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 case vmIntrinsics::_tryUpdateEpochField: return inline_native_try_update_epoch();
520 #endif
521 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
522 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
523 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
524 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
525 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
526 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
527 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
528 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
529 case vmIntrinsics::_getLength: return inline_native_getLength();
530 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
531 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
532 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
533 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
534 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
535 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
536 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
537
538 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
539 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
540 case vmIntrinsics::_newNullRestrictedNonAtomicArray: return inline_newArray(/* null_free */ true, /* atomic */ false);
541 case vmIntrinsics::_newNullRestrictedAtomicArray: return inline_newArray(/* null_free */ true, /* atomic */ true);
542 case vmIntrinsics::_newNullableAtomicArray: return inline_newArray(/* null_free */ false, /* atomic */ true);
543 case vmIntrinsics::_isFlatArray: return inline_getArrayProperties(IsFlat);
544 case vmIntrinsics::_isNullRestrictedArray: return inline_getArrayProperties(IsNullRestricted);
545 case vmIntrinsics::_isAtomicArray: return inline_getArrayProperties(IsAtomic);
546
547 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
548
549 case vmIntrinsics::_isInstance:
550 case vmIntrinsics::_isHidden:
551 case vmIntrinsics::_getSuperclass: return inline_native_Class_query(intrinsic_id());
552
553 case vmIntrinsics::_floatToRawIntBits:
554 case vmIntrinsics::_floatToIntBits:
555 case vmIntrinsics::_intBitsToFloat:
556 case vmIntrinsics::_doubleToRawLongBits:
557 case vmIntrinsics::_doubleToLongBits:
558 case vmIntrinsics::_longBitsToDouble:
559 case vmIntrinsics::_floatToFloat16:
560 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
561 case vmIntrinsics::_sqrt_float16: return inline_fp16_operations(intrinsic_id(), 1);
562 case vmIntrinsics::_fma_float16: return inline_fp16_operations(intrinsic_id(), 3);
563 case vmIntrinsics::_floatIsFinite:
564 case vmIntrinsics::_floatIsInfinite:
565 case vmIntrinsics::_doubleIsFinite:
566 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
567
568 case vmIntrinsics::_numberOfLeadingZeros_i:
569 case vmIntrinsics::_numberOfLeadingZeros_l:
570 case vmIntrinsics::_numberOfTrailingZeros_i:
571 case vmIntrinsics::_numberOfTrailingZeros_l:
572 case vmIntrinsics::_bitCount_i:
573 case vmIntrinsics::_bitCount_l:
574 case vmIntrinsics::_reverse_i:
575 case vmIntrinsics::_reverse_l:
576 case vmIntrinsics::_reverseBytes_i:
577 case vmIntrinsics::_reverseBytes_l:
578 case vmIntrinsics::_reverseBytes_s:
579 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
580
581 case vmIntrinsics::_compress_i:
582 case vmIntrinsics::_compress_l:
583 case vmIntrinsics::_expand_i:
584 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
585
586 case vmIntrinsics::_compareUnsigned_i:
587 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
588
589 case vmIntrinsics::_divideUnsigned_i:
590 case vmIntrinsics::_divideUnsigned_l:
591 case vmIntrinsics::_remainderUnsigned_i:
592 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
593
594 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
595
596 case vmIntrinsics::_Reference_get0: return inline_reference_get0();
597 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
598 case vmIntrinsics::_Reference_reachabilityFence: return inline_reference_reachabilityFence();
599 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
600 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
601 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
602
603 case vmIntrinsics::_Class_cast: return inline_Class_cast();
604
605 case vmIntrinsics::_aescrypt_encryptBlock:
606 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
607
608 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
609 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
610 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
611
612 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
613 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
614 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
615
616 case vmIntrinsics::_counterMode_AESCrypt:
617 return inline_counterMode_AESCrypt(intrinsic_id());
618
619 case vmIntrinsics::_galoisCounterMode_AESCrypt:
620 return inline_galoisCounterMode_AESCrypt();
621
622 case vmIntrinsics::_md5_implCompress:
623 case vmIntrinsics::_sha_implCompress:
624 case vmIntrinsics::_sha2_implCompress:
625 case vmIntrinsics::_sha5_implCompress:
626 case vmIntrinsics::_sha3_implCompress:
627 return inline_digestBase_implCompress(intrinsic_id());
628 case vmIntrinsics::_double_keccak:
629 case vmIntrinsics::_quad_keccak:
630 return inline_keccak(intrinsic_id());
631
632 case vmIntrinsics::_digestBase_implCompressMB:
633 return inline_digestBase_implCompressMB(predicate);
634
635 case vmIntrinsics::_multiplyToLen:
636 return inline_multiplyToLen();
637
638 case vmIntrinsics::_squareToLen:
639 return inline_squareToLen();
640
641 case vmIntrinsics::_mulAdd:
642 return inline_mulAdd();
643
644 case vmIntrinsics::_montgomeryMultiply:
645 return inline_montgomeryMultiply();
646 case vmIntrinsics::_montgomerySquare:
647 return inline_montgomerySquare();
648
649 case vmIntrinsics::_bigIntegerRightShiftWorker:
650 return inline_bigIntegerShift(true);
651 case vmIntrinsics::_bigIntegerLeftShiftWorker:
652 return inline_bigIntegerShift(false);
653
654 case vmIntrinsics::_vectorizedMismatch:
655 return inline_vectorizedMismatch();
656
657 case vmIntrinsics::_ghash_processBlocks:
658 return inline_ghash_processBlocks();
659 case vmIntrinsics::_chacha20Block:
660 return inline_chacha20Block();
661 case vmIntrinsics::_kyberNtt:
662 return inline_kyberNtt();
663 case vmIntrinsics::_kyberInverseNtt:
664 return inline_kyberInverseNtt();
665 case vmIntrinsics::_kyberNttMult:
666 return inline_kyberNttMult();
667 case vmIntrinsics::_kyberAddPoly_2:
668 return inline_kyberAddPoly_2();
669 case vmIntrinsics::_kyberAddPoly_3:
670 return inline_kyberAddPoly_3();
671 case vmIntrinsics::_kyber12To16:
672 return inline_kyber12To16();
673 case vmIntrinsics::_kyberBarrettReduce:
674 return inline_kyberBarrettReduce();
675 case vmIntrinsics::_dilithiumAlmostNtt:
676 return inline_dilithiumAlmostNtt();
677 case vmIntrinsics::_dilithiumAlmostInverseNtt:
678 return inline_dilithiumAlmostInverseNtt();
679 case vmIntrinsics::_dilithiumNttMult:
680 return inline_dilithiumNttMult();
681 case vmIntrinsics::_dilithiumMontMulByConstant:
682 return inline_dilithiumMontMulByConstant();
683 case vmIntrinsics::_dilithiumDecomposePoly:
684 return inline_dilithiumDecomposePoly();
685 case vmIntrinsics::_base64_encodeBlock:
686 return inline_base64_encodeBlock();
687 case vmIntrinsics::_base64_decodeBlock:
688 return inline_base64_decodeBlock();
689 case vmIntrinsics::_poly1305_processBlocks:
690 return inline_poly1305_processBlocks();
691 case vmIntrinsics::_intpoly_montgomeryMult_P256:
692 return inline_intpoly_montgomeryMult_P256();
693 case vmIntrinsics::_intpoly_assign:
694 return inline_intpoly_assign();
695 case vmIntrinsics::_encodeISOArray:
696 case vmIntrinsics::_encodeByteISOArray:
697 return inline_encodeISOArray(false);
698 case vmIntrinsics::_encodeAsciiArray:
699 return inline_encodeISOArray(true);
700
701 case vmIntrinsics::_updateCRC32:
702 return inline_updateCRC32();
703 case vmIntrinsics::_updateBytesCRC32:
704 return inline_updateBytesCRC32();
705 case vmIntrinsics::_updateByteBufferCRC32:
706 return inline_updateByteBufferCRC32();
707
708 case vmIntrinsics::_updateBytesCRC32C:
709 return inline_updateBytesCRC32C();
710 case vmIntrinsics::_updateDirectByteBufferCRC32C:
711 return inline_updateDirectByteBufferCRC32C();
712
713 case vmIntrinsics::_updateBytesAdler32:
714 return inline_updateBytesAdler32();
715 case vmIntrinsics::_updateByteBufferAdler32:
716 return inline_updateByteBufferAdler32();
717
718 case vmIntrinsics::_profileBoolean:
719 return inline_profileBoolean();
720 case vmIntrinsics::_isCompileConstant:
721 return inline_isCompileConstant();
722
723 case vmIntrinsics::_countPositives:
724 return inline_countPositives();
725
726 case vmIntrinsics::_fmaD:
727 case vmIntrinsics::_fmaF:
728 return inline_fma(intrinsic_id());
729
730 case vmIntrinsics::_isDigit:
731 case vmIntrinsics::_isLowerCase:
732 case vmIntrinsics::_isUpperCase:
733 case vmIntrinsics::_isWhitespace:
734 return inline_character_compare(intrinsic_id());
735
736 case vmIntrinsics::_min:
737 case vmIntrinsics::_max:
738 case vmIntrinsics::_min_strict:
739 case vmIntrinsics::_max_strict:
740 case vmIntrinsics::_minL:
741 case vmIntrinsics::_maxL:
742 case vmIntrinsics::_minF:
743 case vmIntrinsics::_maxF:
744 case vmIntrinsics::_minD:
745 case vmIntrinsics::_maxD:
746 case vmIntrinsics::_minF_strict:
747 case vmIntrinsics::_maxF_strict:
748 case vmIntrinsics::_minD_strict:
749 case vmIntrinsics::_maxD_strict:
750 return inline_min_max(intrinsic_id());
751
752 case vmIntrinsics::_VectorUnaryOp:
753 return inline_vector_nary_operation(1);
754 case vmIntrinsics::_VectorBinaryOp:
755 return inline_vector_nary_operation(2);
756 case vmIntrinsics::_VectorUnaryLibOp:
757 return inline_vector_call(1);
758 case vmIntrinsics::_VectorBinaryLibOp:
759 return inline_vector_call(2);
760 case vmIntrinsics::_VectorTernaryOp:
761 return inline_vector_nary_operation(3);
762 case vmIntrinsics::_VectorFromBitsCoerced:
763 return inline_vector_frombits_coerced();
764 case vmIntrinsics::_VectorMaskOp:
765 return inline_vector_mask_operation();
766 case vmIntrinsics::_VectorLoadOp:
767 return inline_vector_mem_operation(/*is_store=*/false);
768 case vmIntrinsics::_VectorLoadMaskedOp:
769 return inline_vector_mem_masked_operation(/*is_store*/false);
770 case vmIntrinsics::_VectorStoreOp:
771 return inline_vector_mem_operation(/*is_store=*/true);
772 case vmIntrinsics::_VectorStoreMaskedOp:
773 return inline_vector_mem_masked_operation(/*is_store=*/true);
774 case vmIntrinsics::_VectorGatherOp:
775 return inline_vector_gather_scatter(/*is_scatter*/ false);
776 case vmIntrinsics::_VectorScatterOp:
777 return inline_vector_gather_scatter(/*is_scatter*/ true);
778 case vmIntrinsics::_VectorReductionCoerced:
779 return inline_vector_reduction();
780 case vmIntrinsics::_VectorTest:
781 return inline_vector_test();
782 case vmIntrinsics::_VectorBlend:
783 return inline_vector_blend();
784 case vmIntrinsics::_VectorRearrange:
785 return inline_vector_rearrange();
786 case vmIntrinsics::_VectorSelectFrom:
787 return inline_vector_select_from();
788 case vmIntrinsics::_VectorCompare:
789 return inline_vector_compare();
790 case vmIntrinsics::_VectorBroadcastInt:
791 return inline_vector_broadcast_int();
792 case vmIntrinsics::_VectorConvert:
793 return inline_vector_convert();
794 case vmIntrinsics::_VectorInsert:
795 return inline_vector_insert();
796 case vmIntrinsics::_VectorExtract:
797 return inline_vector_extract();
798 case vmIntrinsics::_VectorCompressExpand:
799 return inline_vector_compress_expand();
800 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
801 return inline_vector_select_from_two_vectors();
802 case vmIntrinsics::_IndexVector:
803 return inline_index_vector();
804 case vmIntrinsics::_IndexPartiallyInUpperRange:
805 return inline_index_partially_in_upper_range();
806
807 case vmIntrinsics::_getObjectSize:
808 return inline_getObjectSize();
809
810 case vmIntrinsics::_blackhole:
811 return inline_blackhole();
812
813 default:
814 // If you get here, it may be that someone has added a new intrinsic
815 // to the list in vmIntrinsics.hpp without implementing it here.
816 #ifndef PRODUCT
817 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
818 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
819 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
820 }
821 #endif
822 return false;
823 }
824 }
825
826 Node* LibraryCallKit::try_to_predicate(int predicate) {
827 if (!jvms()->has_method()) {
828 // Root JVMState has a null method.
829 assert(map()->memory()->Opcode() == Op_Parm, "");
830 // Insert the memory aliasing node
831 set_all_memory(reset_memory());
832 }
833 assert(merged_memory(), "");
834
835 switch (intrinsic_id()) {
836 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
837 return inline_cipherBlockChaining_AESCrypt_predicate(false);
838 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
839 return inline_cipherBlockChaining_AESCrypt_predicate(true);
840 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
841 return inline_electronicCodeBook_AESCrypt_predicate(false);
842 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
843 return inline_electronicCodeBook_AESCrypt_predicate(true);
844 case vmIntrinsics::_counterMode_AESCrypt:
845 return inline_counterMode_AESCrypt_predicate();
846 case vmIntrinsics::_digestBase_implCompressMB:
847 return inline_digestBase_implCompressMB_predicate(predicate);
848 case vmIntrinsics::_galoisCounterMode_AESCrypt:
849 return inline_galoisCounterMode_AESCrypt_predicate();
850
851 default:
852 // If you get here, it may be that someone has added a new intrinsic
853 // to the list in vmIntrinsics.hpp without implementing it here.
854 #ifndef PRODUCT
855 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
856 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
857 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
858 }
859 #endif
860 Node* slow_ctl = control();
861 set_control(top()); // No fast path intrinsic
862 return slow_ctl;
863 }
864 }
865
866 //------------------------------set_result-------------------------------
867 // Helper function for finishing intrinsics.
868 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
869 record_for_igvn(region);
870 set_control(_gvn.transform(region));
871 set_result( _gvn.transform(value));
872 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
873 }
874
875 RegionNode* LibraryCallKit::create_bailout() {
876 RegionNode* bailout = new RegionNode(1);
877 record_for_igvn(bailout);
878 return bailout;
879 }
880
881 bool LibraryCallKit::check_bailout(RegionNode* bailout) {
882 if (bailout->req() > 1) {
883 bailout = _gvn.transform(bailout)->as_Region();
884 Node* frame = _gvn.transform(new ParmNode(C->start(), TypeFunc::FramePtr));
885 Node* halt = _gvn.transform(new HaltNode(bailout, frame, "unexpected guard failure in intrinsic"));
886 C->root()->add_req(halt);
887 }
888 return stopped();
889 }
890
891 //------------------------------generate_guard---------------------------
892 // Helper function for generating guarded fast-slow graph structures.
893 // The given 'test', if true, guards a slow path. If the test fails
894 // then a fast path can be taken. (We generally hope it fails.)
895 // In all cases, GraphKit::control() is updated to the fast path.
896 // The returned value represents the control for the slow path.
897 // The return value is never 'top'; it is either a valid control
898 // or null if it is obvious that the slow path can never be taken.
899 // Also, if region and the slow control are not null, the slow edge
900 // is appended to the region.
901 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
902 if (stopped()) {
903 // Already short circuited.
904 return nullptr;
905 }
906
907 // Build an if node and its projections.
908 // If test is true we take the slow path, which we assume is uncommon.
909 if (_gvn.type(test) == TypeInt::ZERO) {
910 // The slow branch is never taken. No need to build this guard.
911 return nullptr;
912 }
913
914 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
915
916 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
917 if (if_slow == top()) {
918 // The slow branch is never taken. No need to build this guard.
919 return nullptr;
920 }
921
922 if (region != nullptr)
923 region->add_req(if_slow);
924
925 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
926 set_control(if_fast);
927
928 return if_slow;
929 }
930
931 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
932 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
933 }
934 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
935 return generate_guard(test, region, PROB_FAIR);
936 }
937
938 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
939 Node** pos_index, bool with_opaque) {
940 if (stopped())
941 return nullptr; // already stopped
942 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
943 return nullptr; // index is already adequately typed
944 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
945 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
946 if (with_opaque) {
947 bol_lt = _gvn.transform(new OpaqueConstantBoolNode(C, bol_lt, false));
948 }
949 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
950 if (is_neg != nullptr && pos_index != nullptr) {
951 // Emulate effect of Parse::adjust_map_after_if.
952 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
953 (*pos_index) = _gvn.transform(ccast);
954 }
955 return is_neg;
956 }
957
958 // Make sure that 'position' is a valid limit index, in [0..length].
959 // There are two equivalent plans for checking this:
960 // A. (offset + copyLength) unsigned<= arrayLength
961 // B. offset <= (arrayLength - copyLength)
962 // We require that all of the values above, except for the sum and
963 // difference, are already known to be non-negative.
964 // Plan A is robust in the face of overflow, if offset and copyLength
965 // are both hugely positive.
966 //
967 // Plan B is less direct and intuitive, but it does not overflow at
968 // all, since the difference of two non-negatives is always
969 // representable. Whenever Java methods must perform the equivalent
970 // check they generally use Plan B instead of Plan A.
971 // For the moment we use Plan A.
972 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
973 Node* subseq_length,
974 Node* array_length,
975 RegionNode* region,
976 bool with_opaque) {
977 if (stopped())
978 return nullptr; // already stopped
979 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
980 if (zero_offset && subseq_length->eqv_uncast(array_length))
981 return nullptr; // common case of whole-array copy
982 Node* last = subseq_length;
983 if (!zero_offset) // last += offset
984 last = _gvn.transform(new AddINode(last, offset));
985 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
986 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
987 if (with_opaque) {
988 bol_lt = _gvn.transform(new OpaqueConstantBoolNode(C, bol_lt, false));
989 }
990 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
991 return is_over;
992 }
993
994 // Emit range checks for the given String.value byte array
995 void LibraryCallKit::generate_string_range_check(Node* array,
996 Node* offset,
997 Node* count,
998 bool char_count,
999 RegionNode* region) {
1000 if (stopped()) {
1001 return; // already stopped
1002 }
1003 if (char_count) {
1004 // Convert char count to byte count
1005 count = _gvn.transform(new LShiftINode(count, intcon(1)));
1006 }
1007 // Offset and count must not be negative
1008 generate_negative_guard(offset, region, nullptr, true);
1009 generate_negative_guard(count, region, nullptr, true);
1010 // Offset + count must not exceed length of array
1011 generate_limit_guard(offset, count, load_array_length(array), region, true);
1012 }
1013
1014 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
1015 bool is_immutable) {
1016 ciKlass* thread_klass = env()->Thread_klass();
1017 const Type* thread_type
1018 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1019
1020 Node* thread = _gvn.transform(new ThreadLocalNode());
1021 Node* p = off_heap_plus_addr(thread, in_bytes(handle_offset));
1022 tls_output = thread;
1023
1024 Node* thread_obj_handle
1025 = (is_immutable
1026 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
1027 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
1028 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
1029 thread_obj_handle = _gvn.transform(thread_obj_handle);
1030
1031 DecoratorSet decorators = IN_NATIVE;
1032 if (is_immutable) {
1033 decorators |= C2_IMMUTABLE_MEMORY;
1034 }
1035 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
1036 }
1037
1038 //--------------------------generate_current_thread--------------------
1039 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1040 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
1041 /*is_immutable*/false);
1042 }
1043
1044 //--------------------------generate_virtual_thread--------------------
1045 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
1046 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
1047 !C->method()->changes_current_thread());
1048 }
1049
1050 //------------------------------make_string_method_node------------------------
1051 // Helper method for String intrinsic functions. This version is called with
1052 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1053 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1054 // containing the lengths of str1 and str2.
1055 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1056 Node* result = nullptr;
1057 switch (opcode) {
1058 case Op_StrIndexOf:
1059 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1060 str1_start, cnt1, str2_start, cnt2, ae);
1061 break;
1062 case Op_StrComp:
1063 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1064 str1_start, cnt1, str2_start, cnt2, ae);
1065 break;
1066 case Op_StrEquals:
1067 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1068 // Use the constant length if there is one because optimized match rule may exist.
1069 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1070 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1071 break;
1072 default:
1073 ShouldNotReachHere();
1074 return nullptr;
1075 }
1076
1077 // All these intrinsics have checks.
1078 C->set_has_split_ifs(true); // Has chance for split-if optimization
1079 clear_upper_avx();
1080
1081 return _gvn.transform(result);
1082 }
1083
1084 //------------------------------inline_string_compareTo------------------------
1085 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1086 Node* arg1 = argument(0);
1087 Node* arg2 = argument(1);
1088
1089 arg1 = must_be_not_null(arg1, true);
1090 arg2 = must_be_not_null(arg2, true);
1091
1092 // Get start addr and length of first argument
1093 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1094 Node* arg1_cnt = load_array_length(arg1);
1095
1096 // Get start addr and length of second argument
1097 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1098 Node* arg2_cnt = load_array_length(arg2);
1099
1100 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1101 set_result(result);
1102 return true;
1103 }
1104
1105 //------------------------------inline_string_equals------------------------
1106 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1107 Node* arg1 = argument(0);
1108 Node* arg2 = argument(1);
1109
1110 // paths (plus control) merge
1111 RegionNode* region = new RegionNode(3);
1112 Node* phi = new PhiNode(region, TypeInt::BOOL);
1113
1114 if (!stopped()) {
1115
1116 arg1 = must_be_not_null(arg1, true);
1117 arg2 = must_be_not_null(arg2, true);
1118
1119 // Get start addr and length of first argument
1120 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1121 Node* arg1_cnt = load_array_length(arg1);
1122
1123 // Get start addr and length of second argument
1124 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1125 Node* arg2_cnt = load_array_length(arg2);
1126
1127 // Check for arg1_cnt != arg2_cnt
1128 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1129 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1130 Node* if_ne = generate_slow_guard(bol, nullptr);
1131 if (if_ne != nullptr) {
1132 phi->init_req(2, intcon(0));
1133 region->init_req(2, if_ne);
1134 }
1135
1136 // Check for count == 0 is done by assembler code for StrEquals.
1137
1138 if (!stopped()) {
1139 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1140 phi->init_req(1, equals);
1141 region->init_req(1, control());
1142 }
1143 }
1144
1145 // post merge
1146 set_control(_gvn.transform(region));
1147 record_for_igvn(region);
1148
1149 set_result(_gvn.transform(phi));
1150 return true;
1151 }
1152
1153 //------------------------------inline_array_equals----------------------------
1154 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1155 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1156 Node* arg1 = argument(0);
1157 Node* arg2 = argument(1);
1158
1159 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1160 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), mtype, arg1, arg2, ae)));
1161 clear_upper_avx();
1162
1163 return true;
1164 }
1165
1166
1167 //------------------------------inline_countPositives------------------------------
1168 // int java.lang.StringCoding#countPositives0(byte[] ba, int off, int len)
1169 bool LibraryCallKit::inline_countPositives() {
1170 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1171 // no receiver since it is static method
1172 Node* ba = argument(0);
1173 Node* offset = argument(1);
1174 Node* len = argument(2);
1175
1176 ba = must_be_not_null(ba, true);
1177 RegionNode* bailout = create_bailout();
1178 generate_string_range_check(ba, offset, len, false, bailout);
1179 if (check_bailout(bailout)) {
1180 return true;
1181 }
1182
1183 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1184 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1185 set_result(_gvn.transform(result));
1186 clear_upper_avx();
1187 return true;
1188 }
1189
1190 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1191 Node* index = argument(0);
1192 Node* length = bt == T_INT ? argument(1) : argument(2);
1193 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1194 return false;
1195 }
1196
1197 // check that length is positive
1198 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1199 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1200
1201 {
1202 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1203 uncommon_trap(Deoptimization::Reason_intrinsic,
1204 Deoptimization::Action_make_not_entrant);
1205 }
1206
1207 if (stopped()) {
1208 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1209 return true;
1210 }
1211
1212 // length is now known positive, add a cast node to make this explicit
1213 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1214 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1215 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1216 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1217 casted_length = _gvn.transform(casted_length);
1218 replace_in_map(length, casted_length);
1219 length = casted_length;
1220
1221 // Use an unsigned comparison for the range check itself
1222 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1223 BoolTest::mask btest = BoolTest::lt;
1224 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1225 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1226 _gvn.set_type(rc, rc->Value(&_gvn));
1227 if (!rc_bool->is_Con()) {
1228 record_for_igvn(rc);
1229 }
1230 set_control(_gvn.transform(new IfTrueNode(rc)));
1231 {
1232 PreserveJVMState pjvms(this);
1233 set_control(_gvn.transform(new IfFalseNode(rc)));
1234 uncommon_trap(Deoptimization::Reason_range_check,
1235 Deoptimization::Action_make_not_entrant);
1236 }
1237
1238 if (stopped()) {
1239 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1240 return true;
1241 }
1242
1243 // index is now known to be >= 0 and < length, cast it
1244 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1245 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1246 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1247 result = _gvn.transform(result);
1248 set_result(result);
1249 replace_in_map(index, result);
1250 return true;
1251 }
1252
1253 //------------------------------inline_string_indexOf------------------------
1254 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1255 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1256 return false;
1257 }
1258 Node* src = argument(0);
1259 Node* tgt = argument(1);
1260
1261 // Make the merge point
1262 RegionNode* result_rgn = new RegionNode(4);
1263 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1264
1265 src = must_be_not_null(src, true);
1266 tgt = must_be_not_null(tgt, true);
1267
1268 // Get start addr and length of source string
1269 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1270 Node* src_count = load_array_length(src);
1271
1272 // Get start addr and length of substring
1273 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1274 Node* tgt_count = load_array_length(tgt);
1275
1276 Node* result = nullptr;
1277 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1278
1279 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1280 // Divide src size by 2 if String is UTF16 encoded
1281 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1282 }
1283 if (ae == StrIntrinsicNode::UU) {
1284 // Divide substring size by 2 if String is UTF16 encoded
1285 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1286 }
1287
1288 if (call_opt_stub) {
1289 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1290 StubRoutines::_string_indexof_array[ae],
1291 "stringIndexOf", TypePtr::BOTTOM, src_start,
1292 src_count, tgt_start, tgt_count);
1293 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1294 } else {
1295 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1296 result_rgn, result_phi, ae);
1297 }
1298 if (result != nullptr) {
1299 result_phi->init_req(3, result);
1300 result_rgn->init_req(3, control());
1301 }
1302 set_control(_gvn.transform(result_rgn));
1303 record_for_igvn(result_rgn);
1304 set_result(_gvn.transform(result_phi));
1305
1306 return true;
1307 }
1308
1309 //-----------------------------inline_string_indexOfI-----------------------
1310 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1311 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1312 return false;
1313 }
1314
1315 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1316 Node* src = argument(0); // byte[]
1317 Node* src_count = argument(1); // char count
1318 Node* tgt = argument(2); // byte[]
1319 Node* tgt_count = argument(3); // char count
1320 Node* from_index = argument(4); // char index
1321
1322 src = must_be_not_null(src, true);
1323 tgt = must_be_not_null(tgt, true);
1324
1325 // Multiply byte array index by 2 if String is UTF16 encoded
1326 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1327 src_count = _gvn.transform(new SubINode(src_count, from_index));
1328 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1329 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1330
1331 // Range checks
1332 RegionNode* bailout = create_bailout();
1333 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL, bailout);
1334 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU, bailout);
1335 if (check_bailout(bailout)) {
1336 return true;
1337 }
1338
1339 RegionNode* region = new RegionNode(5);
1340 Node* phi = new PhiNode(region, TypeInt::INT);
1341 Node* result = nullptr;
1342
1343 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1344
1345 if (call_opt_stub) {
1346 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1347 StubRoutines::_string_indexof_array[ae],
1348 "stringIndexOf", TypePtr::BOTTOM, src_start,
1349 src_count, tgt_start, tgt_count);
1350 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1351 } else {
1352 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1353 region, phi, ae);
1354 }
1355 if (result != nullptr) {
1356 // The result is index relative to from_index if substring was found, -1 otherwise.
1357 // Generate code which will fold into cmove.
1358 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1359 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1360
1361 Node* if_lt = generate_slow_guard(bol, nullptr);
1362 if (if_lt != nullptr) {
1363 // result == -1
1364 phi->init_req(3, result);
1365 region->init_req(3, if_lt);
1366 }
1367 if (!stopped()) {
1368 result = _gvn.transform(new AddINode(result, from_index));
1369 phi->init_req(4, result);
1370 region->init_req(4, control());
1371 }
1372 }
1373
1374 set_control(_gvn.transform(region));
1375 record_for_igvn(region);
1376 set_result(_gvn.transform(phi));
1377 clear_upper_avx();
1378
1379 return true;
1380 }
1381
1382 // Create StrIndexOfNode with fast path checks
1383 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1384 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1385 // Check for substr count > string count
1386 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1387 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1388 Node* if_gt = generate_slow_guard(bol, nullptr);
1389 if (if_gt != nullptr) {
1390 phi->init_req(1, intcon(-1));
1391 region->init_req(1, if_gt);
1392 }
1393 if (!stopped()) {
1394 // Check for substr count == 0
1395 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1396 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1397 Node* if_zero = generate_slow_guard(bol, nullptr);
1398 if (if_zero != nullptr) {
1399 phi->init_req(2, intcon(0));
1400 region->init_req(2, if_zero);
1401 }
1402 }
1403 if (!stopped()) {
1404 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1405 }
1406 return nullptr;
1407 }
1408
1409 //-----------------------------inline_string_indexOfChar-----------------------
1410 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1411 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1412 return false;
1413 }
1414 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1415 return false;
1416 }
1417 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1418 Node* src = argument(0); // byte[]
1419 Node* int_ch = argument(1);
1420 Node* from_index = argument(2);
1421 Node* max = argument(3);
1422
1423 src = must_be_not_null(src, true);
1424
1425 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1426 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1427 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1428
1429 // Range checks
1430 RegionNode* bailout = create_bailout();
1431 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U, bailout);
1432 if (check_bailout(bailout)) {
1433 return true;
1434 }
1435
1436 // Check for int_ch >= 0
1437 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1438 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1439 {
1440 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1441 uncommon_trap(Deoptimization::Reason_intrinsic,
1442 Deoptimization::Action_maybe_recompile);
1443 }
1444 if (stopped()) {
1445 return true;
1446 }
1447
1448 RegionNode* region = new RegionNode(3);
1449 Node* phi = new PhiNode(region, TypeInt::INT);
1450
1451 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1452 C->set_has_split_ifs(true); // Has chance for split-if optimization
1453 _gvn.transform(result);
1454
1455 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1456 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1457
1458 Node* if_lt = generate_slow_guard(bol, nullptr);
1459 if (if_lt != nullptr) {
1460 // result == -1
1461 phi->init_req(2, result);
1462 region->init_req(2, if_lt);
1463 }
1464 if (!stopped()) {
1465 result = _gvn.transform(new AddINode(result, from_index));
1466 phi->init_req(1, result);
1467 region->init_req(1, control());
1468 }
1469 set_control(_gvn.transform(region));
1470 record_for_igvn(region);
1471 set_result(_gvn.transform(phi));
1472 clear_upper_avx();
1473
1474 return true;
1475 }
1476 //---------------------------inline_string_copy---------------------
1477 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1478 // int StringUTF16.compress0(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1479 // int StringUTF16.compress0(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1480 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1481 // void StringLatin1.inflate0(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1482 // void StringLatin1.inflate0(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1483 bool LibraryCallKit::inline_string_copy(bool compress) {
1484 int nargs = 5; // 2 oops, 3 ints
1485 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1486
1487 Node* src = argument(0);
1488 Node* src_offset = argument(1);
1489 Node* dst = argument(2);
1490 Node* dst_offset = argument(3);
1491 Node* length = argument(4);
1492
1493 // Check for allocation before we add nodes that would confuse
1494 // tightly_coupled_allocation()
1495 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1496
1497 // Figure out the size and type of the elements we will be copying.
1498 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1499 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1500 if (src_type == nullptr || dst_type == nullptr) {
1501 return false;
1502 }
1503 BasicType src_elem = src_type->elem()->array_element_basic_type();
1504 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1505 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1506 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1507 "Unsupported array types for inline_string_copy");
1508
1509 src = must_be_not_null(src, true);
1510 dst = must_be_not_null(dst, true);
1511
1512 // Convert char[] offsets to byte[] offsets
1513 bool convert_src = (compress && src_elem == T_BYTE);
1514 bool convert_dst = (!compress && dst_elem == T_BYTE);
1515 if (convert_src) {
1516 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1517 } else if (convert_dst) {
1518 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1519 }
1520
1521 // Range checks
1522 RegionNode* bailout = create_bailout();
1523 generate_string_range_check(src, src_offset, length, convert_src, bailout);
1524 generate_string_range_check(dst, dst_offset, length, convert_dst, bailout);
1525 if (check_bailout(bailout)) {
1526 return true;
1527 }
1528
1529 Node* src_start = array_element_address(src, src_offset, src_elem);
1530 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1531 // 'src_start' points to src array + scaled offset
1532 // 'dst_start' points to dst array + scaled offset
1533 Node* count = nullptr;
1534 if (compress) {
1535 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1536 } else {
1537 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1538 }
1539
1540 if (alloc != nullptr) {
1541 if (alloc->maybe_set_complete(&_gvn)) {
1542 // "You break it, you buy it."
1543 InitializeNode* init = alloc->initialization();
1544 assert(init->is_complete(), "we just did this");
1545 init->set_complete_with_arraycopy();
1546 assert(dst->is_CheckCastPP(), "sanity");
1547 assert(dst->in(0)->in(0) == init, "dest pinned");
1548 }
1549 // Do not let stores that initialize this object be reordered with
1550 // a subsequent store that would make this object accessible by
1551 // other threads.
1552 // Record what AllocateNode this StoreStore protects so that
1553 // escape analysis can go from the MemBarStoreStoreNode to the
1554 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1555 // based on the escape status of the AllocateNode.
1556 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1557 }
1558 if (compress) {
1559 set_result(_gvn.transform(count));
1560 }
1561 clear_upper_avx();
1562
1563 return true;
1564 }
1565
1566 #ifdef _LP64
1567 #define XTOP ,top() /*additional argument*/
1568 #else //_LP64
1569 #define XTOP /*no additional argument*/
1570 #endif //_LP64
1571
1572 //------------------------inline_string_toBytesU--------------------------
1573 // public static byte[] StringUTF16.toBytes0(char[] value, int off, int len)
1574 bool LibraryCallKit::inline_string_toBytesU() {
1575 // Get the arguments.
1576 assert(callee()->signature()->size() == 3, "character array encoder requires 3 arguments");
1577 Node* value = argument(0);
1578 Node* offset = argument(1);
1579 Node* length = argument(2);
1580
1581 Node* newcopy = nullptr;
1582
1583 // Set the original stack and the reexecute bit for the interpreter to reexecute
1584 // the bytecode that invokes StringUTF16.toBytes0() if deoptimization happens.
1585 { PreserveReexecuteState preexecs(this);
1586 jvms()->set_should_reexecute(true);
1587
1588 value = must_be_not_null(value, true);
1589 RegionNode* bailout = create_bailout();
1590 generate_negative_guard(offset, bailout, nullptr, true);
1591 generate_negative_guard(length, bailout, nullptr, true);
1592 generate_limit_guard(offset, length, load_array_length(value), bailout, true);
1593 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1594 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout, true);
1595 if (check_bailout(bailout)) {
1596 return true;
1597 }
1598
1599 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1600 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1601 newcopy = new_array(klass_node, size, 0); // no arguments to push
1602 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1603 guarantee(alloc != nullptr, "created above");
1604
1605 // Calculate starting addresses.
1606 Node* src_start = array_element_address(value, offset, T_CHAR);
1607 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1608
1609 // Check if dst array address is aligned to HeapWordSize
1610 bool aligned = (arrayOopDesc::base_offset_in_bytes(T_BYTE) % HeapWordSize == 0);
1611 // If true, then check if src array address is aligned to HeapWordSize
1612 if (aligned) {
1613 const TypeInt* toffset = gvn().type(offset)->is_int();
1614 aligned = toffset->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) +
1615 toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1616 }
1617
1618 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1619 const char* copyfunc_name = "arraycopy";
1620 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1621 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1622 OptoRuntime::fast_arraycopy_Type(),
1623 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1624 src_start, dst_start, ConvI2X(length) XTOP);
1625 // Do not let reads from the cloned object float above the arraycopy.
1626 if (alloc->maybe_set_complete(&_gvn)) {
1627 // "You break it, you buy it."
1628 InitializeNode* init = alloc->initialization();
1629 assert(init->is_complete(), "we just did this");
1630 init->set_complete_with_arraycopy();
1631 assert(newcopy->is_CheckCastPP(), "sanity");
1632 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1633 }
1634 // Do not let stores that initialize this object be reordered with
1635 // a subsequent store that would make this object accessible by
1636 // other threads.
1637 // Record what AllocateNode this StoreStore protects so that
1638 // escape analysis can go from the MemBarStoreStoreNode to the
1639 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1640 // based on the escape status of the AllocateNode.
1641 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1642 } // original reexecute is set back here
1643
1644 C->set_has_split_ifs(true); // Has chance for split-if optimization
1645 if (!stopped()) {
1646 set_result(newcopy);
1647 }
1648 clear_upper_avx();
1649
1650 return true;
1651 }
1652
1653 //------------------------inline_string_getCharsU--------------------------
1654 // public void StringUTF16.getChars0(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1655 bool LibraryCallKit::inline_string_getCharsU() {
1656 assert(callee()->signature()->size() == 5, "StringUTF16.getChars0() has 5 arguments");
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 = must_be_not_null(src, true);
1670 dst = must_be_not_null(dst, true);
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 RegionNode* bailout = create_bailout();
1681 generate_string_range_check(src, src_begin, length, true, bailout);
1682 generate_string_range_check(dst, dst_begin, length, false, bailout);
1683 if (check_bailout(bailout)) {
1684 return true;
1685 }
1686
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 C->set_has_split_ifs(true); // Has chance for split-if optimization
1727 return true;
1728 }
1729
1730 //----------------------inline_string_char_access----------------------------
1731 // Store/Load char to/from byte[] array.
1732 // static void StringUTF16.putChar(byte[] val, int index, int c)
1733 // static char StringUTF16.getChar(byte[] val, int index)
1734 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1735 Node* ch;
1736 if (is_store) {
1737 assert(callee()->signature()->size() == 3, "StringUTF16.putChar() has 3 arguments");
1738 ch = argument(2);
1739 } else {
1740 assert(callee()->signature()->size() == 2, "StringUTF16.getChar() has 2 arguments");
1741 ch = nullptr;
1742 }
1743 Node* value = argument(0);
1744 Node* index = argument(1);
1745
1746 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1747 // correctly requires matched array shapes.
