 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3677
3678 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3679 record_for_igvn(lease_compare_rgn);
3680 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3681 record_for_igvn(lease_compare_mem);
3682 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3683 record_for_igvn(lease_compare_io);
3684 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3685 record_for_igvn(lease_result_value);
3686
3687 // Update control and phi nodes.
3688 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3689 lease_compare_rgn->init_req(_false_path, not_lease);
3690
3691 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3692 lease_compare_mem->init_req(_false_path, commit_memory);
3693
3694 lease_compare_io->init_req(_true_path, i_o());
3695 lease_compare_io->init_req(_false_path, input_io_state);
3696
3697 lease_result_value->init_req(_true_path, _gvn.longcon(0)); // if the lease was returned, return 0L.
3698 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3699
3700 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3701 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3702 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3703 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3704
3705 // Update control and phi nodes.
3706 result_rgn->init_req(_true_path, is_notified);
3707 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3708
3709 result_mem->init_req(_true_path, notified_reset_memory);
3710 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3711
3712 result_io->init_req(_true_path, input_io_state);
3713 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3714
3715 result_value->init_req(_true_path, current_pos);
3716 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3717
3718 // Set output state.
3719 set_control(_gvn.transform(result_rgn));
3720 set_all_memory(_gvn.transform(result_mem));
3721 set_i_o(_gvn.transform(result_io));
3722 set_result(result_rgn, result_value);
3723 return true;
3724 }
3725
3726 /*
3727 * The intrinsic is a model of this pseudo-code:
3728 *
3729 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3730 * jobject h_event_writer = tl->java_event_writer();
3731 * if (h_event_writer == nullptr) {
3732 * return nullptr;
3733 * }
3734 * oop threadObj = Thread::threadObj();
3735 * oop vthread = java_lang_Thread::vthread(threadObj);
3736 * traceid tid;
3737 * bool pinVirtualThread;
3738 * bool excluded;
3739 * if (vthread != threadObj) { // i.e. current thread is virtual
3740 * tid = java_lang_Thread::tid(vthread);
3741 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3742 * pinVirtualThread = VMContinuations;
3743 * excluded = vthread_epoch_raw & excluded_mask;
3744 * if (!excluded) {
3745 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3746 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3747 * if (vthread_epoch != current_epoch) {
3748 * write_checkpoint();
3749 * }
3750 * }
3751 * } else {
3752 * tid = java_lang_Thread::tid(threadObj);
3753 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3754 * pinVirtualThread = false;
3755 * excluded = thread_epoch_raw & excluded_mask;
3756 * }
3757 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3758 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3759 * if (tid_in_event_writer != tid) {
3760 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3761 * setField(event_writer, "excluded", excluded);
3762 * setField(event_writer, "threadID", tid);
3763 * }
3764 * return event_writer
3765 */
3766 bool LibraryCallKit::inline_native_getEventWriter() {
3767 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3768
3769 // Save input memory and i_o state.
3770 Node* input_memory_state = reset_memory();
3771 set_all_memory(input_memory_state);
3772 Node* input_io_state = i_o();
3773
3774 // The most significant bit of the u2 is used to denote thread exclusion
3775 Node* excluded_shift = _gvn.intcon(15);
3776 Node* excluded_mask = _gvn.intcon(1 << 15);
3777 // The epoch generation is the range [1-32767]
3778 Node* epoch_mask = _gvn.intcon(32767);
3779
3780 // TLS
3781 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3782
3783 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3784 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3785
3786 // Load the eventwriter jobject handle.
3787 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3788
3789 // Null check the jobject handle.
3790 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3791 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3792 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3793
3794 // False path, jobj is null.
3795 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3796
3797 // True path, jobj is not null.
3798 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3799
3800 set_control(jobj_is_not_null);
3801
3802 // Load the threadObj for the CarrierThread.
3803 Node* threadObj = generate_current_thread(tls_ptr);
3804
3805 // Load the vthread.
3806 Node* vthread = generate_virtual_thread(tls_ptr);
3807
3808 // If vthread != threadObj, this is a virtual thread.
3809 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3810 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3811 IfNode* iff_vthread_not_equal_threadObj =
3812 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3813
3814 // False branch, fallback to threadObj.
3815 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3816 set_control(vthread_equal_threadObj);
3817
3818 // Load the tid field from the vthread object.
3819 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3820
3821 // Load the raw epoch value from the threadObj.
3822 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3823 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3824 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3825 TypeInt::CHAR, T_CHAR,
3826 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3827
3828 // Mask off the excluded information from the epoch.
3829 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3830
3831 // True branch, this is a virtual thread.
3832 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3833 set_control(vthread_not_equal_threadObj);
3834
3835 // Load the tid field from the vthread object.
3836 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3837
3838 // Continuation support determines if a virtual thread should be pinned.
3839 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3840 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3841
3842 // Load the raw epoch value from the vthread.
3843 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3844 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3845 TypeInt::CHAR, T_CHAR,
3846 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3847
3848 // Mask off the excluded information from the epoch.
3849 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3850
3851 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3852 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3853 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3854 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3855
3856 // False branch, vthread is excluded, no need to write epoch info.
3857 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3858
3859 // True branch, vthread is included, update epoch info.
3860 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3861 set_control(included);
3862
3863 // Get epoch value.
3864 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3865
3866 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3867 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3868 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3869
3870 // Compare the epoch in the vthread to the current epoch generation.
3871 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3872 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3873 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3874
3875 // False path, epoch is equal, checkpoint information is valid.
3876 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3877
3878 // True path, epoch is not equal, write a checkpoint for the vthread.
3879 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3880
3881 set_control(epoch_is_not_equal);
3882
3883 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3884 // The call also updates the native thread local thread id and the vthread with the current epoch.
3885 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3886 OptoRuntime::jfr_write_checkpoint_Type(),
3887 SharedRuntime::jfr_write_checkpoint(),
3888 "write_checkpoint", TypePtr::BOTTOM);
3889 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3890
3891 // vthread epoch != current epoch
3892 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3893 record_for_igvn(epoch_compare_rgn);
3894 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3895 record_for_igvn(epoch_compare_mem);
3896 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3897 record_for_igvn(epoch_compare_io);
3898
3899 // Update control and phi nodes.
3900 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3901 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3902 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3903 epoch_compare_mem->init_req(_false_path, input_memory_state);
3904 epoch_compare_io->init_req(_true_path, i_o());
3905 epoch_compare_io->init_req(_false_path, input_io_state);
3906
3907 // excluded != true
3908 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3909 record_for_igvn(exclude_compare_rgn);
3910 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3911 record_for_igvn(exclude_compare_mem);
3912 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3913 record_for_igvn(exclude_compare_io);
3914
3915 // Update control and phi nodes.
3916 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3917 exclude_compare_rgn->init_req(_false_path, excluded);
3918 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3919 exclude_compare_mem->init_req(_false_path, input_memory_state);
3920 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3921 exclude_compare_io->init_req(_false_path, input_io_state);
3922
3923 // vthread != threadObj
3924 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3925 record_for_igvn(vthread_compare_rgn);
3926 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3927 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3928 record_for_igvn(vthread_compare_io);
3929 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3930 record_for_igvn(tid);
3931 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3932 record_for_igvn(exclusion);
3933 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3934 record_for_igvn(pinVirtualThread);
3935
3936 // Update control and phi nodes.
3937 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3938 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3939 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3940 vthread_compare_mem->init_req(_false_path, input_memory_state);
3941 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3942 vthread_compare_io->init_req(_false_path, input_io_state);
3943 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3944 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3945 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3946 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3947 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3948 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3949
3950 // Update branch state.
3951 set_control(_gvn.transform(vthread_compare_rgn));
3952 set_all_memory(_gvn.transform(vthread_compare_mem));
3953 set_i_o(_gvn.transform(vthread_compare_io));
3954
3955 // Load the event writer oop by dereferencing the jobject handle.
3956 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3957 assert(klass_EventWriter->is_loaded(), "invariant");
3958 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3959 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3960 const TypeOopPtr* const xtype = aklass->as_instance_type();
3961 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3962 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3963
3964 // Load the current thread id from the event writer object.
3965 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3966 // Get the field offset to, conditionally, store an updated tid value later.
3967 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3968 // Get the field offset to, conditionally, store an updated exclusion value later.
3969 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3970 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3971 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3972
3973 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3974 record_for_igvn(event_writer_tid_compare_rgn);
3975 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3976 record_for_igvn(event_writer_tid_compare_mem);
3977 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3978 record_for_igvn(event_writer_tid_compare_io);
3979
3980 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3981 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3982 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3983 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3984
3985 // False path, tids are the same.
3986 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3987
3988 // True path, tid is not equal, need to update the tid in the event writer.
3989 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3990 record_for_igvn(tid_is_not_equal);
3991
3992 // Store the pin state to the event writer.
3993 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3994
3995 // Store the exclusion state to the event writer.
3996 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3997 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3998
3999 // Store the tid to the event writer.
4000 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
4001
4002 // Update control and phi nodes.
4003 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
4004 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
4005 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
4006 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
4007 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
4008 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
4009
4010 // Result of top level CFG, Memory, IO and Value.
4011 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
4012 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
4013 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
4014 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
4015
4016 // Result control.
4017 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
4018 result_rgn->init_req(_false_path, jobj_is_null);
4019
4020 // Result memory.
4021 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
4022 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
4023
4024 // Result IO.
4025 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
4026 result_io->init_req(_false_path, _gvn.transform(input_io_state));
4027
4028 // Result value.
4029 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
4030 result_value->init_req(_false_path, null()); // return null
4031
4032 // Set output state.
4033 set_control(_gvn.transform(result_rgn));
4034 set_all_memory(_gvn.transform(result_mem));
4035 set_i_o(_gvn.transform(result_io));
4036 set_result(result_rgn, result_value);
4037 return true;
4038 }
4039
4040 /*
4041 * The intrinsic is a model of this pseudo-code:
4042 *
4043 * JfrThreadLocal* const tl = thread->jfr_thread_local();
4044 * if (carrierThread != thread) { // is virtual thread
4045 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
4046 * bool excluded = vthread_epoch_raw & excluded_mask;
4047 * AtomicAccess::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
4048 * AtomicAccess::store(&tl->_contextual_thread_excluded, is_excluded);
4049 * if (!excluded) {
4050 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
4051 * AtomicAccess::store(&tl->_vthread_epoch, vthread_epoch);
4052 * }
4053 * AtomicAccess::release_store(&tl->_vthread, true);
4054 * return;
4055 * }
4056 * AtomicAccess::release_store(&tl->_vthread, false);
4057 */
4058 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
4059 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
4060
4061 Node* input_memory_state = reset_memory();
4062 set_all_memory(input_memory_state);
4063
4064 // The most significant bit of the u2 is used to denote thread exclusion
4065 Node* excluded_mask = _gvn.intcon(1 << 15);
4066 // The epoch generation is the range [1-32767]
4067 Node* epoch_mask = _gvn.intcon(32767);
4068
4069 Node* const carrierThread = generate_current_thread(jt);
4070 // If thread != carrierThread, this is a virtual thread.
4071 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
4072 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
4073 IfNode* iff_thread_not_equal_carrierThread =
4074 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
4075
4076 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
4077
4078 // False branch, is carrierThread.
4079 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
4080 // Store release
4081 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
4082
4083 set_all_memory(input_memory_state);
4084
4085 // True branch, is virtual thread.
4086 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
4087 set_control(thread_not_equal_carrierThread);
4088
4089 // Load the raw epoch value from the vthread.
4090 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
4091 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
4092 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
4093
4094 // Mask off the excluded information from the epoch.
4095 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
4096
4097 // Load the tid field from the thread.
4098 Node* tid = load_field_from_object(thread, "tid", "J");
4099
4100 // Store the vthread tid to the jfr thread local.
4101 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
4102 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
4103
4104 // Branch is_excluded to conditionalize updating the epoch .
4105 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
4106 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
4107 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
4108
4109 // True branch, vthread is excluded, no need to write epoch info.
4110 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
4111 set_control(excluded);
4112 Node* vthread_is_excluded = _gvn.intcon(1);
4113
4114 // False branch, vthread is included, update epoch info.
4115 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
4116 set_control(included);
4117 Node* vthread_is_included = _gvn.intcon(0);
4118
4119 // Get epoch value.
4120 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
4121
4122 // Store the vthread epoch to the jfr thread local.
4123 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
4124 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
4125
4126 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
4127 record_for_igvn(excluded_rgn);
4128 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
4129 record_for_igvn(excluded_mem);
4130 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
4131 record_for_igvn(exclusion);
4132
4133 // Merge the excluded control and memory.
4134 excluded_rgn->init_req(_true_path, excluded);
4135 excluded_rgn->init_req(_false_path, included);
4136 excluded_mem->init_req(_true_path, tid_memory);
4137 excluded_mem->init_req(_false_path, included_memory);
4138 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
4139 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
4140
4141 // Set intermediate state.
4142 set_control(_gvn.transform(excluded_rgn));
4143 set_all_memory(excluded_mem);
4144
4145 // Store the vthread exclusion state to the jfr thread local.
4146 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
4147 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
4148
4149 // Store release
4150 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
4151
4152 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
4153 record_for_igvn(thread_compare_rgn);
4154 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
4155 record_for_igvn(thread_compare_mem);
4156 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
4157 record_for_igvn(vthread);
4158
4159 // Merge the thread_compare control and memory.
4160 thread_compare_rgn->init_req(_true_path, control());
4161 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
4162 thread_compare_mem->init_req(_true_path, vthread_true_memory);
4163 thread_compare_mem->init_req(_false_path, vthread_false_memory);
4164
4165 // Set output state.
4166 set_control(_gvn.transform(thread_compare_rgn));
4167 set_all_memory(_gvn.transform(thread_compare_mem));
4168 }
4169
4170 #endif // JFR_HAVE_INTRINSICS
4171
4172 //------------------------inline_native_currentCarrierThread------------------
4173 bool LibraryCallKit::inline_native_currentCarrierThread() {
4174 Node* junk = nullptr;
4175 set_result(generate_current_thread(junk));
4176 return true;
4177 }
4178
4179 //------------------------inline_native_currentThread------------------
4180 bool LibraryCallKit::inline_native_currentThread() {
4181 Node* junk = nullptr;
4182 set_result(generate_virtual_thread(junk));
4183 return true;
4184 }
4185
4186 //------------------------inline_native_setVthread------------------
4187 bool LibraryCallKit::inline_native_setCurrentThread() {
4188 assert(C->method()->changes_current_thread(),
4189 "method changes current Thread but is not annotated ChangesCurrentThread");
4190 Node* arr = argument(1);
4191 Node* thread = _gvn.transform(new ThreadLocalNode());
4192 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
4193 Node* thread_obj_handle
4194 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
4195 thread_obj_handle = _gvn.transform(thread_obj_handle);
4196 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
4197 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
4198
4199 // Change the _monitor_owner_id of the JavaThread
4200 Node* tid = load_field_from_object(arr, "tid", "J");
4201 Node* monitor_owner_id_offset = basic_plus_adr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
4202 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
4203
4204 JFR_ONLY(extend_setCurrentThread(thread, arr);)
4205 return true;
4206 }
4207
4208 const Type* LibraryCallKit::scopedValueCache_type() {
4209 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
4210 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
4211 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS, /* stable= */ false, /* flat= */ false, /* not_flat= */ true, /* not_null_free= */ true, true);
4212
4213 // Because we create the scopedValue cache lazily we have to make the
4214 // type of the result BotPTR.
4215 bool xk = etype->klass_is_exact();
4216 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, TypeAryPtr::Offset(0));
4217 return objects_type;
4218 }
4219
4220 Node* LibraryCallKit::scopedValueCache_helper() {
4221 Node* thread = _gvn.transform(new ThreadLocalNode());
4222 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
4223 // We cannot use immutable_memory() because we might flip onto a
4224 // different carrier thread, at which point we'll need to use that
4225 // carrier thread's cache.
4226 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
4227 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
4228 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
4229 }
4230
4231 //------------------------inline_native_scopedValueCache------------------
4232 bool LibraryCallKit::inline_native_scopedValueCache() {
4233 Node* cache_obj_handle = scopedValueCache_helper();
4234 const Type* objects_type = scopedValueCache_type();
4235 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
4236
4237 return true;
4238 }
4239
4240 //------------------------inline_native_setScopedValueCache------------------
4241 bool LibraryCallKit::inline_native_setScopedValueCache() {
4242 Node* arr = argument(0);
4243 Node* cache_obj_handle = scopedValueCache_helper();
4244 const Type* objects_type = scopedValueCache_type();
4245
4246 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
4247 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
4248
4249 return true;
4250 }
4251
4252 //------------------------inline_native_Continuation_pin and unpin-----------
4253
4254 // Shared implementation routine for both pin and unpin.
4255 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
4256 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
4257
4258 // Save input memory.
4259 Node* input_memory_state = reset_memory();
4260 set_all_memory(input_memory_state);
4261
4262 // TLS
4263 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
4264 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
4265 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
4266
4267 // Null check the last continuation object.
4268 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
4269 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
4270 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
4271
4272 // False path, last continuation is null.
4273 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
4274
4275 // True path, last continuation is not null.
4276 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
4277
4278 set_control(continuation_is_not_null);
4279
4280 // Load the pin count from the last continuation.
4281 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
4282 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
4283
4284 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
4285 Node* pin_count_rhs;
4286 if (unpin) {
4287 pin_count_rhs = _gvn.intcon(0);
4288 } else {
4289 pin_count_rhs = _gvn.intcon(UINT32_MAX);
4290 }
4291 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
4292 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
4293 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
4294
4295 // True branch, pin count over/underflow.
4296 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
4297 {
4298 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
4299 // which will throw IllegalStateException for pin count over/underflow.
4300 // No memory changed so far - we can use memory create by reset_memory()
4301 // at the beginning of this intrinsic. No need to call reset_memory() again.
4302 PreserveJVMState pjvms(this);
4303 set_control(pin_count_over_underflow);
4304 uncommon_trap(Deoptimization::Reason_intrinsic,
4305 Deoptimization::Action_none);
4306 assert(stopped(), "invariant");
4307 }
4308
4309 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
4310 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
4311 set_control(valid_pin_count);
4312
4313 Node* next_pin_count;
4314 if (unpin) {
4315 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
4316 } else {
4317 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
4318 }
4319
4320 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
4321
4322 // Result of top level CFG and Memory.
4323 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
4324 record_for_igvn(result_rgn);
4325 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
4326 record_for_igvn(result_mem);
4327
4328 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
4329 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
4330 result_mem->init_req(_true_path, _gvn.transform(reset_memory()));
4331 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
4332
4333 // Set output state.
4334 set_control(_gvn.transform(result_rgn));
4335 set_all_memory(_gvn.transform(result_mem));
4336
4337 return true;
4338 }
4339
4340 //---------------------------load_mirror_from_klass----------------------------
4341 // Given a klass oop, load its java mirror (a java.lang.Class oop).
4342 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
4343 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
4344 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4345 // mirror = ((OopHandle)mirror)->resolve();
4346 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
4347 }
4348
4349 //-----------------------load_klass_from_mirror_common-------------------------
4350 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
4351 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
4352 // and branch to the given path on the region.
4353 // If never_see_null, take an uncommon trap on null, so we can optimistically
4354 // compile for the non-null case.
4355 // If the region is null, force never_see_null = true.
4356 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
4357 bool never_see_null,
4358 RegionNode* region,
4359 int null_path,
4360 int offset) {
4361 if (region == nullptr) never_see_null = true;
4362 Node* p = basic_plus_adr(mirror, offset);
4363 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4364 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
4365 Node* null_ctl = top();
4366 kls = null_check_oop(kls, &null_ctl, never_see_null);
4367 if (region != nullptr) {
4368 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
4369 region->init_req(null_path, null_ctl);
4370 } else {
4371 assert(null_ctl == top(), "no loose ends");
4372 }
4373 return kls;
4374 }
4375
4376 //--------------------(inline_native_Class_query helpers)---------------------
4377 // Use this for JVM_ACC_INTERFACE.
4378 // Fall through if (mods & mask) == bits, take the guard otherwise.
4379 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
4380 ByteSize offset, const Type* type, BasicType bt) {
4381 // Branch around if the given klass has the given modifier bit set.
4382 // Like generate_guard, adds a new path onto the region.
4383 Node* modp = basic_plus_adr(kls, in_bytes(offset));
4384 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
4385 Node* mask = intcon(modifier_mask);
4386 Node* bits = intcon(modifier_bits);
4387 Node* mbit = _gvn.transform(new AndINode(mods, mask));
4388 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
4389 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
4390 return generate_fair_guard(bol, region);
4391 }
4392
4393 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
4394 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
4395 InstanceKlass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
4396 }
4397
4398 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
4399 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
4400 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
4401 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
4402 }
4403
4404 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
4405 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
4406 }
4407
4408 //-------------------------inline_native_Class_query-------------------
4409 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
4410 const Type* return_type = TypeInt::BOOL;
4411 Node* prim_return_value = top(); // what happens if it's a primitive class?
4412 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4413 bool expect_prim = false; // most of these guys expect to work on refs
4414
4415 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
4416
4417 Node* mirror = argument(0);
4418 Node* obj = top();
4419
4420 switch (id) {
4421 case vmIntrinsics::_isInstance:
4422 // nothing is an instance of a primitive type
4423 prim_return_value = intcon(0);
4424 obj = argument(1);
4425 break;
4426 case vmIntrinsics::_isHidden:
4427 prim_return_value = intcon(0);
4428 break;
4429 case vmIntrinsics::_getSuperclass:
4430 prim_return_value = null();
4431 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
4432 break;
4433 default:
4434 fatal_unexpected_iid(id);
4435 break;
4436 }
4437
4438 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4439 if (mirror_con == nullptr) return false; // cannot happen?
4440
4441 #ifndef PRODUCT
4442 if (C->print_intrinsics() || C->print_inlining()) {
4443 ciType* k = mirror_con->java_mirror_type();
4444 if (k) {
4445 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4446 k->print_name();
4447 tty->cr();
4448 }
4449 }
4450 #endif
4451
4452 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4453 RegionNode* region = new RegionNode(PATH_LIMIT);
4454 record_for_igvn(region);
4455 PhiNode* phi = new PhiNode(region, return_type);
4456
4457 // The mirror will never be null of Reflection.getClassAccessFlags, however
4458 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4459 // if it is. See bug 4774291.
4460
4461 // For Reflection.getClassAccessFlags(), the null check occurs in
4462 // the wrong place; see inline_unsafe_access(), above, for a similar
4463 // situation.
4464 mirror = null_check(mirror);
4465 // If mirror or obj is dead, only null-path is taken.
4466 if (stopped()) return true;
4467
4468 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4469
4470 // Now load the mirror's klass metaobject, and null-check it.
4471 // Side-effects region with the control path if the klass is null.
4472 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4473 // If kls is null, we have a primitive mirror.
4474 phi->init_req(_prim_path, prim_return_value);
4475 if (stopped()) { set_result(region, phi); return true; }
4476 bool safe_for_replace = (region->in(_prim_path) == top());
4477
4478 Node* p; // handy temp
4479 Node* null_ctl;
4480
4481 // Now that we have the non-null klass, we can perform the real query.
4482 // For constant classes, the query will constant-fold in LoadNode::Value.
4483 Node* query_value = top();
4484 switch (id) {
4485 case vmIntrinsics::_isInstance:
4486 // nothing is an instance of a primitive type
4487 query_value = gen_instanceof(obj, kls, safe_for_replace);
4488 break;
4489
4490 case vmIntrinsics::_isHidden:
4491 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4492 if (generate_hidden_class_guard(kls, region) != nullptr)
4493 // A guard was added. If the guard is taken, it was an hidden class.
4494 phi->add_req(intcon(1));
4495 // If we fall through, it's a plain class.
4496 query_value = intcon(0);
4497 break;
4498
4499
4500 case vmIntrinsics::_getSuperclass:
4501 // The rules here are somewhat unfortunate, but we can still do better
4502 // with random logic than with a JNI call.
4503 // Interfaces store null or Object as _super, but must report null.
4504 // Arrays store an intermediate super as _super, but must report Object.
4505 // Other types can report the actual _super.
4506 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4507 if (generate_array_guard(kls, region) != nullptr) {
4508 // A guard was added. If the guard is taken, it was an array.
4509 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4510 }
4511 // Check for interface after array since this checks AccessFlags offset into InstanceKlass.
4512 // In other words, we are accessing subtype-specific information, so we need to determine the subtype first.
4513 if (generate_interface_guard(kls, region) != nullptr) {
4514 // A guard was added. If the guard is taken, it was an interface.
4515 phi->add_req(null());
4516 }
4517 // If we fall through, it's a plain class. Get its _super.
