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::_intpoly_mult_25519:
696 return inline_intpoly_mult_25519();
697 case vmIntrinsics::_intpoly_square_25519:
698 return inline_intpoly_square_25519();
699 case vmIntrinsics::_encodeISOArray:
700 case vmIntrinsics::_encodeByteISOArray:
701 return inline_encodeISOArray(false);
702 case vmIntrinsics::_encodeAsciiArray:
703 return inline_encodeISOArray(true);
704
705 case vmIntrinsics::_updateCRC32:
706 return inline_updateCRC32();
707 case vmIntrinsics::_updateBytesCRC32:
708 return inline_updateBytesCRC32();
709 case vmIntrinsics::_updateByteBufferCRC32:
710 return inline_updateByteBufferCRC32();
711
712 case vmIntrinsics::_updateBytesCRC32C:
713 return inline_updateBytesCRC32C();
714 case vmIntrinsics::_updateDirectByteBufferCRC32C:
715 return inline_updateDirectByteBufferCRC32C();
716
717 case vmIntrinsics::_updateBytesAdler32:
718 return inline_updateBytesAdler32();
719 case vmIntrinsics::_updateByteBufferAdler32:
720 return inline_updateByteBufferAdler32();
721
722 case vmIntrinsics::_profileBoolean:
723 return inline_profileBoolean();
724 case vmIntrinsics::_isCompileConstant:
725 return inline_isCompileConstant();
726
727 case vmIntrinsics::_countPositives:
728 return inline_countPositives();
729
730 case vmIntrinsics::_fmaD:
731 case vmIntrinsics::_fmaF:
732 return inline_fma(intrinsic_id());
733
734 case vmIntrinsics::_isDigit:
735 case vmIntrinsics::_isLowerCase:
736 case vmIntrinsics::_isUpperCase:
737 case vmIntrinsics::_isWhitespace:
738 return inline_character_compare(intrinsic_id());
739
740 case vmIntrinsics::_min:
741 case vmIntrinsics::_max:
742 case vmIntrinsics::_min_strict:
743 case vmIntrinsics::_max_strict:
744 case vmIntrinsics::_minL:
745 case vmIntrinsics::_maxL:
746 case vmIntrinsics::_minF:
747 case vmIntrinsics::_maxF:
748 case vmIntrinsics::_minD:
749 case vmIntrinsics::_maxD:
750 case vmIntrinsics::_minF_strict:
751 case vmIntrinsics::_maxF_strict:
752 case vmIntrinsics::_minD_strict:
753 case vmIntrinsics::_maxD_strict:
754 return inline_min_max(intrinsic_id());
755
756 case vmIntrinsics::_VectorUnaryOp:
757 return inline_vector_nary_operation(1);
758 case vmIntrinsics::_VectorBinaryOp:
759 return inline_vector_nary_operation(2);
760 case vmIntrinsics::_VectorUnaryLibOp:
761 return inline_vector_call(1);
762 case vmIntrinsics::_VectorBinaryLibOp:
763 return inline_vector_call(2);
764 case vmIntrinsics::_VectorTernaryOp:
765 return inline_vector_nary_operation(3);
766 case vmIntrinsics::_VectorFromBitsCoerced:
767 return inline_vector_frombits_coerced();
768 case vmIntrinsics::_VectorMaskOp:
769 return inline_vector_mask_operation();
770 case vmIntrinsics::_VectorLoadOp:
771 return inline_vector_mem_operation(/*is_store=*/false);
772 case vmIntrinsics::_VectorLoadMaskedOp:
773 return inline_vector_mem_masked_operation(/*is_store*/false);
774 case vmIntrinsics::_VectorStoreOp:
775 return inline_vector_mem_operation(/*is_store=*/true);
776 case vmIntrinsics::_VectorStoreMaskedOp:
777 return inline_vector_mem_masked_operation(/*is_store=*/true);
778 case vmIntrinsics::_VectorGatherOp:
779 return inline_vector_gather_scatter(/*is_scatter*/ false);
780 case vmIntrinsics::_VectorScatterOp:
781 return inline_vector_gather_scatter(/*is_scatter*/ true);
782 case vmIntrinsics::_VectorReductionCoerced:
783 return inline_vector_reduction();
784 case vmIntrinsics::_VectorTest:
785 return inline_vector_test();
786 case vmIntrinsics::_VectorBlend:
787 return inline_vector_blend();
788 case vmIntrinsics::_VectorRearrange:
789 return inline_vector_rearrange();
790 case vmIntrinsics::_VectorSelectFrom:
791 return inline_vector_select_from();
792 case vmIntrinsics::_VectorCompare:
793 return inline_vector_compare();
794 case vmIntrinsics::_VectorBroadcastInt:
795 return inline_vector_broadcast_int();
796 case vmIntrinsics::_VectorConvert:
797 return inline_vector_convert();
798 case vmIntrinsics::_VectorInsert:
799 return inline_vector_insert();
800 case vmIntrinsics::_VectorExtract:
801 return inline_vector_extract();
802 case vmIntrinsics::_VectorCompressExpand:
803 return inline_vector_compress_expand();
804 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
805 return inline_vector_select_from_two_vectors();
806 case vmIntrinsics::_IndexVector:
807 return inline_index_vector();
808 case vmIntrinsics::_IndexPartiallyInUpperRange:
809 return inline_index_partially_in_upper_range();
810
811 case vmIntrinsics::_getObjectSize:
812 return inline_getObjectSize();
813
814 case vmIntrinsics::_blackhole:
815 return inline_blackhole();
816
817 default:
818 // If you get here, it may be that someone has added a new intrinsic
819 // to the list in vmIntrinsics.hpp without implementing it here.