1748 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1749 "sanity: byte[] and char[] bases agree");
1750 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1751 "sanity: byte[] and char[] scales agree");
1752
1753 // Bail when getChar over constants is requested: constant folding would
1754 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1755 // Java method would constant fold nicely instead.
1756 if (!is_store && value->is_Con() && index->is_Con()) {
1757 return false;
1758 }
1759
1760 // Save state and restore on bailout
1761 SavedState old_state(this);
1762
1763 value = must_be_not_null(value, true);
1764
1765 Node* adr = array_element_address(value, index, T_CHAR);
1766 if (adr->is_top()) {
1767 return false;
1768 }
1769 old_state.discard();
1770 if (is_store) {
1771 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1772 } else {
1773 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);
1774 set_result(ch);
1775 }
1776 return true;
1777 }
1778
1779
1780 //------------------------------inline_math-----------------------------------
1781 // public static double Math.abs(double)
1782 // public static double Math.sqrt(double)
1783 // public static double Math.log(double)
1784 // public static double Math.log10(double)
1785 // public static double Math.round(double)
1786 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1787 Node* arg = argument(0);
1788 Node* n = nullptr;
1789 switch (id) {
1790 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1791 case vmIntrinsics::_dsqrt:
1792 case vmIntrinsics::_dsqrt_strict:
1793 n = new SqrtDNode(C, control(), arg); break;
1794 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1795 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1796 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1797 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1798 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1799 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1800 default: fatal_unexpected_iid(id); break;
1801 }
1802 set_result(_gvn.transform(n));
1803 return true;
1804 }
1805
1806 //------------------------------inline_math-----------------------------------
1807 // public static float Math.abs(float)
1808 // public static int Math.abs(int)
1809 // public static long Math.abs(long)
1810 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1811 Node* arg = argument(0);
1812 Node* n = nullptr;
1813 switch (id) {
1814 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1815 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1816 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1817 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1818 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1819 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1820 default: fatal_unexpected_iid(id); break;
1821 }
1822 set_result(_gvn.transform(n));
1823 return true;
1824 }
1825
1826 //------------------------------runtime_math-----------------------------
1827 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1828 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1829 "must be (DD)D or (D)D type");
1830
1831 // Inputs
1832 Node* a = argument(0);
1833 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1834
1835 const TypePtr* no_memory_effects = nullptr;
1836 Node* trig = make_runtime_call(RC_LEAF | RC_PURE, call_type, funcAddr, funcName,
1837 no_memory_effects,
1838 a, top(), b, b ? top() : nullptr);
1839 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1840 #ifdef ASSERT
1841 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1842 assert(value_top == top(), "second value must be top");
1843 #endif
1844
1845 set_result(value);
1846 return true;
1847 }
1848
1849 //------------------------------inline_math_pow-----------------------------
1850 bool LibraryCallKit::inline_math_pow() {
1851 Node* base = argument(0);
1852 Node* exp = argument(2);
1853
1854 CallNode* pow = new PowDNode(C, base, exp);
1855 set_predefined_input_for_runtime_call(pow);
1856 pow = _gvn.transform(pow)->as_CallLeafPure();
1857 set_predefined_output_for_runtime_call(pow);
1858 Node* result = _gvn.transform(new ProjNode(pow, TypeFunc::Parms + 0));
1859 record_for_igvn(pow);
1860 set_result(result);
1861 return true;
1862 }
1863
1864 //------------------------------inline_math_native-----------------------------
1865 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1866 switch (id) {
1867 case vmIntrinsics::_dsin:
1868 return StubRoutines::dsin() != nullptr ?
1869 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1870 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1871 case vmIntrinsics::_dcos:
1872 return StubRoutines::dcos() != nullptr ?
1873 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1874 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1875 case vmIntrinsics::_dtan:
1876 return StubRoutines::dtan() != nullptr ?
1877 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1878 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1879 case vmIntrinsics::_dsinh:
1880 return StubRoutines::dsinh() != nullptr ?
1881 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsinh(), "dsinh") : false;
1882 case vmIntrinsics::_dtanh:
1883 return StubRoutines::dtanh() != nullptr ?
1884 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1885 case vmIntrinsics::_dcbrt:
1886 return StubRoutines::dcbrt() != nullptr ?
1887 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcbrt(), "dcbrt") : false;
1888 case vmIntrinsics::_dexp:
1889 return StubRoutines::dexp() != nullptr ?
1890 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1891 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1892 case vmIntrinsics::_dlog:
1893 return StubRoutines::dlog() != nullptr ?
1894 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1895 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1896 case vmIntrinsics::_dlog10:
1897 return StubRoutines::dlog10() != nullptr ?
1898 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1899 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1900
1901 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1902 case vmIntrinsics::_ceil:
1903 case vmIntrinsics::_floor:
1904 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1905
1906 case vmIntrinsics::_dsqrt:
1907 case vmIntrinsics::_dsqrt_strict:
1908 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1909 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1910 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1911 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1912 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1913
1914 case vmIntrinsics::_dpow: return inline_math_pow();
1915 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1916 case vmIntrinsics::_fcopySign: return inline_math(id);
1917 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1918 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1919 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1920
1921 // These intrinsics are not yet correctly implemented
1922 case vmIntrinsics::_datan2:
1923 return false;
1924
1925 default:
1926 fatal_unexpected_iid(id);
1927 return false;
1928 }
1929 }
1930
1931 //----------------------------inline_notify-----------------------------------*
1932 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1933 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1934 address func;
1935 if (id == vmIntrinsics::_notify) {
1936 func = OptoRuntime::monitor_notify_Java();
1937 } else {
1938 func = OptoRuntime::monitor_notifyAll_Java();
1939 }
1940 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1941 make_slow_call_ex(call, env()->Throwable_klass(), false);
1942 return true;
1943 }
1944
1945
1946 //----------------------------inline_min_max-----------------------------------
1947 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1948 Node* a = nullptr;
1949 Node* b = nullptr;
1950 Node* n = nullptr;
1951 switch (id) {
1952 case vmIntrinsics::_min:
1953 case vmIntrinsics::_max:
1954 case vmIntrinsics::_minF:
1955 case vmIntrinsics::_maxF:
1956 case vmIntrinsics::_minF_strict:
1957 case vmIntrinsics::_maxF_strict:
1958 case vmIntrinsics::_min_strict:
1959 case vmIntrinsics::_max_strict:
1960 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
1961 a = argument(0);
1962 b = argument(1);
1963 break;
1964 case vmIntrinsics::_minD:
1965 case vmIntrinsics::_maxD:
1966 case vmIntrinsics::_minD_strict:
1967 case vmIntrinsics::_maxD_strict:
1968 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
1969 a = argument(0);
1970 b = argument(2);
1971 break;
1972 case vmIntrinsics::_minL:
1973 case vmIntrinsics::_maxL:
1974 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
1975 a = argument(0);
1976 b = argument(2);
1977 break;
1978 default:
1979 fatal_unexpected_iid(id);
1980 break;
1981 }
1982
1983 switch (id) {
1984 case vmIntrinsics::_min:
1985 case vmIntrinsics::_min_strict:
1986 n = new MinINode(a, b);
1987 break;
1988 case vmIntrinsics::_max:
1989 case vmIntrinsics::_max_strict:
1990 n = new MaxINode(a, b);
1991 break;
1992 case vmIntrinsics::_minF:
1993 case vmIntrinsics::_minF_strict:
1994 n = new MinFNode(a, b);
1995 break;
1996 case vmIntrinsics::_maxF:
1997 case vmIntrinsics::_maxF_strict:
1998 n = new MaxFNode(a, b);
1999 break;
2000 case vmIntrinsics::_minD:
2001 case vmIntrinsics::_minD_strict:
2002 n = new MinDNode(a, b);
2003 break;
2004 case vmIntrinsics::_maxD:
2005 case vmIntrinsics::_maxD_strict:
2006 n = new MaxDNode(a, b);
2007 break;
2008 case vmIntrinsics::_minL:
2009 n = new MinLNode(_gvn.C, a, b);
2010 break;
2011 case vmIntrinsics::_maxL:
2012 n = new MaxLNode(_gvn.C, a, b);
2013 break;
2014 default:
2015 fatal_unexpected_iid(id);
2016 break;
2017 }
2018
2019 set_result(_gvn.transform(n));
2020 return true;
2021 }
2022
2023 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
2024 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
2025 env()->ArithmeticException_instance())) {
2026 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
2027 // so let's bail out intrinsic rather than risking deopting again.
2028 return false;
2029 }
2030
2031 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2032 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2033 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2034 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2035
2036 {
2037 PreserveJVMState pjvms(this);
2038 PreserveReexecuteState preexecs(this);
2039 jvms()->set_should_reexecute(true);
2040
2041 set_control(slow_path);
2042 set_i_o(i_o());
2043
2044 builtin_throw(Deoptimization::Reason_intrinsic,
2045 env()->ArithmeticException_instance(),
2046 /*allow_too_many_traps*/ false);
2047 }
2048
2049 set_control(fast_path);
2050 set_result(math);
2051 return true;
2052 }
2053
2054 template <typename OverflowOp>
2055 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2056 typedef typename OverflowOp::MathOp MathOp;
2057
2058 MathOp* mathOp = new MathOp(arg1, arg2);
2059 Node* operation = _gvn.transform( mathOp );
2060 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2061 return inline_math_mathExact(operation, ofcheck);
2062 }
2063
2064 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2065 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2066 }
2067
2068 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2069 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2070 }
2071
2072 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2073 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2074 }
2075
2076 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2077 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2078 }
2079
2080 bool LibraryCallKit::inline_math_negateExactI() {
2081 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2082 }
2083
2084 bool LibraryCallKit::inline_math_negateExactL() {
2085 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2086 }
2087
2088 bool LibraryCallKit::inline_math_multiplyExactI() {
2089 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2090 }
2091
2092 bool LibraryCallKit::inline_math_multiplyExactL() {
2093 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2094 }
2095
2096 bool LibraryCallKit::inline_math_multiplyHigh() {
2097 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2098 return true;
2099 }
2100
2101 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2102 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2103 return true;
2104 }
2105
2106 inline int
2107 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2108 const TypePtr* base_type = TypePtr::NULL_PTR;
2109 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2110 if (base_type == nullptr) {
2111 // Unknown type.
2112 return Type::AnyPtr;
2113 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2114 // Since this is a null+long form, we have to switch to a rawptr.
2115 base = _gvn.transform(new CastX2PNode(offset));
2116 offset = MakeConX(0);
2117 return Type::RawPtr;
2118 } else if (base_type->base() == Type::RawPtr) {
2119 return Type::RawPtr;
2120 } else if (base_type->isa_oopptr()) {
2121 // Base is never null => always a heap address.
2122 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2123 return Type::OopPtr;
2124 }
2125 // Offset is small => always a heap address.
2126 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2127 if (offset_type != nullptr &&
2128 base_type->offset() == 0 && // (should always be?)
2129 offset_type->_lo >= 0 &&
2130 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2131 return Type::OopPtr;
2132 } else if (type == T_OBJECT) {
2133 // off heap access to an oop doesn't make any sense. Has to be on
2134 // heap.
2135 return Type::OopPtr;
2136 }
2137 // Otherwise, it might either be oop+off or null+addr.
2138 return Type::AnyPtr;
2139 } else {
2140 // No information:
2141 return Type::AnyPtr;
2142 }
2143 }
2144
2145 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2146 Node* uncasted_base = base;
2147 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2148 if (kind == Type::RawPtr) {
2149 return off_heap_plus_addr(uncasted_base, offset);
2150 } else if (kind == Type::AnyPtr) {
2151 assert(base == uncasted_base, "unexpected base change");
2152 if (can_cast) {
2153 if (!_gvn.type(base)->speculative_maybe_null() &&
2154 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2155 // According to profiling, this access is always on
2156 // heap. Casting the base to not null and thus avoiding membars
2157 // around the access should allow better optimizations
2158 Node* null_ctl = top();
2159 base = null_check_oop(base, &null_ctl, true, true, true);
2160 assert(null_ctl->is_top(), "no null control here");
2161 return basic_plus_adr(base, offset);
2162 } else if (_gvn.type(base)->speculative_always_null() &&
2163 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2164 // According to profiling, this access is always off
2165 // heap.
2166 base = null_assert(base);
2167 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2168 offset = MakeConX(0);
2169 return off_heap_plus_addr(raw_base, offset);
2170 }
2171 }
2172 // We don't know if it's an on heap or off heap access. Fall back
2173 // to raw memory access.
2174 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2175 return off_heap_plus_addr(raw, offset);
2176 } else {
2177 assert(base == uncasted_base, "unexpected base change");
2178 // We know it's an on heap access so base can't be null
2179 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2180 base = must_be_not_null(base, true);
2181 }
2182 return basic_plus_adr(base, offset);
2183 }
2184 }
2185
2186 //--------------------------inline_number_methods-----------------------------
2187 // inline int Integer.numberOfLeadingZeros(int)
2188 // inline int Long.numberOfLeadingZeros(long)
2189 //
2190 // inline int Integer.numberOfTrailingZeros(int)
2191 // inline int Long.numberOfTrailingZeros(long)
2192 //
2193 // inline int Integer.bitCount(int)
2194 // inline int Long.bitCount(long)
2195 //
2196 // inline char Character.reverseBytes(char)
2197 // inline short Short.reverseBytes(short)
2198 // inline int Integer.reverseBytes(int)
2199 // inline long Long.reverseBytes(long)
2200 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2201 Node* arg = argument(0);
2202 Node* n = nullptr;
2203 switch (id) {
2204 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2205 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2206 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2207 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2208 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2209 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2210 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2211 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2212 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2213 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2214 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2215 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2216 default: fatal_unexpected_iid(id); break;
2217 }
2218 set_result(_gvn.transform(n));
2219 return true;
2220 }
2221
2222 //--------------------------inline_bitshuffle_methods-----------------------------
2223 // inline int Integer.compress(int, int)
2224 // inline int Integer.expand(int, int)
2225 // inline long Long.compress(long, long)
2226 // inline long Long.expand(long, long)
2227 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2228 Node* n = nullptr;
2229 switch (id) {
2230 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2231 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2232 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2233 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2234 default: fatal_unexpected_iid(id); break;
2235 }
2236 set_result(_gvn.transform(n));
2237 return true;
2238 }
2239
2240 //--------------------------inline_number_methods-----------------------------
2241 // inline int Integer.compareUnsigned(int, int)
2242 // inline int Long.compareUnsigned(long, long)
2243 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2244 Node* arg1 = argument(0);
2245 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2246 Node* n = nullptr;
2247 switch (id) {
2248 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2249 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2250 default: fatal_unexpected_iid(id); break;
2251 }
2252 set_result(_gvn.transform(n));
2253 return true;
2254 }
2255
2256 //--------------------------inline_unsigned_divmod_methods-----------------------------
2257 // inline int Integer.divideUnsigned(int, int)
2258 // inline int Integer.remainderUnsigned(int, int)
2259 // inline long Long.divideUnsigned(long, long)
2260 // inline long Long.remainderUnsigned(long, long)
2261 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2262 Node* n = nullptr;
2263 switch (id) {
2264 case vmIntrinsics::_divideUnsigned_i: {
2265 zero_check_int(argument(1));
2266 // Compile-time detect of null-exception
2267 if (stopped()) {
2268 return true; // keep the graph constructed so far
2269 }
2270 n = new UDivINode(control(), argument(0), argument(1));
2271 break;
2272 }
2273 case vmIntrinsics::_divideUnsigned_l: {
2274 zero_check_long(argument(2));
2275 // Compile-time detect of null-exception
2276 if (stopped()) {
2277 return true; // keep the graph constructed so far
2278 }
2279 n = new UDivLNode(control(), argument(0), argument(2));
2280 break;
2281 }
2282 case vmIntrinsics::_remainderUnsigned_i: {
2283 zero_check_int(argument(1));
2284 // Compile-time detect of null-exception
2285 if (stopped()) {
2286 return true; // keep the graph constructed so far
2287 }
2288 n = new UModINode(control(), argument(0), argument(1));
2289 break;
2290 }
2291 case vmIntrinsics::_remainderUnsigned_l: {
2292 zero_check_long(argument(2));
2293 // Compile-time detect of null-exception
2294 if (stopped()) {
2295 return true; // keep the graph constructed so far
2296 }
2297 n = new UModLNode(control(), argument(0), argument(2));
2298 break;
2299 }
2300 default: fatal_unexpected_iid(id); break;
2301 }
2302 set_result(_gvn.transform(n));
2303 return true;
2304 }
2305
2306 //----------------------------inline_unsafe_access----------------------------
2307
2308 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2309 // Attempt to infer a sharper value type from the offset and base type.
2310 ciKlass* sharpened_klass = nullptr;
2311 bool null_free = false;
2312
2313 // See if it is an instance field, with an object type.
2314 if (alias_type->field() != nullptr) {
2315 if (alias_type->field()->type()->is_klass()) {
2316 sharpened_klass = alias_type->field()->type()->as_klass();
2317 null_free = alias_type->field()->is_null_free();
2318 }
2319 }
2320
2321 const TypeOopPtr* result = nullptr;
2322 // See if it is a narrow oop array.
2323 if (adr_type->isa_aryptr()) {
2324 if (adr_type->offset() >= refArrayOopDesc::base_offset_in_bytes()) {
2325 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2326 null_free = adr_type->is_aryptr()->is_null_free();
2327 if (elem_type != nullptr && elem_type->is_loaded()) {
2328 // Sharpen the value type.
2329 result = elem_type;
2330 }
2331 }
2332 }
2333
2334 // The sharpened class might be unloaded if there is no class loader
2335 // contraint in place.
2336 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2337 // Sharpen the value type.
2338 result = TypeOopPtr::make_from_klass(sharpened_klass);
2339 if (null_free) {
2340 result = result->join_speculative(TypePtr::NOTNULL)->is_oopptr();
2341 }
2342 }
2343 if (result != nullptr) {
2344 #ifndef PRODUCT
2345 if (C->print_intrinsics() || C->print_inlining()) {
2346 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2347 tty->print(" sharpened value: "); result->dump(); tty->cr();
2348 }
2349 #endif
2350 }
2351 return result;
2352 }
2353
2354 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2355 switch (kind) {
2356 case Relaxed:
2357 return MO_UNORDERED;
2358 case Opaque:
2359 return MO_RELAXED;
2360 case Acquire:
2361 return MO_ACQUIRE;
2362 case Release:
2363 return MO_RELEASE;
2364 case Volatile:
2365 return MO_SEQ_CST;
2366 default:
2367 ShouldNotReachHere();
2368 return 0;
2369 }
2370 }
2371
2372 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2373 if (callee()->is_static()) return false; // caller must have the capability!
2374 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2375 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2376 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2377 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2378
2379 if (is_reference_type(type)) {
2380 decorators |= ON_UNKNOWN_OOP_REF;
2381 }
2382
2383 if (unaligned) {
2384 decorators |= C2_UNALIGNED;
2385 }
2386
2387 #ifndef PRODUCT
2388 {
2389 ResourceMark rm;
2390 // Check the signatures.
2391 ciSignature* sig = callee()->signature();
2392 #ifdef ASSERT
2393 if (!is_store) {
2394 // Object getReference(Object base, int/long offset), etc.
2395 BasicType rtype = sig->return_type()->basic_type();
2396 assert(rtype == type, "getter must return the expected value");
2397 assert(sig->count() == 2, "oop getter has 2 arguments");
2398 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2399 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2400 } else {
2401 // void putReference(Object base, int/long offset, Object x), etc.
2402 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2403 assert(sig->count() == 3, "oop putter has 3 arguments");
2404 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2405 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2406 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2407 assert(vtype == type, "putter must accept the expected value");
2408 }
2409 #endif // ASSERT
2410 }
2411 #endif //PRODUCT
2412
2413 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2414
2415 Node* receiver = argument(0); // type: oop
2416
2417 // Build address expression.
2418 Node* heap_base_oop = top();
2419
2420 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2421 Node* base = argument(1); // type: oop
2422 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2423 Node* offset = argument(2); // type: long
2424 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2425 // to be plain byte offsets, which are also the same as those accepted
2426 // by oopDesc::field_addr.
2427 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2428 "fieldOffset must be byte-scaled");
2429
2430 if (base->is_InlineType()) {
2431 assert(!is_store, "InlineTypeNodes are non-larval value objects");
2432 InlineTypeNode* vt = base->as_InlineType();
2433 if (offset->is_Con()) {
2434 long off = find_long_con(offset, 0);
2435 ciInlineKlass* vk = vt->type()->inline_klass();
2436 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2437 return false;
2438 }
2439
2440 ciField* field = vk->get_non_flat_field_by_offset(off);
2441 if (field != nullptr) {
2442 BasicType bt = type2field[field->type()->basic_type()];
2443 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2444 bt = T_OBJECT;
2445 }
2446 if (bt == type && !field->is_flat()) {
2447 Node* value = vt->field_value_by_offset(off, false);
2448 const Type* value_type = _gvn.type(value);
2449 if (value_type->is_inlinetypeptr()) {
2450 value = InlineTypeNode::make_from_oop(this, value, value_type->inline_klass());
2451 }
2452 set_result(value);
2453 return true;
2454 }
2455 }
2456 }
2457 {
2458 // Re-execute the unsafe access if allocation triggers deoptimization.
2459 PreserveReexecuteState preexecs(this);
2460 jvms()->set_should_reexecute(true);
2461 vt = vt->buffer(this);
2462 }
2463 base = vt->get_oop();
2464 }
2465
2466 // 32-bit machines ignore the high half!
2467 offset = ConvL2X(offset);
2468
2469 // Save state and restore on bailout
2470 SavedState old_state(this);
2471
2472 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2473 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2474
2475 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2476 if (type != T_OBJECT) {
2477 decorators |= IN_NATIVE; // off-heap primitive access
2478 } else {
2479 return false; // off-heap oop accesses are not supported
2480 }
2481 } else {
2482 heap_base_oop = base; // on-heap or mixed access
2483 }
2484
2485 // Can base be null? Otherwise, always on-heap access.
2486 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2487
2488 if (!can_access_non_heap) {
2489 decorators |= IN_HEAP;
2490 }
2491
2492 Node* val = is_store ? argument(4) : nullptr;
2493
2494 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2495 if (adr_type == TypePtr::NULL_PTR) {
2496 return false; // off-heap access with zero address
2497 }
2498
2499 // Try to categorize the address.
2500 Compile::AliasType* alias_type = C->alias_type(adr_type);
2501 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2502
2503 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2504 alias_type->adr_type() == TypeAryPtr::RANGE) {
2505 return false; // not supported
2506 }
2507
2508 bool mismatched = false;
2509 BasicType bt = T_ILLEGAL;
2510 ciField* field = nullptr;
2511 if (adr_type->isa_instptr()) {
2512 const TypeInstPtr* instptr = adr_type->is_instptr();
2513 ciInstanceKlass* k = instptr->instance_klass();
2514 int off = instptr->offset();
2515 if (instptr->const_oop() != nullptr &&
2516 k == ciEnv::current()->Class_klass() &&
2517 instptr->offset() >= (k->size_helper() * wordSize)) {
2518 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2519 field = k->get_field_by_offset(off, true);
2520 } else {
2521 field = k->get_non_flat_field_by_offset(off);
2522 }
2523 if (field != nullptr) {
2524 bt = type2field[field->type()->basic_type()];
2525 }
2526 if (bt != alias_type->basic_type()) {
2527 // Type mismatch. Is it an access to a nested flat field?
2528 field = k->get_field_by_offset(off, false);
2529 if (field != nullptr) {
2530 bt = type2field[field->type()->basic_type()];
2531 }
2532 }
2533 assert(bt == alias_type->basic_type(), "should match");
2534 } else {
2535 bt = alias_type->basic_type();
2536 }
2537
2538 if (bt != T_ILLEGAL) {
2539 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2540 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2541 // Alias type doesn't differentiate between byte[] and boolean[]).
2542 // Use address type to get the element type.
2543 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2544 }
2545 if (is_reference_type(bt, true)) {
2546 // accessing an array field with getReference is not a mismatch
2547 bt = T_OBJECT;
2548 }
2549 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2550 // Don't intrinsify mismatched object accesses
2551 return false;
2552 }
2553 mismatched = (bt != type);
2554 } else if (alias_type->adr_type()->isa_oopptr()) {
2555 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2556 }
2557
2558 old_state.discard();
2559 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2560
2561 if (mismatched) {
2562 decorators |= C2_MISMATCHED;
2563 }
2564
2565 // First guess at the value type.
2566 const Type *value_type = Type::get_const_basic_type(type);
2567
2568 // Figure out the memory ordering.
2569 decorators |= mo_decorator_for_access_kind(kind);
2570
2571 if (!is_store) {
2572 if (type == T_OBJECT) {
2573 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2574 if (tjp != nullptr) {
2575 value_type = tjp;
2576 }
2577 }
2578 }
2579
2580 receiver = null_check(receiver);
2581 if (stopped()) {
2582 return true;
2583 }
2584 // Heap pointers get a null-check from the interpreter,
2585 // as a courtesy. However, this is not guaranteed by Unsafe,
2586 // and it is not possible to fully distinguish unintended nulls
2587 // from intended ones in this API.
2588
2589 if (!is_store) {
2590 Node* p = nullptr;
2591 // Try to constant fold a load from a constant field
2592
2593 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !field->is_flat() && !mismatched) {
2594 // final or stable field
2595 p = make_constant_from_field(field, heap_base_oop);
2596 }
2597
2598 if (p == nullptr) { // Could not constant fold the load
2599 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2600 const TypeOopPtr* ptr = value_type->make_oopptr();
2601 if (ptr != nullptr && ptr->is_inlinetypeptr()) {
2602 // Load a non-flattened inline type from memory
2603 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass());
2604 }
2605 // Normalize the value returned by getBoolean in the following cases
2606 if (type == T_BOOLEAN &&
2607 (mismatched ||
2608 heap_base_oop == top() || // - heap_base_oop is null or
2609 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2610 // and the unsafe access is made to large offset
2611 // (i.e., larger than the maximum offset necessary for any
2612 // field access)
2613 ) {
2614 IdealKit ideal = IdealKit(this);
2615 #define __ ideal.
2616 IdealVariable normalized_result(ideal);
2617 __ declarations_done();
2618 __ set(normalized_result, p);
2619 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2620 __ set(normalized_result, ideal.ConI(1));
2621 ideal.end_if();
2622 final_sync(ideal);
2623 p = __ value(normalized_result);
2624 #undef __
2625 }
2626 }
2627 if (type == T_ADDRESS) {
2628 p = gvn().transform(new CastP2XNode(nullptr, p));
2629 p = ConvX2UL(p);
2630 }
2631 // The load node has the control of the preceding MemBarCPUOrder. All
2632 // following nodes will have the control of the MemBarCPUOrder inserted at
2633 // the end of this method. So, pushing the load onto the stack at a later
2634 // point is fine.
2635 set_result(p);
2636 } else {
2637 if (bt == T_ADDRESS) {
2638 // Repackage the long as a pointer.
2639 val = ConvL2X(val);
2640 val = gvn().transform(new CastX2PNode(val));
2641 }
2642 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2643 }
2644
2645 return true;
2646 }
2647
2648 bool LibraryCallKit::inline_unsafe_flat_access(bool is_store, AccessKind kind) {
2649 #ifdef ASSERT
2650 {
2651 ResourceMark rm;
2652 // Check the signatures.
2653 ciSignature* sig = callee()->signature();
2654 assert(sig->type_at(0)->basic_type() == T_OBJECT, "base should be object, but is %s", type2name(sig->type_at(0)->basic_type()));
2655 assert(sig->type_at(1)->basic_type() == T_LONG, "offset should be long, but is %s", type2name(sig->type_at(1)->basic_type()));
2656 assert(sig->type_at(2)->basic_type() == T_INT, "layout kind should be int, but is %s", type2name(sig->type_at(3)->basic_type()));
2657 assert(sig->type_at(3)->basic_type() == T_OBJECT, "value klass should be object, but is %s", type2name(sig->type_at(4)->basic_type()));
2658 if (is_store) {
2659 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value, but returns %s", type2name(sig->return_type()->basic_type()));
2660 assert(sig->count() == 5, "flat putter should have 5 arguments, but has %d", sig->count());
2661 assert(sig->type_at(4)->basic_type() == T_OBJECT, "put value should be object, but is %s", type2name(sig->type_at(5)->basic_type()));
2662 } else {
2663 assert(sig->return_type()->basic_type() == T_OBJECT, "getter must return an object, but returns %s", type2name(sig->return_type()->basic_type()));
2664 assert(sig->count() == 4, "flat getter should have 4 arguments, but has %d", sig->count());
2665 }
2666 }
2667 #endif // ASSERT
2668
2669 assert(kind == Relaxed, "Only plain accesses for now");
2670 if (callee()->is_static()) {
2671 // caller must have the capability!
2672 return false;
2673 }
2674 C->set_has_unsafe_access(true);
2675
2676 const TypeInstPtr* value_klass_node = _gvn.type(argument(5))->isa_instptr();
2677 if (value_klass_node == nullptr || value_klass_node->const_oop() == nullptr) {
2678 // parameter valueType is not a constant
2679 return false;
2680 }
2681 ciType* mirror_type = value_klass_node->const_oop()->as_instance()->java_mirror_type();
2682 if (!mirror_type->is_inlinetype()) {
2683 // Dead code
2684 return false;
2685 }
2686 ciInlineKlass* value_klass = mirror_type->as_inline_klass();
2687
2688 const TypeInt* layout_type = _gvn.type(argument(4))->isa_int();
2689 if (layout_type == nullptr || !layout_type->is_con()) {
2690 // parameter layoutKind is not a constant
2691 return false;
2692 }
2693 assert(layout_type->get_con() >= static_cast<int>(LayoutKind::REFERENCE) &&
2694 layout_type->get_con() < static_cast<int>(LayoutKind::UNKNOWN),
2695 "invalid layoutKind %d", layout_type->get_con());
2696 LayoutKind layout = static_cast<LayoutKind>(layout_type->get_con());
2697 assert(layout == LayoutKind::REFERENCE || layout == LayoutKind::NULL_FREE_NON_ATOMIC_FLAT ||
2698 layout == LayoutKind::NULL_FREE_ATOMIC_FLAT || layout == LayoutKind::NULLABLE_ATOMIC_FLAT,
2699 "unexpected layoutKind %d", layout_type->get_con());
2700
2701 null_check(argument(0));
2702 if (stopped()) {
2703 return true;
2704 }
2705
2706 Node* base = must_be_not_null(argument(1), true);
2707 Node* offset = argument(2);
2708 const Type* base_type = _gvn.type(base);
2709
2710 Node* ptr;
2711 bool immutable_memory = false;
2712 DecoratorSet decorators = C2_UNSAFE_ACCESS | IN_HEAP | MO_UNORDERED;
2713 if (base_type->isa_instptr()) {
2714 const TypeLong* offset_type = _gvn.type(offset)->isa_long();
2715 if (offset_type == nullptr || !offset_type->is_con()) {
2716 // Offset into a non-array should be a constant
2717 decorators |= C2_MISMATCHED;
2718 } else {
2719 int offset_con = checked_cast<int>(offset_type->get_con());
2720 ciInstanceKlass* base_klass = base_type->is_instptr()->instance_klass();
2721 ciField* field = base_klass->get_non_flat_field_by_offset(offset_con);
2722 if (field == nullptr) {
2723 assert(!base_klass->is_final(), "non-existence field at offset %d of class %s", offset_con, base_klass->name()->as_utf8());
2724 decorators |= C2_MISMATCHED;
2725 } else {
2726 assert(field->type() == value_klass, "field at offset %d of %s is of type %s, but valueType is %s",
2727 offset_con, base_klass->name()->as_utf8(), field->type()->name(), value_klass->name()->as_utf8());
2728 immutable_memory = field->is_strict() && field->is_final();
2729
2730 if (base->is_InlineType()) {
2731 assert(!is_store, "Cannot store into a non-larval value object");
2732 set_result(base->as_InlineType()->field_value_by_offset(offset_con, false));
2733 return true;
2734 }
2735 }
2736 }
2737
2738 if (base->is_InlineType()) {
2739 assert(!is_store, "Cannot store into a non-larval value object");
2740 base = base->as_InlineType()->buffer(this, true);
2741 }
2742 ptr = basic_plus_adr(base, ConvL2X(offset));
2743 } else if (base_type->isa_aryptr()) {
2744 decorators |= IS_ARRAY;
2745 if (layout == LayoutKind::REFERENCE) {
2746 if (!base_type->is_aryptr()->is_not_flat()) {
2747 const TypeAryPtr* array_type = base_type->is_aryptr()->cast_to_not_flat();
2748 // TODO 8350865 This should be a CheckCastPP, can we add a test?
2749 Node* new_base = _gvn.transform(new CastPPNode(control(), base, array_type, ConstraintCastNode::DependencyType::NonFloatingNarrowing));
2750 replace_in_map(base, new_base);
2751 base = new_base;
2752 }
2753 ptr = basic_plus_adr(base, ConvL2X(offset));
2754 } else {
2755 if (UseArrayFlattening) {
2756 // Flat array must have an exact type
2757 bool is_null_free = !LayoutKindHelper::is_nullable_flat(layout);
2758 bool is_atomic = LayoutKindHelper::is_atomic_flat(layout);
2759 Node* new_base = cast_to_flat_array_exact(base, value_klass, is_null_free, is_atomic);
2760 replace_in_map(base, new_base);
2761 base = new_base;
2762 ptr = basic_plus_adr(base, ConvL2X(offset));
2763 const TypeAryPtr* ptr_type = _gvn.type(ptr)->is_aryptr();
2764 if (ptr_type->field_offset().get() != 0) {
2765 // TODO 8350865 This should be a CheckCastPP, can we add a test?
2766 ptr = _gvn.transform(new CastPPNode(control(), ptr, ptr_type->with_field_offset(0), ConstraintCastNode::DependencyType::NonFloatingNarrowing));
2767 }
2768 } else {
2769 uncommon_trap(Deoptimization::Reason_intrinsic,
2770 Deoptimization::Action_none);
2771 return true;
2772 }
2773 }
2774 } else {
2775 decorators |= C2_MISMATCHED;
2776 ptr = basic_plus_adr(base, ConvL2X(offset));
2777 }
2778
2779 if (is_store) {
2780 Node* value = argument(6);
2781 const Type* value_type = _gvn.type(value);
2782 if (!value_type->is_inlinetypeptr()) {
2783 value_type = Type::get_const_type(value_klass)->filter_speculative(value_type);
2784 Node* new_value = _gvn.transform(new CheckCastPPNode(control(), value, value_type, ConstraintCastNode::DependencyType::NonFloatingNarrowing));
2785 new_value = InlineTypeNode::make_from_oop(this, new_value, value_klass);
2786 replace_in_map(value, new_value);
2787 value = new_value;
2788 }
2789
2790 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());
2791 if (layout == LayoutKind::REFERENCE) {
2792 const TypePtr* ptr_type = (decorators & C2_MISMATCHED) != 0 ? TypeRawPtr::BOTTOM : _gvn.type(ptr)->is_ptr();
2793 access_store_at(base, ptr, ptr_type, value, value_type, T_OBJECT, decorators);
2794 } else {
2795 bool atomic = LayoutKindHelper::is_atomic_flat(layout);
2796 bool null_free = !LayoutKindHelper::is_nullable_flat(layout);
2797 value->as_InlineType()->store_flat(this, base, ptr, atomic, immutable_memory, null_free, decorators);
2798 }
2799
2800 return true;
2801 } else {
2802 decorators |= (C2_CONTROL_DEPENDENT_LOAD | C2_UNKNOWN_CONTROL_LOAD);
2803 InlineTypeNode* result;
2804 if (layout == LayoutKind::REFERENCE) {
2805 const TypePtr* ptr_type = (decorators & C2_MISMATCHED) != 0 ? TypeRawPtr::BOTTOM : _gvn.type(ptr)->is_ptr();
2806 Node* oop = access_load_at(base, ptr, ptr_type, Type::get_const_type(value_klass), T_OBJECT, decorators);
2807 result = InlineTypeNode::make_from_oop(this, oop, value_klass);
2808 } else {
2809 bool atomic = LayoutKindHelper::is_atomic_flat(layout);
2810 bool null_free = !LayoutKindHelper::is_nullable_flat(layout);
2811 result = InlineTypeNode::make_from_flat(this, value_klass, base, ptr, atomic, immutable_memory, null_free, decorators);
2812 }
2813
2814 set_result(result);
2815 return true;
2816 }
2817 }
2818
2819 //----------------------------inline_unsafe_load_store----------------------------
2820 // This method serves a couple of different customers (depending on LoadStoreKind):
2821 //
2822 // LS_cmp_swap:
2823 //
2824 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2825 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2826 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2827 //
2828 // LS_cmp_swap_weak:
2829 //
2830 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2831 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2832 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2833 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2834 //
2835 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2836 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2837 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2838 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2839 //
2840 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2841 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2842 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2843 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2844 //
2845 // LS_cmp_exchange:
2846 //
2847 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2848 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2849 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2850 //
2851 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2852 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2853 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2854 //
2855 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2856 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2857 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2858 //
2859 // LS_get_add:
2860 //
2861 // int getAndAddInt( Object o, long offset, int delta)
2862 // long getAndAddLong(Object o, long offset, long delta)
2863 //
2864 // LS_get_set:
2865 //
2866 // int getAndSet(Object o, long offset, int newValue)
2867 // long getAndSet(Object o, long offset, long newValue)
2868 // Object getAndSet(Object o, long offset, Object newValue)
2869 //
2870 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2871 // This basic scheme here is the same as inline_unsafe_access, but
2872 // differs in enough details that combining them would make the code
2873 // overly confusing. (This is a true fact! I originally combined
2874 // them, but even I was confused by it!) As much code/comments as
2875 // possible are retained from inline_unsafe_access though to make
2876 // the correspondences clearer. - dl
2877
2878 if (callee()->is_static()) return false; // caller must have the capability!
2879
2880 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2881 decorators |= mo_decorator_for_access_kind(access_kind);
2882
2883 #ifndef PRODUCT
2884 BasicType rtype;
2885 {
2886 ResourceMark rm;
2887 // Check the signatures.
2888 ciSignature* sig = callee()->signature();
2889 rtype = sig->return_type()->basic_type();
2890 switch(kind) {
2891 case LS_get_add:
2892 case LS_get_set: {
2893 // Check the signatures.
2894 #ifdef ASSERT
2895 assert(rtype == type, "get and set must return the expected type");
2896 assert(sig->count() == 3, "get and set has 3 arguments");
2897 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2898 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2899 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2900 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2901 #endif // ASSERT
2902 break;
2903 }
2904 case LS_cmp_swap:
2905 case LS_cmp_swap_weak: {
2906 // Check the signatures.
2907 #ifdef ASSERT
2908 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2909 assert(sig->count() == 4, "CAS has 4 arguments");
2910 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2911 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2912 #endif // ASSERT
2913 break;
2914 }
2915 case LS_cmp_exchange: {
2916 // Check the signatures.