4518 if (!stopped()) {
4519 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4520 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4521 null_ctl = top();
4522 kls = null_check_oop(kls, &null_ctl);
4523 if (null_ctl != top()) {
4524 // If the guard is taken, Object.superClass is null (both klass and mirror).
4525 region->add_req(null_ctl);
4526 phi ->add_req(null());
4527 }
4528 if (!stopped()) {
4529 query_value = load_mirror_from_klass(kls);
4530 }
4531 }
4532 break;
4533
4534 default:
4535 fatal_unexpected_iid(id);
4536 break;
4537 }
4538
4539 // Fall-through is the normal case of a query to a real class.
4540 phi->init_req(1, query_value);
4541 region->init_req(1, control());
4542
4543 C->set_has_split_ifs(true); // Has chance for split-if optimization
4544 set_result(region, phi);
4545 return true;
4546 }
4547
4548
4549 //-------------------------inline_Class_cast-------------------
4550 bool LibraryCallKit::inline_Class_cast() {
4551 Node* mirror = argument(0); // Class
4552 Node* obj = argument(1);
4553 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4554 if (mirror_con == nullptr) {
4555 return false; // dead path (mirror->is_top()).
4556 }
4557 if (obj == nullptr || obj->is_top()) {
4558 return false; // dead path
4559 }
4560 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4561
4562 // First, see if Class.cast() can be folded statically.
4563 // java_mirror_type() returns non-null for compile-time Class constants.
4564 ciType* tm = mirror_con->java_mirror_type();
4565 if (tm != nullptr && tm->is_klass() &&
4566 tp != nullptr) {
4567 if (!tp->is_loaded()) {
4568 // Don't use intrinsic when class is not loaded.
4569 return false;
4570 } else {
4571 const TypeKlassPtr* tklass = TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces);
4572 int static_res = C->static_subtype_check(tklass, tp->as_klass_type());
4573 if (static_res == Compile::SSC_always_true) {
4574 // isInstance() is true - fold the code.
4575 set_result(obj);
4576 return true;
4577 } else if (static_res == Compile::SSC_always_false) {
4578 // Don't use intrinsic, have to throw ClassCastException.
4579 // If the reference is null, the non-intrinsic bytecode will
4580 // be optimized appropriately.
4581 return false;
4582 }
4583 }
4584 }
4585
4586 // Bailout intrinsic and do normal inlining if exception path is frequent.
4587 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4588 return false;
4589 }
4590
4591 // Generate dynamic checks.
4592 // Class.cast() is java implementation of _checkcast bytecode.
4593 // Do checkcast (Parse::do_checkcast()) optimizations here.
4594
4595 mirror = null_check(mirror);
4596 // If mirror is dead, only null-path is taken.
4597 if (stopped()) {
4598 return true;
4599 }
4600
4601 // Not-subtype or the mirror's klass ptr is nullptr (in case it is a primitive).
4602 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
4603 RegionNode* region = new RegionNode(PATH_LIMIT);
4604 record_for_igvn(region);
4605
4606 // Now load the mirror's klass metaobject, and null-check it.
4607 // If kls is null, we have a primitive mirror and
4608 // nothing is an instance of a primitive type.
4609 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4610
4611 Node* res = top();
4612 Node* io = i_o();
4613 Node* mem = merged_memory();
4614 if (!stopped()) {
4615
4616 Node* bad_type_ctrl = top();
4617 // Do checkcast optimizations.
4618 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4619 region->init_req(_bad_type_path, bad_type_ctrl);
4620 }
4621 if (region->in(_prim_path) != top() ||
4622 region->in(_bad_type_path) != top() ||
4623 region->in(_npe_path) != top()) {
4624 // Let Interpreter throw ClassCastException.
4625 PreserveJVMState pjvms(this);
4626 set_control(_gvn.transform(region));
4627 // Set IO and memory because gen_checkcast may override them when buffering inline types
4628 set_i_o(io);
4629 set_all_memory(mem);
4630 uncommon_trap(Deoptimization::Reason_intrinsic,
4631 Deoptimization::Action_maybe_recompile);
4632 }
4633 if (!stopped()) {
4634 set_result(res);
4635 }
4636 return true;
4637 }
4638
4639
4640 //--------------------------inline_native_subtype_check------------------------
4641 // This intrinsic takes the JNI calls out of the heart of
4642 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4643 bool LibraryCallKit::inline_native_subtype_check() {
4644 // Pull both arguments off the stack.
4645 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4646 args[0] = argument(0);
4647 args[1] = argument(1);
4648 Node* klasses[2]; // corresponding Klasses: superk, subk
4649 klasses[0] = klasses[1] = top();
4650
4651 enum {
4652 // A full decision tree on {superc is prim, subc is prim}:
4653 _prim_0_path = 1, // {P,N} => false
4654 // {P,P} & superc!=subc => false
4655 _prim_same_path, // {P,P} & superc==subc => true
4656 _prim_1_path, // {N,P} => false
4657 _ref_subtype_path, // {N,N} & subtype check wins => true
4658 _both_ref_path, // {N,N} & subtype check loses => false
4659 PATH_LIMIT
4660 };
4661
4662 RegionNode* region = new RegionNode(PATH_LIMIT);
4663 RegionNode* prim_region = new RegionNode(2);
4664 Node* phi = new PhiNode(region, TypeInt::BOOL);
4665 record_for_igvn(region);
4666 record_for_igvn(prim_region);
4667
4668 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4669 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4670 int class_klass_offset = java_lang_Class::klass_offset();
4671
4672 // First null-check both mirrors and load each mirror's klass metaobject.
4673 int which_arg;
4674 for (which_arg = 0; which_arg <= 1; which_arg++) {
4675 Node* arg = args[which_arg];
4676 arg = null_check(arg);
4677 if (stopped()) break;
4678 args[which_arg] = arg;
4679
4680 Node* p = basic_plus_adr(arg, class_klass_offset);
4681 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4682 klasses[which_arg] = _gvn.transform(kls);
4683 }
4684
4685 // Having loaded both klasses, test each for null.
4686 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4687 for (which_arg = 0; which_arg <= 1; which_arg++) {
4688 Node* kls = klasses[which_arg];
4689 Node* null_ctl = top();
4690 kls = null_check_oop(kls, &null_ctl, never_see_null);
4691 if (which_arg == 0) {
4692 prim_region->init_req(1, null_ctl);
4693 } else {
4694 region->init_req(_prim_1_path, null_ctl);
4695 }
4696 if (stopped()) break;
4697 klasses[which_arg] = kls;
4698 }
4699
4700 if (!stopped()) {
4701 // now we have two reference types, in klasses[0..1]
4702 Node* subk = klasses[1]; // the argument to isAssignableFrom
4703 Node* superk = klasses[0]; // the receiver
4704 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4705 region->set_req(_ref_subtype_path, control());
4706 }
4707
4708 // If both operands are primitive (both klasses null), then
4709 // we must return true when they are identical primitives.
4710 // It is convenient to test this after the first null klass check.
4711 // This path is also used if superc is a value mirror.
4712 set_control(_gvn.transform(prim_region));
4713 if (!stopped()) {
4714 // Since superc is primitive, make a guard for the superc==subc case.
4715 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4716 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4717 generate_fair_guard(bol_eq, region);
4718 if (region->req() == PATH_LIMIT+1) {
4719 // A guard was added. If the added guard is taken, superc==subc.
4720 region->swap_edges(PATH_LIMIT, _prim_same_path);
4721 region->del_req(PATH_LIMIT);
4722 }
4723 region->set_req(_prim_0_path, control()); // Not equal after all.
4724 }
4725
4726 // these are the only paths that produce 'true':
4727 phi->set_req(_prim_same_path, intcon(1));
4728 phi->set_req(_ref_subtype_path, intcon(1));
4729
4730 // pull together the cases:
4731 assert(region->req() == PATH_LIMIT, "sane region");
4732 for (uint i = 1; i < region->req(); i++) {
4733 Node* ctl = region->in(i);
4734 if (ctl == nullptr || ctl == top()) {
4735 region->set_req(i, top());
4736 phi ->set_req(i, top());
4737 } else if (phi->in(i) == nullptr) {
4738 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4739 }
4740 }
4741
4742 set_control(_gvn.transform(region));
4743 set_result(_gvn.transform(phi));
4744 return true;
4745 }
4746
4747 //---------------------generate_array_guard_common------------------------
4748 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind, Node** obj) {
4749
4750 if (stopped()) {
4751 return nullptr;
4752 }
4753
4754 // Like generate_guard, adds a new path onto the region.
4755 jint layout_con = 0;
4756 Node* layout_val = get_layout_helper(kls, layout_con);
4757 if (layout_val == nullptr) {
4758 bool query = 0;
4759 switch(kind) {
4760 case RefArray: query = Klass::layout_helper_is_refArray(layout_con); break;
4761 case NonRefArray: query = !Klass::layout_helper_is_refArray(layout_con); break;
4762 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
4763 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
4764 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
4765 default:
4766 ShouldNotReachHere();
4767 }
4768 if (!query) {
4769 return nullptr; // never a branch
4770 } else { // always a branch
4771 Node* always_branch = control();
4772 if (region != nullptr)
4773 region->add_req(always_branch);
4774 set_control(top());
4775 return always_branch;
4776 }
4777 }
4778 unsigned int value = 0;
4779 BoolTest::mask btest = BoolTest::illegal;
4780 switch(kind) {
4781 case RefArray:
4782 case NonRefArray: {
4783 value = Klass::_lh_array_tag_ref_value;
4784 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4785 btest = (kind == RefArray) ? BoolTest::eq : BoolTest::ne;
4786 break;
4787 }
4788 case TypeArray: {
4789 value = Klass::_lh_array_tag_type_value;
4790 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4791 btest = BoolTest::eq;
4792 break;
4793 }
4794 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
4795 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
4796 default:
4797 ShouldNotReachHere();
4798 }
4799 // Now test the correct condition.
4800 jint nval = (jint)value;
4801 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4802 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4803 Node* ctrl = generate_fair_guard(bol, region);
4804 Node* is_array_ctrl = kind == NonArray ? control() : ctrl;
4805 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4806 // Keep track of the fact that 'obj' is an array to prevent
4807 // array specific accesses from floating above the guard.
4808 *obj = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4809 }
4810 return ctrl;
4811 }
4812
4813 // public static native Object[] ValueClass::newNullRestrictedAtomicArray(Class<?> componentType, int length, Object initVal);
4814 // public static native Object[] ValueClass::newNullRestrictedNonAtomicArray(Class<?> componentType, int length, Object initVal);
4815 // public static native Object[] ValueClass::newNullableAtomicArray(Class<?> componentType, int length);
4816 bool LibraryCallKit::inline_newArray(bool null_free, bool atomic) {
4817 assert(null_free || atomic, "nullable implies atomic");
4818 Node* componentType = argument(0);
4819 Node* length = argument(1);
4820 Node* init_val = null_free ? argument(2) : nullptr;
4821
4822 const TypeInstPtr* tp = _gvn.type(componentType)->isa_instptr();
4823 if (tp != nullptr) {
4824 ciInstanceKlass* ik = tp->instance_klass();
4825 if (ik == C->env()->Class_klass()) {
4826 ciType* t = tp->java_mirror_type();
4827 if (t != nullptr && t->is_inlinetype()) {
4828
4829 ciArrayKlass* array_klass = ciArrayKlass::make(t, null_free, atomic, true);
4830 assert(array_klass->is_elem_null_free() == null_free, "inconsistency");
4831
4832 // TOOD 8350865 ZGC needs card marks on initializing oop stores
4833 if (UseZGC && null_free && !array_klass->is_flat_array_klass()) {
4834 return false;
4835 }
4836
4837 if (array_klass->is_loaded() && array_klass->element_klass()->as_inline_klass()->is_initialized()) {
4838 const TypeAryKlassPtr* array_klass_type = TypeAryKlassPtr::make(array_klass, Type::trust_interfaces);
4839 if (null_free) {
4840 if (init_val->is_InlineType()) {
4841 if (array_klass_type->is_flat() && init_val->as_InlineType()->is_all_zero(&gvn(), /* flat */ true)) {
4842 // Zeroing is enough because the init value is the all-zero value
4843 init_val = nullptr;
4844 } else {
4845 init_val = init_val->as_InlineType()->buffer(this);
4846 }
4847 }
4848 // TODO 8350865 Should we add a check of the init_val type (maybe in debug only + halt)?
4849 // If we insert a checkcast here, we can be sure that init_val is an InlineTypeNode, so
4850 // when we folded a field load from an allocation (e.g. during escape analysis), we can
4851 // remove the check init_val->is_InlineType().
4852 }
4853 Node* obj = new_array(makecon(array_klass_type), length, 0, nullptr, false, init_val);
4854 const TypeAryPtr* arytype = gvn().type(obj)->is_aryptr();
4855 assert(arytype->is_null_free() == null_free, "inconsistency");
4856 assert(arytype->is_not_null_free() == !null_free, "inconsistency");
4857 set_result(obj);
4858 return true;
4859 }
4860 }
4861 }
4862 }
4863 return false;
4864 }
4865
4866 // public static native boolean ValueClass::isFlatArray(Object array);
4867 // public static native boolean ValueClass::isNullRestrictedArray(Object array);
4868 // public static native boolean ValueClass::isAtomicArray(Object array);
4869 bool LibraryCallKit::inline_getArrayProperties(ArrayPropertiesCheck check) {
4870 Node* array = argument(0);
4871
4872 Node* bol;
4873 switch(check) {
4874 case IsFlat:
4875 // TODO 8350865 Use the object version here instead of loading the klass
4876 // 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
4877 bol = flat_array_test(load_object_klass(array));
4878 break;
4879 case IsNullRestricted:
4880 bol = null_free_array_test(array);
4881 break;
4882 case IsAtomic:
4883 // TODO 8350865 Implement this. It's a bit more complicated, see conditions in JVM_IsAtomicArray
4884 // Enable TestIntrinsics::test87/88 once this is implemented
4885 // bol = null_free_atomic_array_test
4886 return false;
4887 default:
4888 ShouldNotReachHere();
4889 }
4890
4891 Node* res = gvn().transform(new CMoveINode(bol, intcon(0), intcon(1), TypeInt::BOOL));
4892 set_result(res);
4893 return true;
4894 }
4895
4896 // Load the default refined array klass from an ObjArrayKlass. This relies on the first entry in the
4897 // '_next_refined_array_klass' linked list being the default (see ObjArrayKlass::klass_with_properties).
4898 Node* LibraryCallKit::load_default_refined_array_klass(Node* klass_node, bool type_array_guard) {
4899 RegionNode* region = new RegionNode(2);
4900 Node* phi = new PhiNode(region, TypeInstKlassPtr::OBJECT_OR_NULL);
4901
4902 if (type_array_guard) {
4903 generate_typeArray_guard(klass_node, region);
4904 if (region->req() == 3) {
4905 phi->add_req(klass_node);
4906 }
4907 }
4908 Node* adr_refined_klass = basic_plus_adr(klass_node, in_bytes(ObjArrayKlass::next_refined_array_klass_offset()));
4909 Node* refined_klass = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), adr_refined_klass, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4910
4911 // Can be null if not initialized yet, just deopt
4912 Node* null_ctl = top();
4913 refined_klass = null_check_oop(refined_klass, &null_ctl, /* never_see_null= */ true);
4914
4915 region->init_req(1, control());
4916 phi->init_req(1, refined_klass);
4917
4918 set_control(_gvn.transform(region));
4919 return _gvn.transform(phi);
4920 }
4921
4922 // Load the non-refined array klass from an ObjArrayKlass.
4923 Node* LibraryCallKit::load_non_refined_array_klass(Node* klass_node) {
4924 const TypeAryKlassPtr* ary_klass_ptr = _gvn.type(klass_node)->isa_aryklassptr();
4925 if (ary_klass_ptr != nullptr && ary_klass_ptr->klass_is_exact()) {
4926 return _gvn.makecon(ary_klass_ptr->cast_to_refined_array_klass_ptr(false));
4927 }
4928
4929 RegionNode* region = new RegionNode(2);
4930 Node* phi = new PhiNode(region, TypeInstKlassPtr::OBJECT);
4931
4932 generate_typeArray_guard(klass_node, region);
4933 if (region->req() == 3) {
4934 phi->add_req(klass_node);
4935 }
4936 Node* super_adr = basic_plus_adr(klass_node, in_bytes(Klass::super_offset()));
4937 Node* super_klass = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), super_adr, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT));
4938
4939 region->init_req(1, control());
4940 phi->init_req(1, super_klass);
4941
4942 set_control(_gvn.transform(region));
4943 return _gvn.transform(phi);
4944 }
4945
4946 //-----------------------inline_native_newArray--------------------------
4947 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
4948 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4949 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4950 Node* mirror;
4951 Node* count_val;
4952 if (uninitialized) {
4953 null_check_receiver();
4954 mirror = argument(1);
4955 count_val = argument(2);
4956 } else {
4957 mirror = argument(0);
4958 count_val = argument(1);
4959 }
4960
4961 mirror = null_check(mirror);
4962 // If mirror or obj is dead, only null-path is taken.
4963 if (stopped()) return true;
4964
4965 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4966 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4967 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4968 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4969 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4970
4971 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4972 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4973 result_reg, _slow_path);
4974 Node* normal_ctl = control();
4975 Node* no_array_ctl = result_reg->in(_slow_path);
4976
4977 // Generate code for the slow case. We make a call to newArray().
4978 set_control(no_array_ctl);
4979 if (!stopped()) {
4980 // Either the input type is void.class, or else the
4981 // array klass has not yet been cached. Either the
4982 // ensuing call will throw an exception, or else it
4983 // will cache the array klass for next time.
4984 PreserveJVMState pjvms(this);
4985 CallJavaNode* slow_call = nullptr;
4986 if (uninitialized) {
4987 // Generate optimized virtual call (holder class 'Unsafe' is final)
4988 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4989 } else {
4990 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4991 }
4992 Node* slow_result = set_results_for_java_call(slow_call);
4993 // this->control() comes from set_results_for_java_call
4994 result_reg->set_req(_slow_path, control());
4995 result_val->set_req(_slow_path, slow_result);
4996 result_io ->set_req(_slow_path, i_o());
4997 result_mem->set_req(_slow_path, reset_memory());
4998 }
4999
5000 set_control(normal_ctl);
5001 if (!stopped()) {
5002 // Normal case: The array type has been cached in the java.lang.Class.
5003 // The following call works fine even if the array type is polymorphic.
5004 // It could be a dynamic mix of int[], boolean[], Object[], etc.
5005
5006 klass_node = load_default_refined_array_klass(klass_node);
5007
5008 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
5009 result_reg->init_req(_normal_path, control());
5010 result_val->init_req(_normal_path, obj);
5011 result_io ->init_req(_normal_path, i_o());
5012 result_mem->init_req(_normal_path, reset_memory());
5013
5014 if (uninitialized) {
5015 // Mark the allocation so that zeroing is skipped
5016 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
5017 alloc->maybe_set_complete(&_gvn);
5018 }
5019 }
5020
5021 // Return the combined state.
5022 set_i_o( _gvn.transform(result_io) );
5023 set_all_memory( _gvn.transform(result_mem));
5024
5025 C->set_has_split_ifs(true); // Has chance for split-if optimization
5026 set_result(result_reg, result_val);
5027 return true;
5028 }
5029
5030 //----------------------inline_native_getLength--------------------------
5031 // public static native int java.lang.reflect.Array.getLength(Object array);
5032 bool LibraryCallKit::inline_native_getLength() {
5033 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
5034
5035 Node* array = null_check(argument(0));
5036 // If array is dead, only null-path is taken.
5037 if (stopped()) return true;
5038
5039 // Deoptimize if it is a non-array.
5040 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
5041
5042 if (non_array != nullptr) {
5043 PreserveJVMState pjvms(this);
5044 set_control(non_array);
5045 uncommon_trap(Deoptimization::Reason_intrinsic,
5046 Deoptimization::Action_maybe_recompile);
5047 }
5048
5049 // If control is dead, only non-array-path is taken.
5050 if (stopped()) return true;
5051
5052 // The works fine even if the array type is polymorphic.
5053 // It could be a dynamic mix of int[], boolean[], Object[], etc.
5054 Node* result = load_array_length(array);
5055
5056 C->set_has_split_ifs(true); // Has chance for split-if optimization
5057 set_result(result);
5058 return true;
5059 }
5060
5061 //------------------------inline_array_copyOf----------------------------
5062 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
5063 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
5064 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
5065 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
5066
5067 // Get the arguments.
5068 Node* original = argument(0);
5069 Node* start = is_copyOfRange? argument(1): intcon(0);
5070 Node* end = is_copyOfRange? argument(2): argument(1);
5071 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
5072
5073 Node* newcopy = nullptr;
5074
5075 // Set the original stack and the reexecute bit for the interpreter to reexecute
5076 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
5077 { PreserveReexecuteState preexecs(this);
5078 jvms()->set_should_reexecute(true);
5079
5080 array_type_mirror = null_check(array_type_mirror);
5081 original = null_check(original);
5082
5083 // Check if a null path was taken unconditionally.
5084 if (stopped()) return true;
5085
5086 Node* orig_length = load_array_length(original);
5087
5088 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
5089 klass_node = null_check(klass_node);
5090
5091 RegionNode* bailout = new RegionNode(1);
5092 record_for_igvn(bailout);
5093
5094 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
5095 // Bail out if that is so.
5096 // Inline type array may have object field that would require a
5097 // write barrier. Conservatively, go to slow path.
5098 // TODO 8251971: Optimize for the case when flat src/dst are later found
5099 // to not contain oops (i.e., move this check to the macro expansion phase).
5100 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5101 const TypeAryPtr* orig_t = _gvn.type(original)->isa_aryptr();
5102 const TypeKlassPtr* tklass = _gvn.type(klass_node)->is_klassptr();
5103 bool exclude_flat = UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing) &&
5104 // Can src array be flat and contain oops?
5105 (orig_t == nullptr || (!orig_t->is_not_flat() && (!orig_t->is_flat() || orig_t->elem()->inline_klass()->contains_oops()))) &&
5106 // Can dest array be flat and contain oops?
5107 tklass->can_be_inline_array() && (!tklass->is_flat() || tklass->is_aryklassptr()->elem()->is_instklassptr()->instance_klass()->as_inline_klass()->contains_oops());
5108 Node* not_objArray = exclude_flat ? generate_non_refArray_guard(klass_node, bailout) : generate_typeArray_guard(klass_node, bailout);
5109
5110 Node* refined_klass_node = load_default_refined_array_klass(klass_node, /* type_array_guard= */ false);
5111
5112 if (not_objArray != nullptr) {
5113 // Improve the klass node's type from the new optimistic assumption:
5114 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
5115 bool not_flat = !UseArrayFlattening;
5116 bool not_null_free = !Arguments::is_valhalla_enabled();
5117 const Type* akls = TypeAryKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0), Type::trust_interfaces, not_flat, not_null_free, false, false, not_flat, true);
5118 Node* cast = new CastPPNode(control(), refined_klass_node, akls);
5119 refined_klass_node = _gvn.transform(cast);
5120 }
5121
5122 // Bail out if either start or end is negative.
5123 generate_negative_guard(start, bailout, &start);
5124 generate_negative_guard(end, bailout, &end);
5125
5126 Node* length = end;
5127 if (_gvn.type(start) != TypeInt::ZERO) {
5128 length = _gvn.transform(new SubINode(end, start));
5129 }
5130
5131 // Bail out if length is negative (i.e., if start > end).
5132 // Without this the new_array would throw
5133 // NegativeArraySizeException but IllegalArgumentException is what
5134 // should be thrown
5135 generate_negative_guard(length, bailout, &length);
5136
5137 // Handle inline type arrays
5138 bool can_validate = !too_many_traps(Deoptimization::Reason_class_check);
5139 if (!stopped()) {
5140 // TODO 8251971
5141 if (!orig_t->is_null_free()) {
5142 // Not statically known to be null free, add a check
5143 generate_fair_guard(null_free_array_test(original), bailout);
5144 }
5145 orig_t = _gvn.type(original)->isa_aryptr();
5146 if (orig_t != nullptr && orig_t->is_flat()) {
5147 // Src is flat, check that dest is flat as well
5148 if (exclude_flat) {
5149 // Dest can't be flat, bail out
5150 bailout->add_req(control());
5151 set_control(top());
5152 } else {
5153 generate_fair_guard(flat_array_test(refined_klass_node, /* flat = */ false), bailout);
5154 }
5155 // TODO 8350865 This is not correct anymore. Write tests and fix logic similar to arraycopy.
5156 } else if (UseArrayFlattening && (orig_t == nullptr || !orig_t->is_not_flat()) &&
5157 // If dest is flat, src must be flat as well (guaranteed by src <: dest check if validated).