820 #ifndef PRODUCT
821 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
822 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
823 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
824 }
825 #endif
826 return false;
827 }
828 }
829
830 Node* LibraryCallKit::try_to_predicate(int predicate) {
831 if (!jvms()->has_method()) {
832 // Root JVMState has a null method.
833 assert(map()->memory()->Opcode() == Op_Parm, "");
834 // Insert the memory aliasing node
835 set_all_memory(reset_memory());
836 }
837 assert(merged_memory(), "");
838
839 switch (intrinsic_id()) {
840 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
841 return inline_cipherBlockChaining_AESCrypt_predicate(false);
842 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
843 return inline_cipherBlockChaining_AESCrypt_predicate(true);
844 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
845 return inline_electronicCodeBook_AESCrypt_predicate(false);
846 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
847 return inline_electronicCodeBook_AESCrypt_predicate(true);
848 case vmIntrinsics::_counterMode_AESCrypt:
849 return inline_counterMode_AESCrypt_predicate();
850 case vmIntrinsics::_digestBase_implCompressMB:
851 return inline_digestBase_implCompressMB_predicate(predicate);
852 case vmIntrinsics::_galoisCounterMode_AESCrypt:
853 return inline_galoisCounterMode_AESCrypt_predicate();
854
855 default:
856 // If you get here, it may be that someone has added a new intrinsic
857 // to the list in vmIntrinsics.hpp without implementing it here.
858 #ifndef PRODUCT
859 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
860 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
861 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
862 }
863 #endif
864 Node* slow_ctl = control();
865 set_control(top()); // No fast path intrinsic
866 return slow_ctl;
867 }
868 }
869
870 //------------------------------set_result-------------------------------
871 // Helper function for finishing intrinsics.
872 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
873 record_for_igvn(region);
874 set_control(_gvn.transform(region));
875 set_result( _gvn.transform(value));
876 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
877 }
878
879 RegionNode* LibraryCallKit::create_bailout() {
880 RegionNode* bailout = new RegionNode(1);
881 record_for_igvn(bailout);
882 return bailout;
883 }
884
885 bool LibraryCallKit::check_bailout(RegionNode* bailout) {
886 if (bailout->req() > 1) {
887 bailout = _gvn.transform(bailout)->as_Region();
888 Node* frame = _gvn.transform(new ParmNode(C->start(), TypeFunc::FramePtr));
889 Node* halt = _gvn.transform(new HaltNode(bailout, frame, "unexpected guard failure in intrinsic"));
890 C->root()->add_req(halt);
891 }
892 return stopped();
893 }
894
895 //------------------------------generate_guard---------------------------
896 // Helper function for generating guarded fast-slow graph structures.
897 // The given 'test', if true, guards a slow path. If the test fails
898 // then a fast path can be taken. (We generally hope it fails.)
899 // In all cases, GraphKit::control() is updated to the fast path.
900 // The returned value represents the control for the slow path.
901 // The return value is never 'top'; it is either a valid control
902 // or null if it is obvious that the slow path can never be taken.
903 // Also, if region and the slow control are not null, the slow edge
904 // is appended to the region.
905 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
906 if (stopped()) {
907 // Already short circuited.
908 return nullptr;
909 }
910
911 // Build an if node and its projections.
912 // If test is true we take the slow path, which we assume is uncommon.
913 if (_gvn.type(test) == TypeInt::ZERO) {
914 // The slow branch is never taken. No need to build this guard.
915 return nullptr;
916 }
917
918 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
919
920 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
921 if (if_slow == top()) {
922 // The slow branch is never taken. No need to build this guard.