2917 #ifdef ASSERT
2918 assert(rtype == type, "CAS must return the expected type");
2919 assert(sig->count() == 4, "CAS has 4 arguments");
2920 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2921 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2922 #endif // ASSERT
2923 break;
2924 }
2925 default:
2926 ShouldNotReachHere();
2927 }
2928 }
2929 #endif //PRODUCT
2930
2931 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2932
2933 // Get arguments:
2934 Node* receiver = nullptr;
2935 Node* base = nullptr;
2936 Node* offset = nullptr;
2937 Node* oldval = nullptr;
2938 Node* newval = nullptr;
2939 switch(kind) {
2940 case LS_cmp_swap:
2941 case LS_cmp_swap_weak:
2942 case LS_cmp_exchange: {
2943 const bool two_slot_type = type2size[type] == 2;
2944 receiver = argument(0); // type: oop
2945 base = argument(1); // type: oop
2946 offset = argument(2); // type: long
2947 oldval = argument(4); // type: oop, int, or long
2948 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2949 break;
2950 }
2951 case LS_get_add:
2952 case LS_get_set: {
2953 receiver = argument(0); // type: oop
2954 base = argument(1); // type: oop
2955 offset = argument(2); // type: long
2956 oldval = nullptr;
2957 newval = argument(4); // type: oop, int, or long
2958 break;
2959 }
2960 default:
2961 ShouldNotReachHere();
2962 }
2963
2964 // Build field offset expression.
2965 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2966 // to be plain byte offsets, which are also the same as those accepted
2967 // by oopDesc::field_addr.
2968 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2969 // 32-bit machines ignore the high half of long offsets
2970 offset = ConvL2X(offset);
2971 // Save state and restore on bailout
2972 SavedState old_state(this);
2973 Node* adr = make_unsafe_address(base, offset,type, false);
2974 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2975
2976 Compile::AliasType* alias_type = C->alias_type(adr_type);
2977 BasicType bt = alias_type->basic_type();
2978 if (bt != T_ILLEGAL &&
2979 (is_reference_type(bt) != (type == T_OBJECT))) {
2980 // Don't intrinsify mismatched object accesses.
2981 return false;
2982 }
2983
2984 old_state.discard();
2985
2986 // For CAS, unlike inline_unsafe_access, there seems no point in
2987 // trying to refine types. Just use the coarse types here.
2988 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2989 const Type *value_type = Type::get_const_basic_type(type);
2990
2991 switch (kind) {
2992 case LS_get_set:
2993 case LS_cmp_exchange: {
2994 if (type == T_OBJECT) {
2995 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2996 if (tjp != nullptr) {
2997 value_type = tjp;
2998 }
2999 }
3000 break;
3001 }
3002 case LS_cmp_swap:
3003 case LS_cmp_swap_weak:
3004 case LS_get_add:
3005 break;
3006 default:
3007 ShouldNotReachHere();
3008 }
3009
3010 // Null check receiver.
3011 receiver = null_check(receiver);
3012 if (stopped()) {
3013 return true;
3014 }
3015
3016 int alias_idx = C->get_alias_index(adr_type);
3017
3018 if (is_reference_type(type)) {
3019 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
3020
3021 if (oldval != nullptr && oldval->is_InlineType()) {
3022 // Re-execute the unsafe access if allocation triggers deoptimization.
3023 PreserveReexecuteState preexecs(this);
3024 jvms()->set_should_reexecute(true);
3025 oldval = oldval->as_InlineType()->buffer(this)->get_oop();
3026 }
3027 if (newval != nullptr && newval->is_InlineType()) {
3028 // Re-execute the unsafe access if allocation triggers deoptimization.
3029 PreserveReexecuteState preexecs(this);
3030 jvms()->set_should_reexecute(true);
3031 newval = newval->as_InlineType()->buffer(this)->get_oop();
3032 }
3033
3034 // Transformation of a value which could be null pointer (CastPP #null)
3035 // could be delayed during Parse (for example, in adjust_map_after_if()).
3036 // Execute transformation here to avoid barrier generation in such case.
3037 if (_gvn.type(newval) == TypePtr::NULL_PTR)
3038 newval = _gvn.makecon(TypePtr::NULL_PTR);
3039
3040 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
3041 // Refine the value to a null constant, when it is known to be null
3042 oldval = _gvn.makecon(TypePtr::NULL_PTR);
3043 }
3044 }
3045
3046 Node* result = nullptr;
3047 switch (kind) {
3048 case LS_cmp_exchange: {
3049 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
3050 oldval, newval, value_type, type, decorators);
3051 break;
3052 }
3053 case LS_cmp_swap_weak:
3054 decorators |= C2_WEAK_CMPXCHG;
3055 case LS_cmp_swap: {
3056 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
3057 oldval, newval, value_type, type, decorators);
3058 break;
3059 }
3060 case LS_get_set: {
3061 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
3062 newval, value_type, type, decorators);
3063 break;
3064 }
3065 case LS_get_add: {
3066 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
3067 newval, value_type, type, decorators);
3068 break;
3069 }
3070 default:
3071 ShouldNotReachHere();
3072 }
3073
3074 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3075 set_result(result);
3076 return true;
3077 }
3078
3079 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3080 // Regardless of form, don't allow previous ld/st to move down,
3081 // then issue acquire, release, or volatile mem_bar.
3082 insert_mem_bar(Op_MemBarCPUOrder);
3083 switch(id) {
3084 case vmIntrinsics::_loadFence:
3085 insert_mem_bar(Op_LoadFence);
3086 return true;
3087 case vmIntrinsics::_storeFence:
3088 insert_mem_bar(Op_StoreFence);
3089 return true;
3090 case vmIntrinsics::_storeStoreFence:
3091 insert_mem_bar(Op_StoreStoreFence);
3092 return true;
3093 case vmIntrinsics::_fullFence:
3094 insert_mem_bar(Op_MemBarFull);
3095 return true;
3096 default:
3097 fatal_unexpected_iid(id);
3098 return false;
3099 }
3100 }
3101
3102 // private native int arrayInstanceBaseOffset0(Object[] array);
3103 bool LibraryCallKit::inline_arrayInstanceBaseOffset() {
3104 Node* array = argument(1);
3105 Node* klass_node = load_object_klass(array);
3106
3107 jint layout_con = Klass::_lh_neutral_value;
3108 Node* layout_val = get_layout_helper(klass_node, layout_con);
3109 int layout_is_con = (layout_val == nullptr);
3110
3111 Node* header_size = nullptr;
3112 if (layout_is_con) {
3113 int hsize = Klass::layout_helper_header_size(layout_con);
3114 header_size = intcon(hsize);
3115 } else {
3116 Node* hss = intcon(Klass::_lh_header_size_shift);
3117 Node* hsm = intcon(Klass::_lh_header_size_mask);
3118 header_size = _gvn.transform(new URShiftINode(layout_val, hss));
3119 header_size = _gvn.transform(new AndINode(header_size, hsm));
3120 }
3121 set_result(header_size);
3122 return true;
3123 }
3124
3125 // private native int arrayInstanceIndexScale0(Object[] array);
3126 bool LibraryCallKit::inline_arrayInstanceIndexScale() {
3127 Node* array = argument(1);
3128 Node* klass_node = load_object_klass(array);
3129
3130 jint layout_con = Klass::_lh_neutral_value;
3131 Node* layout_val = get_layout_helper(klass_node, layout_con);
3132 int layout_is_con = (layout_val == nullptr);
3133
3134 Node* element_size = nullptr;
3135 if (layout_is_con) {
3136 int log_element_size = Klass::layout_helper_log2_element_size(layout_con);
3137 int elem_size = 1 << log_element_size;
3138 element_size = intcon(elem_size);
3139 } else {
3140 Node* ess = intcon(Klass::_lh_log2_element_size_shift);
3141 Node* esm = intcon(Klass::_lh_log2_element_size_mask);
3142 Node* log_element_size = _gvn.transform(new URShiftINode(layout_val, ess));
3143 log_element_size = _gvn.transform(new AndINode(log_element_size, esm));
3144 element_size = _gvn.transform(new LShiftINode(intcon(1), log_element_size));
3145 }
3146 set_result(element_size);
3147 return true;
3148 }
3149
3150 // private native int arrayLayout0(Object[] array);
3151 bool LibraryCallKit::inline_arrayLayout() {
3152 RegionNode* region = new RegionNode(2);
3153 Node* phi = new PhiNode(region, TypeInt::POS);
3154
3155 Node* array = argument(1);
3156 Node* klass_node = load_object_klass(array);
3157 generate_refArray_guard(klass_node, region);
3158 if (region->req() == 3) {
3159 phi->add_req(intcon((jint)LayoutKind::REFERENCE));
3160 }
3161
3162 int layout_kind_offset = in_bytes(FlatArrayKlass::layout_kind_offset());
3163 Node* layout_kind_addr = basic_plus_adr(top(), klass_node, layout_kind_offset);
3164 Node* layout_kind = make_load(nullptr, layout_kind_addr, TypeInt::POS, T_INT, MemNode::unordered);
3165
3166 region->init_req(1, control());
3167 phi->init_req(1, layout_kind);
3168
3169 set_control(_gvn.transform(region));
3170 set_result(_gvn.transform(phi));
3171 return true;
3172 }
3173
3174 // private native int[] getFieldMap0(Class <?> c);
3175 // int offset = c._klass._acmp_maps_offset;
3176 // return (int[])c.obj_field(offset);
3177 bool LibraryCallKit::inline_getFieldMap() {
3178 Node* mirror = argument(1);
3179 Node* klass = load_klass_from_mirror(mirror, false, nullptr, 0);
3180
3181 int field_map_offset_offset = in_bytes(InstanceKlass::acmp_maps_offset_offset());
3182 Node* field_map_offset_addr = basic_plus_adr(top(), klass, field_map_offset_offset);
3183 Node* field_map_offset = make_load(nullptr, field_map_offset_addr, TypeInt::INT, T_INT, MemNode::unordered);
3184 field_map_offset = _gvn.transform(ConvI2L(field_map_offset));
3185
3186 Node* map_addr = basic_plus_adr(mirror, field_map_offset);
3187 const TypeAryPtr* val_type = TypeAryPtr::INTS->cast_to_ptr_type(TypePtr::NotNull)->with_offset(0);
3188 Node* map = access_load_at(mirror, map_addr, TypeAryPtr::INTS, val_type, T_ARRAY, IN_HEAP | MO_UNORDERED);
3189
3190 set_result(map);
3191 return true;
3192 }
3193
3194 bool LibraryCallKit::inline_onspinwait() {
3195 insert_mem_bar(Op_OnSpinWait);
3196 return true;
3197 }
3198
3199 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3200 if (!kls->is_Con()) {
3201 return true;
3202 }
3203 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
3204 if (klsptr == nullptr) {
3205 return true;
3206 }
3207 ciInstanceKlass* ik = klsptr->instance_klass();
3208 // don't need a guard for a klass that is already initialized
3209 return !ik->is_initialized();
3210 }
3211
3212 //----------------------------inline_unsafe_writeback0-------------------------
3213 // public native void Unsafe.writeback0(long address)
3214 bool LibraryCallKit::inline_unsafe_writeback0() {
3215 if (!Matcher::has_match_rule(Op_CacheWB)) {
3216 return false;
3217 }
3218 #ifndef PRODUCT
3219 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
3220 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
3221 ciSignature* sig = callee()->signature();
3222 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
3223 #endif
3224 null_check_receiver(); // null-check, then ignore
3225 Node *addr = argument(1);
3226 addr = new CastX2PNode(addr);
3227 addr = _gvn.transform(addr);
3228 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
3229 flush = _gvn.transform(flush);
3230 set_memory(flush, TypeRawPtr::BOTTOM);
3231 return true;
3232 }
3233
3234 //----------------------------inline_unsafe_writeback0-------------------------
3235 // public native void Unsafe.writeback0(long address)
3236 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
3237 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
3238 return false;
3239 }
3240 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
3241 return false;
3242 }
3243 #ifndef PRODUCT
3244 assert(Matcher::has_match_rule(Op_CacheWB),
3245 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
3246 : "found match rule for CacheWBPostSync but not CacheWB"));
3247
3248 #endif
3249 null_check_receiver(); // null-check, then ignore
3250 Node *sync;
3251 if (is_pre) {
3252 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3253 } else {
3254 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3255 }
3256 sync = _gvn.transform(sync);
3257 set_memory(sync, TypeRawPtr::BOTTOM);
3258 return true;
3259 }
3260
3261 //----------------------------inline_unsafe_allocate---------------------------
3262 // public native Object Unsafe.allocateInstance(Class<?> cls);
3263 bool LibraryCallKit::inline_unsafe_allocate() {
3264
3265 #if INCLUDE_JVMTI
3266 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3267 return false;
3268 }
3269 #endif //INCLUDE_JVMTI
3270
3271 if (callee()->is_static()) return false; // caller must have the capability!
3272
3273 null_check_receiver(); // null-check, then ignore
3274 Node* cls = null_check(argument(1));
3275 if (stopped()) return true;
3276
3277 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
3278 kls = null_check(kls);
3279 if (stopped()) return true; // argument was like int.class
3280
3281 #if INCLUDE_JVMTI
3282 // Don't try to access new allocated obj in the intrinsic.
3283 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
3284 // Deoptimize and allocate in interpreter instead.
3285 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
3286 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
3287 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
3288 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
3289 {
3290 BuildCutout unless(this, tst, PROB_MAX);
3291 uncommon_trap(Deoptimization::Reason_intrinsic,
3292 Deoptimization::Action_make_not_entrant);
3293 }
3294 if (stopped()) {
3295 return true;
3296 }
3297 #endif //INCLUDE_JVMTI
3298
3299 Node* test = nullptr;
3300 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3301 // Note: The argument might still be an illegal value like
3302 // Serializable.class or Object[].class. The runtime will handle it.
3303 // But we must make an explicit check for initialization.
3304 Node* insp = off_heap_plus_addr(kls, in_bytes(InstanceKlass::init_state_offset()));
3305 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3306 // can generate code to load it as unsigned byte.
3307 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
3308 Node* bits = intcon(InstanceKlass::fully_initialized);
3309 test = _gvn.transform(new SubINode(inst, bits));
3310 // The 'test' is non-zero if we need to take a slow path.
3311 }
3312 Node* obj = new_instance(kls, test);
3313 set_result(obj);
3314 return true;
3315 }
3316
3317 //------------------------inline_native_time_funcs--------------
3318 // inline code for System.currentTimeMillis() and System.nanoTime()
3319 // these have the same type and signature
3320 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3321 const TypeFunc* tf = OptoRuntime::void_long_Type();
3322 const TypePtr* no_memory_effects = nullptr;
3323 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3324 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3325 #ifdef ASSERT
3326 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3327 assert(value_top == top(), "second value must be top");
3328 #endif
3329 set_result(value);
3330 return true;
3331 }
3332
3333 //--------------------inline_native_vthread_start_transition--------------------
3334 // inline void startTransition(boolean is_mount);
3335 // inline void startFinalTransition();
3336 // Pseudocode of implementation:
3337 //
3338 // java_lang_Thread::set_is_in_vthread_transition(vthread, true);
3339 // carrier->set_is_in_vthread_transition(true);
3340 // OrderAccess::storeload();
3341 // int disable_requests = java_lang_Thread::vthread_transition_disable_count(vthread)
3342 // + global_vthread_transition_disable_count();
3343 // if (disable_requests > 0) {
3344 // slow path: runtime call
3345 // }
3346 bool LibraryCallKit::inline_native_vthread_start_transition(address funcAddr, const char* funcName, bool is_final_transition) {
3347 Node* vt_oop = must_be_not_null(argument(0), true); // VirtualThread this argument
3348 IdealKit ideal(this);
3349
3350 Node* thread = ideal.thread();
3351 Node* jt_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3352 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3353 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3354 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3355 insert_mem_bar(Op_MemBarStoreLoad);
3356 ideal.sync_kit(this);
3357
3358 Node* global_disable_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::global_vthread_transition_disable_count_address()));
3359 Node* global_disable = ideal.load(ideal.ctrl(), global_disable_addr, TypeInt::INT, T_INT, Compile::AliasIdxRaw, true /*require_atomic_access*/);
3360 Node* vt_disable_addr = basic_plus_adr(vt_oop, java_lang_Thread::vthread_transition_disable_count_offset());
3361 const TypePtr* vt_disable_addr_t = _gvn.type(vt_disable_addr)->is_ptr();
3362 Node* vt_disable = ideal.load(ideal.ctrl(), vt_disable_addr, TypeInt::INT, T_INT, C->get_alias_index(vt_disable_addr_t), true /*require_atomic_access*/);
3363 Node* disabled = _gvn.transform(new AddINode(global_disable, vt_disable));
3364
3365 ideal.if_then(disabled, BoolTest::ne, ideal.ConI(0)); {
3366 sync_kit(ideal);
3367 Node* is_mount = is_final_transition ? ideal.ConI(0) : argument(1);
3368 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3369 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3370 ideal.sync_kit(this);
3371 }
3372 ideal.end_if();
3373
3374 final_sync(ideal);
3375 return true;
3376 }
3377
3378 bool LibraryCallKit::inline_native_vthread_end_transition(address funcAddr, const char* funcName, bool is_first_transition) {
3379 Node* vt_oop = must_be_not_null(argument(0), true); // VirtualThread this argument
3380 IdealKit ideal(this);
3381
3382 Node* _notify_jvmti_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::notify_jvmti_events_address()));
3383 Node* _notify_jvmti = ideal.load(ideal.ctrl(), _notify_jvmti_addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3384
3385 ideal.if_then(_notify_jvmti, BoolTest::eq, ideal.ConI(1)); {
3386 sync_kit(ideal);
3387 Node* is_mount = is_first_transition ? ideal.ConI(1) : argument(1);
3388 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3389 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3390 ideal.sync_kit(this);
3391 } ideal.else_(); {
3392 Node* thread = ideal.thread();
3393 Node* jt_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3394 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3395
3396 sync_kit(ideal);
3397 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3398 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3399 ideal.sync_kit(this);
3400 } ideal.end_if();
3401
3402 final_sync(ideal);
3403 return true;
3404 }
3405
3406 #if INCLUDE_JVMTI
3407
3408 // Always update the is_disable_suspend bit.
3409 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3410 if (!DoJVMTIVirtualThreadTransitions) {
3411 return true;
3412 }
3413 IdealKit ideal(this);
3414
3415 {
3416 // unconditionally update the is_disable_suspend bit in current JavaThread
3417 Node* thread = ideal.thread();
3418 Node* arg = argument(0); // argument for notification
3419 Node* addr = off_heap_plus_addr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3420 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3421
3422 sync_kit(ideal);
3423 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3424 ideal.sync_kit(this);
3425 }
3426 final_sync(ideal);
3427
3428 return true;
3429 }
3430
3431 #endif // INCLUDE_JVMTI
3432
3433 #ifdef JFR_HAVE_INTRINSICS
3434
3435 /**
3436 * if oop->klass != null
3437 * // normal class
3438 * epoch = _epoch_state ? 2 : 1
3439 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3440 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3441 * }
3442 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3443 * else
3444 * // primitive class
3445 * if oop->array_klass != null
3446 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3447 * else
3448 * id = LAST_TYPE_ID + 1 // void class path
3449 * if (!signaled)
3450 * signaled = true
3451 */
3452 bool LibraryCallKit::inline_native_classID() {
3453 Node* cls = argument(0);
3454
3455 IdealKit ideal(this);
3456 #define __ ideal.
3457 IdealVariable result(ideal); __ declarations_done();
3458 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3459 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3460 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3461
3462
3463 __ if_then(kls, BoolTest::ne, null()); {
3464 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3465 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3466
3467 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3468 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3469 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3470 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3471 mask = _gvn.transform(new OrLNode(mask, epoch));
3472 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3473
3474 float unlikely = PROB_UNLIKELY(0.999);
3475 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3476 sync_kit(ideal);
3477 make_runtime_call(RC_LEAF,
3478 OptoRuntime::class_id_load_barrier_Type(),
3479 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3480 "class id load barrier",
3481 TypePtr::BOTTOM,
3482 kls);
3483 ideal.sync_kit(this);
3484 } __ end_if();
3485
3486 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3487 } __ else_(); {
3488 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3489 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3490 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3491 __ if_then(array_kls, BoolTest::ne, null()); {
3492 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3493 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3494 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3495 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3496 } __ else_(); {
3497 // void class case
3498 ideal.set(result, longcon(LAST_TYPE_ID + 1));
3499 } __ end_if();
3500
3501 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3502 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3503 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3504 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3505 } __ end_if();
3506 } __ end_if();
3507
3508 final_sync(ideal);
3509 set_result(ideal.value(result));
3510 #undef __
3511 return true;
3512 }
3513
3514 //------------------------inline_native_jvm_commit------------------
3515 bool LibraryCallKit::inline_native_jvm_commit() {
3516 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3517
3518 // Save input memory and i_o state.
3519 Node* input_memory_state = reset_memory();
3520 set_all_memory(input_memory_state);
3521 Node* input_io_state = i_o();
3522
3523 // TLS.
3524 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3525 // Jfr java buffer.
3526 Node* java_buffer_offset = _gvn.transform(AddPNode::make_off_heap(tls_ptr, MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR))));
3527 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3528 Node* java_buffer_pos_offset = _gvn.transform(AddPNode::make_off_heap(java_buffer, MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET))));
3529
3530 // Load the current value of the notified field in the JfrThreadLocal.
3531 Node* notified_offset = off_heap_plus_addr(tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3532 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3533
3534 // Test for notification.
3535 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3536 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3537 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3538
3539 // True branch, is notified.
3540 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3541 set_control(is_notified);
3542
3543 // Reset notified state.
3544 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3545 Node* notified_reset_memory = reset_memory();
3546
3547 // Iff notified, the return address of the commit method is the current position of the backing java buffer. This is used to reset the event writer.
3548 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3549 // Convert the machine-word to a long.
3550 Node* current_pos = ConvX2L(current_pos_X);
3551
3552 // False branch, not notified.
3553 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3554 set_control(not_notified);
3555 set_all_memory(input_memory_state);
3556
3557 // Arg is the next position as a long.
3558 Node* arg = argument(0);
3559 // Convert long to machine-word.
3560 Node* next_pos_X = ConvL2X(arg);
3561
3562 // Store the next_position to the underlying jfr java buffer.
3563 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3564
3565 Node* commit_memory = reset_memory();
3566 set_all_memory(commit_memory);
3567
3568 // Now load the flags from off the java buffer and decide if the buffer is a lease. If so, it needs to be returned post-commit.
3569 Node* java_buffer_flags_offset = _gvn.transform(AddPNode::make_off_heap(java_buffer, MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET))));
3570 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3571 Node* lease_constant = _gvn.intcon(4);
3572
3573 // And flags with lease constant.
3574 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3575
3576 // Branch on lease to conditionalize returning the leased java buffer.
3577 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3578 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3579 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3580
3581 // False branch, not a lease.
3582 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3583
3584 // True branch, is lease.
3585 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3586 set_control(is_lease);
3587
3588 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3589 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3590 OptoRuntime::void_void_Type(),
3591 SharedRuntime::jfr_return_lease(),
3592 "return_lease", TypePtr::BOTTOM);
3593 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3594
3595 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3596 record_for_igvn(lease_compare_rgn);
3597 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3598 record_for_igvn(lease_compare_mem);
3599 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3600 record_for_igvn(lease_compare_io);
3601 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3602 record_for_igvn(lease_result_value);
3603
3604 // Update control and phi nodes.
3605 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3606 lease_compare_rgn->init_req(_false_path, not_lease);
3607
3608 lease_compare_mem->init_req(_true_path, reset_memory());
3609 lease_compare_mem->init_req(_false_path, commit_memory);
3610
3611 lease_compare_io->init_req(_true_path, i_o());
3612 lease_compare_io->init_req(_false_path, input_io_state);
3613
3614 lease_result_value->init_req(_true_path, _gvn.longcon(0)); // if the lease was returned, return 0L.
3615 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3616
3617 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3618 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3619 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3620 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3621
3622 // Update control and phi nodes.
3623 result_rgn->init_req(_true_path, is_notified);
3624 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3625
3626 result_mem->init_req(_true_path, notified_reset_memory);
3627 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3628
3629 result_io->init_req(_true_path, input_io_state);
3630 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3631
3632 result_value->init_req(_true_path, current_pos);
3633 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3634
3635 // Set output state.
3636 set_control(_gvn.transform(result_rgn));
3637 set_all_memory(_gvn.transform(result_mem));
3638 set_i_o(_gvn.transform(result_io));
3639 set_result(result_rgn, result_value);
3640 return true;
3641 }
3642
3643 /*
3644 * The intrinsic is a model of this pseudo-code:
3645 *
3646 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3647 * jobject h_event_writer = tl->java_event_writer();
3648 * if (h_event_writer == nullptr) {
3649 * return nullptr;
3650 * }
3651 * oop threadObj = Thread::threadObj();
3652 * oop vthread = java_lang_Thread::vthread(threadObj);
3653 * traceid tid;
3654 * bool pinVirtualThread;
3655 * bool excluded;
3656 * if (vthread != threadObj) { // i.e. current thread is virtual
3657 * tid = java_lang_Thread::tid(vthread);
3658 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3659 * pinVirtualThread = VMContinuations;
3660 * excluded = vthread_epoch_raw & excluded_mask;
3661 * if (!excluded) {
3662 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3663 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3664 * if (vthread_epoch != current_epoch) {
3665 * write_checkpoint();
3666 * }
3667 * }
3668 * } else {
3669 * tid = java_lang_Thread::tid(threadObj);
3670 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3671 * pinVirtualThread = false;
3672 * excluded = thread_epoch_raw & excluded_mask;
3673 * }
3674 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3675 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3676 * if (tid_in_event_writer != tid) {
3677 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3678 * setField(event_writer, "excluded", excluded);
3679 * setField(event_writer, "threadID", tid);
3680 * }
3681 * return event_writer
3682 */
3683 bool LibraryCallKit::inline_native_getEventWriter() {
3684 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3685
3686 // Save input memory and i_o state.
3687 Node* input_memory_state = reset_memory();
3688 set_all_memory(input_memory_state);
3689 Node* input_io_state = i_o();
3690
3691 // The most significant bit of the u2 is used to denote thread exclusion
3692 Node* excluded_shift = _gvn.intcon(15);
3693 Node* excluded_mask = _gvn.intcon(1 << 15);
3694 // The epoch generation is the range [1-32767]
3695 Node* epoch_mask = _gvn.intcon(32767);
3696
3697 // TLS
3698 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3699
3700 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3701 Node* jobj_ptr = off_heap_plus_addr(tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3702
3703 // Load the eventwriter jobject handle.
3704 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3705
3706 // Null check the jobject handle.
3707 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3708 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3709 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3710
3711 // False path, jobj is null.
3712 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3713
3714 // True path, jobj is not null.
3715 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3716
3717 set_control(jobj_is_not_null);
3718
3719 // Load the threadObj for the CarrierThread.
3720 Node* threadObj = generate_current_thread(tls_ptr);
3721
3722 // Load the vthread.
3723 Node* vthread = generate_virtual_thread(tls_ptr);
3724
3725 // If vthread != threadObj, this is a virtual thread.
3726 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3727 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3728 IfNode* iff_vthread_not_equal_threadObj =
3729 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3730
3731 // False branch, fallback to threadObj.
3732 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3733 set_control(vthread_equal_threadObj);
3734
3735 // Load the tid field from the vthread object.
3736 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3737
3738 // Load the raw epoch value from the threadObj.
3739 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3740 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3741 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3742 TypeInt::CHAR, T_CHAR,
3743 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3744
3745 // Mask off the excluded information from the epoch.
3746 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3747
3748 // True branch, this is a virtual thread.
3749 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3750 set_control(vthread_not_equal_threadObj);
3751
3752 // Load the tid field from the vthread object.
3753 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3754
3755 // Continuation support determines if a virtual thread should be pinned.
3756 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3757 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3758
3759 // Load the raw epoch value from the vthread.
3760 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3761 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3762 TypeInt::CHAR, T_CHAR,
3763 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3764
3765 // Mask off the excluded information from the epoch.
3766 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, excluded_mask));
3767
3768 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3769 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, excluded_mask));
3770 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3771 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3772
3773 // False branch, vthread is excluded, no need to write epoch info.
3774 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3775
3776 // True branch, vthread is included, update epoch info.
3777 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3778 set_control(included);
3779
3780 // Get epoch value.
3781 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, epoch_mask));
3782
3783 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3784 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3785 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3786
3787 // Compare the epoch in the vthread to the current epoch generation.
3788 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3789 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3790 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3791
3792 // False path, epoch is equal, checkpoint information is valid.
3793 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3794
3795 // True path, epoch is not equal, write a checkpoint for the vthread.
3796 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3797
3798 set_control(epoch_is_not_equal);
3799
3800 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3801 // The call also updates the native thread local thread id and the vthread with the current epoch.
3802 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3803 OptoRuntime::jfr_write_checkpoint_Type(),
3804 SharedRuntime::jfr_write_checkpoint(),
3805 "write_checkpoint", TypePtr::BOTTOM);
3806 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3807
3808 // vthread epoch != current epoch
3809 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3810 record_for_igvn(epoch_compare_rgn);
3811 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3812 record_for_igvn(epoch_compare_mem);
3813 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3814 record_for_igvn(epoch_compare_io);
3815
3816 // Update control and phi nodes.
3817 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3818 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3819 epoch_compare_mem->init_req(_true_path, reset_memory());
3820 epoch_compare_mem->init_req(_false_path, input_memory_state);
3821 epoch_compare_io->init_req(_true_path, i_o());
3822 epoch_compare_io->init_req(_false_path, input_io_state);
3823
3824 // excluded != true
3825 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3826 record_for_igvn(exclude_compare_rgn);
3827 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3828 record_for_igvn(exclude_compare_mem);
3829 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3830 record_for_igvn(exclude_compare_io);
3831
3832 // Update control and phi nodes.
3833 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3834 exclude_compare_rgn->init_req(_false_path, excluded);
3835 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3836 exclude_compare_mem->init_req(_false_path, input_memory_state);
3837 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3838 exclude_compare_io->init_req(_false_path, input_io_state);
3839
3840 // vthread != threadObj
3841 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3842 record_for_igvn(vthread_compare_rgn);
3843 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3844 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3845 record_for_igvn(vthread_compare_io);
3846 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3847 record_for_igvn(tid);
3848 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3849 record_for_igvn(exclusion);
3850 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3851 record_for_igvn(pinVirtualThread);
3852
3853 // Update control and phi nodes.
3854 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3855 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3856 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3857 vthread_compare_mem->init_req(_false_path, input_memory_state);
3858 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3859 vthread_compare_io->init_req(_false_path, input_io_state);
3860 tid->init_req(_true_path, vthread_tid);
3861 tid->init_req(_false_path, thread_obj_tid);
3862 exclusion->init_req(_true_path, vthread_is_excluded);
3863 exclusion->init_req(_false_path, threadObj_is_excluded);
3864 pinVirtualThread->init_req(_true_path, continuation_support);
3865 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3866
3867 // Update branch state.
3868 set_control(_gvn.transform(vthread_compare_rgn));
3869 set_all_memory(_gvn.transform(vthread_compare_mem));
3870 set_i_o(_gvn.transform(vthread_compare_io));
3871
3872 // Load the event writer oop by dereferencing the jobject handle.
3873 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3874 assert(klass_EventWriter->is_loaded(), "invariant");
3875 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3876 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3877 const TypeOopPtr* const xtype = aklass->as_instance_type();
3878 Node* jobj_untagged = _gvn.transform(AddPNode::make_off_heap(jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3879 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3880
3881 // Load the current thread id from the event writer object.
3882 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3883 // Get the field offset to, conditionally, store an updated tid value later.
3884 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3885 // Get the field offset to, conditionally, store an updated exclusion value later.
3886 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3887 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3888 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3889
3890 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3891 record_for_igvn(event_writer_tid_compare_rgn);
3892 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3893 record_for_igvn(event_writer_tid_compare_mem);
3894 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3895 record_for_igvn(event_writer_tid_compare_io);
3896
3897 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3898 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3899 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3900 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3901
3902 // False path, tids are the same.
3903 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3904
3905 // True path, tid is not equal, need to update the tid in the event writer.
3906 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3907 record_for_igvn(tid_is_not_equal);
3908
3909 // Store the pin state to the event writer.
3910 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3911
3912 // Store the exclusion state to the event writer.
3913 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3914 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3915
3916 // Store the tid to the event writer.
3917 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3918
3919 // Update control and phi nodes.
3920 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3921 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3922 event_writer_tid_compare_mem->init_req(_true_path, reset_memory());
3923 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3924 event_writer_tid_compare_io->init_req(_true_path, i_o());
3925 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3926
3927 // Result of top level CFG, Memory, IO and Value.
3928 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3929 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3930 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3931 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3932
3933 // Result control.
3934 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3935 result_rgn->init_req(_false_path, jobj_is_null);
3936
3937 // Result memory.
3938 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3939 result_mem->init_req(_false_path, input_memory_state);
3940
3941 // Result IO.
3942 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3943 result_io->init_req(_false_path, input_io_state);
3944
3945 // Result value.
3946 result_value->init_req(_true_path, event_writer); // return event writer oop
3947 result_value->init_req(_false_path, null()); // return null
3948
3949 // Set output state.
3950 set_control(_gvn.transform(result_rgn));
3951 set_all_memory(_gvn.transform(result_mem));
3952 set_i_o(_gvn.transform(result_io));
3953 set_result(result_rgn, result_value);
3954 return true;
3955 }
3956
3957 /*
3958 * The intrinsic is a model of this pseudo-code:
3959 *
3960 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3961 * if (carrierThread != thread) { // is virtual thread
3962 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3963 * bool excluded = vthread_epoch_raw & excluded_mask;
3964 * AtomicAccess::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3965 * AtomicAccess::store(&tl->_contextual_thread_excluded, is_excluded);
3966 * if (!excluded) {
3967 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3968 * AtomicAccess::store(&tl->_vthread_epoch, vthread_epoch);
3969 * }
3970 * AtomicAccess::release_store(&tl->_vthread, true);
3971 * return;
3972 * }
3973 * AtomicAccess::release_store(&tl->_vthread, false);
3974 */
3975 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3976 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3977
3978 Node* input_memory_state = reset_memory();
3979 set_all_memory(input_memory_state);
3980
3981 // The most significant bit of the u2 is used to denote thread exclusion
3982 Node* excluded_mask = _gvn.intcon(1 << 15);
3983 // The epoch generation is the range [1-32767]
3984 Node* epoch_mask = _gvn.intcon(32767);
3985
3986 Node* const carrierThread = generate_current_thread(jt);
3987 // If thread != carrierThread, this is a virtual thread.
3988 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3989 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3990 IfNode* iff_thread_not_equal_carrierThread =
3991 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3992
3993 Node* vthread_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3994
3995 // False branch, is carrierThread.
3996 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3997 // Store release
3998 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3999
4000 set_all_memory(input_memory_state);
4001
4002 // True branch, is virtual thread.
4003 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
4004 set_control(thread_not_equal_carrierThread);
4005
4006 // Load the raw epoch value from the vthread.
4007 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
4008 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
4009 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
4010
4011 // Mask off the excluded information from the epoch.
4012 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, excluded_mask));
4013
4014 // Load the tid field from the thread.
4015 Node* tid = load_field_from_object(thread, "tid", "J");
4016
4017 // Store the vthread tid to the jfr thread local.
4018 Node* thread_id_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
4019 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
4020
4021 // Branch is_excluded to conditionalize updating the epoch .
4022 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, excluded_mask));
4023 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
4024 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
4025
4026 // True branch, vthread is excluded, no need to write epoch info.
4027 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
4028 set_control(excluded);
4029 Node* vthread_is_excluded = _gvn.intcon(1);
4030
4031 // False branch, vthread is included, update epoch info.
4032 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
4033 set_control(included);
4034 Node* vthread_is_included = _gvn.intcon(0);
4035
4036 // Get epoch value.
4037 Node* epoch = _gvn.transform(new AndINode(epoch_raw, epoch_mask));
4038
4039 // Store the vthread epoch to the jfr thread local.
4040 Node* vthread_epoch_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
4041 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
4042
4043 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
4044 record_for_igvn(excluded_rgn);
4045 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
4046 record_for_igvn(excluded_mem);
4047 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
4048 record_for_igvn(exclusion);
4049
4050 // Merge the excluded control and memory.
4051 excluded_rgn->init_req(_true_path, excluded);
4052 excluded_rgn->init_req(_false_path, included);
4053 excluded_mem->init_req(_true_path, tid_memory);
4054 excluded_mem->init_req(_false_path, included_memory);
4055 exclusion->init_req(_true_path, vthread_is_excluded);
4056 exclusion->init_req(_false_path, vthread_is_included);
4057
4058 // Set intermediate state.
4059 set_control(_gvn.transform(excluded_rgn));
4060 set_all_memory(excluded_mem);
4061
4062 // Store the vthread exclusion state to the jfr thread local.
4063 Node* thread_local_excluded_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
4064 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
4065
4066 // Store release
4067 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
4068
4069 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
4070 record_for_igvn(thread_compare_rgn);
4071 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
4072 record_for_igvn(thread_compare_mem);
4073 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
4074 record_for_igvn(vthread);
4075
4076 // Merge the thread_compare control and memory.
4077 thread_compare_rgn->init_req(_true_path, control());
4078 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
4079 thread_compare_mem->init_req(_true_path, vthread_true_memory);
4080 thread_compare_mem->init_req(_false_path, vthread_false_memory);
4081
4082 // Set output state.
4083 set_control(_gvn.transform(thread_compare_rgn));
4084 set_all_memory(_gvn.transform(thread_compare_mem));
4085 }
4086
4087 //------------------------inline_native_try_update_epoch------------------
4088 //
4089 // The generated code is a function of the argument type.
4090 //
4091 bool LibraryCallKit::inline_native_try_update_epoch() {
4092 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
4093
4094 // Save input memory.
4095 Node* input_memory_state = reset_memory();
4096 set_all_memory(input_memory_state);
4097
4098 // Argument is an oop whose class has an injected instance field,
4099 // called 'jfr_epoch' of type T_INT, used for holding a jfr epoch value.
4100 Node* oop = argument(0);
4101 const TypeInstPtr* tinst = _gvn.type(oop)->isa_instptr();
4102 assert(tinst != nullptr, "oop is null");
4103 assert(tinst->is_loaded(), "klass is not loaded");
4104 ciInstanceKlass* const ik = tinst->instance_klass();
4105
4106 ciField* const field = ik->get_injected_instance_field_by_name(ciSymbol::make("jfr_epoch"),
4107 ciSymbol::make("I"));
4108
4109 assert(field != nullptr, "field 'jfr_epoch' of type I not injected in klass %s", ik->name()->as_utf8());
4110
4111 const int jfr_epoch_field_offset = field->offset_in_bytes();
4112 Node* oop_epoch_field_offset = basic_plus_adr(oop, jfr_epoch_field_offset);
4113 const TypePtr* adr_type = _gvn.type(oop_epoch_field_offset)->isa_ptr();
4114 const int alias_idx = C->get_alias_index(adr_type);
4115 BasicType bt = field->layout_type();
4116 const Type * oop_epoch_field_type = Type::get_const_basic_type(bt);
4117
4118 // Load the epoch value from the oop.