5158 ((!tklass->is_flat() && tklass->can_be_inline_array()) || !can_validate)) {
5159 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
5160 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
5161 generate_fair_guard(flat_array_test(load_object_klass(original)), bailout);
5162 if (orig_t != nullptr) {
5163 orig_t = orig_t->cast_to_not_flat();
5164 original = _gvn.transform(new CheckCastPPNode(control(), original, orig_t));
5165 }
5166 }
5167 if (!can_validate) {
5168 // No validation. The subtype check emitted at macro expansion time will not go to the slow
5169 // path but call checkcast_arraycopy which can not handle flat/null-free inline type arrays.
5170 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat/null-free.
5171 generate_fair_guard(flat_array_test(refined_klass_node), bailout);
5172 generate_fair_guard(null_free_array_test(original), bailout);
5173 }
5174 }
5175
5176 // Bail out if start is larger than the original length
5177 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
5178 generate_negative_guard(orig_tail, bailout, &orig_tail);
5179
5180 if (bailout->req() > 1) {
5181 PreserveJVMState pjvms(this);
5182 set_control(_gvn.transform(bailout));
5183 uncommon_trap(Deoptimization::Reason_intrinsic,
5184 Deoptimization::Action_maybe_recompile);
5185 }
5186
5187 if (!stopped()) {
5188 // How many elements will we copy from the original?
5189 // The answer is MinI(orig_tail, length).
5190 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
5191
5192 // Generate a direct call to the right arraycopy function(s).
5193 // We know the copy is disjoint but we might not know if the
5194 // oop stores need checking.
5195 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
5196 // This will fail a store-check if x contains any non-nulls.
5197
5198 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
5199 // loads/stores but it is legal only if we're sure the
5200 // Arrays.copyOf would succeed. So we need all input arguments
5201 // to the copyOf to be validated, including that the copy to the
5202 // new array won't trigger an ArrayStoreException. That subtype
5203 // check can be optimized if we know something on the type of
5204 // the input array from type speculation.
5205 if (_gvn.type(klass_node)->singleton()) {
5206 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
5207 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
5208
5209 int test = C->static_subtype_check(superk, subk);
5210 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
5211 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
5212 if (t_original->speculative_type() != nullptr) {
5213 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
5214 }
5215 }
5216 }
5217
5218 bool validated = false;
5219 // Reason_class_check rather than Reason_intrinsic because we
5220 // want to intrinsify even if this traps.
5221 if (can_validate) {
5222 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
5223
5224 if (not_subtype_ctrl != top()) {
5225 PreserveJVMState pjvms(this);
5226 set_control(not_subtype_ctrl);
5227 uncommon_trap(Deoptimization::Reason_class_check,
5228 Deoptimization::Action_make_not_entrant);
5229 assert(stopped(), "Should be stopped");
5230 }
5231 validated = true;
5232 }
5233
5234 if (!stopped()) {
5235 newcopy = new_array(refined_klass_node, length, 0); // no arguments to push
5236
5237 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
5238 load_object_klass(original), klass_node);
5239 if (!is_copyOfRange) {
5240 ac->set_copyof(validated);
5241 } else {
5242 ac->set_copyofrange(validated);
5243 }
5244 Node* n = _gvn.transform(ac);
5245 if (n == ac) {
5246 ac->connect_outputs(this);
5247 } else {
5248 assert(validated, "shouldn't transform if all arguments not validated");
5249 set_all_memory(n);
5250 }
5251 }
5252 }
5253 } // original reexecute is set back here
5254
5255 C->set_has_split_ifs(true); // Has chance for split-if optimization
5256 if (!stopped()) {
5257 set_result(newcopy);
5258 }
5259 return true;
5260 }
5261
5262
5263 //----------------------generate_virtual_guard---------------------------
5264 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
5265 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
5266 RegionNode* slow_region) {
5267 ciMethod* method = callee();
5268 int vtable_index = method->vtable_index();
5269 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5270 "bad index %d", vtable_index);
5271 // Get the Method* out of the appropriate vtable entry.
5272 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
5273 vtable_index*vtableEntry::size_in_bytes() +
5274 in_bytes(vtableEntry::method_offset());
5275 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
5276 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
5277
5278 // Compare the target method with the expected method (e.g., Object.hashCode).
5279 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
5280
5281 Node* native_call = makecon(native_call_addr);
5282 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
5283 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
5284
5285 return generate_slow_guard(test_native, slow_region);
5286 }
5287
5288 //-----------------------generate_method_call----------------------------
5289 // Use generate_method_call to make a slow-call to the real
5290 // method if the fast path fails. An alternative would be to
5291 // use a stub like OptoRuntime::slow_arraycopy_Java.
5292 // This only works for expanding the current library call,
5293 // not another intrinsic. (E.g., don't use this for making an
5294 // arraycopy call inside of the copyOf intrinsic.)
5295 CallJavaNode*
5296 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
5297 // When compiling the intrinsic method itself, do not use this technique.
5298 guarantee(callee() != C->method(), "cannot make slow-call to self");
5299
5300 ciMethod* method = callee();
5301 // ensure the JVMS we have will be correct for this call
5302 guarantee(method_id == method->intrinsic_id(), "must match");
5303
5304 const TypeFunc* tf = TypeFunc::make(method);
5305 if (res_not_null) {
5306 assert(tf->return_type() == T_OBJECT, "");
5307 const TypeTuple* range = tf->range_cc();
5308 const Type** fields = TypeTuple::fields(range->cnt());
5309 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
5310 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
5311 tf = TypeFunc::make(tf->domain_cc(), new_range);
5312 }
5313 CallJavaNode* slow_call;
5314 if (is_static) {
5315 assert(!is_virtual, "");
5316 slow_call = new CallStaticJavaNode(C, tf,
5317 SharedRuntime::get_resolve_static_call_stub(), method);
5318 } else if (is_virtual) {
5319 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5320 int vtable_index = Method::invalid_vtable_index;
5321 if (UseInlineCaches) {
5322 // Suppress the vtable call
5323 } else {
5324 // hashCode and clone are not a miranda methods,
5325 // so the vtable index is fixed.
5326 // No need to use the linkResolver to get it.
5327 vtable_index = method->vtable_index();
5328 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5329 "bad index %d", vtable_index);
5330 }
5331 slow_call = new CallDynamicJavaNode(tf,
5332 SharedRuntime::get_resolve_virtual_call_stub(),
5333 method, vtable_index);
5334 } else { // neither virtual nor static: opt_virtual
5335 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5336 slow_call = new CallStaticJavaNode(C, tf,
5337 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
5338 slow_call->set_optimized_virtual(true);
5339 }
5340 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
5341 // To be able to issue a direct call (optimized virtual or virtual)
5342 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
5343 // about the method being invoked should be attached to the call site to
5344 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
5345 slow_call->set_override_symbolic_info(true);
5346 }
5347 set_arguments_for_java_call(slow_call);
5348 set_edges_for_java_call(slow_call);
5349 return slow_call;
5350 }
5351
5352
5353 /**
5354 * Build special case code for calls to hashCode on an object. This call may
5355 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
5356 * slightly different code.
5357 */
5358 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
5359 assert(is_static == callee()->is_static(), "correct intrinsic selection");
5360 assert(!(is_virtual && is_static), "either virtual, special, or static");
5361
5362 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
5363
5364 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5365 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
5366 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
5367 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5368 Node* obj = argument(0);
5369
5370 // Don't intrinsify hashcode on inline types for now.
5371 // The "is locked" runtime check also subsumes the inline type check (as inline types cannot be locked) and goes to the slow path.
5372 if (gvn().type(obj)->is_inlinetypeptr()) {
5373 return false;
5374 }
5375
5376 if (!is_static) {
5377 // Check for hashing null object
5378 obj = null_check_receiver();
5379 if (stopped()) return true; // unconditionally null
5380 result_reg->init_req(_null_path, top());
5381 result_val->init_req(_null_path, top());
5382 } else {
5383 // Do a null check, and return zero if null.
5384 // System.identityHashCode(null) == 0
5385 Node* null_ctl = top();
5386 obj = null_check_oop(obj, &null_ctl);
5387 result_reg->init_req(_null_path, null_ctl);
5388 result_val->init_req(_null_path, _gvn.intcon(0));
5389 }
5390
5391 // Unconditionally null? Then return right away.
5392 if (stopped()) {
5393 set_control( result_reg->in(_null_path));
5394 if (!stopped())
5395 set_result(result_val->in(_null_path));
5396 return true;
5397 }
5398
5399 // We only go to the fast case code if we pass a number of guards. The
5400 // paths which do not pass are accumulated in the slow_region.
5401 RegionNode* slow_region = new RegionNode(1);
5402 record_for_igvn(slow_region);
5403
5404 // If this is a virtual call, we generate a funny guard. We pull out
5405 // the vtable entry corresponding to hashCode() from the target object.
5406 // If the target method which we are calling happens to be the native
5407 // Object hashCode() method, we pass the guard. We do not need this
5408 // guard for non-virtual calls -- the caller is known to be the native
5409 // Object hashCode().
5410 if (is_virtual) {
5411 // After null check, get the object's klass.
5412 Node* obj_klass = load_object_klass(obj);
5413 generate_virtual_guard(obj_klass, slow_region);
5414 }
5415
5416 // Get the header out of the object, use LoadMarkNode when available
5417 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
5418 // The control of the load must be null. Otherwise, the load can move before
5419 // the null check after castPP removal.
5420 Node* no_ctrl = nullptr;
5421 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
5422
5423 if (!UseObjectMonitorTable) {
5424 // Test the header to see if it is safe to read w.r.t. locking.
5425 // We cannot use the inline type mask as this may check bits that are overriden
5426 // by an object monitor's pointer when inflating locking.
5427 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
5428 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
5429 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
5430 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
5431 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
5432
5433 generate_slow_guard(test_monitor, slow_region);
5434 }
5435
5436 // Get the hash value and check to see that it has been properly assigned.
5437 // We depend on hash_mask being at most 32 bits and avoid the use of
5438 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
5439 // vm: see markWord.hpp.
5440 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
5441 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
5442 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
5443 // This hack lets the hash bits live anywhere in the mark object now, as long
5444 // as the shift drops the relevant bits into the low 32 bits. Note that
5445 // Java spec says that HashCode is an int so there's no point in capturing
5446 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
5447 hshifted_header = ConvX2I(hshifted_header);
5448 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
5449
5450 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
5451 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
5452 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
5453
5454 generate_slow_guard(test_assigned, slow_region);
5455
5456 Node* init_mem = reset_memory();
5457 // fill in the rest of the null path:
5458 result_io ->init_req(_null_path, i_o());
5459 result_mem->init_req(_null_path, init_mem);
5460
5461 result_val->init_req(_fast_path, hash_val);
5462 result_reg->init_req(_fast_path, control());
5463 result_io ->init_req(_fast_path, i_o());
5464 result_mem->init_req(_fast_path, init_mem);
5465
5466 // Generate code for the slow case. We make a call to hashCode().
5467 set_control(_gvn.transform(slow_region));
5468 if (!stopped()) {
5469 // No need for PreserveJVMState, because we're using up the present state.
5470 set_all_memory(init_mem);
5471 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
5472 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
5473 Node* slow_result = set_results_for_java_call(slow_call);
5474 // this->control() comes from set_results_for_java_call
5475 result_reg->init_req(_slow_path, control());
5476 result_val->init_req(_slow_path, slow_result);
5477 result_io ->set_req(_slow_path, i_o());
5478 result_mem ->set_req(_slow_path, reset_memory());
5479 }
5480
5481 // Return the combined state.
5482 set_i_o( _gvn.transform(result_io) );
5483 set_all_memory( _gvn.transform(result_mem));
5484
5485 set_result(result_reg, result_val);
5486 return true;
5487 }
5488
5489 //---------------------------inline_native_getClass----------------------------
5490 // public final native Class<?> java.lang.Object.getClass();
5491 //
5492 // Build special case code for calls to getClass on an object.
5493 bool LibraryCallKit::inline_native_getClass() {
5494 Node* obj = argument(0);
5495 if (obj->is_InlineType()) {
5496 const Type* t = _gvn.type(obj);
5497 if (t->maybe_null()) {
5498 null_check(obj);
5499 }
5500 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror())));
5501 return true;
5502 }
5503 obj = null_check_receiver();
5504 if (stopped()) return true;
5505 set_result(load_mirror_from_klass(load_object_klass(obj)));
5506 return true;
5507 }
5508
5509 //-----------------inline_native_Reflection_getCallerClass---------------------
5510 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
5511 //
5512 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
5513 //
5514 // NOTE: This code must perform the same logic as JVM_GetCallerClass
5515 // in that it must skip particular security frames and checks for
5516 // caller sensitive methods.
5517 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
5518 #ifndef PRODUCT
5519 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5520 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
5521 }
5522 #endif
5523
5524 if (!jvms()->has_method()) {
5525 #ifndef PRODUCT
5526 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5527 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
5528 }
5529 #endif
5530 return false;
5531 }
5532
5533 // Walk back up the JVM state to find the caller at the required
5534 // depth.
5535 JVMState* caller_jvms = jvms();
5536
5537 // Cf. JVM_GetCallerClass
5538 // NOTE: Start the loop at depth 1 because the current JVM state does
5539 // not include the Reflection.getCallerClass() frame.
5540 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
5541 ciMethod* m = caller_jvms->method();
5542 switch (n) {
5543 case 0:
5544 fatal("current JVM state does not include the Reflection.getCallerClass frame");
5545 break;
5546 case 1:
5547 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
5548 if (!m->caller_sensitive()) {
5549 #ifndef PRODUCT
5550 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5551 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
5552 }
5553 #endif
5554 return false; // bail-out; let JVM_GetCallerClass do the work
5555 }
5556 break;
5557 default:
5558 if (!m->is_ignored_by_security_stack_walk()) {
5559 // We have reached the desired frame; return the holder class.
5560 // Acquire method holder as java.lang.Class and push as constant.
5561 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
5562 ciInstance* caller_mirror = caller_klass->java_mirror();
5563 set_result(makecon(TypeInstPtr::make(caller_mirror)));
5564
5565 #ifndef PRODUCT
5566 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5567 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());
5568 tty->print_cr(" JVM state at this point:");
5569 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5570 ciMethod* m = jvms()->of_depth(i)->method();
5571 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5572 }
5573 }
5574 #endif
5575 return true;
5576 }
5577 break;
5578 }
5579 }
5580
5581 #ifndef PRODUCT
5582 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5583 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
5584 tty->print_cr(" JVM state at this point:");
5585 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5586 ciMethod* m = jvms()->of_depth(i)->method();
5587 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5588 }
5589 }
5590 #endif
5591
5592 return false; // bail-out; let JVM_GetCallerClass do the work
5593 }
5594
5595 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
5596 Node* arg = argument(0);
5597 Node* result = nullptr;
5598
5599 switch (id) {
5600 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
5601 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
5602 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5603 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5604 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5605 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5606
5607 case vmIntrinsics::_doubleToLongBits: {
5608 // two paths (plus control) merge in a wood
5609 RegionNode *r = new RegionNode(3);
5610 Node *phi = new PhiNode(r, TypeLong::LONG);
5611
5612 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5613 // Build the boolean node
5614 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5615
5616 // Branch either way.
5617 // NaN case is less traveled, which makes all the difference.
5618 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5619 Node *opt_isnan = _gvn.transform(ifisnan);
5620 assert( opt_isnan->is_If(), "Expect an IfNode");
5621 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5622 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5623
5624 set_control(iftrue);
5625
5626 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5627 Node *slow_result = longcon(nan_bits); // return NaN
5628 phi->init_req(1, _gvn.transform( slow_result ));
5629 r->init_req(1, iftrue);
5630
5631 // Else fall through
5632 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5633 set_control(iffalse);
5634
5635 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5636 r->init_req(2, iffalse);
5637
5638 // Post merge
5639 set_control(_gvn.transform(r));
5640 record_for_igvn(r);
5641
5642 C->set_has_split_ifs(true); // Has chance for split-if optimization
5643 result = phi;
5644 assert(result->bottom_type()->isa_long(), "must be");
5645 break;
5646 }
5647
5648 case vmIntrinsics::_floatToIntBits: {
5649 // two paths (plus control) merge in a wood
5650 RegionNode *r = new RegionNode(3);
5651 Node *phi = new PhiNode(r, TypeInt::INT);
5652
5653 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5654 // Build the boolean node
5655 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5656
5657 // Branch either way.
5658 // NaN case is less traveled, which makes all the difference.
5659 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5660 Node *opt_isnan = _gvn.transform(ifisnan);
5661 assert( opt_isnan->is_If(), "Expect an IfNode");
5662 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5663 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5664
5665 set_control(iftrue);
5666
5667 static const jint nan_bits = 0x7fc00000;
5668 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5669 phi->init_req(1, _gvn.transform( slow_result ));
5670 r->init_req(1, iftrue);
5671
5672 // Else fall through
5673 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5674 set_control(iffalse);
5675
5676 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5677 r->init_req(2, iffalse);
5678
5679 // Post merge
5680 set_control(_gvn.transform(r));
5681 record_for_igvn(r);
5682
5683 C->set_has_split_ifs(true); // Has chance for split-if optimization
5684 result = phi;
5685 assert(result->bottom_type()->isa_int(), "must be");
5686 break;
5687 }
5688
5689 default:
5690 fatal_unexpected_iid(id);
5691 break;
5692 }
5693 set_result(_gvn.transform(result));
5694 return true;
5695 }
5696
5697 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5698 Node* arg = argument(0);
5699 Node* result = nullptr;
5700
5701 switch (id) {
5702 case vmIntrinsics::_floatIsInfinite:
5703 result = new IsInfiniteFNode(arg);
5704 break;
5705 case vmIntrinsics::_floatIsFinite:
5706 result = new IsFiniteFNode(arg);
5707 break;
5708 case vmIntrinsics::_doubleIsInfinite:
5709 result = new IsInfiniteDNode(arg);
5710 break;
5711 case vmIntrinsics::_doubleIsFinite:
5712 result = new IsFiniteDNode(arg);
5713 break;
5714 default:
5715 fatal_unexpected_iid(id);
5716 break;
5717 }
5718 set_result(_gvn.transform(result));
5719 return true;
5720 }
5721
5722 //----------------------inline_unsafe_copyMemory-------------------------
5723 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5724
5725 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5726 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5727 const Type* base_t = gvn.type(base);
5728
5729 bool in_native = (base_t == TypePtr::NULL_PTR);
5730 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5731 bool is_mixed = !in_heap && !in_native;
5732
5733 if (is_mixed) {
5734 return true; // mixed accesses can touch both on-heap and off-heap memory
5735 }
5736 if (in_heap) {
5737 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5738 if (!is_prim_array) {
5739 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5740 // there's not enough type information available to determine proper memory slice for it.
5741 return true;
5742 }
5743 }
5744 return false;
5745 }
5746
5747 bool LibraryCallKit::inline_unsafe_copyMemory() {
5748 if (callee()->is_static()) return false; // caller must have the capability!
5749 null_check_receiver(); // null-check receiver
5750 if (stopped()) return true;
5751
5752 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5753
5754 Node* src_base = argument(1); // type: oop
5755 Node* src_off = ConvL2X(argument(2)); // type: long
5756 Node* dst_base = argument(4); // type: oop
5757 Node* dst_off = ConvL2X(argument(5)); // type: long
5758 Node* size = ConvL2X(argument(7)); // type: long
5759
5760 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5761 "fieldOffset must be byte-scaled");
5762
5763 Node* src_addr = make_unsafe_address(src_base, src_off);
5764 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5765
5766 Node* thread = _gvn.transform(new ThreadLocalNode());
5767 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5768 BasicType doing_unsafe_access_bt = T_BYTE;
5769 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5770
5771 // update volatile field
5772 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5773
5774 int flags = RC_LEAF | RC_NO_FP;
5775
5776 const TypePtr* dst_type = TypePtr::BOTTOM;
5777
5778 // Adjust memory effects of the runtime call based on input values.
5779 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5780 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5781 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5782
5783 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5784 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5785 flags |= RC_NARROW_MEM; // narrow in memory
5786 }
5787 }
5788
5789 // Call it. Note that the length argument is not scaled.
5790 make_runtime_call(flags,
5791 OptoRuntime::fast_arraycopy_Type(),
5792 StubRoutines::unsafe_arraycopy(),
5793 "unsafe_arraycopy",
5794 dst_type,
5795 src_addr, dst_addr, size XTOP);
5796
5797 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5798
5799 return true;
5800 }
5801
5802 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5803 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5804 bool LibraryCallKit::inline_unsafe_setMemory() {
5805 if (callee()->is_static()) return false; // caller must have the capability!
5806 null_check_receiver(); // null-check receiver
5807 if (stopped()) return true;
5808
5809 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5810
5811 Node* dst_base = argument(1); // type: oop
5812 Node* dst_off = ConvL2X(argument(2)); // type: long
5813 Node* size = ConvL2X(argument(4)); // type: long
5814 Node* byte = argument(6); // type: byte
5815
5816 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5817 "fieldOffset must be byte-scaled");
5818
5819 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5820
5821 Node* thread = _gvn.transform(new ThreadLocalNode());
5822 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5823 BasicType doing_unsafe_access_bt = T_BYTE;
5824 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5825
5826 // update volatile field
5827 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5828
5829 int flags = RC_LEAF | RC_NO_FP;
5830
5831 const TypePtr* dst_type = TypePtr::BOTTOM;
5832
5833 // Adjust memory effects of the runtime call based on input values.
5834 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5835 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5836
5837 flags |= RC_NARROW_MEM; // narrow in memory
5838 }
5839
5840 // Call it. Note that the length argument is not scaled.
5841 make_runtime_call(flags,
5842 OptoRuntime::unsafe_setmemory_Type(),
5843 StubRoutines::unsafe_setmemory(),
5844 "unsafe_setmemory",
5845 dst_type,
5846 dst_addr, size XTOP, byte);
5847
5848 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5849
5850 return true;
5851 }
5852
5853 #undef XTOP
5854
5855 //------------------------clone_coping-----------------------------------
5856 // Helper function for inline_native_clone.
5857 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5858 assert(obj_size != nullptr, "");
5859 Node* raw_obj = alloc_obj->in(1);
5860 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5861
5862 AllocateNode* alloc = nullptr;
5863 if (ReduceBulkZeroing &&
5864 // If we are implementing an array clone without knowing its source type
5865 // (can happen when compiling the array-guarded branch of a reflective
5866 // Object.clone() invocation), initialize the array within the allocation.
5867 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5868 // to a runtime clone call that assumes fully initialized source arrays.
5869 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5870 // We will be completely responsible for initializing this object -
5871 // mark Initialize node as complete.
5872 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5873 // The object was just allocated - there should be no any stores!
5874 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5875 // Mark as complete_with_arraycopy so that on AllocateNode
5876 // expansion, we know this AllocateNode is initialized by an array
5877 // copy and a StoreStore barrier exists after the array copy.
5878 alloc->initialization()->set_complete_with_arraycopy();
5879 }
5880
5881 Node* size = _gvn.transform(obj_size);
5882 access_clone(obj, alloc_obj, size, is_array);
5883
5884 // Do not let reads from the cloned object float above the arraycopy.
5885 if (alloc != nullptr) {
5886 // Do not let stores that initialize this object be reordered with
5887 // a subsequent store that would make this object accessible by
5888 // other threads.
5889 // Record what AllocateNode this StoreStore protects so that
5890 // escape analysis can go from the MemBarStoreStoreNode to the
5891 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5892 // based on the escape status of the AllocateNode.
5893 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5894 } else {
5895 insert_mem_bar(Op_MemBarCPUOrder);
5896 }
5897 }
5898
5899 //------------------------inline_native_clone----------------------------
5900 // protected native Object java.lang.Object.clone();
5901 //
5902 // Here are the simple edge cases:
5903 // null receiver => normal trap
5904 // virtual and clone was overridden => slow path to out-of-line clone
5905 // not cloneable or finalizer => slow path to out-of-line Object.clone
5906 //
5907 // The general case has two steps, allocation and copying.
5908 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5909 //
5910 // Copying also has two cases, oop arrays and everything else.
5911 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5912 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5913 //
5914 // These steps fold up nicely if and when the cloned object's klass
5915 // can be sharply typed as an object array, a type array, or an instance.
5916 //
5917 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5918 PhiNode* result_val;
5919
5920 // Set the reexecute bit for the interpreter to reexecute
5921 // the bytecode that invokes Object.clone if deoptimization happens.