923 return nullptr;
924 }
925
926 if (region != nullptr)
927 region->add_req(if_slow);
928
929 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
930 set_control(if_fast);
931
932 return if_slow;
933 }
934
935 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
936 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
937 }
938 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
939 return generate_guard(test, region, PROB_FAIR);
940 }
941
942 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
943 Node** pos_index, bool with_opaque) {
944 if (stopped())
945 return nullptr; // already stopped
946 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
947 return nullptr; // index is already adequately typed
948 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
949 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
950 if (with_opaque) {
951 bol_lt = _gvn.transform(new OpaqueConstantBoolNode(C, bol_lt, false));
952 }
953 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
954 if (is_neg != nullptr && pos_index != nullptr) {
955 // Emulate effect of Parse::adjust_map_after_if.
956 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
957 (*pos_index) = _gvn.transform(ccast);
958 }
959 return is_neg;
960 }
961
962 // Make sure that 'position' is a valid limit index, in [0..length].
963 // There are two equivalent plans for checking this:
964 // A. (offset + copyLength) unsigned<= arrayLength
965 // B. offset <= (arrayLength - copyLength)
966 // We require that all of the values above, except for the sum and
967 // difference, are already known to be non-negative.
968 // Plan A is robust in the face of overflow, if offset and copyLength
969 // are both hugely positive.
970 //
971 // Plan B is less direct and intuitive, but it does not overflow at
972 // all, since the difference of two non-negatives is always
973 // representable. Whenever Java methods must perform the equivalent
974 // check they generally use Plan B instead of Plan A.
975 // For the moment we use Plan A.
976 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
977 Node* subseq_length,
978 Node* array_length,
979 RegionNode* region,
980 bool with_opaque) {
981 if (stopped())
982 return nullptr; // already stopped
983 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
984 if (zero_offset && subseq_length->eqv_uncast(array_length))
985 return nullptr; // common case of whole-array copy
986 Node* last = subseq_length;
987 if (!zero_offset) // last += offset
988 last = _gvn.transform(new AddINode(last, offset));
989 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
990 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
991 if (with_opaque) {
992 bol_lt = _gvn.transform(new OpaqueConstantBoolNode(C, bol_lt, false));
993 }
994 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
995 return is_over;
996 }
997
998 // Emit range checks for the given String.value byte array
999 void LibraryCallKit::generate_string_range_check(Node* array,
1000 Node* offset,
1001 Node* count,
1002 bool char_count,
1003 RegionNode* region) {
1004 if (stopped()) {
1005 return; // already stopped
1006 }
1007 if (char_count) {
1008 // Convert char count to byte count
1009 count = _gvn.transform(new LShiftINode(count, intcon(1)));
1010 }
1011 // Offset and count must not be negative
1012 generate_negative_guard(offset, region, nullptr, true);
1013 generate_negative_guard(count, region, nullptr, true);
1014 // Offset + count must not exceed length of array
1015 generate_limit_guard(offset, count, load_array_length(array), region, true);
1016 }
1017
1018 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
1019 bool is_immutable) {
1020 ciKlass* thread_klass = env()->Thread_klass();
1021 const Type* thread_type
1022 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1023
1024 Node* thread = _gvn.transform(new ThreadLocalNode());
1025 Node* p = off_heap_plus_addr(thread, in_bytes(handle_offset));
1026 tls_output = thread;
1027
1028 Node* thread_obj_handle
1029 = (is_immutable
1030 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
1031 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
1032 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
1033 thread_obj_handle = _gvn.transform(thread_obj_handle);
1034
1035 DecoratorSet decorators = IN_NATIVE;
1036 if (is_immutable) {
1037 decorators |= C2_IMMUTABLE_MEMORY;
1038 }
1039 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
1040 }
1041
1042 //--------------------------generate_current_thread--------------------
1043 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1044 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
1045 /*is_immutable*/false);
1046 }
1047
1048 //--------------------------generate_virtual_thread--------------------
1049 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
1050 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
1051 !C->method()->changes_current_thread());
1052 }
1053
1054 //------------------------------make_string_method_node------------------------
1055 // Helper method for String intrinsic functions. This version is called with
1056 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1057 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1058 // containing the lengths of str1 and str2.
1059 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1060 Node* result = nullptr;
1061 switch (opcode) {
1062 case Op_StrIndexOf:
1063 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1064 str1_start, cnt1, str2_start, cnt2, ae);
1065 break;
1066 case Op_StrComp:
1067 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1068 str1_start, cnt1, str2_start, cnt2, ae);
1069 break;
1070 case Op_StrEquals:
1071 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1072 // Use the constant length if there is one because optimized match rule may exist.
1073 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1074 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1075 break;
1076 default:
1077 ShouldNotReachHere();
1078 return nullptr;
1079 }
1080
1081 // All these intrinsics have checks.