4119 Node* oop_epoch = access_load_at(oop,
4120 oop_epoch_field_offset,
4121 adr_type, oop_epoch_field_type,
4122 bt, IN_HEAP | MO_UNORDERED);
4123
4124 // Load the current JFR epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
4125 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
4126 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
4127
4128 // Compare the epoch in the oop against the current JFR epoch generation.
4129 Node* const epochs_cmp = _gvn.transform(new CmpINode(current_epoch_generation, oop_epoch));
4130 Node* epochs_equal_test = _gvn.transform(new BoolNode(epochs_cmp, BoolTest::eq));
4131 IfNode* iff_epochs_equal = create_and_map_if(control(), epochs_equal_test, PROB_LIKELY(0.999), COUNT_UNKNOWN);
4132
4133 // True path.
4134 Node* epochs_are_equal = _gvn.transform(new IfTrueNode(iff_epochs_equal));
4135
4136 // False path.
4137 Node* epochs_are_not_equal = _gvn.transform(new IfFalseNode(iff_epochs_equal));
4138
4139 set_control(_gvn.transform(epochs_are_not_equal));
4140
4141 // Attempt to cas the current JFR epoch generation into the oop epoch field.
4142 DecoratorSet decorators = IN_HEAP;
4143 decorators |= mo_decorator_for_access_kind(Volatile);
4144
4145 Node* result = access_atomic_cmpxchg_val_at(oop,
4146 oop_epoch_field_offset,
4147 adr_type, alias_idx,
4148 oop_epoch, // expected value
4149 current_epoch_generation, // new value
4150 oop_epoch_field_type,
4151 bt,
4152 decorators);
4153
4154 // Compare the result of the cas operation to the expected value.
4155 Node* const cas_cmp_to_expected_value = _gvn.transform(new CmpINode(result, oop_epoch));
4156 Node* cas_operation_test = _gvn.transform(new BoolNode(cas_cmp_to_expected_value, BoolTest::eq));
4157 IfNode* iff_cas_success = create_and_map_if(control(), cas_operation_test, PROB_LIKELY(0.999), COUNT_UNKNOWN);
4158
4159 // True path.
4160 Node* cas_success = _gvn.transform(new IfTrueNode(iff_cas_success));
4161
4162 // False path.
4163 Node* cas_failure = _gvn.transform(new IfFalseNode(iff_cas_success));
4164
4165 // Cas result region and phi nodes.
4166 RegionNode* cas_operation_rgn = new RegionNode(PATH_LIMIT);
4167 record_for_igvn(cas_operation_rgn);
4168 PhiNode* cas_operation_mem = new PhiNode(cas_operation_rgn, Type::MEMORY, TypePtr::BOTTOM);
4169 record_for_igvn(cas_operation_mem);
4170 PhiNode* cas_result = new PhiNode(cas_operation_rgn, TypeInt::BOOL);
4171 record_for_igvn(cas_result);
4172
4173 cas_operation_rgn->init_req(_true_path, _gvn.transform(cas_success));
4174 cas_operation_rgn->init_req(_false_path, _gvn.transform(cas_failure));
4175 cas_operation_mem->init_req(_true_path, reset_memory());
4176 cas_operation_mem->init_req(_false_path, input_memory_state);
4177 cas_result->init_req(_true_path, _gvn.intcon(1));
4178 cas_result->init_req(_false_path, _gvn.intcon(0));
4179
4180 // Epoch compare region and phi nodes.
4181 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
4182 record_for_igvn(epoch_compare_rgn);
4183 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
4184 record_for_igvn(epoch_compare_mem);
4185 PhiNode* result_value = new PhiNode(epoch_compare_rgn, TypeInt::BOOL);
4186 record_for_igvn(result_value);
4187
4188 epoch_compare_rgn->init_req(_true_path, _gvn.transform(epochs_are_equal));
4189 epoch_compare_rgn->init_req(_false_path, _gvn.transform(cas_operation_rgn));
4190 epoch_compare_mem->init_req(_true_path, _gvn.transform(input_memory_state));
4191 epoch_compare_mem->init_req(_false_path, _gvn.transform(cas_operation_mem));
4192 result_value->init_req(_true_path, _gvn.intcon(0));
4193 result_value->init_req(_false_path, _gvn.transform(cas_result));
4194
4195 // Set output state.
4196 set_result(epoch_compare_rgn, result_value);
4197 set_all_memory(_gvn.transform(epoch_compare_mem));
4198
4199 return true;
4200 }
4201
4202 #endif // JFR_HAVE_INTRINSICS
4203
4204 //------------------------inline_native_currentCarrierThread------------------
4205 bool LibraryCallKit::inline_native_currentCarrierThread() {
4206 Node* junk = nullptr;
4207 set_result(generate_current_thread(junk));
4208 return true;
4209 }
4210
4211 //------------------------inline_native_currentThread------------------
4212 bool LibraryCallKit::inline_native_currentThread() {
4213 Node* junk = nullptr;
4214 set_result(generate_virtual_thread(junk));
4215 return true;
4216 }
4217
4218 //------------------------inline_native_setVthread------------------
4219 bool LibraryCallKit::inline_native_setCurrentThread() {
4220 assert(C->method()->changes_current_thread(),
4221 "method changes current Thread but is not annotated ChangesCurrentThread");
4222 Node* arr = argument(1);
4223 Node* thread = _gvn.transform(new ThreadLocalNode());
4224 Node* p = off_heap_plus_addr(thread, in_bytes(JavaThread::vthread_offset()));
4225 Node* thread_obj_handle
4226 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
4227 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
4228 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
4229
4230 // Change the _monitor_owner_id of the JavaThread
4231 Node* tid = load_field_from_object(arr, "tid", "J");
4232 Node* monitor_owner_id_offset = off_heap_plus_addr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
4233 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
4234
4235 JFR_ONLY(extend_setCurrentThread(thread, arr);)
4236 return true;
4237 }
4238
4239 const Type* LibraryCallKit::scopedValueCache_type() {
4240 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
4241 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
4242 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS, /* stable= */ false, /* flat= */ false, /* not_flat= */ true, /* not_null_free= */ true, true);
4243
4244 // Because we create the scopedValue cache lazily we have to make the
4245 // type of the result BotPTR.
4246 bool xk = etype->klass_is_exact();
4247 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, TypeAryPtr::Offset(0));
4248 return objects_type;
4249 }
4250
4251 Node* LibraryCallKit::scopedValueCache_helper() {
4252 Node* thread = _gvn.transform(new ThreadLocalNode());
4253 Node* p = off_heap_plus_addr(thread, in_bytes(JavaThread::scopedValueCache_offset()));
4254 // We cannot use immutable_memory() because we might flip onto a
4255 // different carrier thread, at which point we'll need to use that
4256 // carrier thread's cache.
4257 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
4258 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
4259 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
4260 }
4261
4262 //------------------------inline_native_scopedValueCache------------------
4263 bool LibraryCallKit::inline_native_scopedValueCache() {
4264 Node* cache_obj_handle = scopedValueCache_helper();
4265 const Type* objects_type = scopedValueCache_type();
4266 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
4267
4268 return true;
4269 }
4270
4271 //------------------------inline_native_setScopedValueCache------------------
4272 bool LibraryCallKit::inline_native_setScopedValueCache() {
4273 Node* arr = argument(0);
4274 Node* cache_obj_handle = scopedValueCache_helper();
4275 const Type* objects_type = scopedValueCache_type();
4276
4277 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
4278 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
4279
4280 return true;
4281 }
4282
4283 //------------------------inline_native_Continuation_pin and unpin-----------
4284
4285 // Shared implementation routine for both pin and unpin.
4286 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
4287 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
4288
4289 // Save input memory.
4290 Node* input_memory_state = reset_memory();
4291 set_all_memory(input_memory_state);
4292
4293 // TLS
4294 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
4295 Node* last_continuation_offset = off_heap_plus_addr(tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
4296 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
4297
4298 // Null check the last continuation object.
4299 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
4300 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
4301 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
4302
4303 // False path, last continuation is null.
4304 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
4305
4306 // True path, last continuation is not null.
4307 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
4308
4309 set_control(continuation_is_not_null);
4310
4311 // Load the pin count from the last continuation.
4312 Node* pin_count_offset = off_heap_plus_addr(last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
4313 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
4314
4315 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
4316 Node* pin_count_rhs;
4317 if (unpin) {
4318 pin_count_rhs = _gvn.intcon(0);
4319 } else {
4320 pin_count_rhs = _gvn.intcon(UINT32_MAX);
4321 }
4322 Node* pin_count_cmp = _gvn.transform(new CmpUNode(pin_count, pin_count_rhs));
4323 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
4324 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
4325
4326 // True branch, pin count over/underflow.
4327 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
4328 {
4329 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
4330 // which will throw IllegalStateException for pin count over/underflow.
4331 // No memory changed so far - we can use memory create by reset_memory()
4332 // at the beginning of this intrinsic. No need to call reset_memory() again.
4333 PreserveJVMState pjvms(this);
4334 set_control(pin_count_over_underflow);
4335 uncommon_trap(Deoptimization::Reason_intrinsic,
4336 Deoptimization::Action_none);
4337 assert(stopped(), "invariant");
4338 }
4339
4340 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
4341 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
4342 set_control(valid_pin_count);
4343
4344 Node* next_pin_count;
4345 if (unpin) {
4346 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
4347 } else {
4348 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
4349 }
4350
4351 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
4352
4353 // Result of top level CFG and Memory.
4354 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
4355 record_for_igvn(result_rgn);
4356 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
4357 record_for_igvn(result_mem);
4358
4359 result_rgn->init_req(_true_path, valid_pin_count);
4360 result_rgn->init_req(_false_path, continuation_is_null);
4361 result_mem->init_req(_true_path, reset_memory());
4362 result_mem->init_req(_false_path, input_memory_state);
4363
4364 // Set output state.
4365 set_control(_gvn.transform(result_rgn));
4366 set_all_memory(_gvn.transform(result_mem));
4367
4368 return true;
4369 }
4370
4371 //---------------------------load_mirror_from_klass----------------------------
4372 // Given a klass oop, load its java mirror (a java.lang.Class oop).
4373 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
4374 Node* p = off_heap_plus_addr(klass, in_bytes(Klass::java_mirror_offset()));
4375 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4376 // mirror = ((OopHandle)mirror)->resolve();
4377 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
4378 }
4379
4380 //-----------------------load_klass_from_mirror_common-------------------------
4381 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
4382 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
4383 // and branch to the given path on the region.
4384 // If never_see_null, take an uncommon trap on null, so we can optimistically
4385 // compile for the non-null case.
4386 // If the region is null, force never_see_null = true.
4387 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
4388 bool never_see_null,
4389 RegionNode* region,
4390 int null_path,
4391 int offset) {
4392 if (region == nullptr) never_see_null = true;
4393 Node* p = basic_plus_adr(mirror, offset);
4394 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4395 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
4396 Node* null_ctl = top();
4397 kls = null_check_oop(kls, &null_ctl, never_see_null);
4398 if (region != nullptr) {
4399 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
4400 region->init_req(null_path, null_ctl);
4401 } else {
4402 assert(null_ctl == top(), "no loose ends");
4403 }
4404 return kls;
4405 }
4406
4407 //--------------------(inline_native_Class_query helpers)---------------------
4408 // Use this for JVM_ACC_INTERFACE.
4409 // Fall through if (mods & mask) == bits, take the guard otherwise.
4410 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
4411 ByteSize offset, const Type* type, BasicType bt) {
4412 // Branch around if the given klass has the given modifier bit set.
4413 // Like generate_guard, adds a new path onto the region.
4414 Node* modp = off_heap_plus_addr(kls, in_bytes(offset));
4415 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
4416 Node* mask = intcon(modifier_mask);
4417 Node* bits = intcon(modifier_bits);
4418 Node* mbit = _gvn.transform(new AndINode(mods, mask));
4419 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
4420 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
4421 return generate_fair_guard(bol, region);
4422 }
4423
4424 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
4425 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
4426 InstanceKlass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
4427 }
4428
4429 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
4430 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
4431 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
4432 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
4433 }
4434
4435 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
4436 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
4437 }
4438
4439 //-------------------------inline_native_Class_query-------------------
4440 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
4441 const Type* return_type = TypeInt::BOOL;
4442 Node* prim_return_value = top(); // what happens if it's a primitive class?
4443 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4444 bool expect_prim = false; // most of these guys expect to work on refs
4445
4446 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
4447
4448 Node* mirror = argument(0);
4449 Node* obj = top();
4450
4451 switch (id) {
4452 case vmIntrinsics::_isInstance:
4453 // nothing is an instance of a primitive type
4454 prim_return_value = intcon(0);
4455 obj = argument(1);
4456 break;
4457 case vmIntrinsics::_isHidden:
4458 prim_return_value = intcon(0);
4459 break;
4460 case vmIntrinsics::_getSuperclass:
4461 prim_return_value = null();
4462 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
4463 break;
4464 default:
4465 fatal_unexpected_iid(id);
4466 break;
4467 }
4468
4469 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4470 if (mirror_con == nullptr) return false; // cannot happen?
4471
4472 #ifndef PRODUCT
4473 if (C->print_intrinsics() || C->print_inlining()) {
4474 ciType* k = mirror_con->java_mirror_type();
4475 if (k) {
4476 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4477 k->print_name();
4478 tty->cr();
4479 }
4480 }
4481 #endif
4482
4483 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4484 RegionNode* region = new RegionNode(PATH_LIMIT);
4485 record_for_igvn(region);
4486 PhiNode* phi = new PhiNode(region, return_type);
4487
4488 // The mirror will never be null of Reflection.getClassAccessFlags, however
4489 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4490 // if it is. See bug 4774291.
4491
4492 // For Reflection.getClassAccessFlags(), the null check occurs in
4493 // the wrong place; see inline_unsafe_access(), above, for a similar
4494 // situation.
4495 mirror = null_check(mirror);
4496 // If mirror or obj is dead, only null-path is taken.
4497 if (stopped()) return true;
4498
4499 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4500
4501 // Now load the mirror's klass metaobject, and null-check it.
4502 // Side-effects region with the control path if the klass is null.
4503 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4504 // If kls is null, we have a primitive mirror.
4505 phi->init_req(_prim_path, prim_return_value);
4506 if (stopped()) { set_result(region, phi); return true; }
4507 bool safe_for_replace = (region->in(_prim_path) == top());
4508
4509 Node* p; // handy temp
4510 Node* null_ctl;
4511
4512 // Now that we have the non-null klass, we can perform the real query.
4513 // For constant classes, the query will constant-fold in LoadNode::Value.
4514 Node* query_value = top();
4515 switch (id) {
4516 case vmIntrinsics::_isInstance:
4517 // nothing is an instance of a primitive type
4518 query_value = gen_instanceof(obj, kls, safe_for_replace);
4519 break;
4520
4521 case vmIntrinsics::_isHidden:
4522 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4523 if (generate_hidden_class_guard(kls, region) != nullptr)
4524 // A guard was added. If the guard is taken, it was an hidden class.
4525 phi->add_req(intcon(1));
4526 // If we fall through, it's a plain class.
4527 query_value = intcon(0);
4528 break;
4529
4530
4531 case vmIntrinsics::_getSuperclass:
4532 // The rules here are somewhat unfortunate, but we can still do better
4533 // with random logic than with a JNI call.
4534 // Interfaces store null or Object as _super, but must report null.
4535 // Arrays store an intermediate super as _super, but must report Object.
4536 // Other types can report the actual _super.
4537 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4538 if (generate_array_guard(kls, region) != nullptr) {
4539 // A guard was added. If the guard is taken, it was an array.
4540 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4541 }
4542 // Check for interface after array since this checks AccessFlags offset into InstanceKlass.
4543 // In other words, we are accessing subtype-specific information, so we need to determine the subtype first.
4544 if (generate_interface_guard(kls, region) != nullptr) {
4545 // A guard was added. If the guard is taken, it was an interface.
4546 phi->add_req(null());
4547 }
4548 // If we fall through, it's a plain class. Get its _super.
4549 if (!stopped()) {
4550 p = basic_plus_adr(top(), kls, in_bytes(Klass::super_offset()));
4551 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4552 null_ctl = top();
4553 kls = null_check_oop(kls, &null_ctl);
4554 if (null_ctl != top()) {
4555 // If the guard is taken, Object.superClass is null (both klass and mirror).
4556 region->add_req(null_ctl);
4557 phi ->add_req(null());
4558 }
4559 if (!stopped()) {
4560 query_value = load_mirror_from_klass(kls);
4561 }
4562 }
4563 break;
4564
4565 default:
4566 fatal_unexpected_iid(id);
4567 break;
4568 }
4569
4570 // Fall-through is the normal case of a query to a real class.
4571 phi->init_req(1, query_value);
4572 region->init_req(1, control());
4573
4574 C->set_has_split_ifs(true); // Has chance for split-if optimization
4575 set_result(region, phi);
4576 return true;
4577 }
4578
4579
4580 //-------------------------inline_Class_cast-------------------
4581 bool LibraryCallKit::inline_Class_cast() {
4582 Node* mirror = argument(0); // Class
4583 Node* obj = argument(1);
4584 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4585 if (mirror_con == nullptr) {
4586 return false; // dead path (mirror->is_top()).
4587 }
4588 if (obj == nullptr || obj->is_top()) {
4589 return false; // dead path
4590 }
4591 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4592
4593 // First, see if Class.cast() can be folded statically.
4594 // java_mirror_type() returns non-null for compile-time Class constants.
4595 ciType* tm = mirror_con->java_mirror_type();
4596 if (tm != nullptr && tm->is_klass() &&
4597 tp != nullptr) {
4598 if (!tp->is_loaded()) {
4599 // Don't use intrinsic when class is not loaded.
4600 return false;
4601 } else {
4602 const TypeKlassPtr* tklass = TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces);
4603 int static_res = C->static_subtype_check(tklass, tp->as_klass_type());
4604 if (static_res == Compile::SSC_always_true) {
4605 // isInstance() is true - fold the code.
4606 set_result(obj);
4607 return true;
4608 } else if (static_res == Compile::SSC_always_false) {
4609 // Don't use intrinsic, have to throw ClassCastException.
4610 // If the reference is null, the non-intrinsic bytecode will
4611 // be optimized appropriately.
4612 return false;
4613 }
4614 }
4615 }
4616
4617 // Bailout intrinsic and do normal inlining if exception path is frequent.
4618 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4619 return false;
4620 }
4621
4622 // Generate dynamic checks.
4623 // Class.cast() is java implementation of _checkcast bytecode.
4624 // Do checkcast (Parse::do_checkcast()) optimizations here.
4625
4626 mirror = null_check(mirror);
4627 // If mirror is dead, only null-path is taken.
4628 if (stopped()) {
4629 return true;
4630 }
4631
4632 // Not-subtype or the mirror's klass ptr is nullptr (in case it is a primitive).
4633 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
4634 RegionNode* region = new RegionNode(PATH_LIMIT);
4635 record_for_igvn(region);
4636
4637 // Now load the mirror's klass metaobject, and null-check it.
4638 // If kls is null, we have a primitive mirror and
4639 // nothing is an instance of a primitive type.
4640 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4641
4642 Node* res = top();
4643 Node* io = i_o();
4644 Node* mem = merged_memory();
4645 SafePointNode* new_cast_failure_map = nullptr;
4646
4647 if (!stopped()) {
4648
4649 Node* bad_type_ctrl = top();
4650 // Do checkcast optimizations.
4651 res = gen_checkcast(obj, kls, &bad_type_ctrl, &new_cast_failure_map);
4652 region->init_req(_bad_type_path, bad_type_ctrl);
4653 }
4654 if (region->in(_prim_path) != top() ||
4655 region->in(_bad_type_path) != top() ||
4656 region->in(_npe_path) != top()) {
4657 // Let Interpreter throw ClassCastException.
4658 PreserveJVMState pjvms(this);
4659 if (new_cast_failure_map != nullptr) {
4660 // The current map on the success path could have been modified. Use the dedicated failure path map.
4661 set_map(new_cast_failure_map);
4662 }
4663 set_control(_gvn.transform(region));
4664 // Set IO and memory because gen_checkcast may override them when buffering inline types
4665 set_i_o(io);
4666 set_all_memory(mem);
4667 uncommon_trap(Deoptimization::Reason_intrinsic,
4668 Deoptimization::Action_maybe_recompile);
4669 }
4670 if (!stopped()) {
4671 set_result(res);
4672 }
4673 return true;
4674 }
4675
4676
4677 //--------------------------inline_native_subtype_check------------------------
4678 // This intrinsic takes the JNI calls out of the heart of
4679 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4680 bool LibraryCallKit::inline_native_subtype_check() {
4681 // Pull both arguments off the stack.
4682 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4683 args[0] = argument(0);
4684 args[1] = argument(1);
4685 Node* klasses[2]; // corresponding Klasses: superk, subk
4686 klasses[0] = klasses[1] = top();
4687
4688 enum {
4689 // A full decision tree on {superc is prim, subc is prim}:
4690 _prim_0_path = 1, // {P,N} => false
4691 // {P,P} & superc!=subc => false
4692 _prim_same_path, // {P,P} & superc==subc => true
4693 _prim_1_path, // {N,P} => false
4694 _ref_subtype_path, // {N,N} & subtype check wins => true
4695 _both_ref_path, // {N,N} & subtype check loses => false
4696 PATH_LIMIT
4697 };
4698
4699 RegionNode* region = new RegionNode(PATH_LIMIT);
4700 RegionNode* prim_region = new RegionNode(2);
4701 Node* phi = new PhiNode(region, TypeInt::BOOL);
4702 record_for_igvn(region);
4703 record_for_igvn(prim_region);
4704
4705 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4706 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4707 int class_klass_offset = java_lang_Class::klass_offset();
4708
4709 // First null-check both mirrors and load each mirror's klass metaobject.
4710 int which_arg;
4711 for (which_arg = 0; which_arg <= 1; which_arg++) {
4712 Node* arg = args[which_arg];
4713 arg = null_check(arg);
4714 if (stopped()) break;
4715 args[which_arg] = arg;
4716
4717 Node* p = basic_plus_adr(arg, class_klass_offset);
4718 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4719 klasses[which_arg] = _gvn.transform(kls);
4720 }
4721
4722 // Having loaded both klasses, test each for null.
4723 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4724 for (which_arg = 0; which_arg <= 1; which_arg++) {
4725 Node* kls = klasses[which_arg];
4726 Node* null_ctl = top();
4727 kls = null_check_oop(kls, &null_ctl, never_see_null);
4728 if (which_arg == 0) {
4729 prim_region->init_req(1, null_ctl);
4730 } else {
4731 region->init_req(_prim_1_path, null_ctl);
4732 }
4733 if (stopped()) break;
4734 klasses[which_arg] = kls;
4735 }
4736
4737 if (!stopped()) {
4738 // now we have two reference types, in klasses[0..1]
4739 Node* subk = klasses[1]; // the argument to isAssignableFrom
4740 Node* superk = klasses[0]; // the receiver
4741 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4742 region->set_req(_ref_subtype_path, control());
4743 }
4744
4745 // If both operands are primitive (both klasses null), then
4746 // we must return true when they are identical primitives.
4747 // It is convenient to test this after the first null klass check.
4748 // This path is also used if superc is a value mirror.
4749 set_control(_gvn.transform(prim_region));
4750 if (!stopped()) {
4751 // Since superc is primitive, make a guard for the superc==subc case.
4752 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4753 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4754 generate_fair_guard(bol_eq, region);
4755 if (region->req() == PATH_LIMIT+1) {
4756 // A guard was added. If the added guard is taken, superc==subc.
4757 region->swap_edges(PATH_LIMIT, _prim_same_path);
4758 region->del_req(PATH_LIMIT);
4759 }
4760 region->set_req(_prim_0_path, control()); // Not equal after all.
4761 }
4762
4763 // these are the only paths that produce 'true':
4764 phi->set_req(_prim_same_path, intcon(1));
4765 phi->set_req(_ref_subtype_path, intcon(1));
4766
4767 // pull together the cases:
4768 assert(region->req() == PATH_LIMIT, "sane region");
4769 for (uint i = 1; i < region->req(); i++) {
4770 Node* ctl = region->in(i);
4771 if (ctl == nullptr || ctl == top()) {
4772 region->set_req(i, top());
4773 phi ->set_req(i, top());
4774 } else if (phi->in(i) == nullptr) {
4775 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4776 }
4777 }
4778
4779 set_control(_gvn.transform(region));
4780 set_result(_gvn.transform(phi));
4781 return true;
4782 }
4783
4784 //---------------------generate_array_guard_common------------------------
4785 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind, Node** obj) {
4786
4787 if (stopped()) {
4788 return nullptr;
4789 }
4790
4791 // Like generate_guard, adds a new path onto the region.
4792 jint layout_con = 0;
4793 Node* layout_val = get_layout_helper(kls, layout_con);
4794 if (layout_val == nullptr) {
4795 bool query = 0;
4796 switch(kind) {
4797 case RefArray: query = Klass::layout_helper_is_refArray(layout_con); break;
4798 case NonRefArray: query = !Klass::layout_helper_is_refArray(layout_con); break;
4799 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
4800 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
4801 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
4802 default:
4803 ShouldNotReachHere();
4804 }
4805 if (!query) {
4806 return nullptr; // never a branch
4807 } else { // always a branch
4808 Node* always_branch = control();
4809 if (region != nullptr)
4810 region->add_req(always_branch);
4811 set_control(top());
4812 return always_branch;
4813 }
4814 }
4815 unsigned int value = 0;
4816 BoolTest::mask btest = BoolTest::illegal;
4817 switch(kind) {
4818 case RefArray:
4819 case NonRefArray: {
4820 value = Klass::_lh_array_tag_ref_value;
4821 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4822 btest = (kind == RefArray) ? BoolTest::eq : BoolTest::ne;
4823 break;
4824 }
4825 case TypeArray: {
4826 value = Klass::_lh_array_tag_type_value;
4827 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4828 btest = BoolTest::eq;
4829 break;
4830 }
4831 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
4832 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
4833 default:
4834 ShouldNotReachHere();
4835 }
4836 // Now test the correct condition.
4837 jint nval = (jint)value;
4838 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4839 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4840 Node* ctrl = generate_fair_guard(bol, region);
4841 Node* is_array_ctrl = kind == NonArray ? control() : ctrl;
4842 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4843 // Keep track of the fact that 'obj' is an array to prevent
4844 // array specific accesses from floating above the guard.
4845 *obj = _gvn.transform(new CheckCastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4846 }
4847 return ctrl;
4848 }
4849
4850 // public static native Object[] ValueClass::newNullRestrictedAtomicArray(Class<?> componentType, int length, Object initVal);
4851 // public static native Object[] ValueClass::newNullRestrictedNonAtomicArray(Class<?> componentType, int length, Object initVal);
4852 // public static native Object[] ValueClass::newNullableAtomicArray(Class<?> componentType, int length);
4853 bool LibraryCallKit::inline_newArray(bool null_free, bool atomic) {
4854 assert(null_free || atomic, "nullable implies atomic");
4855 Node* componentType = argument(0);
4856 Node* length = argument(1);
4857 Node* init_val = null_free ? argument(2) : nullptr;
4858
4859 const TypeInstPtr* tp = _gvn.type(componentType)->isa_instptr();
4860 if (tp != nullptr) {
4861 ciInstanceKlass* ik = tp->instance_klass();
4862 if (ik == C->env()->Class_klass()) {
4863 ciType* t = tp->java_mirror_type();
4864 if (t != nullptr && t->is_inlinetype()) {
4865
4866 ciArrayKlass* array_klass = ciArrayKlass::make(t, null_free, atomic, true);
4867 assert(array_klass->is_elem_null_free() == null_free, "inconsistency");
4868
4869 // TOOD 8350865 ZGC needs card marks on initializing oop stores
4870 if (UseZGC && null_free && !array_klass->is_flat_array_klass()) {
4871 return false;
4872 }
4873
4874 if (array_klass->is_loaded() && array_klass->element_klass()->as_inline_klass()->is_initialized()) {
4875 const TypeAryKlassPtr* array_klass_type = TypeAryKlassPtr::make(array_klass, Type::trust_interfaces);
4876 if (null_free) {
4877 if (init_val->is_InlineType()) {
4878 if (array_klass_type->is_flat() && init_val->as_InlineType()->is_all_zero(&gvn(), /* flat */ true)) {
4879 // Zeroing is enough because the init value is the all-zero value
4880 init_val = nullptr;
4881 } else {
4882 init_val = init_val->as_InlineType()->buffer(this);
4883 }
4884 }
4885 if (init_val != nullptr) {
4886 #ifdef ASSERT
4887 init_val = null_check(init_val);
4888 Node* wrong_type_ctl = gen_subtype_check(init_val, makecon(TypeKlassPtr::make(array_klass->element_klass())));
4889 {
4890 PreserveJVMState pjvms(this);
4891 set_control(wrong_type_ctl);
4892 halt(control(), frameptr(), "incompatible type for initVal in newArray");
4893 stop_and_kill_map();
4894 }
4895 #endif
4896 init_val = _gvn.transform(new CheckCastPPNode(control(), init_val, TypeOopPtr::make_from_klass(array_klass->element_klass()), ConstraintCastNode::DependencyType::NonFloatingNarrowing));
4897 }
4898 }
4899 Node* obj = new_array(makecon(array_klass_type), length, 0, nullptr, false, init_val);
4900 const TypeAryPtr* arytype = gvn().type(obj)->is_aryptr();
4901 assert(arytype->is_null_free() == null_free, "inconsistency");
4902 assert(arytype->is_not_null_free() == !null_free, "inconsistency");
4903 set_result(obj);
4904 return true;
4905 }
4906 }
4907 }
4908 }
4909 return false;
4910 }
4911
4912 // public static native boolean ValueClass::isFlatArray(Object array);
4913 // public static native boolean ValueClass::isNullRestrictedArray(Object array);
4914 // public static native boolean ValueClass::isAtomicArray(Object array);
4915 bool LibraryCallKit::inline_getArrayProperties(ArrayPropertiesCheck check) {
4916 Node* array = argument(0);
4917
4918 Node* bol;
4919 switch(check) {
4920 case IsFlat:
4921 bol = flat_array_test(load_object_klass(array));
4922 break;
4923 case IsNullRestricted:
4924 bol = null_free_array_test(array);
4925 break;
4926 case IsAtomic: {
4927 // See conditions in JVM_IsAtomicArray
4928 // 1. If not flat, then atomic, or else...
4929 RegionNode* atomic_region = new RegionNode(1);
4930 RegionNode* non_atomic_region = new RegionNode(1);
4931 Node* array_klass = load_object_klass(array);
4932 Node* is_flat_bol = flat_array_test(array_klass);
4933 IfNode* iff_is_flat = create_and_xform_if(control(), is_flat_bol, PROB_FAIR, COUNT_UNKNOWN);
4934 atomic_region->add_req(_gvn.transform(new IfFalseNode(iff_is_flat)));
4935 set_control(_gvn.transform(new IfTrueNode(iff_is_flat)));
4936
4937 // 2. ...if the layout is atomic, then atomic, or else...
4938 Node* layout_kind = atomic_layout_array_test_and_get_layout_kind(array, atomic_region);
4939
4940 // 3. ...if the element type is naturally atomic and null-free OR empty and nullable, then atomic, or else...
4941 int element_klass_offset = in_bytes(ObjArrayKlass::element_klass_offset());
4942 Node* array_element_klass_addr = off_heap_plus_addr(array_klass, element_klass_offset);
4943 Node* array_element_klass = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), array_element_klass_addr, _gvn.type(array_klass)->is_klassptr()));
4944 int klass_flags_offset = in_bytes(InstanceKlass::misc_flags_offset() + InstanceKlassFlags::flags_offset());
4945 Node* array_element_klass_flags_addr = off_heap_plus_addr(array_element_klass, klass_flags_offset);
4946 Node* array_element_klass_flags = make_load(control(), array_element_klass_flags_addr, TypeInt::INT, T_INT, MemNode::unordered);
4947
4948 // Here, layout can only be non-atomic, otherwise atomic_layout_array_test_and_get_layout_kind already decides the array to be atomic.
4949 Node* is_null_free_cmp = _gvn.transform(new CmpINode(layout_kind, intcon(static_cast<jint>(LayoutKind::NULL_FREE_NON_ATOMIC_FLAT))));
4950 Node* is_null_free_bol = _gvn.transform(new BoolNode(is_null_free_cmp, BoolTest::eq));
4951 IfNode* iff_is_null_free_bol = create_and_xform_if(control(), is_null_free_bol, PROB_FAIR, COUNT_UNKNOWN);
4952 Node* is_null_free_ctl = _gvn.transform(new IfTrueNode(iff_is_null_free_bol));
4953 Node* is_nullable_ctl = _gvn.transform(new IfFalseNode(iff_is_null_free_bol));
4954
4955 Node* is_naturally_atomic_flag = _gvn.transform(new AndINode(array_element_klass_flags, intcon(InstanceKlassFlags::_misc_is_naturally_atomic)));
4956 Node* is_naturally_atomic_cmp = _gvn.transform(new CmpINode(is_naturally_atomic_flag, intcon(0)));
4957 Node* is_naturally_atomic_bol = _gvn.transform(new BoolNode(is_naturally_atomic_cmp, BoolTest::ne));
4958 IfNode* iff_is_naturally_atomic = create_and_xform_if(is_null_free_ctl, is_naturally_atomic_bol, PROB_FAIR, COUNT_UNKNOWN);
4959 Node* is_naturally_atomic_ctl = _gvn.transform(new IfTrueNode(iff_is_naturally_atomic));
4960 Node* is_not_naturally_atomic_ctl = _gvn.transform(new IfFalseNode(iff_is_naturally_atomic));
4961 atomic_region->add_req(is_naturally_atomic_ctl);
4962 non_atomic_region->add_req(is_not_naturally_atomic_ctl);
4963
4964 Node* is_empty_inline_type_flag = _gvn.transform(new AndINode(array_element_klass_flags, intcon(InstanceKlassFlags::_misc_is_empty_inline_type)));
4965 Node* is_empty_inline_type_cmp = _gvn.transform(new CmpINode(is_empty_inline_type_flag, intcon(0)));
4966 Node* is_empty_inline_type_bol = _gvn.transform(new BoolNode(is_empty_inline_type_cmp, BoolTest::ne));
4967 IfNode* iff_is_empty_inline_type = create_and_xform_if(is_nullable_ctl, is_empty_inline_type_bol, PROB_FAIR, COUNT_UNKNOWN);
4968 Node* is_empty_inline_type_ctl = _gvn.transform(new IfTrueNode(iff_is_empty_inline_type));
4969 Node* is_nonempty_inline_type_ctl = _gvn.transform(new IfFalseNode(iff_is_empty_inline_type));
4970 atomic_region->add_req(is_empty_inline_type_ctl);
4971 non_atomic_region->add_req(is_nonempty_inline_type_ctl);
4972
4973 // ...non-atomic, but we tried everything.
4974 RegionNode* decision = new RegionNode(3);
4975 decision->set_req(1, _gvn.transform(atomic_region));
4976 decision->set_req(2, _gvn.transform(non_atomic_region));
4977 PhiNode* result = PhiNode::make(decision, intcon(1), TypeInt::BOOL);
4978 result->set_req(2, intcon(0));
4979 set_control(_gvn.transform(decision));
4980 set_result(_gvn.transform(result));
4981 return true;
4982 }
4983 default:
4984 ShouldNotReachHere();
4985 }
4986
4987 Node* res = gvn().transform(new CMoveINode(bol, intcon(0), intcon(1), TypeInt::BOOL));
4988 set_result(res);
4989 return true;
4990 }
4991
4992 // Load the default refined array klass from an ObjArrayKlass. This relies on the first entry in the
4993 // '_next_refined_array_klass' linked list being the default (see ObjArrayKlass::klass_with_properties).
4994 Node* LibraryCallKit::load_default_refined_array_klass(Node* klass_node, bool type_array_guard) {
4995 RegionNode* region = new RegionNode(2);
4996 Node* phi = new PhiNode(region, TypeInstKlassPtr::OBJECT_OR_NULL);
4997
4998 if (type_array_guard) {
4999 generate_typeArray_guard(klass_node, region);
5000 if (region->req() == 3) {
5001 phi->add_req(klass_node);
5002 }
5003 }
5004 Node* adr_refined_klass = basic_plus_adr(top(), klass_node, in_bytes(ObjArrayKlass::next_refined_array_klass_offset()));
5005 Node* refined_klass = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), adr_refined_klass, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
5006
5007 // Can be null if not initialized yet, just deopt
5008 Node* null_ctl = top();
5009 refined_klass = null_check_oop(refined_klass, &null_ctl, /* never_see_null= */ true);
5010
5011 region->init_req(1, control());
5012 phi->init_req(1, refined_klass);
5013
5014 set_control(_gvn.transform(region));
5015 return _gvn.transform(phi);
5016 }
5017
5018 // Load the non-refined array klass from an ObjArrayKlass.
5019 Node* LibraryCallKit::load_non_refined_array_klass(Node* klass_node) {
5020 const TypeAryKlassPtr* ary_klass_ptr = _gvn.type(klass_node)->isa_aryklassptr();
5021 if (ary_klass_ptr != nullptr && ary_klass_ptr->klass_is_exact()) {
5022 return _gvn.makecon(ary_klass_ptr->cast_to_refined_array_klass_ptr(false));
5023 }
5024
5025 RegionNode* region = new RegionNode(2);
5026 Node* phi = new PhiNode(region, TypeInstKlassPtr::OBJECT);
5027
5028 generate_typeArray_guard(klass_node, region);
5029 if (region->req() == 3) {
5030 phi->add_req(klass_node);
5031 }
5032 Node* super_adr = basic_plus_adr(top(), klass_node, in_bytes(Klass::super_offset()));
5033 Node* super_klass = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), super_adr, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT));
5034
5035 region->init_req(1, control());
5036 phi->init_req(1, super_klass);
5037
5038 set_control(_gvn.transform(region));
5039 return _gvn.transform(phi);
5040 }
5041
5042 //-----------------------inline_native_newArray--------------------------
5043 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
5044 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
5045 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
5046 Node* mirror;
5047 Node* count_val;
5048 if (uninitialized) {
5049 null_check_receiver();
5050 mirror = argument(1);
5051 count_val = argument(2);
5052 } else {
5053 mirror = argument(0);
5054 count_val = argument(1);
5055 }
5056
5057 mirror = null_check(mirror);
5058 // If mirror or obj is dead, only null-path is taken.
5059 if (stopped()) return true;
5060
5061 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
5062 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5063 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5064 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
5065 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5066
5067 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
5068 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
5069 result_reg, _slow_path);
5070 Node* normal_ctl = control();
5071 Node* no_array_ctl = result_reg->in(_slow_path);
5072
5073 // Generate code for the slow case. We make a call to newArray().
5074 set_control(no_array_ctl);
5075 if (!stopped()) {
5076 // Either the input type is void.class, or else the
5077 // array klass has not yet been cached. Either the
5078 // ensuing call will throw an exception, or else it
5079 // will cache the array klass for next time.