5922 { PreserveReexecuteState preexecs(this);
5923 jvms()->set_should_reexecute(true);
5924
5925 Node* obj = argument(0);
5926 obj = null_check_receiver();
5927 if (stopped()) return true;
5928
5929 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5930 if (obj_type->is_inlinetypeptr()) {
5931 // If the object to clone is an inline type, we can simply return it (i.e. a nop) since inline types have
5932 // no identity.
5933 set_result(obj);
5934 return true;
5935 }
5936
5937 // If we are going to clone an instance, we need its exact type to
5938 // know the number and types of fields to convert the clone to
5939 // loads/stores. Maybe a speculative type can help us.
5940 if (!obj_type->klass_is_exact() &&
5941 obj_type->speculative_type() != nullptr &&
5942 obj_type->speculative_type()->is_instance_klass() &&
5943 !obj_type->speculative_type()->is_inlinetype()) {
5944 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5945 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5946 !spec_ik->has_injected_fields()) {
5947 if (!obj_type->isa_instptr() ||
5948 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5949 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5950 }
5951 }
5952 }
5953
5954 // Conservatively insert a memory barrier on all memory slices.
5955 // Do not let writes into the original float below the clone.
5956 insert_mem_bar(Op_MemBarCPUOrder);
5957
5958 // paths into result_reg:
5959 enum {
5960 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5961 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5962 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5963 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5964 PATH_LIMIT
5965 };
5966 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5967 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5968 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5969 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5970 record_for_igvn(result_reg);
5971
5972 Node* obj_klass = load_object_klass(obj);
5973 // We only go to the fast case code if we pass a number of guards.
5974 // The paths which do not pass are accumulated in the slow_region.
5975 RegionNode* slow_region = new RegionNode(1);
5976 record_for_igvn(slow_region);
5977
5978 Node* array_obj = obj;
5979 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5980 if (array_ctl != nullptr) {
5981 // It's an array.
5982 PreserveJVMState pjvms(this);
5983 set_control(array_ctl);
5984
5985 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5986 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr();
5987 if (UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) &&
5988 obj_type->can_be_inline_array() &&
5989 (ary_ptr == nullptr || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) {
5990 // Flat inline type array may have object field that would require a
5991 // write barrier. Conservatively, go to slow path.
5992 generate_fair_guard(flat_array_test(obj_klass), slow_region);
5993 }
5994
5995 if (!stopped()) {
5996 Node* obj_length = load_array_length(array_obj);
5997 Node* array_size = nullptr; // Size of the array without object alignment padding.
5998 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5999
6000 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
6001 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
6002 // If it is an oop array, it requires very special treatment,
6003 // because gc barriers are required when accessing the array.
6004 Node* is_obja = generate_refArray_guard(obj_klass, (RegionNode*)nullptr);
6005 if (is_obja != nullptr) {
6006 PreserveJVMState pjvms2(this);
6007 set_control(is_obja);
6008 // Generate a direct call to the right arraycopy function(s).
6009 // Clones are always tightly coupled.
6010 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
6011 ac->set_clone_oop_array();
6012 Node* n = _gvn.transform(ac);
6013 assert(n == ac, "cannot disappear");
6014 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
6015
6016 result_reg->init_req(_objArray_path, control());
6017 result_val->init_req(_objArray_path, alloc_obj);
6018 result_i_o ->set_req(_objArray_path, i_o());
6019 result_mem ->set_req(_objArray_path, reset_memory());
6020 }
6021 }
6022 // Otherwise, there are no barriers to worry about.
6023 // (We can dispense with card marks if we know the allocation
6024 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
6025 // causes the non-eden paths to take compensating steps to
6026 // simulate a fresh allocation, so that no further
6027 // card marks are required in compiled code to initialize
6028 // the object.)
6029
6030 if (!stopped()) {
6031 copy_to_clone(obj, alloc_obj, array_size, true);
6032
6033 // Present the results of the copy.
6034 result_reg->init_req(_array_path, control());
6035 result_val->init_req(_array_path, alloc_obj);
6036 result_i_o ->set_req(_array_path, i_o());
6037 result_mem ->set_req(_array_path, reset_memory());
6038 }
6039 }
6040 }
6041
6042 if (!stopped()) {
6043 // It's an instance (we did array above). Make the slow-path tests.
6044 // If this is a virtual call, we generate a funny guard. We grab
6045 // the vtable entry corresponding to clone() from the target object.
6046 // If the target method which we are calling happens to be the
6047 // Object clone() method, we pass the guard. We do not need this
6048 // guard for non-virtual calls; the caller is known to be the native
6049 // Object clone().
6050 if (is_virtual) {
6051 generate_virtual_guard(obj_klass, slow_region);
6052 }
6053
6054 // The object must be easily cloneable and must not have a finalizer.
6055 // Both of these conditions may be checked in a single test.
6056 // We could optimize the test further, but we don't care.
6057 generate_misc_flags_guard(obj_klass,
6058 // Test both conditions:
6059 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
6060 // Must be cloneable but not finalizer:
6061 KlassFlags::_misc_is_cloneable_fast,
6062 slow_region);
6063 }
6064
6065 if (!stopped()) {
6066 // It's an instance, and it passed the slow-path tests.
6067 PreserveJVMState pjvms(this);
6068 Node* obj_size = nullptr; // Total object size, including object alignment padding.
6069 // Need to deoptimize on exception from allocation since Object.clone intrinsic
6070 // is reexecuted if deoptimization occurs and there could be problems when merging
6071 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
6072 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
6073
6074 copy_to_clone(obj, alloc_obj, obj_size, false);
6075
6076 // Present the results of the slow call.
6077 result_reg->init_req(_instance_path, control());
6078 result_val->init_req(_instance_path, alloc_obj);
6079 result_i_o ->set_req(_instance_path, i_o());
6080 result_mem ->set_req(_instance_path, reset_memory());
6081 }
6082
6083 // Generate code for the slow case. We make a call to clone().
6084 set_control(_gvn.transform(slow_region));
6085 if (!stopped()) {
6086 PreserveJVMState pjvms(this);
6087 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
6088 // We need to deoptimize on exception (see comment above)
6089 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
6090 // this->control() comes from set_results_for_java_call
6091 result_reg->init_req(_slow_path, control());
6092 result_val->init_req(_slow_path, slow_result);
6093 result_i_o ->set_req(_slow_path, i_o());
6094 result_mem ->set_req(_slow_path, reset_memory());
6095 }
6096
6097 // Return the combined state.
6098 set_control( _gvn.transform(result_reg));
6099 set_i_o( _gvn.transform(result_i_o));
6100 set_all_memory( _gvn.transform(result_mem));
6101 } // original reexecute is set back here
6102
6103 set_result(_gvn.transform(result_val));
6104 return true;
6105 }
6106
6107 // If we have a tightly coupled allocation, the arraycopy may take care
6108 // of the array initialization. If one of the guards we insert between
6109 // the allocation and the arraycopy causes a deoptimization, an
6110 // uninitialized array will escape the compiled method. To prevent that
6111 // we set the JVM state for uncommon traps between the allocation and
6112 // the arraycopy to the state before the allocation so, in case of
6113 // deoptimization, we'll reexecute the allocation and the
6114 // initialization.
6115 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
6116 if (alloc != nullptr) {
6117 ciMethod* trap_method = alloc->jvms()->method();
6118 int trap_bci = alloc->jvms()->bci();
6119
6120 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6121 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
6122 // Make sure there's no store between the allocation and the
6123 // arraycopy otherwise visible side effects could be rexecuted
6124 // in case of deoptimization and cause incorrect execution.
6125 bool no_interfering_store = true;
6126 Node* mem = alloc->in(TypeFunc::Memory);
6127 if (mem->is_MergeMem()) {
6128 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
6129 Node* n = mms.memory();
6130 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
6131 assert(n->is_Store(), "what else?");
6132 no_interfering_store = false;
6133 break;
6134 }
6135 }
6136 } else {
6137 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
6138 Node* n = mms.memory();
6139 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
6140 assert(n->is_Store(), "what else?");
6141 no_interfering_store = false;
6142 break;
6143 }
6144 }
6145 }
6146
6147 if (no_interfering_store) {
6148 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6149
6150 JVMState* saved_jvms = jvms();
6151 saved_reexecute_sp = _reexecute_sp;
6152
6153 set_jvms(sfpt->jvms());
6154 _reexecute_sp = jvms()->sp();
6155
6156 return saved_jvms;
6157 }
6158 }
6159 }
6160 return nullptr;
6161 }
6162
6163 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
6164 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
6165 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
6166 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
6167 uint size = alloc->req();
6168 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
6169 old_jvms->set_map(sfpt);
6170 for (uint i = 0; i < size; i++) {
6171 sfpt->init_req(i, alloc->in(i));
6172 }
6173 int adjustment = 1;
6174 const TypeAryKlassPtr* ary_klass_ptr = alloc->in(AllocateNode::KlassNode)->bottom_type()->is_aryklassptr();
6175 if (ary_klass_ptr->is_null_free()) {
6176 // A null-free, tightly coupled array allocation can only come from LibraryCallKit::inline_newArray which
6177 // also requires the componentType and initVal on stack for re-execution.
6178 // Re-create and push the componentType.
6179 ciArrayKlass* klass = ary_klass_ptr->exact_klass()->as_array_klass();
6180 ciInstance* instance = klass->component_mirror_instance();
6181 const TypeInstPtr* t_instance = TypeInstPtr::make(instance);
6182 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), makecon(t_instance));
6183 adjustment++;
6184 }
6185 // re-push array length for deoptimization
6186 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment - 1, alloc->in(AllocateNode::ALength));
6187 if (ary_klass_ptr->is_null_free()) {
6188 // Re-create and push the initVal.
6189 Node* init_val = alloc->in(AllocateNode::InitValue);
6190 if (init_val == nullptr) {
6191 init_val = InlineTypeNode::make_all_zero(_gvn, ary_klass_ptr->elem()->is_instklassptr()->instance_klass()->as_inline_klass());
6192 } else if (UseCompressedOops) {
6193 init_val = _gvn.transform(new DecodeNNode(init_val, init_val->bottom_type()->make_ptr()));
6194 }
6195 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment, init_val);
6196 adjustment++;
6197 }
6198 old_jvms->set_sp(old_jvms->sp() + adjustment);
6199 old_jvms->set_monoff(old_jvms->monoff() + adjustment);
6200 old_jvms->set_scloff(old_jvms->scloff() + adjustment);
6201 old_jvms->set_endoff(old_jvms->endoff() + adjustment);
6202 old_jvms->set_should_reexecute(true);
6203
6204 sfpt->set_i_o(map()->i_o());
6205 sfpt->set_memory(map()->memory());
6206 sfpt->set_control(map()->control());
6207 return sfpt;
6208 }
6209
6210 // In case of a deoptimization, we restart execution at the
6211 // allocation, allocating a new array. We would leave an uninitialized
6212 // array in the heap that GCs wouldn't expect. Move the allocation
6213 // after the traps so we don't allocate the array if we
6214 // deoptimize. This is possible because tightly_coupled_allocation()
6215 // guarantees there's no observer of the allocated array at this point
6216 // and the control flow is simple enough.
6217 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
6218 int saved_reexecute_sp, uint new_idx) {
6219 if (saved_jvms_before_guards != nullptr && !stopped()) {
6220 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
6221
6222 assert(alloc != nullptr, "only with a tightly coupled allocation");
6223 // restore JVM state to the state at the arraycopy
6224 saved_jvms_before_guards->map()->set_control(map()->control());
6225 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
6226 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
6227 // If we've improved the types of some nodes (null check) while
6228 // emitting the guards, propagate them to the current state
6229 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
6230 set_jvms(saved_jvms_before_guards);
6231 _reexecute_sp = saved_reexecute_sp;
6232
6233 // Remove the allocation from above the guards
6234 CallProjections* callprojs = alloc->extract_projections(true);
6235 InitializeNode* init = alloc->initialization();
6236 Node* alloc_mem = alloc->in(TypeFunc::Memory);
6237 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
6238 init->replace_mem_projs_by(alloc_mem, C);
6239
6240 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
6241 // the allocation (i.e. is only valid if the allocation succeeds):
6242 // 1) replace CastIINode with AllocateArrayNode's length here
6243 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
6244 //
6245 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
6246 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
6247 Node* init_control = init->proj_out(TypeFunc::Control);
6248 Node* alloc_length = alloc->Ideal_length();
6249 #ifdef ASSERT
6250 Node* prev_cast = nullptr;
6251 #endif
6252 for (uint i = 0; i < init_control->outcnt(); i++) {
6253 Node* init_out = init_control->raw_out(i);
6254 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
6255 #ifdef ASSERT
6256 if (prev_cast == nullptr) {
6257 prev_cast = init_out;
6258 } else {
6259 if (prev_cast->cmp(*init_out) == false) {
6260 prev_cast->dump();
6261 init_out->dump();
6262 assert(false, "not equal CastIINode");
6263 }
6264 }
6265 #endif
6266 C->gvn_replace_by(init_out, alloc_length);
6267 }
6268 }
6269 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
6270
6271 // move the allocation here (after the guards)
6272 _gvn.hash_delete(alloc);
6273 alloc->set_req(TypeFunc::Control, control());
6274 alloc->set_req(TypeFunc::I_O, i_o());
6275 Node *mem = reset_memory();
6276 set_all_memory(mem);
6277 alloc->set_req(TypeFunc::Memory, mem);
6278 set_control(init->proj_out_or_null(TypeFunc::Control));
6279 set_i_o(callprojs->fallthrough_ioproj);
6280
6281 // Update memory as done in GraphKit::set_output_for_allocation()
6282 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
6283 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
6284 if (ary_type->isa_aryptr() && length_type != nullptr) {
6285 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
6286 }
6287 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
6288 int elemidx = C->get_alias_index(telemref);
6289 // Need to properly move every memory projection for the Initialize
6290 #ifdef ASSERT
6291 int mark_idx = C->get_alias_index(ary_type->add_offset(oopDesc::mark_offset_in_bytes()));
6292 int klass_idx = C->get_alias_index(ary_type->add_offset(oopDesc::klass_offset_in_bytes()));
6293 #endif
6294 auto move_proj = [&](ProjNode* proj) {
6295 int alias_idx = C->get_alias_index(proj->adr_type());
6296 assert(alias_idx == Compile::AliasIdxRaw ||
6297 alias_idx == elemidx ||
6298 alias_idx == mark_idx ||
6299 alias_idx == klass_idx, "should be raw memory or array element type");
6300 set_memory(proj, alias_idx);
6301 };
6302 init->for_each_proj(move_proj, TypeFunc::Memory);
6303
6304 Node* allocx = _gvn.transform(alloc);
6305 assert(allocx == alloc, "where has the allocation gone?");
6306 assert(dest->is_CheckCastPP(), "not an allocation result?");
6307
6308 _gvn.hash_delete(dest);
6309 dest->set_req(0, control());
6310 Node* destx = _gvn.transform(dest);
6311 assert(destx == dest, "where has the allocation result gone?");
6312
6313 array_ideal_length(alloc, ary_type, true);
6314 }
6315 }
6316
6317 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
6318 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
6319 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
6320 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
6321 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
6322 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
6323 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
6324 JVMState* saved_jvms_before_guards) {
6325 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
6326 // There is at least one unrelated uncommon trap which needs to be replaced.
6327 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6328
6329 JVMState* saved_jvms = jvms();
6330 const int saved_reexecute_sp = _reexecute_sp;
6331 set_jvms(sfpt->jvms());
6332 _reexecute_sp = jvms()->sp();
6333
6334 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
6335
6336 // Restore state
6337 set_jvms(saved_jvms);
6338 _reexecute_sp = saved_reexecute_sp;
6339 }
6340 }
6341
6342 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
6343 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
6344 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
6345 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
6346 while (if_proj->is_IfProj()) {
6347 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
6348 if (uncommon_trap != nullptr) {
6349 create_new_uncommon_trap(uncommon_trap);
6350 }
6351 assert(if_proj->in(0)->is_If(), "must be If");
6352 if_proj = if_proj->in(0)->in(0);
6353 }
6354 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
6355 "must have reached control projection of init node");
6356 }
6357
6358 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
6359 const int trap_request = uncommon_trap_call->uncommon_trap_request();
6360 assert(trap_request != 0, "no valid UCT trap request");
6361 PreserveJVMState pjvms(this);
6362 set_control(uncommon_trap_call->in(0));
6363 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
6364 Deoptimization::trap_request_action(trap_request));
6365 assert(stopped(), "Should be stopped");
6366 _gvn.hash_delete(uncommon_trap_call);
6367 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
6368 }
6369
6370 // Common checks for array sorting intrinsics arguments.
6371 // Returns `true` if checks passed.
6372 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
6373 // check address of the class
6374 if (elementType == nullptr || elementType->is_top()) {
6375 return false; // dead path
6376 }
6377 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
6378 if (elem_klass == nullptr) {
6379 return false; // dead path
6380 }
6381 // java_mirror_type() returns non-null for compile-time Class constants only
6382 ciType* elem_type = elem_klass->java_mirror_type();
6383 if (elem_type == nullptr) {
6384 return false;
6385 }
6386 bt = elem_type->basic_type();
6387 // Disable the intrinsic if the CPU does not support SIMD sort
6388 if (!Matcher::supports_simd_sort(bt)) {
6389 return false;
6390 }
6391 // check address of the array
6392 if (obj == nullptr || obj->is_top()) {
6393 return false; // dead path
6394 }
6395 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
6396 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
6397 return false; // failed input validation
6398 }
6399 return true;
6400 }
6401
6402 //------------------------------inline_array_partition-----------------------
6403 bool LibraryCallKit::inline_array_partition() {
6404 address stubAddr = StubRoutines::select_array_partition_function();
6405 if (stubAddr == nullptr) {
6406 return false; // Intrinsic's stub is not implemented on this platform
6407 }
6408 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
6409
6410 // no receiver because it is a static method
6411 Node* elementType = argument(0);
6412 Node* obj = argument(1);
6413 Node* offset = argument(2); // long
6414 Node* fromIndex = argument(4);
6415 Node* toIndex = argument(5);
6416 Node* indexPivot1 = argument(6);
6417 Node* indexPivot2 = argument(7);
6418 // PartitionOperation: argument(8) is ignored
6419
6420 Node* pivotIndices = nullptr;
6421 BasicType bt = T_ILLEGAL;
6422
6423 if (!check_array_sort_arguments(elementType, obj, bt)) {
6424 return false;
6425 }
6426 null_check(obj);
6427 // If obj is dead, only null-path is taken.
6428 if (stopped()) {
6429 return true;
6430 }
6431 // Set the original stack and the reexecute bit for the interpreter to reexecute
6432 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
6433 { PreserveReexecuteState preexecs(this);
6434 jvms()->set_should_reexecute(true);
6435
6436 Node* obj_adr = make_unsafe_address(obj, offset);
6437
6438 // create the pivotIndices array of type int and size = 2
6439 Node* size = intcon(2);
6440 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
6441 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
6442 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
6443 guarantee(alloc != nullptr, "created above");
6444 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
6445
6446 // pass the basic type enum to the stub
6447 Node* elemType = intcon(bt);
6448
6449 // Call the stub
6450 const char *stubName = "array_partition_stub";
6451 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
6452 stubAddr, stubName, TypePtr::BOTTOM,
6453 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
6454 indexPivot1, indexPivot2);
6455
6456 } // original reexecute is set back here
6457
6458 if (!stopped()) {
6459 set_result(pivotIndices);
6460 }
6461
6462 return true;
6463 }
6464
6465
6466 //------------------------------inline_array_sort-----------------------
6467 bool LibraryCallKit::inline_array_sort() {
6468 address stubAddr = StubRoutines::select_arraysort_function();
6469 if (stubAddr == nullptr) {
6470 return false; // Intrinsic's stub is not implemented on this platform
6471 }
6472 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
6473
6474 // no receiver because it is a static method
6475 Node* elementType = argument(0);
6476 Node* obj = argument(1);
6477 Node* offset = argument(2); // long
6478 Node* fromIndex = argument(4);
6479 Node* toIndex = argument(5);
6480 // SortOperation: argument(6) is ignored
6481
6482 BasicType bt = T_ILLEGAL;
6483
6484 if (!check_array_sort_arguments(elementType, obj, bt)) {
6485 return false;
6486 }
6487 null_check(obj);
6488 // If obj is dead, only null-path is taken.
6489 if (stopped()) {
6490 return true;
6491 }
6492 Node* obj_adr = make_unsafe_address(obj, offset);
6493
6494 // pass the basic type enum to the stub
6495 Node* elemType = intcon(bt);
6496
6497 // Call the stub.
6498 const char *stubName = "arraysort_stub";
6499 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
6500 stubAddr, stubName, TypePtr::BOTTOM,
6501 obj_adr, elemType, fromIndex, toIndex);
6502
6503 return true;
6504 }
6505
6506
6507 //------------------------------inline_arraycopy-----------------------
6508 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
6509 // Object dest, int destPos,
6510 // int length);
6511 bool LibraryCallKit::inline_arraycopy() {
6512 // Get the arguments.
6513 Node* src = argument(0); // type: oop
6514 Node* src_offset = argument(1); // type: int
6515 Node* dest = argument(2); // type: oop
6516 Node* dest_offset = argument(3); // type: int
6517 Node* length = argument(4); // type: int
6518
6519 uint new_idx = C->unique();
6520
6521 // Check for allocation before we add nodes that would confuse
6522 // tightly_coupled_allocation()
6523 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
6524
6525 int saved_reexecute_sp = -1;
6526 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
6527 // See arraycopy_restore_alloc_state() comment
6528 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
6529 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
6530 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
6531 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
6532
6533 // The following tests must be performed
6534 // (1) src and dest are arrays.
6535 // (2) src and dest arrays must have elements of the same BasicType
6536 // (3) src and dest must not be null.
6537 // (4) src_offset must not be negative.
6538 // (5) dest_offset must not be negative.
6539 // (6) length must not be negative.
6540 // (7) src_offset + length must not exceed length of src.
6541 // (8) dest_offset + length must not exceed length of dest.
6542 // (9) each element of an oop array must be assignable
6543
6544 // (3) src and dest must not be null.
6545 // always do this here because we need the JVM state for uncommon traps
6546 Node* null_ctl = top();
6547 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
6548 assert(null_ctl->is_top(), "no null control here");
6549 dest = null_check(dest, T_ARRAY);
6550
6551 if (!can_emit_guards) {
6552 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
6553 // guards but the arraycopy node could still take advantage of a
6554 // tightly allocated allocation. tightly_coupled_allocation() is
6555 // called again to make sure it takes the null check above into
6556 // account: the null check is mandatory and if it caused an
6557 // uncommon trap to be emitted then the allocation can't be
6558 // considered tightly coupled in this context.
6559 alloc = tightly_coupled_allocation(dest);
6560 }
6561
6562 bool validated = false;
6563
6564 const Type* src_type = _gvn.type(src);
6565 const Type* dest_type = _gvn.type(dest);
6566 const TypeAryPtr* top_src = src_type->isa_aryptr();
6567 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6568
6569 // Do we have the type of src?
6570 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6571 // Do we have the type of dest?
6572 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6573 // Is the type for src from speculation?
6574 bool src_spec = false;
6575 // Is the type for dest from speculation?
6576 bool dest_spec = false;
6577
6578 if ((!has_src || !has_dest) && can_emit_guards) {
6579 // We don't have sufficient type information, let's see if
6580 // speculative types can help. We need to have types for both src
6581 // and dest so that it pays off.
6582
6583 // Do we already have or could we have type information for src
6584 bool could_have_src = has_src;
6585 // Do we already have or could we have type information for dest
6586 bool could_have_dest = has_dest;
6587
6588 ciKlass* src_k = nullptr;
6589 if (!has_src) {
6590 src_k = src_type->speculative_type_not_null();
6591 if (src_k != nullptr && src_k->is_array_klass()) {
6592 could_have_src = true;
6593 }
6594 }
6595
6596 ciKlass* dest_k = nullptr;
6597 if (!has_dest) {
6598 dest_k = dest_type->speculative_type_not_null();
6599 if (dest_k != nullptr && dest_k->is_array_klass()) {
6600 could_have_dest = true;
6601 }
6602 }
6603
6604 if (could_have_src && could_have_dest) {
6605 // This is going to pay off so emit the required guards
6606 if (!has_src) {
6607 src = maybe_cast_profiled_obj(src, src_k, true);
6608 src_type = _gvn.type(src);
6609 top_src = src_type->isa_aryptr();
6610 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6611 src_spec = true;
6612 }
6613 if (!has_dest) {
6614 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6615 dest_type = _gvn.type(dest);
6616 top_dest = dest_type->isa_aryptr();
6617 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6618 dest_spec = true;
6619 }
6620 }
6621 }
6622
6623 if (has_src && has_dest && can_emit_guards) {
6624 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
6625 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
6626 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
6627 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
6628
6629 if (src_elem == dest_elem && top_src->is_flat() == top_dest->is_flat() && src_elem == T_OBJECT) {
6630 // If both arrays are object arrays then having the exact types
6631 // for both will remove the need for a subtype check at runtime
6632 // before the call and may make it possible to pick a faster copy
6633 // routine (without a subtype check on every element)
6634 // Do we have the exact type of src?