1082 C->set_has_split_ifs(true); // Has chance for split-if optimization
1083 clear_upper_avx();
1084
1085 return _gvn.transform(result);
1086 }
1087
1088 //------------------------------inline_string_compareTo------------------------
1089 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1090 Node* arg1 = argument(0);
1091 Node* arg2 = argument(1);
1092
1093 arg1 = must_be_not_null(arg1, true);
1094 arg2 = must_be_not_null(arg2, true);
1095
1096 // Get start addr and length of first argument
1097 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1098 Node* arg1_cnt = load_array_length(arg1);
1099
1100 // Get start addr and length of second argument
1101 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1102 Node* arg2_cnt = load_array_length(arg2);
1103
1104 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1105 set_result(result);
1106 return true;
1107 }
1108
1109 //------------------------------inline_string_equals------------------------
1110 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1111 Node* arg1 = argument(0);
1112 Node* arg2 = argument(1);
1113
1114 // paths (plus control) merge
1115 RegionNode* region = new RegionNode(3);
1116 Node* phi = new PhiNode(region, TypeInt::BOOL);
1117
1118 if (!stopped()) {
1119
1120 arg1 = must_be_not_null(arg1, true);
1121 arg2 = must_be_not_null(arg2, true);
1122
1123 // Get start addr and length of first argument
1124 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1125 Node* arg1_cnt = load_array_length(arg1);
1126
1127 // Get start addr and length of second argument
1128 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1129 Node* arg2_cnt = load_array_length(arg2);
1130
1131 // Check for arg1_cnt != arg2_cnt
1132 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1133 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1134 Node* if_ne = generate_slow_guard(bol, nullptr);
1135 if (if_ne != nullptr) {
1136 phi->init_req(2, intcon(0));
1137 region->init_req(2, if_ne);
1138 }
1139
1140 // Check for count == 0 is done by assembler code for StrEquals.
1141
1142 if (!stopped()) {
1143 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1144 phi->init_req(1, equals);
1145 region->init_req(1, control());
1146 }
1147 }
1148
1149 // post merge
1150 set_control(_gvn.transform(region));
1151 record_for_igvn(region);
1152
1153 set_result(_gvn.transform(phi));
1154 return true;
1155 }
1156
1157 //------------------------------inline_array_equals----------------------------
1158 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1159 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1160 Node* arg1 = argument(0);
1161 Node* arg2 = argument(1);
1162
1163 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1164 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), mtype, arg1, arg2, ae)));
1165 clear_upper_avx();
1166
1167 return true;
1168 }
1169
1170
1171 //------------------------------inline_countPositives------------------------------
1172 // int java.lang.StringCoding#countPositives0(byte[] ba, int off, int len)
1173 bool LibraryCallKit::inline_countPositives() {
1174 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1175 // no receiver since it is static method
1176 Node* ba = argument(0);
1177 Node* offset = argument(1);
1178 Node* len = argument(2);
1179
1180 ba = must_be_not_null(ba, true);
1181 RegionNode* bailout = create_bailout();
1182 generate_string_range_check(ba, offset, len, false, bailout);
1183 if (check_bailout(bailout)) {
1184 return true;
1185 }
1186
1187 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1188 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1189 set_result(_gvn.transform(result));
1190 clear_upper_avx();
1191 return true;
1192 }
1193
1194 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1195 Node* index = argument(0);
1196 Node* length = bt == T_INT ? argument(1) : argument(2);
1197 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1198 return false;
1199 }
1200
1201 // check that length is positive
1202 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1203 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1204
1205 {
1206 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1207 uncommon_trap(Deoptimization::Reason_intrinsic,
1208 Deoptimization::Action_make_not_entrant);
1209 }
1210
1211 if (stopped()) {
1212 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1213 return true;
1214 }
1215
1216 // length is now known positive, add a cast node to make this explicit
1217 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1218 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1219 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1220 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1221 casted_length = _gvn.transform(casted_length);
1222 replace_in_map(length, casted_length);
1223 length = casted_length;
1224
1225 // Use an unsigned comparison for the range check itself
1226 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1227 BoolTest::mask btest = BoolTest::lt;
1228 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1229 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1230 _gvn.set_type(rc, rc->Value(&_gvn));
1231 if (!rc_bool->is_Con()) {
1232 record_for_igvn(rc);
1233 }
1234 set_control(_gvn.transform(new IfTrueNode(rc)));
1235 {
1236 PreserveJVMState pjvms(this);
1237 set_control(_gvn.transform(new IfFalseNode(rc)));
1238 uncommon_trap(Deoptimization::Reason_range_check,
1239 Deoptimization::Action_make_not_entrant);
1240 }
1241
1242 if (stopped()) {
1243 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1244 return true;
1245 }
1246
1247 // index is now known to be >= 0 and < length, cast it
1248 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1249 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1250 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1251 result = _gvn.transform(result);
1252 set_result(result);
1253 replace_in_map(index, result);