5080 PreserveJVMState pjvms(this);
5081 CallJavaNode* slow_call = nullptr;
5082 if (uninitialized) {
5083 // Generate optimized virtual call (holder class 'Unsafe' is final)
5084 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
5085 } else {
5086 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
5087 }
5088 Node* slow_result = set_results_for_java_call(slow_call);
5089 // this->control() comes from set_results_for_java_call
5090 result_reg->set_req(_slow_path, control());
5091 result_val->set_req(_slow_path, slow_result);
5092 result_io ->set_req(_slow_path, i_o());
5093 result_mem->set_req(_slow_path, reset_memory());
5094 }
5095
5096 set_control(normal_ctl);
5097 if (!stopped()) {
5098 // Normal case: The array type has been cached in the java.lang.Class.
5099 // The following call works fine even if the array type is polymorphic.
5100 // It could be a dynamic mix of int[], boolean[], Object[], etc.
5101
5102 klass_node = load_default_refined_array_klass(klass_node);
5103
5104 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
5105 result_reg->init_req(_normal_path, control());
5106 result_val->init_req(_normal_path, obj);
5107 result_io ->init_req(_normal_path, i_o());
5108 result_mem->init_req(_normal_path, reset_memory());
5109
5110 if (uninitialized) {
5111 // Mark the allocation so that zeroing is skipped
5112 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
5113 alloc->maybe_set_complete(&_gvn);
5114 }
5115 }
5116
5117 // Return the combined state.
5118 set_i_o( _gvn.transform(result_io) );
5119 set_all_memory( _gvn.transform(result_mem));
5120
5121 C->set_has_split_ifs(true); // Has chance for split-if optimization
5122 set_result(result_reg, result_val);
5123 return true;
5124 }
5125
5126 //----------------------inline_native_getLength--------------------------
5127 // public static native int java.lang.reflect.Array.getLength(Object array);
5128 bool LibraryCallKit::inline_native_getLength() {
5129 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
5130
5131 Node* array = null_check(argument(0));
5132 // If array is dead, only null-path is taken.
5133 if (stopped()) return true;
5134
5135 // Deoptimize if it is a non-array.
5136 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
5137
5138 if (non_array != nullptr) {
5139 PreserveJVMState pjvms(this);
5140 set_control(non_array);
5141 uncommon_trap(Deoptimization::Reason_intrinsic,
5142 Deoptimization::Action_maybe_recompile);
5143 }
5144
5145 // If control is dead, only non-array-path is taken.
5146 if (stopped()) return true;
5147
5148 // The works fine even if the array type is polymorphic.
5149 // It could be a dynamic mix of int[], boolean[], Object[], etc.
5150 Node* result = load_array_length(array);
5151
5152 C->set_has_split_ifs(true); // Has chance for split-if optimization
5153 set_result(result);
5154 return true;
5155 }
5156
5157 //------------------------inline_array_copyOf----------------------------
5158 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
5159 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
5160 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
5161 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
5162
5163 // Get the arguments.
5164 Node* original = argument(0);
5165 Node* start = is_copyOfRange? argument(1): intcon(0);
5166 Node* end = is_copyOfRange? argument(2): argument(1);
5167 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
5168
5169 Node* newcopy = nullptr;
5170
5171 // Set the original stack and the reexecute bit for the interpreter to reexecute
5172 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
5173 { PreserveReexecuteState preexecs(this);
5174 jvms()->set_should_reexecute(true);
5175
5176 array_type_mirror = null_check(array_type_mirror);
5177 original = null_check(original);
5178
5179 // Check if a null path was taken unconditionally.
5180 if (stopped()) return true;
5181
5182 Node* orig_length = load_array_length(original);
5183
5184 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
5185 klass_node = null_check(klass_node);
5186
5187 RegionNode* bailout = new RegionNode(1);
5188 record_for_igvn(bailout);
5189
5190 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
5191 // Bail out if that is so.
5192 // Inline type array may have object field that would require a
5193 // write barrier. Conservatively, go to slow path.
5194 // TODO 8251971: Optimize for the case when flat src/dst are later found
5195 // to not contain oops (i.e., move this check to the macro expansion phase).
5196 // TODO 8382226: Revisit for flat abstract value class arrays
5197 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5198 const TypeAryPtr* orig_t = _gvn.type(original)->isa_aryptr();
5199 const TypeKlassPtr* tklass = _gvn.type(klass_node)->is_klassptr();
5200 bool exclude_flat = UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing) &&
5201 // Can src array be flat and contain oops?
5202 (orig_t == nullptr || (!orig_t->is_not_flat() && (!orig_t->is_flat() || orig_t->elem()->inline_klass()->contains_oops()))) &&
5203 // Can dest array be flat and contain oops?
5204 tklass->can_be_inline_array() && (!tklass->is_flat() || tklass->is_aryklassptr()->elem()->is_instklassptr()->instance_klass()->as_inline_klass()->contains_oops());
5205 Node* not_objArray = exclude_flat ? generate_non_refArray_guard(klass_node, bailout) : generate_typeArray_guard(klass_node, bailout);
5206
5207 Node* refined_klass_node = load_default_refined_array_klass(klass_node, /* type_array_guard= */ false);
5208
5209 if (not_objArray != nullptr) {
5210 // Improve the klass node's type from the new optimistic assumption:
5211 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
5212 bool not_flat = !UseArrayFlattening;
5213 bool not_null_free = !Arguments::is_valhalla_enabled();
5214 const Type* akls = TypeAryKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0), Type::trust_interfaces, not_flat, not_null_free, false, false, not_flat, true);
5215 Node* cast = new CastPPNode(control(), refined_klass_node, akls);
5216 refined_klass_node = _gvn.transform(cast);
5217 }
5218
5219 // Bail out if either start or end is negative.
5220 generate_negative_guard(start, bailout, &start);
5221 generate_negative_guard(end, bailout, &end);
5222
5223 Node* length = end;
5224 if (_gvn.type(start) != TypeInt::ZERO) {
5225 length = _gvn.transform(new SubINode(end, start));
5226 }
5227
5228 // Bail out if length is negative (i.e., if start > end).
5229 // Without this the new_array would throw
5230 // NegativeArraySizeException but IllegalArgumentException is what
5231 // should be thrown
5232 generate_negative_guard(length, bailout, &length);
5233
5234 // Handle inline type arrays
5235 // TODO 8251971 This is too strong
5236 generate_fair_guard(flat_array_test(load_object_klass(original)), bailout);
5237 generate_fair_guard(flat_array_test(refined_klass_node), bailout);
5238 generate_fair_guard(null_free_array_test(original), bailout);
5239
5240 // Bail out if start is larger than the original length
5241 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
5242 generate_negative_guard(orig_tail, bailout, &orig_tail);
5243
5244 if (bailout->req() > 1) {
5245 PreserveJVMState pjvms(this);
5246 set_control(_gvn.transform(bailout));
5247 uncommon_trap(Deoptimization::Reason_intrinsic,
5248 Deoptimization::Action_maybe_recompile);
5249 }
5250
5251 if (!stopped()) {
5252 // How many elements will we copy from the original?
5253 // The answer is MinI(orig_tail, length).
5254 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
5255
5256 // Generate a direct call to the right arraycopy function(s).
5257 // We know the copy is disjoint but we might not know if the
5258 // oop stores need checking.
5259 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
5260 // This will fail a store-check if x contains any non-nulls.
5261
5262 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
5263 // loads/stores but it is legal only if we're sure the
5264 // Arrays.copyOf would succeed. So we need all input arguments
5265 // to the copyOf to be validated, including that the copy to the
5266 // new array won't trigger an ArrayStoreException. That subtype
5267 // check can be optimized if we know something on the type of
5268 // the input array from type speculation.
5269 if (_gvn.type(klass_node)->singleton()) {
5270 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
5271 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
5272
5273 int test = C->static_subtype_check(superk, subk);
5274 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
5275 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
5276 if (t_original->speculative_type() != nullptr) {
5277 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
5278 }
5279 }
5280 }
5281
5282 bool validated = false;
5283 // Reason_class_check rather than Reason_intrinsic because we
5284 // want to intrinsify even if this traps.
5285 if (!too_many_traps(Deoptimization::Reason_class_check)) {
5286 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
5287
5288 if (not_subtype_ctrl != top()) {
5289 PreserveJVMState pjvms(this);
5290 set_control(not_subtype_ctrl);
5291 uncommon_trap(Deoptimization::Reason_class_check,
5292 Deoptimization::Action_make_not_entrant);
5293 assert(stopped(), "Should be stopped");
5294 }
5295 validated = true;
5296 }
5297
5298 if (!stopped()) {
5299 newcopy = new_array(refined_klass_node, length, 0); // no arguments to push
5300
5301 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
5302 load_object_klass(original), klass_node);
5303 if (!is_copyOfRange) {
5304 ac->set_copyof(validated);
5305 } else {
5306 ac->set_copyofrange(validated);
5307 }
5308 Node* n = _gvn.transform(ac);
5309 if (n == ac) {
5310 ac->connect_outputs(this);
5311 } else {
5312 assert(validated, "shouldn't transform if all arguments not validated");
5313 set_all_memory(n);
5314 }
5315 }
5316 }
5317 } // original reexecute is set back here
5318
5319 C->set_has_split_ifs(true); // Has chance for split-if optimization
5320 if (!stopped()) {
5321 set_result(newcopy);
5322 }
5323 return true;
5324 }
5325
5326
5327 //----------------------generate_virtual_guard---------------------------
5328 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
5329 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
5330 RegionNode* slow_region) {
5331 ciMethod* method = callee();
5332 int vtable_index = method->vtable_index();
5333 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5334 "bad index %d", vtable_index);
5335 // Get the Method* out of the appropriate vtable entry.
5336 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
5337 vtable_index*vtableEntry::size_in_bytes() +
5338 in_bytes(vtableEntry::method_offset());
5339 Node* entry_addr = off_heap_plus_addr(obj_klass, entry_offset);
5340 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
5341
5342 // Compare the target method with the expected method (e.g., Object.hashCode).
5343 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
5344
5345 Node* native_call = makecon(native_call_addr);
5346 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
5347 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
5348
5349 return generate_slow_guard(test_native, slow_region);
5350 }
5351
5352 //-----------------------generate_method_call----------------------------
5353 // Use generate_method_call to make a slow-call to the real
5354 // method if the fast path fails. An alternative would be to
5355 // use a stub like OptoRuntime::slow_arraycopy_Java.
5356 // This only works for expanding the current library call,
5357 // not another intrinsic. (E.g., don't use this for making an
5358 // arraycopy call inside of the copyOf intrinsic.)
5359 CallJavaNode*
5360 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
5361 // When compiling the intrinsic method itself, do not use this technique.
5362 guarantee(callee() != C->method(), "cannot make slow-call to self");
5363
5364 ciMethod* method = callee();
5365 // ensure the JVMS we have will be correct for this call
5366 guarantee(method_id == method->intrinsic_id(), "must match");
5367
5368 const TypeFunc* tf = TypeFunc::make(method);
5369 if (res_not_null) {
5370 assert(tf->return_type() == T_OBJECT, "");
5371 const TypeTuple* range = tf->range_cc();
5372 const Type** fields = TypeTuple::fields(range->cnt());
5373 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
5374 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
5375 tf = TypeFunc::make(tf->domain_cc(), new_range);
5376 }
5377 CallJavaNode* slow_call;
5378 if (is_static) {
5379 assert(!is_virtual, "");
5380 slow_call = new CallStaticJavaNode(C, tf,
5381 SharedRuntime::get_resolve_static_call_stub(), method);
5382 } else if (is_virtual) {
5383 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5384 int vtable_index = Method::invalid_vtable_index;
5385 if (UseInlineCaches) {
5386 // Suppress the vtable call
5387 } else {
5388 // hashCode and clone are not a miranda methods,
5389 // so the vtable index is fixed.
5390 // No need to use the linkResolver to get it.
5391 vtable_index = method->vtable_index();
5392 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5393 "bad index %d", vtable_index);
5394 }
5395 slow_call = new CallDynamicJavaNode(tf,
5396 SharedRuntime::get_resolve_virtual_call_stub(),
5397 method, vtable_index);
5398 } else { // neither virtual nor static: opt_virtual
5399 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5400 slow_call = new CallStaticJavaNode(C, tf,
5401 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
5402 slow_call->set_optimized_virtual(true);
5403 }
5404 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
5405 // To be able to issue a direct call (optimized virtual or virtual)
5406 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
5407 // about the method being invoked should be attached to the call site to
5408 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
5409 slow_call->set_override_symbolic_info(true);
5410 }
5411 set_arguments_for_java_call(slow_call);
5412 set_edges_for_java_call(slow_call);
5413 return slow_call;
5414 }
5415
5416
5417 /**
5418 * Build special case code for calls to hashCode on an object. This call may
5419 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
5420 * slightly different code.
5421 */
5422 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
5423 assert(is_static == callee()->is_static(), "correct intrinsic selection");
5424 assert(!(is_virtual && is_static), "either virtual, special, or static");
5425
5426 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
5427
5428 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5429 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
5430 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
5431 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5432 Node* obj = argument(0);
5433
5434 // Don't intrinsify hashcode on inline types for now.
5435 // The "is locked" runtime check also subsumes the inline type check (as inline types cannot be locked) and goes to the slow path.
5436 if (gvn().type(obj)->is_inlinetypeptr()) {
5437 return false;
5438 }
5439
5440 if (!is_static) {
5441 // Check for hashing null object
5442 obj = null_check_receiver();
5443 if (stopped()) return true; // unconditionally null
5444 result_reg->init_req(_null_path, top());
5445 result_val->init_req(_null_path, top());
5446 } else {
5447 // Do a null check, and return zero if null.
5448 // System.identityHashCode(null) == 0
5449 Node* null_ctl = top();
5450 obj = null_check_oop(obj, &null_ctl);
5451 result_reg->init_req(_null_path, null_ctl);
5452 result_val->init_req(_null_path, _gvn.intcon(0));
5453 }
5454
5455 // Unconditionally null? Then return right away.
5456 if (stopped()) {
5457 set_control( result_reg->in(_null_path));
5458 if (!stopped())
5459 set_result(result_val->in(_null_path));
5460 return true;
5461 }
5462
5463 // We only go to the fast case code if we pass a number of guards. The
5464 // paths which do not pass are accumulated in the slow_region.
5465 RegionNode* slow_region = new RegionNode(1);
5466 record_for_igvn(slow_region);
5467
5468 // If this is a virtual call, we generate a funny guard. We pull out
5469 // the vtable entry corresponding to hashCode() from the target object.
5470 // If the target method which we are calling happens to be the native
5471 // Object hashCode() method, we pass the guard. We do not need this
5472 // guard for non-virtual calls -- the caller is known to be the native
5473 // Object hashCode().
5474 if (is_virtual) {
5475 // After null check, get the object's klass.
5476 Node* obj_klass = load_object_klass(obj);
5477 generate_virtual_guard(obj_klass, slow_region);
5478 }
5479
5480 // Get the header out of the object, use LoadMarkNode when available
5481 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
5482 // The control of the load must be null. Otherwise, the load can move before
5483 // the null check after castPP removal.
5484 Node* no_ctrl = nullptr;
5485 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
5486
5487 if (!UseObjectMonitorTable) {
5488 // Test the header to see if it is safe to read w.r.t. locking.
5489 // We cannot use the inline type mask as this may check bits that are overridden
5490 // by an object monitor's pointer when inflating locking.
5491 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
5492 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
5493 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
5494 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
5495 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
5496
5497 generate_slow_guard(test_monitor, slow_region);
5498 }
5499
5500 // Get the hash value and check to see that it has been properly assigned.
5501 // We depend on hash_mask being at most 32 bits and avoid the use of
5502 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
5503 // vm: see markWord.hpp.
5504 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
5505 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
5506 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
5507 // This hack lets the hash bits live anywhere in the mark object now, as long
5508 // as the shift drops the relevant bits into the low 32 bits. Note that
5509 // Java spec says that HashCode is an int so there's no point in capturing
5510 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
5511 hshifted_header = ConvX2I(hshifted_header);
5512 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
5513
5514 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
5515 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
5516 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
5517
5518 generate_slow_guard(test_assigned, slow_region);
5519
5520 Node* init_mem = reset_memory();
5521 // fill in the rest of the null path:
5522 result_io ->init_req(_null_path, i_o());
5523 result_mem->init_req(_null_path, init_mem);
5524
5525 result_val->init_req(_fast_path, hash_val);
5526 result_reg->init_req(_fast_path, control());
5527 result_io ->init_req(_fast_path, i_o());
5528 result_mem->init_req(_fast_path, init_mem);
5529
5530 // Generate code for the slow case. We make a call to hashCode().
5531 set_control(_gvn.transform(slow_region));
5532 if (!stopped()) {
5533 // No need for PreserveJVMState, because we're using up the present state.
5534 set_all_memory(init_mem);
5535 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
5536 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
5537 Node* slow_result = set_results_for_java_call(slow_call);
5538 // this->control() comes from set_results_for_java_call
5539 result_reg->init_req(_slow_path, control());
5540 result_val->init_req(_slow_path, slow_result);
5541 result_io ->set_req(_slow_path, i_o());
5542 result_mem ->set_req(_slow_path, reset_memory());
5543 }
5544
5545 // Return the combined state.
5546 set_i_o( _gvn.transform(result_io) );
5547 set_all_memory( _gvn.transform(result_mem));
5548
5549 set_result(result_reg, result_val);
5550 return true;
5551 }
5552
5553 //---------------------------inline_native_getClass----------------------------
5554 // public final native Class<?> java.lang.Object.getClass();
5555 //
5556 // Build special case code for calls to getClass on an object.
5557 bool LibraryCallKit::inline_native_getClass() {
5558 Node* obj = argument(0);
5559 if (obj->is_InlineType()) {
5560 const Type* t = _gvn.type(obj);
5561 if (t->maybe_null()) {
5562 null_check(obj);
5563 }
5564 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror())));
5565 return true;
5566 }
5567 obj = null_check_receiver();
5568 if (stopped()) return true;
5569 set_result(load_mirror_from_klass(load_object_klass(obj)));
5570 return true;
5571 }
5572
5573 //-----------------inline_native_Reflection_getCallerClass---------------------
5574 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
5575 //
5576 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
5577 //
5578 // NOTE: This code must perform the same logic as JVM_GetCallerClass
5579 // in that it must skip particular security frames and checks for
5580 // caller sensitive methods.
5581 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
5582 #ifndef PRODUCT
5583 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5584 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
5585 }
5586 #endif
5587
5588 if (!jvms()->has_method()) {
5589 #ifndef PRODUCT
5590 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5591 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
5592 }
5593 #endif
5594 return false;
5595 }
5596
5597 // Walk back up the JVM state to find the caller at the required
5598 // depth.
5599 JVMState* caller_jvms = jvms();
5600
5601 // Cf. JVM_GetCallerClass
5602 // NOTE: Start the loop at depth 1 because the current JVM state does
5603 // not include the Reflection.getCallerClass() frame.
5604 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
5605 ciMethod* m = caller_jvms->method();
5606 switch (n) {
5607 case 0:
5608 fatal("current JVM state does not include the Reflection.getCallerClass frame");
5609 break;
5610 case 1:
5611 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
5612 if (!m->caller_sensitive()) {
5613 #ifndef PRODUCT
5614 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5615 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
5616 }
5617 #endif
5618 return false; // bail-out; let JVM_GetCallerClass do the work
5619 }
5620 break;
5621 default:
5622 if (!m->is_ignored_by_security_stack_walk()) {
5623 // We have reached the desired frame; return the holder class.
5624 // Acquire method holder as java.lang.Class and push as constant.
5625 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
5626 ciInstance* caller_mirror = caller_klass->java_mirror();
5627 set_result(makecon(TypeInstPtr::make(caller_mirror)));
5628
5629 #ifndef PRODUCT
5630 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5631 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());
5632 tty->print_cr(" JVM state at this point:");
5633 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5634 ciMethod* m = jvms()->of_depth(i)->method();
5635 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5636 }
5637 }
5638 #endif
5639 return true;
5640 }
5641 break;
5642 }
5643 }
5644
5645 #ifndef PRODUCT
5646 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5647 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
5648 tty->print_cr(" JVM state at this point:");
5649 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5650 ciMethod* m = jvms()->of_depth(i)->method();
5651 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5652 }
5653 }
5654 #endif
5655
5656 return false; // bail-out; let JVM_GetCallerClass do the work
5657 }
5658
5659 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
5660 Node* arg = argument(0);
5661 Node* result = nullptr;
5662
5663 switch (id) {
5664 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
5665 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
5666 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5667 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5668 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5669 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5670
5671 case vmIntrinsics::_doubleToLongBits: {
5672 // two paths (plus control) merge in a wood
5673 RegionNode *r = new RegionNode(3);
5674 Node *phi = new PhiNode(r, TypeLong::LONG);
5675
5676 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5677 // Build the boolean node
5678 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5679
5680 // Branch either way.
5681 // NaN case is less traveled, which makes all the difference.
5682 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5683 Node *opt_isnan = _gvn.transform(ifisnan);
5684 assert( opt_isnan->is_If(), "Expect an IfNode");
5685 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5686 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5687
5688 set_control(iftrue);
5689
5690 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5691 Node *slow_result = longcon(nan_bits); // return NaN
5692 phi->init_req(1, _gvn.transform( slow_result ));
5693 r->init_req(1, iftrue);
5694
5695 // Else fall through
5696 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5697 set_control(iffalse);
5698
5699 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5700 r->init_req(2, iffalse);
5701
5702 // Post merge
5703 set_control(_gvn.transform(r));
5704 record_for_igvn(r);
5705
5706 C->set_has_split_ifs(true); // Has chance for split-if optimization
5707 result = phi;
5708 assert(result->bottom_type()->isa_long(), "must be");
5709 break;
5710 }
5711
5712 case vmIntrinsics::_floatToIntBits: {
5713 // two paths (plus control) merge in a wood
5714 RegionNode *r = new RegionNode(3);
5715 Node *phi = new PhiNode(r, TypeInt::INT);
5716
5717 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5718 // Build the boolean node
5719 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5720
5721 // Branch either way.
5722 // NaN case is less traveled, which makes all the difference.
5723 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5724 Node *opt_isnan = _gvn.transform(ifisnan);
5725 assert( opt_isnan->is_If(), "Expect an IfNode");
5726 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5727 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5728
5729 set_control(iftrue);
5730
5731 static const jint nan_bits = 0x7fc00000;
5732 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5733 phi->init_req(1, _gvn.transform( slow_result ));
5734 r->init_req(1, iftrue);
5735
5736 // Else fall through
5737 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5738 set_control(iffalse);
5739
5740 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5741 r->init_req(2, iffalse);
5742
5743 // Post merge
5744 set_control(_gvn.transform(r));
5745 record_for_igvn(r);
5746
5747 C->set_has_split_ifs(true); // Has chance for split-if optimization
5748 result = phi;
5749 assert(result->bottom_type()->isa_int(), "must be");
5750 break;
5751 }
5752
5753 default:
5754 fatal_unexpected_iid(id);
5755 break;
5756 }
5757 set_result(_gvn.transform(result));
5758 return true;
5759 }
5760
5761 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5762 Node* arg = argument(0);
5763 Node* result = nullptr;
5764
5765 switch (id) {
5766 case vmIntrinsics::_floatIsInfinite:
5767 result = new IsInfiniteFNode(arg);
5768 break;
5769 case vmIntrinsics::_floatIsFinite:
5770 result = new IsFiniteFNode(arg);
5771 break;
5772 case vmIntrinsics::_doubleIsInfinite:
5773 result = new IsInfiniteDNode(arg);
5774 break;
5775 case vmIntrinsics::_doubleIsFinite:
5776 result = new IsFiniteDNode(arg);
5777 break;
5778 default:
5779 fatal_unexpected_iid(id);
5780 break;
5781 }
5782 set_result(_gvn.transform(result));
5783 return true;
5784 }
5785
5786 //----------------------inline_unsafe_copyMemory-------------------------
5787 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5788
5789 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5790 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5791 const Type* base_t = gvn.type(base);
5792
5793 bool in_native = (base_t == TypePtr::NULL_PTR);
5794 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5795 bool is_mixed = !in_heap && !in_native;
5796
5797 if (is_mixed) {
5798 return true; // mixed accesses can touch both on-heap and off-heap memory
5799 }
5800 if (in_heap) {
5801 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5802 if (!is_prim_array) {
5803 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5804 // there's not enough type information available to determine proper memory slice for it.
5805 return true;
5806 }
5807 }
5808 return false;
5809 }
5810
5811 bool LibraryCallKit::inline_unsafe_copyMemory() {
5812 if (callee()->is_static()) return false; // caller must have the capability!
5813 null_check_receiver(); // null-check receiver
5814 if (stopped()) return true;
5815
5816 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5817
5818 Node* src_base = argument(1); // type: oop
5819 Node* src_off = ConvL2X(argument(2)); // type: long
5820 Node* dst_base = argument(4); // type: oop
5821 Node* dst_off = ConvL2X(argument(5)); // type: long
5822 Node* size = ConvL2X(argument(7)); // type: long
5823
5824 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5825 "fieldOffset must be byte-scaled");
5826
5827 Node* src_addr = make_unsafe_address(src_base, src_off);
5828 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5829
5830 Node* thread = _gvn.transform(new ThreadLocalNode());
5831 Node* doing_unsafe_access_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5832 BasicType doing_unsafe_access_bt = T_BYTE;
5833 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5834
5835 // update volatile field
5836 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5837
5838 int flags = RC_LEAF | RC_NO_FP;
5839
5840 const TypePtr* dst_type = TypePtr::BOTTOM;
5841
5842 // Adjust memory effects of the runtime call based on input values.
5843 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5844 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5845 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5846
5847 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5848 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5849 flags |= RC_NARROW_MEM; // narrow in memory
5850 }
5851 }
5852
5853 // Call it. Note that the length argument is not scaled.
5854 make_runtime_call(flags,
5855 OptoRuntime::fast_arraycopy_Type(),
5856 StubRoutines::unsafe_arraycopy(),
5857 "unsafe_arraycopy",
5858 dst_type,
5859 src_addr, dst_addr, size XTOP);
5860
5861 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5862
5863 return true;
5864 }
5865
5866 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5867 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5868 bool LibraryCallKit::inline_unsafe_setMemory() {
5869 if (callee()->is_static()) return false; // caller must have the capability!
5870 null_check_receiver(); // null-check receiver
5871 if (stopped()) return true;
5872
5873 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5874
5875 Node* dst_base = argument(1); // type: oop
5876 Node* dst_off = ConvL2X(argument(2)); // type: long
5877 Node* size = ConvL2X(argument(4)); // type: long
5878 Node* byte = argument(6); // type: byte
5879
5880 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5881 "fieldOffset must be byte-scaled");
5882
5883 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5884
5885 Node* thread = _gvn.transform(new ThreadLocalNode());
5886 Node* doing_unsafe_access_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5887 BasicType doing_unsafe_access_bt = T_BYTE;
5888 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5889
5890 // update volatile field
5891 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5892
5893 int flags = RC_LEAF | RC_NO_FP;
5894
5895 const TypePtr* dst_type = TypePtr::BOTTOM;
5896
5897 // Adjust memory effects of the runtime call based on input values.
5898 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5899 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5900
5901 flags |= RC_NARROW_MEM; // narrow in memory
5902 }
5903
5904 // Call it. Note that the length argument is not scaled.
5905 make_runtime_call(flags,
5906 OptoRuntime::unsafe_setmemory_Type(),
5907 StubRoutines::unsafe_setmemory(),
5908 "unsafe_setmemory",
5909 dst_type,
5910 dst_addr, size XTOP, byte);
5911
5912 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5913
5914 return true;
5915 }
5916
5917 #undef XTOP
5918
5919 //------------------------clone_coping-----------------------------------
5920 // Helper function for inline_native_clone.
5921 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5922 assert(obj_size != nullptr, "");
5923 Node* raw_obj = alloc_obj->in(1);
5924 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5925
5926 AllocateNode* alloc = nullptr;
5927 if (ReduceBulkZeroing &&
5928 // If we are implementing an array clone without knowing its source type
5929 // (can happen when compiling the array-guarded branch of a reflective
5930 // Object.clone() invocation), initialize the array within the allocation.
5931 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5932 // to a runtime clone call that assumes fully initialized source arrays.
5933 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5934 // We will be completely responsible for initializing this object -
5935 // mark Initialize node as complete.
5936 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5937 // The object was just allocated - there should be no any stores!
5938 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5939 // Mark as complete_with_arraycopy so that on AllocateNode
5940 // expansion, we know this AllocateNode is initialized by an array
5941 // copy and a StoreStore barrier exists after the array copy.
5942 alloc->initialization()->set_complete_with_arraycopy();
5943 }
5944
5945 Node* size = _gvn.transform(obj_size);
5946 access_clone(obj, alloc_obj, size, is_array);
5947
5948 // Do not let reads from the cloned object float above the arraycopy.
5949 if (alloc != nullptr) {
5950 // Do not let stores that initialize this object be reordered with
5951 // a subsequent store that would make this object accessible by
5952 // other threads.
5953 // Record what AllocateNode this StoreStore protects so that
5954 // escape analysis can go from the MemBarStoreStoreNode to the
5955 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5956 // based on the escape status of the AllocateNode.
5957 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5958 } else {
5959 insert_mem_bar(Op_MemBarCPUOrder);
5960 }
5961 }
5962
5963 //------------------------inline_native_clone----------------------------
5964 // protected native Object java.lang.Object.clone();
5965 //
5966 // Here are the simple edge cases:
5967 // null receiver => normal trap
5968 // virtual and clone was overridden => slow path to out-of-line clone
5969 // not cloneable or finalizer => slow path to out-of-line Object.clone
5970 //
5971 // The general case has two steps, allocation and copying.
5972 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5973 //
5974 // Copying also has two cases, oop arrays and everything else.
5975 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5976 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5977 //
5978 // These steps fold up nicely if and when the cloned object's klass
5979 // can be sharply typed as an object array, a type array, or an instance.
5980 //
5981 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5982 PhiNode* result_val;
5983
5984 // Set the reexecute bit for the interpreter to reexecute
5985 // the bytecode that invokes Object.clone if deoptimization happens.
5986 { PreserveReexecuteState preexecs(this);
5987 jvms()->set_should_reexecute(true);
5988
5989 Node* obj = argument(0);
5990 obj = null_check_receiver();
5991 if (stopped()) return true;
5992
5993 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5994 if (obj_type->is_inlinetypeptr()) {
5995 // If the object to clone is an inline type, we can simply return it (i.e. a nop) since inline types have
5996 // no identity.
5997 set_result(obj);
5998 return true;
5999 }
6000
6001 // If we are going to clone an instance, we need its exact type to
6002 // know the number and types of fields to convert the clone to
6003 // loads/stores. Maybe a speculative type can help us.
6004 if (!obj_type->klass_is_exact() &&
6005 obj_type->speculative_type() != nullptr &&
6006 obj_type->speculative_type()->is_instance_klass() &&
6007 !obj_type->speculative_type()->is_inlinetype()) {
6008 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
6009 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
6010 !spec_ik->has_injected_fields()) {
6011 if (!obj_type->isa_instptr() ||
6012 obj_type->is_instptr()->instance_klass()->has_subklass()) {
6013 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
6014 }
6015 }
6016 }
6017
6018 // Conservatively insert a memory barrier on all memory slices.
6019 // Do not let writes into the original float below the clone.
6020 insert_mem_bar(Op_MemBarCPUOrder);
6021
6022 // paths into result_reg:
6023 enum {
6024 _slow_path = 1, // out-of-line call to clone method (virtual or not)
6025 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
6026 _array_path, // plain array allocation, plus arrayof_long_arraycopy
6027 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
6028 PATH_LIMIT
6029 };
6030 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
6031 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
6032 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
6033 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
6034 record_for_igvn(result_reg);
6035
6036 Node* obj_klass = load_object_klass(obj);
6037 // We only go to the fast case code if we pass a number of guards.
6038 // The paths which do not pass are accumulated in the slow_region.
6039 RegionNode* slow_region = new RegionNode(1);
6040 record_for_igvn(slow_region);
6041
6042 Node* array_obj = obj;
6043 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
6044 if (array_ctl != nullptr) {
6045 // It's an array.
6046 PreserveJVMState pjvms(this);
6047 set_control(array_ctl);
6048
6049 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
6050 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr();
6051 if (UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) &&
6052 obj_type->can_be_inline_array() &&
6053 (ary_ptr == nullptr || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) {
6054 // Flat inline type array may have object field that would require a
6055 // write barrier. Conservatively, go to slow path.
6056 generate_fair_guard(flat_array_test(obj_klass), slow_region);
6057 }
6058
6059 if (!stopped()) {
6060 Node* obj_length = load_array_length(array_obj);
6061 Node* array_size = nullptr; // Size of the array without object alignment padding.
6062 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
6063
6064 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
6065 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
6066 // If it is an oop array, it requires very special treatment,
6067 // because gc barriers are required when accessing the array.
6068 Node* is_obja = generate_refArray_guard(obj_klass, (RegionNode*)nullptr);
6069 if (is_obja != nullptr) {
6070 PreserveJVMState pjvms2(this);
6071 set_control(is_obja);
6072 // Generate a direct call to the right arraycopy function(s).
6073 // Clones are always tightly coupled.
6074 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
6075 ac->set_clone_oop_array();
6076 Node* n = _gvn.transform(ac);
6077 assert(n == ac, "cannot disappear");
6078 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
6079
6080 result_reg->init_req(_objArray_path, control());
6081 result_val->init_req(_objArray_path, alloc_obj);
6082 result_i_o ->set_req(_objArray_path, i_o());
6083 result_mem ->set_req(_objArray_path, reset_memory());
6084 }
6085 }
6086 // Otherwise, there are no barriers to worry about.
6087 // (We can dispense with card marks if we know the allocation
6088 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
6089 // causes the non-eden paths to take compensating steps to
6090 // simulate a fresh allocation, so that no further
6091 // card marks are required in compiled code to initialize
6092 // the object.)
6093
6094 if (!stopped()) {
6095 copy_to_clone(obj, alloc_obj, array_size, true);
6096
6097 // Present the results of the copy.
6098 result_reg->init_req(_array_path, control());
6099 result_val->init_req(_array_path, alloc_obj);
6100 result_i_o ->set_req(_array_path, i_o());
6101 result_mem ->set_req(_array_path, reset_memory());
6102 }
6103 }
6104 }
6105
6106 if (!stopped()) {
6107 // It's an instance (we did array above). Make the slow-path tests.
6108 // If this is a virtual call, we generate a funny guard. We grab
6109 // the vtable entry corresponding to clone() from the target object.
6110 // If the target method which we are calling happens to be the
6111 // Object clone() method, we pass the guard. We do not need this
6112 // guard for non-virtual calls; the caller is known to be the native
6113 // Object clone().
6114 if (is_virtual) {
6115 generate_virtual_guard(obj_klass, slow_region);
6116 }
6117
6118 // The object must be easily cloneable and must not have a finalizer.
6119 // Both of these conditions may be checked in a single test.
6120 // We could optimize the test further, but we don't care.
6121 generate_misc_flags_guard(obj_klass,
6122 // Test both conditions:
6123 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
6124 // Must be cloneable but not finalizer:
6125 KlassFlags::_misc_is_cloneable_fast,
6126 slow_region);
6127 }
6128
6129 if (!stopped()) {
6130 // It's an instance, and it passed the slow-path tests.
6131 PreserveJVMState pjvms(this);
6132 Node* obj_size = nullptr; // Total object size, including object alignment padding.
6133 // Need to deoptimize on exception from allocation since Object.clone intrinsic
6134 // is reexecuted if deoptimization occurs and there could be problems when merging
6135 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
6136 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
6137
6138 copy_to_clone(obj, alloc_obj, obj_size, false);
6139
6140 // Present the results of the slow call.
6141 result_reg->init_req(_instance_path, control());
6142 result_val->init_req(_instance_path, alloc_obj);
6143 result_i_o ->set_req(_instance_path, i_o());
6144 result_mem ->set_req(_instance_path, reset_memory());
6145 }
6146
6147 // Generate code for the slow case. We make a call to clone().
6148 set_control(_gvn.transform(slow_region));
6149 if (!stopped()) {
6150 PreserveJVMState pjvms(this);
6151 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
6152 // We need to deoptimize on exception (see comment above)
6153 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
6154 // this->control() comes from set_results_for_java_call
6155 result_reg->init_req(_slow_path, control());
6156 result_val->init_req(_slow_path, slow_result);
6157 result_i_o ->set_req(_slow_path, i_o());
6158 result_mem ->set_req(_slow_path, reset_memory());
6159 }
6160
6161 // Return the combined state.
6162 set_control( _gvn.transform(result_reg));
6163 set_i_o( _gvn.transform(result_i_o));
6164 set_all_memory( _gvn.transform(result_mem));
6165 } // original reexecute is set back here
6166
6167 set_result(_gvn.transform(result_val));
6168 return true;
6169 }
6170
6171 // If we have a tightly coupled allocation, the arraycopy may take care
6172 // of the array initialization. If one of the guards we insert between
6173 // the allocation and the arraycopy causes a deoptimization, an
6174 // uninitialized array will escape the compiled method. To prevent that
6175 // we set the JVM state for uncommon traps between the allocation and
6176 // the arraycopy to the state before the allocation so, in case of
6177 // deoptimization, we'll reexecute the allocation and the
6178 // initialization.
6179 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
6180 if (alloc != nullptr) {
6181 ciMethod* trap_method = alloc->jvms()->method();
6182 int trap_bci = alloc->jvms()->bci();
6183
6184 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6185 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
6186 // Make sure there's no store between the allocation and the
6187 // arraycopy otherwise visible side effects could be rexecuted
6188 // in case of deoptimization and cause incorrect execution.
6189 bool no_interfering_store = true;
6190 Node* mem = alloc->in(TypeFunc::Memory);
6191 if (mem->is_MergeMem()) {
6192 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
6193 Node* n = mms.memory();
6194 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
6195 assert(n->is_Store(), "what else?");
6196 no_interfering_store = false;
6197 break;
6198 }
6199 }
6200 } else {
6201 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
6202 Node* n = mms.memory();
6203 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
6204 assert(n->is_Store(), "what else?");
6205 no_interfering_store = false;
6206 break;
6207 }
6208 }
6209 }
6210
6211 if (no_interfering_store) {
6212 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6213
6214 JVMState* saved_jvms = jvms();
6215 saved_reexecute_sp = _reexecute_sp;
6216
6217 set_jvms(sfpt->jvms());
6218 _reexecute_sp = jvms()->sp();
6219
6220 return saved_jvms;
6221 }
6222 }
6223 }
6224 return nullptr;
6225 }
6226
6227 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
6228 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
6229 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
6230 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
6231 uint size = alloc->req();
6232 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
6233 old_jvms->set_map(sfpt);
6234 for (uint i = 0; i < size; i++) {
6235 sfpt->init_req(i, alloc->in(i));
6236 }
6237 int adjustment = 1;
6238 const TypeAryKlassPtr* ary_klass_ptr = alloc->in(AllocateNode::KlassNode)->bottom_type()->is_aryklassptr();
6239 if (ary_klass_ptr->is_null_free()) {
6240 // A null-free, tightly coupled array allocation can only come from LibraryCallKit::inline_newArray which
6241 // also requires the componentType and initVal on stack for re-execution.