6635 bool could_have_src = src_spec;
6636 // Do we have the exact type of dest?
6637 bool could_have_dest = dest_spec;
6638 ciKlass* src_k = nullptr;
6639 ciKlass* dest_k = nullptr;
6640 if (!src_spec) {
6641 src_k = src_type->speculative_type_not_null();
6642 if (src_k != nullptr && src_k->is_array_klass()) {
6643 could_have_src = true;
6644 }
6645 }
6646 if (!dest_spec) {
6647 dest_k = dest_type->speculative_type_not_null();
6648 if (dest_k != nullptr && dest_k->is_array_klass()) {
6649 could_have_dest = true;
6650 }
6651 }
6652 if (could_have_src && could_have_dest) {
6653 // If we can have both exact types, emit the missing guards
6654 if (could_have_src && !src_spec) {
6655 src = maybe_cast_profiled_obj(src, src_k, true);
6656 src_type = _gvn.type(src);
6657 top_src = src_type->isa_aryptr();
6658 }
6659 if (could_have_dest && !dest_spec) {
6660 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6661 dest_type = _gvn.type(dest);
6662 top_dest = dest_type->isa_aryptr();
6663 }
6664 }
6665 }
6666 }
6667
6668 ciMethod* trap_method = method();
6669 int trap_bci = bci();
6670 if (saved_jvms_before_guards != nullptr) {
6671 trap_method = alloc->jvms()->method();
6672 trap_bci = alloc->jvms()->bci();
6673 }
6674
6675 bool negative_length_guard_generated = false;
6676
6677 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6678 can_emit_guards && !src->is_top() && !dest->is_top()) {
6679 // validate arguments: enables transformation the ArrayCopyNode
6680 validated = true;
6681
6682 RegionNode* slow_region = new RegionNode(1);
6683 record_for_igvn(slow_region);
6684
6685 // (1) src and dest are arrays.
6686 generate_non_array_guard(load_object_klass(src), slow_region, &src);
6687 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
6688
6689 // (2) src and dest arrays must have elements of the same BasicType
6690 // done at macro expansion or at Ideal transformation time
6691
6692 // (4) src_offset must not be negative.
6693 generate_negative_guard(src_offset, slow_region);
6694
6695 // (5) dest_offset must not be negative.
6696 generate_negative_guard(dest_offset, slow_region);
6697
6698 // (7) src_offset + length must not exceed length of src.
6699 generate_limit_guard(src_offset, length,
6700 load_array_length(src),
6701 slow_region);
6702
6703 // (8) dest_offset + length must not exceed length of dest.
6704 generate_limit_guard(dest_offset, length,
6705 load_array_length(dest),
6706 slow_region);
6707
6708 // (6) length must not be negative.
6709 // This is also checked in generate_arraycopy() during macro expansion, but
6710 // we also have to check it here for the case where the ArrayCopyNode will
6711 // be eliminated by Escape Analysis.
6712 if (EliminateAllocations) {
6713 generate_negative_guard(length, slow_region);
6714 negative_length_guard_generated = true;
6715 }
6716
6717 // (9) each element of an oop array must be assignable
6718 Node* dest_klass = load_object_klass(dest);
6719 Node* refined_dest_klass = dest_klass;
6720 if (src != dest) {
6721 dest_klass = load_non_refined_array_klass(refined_dest_klass);
6722 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6723 slow_region->add_req(not_subtype_ctrl);
6724 }
6725
6726 // TODO 8350865 Improve this. What about atomicity? Make sure this is always folded for type arrays.
6727 // If destination is null-restricted, source must be null-restricted as well: src_null_restricted || !dst_null_restricted
6728 Node* src_klass = load_object_klass(src);
6729 Node* adr_prop_src = basic_plus_adr(src_klass, in_bytes(ArrayKlass::properties_offset()));
6730 Node* prop_src = _gvn.transform(LoadNode::make(_gvn, control(), immutable_memory(), adr_prop_src, TypeRawPtr::BOTTOM, TypeInt::INT, T_INT, MemNode::unordered));
6731 Node* adr_prop_dest = basic_plus_adr(refined_dest_klass, in_bytes(ArrayKlass::properties_offset()));
6732 Node* prop_dest = _gvn.transform(LoadNode::make(_gvn, control(), immutable_memory(), adr_prop_dest, TypeRawPtr::BOTTOM, TypeInt::INT, T_INT, MemNode::unordered));
6733
6734 prop_dest = _gvn.transform(new XorINode(prop_dest, intcon(ArrayKlass::ArrayProperties::NULL_RESTRICTED)));
6735 prop_src = _gvn.transform(new OrINode(prop_dest, prop_src));
6736 prop_src = _gvn.transform(new AndINode(prop_src, intcon(ArrayKlass::ArrayProperties::NULL_RESTRICTED)));
6737
6738 Node* chk = _gvn.transform(new CmpINode(prop_src, intcon(ArrayKlass::ArrayProperties::NULL_RESTRICTED)));
6739 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::ne));
6740 generate_fair_guard(tst, slow_region);
6741
6742 // TODO 8350865 This is too strong
6743 generate_fair_guard(flat_array_test(src), slow_region);
6744 generate_fair_guard(flat_array_test(dest), slow_region);
6745
6746 {
6747 PreserveJVMState pjvms(this);
6748 set_control(_gvn.transform(slow_region));
6749 uncommon_trap(Deoptimization::Reason_intrinsic,
6750 Deoptimization::Action_make_not_entrant);
6751 assert(stopped(), "Should be stopped");
6752 }
6753
6754 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->isa_klassptr();
6755 if (dest_klass_t == nullptr) {
6756 // refined_dest_klass may not be an array, which leads to dest_klass being top. This means we
6757 // are in a dead path.
6758 uncommon_trap(Deoptimization::Reason_intrinsic,
6759 Deoptimization::Action_make_not_entrant);
6760 return true;
6761 }
6762
6763 const Type* toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6764 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6765 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6766 }
6767
6768 if (stopped()) {
6769 return true;
6770 }
6771
6772 Node* dest_klass = load_object_klass(dest);
6773 dest_klass = load_non_refined_array_klass(dest_klass);
6774
6775 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6776 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6777 // so the compiler has a chance to eliminate them: during macro expansion,
6778 // we have to set their control (CastPP nodes are eliminated).
6779 load_object_klass(src), dest_klass,
6780 load_array_length(src), load_array_length(dest));
6781
6782 ac->set_arraycopy(validated);
6783
6784 Node* n = _gvn.transform(ac);
6785 if (n == ac) {
6786 ac->connect_outputs(this);
6787 } else {
6788 assert(validated, "shouldn't transform if all arguments not validated");
6789 set_all_memory(n);
6790 }
6791 clear_upper_avx();
6792
6793
6794 return true;
6795 }
6796
6797
6798 // Helper function which determines if an arraycopy immediately follows
6799 // an allocation, with no intervening tests or other escapes for the object.
6800 AllocateArrayNode*
6801 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6802 if (stopped()) return nullptr; // no fast path
6803 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6804
6805 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6806 if (alloc == nullptr) return nullptr;
6807
6808 Node* rawmem = memory(Compile::AliasIdxRaw);
6809 // Is the allocation's memory state untouched?
6810 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6811 // Bail out if there have been raw-memory effects since the allocation.
6812 // (Example: There might have been a call or safepoint.)
6813 return nullptr;
6814 }
6815 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6816 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6817 return nullptr;
6818 }
6819
6820 // There must be no unexpected observers of this allocation.
6821 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6822 Node* obs = ptr->fast_out(i);
6823 if (obs != this->map()) {
6824 return nullptr;
6825 }
6826 }
6827
6828 // This arraycopy must unconditionally follow the allocation of the ptr.
6829 Node* alloc_ctl = ptr->in(0);
6830 Node* ctl = control();
6831 while (ctl != alloc_ctl) {
6832 // There may be guards which feed into the slow_region.
6833 // Any other control flow means that we might not get a chance
6834 // to finish initializing the allocated object.
6835 // Various low-level checks bottom out in uncommon traps. These
6836 // are considered safe since we've already checked above that
6837 // there is no unexpected observer of this allocation.
6838 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6839 assert(ctl->in(0)->is_If(), "must be If");
6840 ctl = ctl->in(0)->in(0);
6841 } else {
6842 return nullptr;
6843 }
6844 }
6845
6846 // If we get this far, we have an allocation which immediately
6847 // precedes the arraycopy, and we can take over zeroing the new object.
6848 // The arraycopy will finish the initialization, and provide
6849 // a new control state to which we will anchor the destination pointer.
6850
6851 return alloc;
6852 }
6853
6854 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6855 if (node->is_IfProj()) {
6856 IfProjNode* other_proj = node->as_IfProj()->other_if_proj();
6857 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6858 Node* obs = other_proj->fast_out(j);
6859 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6860 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6861 return obs->as_CallStaticJava();
6862 }
6863 }
6864 }
6865 return nullptr;
6866 }
6867
6868 //-------------inline_encodeISOArray-----------------------------------
6869 // encode char[] to byte[] in ISO_8859_1 or ASCII
6870 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6871 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6872 // no receiver since it is static method
6873 Node *src = argument(0);
6874 Node *src_offset = argument(1);
6875 Node *dst = argument(2);
6876 Node *dst_offset = argument(3);
6877 Node *length = argument(4);
6878
6879 src = must_be_not_null(src, true);
6880 dst = must_be_not_null(dst, true);
6881
6882 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6883 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6884 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6885 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6886 // failed array check
6887 return false;
6888 }
6889
6890 // Figure out the size and type of the elements we will be copying.
6891 BasicType src_elem = src_type->elem()->array_element_basic_type();
6892 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6893 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6894 return false;
6895 }
6896
6897 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6898 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6899 // 'src_start' points to src array + scaled offset
6900 // 'dst_start' points to dst array + scaled offset
6901
6902 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6903 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6904 enc = _gvn.transform(enc);
6905 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6906 set_memory(res_mem, mtype);
6907 set_result(enc);
6908 clear_upper_avx();
6909
6910 return true;
6911 }
6912
6913 //-------------inline_multiplyToLen-----------------------------------
6914 bool LibraryCallKit::inline_multiplyToLen() {
6915 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6916
6917 address stubAddr = StubRoutines::multiplyToLen();
6918 if (stubAddr == nullptr) {
6919 return false; // Intrinsic's stub is not implemented on this platform
6920 }
6921 const char* stubName = "multiplyToLen";
6922
6923 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6924
6925 // no receiver because it is a static method
6926 Node* x = argument(0);
6927 Node* xlen = argument(1);
6928 Node* y = argument(2);
6929 Node* ylen = argument(3);
6930 Node* z = argument(4);
6931
6932 x = must_be_not_null(x, true);
6933 y = must_be_not_null(y, true);
6934
6935 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6936 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6937 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6938 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6939 // failed array check
6940 return false;
6941 }
6942
6943 BasicType x_elem = x_type->elem()->array_element_basic_type();
6944 BasicType y_elem = y_type->elem()->array_element_basic_type();
6945 if (x_elem != T_INT || y_elem != T_INT) {
6946 return false;
6947 }
6948
6949 Node* x_start = array_element_address(x, intcon(0), x_elem);
6950 Node* y_start = array_element_address(y, intcon(0), y_elem);
6951 // 'x_start' points to x array + scaled xlen
6952 // 'y_start' points to y array + scaled ylen
6953
6954 Node* z_start = array_element_address(z, intcon(0), T_INT);
6955
6956 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6957 OptoRuntime::multiplyToLen_Type(),
6958 stubAddr, stubName, TypePtr::BOTTOM,
6959 x_start, xlen, y_start, ylen, z_start);
6960
6961 C->set_has_split_ifs(true); // Has chance for split-if optimization
6962 set_result(z);
6963 return true;
6964 }
6965
6966 //-------------inline_squareToLen------------------------------------
6967 bool LibraryCallKit::inline_squareToLen() {
6968 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6969
6970 address stubAddr = StubRoutines::squareToLen();
6971 if (stubAddr == nullptr) {
6972 return false; // Intrinsic's stub is not implemented on this platform
6973 }
6974 const char* stubName = "squareToLen";
6975
6976 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6977
6978 Node* x = argument(0);
6979 Node* len = argument(1);
6980 Node* z = argument(2);
6981 Node* zlen = argument(3);
6982
6983 x = must_be_not_null(x, true);
6984 z = must_be_not_null(z, true);
6985
6986 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6987 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6988 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6989 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6990 // failed array check
6991 return false;
6992 }
6993
6994 BasicType x_elem = x_type->elem()->array_element_basic_type();
6995 BasicType z_elem = z_type->elem()->array_element_basic_type();
6996 if (x_elem != T_INT || z_elem != T_INT) {
6997 return false;
6998 }
6999
7000
7001 Node* x_start = array_element_address(x, intcon(0), x_elem);
7002 Node* z_start = array_element_address(z, intcon(0), z_elem);
7003
7004 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
7005 OptoRuntime::squareToLen_Type(),
7006 stubAddr, stubName, TypePtr::BOTTOM,
7007 x_start, len, z_start, zlen);
7008
7009 set_result(z);
7010 return true;
7011 }
7012
7013 //-------------inline_mulAdd------------------------------------------
7014 bool LibraryCallKit::inline_mulAdd() {
7015 assert(UseMulAddIntrinsic, "not implemented on this platform");
7016
7017 address stubAddr = StubRoutines::mulAdd();
7018 if (stubAddr == nullptr) {
7019 return false; // Intrinsic's stub is not implemented on this platform
7020 }
7021 const char* stubName = "mulAdd";
7022
7023 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
7024
7025 Node* out = argument(0);
7026 Node* in = argument(1);
7027 Node* offset = argument(2);
7028 Node* len = argument(3);
7029 Node* k = argument(4);
7030
7031 in = must_be_not_null(in, true);
7032 out = must_be_not_null(out, true);
7033
7034 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
7035 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
7036 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
7037 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
7038 // failed array check
7039 return false;
7040 }
7041
7042 BasicType out_elem = out_type->elem()->array_element_basic_type();
7043 BasicType in_elem = in_type->elem()->array_element_basic_type();
7044 if (out_elem != T_INT || in_elem != T_INT) {
7045 return false;
7046 }
7047
7048 Node* outlen = load_array_length(out);
7049 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
7050 Node* out_start = array_element_address(out, intcon(0), out_elem);
7051 Node* in_start = array_element_address(in, intcon(0), in_elem);
7052
7053 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
7054 OptoRuntime::mulAdd_Type(),
7055 stubAddr, stubName, TypePtr::BOTTOM,
7056 out_start,in_start, new_offset, len, k);
7057 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7058 set_result(result);
7059 return true;
7060 }
7061
7062 //-------------inline_montgomeryMultiply-----------------------------------
7063 bool LibraryCallKit::inline_montgomeryMultiply() {
7064 address stubAddr = StubRoutines::montgomeryMultiply();
7065 if (stubAddr == nullptr) {
7066 return false; // Intrinsic's stub is not implemented on this platform
7067 }
7068
7069 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
7070 const char* stubName = "montgomery_multiply";
7071
7072 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
7073
7074 Node* a = argument(0);
7075 Node* b = argument(1);
7076 Node* n = argument(2);
7077 Node* len = argument(3);
7078 Node* inv = argument(4);
7079 Node* m = argument(6);
7080
7081 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
7082 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
7083 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
7084 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
7085 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
7086 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
7087 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
7088 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
7089 // failed array check
7090 return false;
7091 }
7092
7093 BasicType a_elem = a_type->elem()->array_element_basic_type();
7094 BasicType b_elem = b_type->elem()->array_element_basic_type();
7095 BasicType n_elem = n_type->elem()->array_element_basic_type();
7096 BasicType m_elem = m_type->elem()->array_element_basic_type();
7097 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
7098 return false;
7099 }
7100
7101 // Make the call
7102 {
7103 Node* a_start = array_element_address(a, intcon(0), a_elem);
7104 Node* b_start = array_element_address(b, intcon(0), b_elem);
7105 Node* n_start = array_element_address(n, intcon(0), n_elem);
7106 Node* m_start = array_element_address(m, intcon(0), m_elem);
7107
7108 Node* call = make_runtime_call(RC_LEAF,
7109 OptoRuntime::montgomeryMultiply_Type(),
7110 stubAddr, stubName, TypePtr::BOTTOM,
7111 a_start, b_start, n_start, len, inv, top(),
7112 m_start);
7113 set_result(m);
7114 }
7115
7116 return true;
7117 }
7118
7119 bool LibraryCallKit::inline_montgomerySquare() {
7120 address stubAddr = StubRoutines::montgomerySquare();
7121 if (stubAddr == nullptr) {
7122 return false; // Intrinsic's stub is not implemented on this platform
7123 }
7124
7125 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
7126 const char* stubName = "montgomery_square";
7127
7128 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
7129
7130 Node* a = argument(0);
7131 Node* n = argument(1);
7132 Node* len = argument(2);
7133 Node* inv = argument(3);
7134 Node* m = argument(5);
7135
7136 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
7137 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
7138 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
7139 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
7140 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
7141 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
7142 // failed array check
7143 return false;
7144 }
7145
7146 BasicType a_elem = a_type->elem()->array_element_basic_type();
7147 BasicType n_elem = n_type->elem()->array_element_basic_type();
7148 BasicType m_elem = m_type->elem()->array_element_basic_type();
7149 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
7150 return false;
7151 }
7152
7153 // Make the call
7154 {
7155 Node* a_start = array_element_address(a, intcon(0), a_elem);
7156 Node* n_start = array_element_address(n, intcon(0), n_elem);
7157 Node* m_start = array_element_address(m, intcon(0), m_elem);
7158
7159 Node* call = make_runtime_call(RC_LEAF,
7160 OptoRuntime::montgomerySquare_Type(),
7161 stubAddr, stubName, TypePtr::BOTTOM,
7162 a_start, n_start, len, inv, top(),
7163 m_start);
7164 set_result(m);
7165 }
7166
7167 return true;
7168 }
7169
7170 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
7171 address stubAddr = nullptr;
7172 const char* stubName = nullptr;
7173
7174 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
7175 if (stubAddr == nullptr) {
7176 return false; // Intrinsic's stub is not implemented on this platform
7177 }
7178
7179 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
7180
7181 assert(callee()->signature()->size() == 5, "expected 5 arguments");
7182
7183 Node* newArr = argument(0);
7184 Node* oldArr = argument(1);
7185 Node* newIdx = argument(2);
7186 Node* shiftCount = argument(3);
7187 Node* numIter = argument(4);
7188
7189 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
7190 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
7191 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
7192 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
7193 return false;
7194 }
7195
7196 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
7197 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
7198 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
7199 return false;
7200 }
7201
7202 // Make the call
7203 {
7204 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
7205 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
7206
7207 Node* call = make_runtime_call(RC_LEAF,
7208 OptoRuntime::bigIntegerShift_Type(),
7209 stubAddr,
7210 stubName,
7211 TypePtr::BOTTOM,
7212 newArr_start,
7213 oldArr_start,
7214 newIdx,
7215 shiftCount,
7216 numIter);
7217 }
7218
7219 return true;
7220 }
7221
7222 //-------------inline_vectorizedMismatch------------------------------
7223 bool LibraryCallKit::inline_vectorizedMismatch() {
7224 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
7225
7226 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
7227 Node* obja = argument(0); // Object
7228 Node* aoffset = argument(1); // long
7229 Node* objb = argument(3); // Object
7230 Node* boffset = argument(4); // long
7231 Node* length = argument(6); // int
7232 Node* scale = argument(7); // int
7233
7234 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
7235 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
7236 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
7237 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
7238 scale == top()) {
7239 return false; // failed input validation
7240 }
7241
7242 Node* obja_adr = make_unsafe_address(obja, aoffset);
7243 Node* objb_adr = make_unsafe_address(objb, boffset);
7244
7245 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
7246 //
7247 // inline_limit = ArrayOperationPartialInlineSize / element_size;
7248 // if (length <= inline_limit) {
7249 // inline_path:
7250 // vmask = VectorMaskGen length
7251 // vload1 = LoadVectorMasked obja, vmask
7252 // vload2 = LoadVectorMasked objb, vmask
7253 // result1 = VectorCmpMasked vload1, vload2, vmask
7254 // } else {
7255 // call_stub_path:
7256 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
7257 // }
7258 // exit_block:
7259 // return Phi(result1, result2);
7260 //
7261 enum { inline_path = 1, // input is small enough to process it all at once
7262 stub_path = 2, // input is too large; call into the VM
7263 PATH_LIMIT = 3
7264 };
7265
7266 Node* exit_block = new RegionNode(PATH_LIMIT);
7267 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
7268 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
7269
7270 Node* call_stub_path = control();
7271
7272 BasicType elem_bt = T_ILLEGAL;
7273
7274 const TypeInt* scale_t = _gvn.type(scale)->is_int();
7275 if (scale_t->is_con()) {
7276 switch (scale_t->get_con()) {
7277 case 0: elem_bt = T_BYTE; break;
7278 case 1: elem_bt = T_SHORT; break;
7279 case 2: elem_bt = T_INT; break;
7280 case 3: elem_bt = T_LONG; break;
7281
7282 default: elem_bt = T_ILLEGAL; break; // not supported
7283 }
7284 }
7285
7286 int inline_limit = 0;
7287 bool do_partial_inline = false;
7288
7289 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
7290 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
7291 do_partial_inline = inline_limit >= 16;
7292 }
7293
7294 if (do_partial_inline) {
7295 assert(elem_bt != T_ILLEGAL, "sanity");
7296
7297 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
7298 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
7299 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
7300
7301 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
7302 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
7303 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
7304
7305 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
7306
7307 if (!stopped()) {
7308 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
7309
7310 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
7311 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
7312 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
7313 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
7314
7315 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
7316 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
7317 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
7318 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
7319
7320 exit_block->init_req(inline_path, control());
7321 memory_phi->init_req(inline_path, map()->memory());
7322 result_phi->init_req(inline_path, result);
7323
7324 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
7325 clear_upper_avx();
7326 }
7327 }
7328 }
7329
7330 if (call_stub_path != nullptr) {
7331 set_control(call_stub_path);
7332
7333 Node* call = make_runtime_call(RC_LEAF,
7334 OptoRuntime::vectorizedMismatch_Type(),
7335 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
7336 obja_adr, objb_adr, length, scale);
7337
7338 exit_block->init_req(stub_path, control());
7339 memory_phi->init_req(stub_path, map()->memory());
7340 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
7341 }
7342
7343 exit_block = _gvn.transform(exit_block);
7344 memory_phi = _gvn.transform(memory_phi);
7345 result_phi = _gvn.transform(result_phi);
7346
7347 record_for_igvn(exit_block);
7348 record_for_igvn(memory_phi);
7349 record_for_igvn(result_phi);
7350
7351 set_control(exit_block);
7352 set_all_memory(memory_phi);
7353 set_result(result_phi);
7354
7355 return true;
7356 }
7357
7358 //------------------------------inline_vectorizedHashcode----------------------------
7359 bool LibraryCallKit::inline_vectorizedHashCode() {
7360 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
7361
7362 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
7363 Node* array = argument(0);
7364 Node* offset = argument(1);
7365 Node* length = argument(2);
7366 Node* initialValue = argument(3);
7367 Node* basic_type = argument(4);
7368
7369 if (basic_type == top()) {
7370 return false; // failed input validation
7371 }
7372
7373 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
7374 if (!basic_type_t->is_con()) {
7375 return false; // Only intrinsify if mode argument is constant
7376 }
7377
7378 array = must_be_not_null(array, true);
7379
7380 BasicType bt = (BasicType)basic_type_t->get_con();
7381
7382 // Resolve address of first element
7383 Node* array_start = array_element_address(array, offset, bt);
7384
7385 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
7386 array_start, length, initialValue, basic_type)));
7387 clear_upper_avx();
7388
7389 return true;
7390 }
7391
7392 /**
7393 * Calculate CRC32 for byte.