1254 return true;
1255 }
1256
1257 //------------------------------inline_string_indexOf------------------------
1258 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1259 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1260 return false;
1261 }
1262 Node* src = argument(0);
1263 Node* tgt = argument(1);
1264
1265 // Make the merge point
1266 RegionNode* result_rgn = new RegionNode(4);
1267 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1268
1269 src = must_be_not_null(src, true);
1270 tgt = must_be_not_null(tgt, true);
1271
1272 // Get start addr and length of source string
1273 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1274 Node* src_count = load_array_length(src);
1275
1276 // Get start addr and length of substring
1277 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1278 Node* tgt_count = load_array_length(tgt);
1279
1280 Node* result = nullptr;
1281 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1282
1283 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1284 // Divide src size by 2 if String is UTF16 encoded
1285 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1286 }
1287 if (ae == StrIntrinsicNode::UU) {
1288 // Divide substring size by 2 if String is UTF16 encoded
1289 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1290 }
1291
1292 if (call_opt_stub) {
1293 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1294 StubRoutines::_string_indexof_array[ae],
1295 "stringIndexOf", TypePtr::BOTTOM, src_start,
1296 src_count, tgt_start, tgt_count);
1297 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1298 } else {
1299 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1300 result_rgn, result_phi, ae);
1301 }
1302 if (result != nullptr) {
1303 result_phi->init_req(3, result);
1304 result_rgn->init_req(3, control());
1305 }
1306 set_control(_gvn.transform(result_rgn));
1307 record_for_igvn(result_rgn);
1308 set_result(_gvn.transform(result_phi));
1309
1310 return true;
1311 }
1312
1313 //-----------------------------inline_string_indexOfI-----------------------
1314 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1315 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1316 return false;
1317 }
1318
1319 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1320 Node* src = argument(0); // byte[]
1321 Node* src_count = argument(1); // char count
1322 Node* tgt = argument(2); // byte[]
1323 Node* tgt_count = argument(3); // char count
1324 Node* from_index = argument(4); // char index
1325
1326 src = must_be_not_null(src, true);
1327 tgt = must_be_not_null(tgt, true);
1328
1329 // Multiply byte array index by 2 if String is UTF16 encoded
1330 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1331 src_count = _gvn.transform(new SubINode(src_count, from_index));
1332 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1333 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1334
1335 // Range checks
1336 RegionNode* bailout = create_bailout();
1337 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL, bailout);
1338 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU, bailout);
1339 if (check_bailout(bailout)) {
1340 return true;
1341 }
1342
1343 RegionNode* region = new RegionNode(5);
1344 Node* phi = new PhiNode(region, TypeInt::INT);
1345 Node* result = nullptr;
1346
1347 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1348
1349 if (call_opt_stub) {
1350 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1351 StubRoutines::_string_indexof_array[ae],
1352 "stringIndexOf", TypePtr::BOTTOM, src_start,
1353 src_count, tgt_start, tgt_count);
1354 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1355 } else {
1356 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1357 region, phi, ae);
1358 }
1359 if (result != nullptr) {
1360 // The result is index relative to from_index if substring was found, -1 otherwise.
1361 // Generate code which will fold into cmove.
1362 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1363 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1364
1365 Node* if_lt = generate_slow_guard(bol, nullptr);
1366 if (if_lt != nullptr) {
1367 // result == -1
1368 phi->init_req(3, result);
1369 region->init_req(3, if_lt);
1370 }
1371 if (!stopped()) {
1372 result = _gvn.transform(new AddINode(result, from_index));
1373 phi->init_req(4, result);
1374 region->init_req(4, control());
1375 }
1376 }
1377
1378 set_control(_gvn.transform(region));
1379 record_for_igvn(region);
1380 set_result(_gvn.transform(phi));
1381 clear_upper_avx();
1382
1383 return true;
1384 }
1385
1386 // Create StrIndexOfNode with fast path checks
1387 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1388 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1389 // Check for substr count > string count
1390 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1391 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1392 Node* if_gt = generate_slow_guard(bol, nullptr);
1393 if (if_gt != nullptr) {
1394 phi->init_req(1, intcon(-1));
1395 region->init_req(1, if_gt);
1396 }
1397 if (!stopped()) {
1398 // Check for substr count == 0
1399 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1400 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1401 Node* if_zero = generate_slow_guard(bol, nullptr);
1402 if (if_zero != nullptr) {
1403 phi->init_req(2, intcon(0));
1404 region->init_req(2, if_zero);
1405 }
1406 }
1407 if (!stopped()) {
1408 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1409 }
1410 return nullptr;
1411 }
1412
1413 //-----------------------------inline_string_indexOfChar-----------------------
1414 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1415 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1416 return false;
1417 }
1418 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1419 return false;
1420 }
1421 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1422 Node* src = argument(0); // byte[]