6242 // Re-create and push the componentType.
6243 ciArrayKlass* klass = ary_klass_ptr->exact_klass()->as_array_klass();
6244 ciInstance* instance = klass->component_mirror_instance();
6245 const TypeInstPtr* t_instance = TypeInstPtr::make(instance);
6246 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), makecon(t_instance));
6247 adjustment++;
6248 }
6249 // re-push array length for deoptimization
6250 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment - 1, alloc->in(AllocateNode::ALength));
6251 if (ary_klass_ptr->is_null_free()) {
6252 // Re-create and push the initVal.
6253 Node* init_val = alloc->in(AllocateNode::InitValue);
6254 if (init_val == nullptr) {
6255 init_val = InlineTypeNode::make_all_zero(_gvn, ary_klass_ptr->elem()->is_instklassptr()->instance_klass()->as_inline_klass());
6256 } else if (UseCompressedOops) {
6257 init_val = _gvn.transform(new DecodeNNode(init_val, init_val->bottom_type()->make_ptr()));
6258 }
6259 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment, init_val);
6260 adjustment++;
6261 }
6262 old_jvms->set_sp(old_jvms->sp() + adjustment);
6263 old_jvms->set_monoff(old_jvms->monoff() + adjustment);
6264 old_jvms->set_scloff(old_jvms->scloff() + adjustment);
6265 old_jvms->set_endoff(old_jvms->endoff() + adjustment);
6266 old_jvms->set_should_reexecute(true);
6267
6268 sfpt->set_i_o(map()->i_o());
6269 sfpt->set_memory(map()->memory());
6270 sfpt->set_control(map()->control());
6271 return sfpt;
6272 }
6273
6274 // In case of a deoptimization, we restart execution at the
6275 // allocation, allocating a new array. We would leave an uninitialized
6276 // array in the heap that GCs wouldn't expect. Move the allocation
6277 // after the traps so we don't allocate the array if we
6278 // deoptimize. This is possible because tightly_coupled_allocation()
6279 // guarantees there's no observer of the allocated array at this point
6280 // and the control flow is simple enough.
6281 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
6282 int saved_reexecute_sp, uint new_idx) {
6283 if (saved_jvms_before_guards != nullptr && !stopped()) {
6284 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
6285
6286 assert(alloc != nullptr, "only with a tightly coupled allocation");
6287 // restore JVM state to the state at the arraycopy
6288 saved_jvms_before_guards->map()->set_control(map()->control());
6289 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
6290 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
6291 // If we've improved the types of some nodes (null check) while
6292 // emitting the guards, propagate them to the current state
6293 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
6294 set_jvms(saved_jvms_before_guards);
6295 _reexecute_sp = saved_reexecute_sp;
6296
6297 // Remove the allocation from above the guards
6298 CallProjections* callprojs = alloc->extract_projections(true);
6299 InitializeNode* init = alloc->initialization();
6300 Node* alloc_mem = alloc->in(TypeFunc::Memory);
6301 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
6302 init->replace_mem_projs_by(alloc_mem, C);
6303
6304 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
6305 // the allocation (i.e. is only valid if the allocation succeeds):
6306 // 1) replace CastIINode with AllocateArrayNode's length here
6307 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
6308 //
6309 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
6310 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
6311 Node* init_control = init->proj_out(TypeFunc::Control);
6312 Node* alloc_length = alloc->Ideal_length();
6313 #ifdef ASSERT
6314 Node* prev_cast = nullptr;
6315 #endif
6316 for (uint i = 0; i < init_control->outcnt(); i++) {
6317 Node* init_out = init_control->raw_out(i);
6318 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
6319 #ifdef ASSERT
6320 if (prev_cast == nullptr) {
6321 prev_cast = init_out;
6322 } else {
6323 if (prev_cast->cmp(*init_out) == false) {
6324 prev_cast->dump();
6325 init_out->dump();
6326 assert(false, "not equal CastIINode");
6327 }
6328 }
6329 #endif
6330 C->gvn_replace_by(init_out, alloc_length);
6331 }
6332 }
6333 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
6334
6335 // move the allocation here (after the guards)
6336 _gvn.hash_delete(alloc);
6337 alloc->set_req(TypeFunc::Control, control());
6338 alloc->set_req(TypeFunc::I_O, i_o());
6339 Node *mem = reset_memory();
6340 set_all_memory(mem);
6341 alloc->set_req(TypeFunc::Memory, mem);
6342 set_control(init->proj_out_or_null(TypeFunc::Control));
6343 set_i_o(callprojs->fallthrough_ioproj);
6344
6345 // Update memory as done in GraphKit::set_output_for_allocation()
6346 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
6347 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
6348 if (ary_type->isa_aryptr() && length_type != nullptr) {
6349 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
6350 }
6351 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
6352 int elemidx = C->get_alias_index(telemref);
6353 // Need to properly move every memory projection for the Initialize
6354 #ifdef ASSERT
6355 int mark_idx = C->get_alias_index(ary_type->add_offset(oopDesc::mark_offset_in_bytes()));
6356 int klass_idx = C->get_alias_index(ary_type->add_offset(oopDesc::klass_offset_in_bytes()));
6357 #endif
6358 auto move_proj = [&](ProjNode* proj) {
6359 int alias_idx = C->get_alias_index(proj->adr_type());
6360 assert(alias_idx == Compile::AliasIdxRaw ||
6361 alias_idx == elemidx ||
6362 alias_idx == mark_idx ||
6363 alias_idx == klass_idx, "should be raw memory or array element type");
6364 set_memory(proj, alias_idx);
6365 };
6366 init->for_each_proj(move_proj, TypeFunc::Memory);
6367
6368 Node* allocx = _gvn.transform(alloc);
6369 assert(allocx == alloc, "where has the allocation gone?");
6370 assert(dest->is_CheckCastPP(), "not an allocation result?");
6371
6372 _gvn.hash_delete(dest);
6373 dest->set_req(0, control());
6374 Node* destx = _gvn.transform(dest);
6375 assert(destx == dest, "where has the allocation result gone?");
6376
6377 array_ideal_length(alloc, ary_type, true);
6378 }
6379 }
6380
6381 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
6382 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
6383 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
6384 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
6385 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
6386 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
6387 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
6388 JVMState* saved_jvms_before_guards) {
6389 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
6390 // There is at least one unrelated uncommon trap which needs to be replaced.
6391 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6392
6393 JVMState* saved_jvms = jvms();
6394 const int saved_reexecute_sp = _reexecute_sp;
6395 set_jvms(sfpt->jvms());
6396 _reexecute_sp = jvms()->sp();
6397
6398 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
6399
6400 // Restore state
6401 set_jvms(saved_jvms);
6402 _reexecute_sp = saved_reexecute_sp;
6403 }
6404 }
6405
6406 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
6407 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
6408 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
6409 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
6410 while (if_proj->is_IfProj()) {
6411 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
6412 if (uncommon_trap != nullptr) {
6413 create_new_uncommon_trap(uncommon_trap);
6414 }
6415 assert(if_proj->in(0)->is_If(), "must be If");
6416 if_proj = if_proj->in(0)->in(0);
6417 }
6418 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
6419 "must have reached control projection of init node");
6420 }
6421
6422 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
6423 const int trap_request = uncommon_trap_call->uncommon_trap_request();
6424 assert(trap_request != 0, "no valid UCT trap request");
6425 PreserveJVMState pjvms(this);
6426 set_control(uncommon_trap_call->in(0));
6427 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
6428 Deoptimization::trap_request_action(trap_request));
6429 assert(stopped(), "Should be stopped");
6430 _gvn.hash_delete(uncommon_trap_call);
6431 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
6432 }
6433
6434 // Common checks for array sorting intrinsics arguments.
6435 // Returns `true` if checks passed.
6436 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
6437 // check address of the class
6438 if (elementType == nullptr || elementType->is_top()) {
6439 return false; // dead path
6440 }
6441 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
6442 if (elem_klass == nullptr) {
6443 return false; // dead path
6444 }
6445 // java_mirror_type() returns non-null for compile-time Class constants only
6446 ciType* elem_type = elem_klass->java_mirror_type();
6447 if (elem_type == nullptr) {
6448 return false;
6449 }
6450 bt = elem_type->basic_type();
6451 // Disable the intrinsic if the CPU does not support SIMD sort
6452 if (!Matcher::supports_simd_sort(bt)) {
6453 return false;
6454 }
6455 // check address of the array
6456 if (obj == nullptr || obj->is_top()) {
6457 return false; // dead path
6458 }
6459 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
6460 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
6461 return false; // failed input validation
6462 }
6463 return true;
6464 }
6465
6466 //------------------------------inline_array_partition-----------------------
6467 bool LibraryCallKit::inline_array_partition() {
6468 address stubAddr = StubRoutines::select_array_partition_function();
6469 if (stubAddr == nullptr) {
6470 return false; // Intrinsic's stub is not implemented on this platform
6471 }
6472 assert(callee()->signature()->size() == 9, "arrayPartition has 8 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 Node* indexPivot1 = argument(6);
6481 Node* indexPivot2 = argument(7);
6482 // PartitionOperation: argument(8) is ignored
6483
6484 Node* pivotIndices = nullptr;
6485 BasicType bt = T_ILLEGAL;
6486
6487 if (!check_array_sort_arguments(elementType, obj, bt)) {
6488 return false;
6489 }
6490 null_check(obj);
6491 // If obj is dead, only null-path is taken.
6492 if (stopped()) {
6493 return true;
6494 }
6495 // Set the original stack and the reexecute bit for the interpreter to reexecute
6496 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
6497 { PreserveReexecuteState preexecs(this);
6498 jvms()->set_should_reexecute(true);
6499
6500 Node* obj_adr = make_unsafe_address(obj, offset);
6501
6502 // create the pivotIndices array of type int and size = 2
6503 Node* size = intcon(2);
6504 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
6505 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
6506 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
6507 guarantee(alloc != nullptr, "created above");
6508 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
6509
6510 // pass the basic type enum to the stub
6511 Node* elemType = intcon(bt);
6512
6513 // Call the stub
6514 const char *stubName = "array_partition_stub";
6515 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
6516 stubAddr, stubName, TypePtr::BOTTOM,
6517 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
6518 indexPivot1, indexPivot2);
6519
6520 } // original reexecute is set back here
6521
6522 if (!stopped()) {
6523 set_result(pivotIndices);
6524 }
6525
6526 return true;
6527 }
6528
6529
6530 //------------------------------inline_array_sort-----------------------
6531 bool LibraryCallKit::inline_array_sort() {
6532 address stubAddr = StubRoutines::select_arraysort_function();
6533 if (stubAddr == nullptr) {
6534 return false; // Intrinsic's stub is not implemented on this platform
6535 }
6536 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
6537
6538 // no receiver because it is a static method
6539 Node* elementType = argument(0);
6540 Node* obj = argument(1);
6541 Node* offset = argument(2); // long
6542 Node* fromIndex = argument(4);
6543 Node* toIndex = argument(5);
6544 // SortOperation: argument(6) is ignored
6545
6546 BasicType bt = T_ILLEGAL;
6547
6548 if (!check_array_sort_arguments(elementType, obj, bt)) {
6549 return false;
6550 }
6551 null_check(obj);
6552 // If obj is dead, only null-path is taken.
6553 if (stopped()) {
6554 return true;
6555 }
6556 Node* obj_adr = make_unsafe_address(obj, offset);
6557
6558 // pass the basic type enum to the stub
6559 Node* elemType = intcon(bt);
6560
6561 // Call the stub.
6562 const char *stubName = "arraysort_stub";
6563 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
6564 stubAddr, stubName, TypePtr::BOTTOM,
6565 obj_adr, elemType, fromIndex, toIndex);
6566
6567 return true;
6568 }
6569
6570
6571 //------------------------------inline_arraycopy-----------------------
6572 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
6573 // Object dest, int destPos,
6574 // int length);
6575 bool LibraryCallKit::inline_arraycopy() {
6576 // Get the arguments.
6577 Node* src = argument(0); // type: oop
6578 Node* src_offset = argument(1); // type: int
6579 Node* dest = argument(2); // type: oop
6580 Node* dest_offset = argument(3); // type: int
6581 Node* length = argument(4); // type: int
6582
6583 uint new_idx = C->unique();
6584
6585 // Check for allocation before we add nodes that would confuse
6586 // tightly_coupled_allocation()
6587 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
6588
6589 int saved_reexecute_sp = -1;
6590 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
6591 // See arraycopy_restore_alloc_state() comment
6592 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
6593 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
6594 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
6595 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
6596
6597 // The following tests must be performed
6598 // (1) src and dest are arrays.
6599 // (2) src and dest arrays must have elements of the same BasicType
6600 // (3) src and dest must not be null.
6601 // (4) src_offset must not be negative.
6602 // (5) dest_offset must not be negative.
6603 // (6) length must not be negative.
6604 // (7) src_offset + length must not exceed length of src.
6605 // (8) dest_offset + length must not exceed length of dest.
6606 // (9) each element of an oop array must be assignable
6607
6608 // (3) src and dest must not be null.
6609 // always do this here because we need the JVM state for uncommon traps
6610 Node* null_ctl = top();
6611 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
6612 assert(null_ctl->is_top(), "no null control here");
6613 dest = null_check(dest, T_ARRAY);
6614
6615 if (!can_emit_guards) {
6616 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
6617 // guards but the arraycopy node could still take advantage of a
6618 // tightly allocated allocation. tightly_coupled_allocation() is
6619 // called again to make sure it takes the null check above into
6620 // account: the null check is mandatory and if it caused an
6621 // uncommon trap to be emitted then the allocation can't be
6622 // considered tightly coupled in this context.
6623 alloc = tightly_coupled_allocation(dest);
6624 }
6625
6626 bool validated = false;
6627
6628 const Type* src_type = _gvn.type(src);
6629 const Type* dest_type = _gvn.type(dest);
6630 const TypeAryPtr* top_src = src_type->isa_aryptr();
6631 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6632
6633 // Do we have the type of src?
6634 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6635 // Do we have the type of dest?
6636 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6637 // Is the type for src from speculation?
6638 bool src_spec = false;
6639 // Is the type for dest from speculation?
6640 bool dest_spec = false;
6641
6642 if ((!has_src || !has_dest) && can_emit_guards) {
6643 // We don't have sufficient type information, let's see if
6644 // speculative types can help. We need to have types for both src
6645 // and dest so that it pays off.
6646
6647 // Do we already have or could we have type information for src
6648 bool could_have_src = has_src;
6649 // Do we already have or could we have type information for dest
6650 bool could_have_dest = has_dest;
6651
6652 ciKlass* src_k = nullptr;
6653 if (!has_src) {
6654 src_k = src_type->speculative_type_not_null();
6655 if (src_k != nullptr && src_k->is_array_klass()) {
6656 could_have_src = true;
6657 }
6658 }
6659
6660 ciKlass* dest_k = nullptr;
6661 if (!has_dest) {
6662 dest_k = dest_type->speculative_type_not_null();
6663 if (dest_k != nullptr && dest_k->is_array_klass()) {
6664 could_have_dest = true;
6665 }
6666 }
6667
6668 if (could_have_src && could_have_dest) {
6669 // This is going to pay off so emit the required guards
6670 if (!has_src) {
6671 src = maybe_cast_profiled_obj(src, src_k, true);
6672 src_type = _gvn.type(src);
6673 top_src = src_type->isa_aryptr();
6674 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6675 src_spec = true;
6676 }
6677 if (!has_dest) {
6678 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6679 dest_type = _gvn.type(dest);
6680 top_dest = dest_type->isa_aryptr();
6681 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6682 dest_spec = true;
6683 }
6684 }
6685 }
6686
6687 if (has_src && has_dest && can_emit_guards) {
6688 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
6689 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
6690 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
6691 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
6692
6693 if (src_elem == dest_elem && top_src->is_flat() == top_dest->is_flat() && src_elem == T_OBJECT) {
6694 // If both arrays are object arrays then having the exact types
6695 // for both will remove the need for a subtype check at runtime
6696 // before the call and may make it possible to pick a faster copy
6697 // routine (without a subtype check on every element)
6698 // Do we have the exact type of src?
6699 bool could_have_src = src_spec;
6700 // Do we have the exact type of dest?
6701 bool could_have_dest = dest_spec;
6702 ciKlass* src_k = nullptr;
6703 ciKlass* dest_k = nullptr;
6704 if (!src_spec) {
6705 src_k = src_type->speculative_type_not_null();
6706 if (src_k != nullptr && src_k->is_array_klass()) {
6707 could_have_src = true;
6708 }
6709 }
6710 if (!dest_spec) {
6711 dest_k = dest_type->speculative_type_not_null();
6712 if (dest_k != nullptr && dest_k->is_array_klass()) {
6713 could_have_dest = true;
6714 }
6715 }
6716 if (could_have_src && could_have_dest) {
6717 // If we can have both exact types, emit the missing guards
6718 if (could_have_src && !src_spec) {
6719 src = maybe_cast_profiled_obj(src, src_k, true);
6720 src_type = _gvn.type(src);
6721 top_src = src_type->isa_aryptr();
6722 }
6723 if (could_have_dest && !dest_spec) {
6724 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6725 dest_type = _gvn.type(dest);
6726 top_dest = dest_type->isa_aryptr();
6727 }
6728 }
6729 }
6730 }
6731
6732 ciMethod* trap_method = method();
6733 int trap_bci = bci();
6734 if (saved_jvms_before_guards != nullptr) {
6735 trap_method = alloc->jvms()->method();
6736 trap_bci = alloc->jvms()->bci();
6737 }
6738
6739 bool negative_length_guard_generated = false;
6740
6741 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6742 can_emit_guards && !src->is_top() && !dest->is_top()) {
6743 // validate arguments: enables transformation the ArrayCopyNode
6744 validated = true;
6745
6746 RegionNode* slow_region = new RegionNode(1);
6747 record_for_igvn(slow_region);
6748
6749 // (1) src and dest are arrays.
6750 generate_non_array_guard(load_object_klass(src), slow_region, &src);
6751 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
6752
6753 // (2) src and dest arrays must have elements of the same BasicType
6754 // done at macro expansion or at Ideal transformation time
6755
6756 // (4) src_offset must not be negative.
6757 generate_negative_guard(src_offset, slow_region);
6758
6759 // (5) dest_offset must not be negative.
6760 generate_negative_guard(dest_offset, slow_region);
6761
6762 // (7) src_offset + length must not exceed length of src.
6763 generate_limit_guard(src_offset, length,
6764 load_array_length(src),
6765 slow_region);
6766
6767 // (8) dest_offset + length must not exceed length of dest.
6768 generate_limit_guard(dest_offset, length,
6769 load_array_length(dest),
6770 slow_region);
6771
6772 // (6) length must not be negative.
6773 // This is also checked in generate_arraycopy() during macro expansion, but
6774 // we also have to check it here for the case where the ArrayCopyNode will
6775 // be eliminated by Escape Analysis.
6776 if (EliminateAllocations) {
6777 generate_negative_guard(length, slow_region);
6778 negative_length_guard_generated = true;
6779 }
6780
6781 // (9) each element of an oop array must be assignable
6782 Node* dest_klass = load_object_klass(dest);
6783 Node* refined_dest_klass = dest_klass;
6784 if (src != dest) {
6785 dest_klass = load_non_refined_array_klass(refined_dest_klass);
6786 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6787 slow_region->add_req(not_subtype_ctrl);
6788 }
6789
6790 // TODO 8251971 Improve this. What about atomicity? Make sure this is always folded for type arrays.
6791 // If destination is null-restricted, source must be null-restricted as well: src_null_restricted || !dst_null_restricted
6792 Node* src_klass = load_object_klass(src);
6793 Node* adr_prop_src = basic_plus_adr(top(), src_klass, in_bytes(ArrayKlass::properties_offset()));
6794 Node* prop_src = _gvn.transform(LoadNode::make(_gvn, control(), immutable_memory(), adr_prop_src,
6795 _gvn.type(adr_prop_src)->is_ptr(), TypeInt::INT, T_INT,
6796 MemNode::unordered));
6797 Node* adr_prop_dest = basic_plus_adr(top(), refined_dest_klass, in_bytes(ArrayKlass::properties_offset()));
6798 Node* prop_dest = _gvn.transform(LoadNode::make(_gvn, control(), immutable_memory(), adr_prop_dest,
6799 _gvn.type(adr_prop_dest)->is_ptr(), TypeInt::INT, T_INT,
6800 MemNode::unordered));
6801
6802 const ArrayProperties props_null_restricted = ArrayProperties::Default().with_null_restricted();
6803 jint props_value = (jint)props_null_restricted.value();
6804
6805 prop_dest = _gvn.transform(new XorINode(prop_dest, intcon(props_value)));
6806 prop_src = _gvn.transform(new OrINode(prop_dest, prop_src));
6807 prop_src = _gvn.transform(new AndINode(prop_src, intcon(props_value)));
6808
6809 Node* chk = _gvn.transform(new CmpINode(prop_src, intcon(props_value)));
6810 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::ne));
6811 generate_fair_guard(tst, slow_region);
6812
6813 // TODO 8251971 This is too strong
6814 generate_fair_guard(flat_array_test(src), slow_region);
6815 generate_fair_guard(flat_array_test(dest), slow_region);
6816
6817 {
6818 PreserveJVMState pjvms(this);
6819 set_control(_gvn.transform(slow_region));
6820 uncommon_trap(Deoptimization::Reason_intrinsic,
6821 Deoptimization::Action_make_not_entrant);
6822 assert(stopped(), "Should be stopped");
6823 }
6824
6825 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->isa_klassptr();
6826 if (dest_klass_t == nullptr) {
6827 // refined_dest_klass may not be an array, which leads to dest_klass being top. This means we
6828 // are in a dead path.
6829 uncommon_trap(Deoptimization::Reason_intrinsic,
6830 Deoptimization::Action_make_not_entrant);
6831 return true;
6832 }
6833
6834 const Type* toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6835 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6836 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6837 }
6838
6839 if (stopped()) {
6840 return true;
6841 }
6842
6843 Node* dest_klass = load_object_klass(dest);
6844 dest_klass = load_non_refined_array_klass(dest_klass);
6845
6846 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6847 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6848 // so the compiler has a chance to eliminate them: during macro expansion,
6849 // we have to set their control (CastPP nodes are eliminated).
6850 load_object_klass(src), dest_klass,
6851 load_array_length(src), load_array_length(dest));
6852
6853 ac->set_arraycopy(validated);
6854
6855 Node* n = _gvn.transform(ac);
6856 if (n == ac) {
6857 ac->connect_outputs(this);
6858 } else {
6859 assert(validated, "shouldn't transform if all arguments not validated");
6860 set_all_memory(n);
6861 }
6862 clear_upper_avx();
6863
6864
6865 return true;
6866 }
6867
6868
6869 // Helper function which determines if an arraycopy immediately follows
6870 // an allocation, with no intervening tests or other escapes for the object.
6871 AllocateArrayNode*
6872 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6873 if (stopped()) return nullptr; // no fast path
6874 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6875
6876 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6877 if (alloc == nullptr) return nullptr;
6878
6879 Node* rawmem = memory(Compile::AliasIdxRaw);
6880 // Is the allocation's memory state untouched?
6881 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6882 // Bail out if there have been raw-memory effects since the allocation.
6883 // (Example: There might have been a call or safepoint.)
6884 return nullptr;
6885 }
6886 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6887 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6888 return nullptr;
6889 }
6890
6891 // There must be no unexpected observers of this allocation.
6892 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6893 Node* obs = ptr->fast_out(i);
6894 if (obs != this->map()) {
6895 return nullptr;
6896 }
6897 }
6898
6899 // This arraycopy must unconditionally follow the allocation of the ptr.
6900 Node* alloc_ctl = ptr->in(0);
6901 Node* ctl = control();
6902 while (ctl != alloc_ctl) {
6903 // There may be guards which feed into the slow_region.
6904 // Any other control flow means that we might not get a chance
6905 // to finish initializing the allocated object.
6906 // Various low-level checks bottom out in uncommon traps. These
6907 // are considered safe since we've already checked above that
6908 // there is no unexpected observer of this allocation.
6909 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6910 assert(ctl->in(0)->is_If(), "must be If");
6911 ctl = ctl->in(0)->in(0);
6912 } else {
6913 return nullptr;
6914 }
6915 }
6916
6917 // If we get this far, we have an allocation which immediately
6918 // precedes the arraycopy, and we can take over zeroing the new object.
6919 // The arraycopy will finish the initialization, and provide
6920 // a new control state to which we will anchor the destination pointer.
6921
6922 return alloc;
6923 }
6924
6925 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6926 if (node->is_IfProj()) {
6927 IfProjNode* other_proj = node->as_IfProj()->other_if_proj();
6928 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6929 Node* obs = other_proj->fast_out(j);
6930 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6931 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6932 return obs->as_CallStaticJava();
6933 }
6934 }
6935 }
6936 return nullptr;
6937 }
6938
6939 //-------------inline_encodeISOArray-----------------------------------
6940 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6941 // int java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6942 // int java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6943 // encode char[] to byte[] in ISO_8859_1 or ASCII
6944 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6945 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6946 // no receiver since it is static method
6947 Node *src = argument(0);
6948 Node *src_offset = argument(1);
6949 Node *dst = argument(2);
6950 Node *dst_offset = argument(3);
6951 Node *length = argument(4);
6952
6953 // Cast source & target arrays to not-null
6954 src = must_be_not_null(src, true);
6955 dst = must_be_not_null(dst, true);
6956 if (stopped()) {
6957 return true;
6958 }
6959
6960 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6961 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6962 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6963 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6964 // failed array check
6965 return false;
6966 }
6967
6968 // Figure out the size and type of the elements we will be copying.
6969 BasicType src_elem = src_type->elem()->array_element_basic_type();
6970 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6971 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6972 return false;
6973 }
6974
6975 // Check source & target bounds
6976 RegionNode* bailout = create_bailout();
6977 generate_string_range_check(src, src_offset, length, src_elem == T_BYTE, bailout);
6978 generate_string_range_check(dst, dst_offset, length, false, bailout);
6979 if (check_bailout(bailout)) {
6980 return true;
6981 }
6982
6983 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6984 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6985 // 'src_start' points to src array + scaled offset
6986 // 'dst_start' points to dst array + scaled offset
6987
6988 // See GraphKit::compress_string
6989 const TypePtr* adr_type;
6990 Node* mem = capture_memory(adr_type, src_type, dst_type);
6991 Node* enc = new EncodeISOArrayNode(control(), mem, adr_type, src_start, dst_start, length, ascii);
6992 enc = _gvn.transform(enc);
6993 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6994 memory_effect(res_mem, src_type, dst_type);
6995
6996 set_result(enc);
6997 clear_upper_avx();
6998
6999 return true;
7000 }
7001
7002 //-------------inline_multiplyToLen-----------------------------------
7003 bool LibraryCallKit::inline_multiplyToLen() {
7004 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
7005
7006 address stubAddr = StubRoutines::multiplyToLen();
7007 if (stubAddr == nullptr) {
7008 return false; // Intrinsic's stub is not implemented on this platform
7009 }
7010 const char* stubName = "multiplyToLen";
7011
7012 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
7013
7014 // no receiver because it is a static method
7015 Node* x = argument(0);
7016 Node* xlen = argument(1);
7017 Node* y = argument(2);
7018 Node* ylen = argument(3);
7019 Node* z = argument(4);
7020
7021 x = must_be_not_null(x, true);
7022 y = must_be_not_null(y, true);
7023
7024 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
7025 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
7026 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
7027 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
7028 // failed array check
7029 return false;
7030 }
7031
7032 BasicType x_elem = x_type->elem()->array_element_basic_type();
7033 BasicType y_elem = y_type->elem()->array_element_basic_type();
7034 if (x_elem != T_INT || y_elem != T_INT) {
7035 return false;
7036 }
7037
7038 Node* x_start = array_element_address(x, intcon(0), x_elem);
7039 Node* y_start = array_element_address(y, intcon(0), y_elem);
7040 // 'x_start' points to x array + scaled xlen
7041 // 'y_start' points to y array + scaled ylen
7042
7043 Node* z_start = array_element_address(z, intcon(0), T_INT);
7044
7045 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
7046 OptoRuntime::multiplyToLen_Type(),
7047 stubAddr, stubName, TypePtr::BOTTOM,
7048 x_start, xlen, y_start, ylen, z_start);
7049
7050 C->set_has_split_ifs(true); // Has chance for split-if optimization
7051 set_result(z);
7052 return true;
7053 }
7054
7055 //-------------inline_squareToLen------------------------------------
7056 bool LibraryCallKit::inline_squareToLen() {
7057 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
7058
7059 address stubAddr = StubRoutines::squareToLen();
7060 if (stubAddr == nullptr) {
7061 return false; // Intrinsic's stub is not implemented on this platform
7062 }
7063 const char* stubName = "squareToLen";
7064
7065 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
7066
7067 Node* x = argument(0);
7068 Node* len = argument(1);
7069 Node* z = argument(2);
7070 Node* zlen = argument(3);
7071
7072 x = must_be_not_null(x, true);
7073 z = must_be_not_null(z, true);
7074
7075 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
7076 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
7077 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
7078 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
7079 // failed array check
7080 return false;
7081 }
7082
7083 BasicType x_elem = x_type->elem()->array_element_basic_type();
7084 BasicType z_elem = z_type->elem()->array_element_basic_type();
7085 if (x_elem != T_INT || z_elem != T_INT) {
7086 return false;
7087 }
7088
7089
7090 Node* x_start = array_element_address(x, intcon(0), x_elem);
7091 Node* z_start = array_element_address(z, intcon(0), z_elem);
7092
7093 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
7094 OptoRuntime::squareToLen_Type(),
7095 stubAddr, stubName, TypePtr::BOTTOM,
7096 x_start, len, z_start, zlen);
7097
7098 set_result(z);
7099 return true;
7100 }
7101
7102 //-------------inline_mulAdd------------------------------------------
7103 bool LibraryCallKit::inline_mulAdd() {
7104 assert(UseMulAddIntrinsic, "not implemented on this platform");
7105
7106 address stubAddr = StubRoutines::mulAdd();
7107 if (stubAddr == nullptr) {
7108 return false; // Intrinsic's stub is not implemented on this platform
7109 }
7110 const char* stubName = "mulAdd";
7111
7112 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
7113
7114 Node* out = argument(0);
7115 Node* in = argument(1);
7116 Node* offset = argument(2);
7117 Node* len = argument(3);
7118 Node* k = argument(4);
7119
7120 in = must_be_not_null(in, true);
7121 out = must_be_not_null(out, true);
7122
7123 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
7124 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
7125 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
7126 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
7127 // failed array check
7128 return false;
7129 }
7130
7131 BasicType out_elem = out_type->elem()->array_element_basic_type();
7132 BasicType in_elem = in_type->elem()->array_element_basic_type();
7133 if (out_elem != T_INT || in_elem != T_INT) {
7134 return false;
7135 }
7136
7137 Node* outlen = load_array_length(out);
7138 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
7139 Node* out_start = array_element_address(out, intcon(0), out_elem);
7140 Node* in_start = array_element_address(in, intcon(0), in_elem);
7141
7142 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
7143 OptoRuntime::mulAdd_Type(),
7144 stubAddr, stubName, TypePtr::BOTTOM,
7145 out_start,in_start, new_offset, len, k);
7146 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7147 set_result(result);
7148 return true;
7149 }
7150
7151 //-------------inline_montgomeryMultiply-----------------------------------
7152 bool LibraryCallKit::inline_montgomeryMultiply() {
7153 address stubAddr = StubRoutines::montgomeryMultiply();
7154 if (stubAddr == nullptr) {
7155 return false; // Intrinsic's stub is not implemented on this platform
7156 }
7157
7158 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
7159 const char* stubName = "montgomery_multiply";
7160
7161 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
7162
7163 Node* a = argument(0);
7164 Node* b = argument(1);
7165 Node* n = argument(2);
7166 Node* len = argument(3);
7167 Node* inv = argument(4);
7168 Node* m = argument(6);
7169
7170 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
7171 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
7172 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
7173 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
7174 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
7175 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
7176 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
7177 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
7178 // failed array check
7179 return false;
7180 }
7181
7182 BasicType a_elem = a_type->elem()->array_element_basic_type();
7183 BasicType b_elem = b_type->elem()->array_element_basic_type();
7184 BasicType n_elem = n_type->elem()->array_element_basic_type();
7185 BasicType m_elem = m_type->elem()->array_element_basic_type();
7186 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
7187 return false;
7188 }
7189
7190 // Make the call
7191 {
7192 Node* a_start = array_element_address(a, intcon(0), a_elem);
7193 Node* b_start = array_element_address(b, intcon(0), b_elem);
7194 Node* n_start = array_element_address(n, intcon(0), n_elem);
7195 Node* m_start = array_element_address(m, intcon(0), m_elem);
7196
7197 Node* call = make_runtime_call(RC_LEAF,
7198 OptoRuntime::montgomeryMultiply_Type(),
7199 stubAddr, stubName, TypePtr::BOTTOM,
7200 a_start, b_start, n_start, len, inv, top(),
7201 m_start);
7202 set_result(m);
7203 }
7204
7205 return true;
7206 }
7207
7208 bool LibraryCallKit::inline_montgomerySquare() {
7209 address stubAddr = StubRoutines::montgomerySquare();
7210 if (stubAddr == nullptr) {
7211 return false; // Intrinsic's stub is not implemented on this platform
7212 }
7213
7214 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
7215 const char* stubName = "montgomery_square";
7216
7217 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
7218
7219 Node* a = argument(0);
7220 Node* n = argument(1);
7221 Node* len = argument(2);
7222 Node* inv = argument(3);
7223 Node* m = argument(5);
7224
7225 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
7226 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
7227 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
7228 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
7229 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
7230 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
7231 // failed array check
7232 return false;
7233 }
7234
7235 BasicType a_elem = a_type->elem()->array_element_basic_type();
7236 BasicType n_elem = n_type->elem()->array_element_basic_type();
7237 BasicType m_elem = m_type->elem()->array_element_basic_type();
7238 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
7239 return false;
7240 }
7241
7242 // Make the call
7243 {
7244 Node* a_start = array_element_address(a, intcon(0), a_elem);
7245 Node* n_start = array_element_address(n, intcon(0), n_elem);
7246 Node* m_start = array_element_address(m, intcon(0), m_elem);
7247
7248 Node* call = make_runtime_call(RC_LEAF,
7249 OptoRuntime::montgomerySquare_Type(),
7250 stubAddr, stubName, TypePtr::BOTTOM,
7251 a_start, n_start, len, inv, top(),
7252 m_start);
7253 set_result(m);
7254 }
7255
7256 return true;
7257 }
7258
7259 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
7260 address stubAddr = nullptr;
7261 const char* stubName = nullptr;
7262
7263 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
7264 if (stubAddr == nullptr) {
7265 return false; // Intrinsic's stub is not implemented on this platform
7266 }
7267
7268 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
7269
7270 assert(callee()->signature()->size() == 5, "expected 5 arguments");
7271
7272 Node* newArr = argument(0);
7273 Node* oldArr = argument(1);
7274 Node* newIdx = argument(2);
7275 Node* shiftCount = argument(3);
7276 Node* numIter = argument(4);
7277
7278 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
7279 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
7280 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
7281 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
7282 return false;
7283 }
7284
7285 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
7286 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
7287 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
7288 return false;
7289 }
7290
7291 // Make the call
7292 {
7293 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
7294 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
7295
7296 Node* call = make_runtime_call(RC_LEAF,
7297 OptoRuntime::bigIntegerShift_Type(),
7298 stubAddr,
7299 stubName,
7300 TypePtr::BOTTOM,
7301 newArr_start,
7302 oldArr_start,
7303 newIdx,
7304 shiftCount,
7305 numIter);
7306 }
7307
7308 return true;
7309 }
7310
7311 //-------------inline_vectorizedMismatch------------------------------
7312 bool LibraryCallKit::inline_vectorizedMismatch() {
7313 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
7314
7315 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
7316 Node* obja = argument(0); // Object
7317 Node* aoffset = argument(1); // long
7318 Node* objb = argument(3); // Object
7319 Node* boffset = argument(4); // long
7320 Node* length = argument(6); // int
7321 Node* scale = argument(7); // int
7322
7323 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
7324 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
7325 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
7326 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
7327 scale == top()) {
7328 return false; // failed input validation
7329 }
7330
7331 Node* obja_adr = make_unsafe_address(obja, aoffset);
7332 Node* objb_adr = make_unsafe_address(objb, boffset);
7333
7334 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
7335 //
7336 // inline_limit = ArrayOperationPartialInlineSize / element_size;
7337 // if (length <= inline_limit) {
7338 // inline_path:
7339 // vmask = VectorMaskGen length
7340 // vload1 = LoadVectorMasked obja, vmask
7341 // vload2 = LoadVectorMasked objb, vmask
7342 // result1 = VectorCmpMasked vload1, vload2, vmask
7343 // } else {
7344 // call_stub_path:
7345 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
7346 // }
7347 // exit_block:
7348 // return Phi(result1, result2);
7349 //
7350 enum { inline_path = 1, // input is small enough to process it all at once
7351 stub_path = 2, // input is too large; call into the VM
7352 PATH_LIMIT = 3
7353 };
7354
7355 Node* exit_block = new RegionNode(PATH_LIMIT);
7356 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
7357 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
7358
7359 Node* call_stub_path = control();
7360
7361 BasicType elem_bt = T_ILLEGAL;
7362
7363 const TypeInt* scale_t = _gvn.type(scale)->is_int();
7364 if (scale_t->is_con()) {
7365 switch (scale_t->get_con()) {
7366 case 0: elem_bt = T_BYTE; break;
7367 case 1: elem_bt = T_SHORT; break;
7368 case 2: elem_bt = T_INT; break;
7369 case 3: elem_bt = T_LONG; break;
7370
7371 default: elem_bt = T_ILLEGAL; break; // not supported
7372 }
7373 }
7374
7375 int inline_limit = 0;
7376 bool do_partial_inline = false;
7377
7378 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
7379 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
7380 do_partial_inline = inline_limit >= 16;
7381 }
7382
7383 if (do_partial_inline) {
7384 assert(elem_bt != T_ILLEGAL, "sanity");
7385
7386 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
7387 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
7388 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
7389
7390 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
7391 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
7392 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
7393
7394 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
7395
7396 if (!stopped()) {
7397 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
7398
7399 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
7400 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
7401 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
7402 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
7403
7404 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
7405 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
7406 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
7407 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
7408
7409 exit_block->init_req(inline_path, control());
7410 memory_phi->init_req(inline_path, map()->memory());
7411 result_phi->init_req(inline_path, result);
7412
7413 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
7414 clear_upper_avx();
7415 }
7416 }
7417 }
7418
7419 if (call_stub_path != nullptr) {
7420 set_control(call_stub_path);
7421
7422 Node* call = make_runtime_call(RC_LEAF,
7423 OptoRuntime::vectorizedMismatch_Type(),
7424 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
7425 obja_adr, objb_adr, length, scale);
7426
7427 exit_block->init_req(stub_path, control());
7428 memory_phi->init_req(stub_path, map()->memory());
7429 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
7430 }
7431
7432 exit_block = _gvn.transform(exit_block);
7433 memory_phi = _gvn.transform(memory_phi);
7434 result_phi = _gvn.transform(result_phi);
7435
7436 record_for_igvn(exit_block);
7437 record_for_igvn(memory_phi);
7438 record_for_igvn(result_phi);
7439
7440 set_control(exit_block);
7441 set_all_memory(memory_phi);
7442 set_result(result_phi);
7443
7444 return true;
7445 }
7446
7447 //------------------------------inline_vectorizedHashcode----------------------------
7448 bool LibraryCallKit::inline_vectorizedHashCode() {
7449 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
7450
7451 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
7452 Node* array = argument(0);
7453 Node* offset = argument(1);
7454 Node* length = argument(2);
7455 Node* initialValue = argument(3);
7456 Node* basic_type = argument(4);
7457
7458 if (basic_type == top()) {
7459 return false; // failed input validation
7460 }
7461
7462 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
7463 if (!basic_type_t->is_con()) {
7464 return false; // Only intrinsify if mode argument is constant
7465 }
7466
7467 array = must_be_not_null(array, true);
7468
7469 BasicType bt = (BasicType)basic_type_t->get_con();
7470
7471 // Resolve address of first element
7472 Node* array_start = array_element_address(array, offset, bt);
7473
7474 const TypeAryPtr* in_adr_type = TypeAryPtr::get_array_body_type(bt);
7475 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(in_adr_type), in_adr_type,
7476 array_start, length, initialValue, basic_type)));
7477 clear_upper_avx();
7478
7479 return true;
7480 }
7481
7482 /**
7483 * Calculate CRC32 for byte.