7394 * int java.util.zip.CRC32.update(int crc, int b)
7395 */
7396 bool LibraryCallKit::inline_updateCRC32() {
7397 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7398 assert(callee()->signature()->size() == 2, "update has 2 parameters");
7399 // no receiver since it is static method
7400 Node* crc = argument(0); // type: int
7401 Node* b = argument(1); // type: int
7402
7403 /*
7404 * int c = ~ crc;
7405 * b = timesXtoThe32[(b ^ c) & 0xFF];
7406 * b = b ^ (c >>> 8);
7407 * crc = ~b;
7408 */
7409
7410 Node* M1 = intcon(-1);
7411 crc = _gvn.transform(new XorINode(crc, M1));
7412 Node* result = _gvn.transform(new XorINode(crc, b));
7413 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
7414
7415 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
7416 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
7417 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
7418 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
7419
7420 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
7421 result = _gvn.transform(new XorINode(crc, result));
7422 result = _gvn.transform(new XorINode(result, M1));
7423 set_result(result);
7424 return true;
7425 }
7426
7427 /**
7428 * Calculate CRC32 for byte[] array.
7429 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
7430 */
7431 bool LibraryCallKit::inline_updateBytesCRC32() {
7432 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7433 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7434 // no receiver since it is static method
7435 Node* crc = argument(0); // type: int
7436 Node* src = argument(1); // type: oop
7437 Node* offset = argument(2); // type: int
7438 Node* length = argument(3); // type: int
7439
7440 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7441 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7442 // failed array check
7443 return false;
7444 }
7445
7446 // Figure out the size and type of the elements we will be copying.
7447 BasicType src_elem = src_type->elem()->array_element_basic_type();
7448 if (src_elem != T_BYTE) {
7449 return false;
7450 }
7451
7452 // 'src_start' points to src array + scaled offset
7453 src = must_be_not_null(src, true);
7454 Node* src_start = array_element_address(src, offset, src_elem);
7455
7456 // We assume that range check is done by caller.
7457 // TODO: generate range check (offset+length < src.length) in debug VM.
7458
7459 // Call the stub.
7460 address stubAddr = StubRoutines::updateBytesCRC32();
7461 const char *stubName = "updateBytesCRC32";
7462
7463 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7464 stubAddr, stubName, TypePtr::BOTTOM,
7465 crc, src_start, length);
7466 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7467 set_result(result);
7468 return true;
7469 }
7470
7471 /**
7472 * Calculate CRC32 for ByteBuffer.
7473 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
7474 */
7475 bool LibraryCallKit::inline_updateByteBufferCRC32() {
7476 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7477 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7478 // no receiver since it is static method
7479 Node* crc = argument(0); // type: int
7480 Node* src = argument(1); // type: long
7481 Node* offset = argument(3); // type: int
7482 Node* length = argument(4); // type: int
7483
7484 src = ConvL2X(src); // adjust Java long to machine word
7485 Node* base = _gvn.transform(new CastX2PNode(src));
7486 offset = ConvI2X(offset);
7487
7488 // 'src_start' points to src array + scaled offset
7489 Node* src_start = basic_plus_adr(top(), base, offset);
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 //------------------------------get_table_from_crc32c_class-----------------------
7504 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
7505 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
7506 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
7507
7508 return table;
7509 }
7510
7511 //------------------------------inline_updateBytesCRC32C-----------------------
7512 //
7513 // Calculate CRC32C for byte[] array.
7514 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
7515 //
7516 bool LibraryCallKit::inline_updateBytesCRC32C() {
7517 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7518 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7519 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7520 // no receiver since it is a static method
7521 Node* crc = argument(0); // type: int
7522 Node* src = argument(1); // type: oop
7523 Node* offset = argument(2); // type: int
7524 Node* end = argument(3); // type: int
7525
7526 Node* length = _gvn.transform(new SubINode(end, offset));
7527
7528 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7529 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7530 // failed array check
7531 return false;
7532 }
7533
7534 // Figure out the size and type of the elements we will be copying.
7535 BasicType src_elem = src_type->elem()->array_element_basic_type();
7536 if (src_elem != T_BYTE) {
7537 return false;
7538 }
7539
7540 // 'src_start' points to src array + scaled offset
7541 src = must_be_not_null(src, true);
7542 Node* src_start = array_element_address(src, offset, src_elem);
7543
7544 // static final int[] byteTable in class CRC32C
7545 Node* table = get_table_from_crc32c_class(callee()->holder());
7546 table = must_be_not_null(table, true);
7547 Node* table_start = array_element_address(table, intcon(0), T_INT);
7548
7549 // We assume that range check is done by caller.
7550 // TODO: generate range check (offset+length < src.length) in debug VM.
7551
7552 // Call the stub.
7553 address stubAddr = StubRoutines::updateBytesCRC32C();
7554 const char *stubName = "updateBytesCRC32C";
7555
7556 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7557 stubAddr, stubName, TypePtr::BOTTOM,
7558 crc, src_start, length, table_start);
7559 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7560 set_result(result);
7561 return true;
7562 }
7563
7564 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
7565 //
7566 // Calculate CRC32C for DirectByteBuffer.
7567 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
7568 //
7569 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
7570 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7571 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
7572 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7573 // no receiver since it is a static method
7574 Node* crc = argument(0); // type: int
7575 Node* src = argument(1); // type: long
7576 Node* offset = argument(3); // type: int
7577 Node* end = argument(4); // type: int
7578
7579 Node* length = _gvn.transform(new SubINode(end, offset));
7580
7581 src = ConvL2X(src); // adjust Java long to machine word
7582 Node* base = _gvn.transform(new CastX2PNode(src));
7583 offset = ConvI2X(offset);
7584
7585 // 'src_start' points to src array + scaled offset
7586 Node* src_start = basic_plus_adr(top(), base, offset);
7587
7588 // static final int[] byteTable in class CRC32C
7589 Node* table = get_table_from_crc32c_class(callee()->holder());
7590 table = must_be_not_null(table, true);
7591 Node* table_start = array_element_address(table, intcon(0), T_INT);
7592
7593 // Call the stub.
7594 address stubAddr = StubRoutines::updateBytesCRC32C();
7595 const char *stubName = "updateBytesCRC32C";
7596
7597 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7598 stubAddr, stubName, TypePtr::BOTTOM,
7599 crc, src_start, length, table_start);
7600 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7601 set_result(result);
7602 return true;
7603 }
7604
7605 //------------------------------inline_updateBytesAdler32----------------------
7606 //
7607 // Calculate Adler32 checksum for byte[] array.
7608 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
7609 //
7610 bool LibraryCallKit::inline_updateBytesAdler32() {
7611 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7612 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7613 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7614 // no receiver since it is static method
7615 Node* crc = argument(0); // type: int
7616 Node* src = argument(1); // type: oop
7617 Node* offset = argument(2); // type: int
7618 Node* length = argument(3); // type: int
7619
7620 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7621 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7622 // failed array check
7623 return false;
7624 }
7625
7626 // Figure out the size and type of the elements we will be copying.
7627 BasicType src_elem = src_type->elem()->array_element_basic_type();
7628 if (src_elem != T_BYTE) {
7629 return false;
7630 }
7631
7632 // 'src_start' points to src array + scaled offset
7633 Node* src_start = array_element_address(src, offset, src_elem);
7634
7635 // We assume that range check is done by caller.
7636 // TODO: generate range check (offset+length < src.length) in debug VM.
7637
7638 // Call the stub.
7639 address stubAddr = StubRoutines::updateBytesAdler32();
7640 const char *stubName = "updateBytesAdler32";
7641
7642 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7643 stubAddr, stubName, TypePtr::BOTTOM,
7644 crc, src_start, length);
7645 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7646 set_result(result);
7647 return true;
7648 }
7649
7650 //------------------------------inline_updateByteBufferAdler32---------------
7651 //
7652 // Calculate Adler32 checksum for DirectByteBuffer.
7653 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
7654 //
7655 bool LibraryCallKit::inline_updateByteBufferAdler32() {
7656 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7657 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7658 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7659 // no receiver since it is static method
7660 Node* crc = argument(0); // type: int
7661 Node* src = argument(1); // type: long
7662 Node* offset = argument(3); // type: int
7663 Node* length = argument(4); // type: int
7664
7665 src = ConvL2X(src); // adjust Java long to machine word
7666 Node* base = _gvn.transform(new CastX2PNode(src));
7667 offset = ConvI2X(offset);
7668
7669 // 'src_start' points to src array + scaled offset
7670 Node* src_start = basic_plus_adr(top(), base, offset);
7671
7672 // Call the stub.
7673 address stubAddr = StubRoutines::updateBytesAdler32();
7674 const char *stubName = "updateBytesAdler32";
7675
7676 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7677 stubAddr, stubName, TypePtr::BOTTOM,
7678 crc, src_start, length);
7679
7680 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7681 set_result(result);
7682 return true;
7683 }
7684
7685 //----------------------------inline_reference_get0----------------------------
7686 // public T java.lang.ref.Reference.get();
7687 bool LibraryCallKit::inline_reference_get0() {
7688 const int referent_offset = java_lang_ref_Reference::referent_offset();
7689
7690 // Get the argument:
7691 Node* reference_obj = null_check_receiver();
7692 if (stopped()) return true;
7693
7694 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7695 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7696 decorators, /*is_static*/ false, nullptr);
7697 if (result == nullptr) return false;
7698
7699 // Add memory barrier to prevent commoning reads from this field
7700 // across safepoint since GC can change its value.
7701 insert_mem_bar(Op_MemBarCPUOrder);
7702
7703 set_result(result);
7704 return true;
7705 }
7706
7707 //----------------------------inline_reference_refersTo0----------------------------
7708 // bool java.lang.ref.Reference.refersTo0();
7709 // bool java.lang.ref.PhantomReference.refersTo0();
7710 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7711 // Get arguments:
7712 Node* reference_obj = null_check_receiver();
7713 Node* other_obj = argument(1);
7714 if (stopped()) return true;
7715
7716 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7717 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7718 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7719 decorators, /*is_static*/ false, nullptr);
7720 if (referent == nullptr) return false;
7721
7722 // Add memory barrier to prevent commoning reads from this field
7723 // across safepoint since GC can change its value.
7724 insert_mem_bar(Op_MemBarCPUOrder);
7725
7726 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7727 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7728 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7729
7730 RegionNode* region = new RegionNode(3);
7731 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7732
7733 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7734 region->init_req(1, if_true);
7735 phi->init_req(1, intcon(1));
7736
7737 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7738 region->init_req(2, if_false);
7739 phi->init_req(2, intcon(0));
7740
7741 set_control(_gvn.transform(region));
7742 record_for_igvn(region);
7743 set_result(_gvn.transform(phi));
7744 return true;
7745 }
7746
7747 //----------------------------inline_reference_clear0----------------------------
7748 // void java.lang.ref.Reference.clear0();
7749 // void java.lang.ref.PhantomReference.clear0();
7750 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7751 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7752
7753 // Get arguments
7754 Node* reference_obj = null_check_receiver();
7755 if (stopped()) return true;
7756
7757 // Common access parameters
7758 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7759 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7760 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7761 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7762 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7763
7764 Node* referent = access_load_at(reference_obj,
7765 referent_field_addr,
7766 referent_field_addr_type,
7767 val_type,
7768 T_OBJECT,
7769 decorators);
7770
7771 IdealKit ideal(this);
7772 #define __ ideal.
7773 __ if_then(referent, BoolTest::ne, null());
7774 sync_kit(ideal);
7775 access_store_at(reference_obj,
7776 referent_field_addr,
7777 referent_field_addr_type,
7778 null(),
7779 val_type,
7780 T_OBJECT,
7781 decorators);
7782 __ sync_kit(this);
7783 __ end_if();
7784 final_sync(ideal);
7785 #undef __
7786
7787 return true;
7788 }
7789
7790 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7791 DecoratorSet decorators, bool is_static,
7792 ciInstanceKlass* fromKls) {
7793 if (fromKls == nullptr) {
7794 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7795 assert(tinst != nullptr, "obj is null");
7796 assert(tinst->is_loaded(), "obj is not loaded");
7797 fromKls = tinst->instance_klass();
7798 } else {
7799 assert(is_static, "only for static field access");
7800 }
7801 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7802 ciSymbol::make(fieldTypeString),
7803 is_static);
7804
7805 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7806 if (field == nullptr) return (Node *) nullptr;
7807
7808 if (is_static) {
7809 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7810 fromObj = makecon(tip);
7811 }
7812
7813 // Next code copied from Parse::do_get_xxx():
7814
7815 // Compute address and memory type.
7816 int offset = field->offset_in_bytes();
7817 bool is_vol = field->is_volatile();
7818 ciType* field_klass = field->type();
7819 assert(field_klass->is_loaded(), "should be loaded");
7820 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7821 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7822 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7823 "slice of address and input slice don't match");
7824 BasicType bt = field->layout_type();
7825
7826 // Build the resultant type of the load
7827 const Type *type;
7828 if (bt == T_OBJECT) {
7829 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7830 } else {
7831 type = Type::get_const_basic_type(bt);
7832 }
7833
7834 if (is_vol) {
7835 decorators |= MO_SEQ_CST;
7836 }
7837
7838 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7839 }
7840
7841 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7842 bool is_exact /* true */, bool is_static /* false */,
7843 ciInstanceKlass * fromKls /* nullptr */) {
7844 if (fromKls == nullptr) {
7845 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7846 assert(tinst != nullptr, "obj is null");
7847 assert(tinst->is_loaded(), "obj is not loaded");
7848 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7849 fromKls = tinst->instance_klass();
7850 }
7851 else {
7852 assert(is_static, "only for static field access");
7853 }
7854 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7855 ciSymbol::make(fieldTypeString),
7856 is_static);
7857
7858 assert(field != nullptr, "undefined field");
7859 assert(!field->is_volatile(), "not defined for volatile fields");
7860
7861 if (is_static) {
7862 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7863 fromObj = makecon(tip);
7864 }
7865
7866 // Next code copied from Parse::do_get_xxx():
7867
7868 // Compute address and memory type.
7869 int offset = field->offset_in_bytes();
7870 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7871
7872 return adr;
7873 }
7874
7875 //------------------------------inline_aescrypt_Block-----------------------
7876 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7877 address stubAddr = nullptr;
7878 const char *stubName;
7879 bool is_decrypt = false;
7880 assert(UseAES, "need AES instruction support");
7881
7882 switch(id) {
7883 case vmIntrinsics::_aescrypt_encryptBlock:
7884 stubAddr = StubRoutines::aescrypt_encryptBlock();
7885 stubName = "aescrypt_encryptBlock";
7886 break;
7887 case vmIntrinsics::_aescrypt_decryptBlock:
7888 stubAddr = StubRoutines::aescrypt_decryptBlock();
7889 stubName = "aescrypt_decryptBlock";
7890 is_decrypt = true;
7891 break;
7892 default:
7893 break;
7894 }
7895 if (stubAddr == nullptr) return false;
7896
7897 Node* aescrypt_object = argument(0);
7898 Node* src = argument(1);
7899 Node* src_offset = argument(2);
7900 Node* dest = argument(3);
7901 Node* dest_offset = argument(4);
7902
7903 src = must_be_not_null(src, true);
7904 dest = must_be_not_null(dest, true);
7905
7906 // (1) src and dest are arrays.
7907 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7908 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7909 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7910 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7911
7912 // for the quick and dirty code we will skip all the checks.
7913 // we are just trying to get the call to be generated.
7914 Node* src_start = src;
7915 Node* dest_start = dest;
7916 if (src_offset != nullptr || dest_offset != nullptr) {
7917 assert(src_offset != nullptr && dest_offset != nullptr, "");
7918 src_start = array_element_address(src, src_offset, T_BYTE);
7919 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7920 }
7921
7922 // now need to get the start of its expanded key array
7923 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7924 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7925 if (k_start == nullptr) return false;
7926
7927 // Call the stub.
7928 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7929 stubAddr, stubName, TypePtr::BOTTOM,
7930 src_start, dest_start, k_start);
7931
7932 return true;
7933 }
7934
7935 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7936 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7937 address stubAddr = nullptr;
7938 const char *stubName = nullptr;
7939 bool is_decrypt = false;
7940 assert(UseAES, "need AES instruction support");
7941
7942 switch(id) {
7943 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7944 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7945 stubName = "cipherBlockChaining_encryptAESCrypt";
7946 break;
7947 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7948 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7949 stubName = "cipherBlockChaining_decryptAESCrypt";
7950 is_decrypt = true;
7951 break;
7952 default:
7953 break;
7954 }
7955 if (stubAddr == nullptr) return false;
7956
7957 Node* cipherBlockChaining_object = argument(0);
7958 Node* src = argument(1);
7959 Node* src_offset = argument(2);
7960 Node* len = argument(3);
7961 Node* dest = argument(4);
7962 Node* dest_offset = argument(5);
7963
7964 src = must_be_not_null(src, false);
7965 dest = must_be_not_null(dest, false);
7966
7967 // (1) src and dest are arrays.
7968 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7969 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7970 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7971 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7972
7973 // checks are the responsibility of the caller
7974 Node* src_start = src;
7975 Node* dest_start = dest;
7976 if (src_offset != nullptr || dest_offset != nullptr) {
7977 assert(src_offset != nullptr && dest_offset != nullptr, "");
7978 src_start = array_element_address(src, src_offset, T_BYTE);
7979 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7980 }
7981
7982 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7983 // (because of the predicated logic executed earlier).
7984 // so we cast it here safely.
7985 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7986
7987 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7988 if (embeddedCipherObj == nullptr) return false;
7989
7990 // cast it to what we know it will be at runtime
7991 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7992 assert(tinst != nullptr, "CBC obj is null");
7993 assert(tinst->is_loaded(), "CBC obj is not loaded");
7994 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7995 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7996
7997 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7998 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7999 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8000 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8001 aescrypt_object = _gvn.transform(aescrypt_object);
8002
8003 // we need to get the start of the aescrypt_object's expanded key array
8004 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
8005 if (k_start == nullptr) return false;
8006
8007 // similarly, get the start address of the r vector
8008 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
8009 if (objRvec == nullptr) return false;
8010 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
8011
8012 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8013 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8014 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
8015 stubAddr, stubName, TypePtr::BOTTOM,
8016 src_start, dest_start, k_start, r_start, len);
8017
8018 // return cipher length (int)
8019 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
8020 set_result(retvalue);
8021 return true;
8022 }
8023
8024 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
8025 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
8026 address stubAddr = nullptr;
8027 const char *stubName = nullptr;
8028 bool is_decrypt = false;
8029 assert(UseAES, "need AES instruction support");
8030
8031 switch (id) {
8032 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
8033 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
8034 stubName = "electronicCodeBook_encryptAESCrypt";
8035 break;
8036 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
8037 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
8038 stubName = "electronicCodeBook_decryptAESCrypt";
8039 is_decrypt = true;
8040 break;
8041 default:
8042 break;
8043 }
8044
8045 if (stubAddr == nullptr) return false;
8046
8047 Node* electronicCodeBook_object = argument(0);
8048 Node* src = argument(1);
8049 Node* src_offset = argument(2);
8050 Node* len = argument(3);
8051 Node* dest = argument(4);
8052 Node* dest_offset = argument(5);
8053
8054 // (1) src and dest are arrays.
8055 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8056 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8057 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8058 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8059
8060 // checks are the responsibility of the caller
8061 Node* src_start = src;
8062 Node* dest_start = dest;
8063 if (src_offset != nullptr || dest_offset != nullptr) {
8064 assert(src_offset != nullptr && dest_offset != nullptr, "");
8065 src_start = array_element_address(src, src_offset, T_BYTE);
8066 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8067 }
8068
8069 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8070 // (because of the predicated logic executed earlier).
8071 // so we cast it here safely.
8072 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8073
8074 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8075 if (embeddedCipherObj == nullptr) return false;
8076
8077 // cast it to what we know it will be at runtime
8078 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
8079 assert(tinst != nullptr, "ECB obj is null");
8080 assert(tinst->is_loaded(), "ECB obj is not loaded");
8081 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8082 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8083
8084 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8085 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8086 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8087 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8088 aescrypt_object = _gvn.transform(aescrypt_object);
8089
8090 // we need to get the start of the aescrypt_object's expanded key array
8091 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
8092 if (k_start == nullptr) return false;
8093
8094 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8095 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
8096 OptoRuntime::electronicCodeBook_aescrypt_Type(),
8097 stubAddr, stubName, TypePtr::BOTTOM,
8098 src_start, dest_start, k_start, len);
8099
8100 // return cipher length (int)
8101 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
8102 set_result(retvalue);
8103 return true;
8104 }
8105
8106 //------------------------------inline_counterMode_AESCrypt-----------------------
8107 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
8108 assert(UseAES, "need AES instruction support");
8109 if (!UseAESCTRIntrinsics) return false;
8110
8111 address stubAddr = nullptr;
8112 const char *stubName = nullptr;
8113 if (id == vmIntrinsics::_counterMode_AESCrypt) {
8114 stubAddr = StubRoutines::counterMode_AESCrypt();
8115 stubName = "counterMode_AESCrypt";
8116 }
8117 if (stubAddr == nullptr) return false;
8118
8119 Node* counterMode_object = argument(0);
8120 Node* src = argument(1);
8121 Node* src_offset = argument(2);
8122 Node* len = argument(3);
8123 Node* dest = argument(4);
8124 Node* dest_offset = argument(5);
8125
8126 // (1) src and dest are arrays.
8127 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8128 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8129 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8130 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8131
8132 // checks are the responsibility of the caller
8133 Node* src_start = src;
8134 Node* dest_start = dest;
8135 if (src_offset != nullptr || dest_offset != nullptr) {
8136 assert(src_offset != nullptr && dest_offset != nullptr, "");
8137 src_start = array_element_address(src, src_offset, T_BYTE);
8138 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8139 }
8140
8141 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8142 // (because of the predicated logic executed earlier).
8143 // so we cast it here safely.
8144 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8145 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8146 if (embeddedCipherObj == nullptr) return false;
8147 // cast it to what we know it will be at runtime
8148 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
8149 assert(tinst != nullptr, "CTR obj is null");
8150 assert(tinst->is_loaded(), "CTR obj is not loaded");
8151 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8152 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8153 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8154 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8155 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8156 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8157 aescrypt_object = _gvn.transform(aescrypt_object);
8158 // we need to get the start of the aescrypt_object's expanded key array
8159 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
8160 if (k_start == nullptr) return false;
8161 // similarly, get the start address of the r vector
8162 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
8163 if (obj_counter == nullptr) return false;
8164 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
8165
8166 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
8167 if (saved_encCounter == nullptr) return false;
8168 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
8169 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
8170
8171 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8172 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8173 OptoRuntime::counterMode_aescrypt_Type(),
8174 stubAddr, stubName, TypePtr::BOTTOM,
8175 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
8176
8177 // return cipher length (int)
8178 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
8179 set_result(retvalue);
8180 return true;
8181 }
8182
8183 //------------------------------get_key_start_from_aescrypt_object-----------------------
8184 Node* LibraryCallKit::get_key_start_from_aescrypt_object(Node* aescrypt_object, bool is_decrypt) {
8185 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
8186 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
8187 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
8188 // The following platform specific stubs of encryption and decryption use the same round keys.
8189 #if defined(PPC64) || defined(S390) || defined(RISCV64)
8190 bool use_decryption_key = false;
8191 #else
8192 bool use_decryption_key = is_decrypt;
8193 #endif
8194 Node* objAESCryptKey = load_field_from_object(aescrypt_object, use_decryption_key ? "sessionKd" : "sessionKe", "[I");
8195 assert(objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AES_Crypt");
8196 if (objAESCryptKey == nullptr) return (Node *) nullptr;
8197
8198 // now have the array, need to get the start address of the selected key array
8199 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
8200 return k_start;
8201 }
8202
8203 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
8204 // Return node representing slow path of predicate check.
8205 // the pseudo code we want to emulate with this predicate is:
8206 // for encryption:
8207 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8208 // for decryption:
8209 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8210 // note cipher==plain is more conservative than the original java code but that's OK
8211 //
8212 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
8213 // The receiver was checked for null already.
8214 Node* objCBC = argument(0);
8215
8216 Node* src = argument(1);
8217 Node* dest = argument(4);
8218
8219 // Load embeddedCipher field of CipherBlockChaining object.
8220 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8221
8222 // get AESCrypt klass for instanceOf check
8223 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8224 // will have same classloader as CipherBlockChaining object
8225 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
8226 assert(tinst != nullptr, "CBCobj is null");
8227 assert(tinst->is_loaded(), "CBCobj is not loaded");
8228
8229 // we want to do an instanceof comparison against the AESCrypt class
8230 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8231 if (!klass_AESCrypt->is_loaded()) {
8232 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8233 Node* ctrl = control();
8234 set_control(top()); // no regular fast path
8235 return ctrl;
8236 }
8237
8238 src = must_be_not_null(src, true);
8239 dest = must_be_not_null(dest, true);
8240
8241 // Resolve oops to stable for CmpP below.
8242 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8243
8244 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8245 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8246 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8247
8248 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8249
8250 // for encryption, we are done
8251 if (!decrypting)
8252 return instof_false; // even if it is null
8253
8254 // for decryption, we need to add a further check to avoid
8255 // taking the intrinsic path when cipher and plain are the same
8256 // see the original java code for why.