1423 Node* int_ch = argument(1);
1424 Node* from_index = argument(2);
1425 Node* max = argument(3);
1426
1427 src = must_be_not_null(src, true);
1428
1429 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1430 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1431 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1432
1433 // Range checks
1434 RegionNode* bailout = create_bailout();
1435 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U, bailout);
1436 if (check_bailout(bailout)) {
1437 return true;
1438 }
1439
1440 // Check for int_ch >= 0
1441 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1442 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1443 {
1444 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1445 uncommon_trap(Deoptimization::Reason_intrinsic,
1446 Deoptimization::Action_maybe_recompile);
1447 }
1448 if (stopped()) {
1449 return true;
1450 }
1451
1452 RegionNode* region = new RegionNode(3);
1453 Node* phi = new PhiNode(region, TypeInt::INT);
1454
1455 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1456 C->set_has_split_ifs(true); // Has chance for split-if optimization
1457 _gvn.transform(result);
1458
1459 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1460 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1461
1462 Node* if_lt = generate_slow_guard(bol, nullptr);
1463 if (if_lt != nullptr) {
1464 // result == -1
1465 phi->init_req(2, result);
1466 region->init_req(2, if_lt);
1467 }
1468 if (!stopped()) {
1469 result = _gvn.transform(new AddINode(result, from_index));
1470 phi->init_req(1, result);
1471 region->init_req(1, control());
1472 }
1473 set_control(_gvn.transform(region));
1474 record_for_igvn(region);
1475 set_result(_gvn.transform(phi));
1476 clear_upper_avx();
1477
1478 return true;
1479 }
1480 //---------------------------inline_string_copy---------------------
1481 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1482 // int StringUTF16.compress0(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1483 // int StringUTF16.compress0(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1484 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1485 // void StringLatin1.inflate0(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1486 // void StringLatin1.inflate0(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1487 bool LibraryCallKit::inline_string_copy(bool compress) {
1488 int nargs = 5; // 2 oops, 3 ints
1489 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1490
1491 Node* src = argument(0);
1492 Node* src_offset = argument(1);
1493 Node* dst = argument(2);
1494 Node* dst_offset = argument(3);
1495 Node* length = argument(4);
1496
1497 // Check for allocation before we add nodes that would confuse
1498 // tightly_coupled_allocation()
1499 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1500
1501 // Figure out the size and type of the elements we will be copying.
1502 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1503 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1504 if (src_type == nullptr || dst_type == nullptr) {
1505 return false;
1506 }
1507 BasicType src_elem = src_type->elem()->array_element_basic_type();
1508 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1509 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1510 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1511 "Unsupported array types for inline_string_copy");
1512
1513 src = must_be_not_null(src, true);
1514 dst = must_be_not_null(dst, true);
1515
1516 // Convert char[] offsets to byte[] offsets
1517 bool convert_src = (compress && src_elem == T_BYTE);
1518 bool convert_dst = (!compress && dst_elem == T_BYTE);
1519 if (convert_src) {
1520 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1521 } else if (convert_dst) {
1522 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1523 }
1524
1525 // Range checks
1526 RegionNode* bailout = create_bailout();
1527 generate_string_range_check(src, src_offset, length, convert_src, bailout);
1528 generate_string_range_check(dst, dst_offset, length, convert_dst, bailout);
1529 if (check_bailout(bailout)) {
1530 return true;
1531 }
1532
1533 Node* src_start = array_element_address(src, src_offset, src_elem);
1534 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1535 // 'src_start' points to src array + scaled offset
1536 // 'dst_start' points to dst array + scaled offset
1537 Node* count = nullptr;
1538 if (compress) {
1539 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1540 } else {
1541 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1542 }
1543
1544 if (alloc != nullptr) {
1545 if (alloc->maybe_set_complete(&_gvn)) {
1546 // "You break it, you buy it."
1547 InitializeNode* init = alloc->initialization();
1548 assert(init->is_complete(), "we just did this");
1549 init->set_complete_with_arraycopy();
1550 assert(dst->is_CheckCastPP(), "sanity");
1551 assert(dst->in(0)->in(0) == init, "dest pinned");
1552 }
1553 // Do not let stores that initialize this object be reordered with
1554 // a subsequent store that would make this object accessible by
1555 // other threads.
1556 // Record what AllocateNode this StoreStore protects so that
1557 // escape analysis can go from the MemBarStoreStoreNode to the
1558 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1559 // based on the escape status of the AllocateNode.
1560 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1561 }
1562 if (compress) {
1563 set_result(_gvn.transform(count));
1564 }
1565 clear_upper_avx();
1566
1567 return true;
1568 }
1569
1570 #ifdef _LP64
1571 #define XTOP ,top() /*additional argument*/
1572 #else //_LP64
1573 #define XTOP /*no additional argument*/
1574 #endif //_LP64
1575
1576 //------------------------inline_string_toBytesU--------------------------
1577 // public static byte[] StringUTF16.toBytes0(char[] value, int off, int len)
1578 bool LibraryCallKit::inline_string_toBytesU() {
1579 // Get the arguments.