7484 * int java.util.zip.CRC32.update(int crc, int b)
7485 */
7486 bool LibraryCallKit::inline_updateCRC32() {
7487 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7488 assert(callee()->signature()->size() == 2, "update has 2 parameters");
7489 // no receiver since it is static method
7490 Node* crc = argument(0); // type: int
7491 Node* b = argument(1); // type: int
7492
7493 /*
7494 * int c = ~ crc;
7495 * b = timesXtoThe32[(b ^ c) & 0xFF];
7496 * b = b ^ (c >>> 8);
7497 * crc = ~b;
7498 */
7499
7500 Node* M1 = intcon(-1);
7501 crc = _gvn.transform(new XorINode(crc, M1));
7502 Node* result = _gvn.transform(new XorINode(crc, b));
7503 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
7504
7505 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
7506 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
7507 Node* adr = off_heap_plus_addr(base, ConvI2X(offset));
7508 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
7509
7510 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
7511 result = _gvn.transform(new XorINode(crc, result));
7512 result = _gvn.transform(new XorINode(result, M1));
7513 set_result(result);
7514 return true;
7515 }
7516
7517 /**
7518 * Calculate CRC32 for byte[] array.
7519 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
7520 */
7521 bool LibraryCallKit::inline_updateBytesCRC32() {
7522 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7523 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7524 // no receiver since it is static method
7525 Node* crc = argument(0); // type: int
7526 Node* src = argument(1); // type: oop
7527 Node* offset = argument(2); // type: int
7528 Node* length = argument(3); // type: int
7529
7530 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7531 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7532 // failed array check
7533 return false;
7534 }
7535
7536 // Figure out the size and type of the elements we will be copying.
7537 BasicType src_elem = src_type->elem()->array_element_basic_type();
7538 if (src_elem != T_BYTE) {
7539 return false;
7540 }
7541
7542 // 'src_start' points to src array + scaled offset
7543 src = must_be_not_null(src, true);
7544 Node* src_start = array_element_address(src, offset, src_elem);
7545
7546 // We assume that range check is done by caller.
7547 // TODO: generate range check (offset+length < src.length) in debug VM.
7548
7549 // Call the stub.
7550 address stubAddr = StubRoutines::updateBytesCRC32();
7551 const char *stubName = "updateBytesCRC32";
7552
7553 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7554 stubAddr, stubName, TypePtr::BOTTOM,
7555 crc, src_start, length);
7556 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7557 set_result(result);
7558 return true;
7559 }
7560
7561 /**
7562 * Calculate CRC32 for ByteBuffer.
7563 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
7564 */
7565 bool LibraryCallKit::inline_updateByteBufferCRC32() {
7566 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7567 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7568 // no receiver since it is static method
7569 Node* crc = argument(0); // type: int
7570 Node* src = argument(1); // type: long
7571 Node* offset = argument(3); // type: int
7572 Node* length = argument(4); // type: int
7573
7574 src = ConvL2X(src); // adjust Java long to machine word
7575 Node* base = _gvn.transform(new CastX2PNode(src));
7576 offset = ConvI2X(offset);
7577
7578 // 'src_start' points to src array + scaled offset
7579 Node* src_start = off_heap_plus_addr(base, offset);
7580
7581 // Call the stub.
7582 address stubAddr = StubRoutines::updateBytesCRC32();
7583 const char *stubName = "updateBytesCRC32";
7584
7585 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7586 stubAddr, stubName, TypePtr::BOTTOM,
7587 crc, src_start, length);
7588 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7589 set_result(result);
7590 return true;
7591 }
7592
7593 //------------------------------get_table_from_crc32c_class-----------------------
7594 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
7595 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
7596 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
7597
7598 return table;
7599 }
7600
7601 //------------------------------inline_updateBytesCRC32C-----------------------
7602 //
7603 // Calculate CRC32C for byte[] array.
7604 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
7605 //
7606 bool LibraryCallKit::inline_updateBytesCRC32C() {
7607 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7608 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7609 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7610 // no receiver since it is a static method
7611 Node* crc = argument(0); // type: int
7612 Node* src = argument(1); // type: oop
7613 Node* offset = argument(2); // type: int
7614 Node* end = argument(3); // type: int
7615
7616 Node* length = _gvn.transform(new SubINode(end, offset));
7617
7618 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7619 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7620 // failed array check
7621 return false;
7622 }
7623
7624 // Figure out the size and type of the elements we will be copying.
7625 BasicType src_elem = src_type->elem()->array_element_basic_type();
7626 if (src_elem != T_BYTE) {
7627 return false;
7628 }
7629
7630 // 'src_start' points to src array + scaled offset
7631 src = must_be_not_null(src, true);
7632 Node* src_start = array_element_address(src, offset, src_elem);
7633
7634 // static final int[] byteTable in class CRC32C
7635 Node* table = get_table_from_crc32c_class(callee()->holder());
7636 table = must_be_not_null(table, true);
7637 Node* table_start = array_element_address(table, intcon(0), T_INT);
7638
7639 // We assume that range check is done by caller.
7640 // TODO: generate range check (offset+length < src.length) in debug VM.
7641
7642 // Call the stub.
7643 address stubAddr = StubRoutines::updateBytesCRC32C();
7644 const char *stubName = "updateBytesCRC32C";
7645
7646 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7647 stubAddr, stubName, TypePtr::BOTTOM,
7648 crc, src_start, length, table_start);
7649 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7650 set_result(result);
7651 return true;
7652 }
7653
7654 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
7655 //
7656 // Calculate CRC32C for DirectByteBuffer.
7657 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
7658 //
7659 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
7660 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7661 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
7662 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7663 // no receiver since it is a static method
7664 Node* crc = argument(0); // type: int
7665 Node* src = argument(1); // type: long
7666 Node* offset = argument(3); // type: int
7667 Node* end = argument(4); // type: int
7668
7669 Node* length = _gvn.transform(new SubINode(end, offset));
7670
7671 src = ConvL2X(src); // adjust Java long to machine word
7672 Node* base = _gvn.transform(new CastX2PNode(src));
7673 offset = ConvI2X(offset);
7674
7675 // 'src_start' points to src array + scaled offset
7676 Node* src_start = off_heap_plus_addr(base, offset);
7677
7678 // static final int[] byteTable in class CRC32C
7679 Node* table = get_table_from_crc32c_class(callee()->holder());
7680 table = must_be_not_null(table, true);
7681 Node* table_start = array_element_address(table, intcon(0), T_INT);
7682
7683 // Call the stub.
7684 address stubAddr = StubRoutines::updateBytesCRC32C();
7685 const char *stubName = "updateBytesCRC32C";
7686
7687 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7688 stubAddr, stubName, TypePtr::BOTTOM,
7689 crc, src_start, length, table_start);
7690 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7691 set_result(result);
7692 return true;
7693 }
7694
7695 //------------------------------inline_updateBytesAdler32----------------------
7696 //
7697 // Calculate Adler32 checksum for byte[] array.
7698 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
7699 //
7700 bool LibraryCallKit::inline_updateBytesAdler32() {
7701 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7702 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7703 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7704 // no receiver since it is static method
7705 Node* crc = argument(0); // type: int
7706 Node* src = argument(1); // type: oop
7707 Node* offset = argument(2); // type: int
7708 Node* length = argument(3); // type: int
7709
7710 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7711 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7712 // failed array check
7713 return false;
7714 }
7715
7716 // Figure out the size and type of the elements we will be copying.
7717 BasicType src_elem = src_type->elem()->array_element_basic_type();
7718 if (src_elem != T_BYTE) {
7719 return false;
7720 }
7721
7722 // 'src_start' points to src array + scaled offset
7723 Node* src_start = array_element_address(src, offset, src_elem);
7724
7725 // We assume that range check is done by caller.
7726 // TODO: generate range check (offset+length < src.length) in debug VM.
7727
7728 // Call the stub.
7729 address stubAddr = StubRoutines::updateBytesAdler32();
7730 const char *stubName = "updateBytesAdler32";
7731
7732 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7733 stubAddr, stubName, TypePtr::BOTTOM,
7734 crc, src_start, length);
7735 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7736 set_result(result);
7737 return true;
7738 }
7739
7740 //------------------------------inline_updateByteBufferAdler32---------------
7741 //
7742 // Calculate Adler32 checksum for DirectByteBuffer.
7743 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
7744 //
7745 bool LibraryCallKit::inline_updateByteBufferAdler32() {
7746 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7747 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7748 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7749 // no receiver since it is static method
7750 Node* crc = argument(0); // type: int
7751 Node* src = argument(1); // type: long
7752 Node* offset = argument(3); // type: int
7753 Node* length = argument(4); // type: int
7754
7755 src = ConvL2X(src); // adjust Java long to machine word
7756 Node* base = _gvn.transform(new CastX2PNode(src));
7757 offset = ConvI2X(offset);
7758
7759 // 'src_start' points to src array + scaled offset
7760 Node* src_start = off_heap_plus_addr(base, offset);
7761
7762 // Call the stub.
7763 address stubAddr = StubRoutines::updateBytesAdler32();
7764 const char *stubName = "updateBytesAdler32";
7765
7766 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7767 stubAddr, stubName, TypePtr::BOTTOM,
7768 crc, src_start, length);
7769
7770 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7771 set_result(result);
7772 return true;
7773 }
7774
7775 //----------------------------inline_reference_get0----------------------------
7776 // public T java.lang.ref.Reference.get();
7777 bool LibraryCallKit::inline_reference_get0() {
7778 const int referent_offset = java_lang_ref_Reference::referent_offset();
7779
7780 // Get the argument:
7781 Node* reference_obj = null_check_receiver();
7782 if (stopped()) return true;
7783
7784 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7785 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7786 decorators, /*is_static*/ false,
7787 env()->Reference_klass());
7788 if (result == nullptr) return false;
7789
7790 // Add memory barrier to prevent commoning reads from this field
7791 // across safepoint since GC can change its value.
7792 insert_mem_bar(Op_MemBarCPUOrder);
7793
7794 set_result(result);
7795 return true;
7796 }
7797
7798 //----------------------------inline_reference_refersTo0----------------------------
7799 // bool java.lang.ref.Reference.refersTo0();
7800 // bool java.lang.ref.PhantomReference.refersTo0();
7801 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7802 // Get arguments:
7803 Node* reference_obj = null_check_receiver();
7804 Node* other_obj = argument(1);
7805 if (stopped()) return true;
7806
7807 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7808 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7809 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7810 decorators, /*is_static*/ false,
7811 env()->Reference_klass());
7812 if (referent == nullptr) return false;
7813
7814 // Add memory barrier to prevent commoning reads from this field
7815 // across safepoint since GC can change its value.
7816 insert_mem_bar(Op_MemBarCPUOrder);
7817
7818 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7819 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7820 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7821
7822 RegionNode* region = new RegionNode(3);
7823 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7824
7825 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7826 region->init_req(1, if_true);
7827 phi->init_req(1, intcon(1));
7828
7829 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7830 region->init_req(2, if_false);
7831 phi->init_req(2, intcon(0));
7832
7833 set_control(_gvn.transform(region));
7834 record_for_igvn(region);
7835 set_result(_gvn.transform(phi));
7836 return true;
7837 }
7838
7839 //----------------------------inline_reference_clear0----------------------------
7840 // void java.lang.ref.Reference.clear0();
7841 // void java.lang.ref.PhantomReference.clear0();
7842 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7843 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7844
7845 // Get arguments
7846 Node* reference_obj = null_check_receiver();
7847 if (stopped()) return true;
7848
7849 // Common access parameters
7850 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7851 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7852 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7853 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7854 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7855
7856 Node* referent = access_load_at(reference_obj,
7857 referent_field_addr,
7858 referent_field_addr_type,
7859 val_type,
7860 T_OBJECT,
7861 decorators);
7862
7863 IdealKit ideal(this);
7864 #define __ ideal.
7865 __ if_then(referent, BoolTest::ne, null());
7866 sync_kit(ideal);
7867 access_store_at(reference_obj,
7868 referent_field_addr,
7869 referent_field_addr_type,
7870 null(),
7871 val_type,
7872 T_OBJECT,
7873 decorators);
7874 __ sync_kit(this);
7875 __ end_if();
7876 final_sync(ideal);
7877 #undef __
7878
7879 return true;
7880 }
7881
7882 //-----------------------inline_reference_reachabilityFence-----------------
7883 // bool java.lang.ref.Reference.reachabilityFence();
7884 bool LibraryCallKit::inline_reference_reachabilityFence() {
7885 Node* referent = argument(0);
7886 insert_reachability_fence(referent);
7887 return true;
7888 }
7889
7890 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7891 DecoratorSet decorators, bool is_static,
7892 ciInstanceKlass* fromKls) {
7893 if (fromKls == nullptr) {
7894 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7895 assert(tinst != nullptr, "obj is null");
7896 assert(tinst->is_loaded(), "obj is not loaded");
7897 fromKls = tinst->instance_klass();
7898 }
7899 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7900 ciSymbol::make(fieldTypeString),
7901 is_static);
7902
7903 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7904 if (field == nullptr) return (Node *) nullptr;
7905
7906 if (is_static) {
7907 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7908 fromObj = makecon(tip);
7909 }
7910
7911 // Next code copied from Parse::do_get_xxx():
7912
7913 // Compute address and memory type.
7914 int offset = field->offset_in_bytes();
7915 bool is_vol = field->is_volatile();
7916 ciType* field_klass = field->type();
7917 assert(field_klass->is_loaded(), "should be loaded");
7918 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7919 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7920 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7921 "slice of address and input slice don't match");
7922 BasicType bt = field->layout_type();
7923
7924 // Build the resultant type of the load
7925 const Type *type;
7926 if (bt == T_OBJECT) {
7927 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7928 } else {
7929 type = Type::get_const_basic_type(bt);
7930 }
7931
7932 if (is_vol) {
7933 decorators |= MO_SEQ_CST;
7934 }
7935
7936 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7937 }
7938
7939 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7940 bool is_exact /* true */, bool is_static /* false */,
7941 ciInstanceKlass * fromKls /* nullptr */) {
7942 if (fromKls == nullptr) {
7943 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7944 assert(tinst != nullptr, "obj is null");
7945 assert(tinst->is_loaded(), "obj is not loaded");
7946 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7947 fromKls = tinst->instance_klass();
7948 }
7949 else {
7950 assert(is_static, "only for static field access");
7951 }
7952 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7953 ciSymbol::make(fieldTypeString),
7954 is_static);
7955
7956 assert(field != nullptr, "undefined field");
7957 assert(!field->is_volatile(), "not defined for volatile fields");
7958
7959 if (is_static) {
7960 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7961 fromObj = makecon(tip);
7962 }
7963
7964 // Next code copied from Parse::do_get_xxx():
7965
7966 // Compute address and memory type.
7967 int offset = field->offset_in_bytes();
7968 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7969
7970 return adr;
7971 }
7972
7973 //------------------------------inline_aescrypt_Block-----------------------
7974 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7975 address stubAddr = nullptr;
7976 const char *stubName;
7977 bool is_decrypt = false;
7978 assert(UseAES, "need AES instruction support");
7979
7980 switch(id) {
7981 case vmIntrinsics::_aescrypt_encryptBlock:
7982 stubAddr = StubRoutines::aescrypt_encryptBlock();
7983 stubName = "aescrypt_encryptBlock";
7984 break;
7985 case vmIntrinsics::_aescrypt_decryptBlock:
7986 stubAddr = StubRoutines::aescrypt_decryptBlock();
7987 stubName = "aescrypt_decryptBlock";
7988 is_decrypt = true;
7989 break;
7990 default:
7991 break;
7992 }
7993 if (stubAddr == nullptr) return false;
7994
7995 Node* aescrypt_object = argument(0);
7996 Node* src = argument(1);
7997 Node* src_offset = argument(2);
7998 Node* dest = argument(3);
7999 Node* dest_offset = argument(4);
8000
8001 src = must_be_not_null(src, true);
8002 dest = must_be_not_null(dest, true);
8003
8004 // (1) src and dest are arrays.
8005 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8006 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8007 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8008 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8009
8010 // for the quick and dirty code we will skip all the checks.
8011 // we are just trying to get the call to be generated.
8012 Node* src_start = src;
8013 Node* dest_start = dest;
8014 if (src_offset != nullptr || dest_offset != nullptr) {
8015 assert(src_offset != nullptr && dest_offset != nullptr, "");
8016 src_start = array_element_address(src, src_offset, T_BYTE);
8017 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8018 }
8019
8020 // now need to get the start of its expanded key array
8021 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8022 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
8023 if (k_start == nullptr) return false;
8024
8025 // Call the stub.
8026 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
8027 stubAddr, stubName, TypePtr::BOTTOM,
8028 src_start, dest_start, k_start);
8029
8030 return true;
8031 }
8032
8033 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
8034 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
8035 address stubAddr = nullptr;
8036 const char *stubName = nullptr;
8037 bool is_decrypt = false;
8038 assert(UseAES, "need AES instruction support");
8039
8040 switch(id) {
8041 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
8042 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
8043 stubName = "cipherBlockChaining_encryptAESCrypt";
8044 break;
8045 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
8046 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
8047 stubName = "cipherBlockChaining_decryptAESCrypt";
8048 is_decrypt = true;
8049 break;
8050 default:
8051 break;
8052 }
8053 if (stubAddr == nullptr) return false;
8054
8055 Node* cipherBlockChaining_object = argument(0);
8056 Node* src = argument(1);
8057 Node* src_offset = argument(2);
8058 Node* len = argument(3);
8059 Node* dest = argument(4);
8060 Node* dest_offset = argument(5);
8061
8062 src = must_be_not_null(src, false);
8063 dest = must_be_not_null(dest, false);
8064
8065 // (1) src and dest are arrays.
8066 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8067 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8068 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8069 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8070
8071 // checks are the responsibility of the caller
8072 Node* src_start = src;
8073 Node* dest_start = dest;
8074 if (src_offset != nullptr || dest_offset != nullptr) {
8075 assert(src_offset != nullptr && dest_offset != nullptr, "");
8076 src_start = array_element_address(src, src_offset, T_BYTE);
8077 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8078 }
8079
8080 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8081 // (because of the predicated logic executed earlier).
8082 // so we cast it here safely.
8083 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8084
8085 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8086 if (embeddedCipherObj == nullptr) return false;
8087
8088 // cast it to what we know it will be at runtime
8089 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
8090 assert(tinst != nullptr, "CBC obj is null");
8091 assert(tinst->is_loaded(), "CBC obj is not loaded");
8092 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8093 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8094
8095 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8096 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8097 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8098 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8099 aescrypt_object = _gvn.transform(aescrypt_object);
8100
8101 // we need to get the start of the aescrypt_object's expanded key array
8102 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
8103 if (k_start == nullptr) return false;
8104
8105 // similarly, get the start address of the r vector
8106 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
8107 if (objRvec == nullptr) return false;
8108 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
8109
8110 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8111 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8112 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
8113 stubAddr, stubName, TypePtr::BOTTOM,
8114 src_start, dest_start, k_start, r_start, len);
8115
8116 // return cipher length (int)
8117 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
8118 set_result(retvalue);
8119 return true;
8120 }
8121
8122 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
8123 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
8124 address stubAddr = nullptr;
8125 const char *stubName = nullptr;
8126 bool is_decrypt = false;
8127 assert(UseAES, "need AES instruction support");
8128
8129 switch (id) {
8130 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
8131 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
8132 stubName = "electronicCodeBook_encryptAESCrypt";
8133 break;
8134 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
8135 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
8136 stubName = "electronicCodeBook_decryptAESCrypt";
8137 is_decrypt = true;
8138 break;
8139 default:
8140 break;
8141 }
8142
8143 if (stubAddr == nullptr) return false;
8144
8145 Node* electronicCodeBook_object = argument(0);
8146 Node* src = argument(1);
8147 Node* src_offset = argument(2);
8148 Node* len = argument(3);
8149 Node* dest = argument(4);
8150 Node* dest_offset = argument(5);
8151
8152 // (1) src and dest are arrays.
8153 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8154 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8155 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8156 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8157
8158 // checks are the responsibility of the caller
8159 Node* src_start = src;
8160 Node* dest_start = dest;
8161 if (src_offset != nullptr || dest_offset != nullptr) {
8162 assert(src_offset != nullptr && dest_offset != nullptr, "");
8163 src_start = array_element_address(src, src_offset, T_BYTE);
8164 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8165 }
8166
8167 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8168 // (because of the predicated logic executed earlier).
8169 // so we cast it here safely.
8170 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8171
8172 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8173 if (embeddedCipherObj == nullptr) return false;
8174
8175 // cast it to what we know it will be at runtime
8176 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
8177 assert(tinst != nullptr, "ECB obj is null");
8178 assert(tinst->is_loaded(), "ECB obj is not loaded");
8179 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8180 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8181
8182 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8183 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8184 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8185 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8186 aescrypt_object = _gvn.transform(aescrypt_object);
8187
8188 // we need to get the start of the aescrypt_object's expanded key array
8189 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
8190 if (k_start == nullptr) return false;
8191
8192 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8193 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
8194 OptoRuntime::electronicCodeBook_aescrypt_Type(),
8195 stubAddr, stubName, TypePtr::BOTTOM,
8196 src_start, dest_start, k_start, len);
8197
8198 // return cipher length (int)
8199 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
8200 set_result(retvalue);
8201 return true;
8202 }
8203
8204 //------------------------------inline_counterMode_AESCrypt-----------------------
8205 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
8206 assert(UseAES, "need AES instruction support");
8207 if (!UseAESCTRIntrinsics) return false;
8208
8209 address stubAddr = nullptr;
8210 const char *stubName = nullptr;
8211 if (id == vmIntrinsics::_counterMode_AESCrypt) {
8212 stubAddr = StubRoutines::counterMode_AESCrypt();
8213 stubName = "counterMode_AESCrypt";
8214 }
8215 if (stubAddr == nullptr) return false;
8216
8217 Node* counterMode_object = argument(0);
8218 Node* src = argument(1);
8219 Node* src_offset = argument(2);
8220 Node* len = argument(3);
8221 Node* dest = argument(4);
8222 Node* dest_offset = argument(5);
8223
8224 // (1) src and dest are arrays.
8225 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8226 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8227 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8228 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8229
8230 // checks are the responsibility of the caller
8231 Node* src_start = src;
8232 Node* dest_start = dest;
8233 if (src_offset != nullptr || dest_offset != nullptr) {
8234 assert(src_offset != nullptr && dest_offset != nullptr, "");
8235 src_start = array_element_address(src, src_offset, T_BYTE);
8236 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8237 }
8238
8239 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8240 // (because of the predicated logic executed earlier).
8241 // so we cast it here safely.
8242 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8243 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8244 if (embeddedCipherObj == nullptr) return false;
8245 // cast it to what we know it will be at runtime
8246 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
8247 assert(tinst != nullptr, "CTR obj is null");
8248 assert(tinst->is_loaded(), "CTR obj is not loaded");
8249 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8250 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8251 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8252 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8253 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8254 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8255 aescrypt_object = _gvn.transform(aescrypt_object);
8256 // we need to get the start of the aescrypt_object's expanded key array
8257 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
8258 if (k_start == nullptr) return false;
8259 // similarly, get the start address of the r vector
8260 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
8261 if (obj_counter == nullptr) return false;
8262 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
8263
8264 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
8265 if (saved_encCounter == nullptr) return false;
8266 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
8267 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
8268
8269 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8270 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8271 OptoRuntime::counterMode_aescrypt_Type(),
8272 stubAddr, stubName, TypePtr::BOTTOM,
8273 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
8274
8275 // return cipher length (int)
8276 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
8277 set_result(retvalue);
8278 return true;
8279 }
8280
8281 //------------------------------get_key_start_from_aescrypt_object-----------------------
8282 Node* LibraryCallKit::get_key_start_from_aescrypt_object(Node* aescrypt_object, bool is_decrypt) {
8283 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
8284 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
8285 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
8286 // The following platform specific stubs of encryption and decryption use the same round keys.
8287 #if defined(PPC64) || defined(S390) || defined(RISCV64)
8288 bool use_decryption_key = false;
8289 #else
8290 bool use_decryption_key = is_decrypt;
8291 #endif
8292 Node* objAESCryptKey = load_field_from_object(aescrypt_object, use_decryption_key ? "sessionKd" : "sessionKe", "[I");
8293 assert(objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AES_Crypt");
8294 if (objAESCryptKey == nullptr) return (Node *) nullptr;
8295
8296 // now have the array, need to get the start address of the selected key array
8297 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
8298 return k_start;
8299 }
8300
8301 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
8302 // Return node representing slow path of predicate check.
8303 // the pseudo code we want to emulate with this predicate is:
8304 // for encryption:
8305 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8306 // for decryption:
8307 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8308 // note cipher==plain is more conservative than the original java code but that's OK
8309 //
8310 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
8311 // The receiver was checked for null already.
8312 Node* objCBC = argument(0);
8313
8314 Node* src = argument(1);
8315 Node* dest = argument(4);
8316
8317 // Load embeddedCipher field of CipherBlockChaining object.
8318 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8319
8320 // get AESCrypt klass for instanceOf check
8321 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8322 // will have same classloader as CipherBlockChaining object
8323 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
8324 assert(tinst != nullptr, "CBCobj is null");
8325 assert(tinst->is_loaded(), "CBCobj is not loaded");
8326
8327 // we want to do an instanceof comparison against the AESCrypt class
8328 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8329 if (!klass_AESCrypt->is_loaded()) {
8330 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8331 Node* ctrl = control();
8332 set_control(top()); // no regular fast path
8333 return ctrl;
8334 }
8335
8336 src = must_be_not_null(src, true);
8337 dest = must_be_not_null(dest, true);
8338
8339 // Resolve oops to stable for CmpP below.
8340 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8341
8342 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8343 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8344 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8345
8346 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8347
8348 // for encryption, we are done
8349 if (!decrypting)
8350 return instof_false; // even if it is null
8351
8352 // for decryption, we need to add a further check to avoid
8353 // taking the intrinsic path when cipher and plain are the same
8354 // see the original java code for why.
8355 RegionNode* region = new RegionNode(3);
8356 region->init_req(1, instof_false);
8357
8358 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8359 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8360 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8361 region->init_req(2, src_dest_conjoint);
8362
8363 record_for_igvn(region);
8364 return _gvn.transform(region);
8365 }
8366
8367 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
8368 // Return node representing slow path of predicate check.
8369 // the pseudo code we want to emulate with this predicate is:
8370 // for encryption:
8371 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8372 // for decryption:
8373 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8374 // note cipher==plain is more conservative than the original java code but that's OK
8375 //
8376 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
8377 // The receiver was checked for null already.
8378 Node* objECB = argument(0);
8379
8380 // Load embeddedCipher field of ElectronicCodeBook object.
8381 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8382
8383 // get AESCrypt klass for instanceOf check
8384 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8385 // will have same classloader as ElectronicCodeBook object
8386 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
8387 assert(tinst != nullptr, "ECBobj is null");
8388 assert(tinst->is_loaded(), "ECBobj is not loaded");
8389
8390 // we want to do an instanceof comparison against the AESCrypt class
8391 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8392 if (!klass_AESCrypt->is_loaded()) {
8393 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8394 Node* ctrl = control();
8395 set_control(top()); // no regular fast path
8396 return ctrl;
8397 }
8398 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8399
8400 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8401 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8402 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8403
8404 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8405
8406 // for encryption, we are done
8407 if (!decrypting)
8408 return instof_false; // even if it is null
8409
8410 // for decryption, we need to add a further check to avoid
8411 // taking the intrinsic path when cipher and plain are the same
8412 // see the original java code for why.
8413 RegionNode* region = new RegionNode(3);
8414 region->init_req(1, instof_false);
8415 Node* src = argument(1);
8416 Node* dest = argument(4);
8417 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8418 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8419 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8420 region->init_req(2, src_dest_conjoint);
8421
8422 record_for_igvn(region);
8423 return _gvn.transform(region);
8424 }
8425
8426 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
8427 // Return node representing slow path of predicate check.
8428 // the pseudo code we want to emulate with this predicate is:
8429 // for encryption:
8430 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8431 // for decryption:
8432 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8433 // note cipher==plain is more conservative than the original java code but that's OK
8434 //
8435
8436 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
8437 // The receiver was checked for null already.
8438 Node* objCTR = argument(0);
8439
8440 // Load embeddedCipher field of CipherBlockChaining object.