8257 RegionNode* region = new RegionNode(3);
8258 region->init_req(1, instof_false);
8259
8260 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8261 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8262 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8263 region->init_req(2, src_dest_conjoint);
8264
8265 record_for_igvn(region);
8266 return _gvn.transform(region);
8267 }
8268
8269 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
8270 // Return node representing slow path of predicate check.
8271 // the pseudo code we want to emulate with this predicate is:
8272 // for encryption:
8273 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8274 // for decryption:
8275 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8276 // note cipher==plain is more conservative than the original java code but that's OK
8277 //
8278 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
8279 // The receiver was checked for null already.
8280 Node* objECB = argument(0);
8281
8282 // Load embeddedCipher field of ElectronicCodeBook object.
8283 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8284
8285 // get AESCrypt klass for instanceOf check
8286 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8287 // will have same classloader as ElectronicCodeBook object
8288 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
8289 assert(tinst != nullptr, "ECBobj is null");
8290 assert(tinst->is_loaded(), "ECBobj is not loaded");
8291
8292 // we want to do an instanceof comparison against the AESCrypt class
8293 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8294 if (!klass_AESCrypt->is_loaded()) {
8295 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8296 Node* ctrl = control();
8297 set_control(top()); // no regular fast path
8298 return ctrl;
8299 }
8300 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8301
8302 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8303 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8304 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8305
8306 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8307
8308 // for encryption, we are done
8309 if (!decrypting)
8310 return instof_false; // even if it is null
8311
8312 // for decryption, we need to add a further check to avoid
8313 // taking the intrinsic path when cipher and plain are the same
8314 // see the original java code for why.
8315 RegionNode* region = new RegionNode(3);
8316 region->init_req(1, instof_false);
8317 Node* src = argument(1);
8318 Node* dest = argument(4);
8319 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8320 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8321 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8322 region->init_req(2, src_dest_conjoint);
8323
8324 record_for_igvn(region);
8325 return _gvn.transform(region);
8326 }
8327
8328 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
8329 // Return node representing slow path of predicate check.
8330 // the pseudo code we want to emulate with this predicate is:
8331 // for encryption:
8332 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8333 // for decryption:
8334 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8335 // note cipher==plain is more conservative than the original java code but that's OK
8336 //
8337
8338 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
8339 // The receiver was checked for null already.
8340 Node* objCTR = argument(0);
8341
8342 // Load embeddedCipher field of CipherBlockChaining object.
8343 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8344
8345 // get AESCrypt klass for instanceOf check
8346 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8347 // will have same classloader as CipherBlockChaining object
8348 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
8349 assert(tinst != nullptr, "CTRobj is null");
8350 assert(tinst->is_loaded(), "CTRobj is not loaded");
8351
8352 // we want to do an instanceof comparison against the AESCrypt class
8353 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8354 if (!klass_AESCrypt->is_loaded()) {
8355 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8356 Node* ctrl = control();
8357 set_control(top()); // no regular fast path
8358 return ctrl;
8359 }
8360
8361 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8362 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8363 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8364 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8365 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8366
8367 return instof_false; // even if it is null
8368 }
8369
8370 //------------------------------inline_ghash_processBlocks
8371 bool LibraryCallKit::inline_ghash_processBlocks() {
8372 address stubAddr;
8373 const char *stubName;
8374 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
8375
8376 stubAddr = StubRoutines::ghash_processBlocks();
8377 stubName = "ghash_processBlocks";
8378
8379 Node* data = argument(0);
8380 Node* offset = argument(1);
8381 Node* len = argument(2);
8382 Node* state = argument(3);
8383 Node* subkeyH = argument(4);
8384
8385 state = must_be_not_null(state, true);
8386 subkeyH = must_be_not_null(subkeyH, true);
8387 data = must_be_not_null(data, true);
8388
8389 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8390 assert(state_start, "state is null");
8391 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
8392 assert(subkeyH_start, "subkeyH is null");
8393 Node* data_start = array_element_address(data, offset, T_BYTE);
8394 assert(data_start, "data is null");
8395
8396 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
8397 OptoRuntime::ghash_processBlocks_Type(),
8398 stubAddr, stubName, TypePtr::BOTTOM,
8399 state_start, subkeyH_start, data_start, len);
8400 return true;
8401 }
8402
8403 //------------------------------inline_chacha20Block
8404 bool LibraryCallKit::inline_chacha20Block() {
8405 address stubAddr;
8406 const char *stubName;
8407 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
8408
8409 stubAddr = StubRoutines::chacha20Block();
8410 stubName = "chacha20Block";
8411
8412 Node* state = argument(0);
8413 Node* result = argument(1);
8414
8415 state = must_be_not_null(state, true);
8416 result = must_be_not_null(result, true);
8417
8418 Node* state_start = array_element_address(state, intcon(0), T_INT);
8419 assert(state_start, "state is null");
8420 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
8421 assert(result_start, "result is null");
8422
8423 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
8424 OptoRuntime::chacha20Block_Type(),
8425 stubAddr, stubName, TypePtr::BOTTOM,
8426 state_start, result_start);
8427 // return key stream length (int)
8428 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
8429 set_result(retvalue);
8430 return true;
8431 }
8432
8433 //------------------------------inline_kyberNtt
8434 bool LibraryCallKit::inline_kyberNtt() {
8435 address stubAddr;
8436 const char *stubName;
8437 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8438 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
8439
8440 stubAddr = StubRoutines::kyberNtt();
8441 stubName = "kyberNtt";
8442 if (!stubAddr) return false;
8443
8444 Node* coeffs = argument(0);
8445 Node* ntt_zetas = argument(1);
8446
8447 coeffs = must_be_not_null(coeffs, true);
8448 ntt_zetas = must_be_not_null(ntt_zetas, true);
8449
8450 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8451 assert(coeffs_start, "coeffs is null");
8452 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
8453 assert(ntt_zetas_start, "ntt_zetas is null");
8454 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8455 OptoRuntime::kyberNtt_Type(),
8456 stubAddr, stubName, TypePtr::BOTTOM,
8457 coeffs_start, ntt_zetas_start);
8458 // return an int
8459 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
8460 set_result(retvalue);
8461 return true;
8462 }
8463
8464 //------------------------------inline_kyberInverseNtt
8465 bool LibraryCallKit::inline_kyberInverseNtt() {
8466 address stubAddr;
8467 const char *stubName;
8468 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8469 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
8470
8471 stubAddr = StubRoutines::kyberInverseNtt();
8472 stubName = "kyberInverseNtt";
8473 if (!stubAddr) return false;
8474
8475 Node* coeffs = argument(0);
8476 Node* zetas = argument(1);
8477
8478 coeffs = must_be_not_null(coeffs, true);
8479 zetas = must_be_not_null(zetas, true);
8480
8481 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8482 assert(coeffs_start, "coeffs is null");
8483 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8484 assert(zetas_start, "inverseNtt_zetas is null");
8485 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8486 OptoRuntime::kyberInverseNtt_Type(),
8487 stubAddr, stubName, TypePtr::BOTTOM,
8488 coeffs_start, zetas_start);
8489
8490 // return an int
8491 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
8492 set_result(retvalue);
8493 return true;
8494 }
8495
8496 //------------------------------inline_kyberNttMult
8497 bool LibraryCallKit::inline_kyberNttMult() {
8498 address stubAddr;
8499 const char *stubName;
8500 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8501 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
8502
8503 stubAddr = StubRoutines::kyberNttMult();
8504 stubName = "kyberNttMult";
8505 if (!stubAddr) return false;
8506
8507 Node* result = argument(0);
8508 Node* ntta = argument(1);
8509 Node* nttb = argument(2);
8510 Node* zetas = argument(3);
8511
8512 result = must_be_not_null(result, true);
8513 ntta = must_be_not_null(ntta, true);
8514 nttb = must_be_not_null(nttb, true);
8515 zetas = must_be_not_null(zetas, true);
8516
8517 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8518 assert(result_start, "result is null");
8519 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
8520 assert(ntta_start, "ntta is null");
8521 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
8522 assert(nttb_start, "nttb is null");
8523 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8524 assert(zetas_start, "nttMult_zetas is null");
8525 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8526 OptoRuntime::kyberNttMult_Type(),
8527 stubAddr, stubName, TypePtr::BOTTOM,
8528 result_start, ntta_start, nttb_start,
8529 zetas_start);
8530
8531 // return an int
8532 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
8533 set_result(retvalue);
8534
8535 return true;
8536 }
8537
8538 //------------------------------inline_kyberAddPoly_2
8539 bool LibraryCallKit::inline_kyberAddPoly_2() {
8540 address stubAddr;
8541 const char *stubName;
8542 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8543 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
8544
8545 stubAddr = StubRoutines::kyberAddPoly_2();
8546 stubName = "kyberAddPoly_2";
8547 if (!stubAddr) return false;
8548
8549 Node* result = argument(0);
8550 Node* a = argument(1);
8551 Node* b = argument(2);
8552
8553 result = must_be_not_null(result, true);
8554 a = must_be_not_null(a, true);
8555 b = must_be_not_null(b, true);
8556
8557 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8558 assert(result_start, "result is null");
8559 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8560 assert(a_start, "a is null");
8561 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8562 assert(b_start, "b is null");
8563 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
8564 OptoRuntime::kyberAddPoly_2_Type(),
8565 stubAddr, stubName, TypePtr::BOTTOM,
8566 result_start, a_start, b_start);
8567 // return an int
8568 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
8569 set_result(retvalue);
8570 return true;
8571 }
8572
8573 //------------------------------inline_kyberAddPoly_3
8574 bool LibraryCallKit::inline_kyberAddPoly_3() {
8575 address stubAddr;
8576 const char *stubName;
8577 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8578 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
8579
8580 stubAddr = StubRoutines::kyberAddPoly_3();
8581 stubName = "kyberAddPoly_3";
8582 if (!stubAddr) return false;
8583
8584 Node* result = argument(0);
8585 Node* a = argument(1);
8586 Node* b = argument(2);
8587 Node* c = argument(3);
8588
8589 result = must_be_not_null(result, true);
8590 a = must_be_not_null(a, true);
8591 b = must_be_not_null(b, true);
8592 c = must_be_not_null(c, true);
8593
8594 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8595 assert(result_start, "result is null");
8596 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8597 assert(a_start, "a is null");
8598 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8599 assert(b_start, "b is null");
8600 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
8601 assert(c_start, "c is null");
8602 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
8603 OptoRuntime::kyberAddPoly_3_Type(),
8604 stubAddr, stubName, TypePtr::BOTTOM,
8605 result_start, a_start, b_start, c_start);
8606 // return an int
8607 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
8608 set_result(retvalue);
8609 return true;
8610 }
8611
8612 //------------------------------inline_kyber12To16
8613 bool LibraryCallKit::inline_kyber12To16() {
8614 address stubAddr;
8615 const char *stubName;
8616 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8617 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
8618
8619 stubAddr = StubRoutines::kyber12To16();
8620 stubName = "kyber12To16";
8621 if (!stubAddr) return false;
8622
8623 Node* condensed = argument(0);
8624 Node* condensedOffs = argument(1);
8625 Node* parsed = argument(2);
8626 Node* parsedLength = argument(3);
8627
8628 condensed = must_be_not_null(condensed, true);
8629 parsed = must_be_not_null(parsed, true);
8630
8631 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
8632 assert(condensed_start, "condensed is null");
8633 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
8634 assert(parsed_start, "parsed is null");
8635 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
8636 OptoRuntime::kyber12To16_Type(),
8637 stubAddr, stubName, TypePtr::BOTTOM,
8638 condensed_start, condensedOffs, parsed_start, parsedLength);
8639 // return an int
8640 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
8641 set_result(retvalue);
8642 return true;
8643
8644 }
8645
8646 //------------------------------inline_kyberBarrettReduce
8647 bool LibraryCallKit::inline_kyberBarrettReduce() {
8648 address stubAddr;
8649 const char *stubName;
8650 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8651 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
8652
8653 stubAddr = StubRoutines::kyberBarrettReduce();
8654 stubName = "kyberBarrettReduce";
8655 if (!stubAddr) return false;
8656
8657 Node* coeffs = argument(0);
8658
8659 coeffs = must_be_not_null(coeffs, true);
8660
8661 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8662 assert(coeffs_start, "coeffs is null");
8663 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
8664 OptoRuntime::kyberBarrettReduce_Type(),
8665 stubAddr, stubName, TypePtr::BOTTOM,
8666 coeffs_start);
8667 // return an int
8668 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
8669 set_result(retvalue);
8670 return true;
8671 }
8672
8673 //------------------------------inline_dilithiumAlmostNtt
8674 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
8675 address stubAddr;
8676 const char *stubName;
8677 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8678 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
8679
8680 stubAddr = StubRoutines::dilithiumAlmostNtt();
8681 stubName = "dilithiumAlmostNtt";
8682 if (!stubAddr) return false;
8683
8684 Node* coeffs = argument(0);
8685 Node* ntt_zetas = argument(1);
8686
8687 coeffs = must_be_not_null(coeffs, true);
8688 ntt_zetas = must_be_not_null(ntt_zetas, true);
8689
8690 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8691 assert(coeffs_start, "coeffs is null");
8692 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
8693 assert(ntt_zetas_start, "ntt_zetas is null");
8694 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8695 OptoRuntime::dilithiumAlmostNtt_Type(),
8696 stubAddr, stubName, TypePtr::BOTTOM,
8697 coeffs_start, ntt_zetas_start);
8698 // return an int
8699 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
8700 set_result(retvalue);
8701 return true;
8702 }
8703
8704 //------------------------------inline_dilithiumAlmostInverseNtt
8705 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
8706 address stubAddr;
8707 const char *stubName;
8708 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8709 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
8710
8711 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
8712 stubName = "dilithiumAlmostInverseNtt";
8713 if (!stubAddr) return false;
8714
8715 Node* coeffs = argument(0);
8716 Node* zetas = argument(1);
8717
8718 coeffs = must_be_not_null(coeffs, true);
8719 zetas = must_be_not_null(zetas, true);
8720
8721 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8722 assert(coeffs_start, "coeffs is null");
8723 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
8724 assert(zetas_start, "inverseNtt_zetas is null");
8725 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8726 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
8727 stubAddr, stubName, TypePtr::BOTTOM,
8728 coeffs_start, zetas_start);
8729 // return an int
8730 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
8731 set_result(retvalue);
8732 return true;
8733 }
8734
8735 //------------------------------inline_dilithiumNttMult
8736 bool LibraryCallKit::inline_dilithiumNttMult() {
8737 address stubAddr;
8738 const char *stubName;
8739 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8740 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
8741
8742 stubAddr = StubRoutines::dilithiumNttMult();
8743 stubName = "dilithiumNttMult";
8744 if (!stubAddr) return false;
8745
8746 Node* result = argument(0);
8747 Node* ntta = argument(1);
8748 Node* nttb = argument(2);
8749 Node* zetas = argument(3);
8750
8751 result = must_be_not_null(result, true);
8752 ntta = must_be_not_null(ntta, true);
8753 nttb = must_be_not_null(nttb, true);
8754 zetas = must_be_not_null(zetas, true);
8755
8756 Node* result_start = array_element_address(result, intcon(0), T_INT);
8757 assert(result_start, "result is null");
8758 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
8759 assert(ntta_start, "ntta is null");
8760 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
8761 assert(nttb_start, "nttb is null");
8762 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8763 OptoRuntime::dilithiumNttMult_Type(),
8764 stubAddr, stubName, TypePtr::BOTTOM,
8765 result_start, ntta_start, nttb_start);
8766
8767 // return an int
8768 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8769 set_result(retvalue);
8770
8771 return true;
8772 }
8773
8774 //------------------------------inline_dilithiumMontMulByConstant
8775 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8776 address stubAddr;
8777 const char *stubName;
8778 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8779 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8780
8781 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8782 stubName = "dilithiumMontMulByConstant";
8783 if (!stubAddr) return false;
8784
8785 Node* coeffs = argument(0);
8786 Node* constant = argument(1);
8787
8788 coeffs = must_be_not_null(coeffs, true);
8789
8790 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8791 assert(coeffs_start, "coeffs is null");
8792 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8793 OptoRuntime::dilithiumMontMulByConstant_Type(),
8794 stubAddr, stubName, TypePtr::BOTTOM,
8795 coeffs_start, constant);
8796
8797 // return an int
8798 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8799 set_result(retvalue);
8800 return true;
8801 }
8802
8803
8804 //------------------------------inline_dilithiumDecomposePoly
8805 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8806 address stubAddr;
8807 const char *stubName;
8808 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8809 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8810
8811 stubAddr = StubRoutines::dilithiumDecomposePoly();
8812 stubName = "dilithiumDecomposePoly";
8813 if (!stubAddr) return false;
8814
8815 Node* input = argument(0);
8816 Node* lowPart = argument(1);
8817 Node* highPart = argument(2);
8818 Node* twoGamma2 = argument(3);
8819 Node* multiplier = argument(4);
8820
8821 input = must_be_not_null(input, true);
8822 lowPart = must_be_not_null(lowPart, true);
8823 highPart = must_be_not_null(highPart, true);
8824
8825 Node* input_start = array_element_address(input, intcon(0), T_INT);
8826 assert(input_start, "input is null");
8827 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8828 assert(lowPart_start, "lowPart is null");
8829 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8830 assert(highPart_start, "highPart is null");
8831
8832 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8833 OptoRuntime::dilithiumDecomposePoly_Type(),
8834 stubAddr, stubName, TypePtr::BOTTOM,
8835 input_start, lowPart_start, highPart_start,
8836 twoGamma2, multiplier);
8837
8838 // return an int
8839 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8840 set_result(retvalue);
8841 return true;
8842 }
8843
8844 bool LibraryCallKit::inline_base64_encodeBlock() {
8845 address stubAddr;
8846 const char *stubName;
8847 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8848 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8849 stubAddr = StubRoutines::base64_encodeBlock();
8850 stubName = "encodeBlock";
8851
8852 if (!stubAddr) return false;
8853 Node* base64obj = argument(0);
8854 Node* src = argument(1);
8855 Node* offset = argument(2);
8856 Node* len = argument(3);
8857 Node* dest = argument(4);
8858 Node* dp = argument(5);
8859 Node* isURL = argument(6);
8860
8861 src = must_be_not_null(src, true);
8862 dest = must_be_not_null(dest, true);
8863
8864 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8865 assert(src_start, "source array is null");
8866 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8867 assert(dest_start, "destination array is null");
8868
8869 Node* base64 = make_runtime_call(RC_LEAF,
8870 OptoRuntime::base64_encodeBlock_Type(),
8871 stubAddr, stubName, TypePtr::BOTTOM,
8872 src_start, offset, len, dest_start, dp, isURL);
8873 return true;
8874 }
8875
8876 bool LibraryCallKit::inline_base64_decodeBlock() {
8877 address stubAddr;
8878 const char *stubName;
8879 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8880 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8881 stubAddr = StubRoutines::base64_decodeBlock();
8882 stubName = "decodeBlock";
8883
8884 if (!stubAddr) return false;
8885 Node* base64obj = argument(0);
8886 Node* src = argument(1);
8887 Node* src_offset = argument(2);
8888 Node* len = argument(3);
8889 Node* dest = argument(4);
8890 Node* dest_offset = argument(5);
8891 Node* isURL = argument(6);
8892 Node* isMIME = argument(7);
8893
8894 src = must_be_not_null(src, true);
8895 dest = must_be_not_null(dest, true);
8896
8897 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8898 assert(src_start, "source array is null");
8899 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8900 assert(dest_start, "destination array is null");
8901
8902 Node* call = make_runtime_call(RC_LEAF,
8903 OptoRuntime::base64_decodeBlock_Type(),
8904 stubAddr, stubName, TypePtr::BOTTOM,
8905 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8906 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8907 set_result(result);
8908 return true;
8909 }
8910
8911 bool LibraryCallKit::inline_poly1305_processBlocks() {
8912 address stubAddr;
8913 const char *stubName;
8914 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8915 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8916 stubAddr = StubRoutines::poly1305_processBlocks();
8917 stubName = "poly1305_processBlocks";
8918
8919 if (!stubAddr) return false;
8920 null_check_receiver(); // null-check receiver
8921 if (stopped()) return true;
8922
8923 Node* input = argument(1);
8924 Node* input_offset = argument(2);
8925 Node* len = argument(3);
8926 Node* alimbs = argument(4);
8927 Node* rlimbs = argument(5);
8928
8929 input = must_be_not_null(input, true);
8930 alimbs = must_be_not_null(alimbs, true);
8931 rlimbs = must_be_not_null(rlimbs, true);
8932
8933 Node* input_start = array_element_address(input, input_offset, T_BYTE);
8934 assert(input_start, "input array is null");
8935 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
8936 assert(acc_start, "acc array is null");
8937 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
8938 assert(r_start, "r array is null");
8939
8940 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8941 OptoRuntime::poly1305_processBlocks_Type(),
8942 stubAddr, stubName, TypePtr::BOTTOM,
8943 input_start, len, acc_start, r_start);
8944 return true;
8945 }
8946
8947 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
8948 address stubAddr;
8949 const char *stubName;
8950 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8951 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
8952 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
8953 stubName = "intpoly_montgomeryMult_P256";
8954
8955 if (!stubAddr) return false;
8956 null_check_receiver(); // null-check receiver
8957 if (stopped()) return true;
8958
8959 Node* a = argument(1);
8960 Node* b = argument(2);
8961 Node* r = argument(3);
8962
8963 a = must_be_not_null(a, true);
8964 b = must_be_not_null(b, true);
8965 r = must_be_not_null(r, true);
8966
8967 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8968 assert(a_start, "a array is null");
8969 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8970 assert(b_start, "b array is null");
8971 Node* r_start = array_element_address(r, intcon(0), T_LONG);
8972 assert(r_start, "r array is null");
8973
8974 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8975 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
8976 stubAddr, stubName, TypePtr::BOTTOM,
8977 a_start, b_start, r_start);
8978 return true;
8979 }
8980
8981 bool LibraryCallKit::inline_intpoly_assign() {
8982 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8983 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
8984 const char *stubName = "intpoly_assign";
8985 address stubAddr = StubRoutines::intpoly_assign();
8986 if (!stubAddr) return false;
8987
8988 Node* set = argument(0);
8989 Node* a = argument(1);
8990 Node* b = argument(2);
8991 Node* arr_length = load_array_length(a);
8992
8993 a = must_be_not_null(a, true);
8994 b = must_be_not_null(b, true);
8995
8996 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8997 assert(a_start, "a array is null");
8998 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8999 assert(b_start, "b array is null");
9000
9001 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
9002 OptoRuntime::intpoly_assign_Type(),
9003 stubAddr, stubName, TypePtr::BOTTOM,
9004 set, a_start, b_start, arr_length);
9005 return true;
9006 }
9007
9008 //------------------------------inline_digestBase_implCompress-----------------------
9009 //
9010 // Calculate MD5 for single-block byte[] array.
9011 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
9012 //
9013 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
9014 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
9015 //
9016 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
9017 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
9018 //
9019 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
9020 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
9021 //
9022 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
9023 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
9024 //
9025 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
9026 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
9027
9028 Node* digestBase_obj = argument(0);
9029 Node* src = argument(1); // type oop
9030 Node* ofs = argument(2); // type int
9031
9032 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
9033 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
9034 // failed array check
9035 return false;
9036 }
9037 // Figure out the size and type of the elements we will be copying.
9038 BasicType src_elem = src_type->elem()->array_element_basic_type();
9039 if (src_elem != T_BYTE) {
9040 return false;
9041 }
9042 // 'src_start' points to src array + offset
9043 src = must_be_not_null(src, true);
9044 Node* src_start = array_element_address(src, ofs, src_elem);
9045 Node* state = nullptr;
9046 Node* block_size = nullptr;
9047 address stubAddr;
9048 const char *stubName;
9049
9050 switch(id) {
9051 case vmIntrinsics::_md5_implCompress:
9052 assert(UseMD5Intrinsics, "need MD5 instruction support");
9053 state = get_state_from_digest_object(digestBase_obj, T_INT);
9054 stubAddr = StubRoutines::md5_implCompress();
9055 stubName = "md5_implCompress";
9056 break;
9057 case vmIntrinsics::_sha_implCompress:
9058 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
9059 state = get_state_from_digest_object(digestBase_obj, T_INT);
9060 stubAddr = StubRoutines::sha1_implCompress();
9061 stubName = "sha1_implCompress";
9062 break;
9063 case vmIntrinsics::_sha2_implCompress:
9064 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
9065 state = get_state_from_digest_object(digestBase_obj, T_INT);
9066 stubAddr = StubRoutines::sha256_implCompress();
9067 stubName = "sha256_implCompress";
9068 break;
9069 case vmIntrinsics::_sha5_implCompress:
9070 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
9071 state = get_state_from_digest_object(digestBase_obj, T_LONG);
9072 stubAddr = StubRoutines::sha512_implCompress();
9073 stubName = "sha512_implCompress";
9074 break;
9075 case vmIntrinsics::_sha3_implCompress:
9076 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
9077 state = get_state_from_digest_object(digestBase_obj, T_LONG);
9078 stubAddr = StubRoutines::sha3_implCompress();
9079 stubName = "sha3_implCompress";
9080 block_size = get_block_size_from_digest_object(digestBase_obj);
9081 if (block_size == nullptr) return false;
9082 break;
9083 default:
9084 fatal_unexpected_iid(id);
9085 return false;
9086 }
9087 if (state == nullptr) return false;
9088
9089 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
9090 if (stubAddr == nullptr) return false;
9091
9092 // Call the stub.