1580 assert(callee()->signature()->size() == 3, "character array encoder requires 3 arguments");
1581 Node* value = argument(0);
1582 Node* offset = argument(1);
1583 Node* length = argument(2);
1584
1585 Node* newcopy = nullptr;
1586
1587 // Set the original stack and the reexecute bit for the interpreter to reexecute
1588 // the bytecode that invokes StringUTF16.toBytes0() if deoptimization happens.
1589 { PreserveReexecuteState preexecs(this);
1590 jvms()->set_should_reexecute(true);
1591
1592 value = must_be_not_null(value, true);
1593 RegionNode* bailout = create_bailout();
1594 generate_negative_guard(offset, bailout, nullptr, true);
1595 generate_negative_guard(length, bailout, nullptr, true);
1596 generate_limit_guard(offset, length, load_array_length(value), bailout, true);
1597 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1598 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout, true);
1599 if (check_bailout(bailout)) {
1600 return true;
1601 }
1602
1603 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1604 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1605 newcopy = new_array(klass_node, size, 0); // no arguments to push
1606 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1607 guarantee(alloc != nullptr, "created above");
1608
1609 // Calculate starting addresses.
1610 Node* src_start = array_element_address(value, offset, T_CHAR);
1611 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1612
1613 // Check if dst array address is aligned to HeapWordSize
1614 bool aligned = (arrayOopDesc::base_offset_in_bytes(T_BYTE) % HeapWordSize == 0);
1615 // If true, then check if src array address is aligned to HeapWordSize
1616 if (aligned) {
1617 const TypeInt* toffset = gvn().type(offset)->is_int();
1618 aligned = toffset->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) +
1619 toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1620 }
1621
1622 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1623 const char* copyfunc_name = "arraycopy";
1624 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1625 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1626 OptoRuntime::fast_arraycopy_Type(),
1627 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1628 src_start, dst_start, ConvI2X(length) XTOP);
1629 // Do not let reads from the cloned object float above the arraycopy.
1630 if (alloc->maybe_set_complete(&_gvn)) {
1631 // "You break it, you buy it."
1632 InitializeNode* init = alloc->initialization();
1633 assert(init->is_complete(), "we just did this");
1634 init->set_complete_with_arraycopy();
1635 assert(newcopy->is_CheckCastPP(), "sanity");
1636 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1637 }
1638 // Do not let stores that initialize this object be reordered with
1639 // a subsequent store that would make this object accessible by
1640 // other threads.
1641 // Record what AllocateNode this StoreStore protects so that
1642 // escape analysis can go from the MemBarStoreStoreNode to the
1643 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1644 // based on the escape status of the AllocateNode.
1645 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1646 } // original reexecute is set back here
1647
1648 C->set_has_split_ifs(true); // Has chance for split-if optimization
1649 if (!stopped()) {
1650 set_result(newcopy);
1651 }
1652 clear_upper_avx();
1653
1654 return true;
1655 }
1656
1657 //------------------------inline_string_getCharsU--------------------------
1658 // public void StringUTF16.getChars0(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1659 bool LibraryCallKit::inline_string_getCharsU() {
1660 assert(callee()->signature()->size() == 5, "StringUTF16.getChars0() has 5 arguments");
1661 // Get the arguments.
1662 Node* src = argument(0);
1663 Node* src_begin = argument(1);
1664 Node* src_end = argument(2); // exclusive offset (i < src_end)
1665 Node* dst = argument(3);
1666 Node* dst_begin = argument(4);
1667
1668 // Check for allocation before we add nodes that would confuse
1669 // tightly_coupled_allocation()
1670 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1671
1672 // Check if a null path was taken unconditionally.
1673 src = must_be_not_null(src, true);
1674 dst = must_be_not_null(dst, true);
1675 if (stopped()) {
1676 return true;
1677 }
1678
1679 // Get length and convert char[] offset to byte[] offset
1680 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1681 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1682
1683 // Range checks
1684 RegionNode* bailout = create_bailout();
1685 generate_string_range_check(src, src_begin, length, true, bailout);
1686 generate_string_range_check(dst, dst_begin, length, false, bailout);
1687 if (check_bailout(bailout)) {
1688 return true;
1689 }
1690
1691 // Calculate starting addresses.
1692 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1693 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1694
1695 // Check if array addresses are aligned to HeapWordSize
1696 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1697 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1698 bool aligned = tsrc->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_BYTE) + tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1699 tdst->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) + tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1700
1701 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1702 const char* copyfunc_name = "arraycopy";
1703 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1704 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1705 OptoRuntime::fast_arraycopy_Type(),
1706 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1707 src_start, dst_start, ConvI2X(length) XTOP);
1708 // Do not let reads from the cloned object float above the arraycopy.