8441 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8442
8443 // get AESCrypt klass for instanceOf check
8444 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8445 // will have same classloader as CipherBlockChaining object
8446 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
8447 assert(tinst != nullptr, "CTRobj is null");
8448 assert(tinst->is_loaded(), "CTRobj is not loaded");
8449
8450 // we want to do an instanceof comparison against the AESCrypt class
8451 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8452 if (!klass_AESCrypt->is_loaded()) {
8453 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8454 Node* ctrl = control();
8455 set_control(top()); // no regular fast path
8456 return ctrl;
8457 }
8458
8459 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8460 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8461 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8462 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8463 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8464
8465 return instof_false; // even if it is null
8466 }
8467
8468 //------------------------------inline_ghash_processBlocks
8469 bool LibraryCallKit::inline_ghash_processBlocks() {
8470 address stubAddr;
8471 const char *stubName;
8472 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
8473
8474 stubAddr = StubRoutines::ghash_processBlocks();
8475 stubName = "ghash_processBlocks";
8476
8477 Node* data = argument(0);
8478 Node* offset = argument(1);
8479 Node* len = argument(2);
8480 Node* state = argument(3);
8481 Node* subkeyH = argument(4);
8482
8483 state = must_be_not_null(state, true);
8484 subkeyH = must_be_not_null(subkeyH, true);
8485 data = must_be_not_null(data, true);
8486
8487 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8488 assert(state_start, "state is null");
8489 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
8490 assert(subkeyH_start, "subkeyH is null");
8491 Node* data_start = array_element_address(data, offset, T_BYTE);
8492 assert(data_start, "data is null");
8493
8494 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
8495 OptoRuntime::ghash_processBlocks_Type(),
8496 stubAddr, stubName, TypePtr::BOTTOM,
8497 state_start, subkeyH_start, data_start, len);
8498 return true;
8499 }
8500
8501 //------------------------------inline_chacha20Block
8502 bool LibraryCallKit::inline_chacha20Block() {
8503 address stubAddr;
8504 const char *stubName;
8505 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
8506
8507 stubAddr = StubRoutines::chacha20Block();
8508 stubName = "chacha20Block";
8509
8510 Node* state = argument(0);
8511 Node* result = argument(1);
8512
8513 state = must_be_not_null(state, true);
8514 result = must_be_not_null(result, true);
8515
8516 Node* state_start = array_element_address(state, intcon(0), T_INT);
8517 assert(state_start, "state is null");
8518 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
8519 assert(result_start, "result is null");
8520
8521 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
8522 OptoRuntime::chacha20Block_Type(),
8523 stubAddr, stubName, TypePtr::BOTTOM,
8524 state_start, result_start);
8525 // return key stream length (int)
8526 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
8527 set_result(retvalue);
8528 return true;
8529 }
8530
8531 //------------------------------inline_kyberNtt
8532 bool LibraryCallKit::inline_kyberNtt() {
8533 address stubAddr;
8534 const char *stubName;
8535 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8536 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
8537
8538 stubAddr = StubRoutines::kyberNtt();
8539 stubName = "kyberNtt";
8540 if (!stubAddr) return false;
8541
8542 Node* coeffs = argument(0);
8543 Node* ntt_zetas = argument(1);
8544
8545 coeffs = must_be_not_null(coeffs, true);
8546 ntt_zetas = must_be_not_null(ntt_zetas, true);
8547
8548 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8549 assert(coeffs_start, "coeffs is null");
8550 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
8551 assert(ntt_zetas_start, "ntt_zetas is null");
8552 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8553 OptoRuntime::kyberNtt_Type(),
8554 stubAddr, stubName, TypePtr::BOTTOM,
8555 coeffs_start, ntt_zetas_start);
8556 // return an int
8557 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
8558 set_result(retvalue);
8559 return true;
8560 }
8561
8562 //------------------------------inline_kyberInverseNtt
8563 bool LibraryCallKit::inline_kyberInverseNtt() {
8564 address stubAddr;
8565 const char *stubName;
8566 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8567 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
8568
8569 stubAddr = StubRoutines::kyberInverseNtt();
8570 stubName = "kyberInverseNtt";
8571 if (!stubAddr) return false;
8572
8573 Node* coeffs = argument(0);
8574 Node* zetas = argument(1);
8575
8576 coeffs = must_be_not_null(coeffs, true);
8577 zetas = must_be_not_null(zetas, true);
8578
8579 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8580 assert(coeffs_start, "coeffs is null");
8581 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8582 assert(zetas_start, "inverseNtt_zetas is null");
8583 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8584 OptoRuntime::kyberInverseNtt_Type(),
8585 stubAddr, stubName, TypePtr::BOTTOM,
8586 coeffs_start, zetas_start);
8587
8588 // return an int
8589 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
8590 set_result(retvalue);
8591 return true;
8592 }
8593
8594 //------------------------------inline_kyberNttMult
8595 bool LibraryCallKit::inline_kyberNttMult() {
8596 address stubAddr;
8597 const char *stubName;
8598 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8599 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
8600
8601 stubAddr = StubRoutines::kyberNttMult();
8602 stubName = "kyberNttMult";
8603 if (!stubAddr) return false;
8604
8605 Node* result = argument(0);
8606 Node* ntta = argument(1);
8607 Node* nttb = argument(2);
8608 Node* zetas = argument(3);
8609
8610 result = must_be_not_null(result, true);
8611 ntta = must_be_not_null(ntta, true);
8612 nttb = must_be_not_null(nttb, true);
8613 zetas = must_be_not_null(zetas, true);
8614
8615 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8616 assert(result_start, "result is null");
8617 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
8618 assert(ntta_start, "ntta is null");
8619 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
8620 assert(nttb_start, "nttb is null");
8621 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8622 assert(zetas_start, "nttMult_zetas is null");
8623 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8624 OptoRuntime::kyberNttMult_Type(),
8625 stubAddr, stubName, TypePtr::BOTTOM,
8626 result_start, ntta_start, nttb_start,
8627 zetas_start);
8628
8629 // return an int
8630 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
8631 set_result(retvalue);
8632
8633 return true;
8634 }
8635
8636 //------------------------------inline_kyberAddPoly_2
8637 bool LibraryCallKit::inline_kyberAddPoly_2() {
8638 address stubAddr;
8639 const char *stubName;
8640 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8641 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
8642
8643 stubAddr = StubRoutines::kyberAddPoly_2();
8644 stubName = "kyberAddPoly_2";
8645 if (!stubAddr) return false;
8646
8647 Node* result = argument(0);
8648 Node* a = argument(1);
8649 Node* b = argument(2);
8650
8651 result = must_be_not_null(result, true);
8652 a = must_be_not_null(a, true);
8653 b = must_be_not_null(b, true);
8654
8655 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8656 assert(result_start, "result is null");
8657 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8658 assert(a_start, "a is null");
8659 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8660 assert(b_start, "b is null");
8661 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
8662 OptoRuntime::kyberAddPoly_2_Type(),
8663 stubAddr, stubName, TypePtr::BOTTOM,
8664 result_start, a_start, b_start);
8665 // return an int
8666 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
8667 set_result(retvalue);
8668 return true;
8669 }
8670
8671 //------------------------------inline_kyberAddPoly_3
8672 bool LibraryCallKit::inline_kyberAddPoly_3() {
8673 address stubAddr;
8674 const char *stubName;
8675 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8676 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
8677
8678 stubAddr = StubRoutines::kyberAddPoly_3();
8679 stubName = "kyberAddPoly_3";
8680 if (!stubAddr) return false;
8681
8682 Node* result = argument(0);
8683 Node* a = argument(1);
8684 Node* b = argument(2);
8685 Node* c = argument(3);
8686
8687 result = must_be_not_null(result, true);
8688 a = must_be_not_null(a, true);
8689 b = must_be_not_null(b, true);
8690 c = must_be_not_null(c, true);
8691
8692 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8693 assert(result_start, "result is null");
8694 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8695 assert(a_start, "a is null");
8696 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8697 assert(b_start, "b is null");
8698 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
8699 assert(c_start, "c is null");
8700 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
8701 OptoRuntime::kyberAddPoly_3_Type(),
8702 stubAddr, stubName, TypePtr::BOTTOM,
8703 result_start, a_start, b_start, c_start);
8704 // return an int
8705 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
8706 set_result(retvalue);
8707 return true;
8708 }
8709
8710 //------------------------------inline_kyber12To16
8711 bool LibraryCallKit::inline_kyber12To16() {
8712 address stubAddr;
8713 const char *stubName;
8714 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8715 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
8716
8717 stubAddr = StubRoutines::kyber12To16();
8718 stubName = "kyber12To16";
8719 if (!stubAddr) return false;
8720
8721 Node* condensed = argument(0);
8722 Node* condensedOffs = argument(1);
8723 Node* parsed = argument(2);
8724 Node* parsedLength = argument(3);
8725
8726 condensed = must_be_not_null(condensed, true);
8727 parsed = must_be_not_null(parsed, true);
8728
8729 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
8730 assert(condensed_start, "condensed is null");
8731 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
8732 assert(parsed_start, "parsed is null");
8733 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
8734 OptoRuntime::kyber12To16_Type(),
8735 stubAddr, stubName, TypePtr::BOTTOM,
8736 condensed_start, condensedOffs, parsed_start, parsedLength);
8737 // return an int
8738 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
8739 set_result(retvalue);
8740 return true;
8741
8742 }
8743
8744 //------------------------------inline_kyberBarrettReduce
8745 bool LibraryCallKit::inline_kyberBarrettReduce() {
8746 address stubAddr;
8747 const char *stubName;
8748 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8749 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
8750
8751 stubAddr = StubRoutines::kyberBarrettReduce();
8752 stubName = "kyberBarrettReduce";
8753 if (!stubAddr) return false;
8754
8755 Node* coeffs = argument(0);
8756
8757 coeffs = must_be_not_null(coeffs, true);
8758
8759 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8760 assert(coeffs_start, "coeffs is null");
8761 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
8762 OptoRuntime::kyberBarrettReduce_Type(),
8763 stubAddr, stubName, TypePtr::BOTTOM,
8764 coeffs_start);
8765 // return an int
8766 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
8767 set_result(retvalue);
8768 return true;
8769 }
8770
8771 //------------------------------inline_dilithiumAlmostNtt
8772 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
8773 address stubAddr;
8774 const char *stubName;
8775 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8776 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
8777
8778 stubAddr = StubRoutines::dilithiumAlmostNtt();
8779 stubName = "dilithiumAlmostNtt";
8780 if (!stubAddr) return false;
8781
8782 Node* coeffs = argument(0);
8783 Node* ntt_zetas = argument(1);
8784
8785 coeffs = must_be_not_null(coeffs, true);
8786 ntt_zetas = must_be_not_null(ntt_zetas, true);
8787
8788 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8789 assert(coeffs_start, "coeffs is null");
8790 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
8791 assert(ntt_zetas_start, "ntt_zetas is null");
8792 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8793 OptoRuntime::dilithiumAlmostNtt_Type(),
8794 stubAddr, stubName, TypePtr::BOTTOM,
8795 coeffs_start, ntt_zetas_start);
8796 // return an int
8797 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
8798 set_result(retvalue);
8799 return true;
8800 }
8801
8802 //------------------------------inline_dilithiumAlmostInverseNtt
8803 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
8804 address stubAddr;
8805 const char *stubName;
8806 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8807 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
8808
8809 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
8810 stubName = "dilithiumAlmostInverseNtt";
8811 if (!stubAddr) return false;
8812
8813 Node* coeffs = argument(0);
8814 Node* zetas = argument(1);
8815
8816 coeffs = must_be_not_null(coeffs, true);
8817 zetas = must_be_not_null(zetas, true);
8818
8819 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8820 assert(coeffs_start, "coeffs is null");
8821 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
8822 assert(zetas_start, "inverseNtt_zetas is null");
8823 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8824 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
8825 stubAddr, stubName, TypePtr::BOTTOM,
8826 coeffs_start, zetas_start);
8827 // return an int
8828 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
8829 set_result(retvalue);
8830 return true;
8831 }
8832
8833 //------------------------------inline_dilithiumNttMult
8834 bool LibraryCallKit::inline_dilithiumNttMult() {
8835 address stubAddr;
8836 const char *stubName;
8837 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8838 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
8839
8840 stubAddr = StubRoutines::dilithiumNttMult();
8841 stubName = "dilithiumNttMult";
8842 if (!stubAddr) return false;
8843
8844 Node* result = argument(0);
8845 Node* ntta = argument(1);
8846 Node* nttb = argument(2);
8847 Node* zetas = argument(3);
8848
8849 result = must_be_not_null(result, true);
8850 ntta = must_be_not_null(ntta, true);
8851 nttb = must_be_not_null(nttb, true);
8852 zetas = must_be_not_null(zetas, true);
8853
8854 Node* result_start = array_element_address(result, intcon(0), T_INT);
8855 assert(result_start, "result is null");
8856 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
8857 assert(ntta_start, "ntta is null");
8858 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
8859 assert(nttb_start, "nttb is null");
8860 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8861 OptoRuntime::dilithiumNttMult_Type(),
8862 stubAddr, stubName, TypePtr::BOTTOM,
8863 result_start, ntta_start, nttb_start);
8864
8865 // return an int
8866 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8867 set_result(retvalue);
8868
8869 return true;
8870 }
8871
8872 //------------------------------inline_dilithiumMontMulByConstant
8873 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8874 address stubAddr;
8875 const char *stubName;
8876 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8877 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8878
8879 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8880 stubName = "dilithiumMontMulByConstant";
8881 if (!stubAddr) return false;
8882
8883 Node* coeffs = argument(0);
8884 Node* constant = argument(1);
8885
8886 coeffs = must_be_not_null(coeffs, true);
8887
8888 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8889 assert(coeffs_start, "coeffs is null");
8890 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8891 OptoRuntime::dilithiumMontMulByConstant_Type(),
8892 stubAddr, stubName, TypePtr::BOTTOM,
8893 coeffs_start, constant);
8894
8895 // return an int
8896 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8897 set_result(retvalue);
8898 return true;
8899 }
8900
8901
8902 //------------------------------inline_dilithiumDecomposePoly
8903 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8904 address stubAddr;
8905 const char *stubName;
8906 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8907 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8908
8909 stubAddr = StubRoutines::dilithiumDecomposePoly();
8910 stubName = "dilithiumDecomposePoly";
8911 if (!stubAddr) return false;
8912
8913 Node* input = argument(0);
8914 Node* lowPart = argument(1);
8915 Node* highPart = argument(2);
8916 Node* twoGamma2 = argument(3);
8917 Node* multiplier = argument(4);
8918
8919 input = must_be_not_null(input, true);
8920 lowPart = must_be_not_null(lowPart, true);
8921 highPart = must_be_not_null(highPart, true);
8922
8923 Node* input_start = array_element_address(input, intcon(0), T_INT);
8924 assert(input_start, "input is null");
8925 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8926 assert(lowPart_start, "lowPart is null");
8927 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8928 assert(highPart_start, "highPart is null");
8929
8930 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8931 OptoRuntime::dilithiumDecomposePoly_Type(),
8932 stubAddr, stubName, TypePtr::BOTTOM,
8933 input_start, lowPart_start, highPart_start,
8934 twoGamma2, multiplier);
8935
8936 // return an int
8937 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8938 set_result(retvalue);
8939 return true;
8940 }
8941
8942 bool LibraryCallKit::inline_base64_encodeBlock() {
8943 address stubAddr;
8944 const char *stubName;
8945 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8946 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8947 stubAddr = StubRoutines::base64_encodeBlock();
8948 stubName = "encodeBlock";
8949
8950 if (!stubAddr) return false;
8951 Node* base64obj = argument(0);
8952 Node* src = argument(1);
8953 Node* offset = argument(2);
8954 Node* len = argument(3);
8955 Node* dest = argument(4);
8956 Node* dp = argument(5);
8957 Node* isURL = argument(6);
8958
8959 src = must_be_not_null(src, true);
8960 dest = must_be_not_null(dest, true);
8961
8962 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8963 assert(src_start, "source array is null");
8964 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8965 assert(dest_start, "destination array is null");
8966
8967 Node* base64 = make_runtime_call(RC_LEAF,
8968 OptoRuntime::base64_encodeBlock_Type(),
8969 stubAddr, stubName, TypePtr::BOTTOM,
8970 src_start, offset, len, dest_start, dp, isURL);
8971 return true;
8972 }
8973
8974 bool LibraryCallKit::inline_base64_decodeBlock() {
8975 address stubAddr;
8976 const char *stubName;
8977 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8978 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8979 stubAddr = StubRoutines::base64_decodeBlock();
8980 stubName = "decodeBlock";
8981
8982 if (!stubAddr) return false;
8983 Node* base64obj = argument(0);
8984 Node* src = argument(1);
8985 Node* src_offset = argument(2);
8986 Node* len = argument(3);
8987 Node* dest = argument(4);
8988 Node* dest_offset = argument(5);
8989 Node* isURL = argument(6);
8990 Node* isMIME = argument(7);
8991
8992 src = must_be_not_null(src, true);
8993 dest = must_be_not_null(dest, true);
8994
8995 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8996 assert(src_start, "source array is null");
8997 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8998 assert(dest_start, "destination array is null");
8999
9000 Node* call = make_runtime_call(RC_LEAF,
9001 OptoRuntime::base64_decodeBlock_Type(),
9002 stubAddr, stubName, TypePtr::BOTTOM,
9003 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
9004 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
9005 set_result(result);
9006 return true;
9007 }
9008
9009 bool LibraryCallKit::inline_poly1305_processBlocks() {
9010 address stubAddr;
9011 const char *stubName;
9012 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
9013 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
9014 stubAddr = StubRoutines::poly1305_processBlocks();
9015 stubName = "poly1305_processBlocks";
9016
9017 if (!stubAddr) return false;
9018 null_check_receiver(); // null-check receiver
9019 if (stopped()) return true;
9020
9021 Node* input = argument(1);
9022 Node* input_offset = argument(2);
9023 Node* len = argument(3);
9024 Node* alimbs = argument(4);
9025 Node* rlimbs = argument(5);
9026
9027 input = must_be_not_null(input, true);
9028 alimbs = must_be_not_null(alimbs, true);
9029 rlimbs = must_be_not_null(rlimbs, true);
9030
9031 Node* input_start = array_element_address(input, input_offset, T_BYTE);
9032 assert(input_start, "input array is null");
9033 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
9034 assert(acc_start, "acc array is null");
9035 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
9036 assert(r_start, "r array is null");
9037
9038 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
9039 OptoRuntime::poly1305_processBlocks_Type(),
9040 stubAddr, stubName, TypePtr::BOTTOM,
9041 input_start, len, acc_start, r_start);
9042 return true;
9043 }
9044
9045 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
9046 address stubAddr;
9047 const char *stubName;
9048 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
9049 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
9050 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
9051 stubName = "intpoly_montgomeryMult_P256";
9052
9053 if (!stubAddr) return false;
9054 null_check_receiver(); // null-check receiver
9055 if (stopped()) return true;
9056
9057 Node* a = argument(1);
9058 Node* b = argument(2);
9059 Node* r = argument(3);
9060
9061 a = must_be_not_null(a, true);
9062 b = must_be_not_null(b, true);
9063 r = must_be_not_null(r, true);
9064
9065 Node* a_start = array_element_address(a, intcon(0), T_LONG);
9066 assert(a_start, "a array is null");
9067 Node* b_start = array_element_address(b, intcon(0), T_LONG);
9068 assert(b_start, "b array is null");
9069 Node* r_start = array_element_address(r, intcon(0), T_LONG);
9070 assert(r_start, "r array is null");
9071
9072 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
9073 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
9074 stubAddr, stubName, TypePtr::BOTTOM,
9075 a_start, b_start, r_start);
9076 return true;
9077 }
9078
9079 bool LibraryCallKit::inline_intpoly_assign() {
9080 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
9081 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
9082 const char *stubName = "intpoly_assign";
9083 address stubAddr = StubRoutines::intpoly_assign();
9084 if (!stubAddr) return false;
9085
9086 Node* set = argument(0);
9087 Node* a = argument(1);
9088 Node* b = argument(2);
9089 Node* arr_length = load_array_length(a);
9090
9091 a = must_be_not_null(a, true);
9092 b = must_be_not_null(b, true);
9093
9094 Node* a_start = array_element_address(a, intcon(0), T_LONG);
9095 assert(a_start, "a array is null");
9096 Node* b_start = array_element_address(b, intcon(0), T_LONG);
9097 assert(b_start, "b array is null");
9098
9099 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
9100 OptoRuntime::intpoly_assign_Type(),
9101 stubAddr, stubName, TypePtr::BOTTOM,
9102 set, a_start, b_start, arr_length);
9103 return true;
9104 }
9105
9106 //------------------------------inline_digestBase_implCompress-----------------------
9107 //
9108 // Calculate MD5 for single-block byte[] array.
9109 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
9110 //
9111 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
9112 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
9113 //
9114 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
9115 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
9116 //
9117 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
9118 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
9119 //
9120 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
9121 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
9122 //
9123 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
9124 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
9125
9126 Node* digestBase_obj = argument(0);
9127 Node* src = argument(1); // type oop
9128 Node* ofs = argument(2); // type int
9129
9130 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
9131 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
9132 // failed array check
9133 return false;
9134 }
9135 // Figure out the size and type of the elements we will be copying.
9136 BasicType src_elem = src_type->elem()->array_element_basic_type();
9137 if (src_elem != T_BYTE) {
9138 return false;
9139 }
9140 // 'src_start' points to src array + offset
9141 src = must_be_not_null(src, true);
9142 Node* src_start = array_element_address(src, ofs, src_elem);
9143 Node* state = nullptr;
9144 Node* block_size = nullptr;
9145 address stubAddr;
9146 const char *stubName;
9147
9148 switch(id) {
9149 case vmIntrinsics::_md5_implCompress:
9150 assert(UseMD5Intrinsics, "need MD5 instruction support");
9151 state = get_state_from_digest_object(digestBase_obj, T_INT);
9152 stubAddr = StubRoutines::md5_implCompress();
9153 stubName = "md5_implCompress";
9154 break;
9155 case vmIntrinsics::_sha_implCompress:
9156 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
9157 state = get_state_from_digest_object(digestBase_obj, T_INT);
9158 stubAddr = StubRoutines::sha1_implCompress();
9159 stubName = "sha1_implCompress";
9160 break;
9161 case vmIntrinsics::_sha2_implCompress:
9162 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
9163 state = get_state_from_digest_object(digestBase_obj, T_INT);
9164 stubAddr = StubRoutines::sha256_implCompress();
9165 stubName = "sha256_implCompress";
9166 break;
9167 case vmIntrinsics::_sha5_implCompress:
9168 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
9169 state = get_state_from_digest_object(digestBase_obj, T_LONG);
9170 stubAddr = StubRoutines::sha512_implCompress();
9171 stubName = "sha512_implCompress";
9172 break;
9173 case vmIntrinsics::_sha3_implCompress:
9174 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
9175 state = get_state_from_digest_object(digestBase_obj, T_LONG);
9176 stubAddr = StubRoutines::sha3_implCompress();
9177 stubName = "sha3_implCompress";
9178 block_size = get_block_size_from_digest_object(digestBase_obj);
9179 if (block_size == nullptr) return false;
9180 break;
9181 default:
9182 fatal_unexpected_iid(id);
9183 return false;
9184 }
9185 if (state == nullptr) return false;
9186
9187 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
9188 if (stubAddr == nullptr) return false;
9189
9190 // Call the stub.
9191 Node* call;
9192 if (block_size == nullptr) {
9193 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
9194 stubAddr, stubName, TypePtr::BOTTOM,
9195 src_start, state);
9196 } else {
9197 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
9198 stubAddr, stubName, TypePtr::BOTTOM,
9199 src_start, state, block_size);
9200 }
9201
9202 return true;
9203 }
9204
9205 //------------------------------inline_keccak
9206 bool LibraryCallKit::inline_keccak(vmIntrinsics::ID id) {
9207 address stubAddr = nullptr;
9208 const char *stubName;
9209 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
9210 assert((id == vmIntrinsics::_double_keccak && callee()->signature()->size() == 2) ||
9211 (id == vmIntrinsics::_quad_keccak && callee()->signature()->size() == 4),
9212 "double_keccak wrong number of parameters");
9213
9214 int parmCnt = 0;
9215 switch (id) {
9216 case vmIntrinsics::_double_keccak:
9217 stubAddr = StubRoutines::double_keccak();
9218 stubName = "double_keccak";
9219 parmCnt = 2;
9220 break;
9221 case vmIntrinsics::_quad_keccak:
9222 stubAddr = StubRoutines::quad_keccak();
9223 stubName = "quad_keccak";
9224 parmCnt = 4;
9225 break;
9226 default:
9227 ShouldNotReachHere();
9228 }
9229
9230 if (!stubAddr) return false;
9231
9232 Node* state[4];
9233 for (int i = 0; i<parmCnt; i++) {
9234 state[i] = must_be_not_null(argument(i), true);
9235 state[i] = array_element_address(state[i], intcon(0), T_LONG);
9236 assert(state[i], "state[%d] is null", i);
9237 }
9238
9239 Node* keccak;
9240 switch (id) {
9241 case vmIntrinsics::_double_keccak:
9242 keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
9243 OptoRuntime::double_keccak_Type(),
9244 stubAddr, stubName, TypePtr::BOTTOM,
9245 state[0], state[1]);
9246 break;
9247 case vmIntrinsics::_quad_keccak:
9248 keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
9249 OptoRuntime::quad_keccak_Type(),
9250 stubAddr, stubName, TypePtr::BOTTOM,
9251 state[0], state[1], state[2], state[3]);
9252 break;
9253 default:
9254 ShouldNotReachHere();
9255 }
9256
9257 // return an int
9258 Node* retvalue = _gvn.transform(new ProjNode(keccak, TypeFunc::Parms));
9259 set_result(retvalue);
9260 return true;
9261 }
9262
9263
9264 //------------------------------inline_digestBase_implCompressMB-----------------------
9265 //
9266 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
9267 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
9268 //
9269 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
9270 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9271 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9272 assert((uint)predicate < 5, "sanity");
9273 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
9274
9275 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
9276 Node* src = argument(1); // byte[] array
9277 Node* ofs = argument(2); // type int
9278 Node* limit = argument(3); // type int
9279
9280 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
9281 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
9282 // failed array check
9283 return false;
9284 }
9285 // Figure out the size and type of the elements we will be copying.
9286 BasicType src_elem = src_type->elem()->array_element_basic_type();
9287 if (src_elem != T_BYTE) {
9288 return false;
9289 }
9290 // 'src_start' points to src array + offset
9291 src = must_be_not_null(src, false);
9292 Node* src_start = array_element_address(src, ofs, src_elem);
9293
9294 const char* klass_digestBase_name = nullptr;
9295 const char* stub_name = nullptr;
9296 address stub_addr = nullptr;
9297 BasicType elem_type = T_INT;
9298
9299 switch (predicate) {
9300 case 0:
9301 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
9302 klass_digestBase_name = "sun/security/provider/MD5";
9303 stub_name = "md5_implCompressMB";
9304 stub_addr = StubRoutines::md5_implCompressMB();
9305 }
9306 break;
9307 case 1:
9308 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
9309 klass_digestBase_name = "sun/security/provider/SHA";
9310 stub_name = "sha1_implCompressMB";
9311 stub_addr = StubRoutines::sha1_implCompressMB();
9312 }
9313 break;
9314 case 2:
9315 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
9316 klass_digestBase_name = "sun/security/provider/SHA2";
9317 stub_name = "sha256_implCompressMB";
9318 stub_addr = StubRoutines::sha256_implCompressMB();
9319 }
9320 break;
9321 case 3:
9322 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
9323 klass_digestBase_name = "sun/security/provider/SHA5";
9324 stub_name = "sha512_implCompressMB";
9325 stub_addr = StubRoutines::sha512_implCompressMB();
9326 elem_type = T_LONG;
9327 }
9328 break;
9329 case 4:
9330 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
9331 klass_digestBase_name = "sun/security/provider/SHA3";
9332 stub_name = "sha3_implCompressMB";
9333 stub_addr = StubRoutines::sha3_implCompressMB();
9334 elem_type = T_LONG;
9335 }
9336 break;
9337 default:
9338 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
9339 }
9340 if (klass_digestBase_name != nullptr) {
9341 assert(stub_addr != nullptr, "Stub is generated");
9342 if (stub_addr == nullptr) return false;
9343
9344 // get DigestBase klass to lookup for SHA klass
9345 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
9346 assert(tinst != nullptr, "digestBase_obj is not instance???");
9347 assert(tinst->is_loaded(), "DigestBase is not loaded");
9348
9349 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
9350 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
9351 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
9352 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
9353 }
9354 return false;
9355 }
9356
9357 //------------------------------inline_digestBase_implCompressMB-----------------------
9358 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
9359 BasicType elem_type, address stubAddr, const char *stubName,
9360 Node* src_start, Node* ofs, Node* limit) {
9361 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
9362 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
9363 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
9364 digest_obj = _gvn.transform(digest_obj);
9365
9366 Node* state = get_state_from_digest_object(digest_obj, elem_type);
9367 if (state == nullptr) return false;
9368
9369 Node* block_size = nullptr;
9370 if (strcmp("sha3_implCompressMB", stubName) == 0) {
9371 block_size = get_block_size_from_digest_object(digest_obj);
9372 if (block_size == nullptr) return false;
9373 }
9374
9375 // Call the stub.
9376 Node* call;
9377 if (block_size == nullptr) {
9378 call = make_runtime_call(RC_LEAF|RC_NO_FP,
9379 OptoRuntime::digestBase_implCompressMB_Type(false),
9380 stubAddr, stubName, TypePtr::BOTTOM,
9381 src_start, state, ofs, limit);
9382 } else {
9383 call = make_runtime_call(RC_LEAF|RC_NO_FP,
9384 OptoRuntime::digestBase_implCompressMB_Type(true),
9385 stubAddr, stubName, TypePtr::BOTTOM,
9386 src_start, state, block_size, ofs, limit);
9387 }
9388
9389 // return ofs (int)
9390 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
9391 set_result(result);
9392
9393 return true;
9394 }
9395
9396 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
9397 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
9398 assert(UseAES, "need AES instruction support");
9399 address stubAddr = nullptr;
9400 const char *stubName = nullptr;
9401 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
9402 stubName = "galoisCounterMode_AESCrypt";
9403
9404 if (stubAddr == nullptr) return false;
9405
9406 Node* in = argument(0);
9407 Node* inOfs = argument(1);
9408 Node* len = argument(2);
9409 Node* ct = argument(3);
9410 Node* ctOfs = argument(4);
9411 Node* out = argument(5);
9412 Node* outOfs = argument(6);
9413 Node* gctr_object = argument(7);
9414 Node* ghash_object = argument(8);
9415
9416 // (1) in, ct and out are arrays.
9417 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
9418 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
9419 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
9420 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
9421 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
9422 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
9423
9424 // checks are the responsibility of the caller
9425 Node* in_start = in;
9426 Node* ct_start = ct;
9427 Node* out_start = out;
9428 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
9429 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
9430 in_start = array_element_address(in, inOfs, T_BYTE);
9431 ct_start = array_element_address(ct, ctOfs, T_BYTE);
9432 out_start = array_element_address(out, outOfs, T_BYTE);
9433 }
9434
9435 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
9436 // (because of the predicated logic executed earlier).
9437 // so we cast it here safely.
9438 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
9439 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9440 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
9441 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
9442 Node* state = load_field_from_object(ghash_object, "state", "[J");
9443
9444 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
9445 return false;
9446 }
9447 // cast it to what we know it will be at runtime
9448 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
9449 assert(tinst != nullptr, "GCTR obj is null");
9450 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9451 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
9452 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
9453 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9454 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
9455 const TypeOopPtr* xtype = aklass->as_instance_type();
9456 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
9457 aescrypt_object = _gvn.transform(aescrypt_object);
9458 // we need to get the start of the aescrypt_object's expanded key array
9459 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
9460 if (k_start == nullptr) return false;
9461 // similarly, get the start address of the r vector
9462 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
9463 Node* state_start = array_element_address(state, intcon(0), T_LONG);
9464 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
9465
9466
9467 // Call the stub, passing params
9468 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
9469 OptoRuntime::galoisCounterMode_aescrypt_Type(),
9470 stubAddr, stubName, TypePtr::BOTTOM,
9471 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
9472
9473 // return cipher length (int)
9474 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
9475 set_result(retvalue);
9476
9477 return true;
9478 }
9479
9480 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
9481 // Return node representing slow path of predicate check.
9482 // the pseudo code we want to emulate with this predicate is:
9483 // for encryption:
9484 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
9485 // for decryption:
9486 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
9487 // note cipher==plain is more conservative than the original java code but that's OK
9488 //
9489
9490 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
9491 // The receiver was checked for null already.
9492 Node* objGCTR = argument(7);
9493 // Load embeddedCipher field of GCTR object.
9494 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9495 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
9496
9497 // get AESCrypt klass for instanceOf check
9498 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
9499 // will have same classloader as CipherBlockChaining object
9500 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
9501 assert(tinst != nullptr, "GCTR obj is null");
9502 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9503
9504 // we want to do an instanceof comparison against the AESCrypt class
9505 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
9506 if (!klass_AESCrypt->is_loaded()) {
9507 // if AESCrypt is not even loaded, we never take the intrinsic fast path
9508 Node* ctrl = control();
9509 set_control(top()); // no regular fast path
9510 return ctrl;
9511 }
9512
9513 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9514 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
9515 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9516 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9517 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9518
9519 return instof_false; // even if it is null
9520 }
9521
9522 //------------------------------get_state_from_digest_object-----------------------
9523 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
9524 const char* state_type;
9525 switch (elem_type) {
9526 case T_BYTE: state_type = "[B"; break;
9527 case T_INT: state_type = "[I"; break;
9528 case T_LONG: state_type = "[J"; break;
9529 default: ShouldNotReachHere();
9530 }
9531 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
9532 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
9533 if (digest_state == nullptr) return (Node *) nullptr;
9534
9535 // now have the array, need to get the start address of the state array
9536 Node* state = array_element_address(digest_state, intcon(0), elem_type);
9537 return state;
9538 }
9539
9540 //------------------------------get_block_size_from_sha3_object----------------------------------
9541 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
9542 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
9543 assert (block_size != nullptr, "sanity");
9544 return block_size;
9545 }
9546
9547 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
9548 // Return node representing slow path of predicate check.
9549 // the pseudo code we want to emulate with this predicate is:
9550 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
9551 //
9552 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
9553 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9554 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9555 assert((uint)predicate < 5, "sanity");
9556
9557 // The receiver was checked for null already.
9558 Node* digestBaseObj = argument(0);
9559
9560 // get DigestBase klass for instanceOf check
9561 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
9562 assert(tinst != nullptr, "digestBaseObj is null");
9563 assert(tinst->is_loaded(), "DigestBase is not loaded");
9564
9565 const char* klass_name = nullptr;
9566 switch (predicate) {
9567 case 0:
9568 if (UseMD5Intrinsics) {
9569 // we want to do an instanceof comparison against the MD5 class
9570 klass_name = "sun/security/provider/MD5";
9571 }
9572 break;
9573 case 1:
9574 if (UseSHA1Intrinsics) {
9575 // we want to do an instanceof comparison against the SHA class
9576 klass_name = "sun/security/provider/SHA";
9577 }
9578 break;
9579 case 2:
9580 if (UseSHA256Intrinsics) {
9581 // we want to do an instanceof comparison against the SHA2 class
9582 klass_name = "sun/security/provider/SHA2";
9583 }
9584 break;
9585 case 3:
9586 if (UseSHA512Intrinsics) {
9587 // we want to do an instanceof comparison against the SHA5 class
9588 klass_name = "sun/security/provider/SHA5";
9589 }
9590 break;
9591 case 4:
9592 if (UseSHA3Intrinsics) {
9593 // we want to do an instanceof comparison against the SHA3 class
9594 klass_name = "sun/security/provider/SHA3";
9595 }
9596 break;
9597 default:
9598 fatal("unknown SHA intrinsic predicate: %d", predicate);
9599 }
9600
9601 ciKlass* klass = nullptr;
9602 if (klass_name != nullptr) {
9603 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
9604 }
9605 if ((klass == nullptr) || !klass->is_loaded()) {
9606 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
9607 Node* ctrl = control();
9608 set_control(top()); // no intrinsic path
9609 return ctrl;
9610 }
9611 ciInstanceKlass* instklass = klass->as_instance_klass();
9612
9613 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
9614 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9615 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9616 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9617
9618 return instof_false; // even if it is null
9619 }
9620
9621 //-------------inline_fma-----------------------------------
9622 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
9623 Node *a = nullptr;
9624 Node *b = nullptr;
9625 Node *c = nullptr;
9626 Node* result = nullptr;
9627 switch (id) {
9628 case vmIntrinsics::_fmaD:
9629 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
9630 // no receiver since it is static method
9631 a = argument(0);
9632 b = argument(2);
9633 c = argument(4);
9634 result = _gvn.transform(new FmaDNode(a, b, c));
9635 break;
9636 case vmIntrinsics::_fmaF:
9637 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
9638 a = argument(0);
9639 b = argument(1);
9640 c = argument(2);
9641 result = _gvn.transform(new FmaFNode(a, b, c));
9642 break;
9643 default:
9644 fatal_unexpected_iid(id); break;
9645 }
9646 set_result(result);
9647 return true;
9648 }
9649
9650 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
9651 // argument(0) is receiver
9652 Node* codePoint = argument(1);
9653 Node* n = nullptr;
9654
9655 switch (id) {
9656 case vmIntrinsics::_isDigit :
9657 n = new DigitNode(control(), codePoint);
9658 break;
9659 case vmIntrinsics::_isLowerCase :
9660 n = new LowerCaseNode(control(), codePoint);
9661 break;
9662 case vmIntrinsics::_isUpperCase :
9663 n = new UpperCaseNode(control(), codePoint);
9664 break;
9665 case vmIntrinsics::_isWhitespace :
9666 n = new WhitespaceNode(control(), codePoint);
9667 break;
9668 default:
9669 fatal_unexpected_iid(id);
9670 }
9671
9672 set_result(_gvn.transform(n));
9673 return true;
9674 }
9675
9676 bool LibraryCallKit::inline_profileBoolean() {
9677 Node* counts = argument(1);
9678 const TypeAryPtr* ary = nullptr;
9679 ciArray* aobj = nullptr;
9680 if (counts->is_Con()
9681 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
9682 && (aobj = ary->const_oop()->as_array()) != nullptr
9683 && (aobj->length() == 2)) {
9684 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
9685 jint false_cnt = aobj->element_value(0).as_int();
9686 jint true_cnt = aobj->element_value(1).as_int();
9687
9688 if (C->log() != nullptr) {
9689 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
9690 false_cnt, true_cnt);
9691 }
9692
9693 if (false_cnt + true_cnt == 0) {
9694 // According to profile, never executed.
9695 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9696 Deoptimization::Action_reinterpret);
9697 return true;
9698 }
9699
9700 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
9701 // is a number of each value occurrences.
9702 Node* result = argument(0);
9703 if (false_cnt == 0 || true_cnt == 0) {
9704 // According to profile, one value has been never seen.
9705 int expected_val = (false_cnt == 0) ? 1 : 0;
9706
9707 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
9708 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
9709
9710 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
9711 Node* fast_path = _gvn.transform(new IfTrueNode(check));
9712 Node* slow_path = _gvn.transform(new IfFalseNode(check));
9713
9714 { // Slow path: uncommon trap for never seen value and then reexecute
9715 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
9716 // the value has been seen at least once.
9717 PreserveJVMState pjvms(this);
9718 PreserveReexecuteState preexecs(this);
9719 jvms()->set_should_reexecute(true);
9720
9721 set_control(slow_path);
9722 set_i_o(i_o());
9723
9724 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9725 Deoptimization::Action_reinterpret);
9726 }
9727 // The guard for never seen value enables sharpening of the result and
9728 // returning a constant. It allows to eliminate branches on the same value
9729 // later on.
9730 set_control(fast_path);
9731 result = intcon(expected_val);
9732 }
9733 // Stop profiling.
9734 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
9735 // By replacing method body with profile data (represented as ProfileBooleanNode
9736 // on IR level) we effectively disable profiling.
9737 // It enables full speed execution once optimized code is generated.
9738 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
9739 C->record_for_igvn(profile);
9740 set_result(profile);
9741 return true;
9742 } else {
9743 // Continue profiling.
9744 // Profile data isn't available at the moment. So, execute method's bytecode version.
9745 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
9746 // is compiled and counters aren't available since corresponding MethodHandle
9747 // isn't a compile-time constant.
9748 return false;
9749 }
9750 }
9751
9752 bool LibraryCallKit::inline_isCompileConstant() {
9753 Node* n = argument(0);
9754 set_result(n->is_Con() ? intcon(1) : intcon(0));
9755 return true;
9756 }
9757
9758 //------------------------------- inline_getObjectSize --------------------------------------
9759 //
9760 // Calculate the runtime size of the object/array.
9761 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
9762 //
9763 bool LibraryCallKit::inline_getObjectSize() {
9764 Node* obj = argument(3);
9765 Node* klass_node = load_object_klass(obj);
9766
9767 jint layout_con = Klass::_lh_neutral_value;
9768 Node* layout_val = get_layout_helper(klass_node, layout_con);
9769 int layout_is_con = (layout_val == nullptr);
9770
9771 if (layout_is_con) {
9772 // Layout helper is constant, can figure out things at compile time.
9773
9774 if (Klass::layout_helper_is_instance(layout_con)) {
9775 // Instance case: layout_con contains the size itself.
9776 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
9777 set_result(size);
9778 } else {
9779 // Array case: size is round(header + element_size*arraylength).
9780 // Since arraylength is different for every array instance, we have to
9781 // compute the whole thing at runtime.
9782
9783 Node* arr_length = load_array_length(obj);
9784
9785 int round_mask = MinObjAlignmentInBytes - 1;
9786 int hsize = Klass::layout_helper_header_size(layout_con);
9787 int eshift = Klass::layout_helper_log2_element_size(layout_con);
9788
9789 if ((round_mask & ~right_n_bits(eshift)) == 0) {
9790 round_mask = 0; // strength-reduce it if it goes away completely
9791 }
9792 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
9793 Node* header_size = intcon(hsize + round_mask);
9794
9795 Node* lengthx = ConvI2X(arr_length);
9796 Node* headerx = ConvI2X(header_size);
9797
9798 Node* abody = lengthx;
9799 if (eshift != 0) {
9800 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
9801 }
9802 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
9803 if (round_mask != 0) {
9804 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
9805 }
9806 size = ConvX2L(size);
9807 set_result(size);
9808 }
9809 } else {
9810 // Layout helper is not constant, need to test for array-ness at runtime.
9811
9812 enum { _instance_path = 1, _array_path, PATH_LIMIT };
9813 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
9814 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
9815 record_for_igvn(result_reg);
9816
9817 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
9818 if (array_ctl != nullptr) {
9819 // Array case: size is round(header + element_size*arraylength).
9820 // Since arraylength is different for every array instance, we have to
9821 // compute the whole thing at runtime.
9822
9823 PreserveJVMState pjvms(this);
9824 set_control(array_ctl);
9825 Node* arr_length = load_array_length(obj);
9826
9827 int round_mask = MinObjAlignmentInBytes - 1;
9828 Node* mask = intcon(round_mask);
9829
9830 Node* hss = intcon(Klass::_lh_header_size_shift);
9831 Node* hsm = intcon(Klass::_lh_header_size_mask);
9832 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
9833 header_size = _gvn.transform(new AndINode(header_size, hsm));
9834 header_size = _gvn.transform(new AddINode(header_size, mask));
9835
9836 // There is no need to mask or shift this value.
9837 // The semantics of LShiftINode include an implicit mask to 0x1F.
9838 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
9839 Node* elem_shift = layout_val;
9840
9841 Node* lengthx = ConvI2X(arr_length);
9842 Node* headerx = ConvI2X(header_size);
9843
9844 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
9845 Node* size = _gvn.transform(new AddXNode(headerx, abody));
9846 if (round_mask != 0) {
9847 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
9848 }
9849 size = ConvX2L(size);
9850
9851 result_reg->init_req(_array_path, control());
9852 result_val->init_req(_array_path, size);
9853 }
9854
9855 if (!stopped()) {
9856 // Instance case: the layout helper gives us instance size almost directly,
9857 // but we need to mask out the _lh_instance_slow_path_bit.
9858 Node* size = ConvI2X(layout_val);
9859 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
9860 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
9861 size = _gvn.transform(new AndXNode(size, mask));
9862 size = ConvX2L(size);
9863
9864 result_reg->init_req(_instance_path, control());
9865 result_val->init_req(_instance_path, size);
9866 }
9867
9868 set_result(result_reg, result_val);
9869 }
9870
9871 return true;
9872 }
9873
9874 //------------------------------- inline_blackhole --------------------------------------
9875 //
9876 // Make sure all arguments to this node are alive.
9877 // This matches methods that were requested to be blackholed through compile commands.
9878 //
9879 bool LibraryCallKit::inline_blackhole() {
9880 assert(callee()->is_static(), "Should have been checked before: only static methods here");
9881 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
9882 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
9883
9884 // Blackhole node pinches only the control, not memory. This allows
9885 // the blackhole to be pinned in the loop that computes blackholed
9886 // values, but have no other side effects, like breaking the optimizations
9887 // across the blackhole.
9888
9889 Node* bh = _gvn.transform(new BlackholeNode(control()));
9890 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
9891
9892 // Bind call arguments as blackhole arguments to keep them alive
9893 uint nargs = callee()->arg_size();
9894 for (uint i = 0; i < nargs; i++) {
9895 bh->add_req(argument(i));
9896 }
9897
9898 return true;
9899 }
9900
9901 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9902 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9903 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9904 return nullptr; // box klass is not Float16
9905 }
9906
9907 // Null check; get notnull casted pointer
9908 Node* null_ctl = top();
9909 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9910 // If not_null_box is dead, only null-path is taken
9911 if (stopped()) {
9912 set_control(null_ctl);
9913 return nullptr;
9914 }
9915 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9916 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9917 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9918 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9919 }
9920
9921 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9922 PreserveReexecuteState preexecs(this);
9923 jvms()->set_should_reexecute(true);
9924
9925 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9926 Node* klass_node = makecon(klass_type);
9927 Node* box = new_instance(klass_node);
9928
9929 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9930 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9931
9932 Node* field_store = _gvn.transform(access_store_at(box,
9933 value_field,
9934 value_adr_type,
9935 value,
9936 TypeInt::SHORT,
9937 T_SHORT,
9938 IN_HEAP));
9939 set_memory(field_store, value_adr_type);
9940 return box;
9941 }
9942
9943 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9944 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9945 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9946 return false;
9947 }
9948
9949 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9950 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9951 return false;
9952 }
9953
9954 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9955 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9956 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9957 ciSymbols::short_signature(),
9958 false);
9959 assert(field != nullptr, "");
9960
9961 // Transformed nodes
9962 Node* fld1 = nullptr;
9963 Node* fld2 = nullptr;
9964 Node* fld3 = nullptr;
9965 switch(num_args) {
9966 case 3:
9967 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9968 if (fld3 == nullptr) {
9969 return false;
9970 }
9971 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9972 // fall-through
9973 case 2:
9974 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9975 if (fld2 == nullptr) {
9976 return false;
9977 }
9978 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9979 // fall-through
9980 case 1:
9981 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9982 if (fld1 == nullptr) {
9983 return false;
9984 }
9985 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9986 break;
9987 default: fatal("Unsupported number of arguments %d", num_args);
9988 }
9989
9990 Node* result = nullptr;
9991 switch (id) {
9992 // Unary operations
9993 case vmIntrinsics::_sqrt_float16:
9994 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9995 break;
9996 // Ternary operations
9997 case vmIntrinsics::_fma_float16:
9998 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9999 break;
10000 default:
10001 fatal_unexpected_iid(id);
10002 break;
10003 }
10004 result = _gvn.transform(new ReinterpretHF2SNode(result));
10005 set_result(box_fp16_value(float16_box_type, field, result));
10006 return true;
10007 }
10008