9093 Node* call;
9094 if (block_size == nullptr) {
9095 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
9096 stubAddr, stubName, TypePtr::BOTTOM,
9097 src_start, state);
9098 } else {
9099 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
9100 stubAddr, stubName, TypePtr::BOTTOM,
9101 src_start, state, block_size);
9102 }
9103
9104 return true;
9105 }
9106
9107 //------------------------------inline_double_keccak
9108 bool LibraryCallKit::inline_double_keccak() {
9109 address stubAddr;
9110 const char *stubName;
9111 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
9112 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
9113
9114 stubAddr = StubRoutines::double_keccak();
9115 stubName = "double_keccak";
9116 if (!stubAddr) return false;
9117
9118 Node* status0 = argument(0);
9119 Node* status1 = argument(1);
9120
9121 status0 = must_be_not_null(status0, true);
9122 status1 = must_be_not_null(status1, true);
9123
9124 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
9125 assert(status0_start, "status0 is null");
9126 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
9127 assert(status1_start, "status1 is null");
9128 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
9129 OptoRuntime::double_keccak_Type(),
9130 stubAddr, stubName, TypePtr::BOTTOM,
9131 status0_start, status1_start);
9132 // return an int
9133 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
9134 set_result(retvalue);
9135 return true;
9136 }
9137
9138
9139 //------------------------------inline_digestBase_implCompressMB-----------------------
9140 //
9141 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
9142 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
9143 //
9144 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
9145 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9146 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9147 assert((uint)predicate < 5, "sanity");
9148 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
9149
9150 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
9151 Node* src = argument(1); // byte[] array
9152 Node* ofs = argument(2); // type int
9153 Node* limit = argument(3); // type int
9154
9155 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
9156 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
9157 // failed array check
9158 return false;
9159 }
9160 // Figure out the size and type of the elements we will be copying.
9161 BasicType src_elem = src_type->elem()->array_element_basic_type();
9162 if (src_elem != T_BYTE) {
9163 return false;
9164 }
9165 // 'src_start' points to src array + offset
9166 src = must_be_not_null(src, false);
9167 Node* src_start = array_element_address(src, ofs, src_elem);
9168
9169 const char* klass_digestBase_name = nullptr;
9170 const char* stub_name = nullptr;
9171 address stub_addr = nullptr;
9172 BasicType elem_type = T_INT;
9173
9174 switch (predicate) {
9175 case 0:
9176 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
9177 klass_digestBase_name = "sun/security/provider/MD5";
9178 stub_name = "md5_implCompressMB";
9179 stub_addr = StubRoutines::md5_implCompressMB();
9180 }
9181 break;
9182 case 1:
9183 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
9184 klass_digestBase_name = "sun/security/provider/SHA";
9185 stub_name = "sha1_implCompressMB";
9186 stub_addr = StubRoutines::sha1_implCompressMB();
9187 }
9188 break;
9189 case 2:
9190 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
9191 klass_digestBase_name = "sun/security/provider/SHA2";
9192 stub_name = "sha256_implCompressMB";
9193 stub_addr = StubRoutines::sha256_implCompressMB();
9194 }
9195 break;
9196 case 3:
9197 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
9198 klass_digestBase_name = "sun/security/provider/SHA5";
9199 stub_name = "sha512_implCompressMB";
9200 stub_addr = StubRoutines::sha512_implCompressMB();
9201 elem_type = T_LONG;
9202 }
9203 break;
9204 case 4:
9205 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
9206 klass_digestBase_name = "sun/security/provider/SHA3";
9207 stub_name = "sha3_implCompressMB";
9208 stub_addr = StubRoutines::sha3_implCompressMB();
9209 elem_type = T_LONG;
9210 }
9211 break;
9212 default:
9213 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
9214 }
9215 if (klass_digestBase_name != nullptr) {
9216 assert(stub_addr != nullptr, "Stub is generated");
9217 if (stub_addr == nullptr) return false;
9218
9219 // get DigestBase klass to lookup for SHA klass
9220 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
9221 assert(tinst != nullptr, "digestBase_obj is not instance???");
9222 assert(tinst->is_loaded(), "DigestBase is not loaded");
9223
9224 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
9225 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
9226 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
9227 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
9228 }
9229 return false;
9230 }
9231
9232 //------------------------------inline_digestBase_implCompressMB-----------------------
9233 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
9234 BasicType elem_type, address stubAddr, const char *stubName,
9235 Node* src_start, Node* ofs, Node* limit) {
9236 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
9237 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
9238 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
9239 digest_obj = _gvn.transform(digest_obj);
9240
9241 Node* state = get_state_from_digest_object(digest_obj, elem_type);
9242 if (state == nullptr) return false;
9243
9244 Node* block_size = nullptr;
9245 if (strcmp("sha3_implCompressMB", stubName) == 0) {
9246 block_size = get_block_size_from_digest_object(digest_obj);
9247 if (block_size == nullptr) return false;
9248 }
9249
9250 // Call the stub.
9251 Node* call;
9252 if (block_size == nullptr) {
9253 call = make_runtime_call(RC_LEAF|RC_NO_FP,
9254 OptoRuntime::digestBase_implCompressMB_Type(false),
9255 stubAddr, stubName, TypePtr::BOTTOM,
9256 src_start, state, ofs, limit);
9257 } else {
9258 call = make_runtime_call(RC_LEAF|RC_NO_FP,
9259 OptoRuntime::digestBase_implCompressMB_Type(true),
9260 stubAddr, stubName, TypePtr::BOTTOM,
9261 src_start, state, block_size, ofs, limit);
9262 }
9263
9264 // return ofs (int)
9265 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
9266 set_result(result);
9267
9268 return true;
9269 }
9270
9271 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
9272 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
9273 assert(UseAES, "need AES instruction support");
9274 address stubAddr = nullptr;
9275 const char *stubName = nullptr;
9276 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
9277 stubName = "galoisCounterMode_AESCrypt";
9278
9279 if (stubAddr == nullptr) return false;
9280
9281 Node* in = argument(0);
9282 Node* inOfs = argument(1);
9283 Node* len = argument(2);
9284 Node* ct = argument(3);
9285 Node* ctOfs = argument(4);
9286 Node* out = argument(5);
9287 Node* outOfs = argument(6);
9288 Node* gctr_object = argument(7);
9289 Node* ghash_object = argument(8);
9290
9291 // (1) in, ct and out are arrays.
9292 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
9293 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
9294 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
9295 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
9296 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
9297 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
9298
9299 // checks are the responsibility of the caller
9300 Node* in_start = in;
9301 Node* ct_start = ct;
9302 Node* out_start = out;
9303 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
9304 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
9305 in_start = array_element_address(in, inOfs, T_BYTE);
9306 ct_start = array_element_address(ct, ctOfs, T_BYTE);
9307 out_start = array_element_address(out, outOfs, T_BYTE);
9308 }
9309
9310 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
9311 // (because of the predicated logic executed earlier).
9312 // so we cast it here safely.
9313 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
9314 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9315 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
9316 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
9317 Node* state = load_field_from_object(ghash_object, "state", "[J");
9318
9319 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
9320 return false;
9321 }
9322 // cast it to what we know it will be at runtime
9323 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
9324 assert(tinst != nullptr, "GCTR obj is null");
9325 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9326 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
9327 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
9328 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9329 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
9330 const TypeOopPtr* xtype = aklass->as_instance_type();
9331 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
9332 aescrypt_object = _gvn.transform(aescrypt_object);
9333 // we need to get the start of the aescrypt_object's expanded key array
9334 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
9335 if (k_start == nullptr) return false;
9336 // similarly, get the start address of the r vector
9337 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
9338 Node* state_start = array_element_address(state, intcon(0), T_LONG);
9339 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
9340
9341
9342 // Call the stub, passing params
9343 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
9344 OptoRuntime::galoisCounterMode_aescrypt_Type(),
9345 stubAddr, stubName, TypePtr::BOTTOM,
9346 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
9347
9348 // return cipher length (int)
9349 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
9350 set_result(retvalue);
9351
9352 return true;
9353 }
9354
9355 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
9356 // Return node representing slow path of predicate check.
9357 // the pseudo code we want to emulate with this predicate is:
9358 // for encryption:
9359 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
9360 // for decryption:
9361 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
9362 // note cipher==plain is more conservative than the original java code but that's OK
9363 //
9364
9365 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
9366 // The receiver was checked for null already.
9367 Node* objGCTR = argument(7);
9368 // Load embeddedCipher field of GCTR object.
9369 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9370 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
9371
9372 // get AESCrypt klass for instanceOf check
9373 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
9374 // will have same classloader as CipherBlockChaining object
9375 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
9376 assert(tinst != nullptr, "GCTR obj is null");
9377 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9378
9379 // we want to do an instanceof comparison against the AESCrypt class
9380 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
9381 if (!klass_AESCrypt->is_loaded()) {
9382 // if AESCrypt is not even loaded, we never take the intrinsic fast path
9383 Node* ctrl = control();
9384 set_control(top()); // no regular fast path
9385 return ctrl;
9386 }
9387
9388 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9389 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
9390 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9391 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9392 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9393
9394 return instof_false; // even if it is null
9395 }
9396
9397 //------------------------------get_state_from_digest_object-----------------------
9398 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
9399 const char* state_type;
9400 switch (elem_type) {
9401 case T_BYTE: state_type = "[B"; break;
9402 case T_INT: state_type = "[I"; break;
9403 case T_LONG: state_type = "[J"; break;
9404 default: ShouldNotReachHere();
9405 }
9406 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
9407 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
9408 if (digest_state == nullptr) return (Node *) nullptr;
9409
9410 // now have the array, need to get the start address of the state array
9411 Node* state = array_element_address(digest_state, intcon(0), elem_type);
9412 return state;
9413 }
9414
9415 //------------------------------get_block_size_from_sha3_object----------------------------------
9416 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
9417 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
9418 assert (block_size != nullptr, "sanity");
9419 return block_size;
9420 }
9421
9422 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
9423 // Return node representing slow path of predicate check.
9424 // the pseudo code we want to emulate with this predicate is:
9425 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
9426 //
9427 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
9428 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9429 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9430 assert((uint)predicate < 5, "sanity");
9431
9432 // The receiver was checked for null already.
9433 Node* digestBaseObj = argument(0);
9434
9435 // get DigestBase klass for instanceOf check
9436 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
9437 assert(tinst != nullptr, "digestBaseObj is null");
9438 assert(tinst->is_loaded(), "DigestBase is not loaded");
9439
9440 const char* klass_name = nullptr;
9441 switch (predicate) {
9442 case 0:
9443 if (UseMD5Intrinsics) {
9444 // we want to do an instanceof comparison against the MD5 class
9445 klass_name = "sun/security/provider/MD5";
9446 }
9447 break;
9448 case 1:
9449 if (UseSHA1Intrinsics) {
9450 // we want to do an instanceof comparison against the SHA class
9451 klass_name = "sun/security/provider/SHA";
9452 }
9453 break;
9454 case 2:
9455 if (UseSHA256Intrinsics) {
9456 // we want to do an instanceof comparison against the SHA2 class
9457 klass_name = "sun/security/provider/SHA2";
9458 }
9459 break;
9460 case 3:
9461 if (UseSHA512Intrinsics) {
9462 // we want to do an instanceof comparison against the SHA5 class
9463 klass_name = "sun/security/provider/SHA5";
9464 }
9465 break;
9466 case 4:
9467 if (UseSHA3Intrinsics) {
9468 // we want to do an instanceof comparison against the SHA3 class
9469 klass_name = "sun/security/provider/SHA3";
9470 }
9471 break;
9472 default:
9473 fatal("unknown SHA intrinsic predicate: %d", predicate);
9474 }
9475
9476 ciKlass* klass = nullptr;
9477 if (klass_name != nullptr) {
9478 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
9479 }
9480 if ((klass == nullptr) || !klass->is_loaded()) {
9481 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
9482 Node* ctrl = control();
9483 set_control(top()); // no intrinsic path
9484 return ctrl;
9485 }
9486 ciInstanceKlass* instklass = klass->as_instance_klass();
9487
9488 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
9489 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9490 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9491 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9492
9493 return instof_false; // even if it is null
9494 }
9495
9496 //-------------inline_fma-----------------------------------
9497 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
9498 Node *a = nullptr;
9499 Node *b = nullptr;
9500 Node *c = nullptr;
9501 Node* result = nullptr;
9502 switch (id) {
9503 case vmIntrinsics::_fmaD:
9504 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
9505 // no receiver since it is static method
9506 a = argument(0);
9507 b = argument(2);
9508 c = argument(4);
9509 result = _gvn.transform(new FmaDNode(a, b, c));
9510 break;
9511 case vmIntrinsics::_fmaF:
9512 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
9513 a = argument(0);
9514 b = argument(1);
9515 c = argument(2);
9516 result = _gvn.transform(new FmaFNode(a, b, c));
9517 break;
9518 default:
9519 fatal_unexpected_iid(id); break;
9520 }
9521 set_result(result);
9522 return true;
9523 }
9524
9525 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
9526 // argument(0) is receiver
9527 Node* codePoint = argument(1);
9528 Node* n = nullptr;
9529
9530 switch (id) {
9531 case vmIntrinsics::_isDigit :
9532 n = new DigitNode(control(), codePoint);
9533 break;
9534 case vmIntrinsics::_isLowerCase :
9535 n = new LowerCaseNode(control(), codePoint);
9536 break;
9537 case vmIntrinsics::_isUpperCase :
9538 n = new UpperCaseNode(control(), codePoint);
9539 break;
9540 case vmIntrinsics::_isWhitespace :
9541 n = new WhitespaceNode(control(), codePoint);
9542 break;
9543 default:
9544 fatal_unexpected_iid(id);
9545 }
9546
9547 set_result(_gvn.transform(n));
9548 return true;
9549 }
9550
9551 bool LibraryCallKit::inline_profileBoolean() {
9552 Node* counts = argument(1);
9553 const TypeAryPtr* ary = nullptr;
9554 ciArray* aobj = nullptr;
9555 if (counts->is_Con()
9556 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
9557 && (aobj = ary->const_oop()->as_array()) != nullptr
9558 && (aobj->length() == 2)) {
9559 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
9560 jint false_cnt = aobj->element_value(0).as_int();
9561 jint true_cnt = aobj->element_value(1).as_int();
9562
9563 if (C->log() != nullptr) {
9564 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
9565 false_cnt, true_cnt);
9566 }
9567
9568 if (false_cnt + true_cnt == 0) {
9569 // According to profile, never executed.
9570 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9571 Deoptimization::Action_reinterpret);
9572 return true;
9573 }
9574
9575 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
9576 // is a number of each value occurrences.
9577 Node* result = argument(0);
9578 if (false_cnt == 0 || true_cnt == 0) {
9579 // According to profile, one value has been never seen.
9580 int expected_val = (false_cnt == 0) ? 1 : 0;
9581
9582 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
9583 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
9584
9585 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
9586 Node* fast_path = _gvn.transform(new IfTrueNode(check));
9587 Node* slow_path = _gvn.transform(new IfFalseNode(check));
9588
9589 { // Slow path: uncommon trap for never seen value and then reexecute
9590 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
9591 // the value has been seen at least once.
9592 PreserveJVMState pjvms(this);
9593 PreserveReexecuteState preexecs(this);
9594 jvms()->set_should_reexecute(true);
9595
9596 set_control(slow_path);
9597 set_i_o(i_o());
9598
9599 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9600 Deoptimization::Action_reinterpret);
9601 }
9602 // The guard for never seen value enables sharpening of the result and
9603 // returning a constant. It allows to eliminate branches on the same value
9604 // later on.
9605 set_control(fast_path);
9606 result = intcon(expected_val);
9607 }
9608 // Stop profiling.
9609 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
9610 // By replacing method body with profile data (represented as ProfileBooleanNode
9611 // on IR level) we effectively disable profiling.
9612 // It enables full speed execution once optimized code is generated.
9613 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
9614 C->record_for_igvn(profile);
9615 set_result(profile);
9616 return true;
9617 } else {
9618 // Continue profiling.
9619 // Profile data isn't available at the moment. So, execute method's bytecode version.
9620 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
9621 // is compiled and counters aren't available since corresponding MethodHandle
9622 // isn't a compile-time constant.
9623 return false;
9624 }
9625 }
9626
9627 bool LibraryCallKit::inline_isCompileConstant() {
9628 Node* n = argument(0);
9629 set_result(n->is_Con() ? intcon(1) : intcon(0));
9630 return true;
9631 }
9632
9633 //------------------------------- inline_getObjectSize --------------------------------------
9634 //
9635 // Calculate the runtime size of the object/array.
9636 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
9637 //
9638 bool LibraryCallKit::inline_getObjectSize() {
9639 Node* obj = argument(3);
9640 Node* klass_node = load_object_klass(obj);
9641
9642 jint layout_con = Klass::_lh_neutral_value;
9643 Node* layout_val = get_layout_helper(klass_node, layout_con);
9644 int layout_is_con = (layout_val == nullptr);
9645
9646 if (layout_is_con) {
9647 // Layout helper is constant, can figure out things at compile time.
9648
9649 if (Klass::layout_helper_is_instance(layout_con)) {
9650 // Instance case: layout_con contains the size itself.
9651 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
9652 set_result(size);
9653 } else {
9654 // Array case: size is round(header + element_size*arraylength).
9655 // Since arraylength is different for every array instance, we have to
9656 // compute the whole thing at runtime.
9657
9658 Node* arr_length = load_array_length(obj);
9659
9660 int round_mask = MinObjAlignmentInBytes - 1;
9661 int hsize = Klass::layout_helper_header_size(layout_con);
9662 int eshift = Klass::layout_helper_log2_element_size(layout_con);
9663
9664 if ((round_mask & ~right_n_bits(eshift)) == 0) {
9665 round_mask = 0; // strength-reduce it if it goes away completely
9666 }
9667 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
9668 Node* header_size = intcon(hsize + round_mask);
9669
9670 Node* lengthx = ConvI2X(arr_length);
9671 Node* headerx = ConvI2X(header_size);
9672
9673 Node* abody = lengthx;
9674 if (eshift != 0) {
9675 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
9676 }
9677 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
9678 if (round_mask != 0) {
9679 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
9680 }
9681 size = ConvX2L(size);
9682 set_result(size);
9683 }
9684 } else {
9685 // Layout helper is not constant, need to test for array-ness at runtime.
9686
9687 enum { _instance_path = 1, _array_path, PATH_LIMIT };
9688 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
9689 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
9690 record_for_igvn(result_reg);
9691
9692 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
9693 if (array_ctl != nullptr) {
9694 // Array case: size is round(header + element_size*arraylength).
9695 // Since arraylength is different for every array instance, we have to
9696 // compute the whole thing at runtime.
9697
9698 PreserveJVMState pjvms(this);
9699 set_control(array_ctl);
9700 Node* arr_length = load_array_length(obj);
9701
9702 int round_mask = MinObjAlignmentInBytes - 1;
9703 Node* mask = intcon(round_mask);
9704
9705 Node* hss = intcon(Klass::_lh_header_size_shift);
9706 Node* hsm = intcon(Klass::_lh_header_size_mask);
9707 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
9708 header_size = _gvn.transform(new AndINode(header_size, hsm));
9709 header_size = _gvn.transform(new AddINode(header_size, mask));
9710
9711 // There is no need to mask or shift this value.
9712 // The semantics of LShiftINode include an implicit mask to 0x1F.
9713 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
9714 Node* elem_shift = layout_val;
9715
9716 Node* lengthx = ConvI2X(arr_length);
9717 Node* headerx = ConvI2X(header_size);
9718
9719 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
9720 Node* size = _gvn.transform(new AddXNode(headerx, abody));
9721 if (round_mask != 0) {
9722 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
9723 }
9724 size = ConvX2L(size);
9725
9726 result_reg->init_req(_array_path, control());
9727 result_val->init_req(_array_path, size);
9728 }
9729
9730 if (!stopped()) {
9731 // Instance case: the layout helper gives us instance size almost directly,
9732 // but we need to mask out the _lh_instance_slow_path_bit.
9733 Node* size = ConvI2X(layout_val);
9734 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
9735 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
9736 size = _gvn.transform(new AndXNode(size, mask));
9737 size = ConvX2L(size);
9738
9739 result_reg->init_req(_instance_path, control());
9740 result_val->init_req(_instance_path, size);
9741 }
9742
9743 set_result(result_reg, result_val);
9744 }
9745
9746 return true;
9747 }
9748
9749 //------------------------------- inline_blackhole --------------------------------------
9750 //
9751 // Make sure all arguments to this node are alive.
9752 // This matches methods that were requested to be blackholed through compile commands.
9753 //
9754 bool LibraryCallKit::inline_blackhole() {
9755 assert(callee()->is_static(), "Should have been checked before: only static methods here");
9756 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
9757 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
9758
9759 // Blackhole node pinches only the control, not memory. This allows
9760 // the blackhole to be pinned in the loop that computes blackholed
9761 // values, but have no other side effects, like breaking the optimizations
9762 // across the blackhole.
9763
9764 Node* bh = _gvn.transform(new BlackholeNode(control()));
9765 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
9766
9767 // Bind call arguments as blackhole arguments to keep them alive
9768 uint nargs = callee()->arg_size();
9769 for (uint i = 0; i < nargs; i++) {
9770 bh->add_req(argument(i));
9771 }
9772
9773 return true;
9774 }
9775
9776 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9777 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9778 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9779 return nullptr; // box klass is not Float16
9780 }
9781
9782 // Null check; get notnull casted pointer
9783 Node* null_ctl = top();
9784 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9785 // If not_null_box is dead, only null-path is taken
9786 if (stopped()) {
9787 set_control(null_ctl);
9788 return nullptr;
9789 }
9790 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9791 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9792 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9793 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9794 }
9795
9796 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9797 PreserveReexecuteState preexecs(this);
9798 jvms()->set_should_reexecute(true);
9799
9800 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9801 Node* klass_node = makecon(klass_type);
9802 Node* box = new_instance(klass_node);
9803
9804 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9805 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9806
9807 Node* field_store = _gvn.transform(access_store_at(box,
9808 value_field,
9809 value_adr_type,
9810 value,
9811 TypeInt::SHORT,
9812 T_SHORT,
9813 IN_HEAP));
9814 set_memory(field_store, value_adr_type);
9815 return box;
9816 }
9817
9818 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9819 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9820 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9821 return false;
9822 }
9823
9824 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9825 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9826 return false;
9827 }
9828
9829 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9830 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9831 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9832 ciSymbols::short_signature(),
9833 false);
9834 assert(field != nullptr, "");
9835
9836 // Transformed nodes
9837 Node* fld1 = nullptr;
9838 Node* fld2 = nullptr;
9839 Node* fld3 = nullptr;
9840 switch(num_args) {
9841 case 3:
9842 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9843 if (fld3 == nullptr) {
9844 return false;
9845 }
9846 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9847 // fall-through
9848 case 2:
9849 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9850 if (fld2 == nullptr) {
9851 return false;
9852 }
9853 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9854 // fall-through
9855 case 1:
9856 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9857 if (fld1 == nullptr) {
9858 return false;
9859 }
9860 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9861 break;
9862 default: fatal("Unsupported number of arguments %d", num_args);
9863 }
9864
9865 Node* result = nullptr;
9866 switch (id) {
9867 // Unary operations
9868 case vmIntrinsics::_sqrt_float16:
9869 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9870 break;
9871 // Ternary operations
9872 case vmIntrinsics::_fma_float16:
9873 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9874 break;
9875 default:
9876 fatal_unexpected_iid(id);
9877 break;
9878 }
9879 result = _gvn.transform(new ReinterpretHF2SNode(result));
9880 set_result(box_fp16_value(float16_box_type, field, result));
9881 return true;
9882 }
9883