1709 if (alloc != nullptr) {
1710 if (alloc->maybe_set_complete(&_gvn)) {
1711 // "You break it, you buy it."
1712 InitializeNode* init = alloc->initialization();
1713 assert(init->is_complete(), "we just did this");
1714 init->set_complete_with_arraycopy();
1715 assert(dst->is_CheckCastPP(), "sanity");
1716 assert(dst->in(0)->in(0) == init, "dest pinned");
1717 }
1718 // Do not let stores that initialize this object be reordered with
1719 // a subsequent store that would make this object accessible by
1720 // other threads.
1721 // Record what AllocateNode this StoreStore protects so that
1722 // escape analysis can go from the MemBarStoreStoreNode to the
1723 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1724 // based on the escape status of the AllocateNode.
1725 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1726 } else {
1727 insert_mem_bar(Op_MemBarCPUOrder);
1728 }
1729
1730 C->set_has_split_ifs(true); // Has chance for split-if optimization
1731 return true;
1732 }
1733
1734 //----------------------inline_string_char_access----------------------------
1735 // Store/Load char to/from byte[] array.
1736 // static void StringUTF16.putChar(byte[] val, int index, int c)
1737 // static char StringUTF16.getChar(byte[] val, int index)
1738 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1739 Node* ch;
1740 if (is_store) {
1741 assert(callee()->signature()->size() == 3, "StringUTF16.putChar() has 3 arguments");
1742 ch = argument(2);
1743 } else {
1744 assert(callee()->signature()->size() == 2, "StringUTF16.getChar() has 2 arguments");
1745 ch = nullptr;
1746 }
1747 Node* value = argument(0);
1748 Node* index = argument(1);
1749
1750 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1751 // correctly requires matched array shapes.
1752 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1753 "sanity: byte[] and char[] bases agree");
1754 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1755 "sanity: byte[] and char[] scales agree");
1756
1757 // Bail when getChar over constants is requested: constant folding would
1758 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1759 // Java method would constant fold nicely instead.
1760 if (!is_store && value->is_Con() && index->is_Con()) {
1761 return false;
1762 }
1763
1764 // Save state and restore on bailout
1765 SavedState old_state(this);
1766
1767 value = must_be_not_null(value, true);
1768
1769 Node* adr = array_element_address(value, index, T_CHAR);
1770 if (adr->is_top()) {
1771 return false;
1772 }
1773 old_state.discard();
1774 if (is_store) {
1775 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1776 } else {
1777 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);
1778 set_result(ch);
1779 }
1780 return true;
1781 }
1782
1783
1784 //------------------------------inline_math-----------------------------------
1785 // public static double Math.abs(double)
1786 // public static double Math.sqrt(double)
1787 // public static double Math.log(double)
1788 // public static double Math.log10(double)
1789 // public static double Math.round(double)
1790 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1791 Node* arg = argument(0);
1792 Node* n = nullptr;
1793 switch (id) {
1794 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1795 case vmIntrinsics::_dsqrt:
1796 case vmIntrinsics::_dsqrt_strict:
1797 n = new SqrtDNode(C, control(), arg); break;
1798 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1799 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1800 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1801 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1802 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1803 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1804 default: fatal_unexpected_iid(id); break;
1805 }
1806 set_result(_gvn.transform(n));
1807 return true;
1808 }
1809
1810 //------------------------------inline_math-----------------------------------
1811 // public static float Math.abs(float)
1812 // public static int Math.abs(int)
1813 // public static long Math.abs(long)
1814 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1815 Node* arg = argument(0);
1816 Node* n = nullptr;
1817 switch (id) {
1818 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1819 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1820 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1821 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1822 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1823 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1824 default: fatal_unexpected_iid(id); break;
1825 }
1826 set_result(_gvn.transform(n));
1827 return true;
1828 }
1829
1830 //------------------------------runtime_math-----------------------------
1831 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1832 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1833 "must be (DD)D or (D)D type");
1834
1835 // Inputs
1836 Node* a = argument(0);
1837 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1838
1839 const TypePtr* no_memory_effects = nullptr;
1840 Node* trig = make_runtime_call(RC_LEAF | RC_PURE, call_type, funcAddr, funcName,
1841 no_memory_effects,
1842 a, top(), b, b ? top() : nullptr);
1843 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1844 #ifdef ASSERT
1845 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1846 assert(value_top == top(), "second value must be top");
1847 #endif
1848
1849 set_result(value);
1850 return true;
1851 }
1852
1853 //------------------------------inline_math_pow-----------------------------
1854 bool LibraryCallKit::inline_math_pow() {
1855 Node* base = argument(0);
1856 Node* exp = argument(2);
1857
1858 CallNode* pow = new PowDNode(C, base, exp);
1859 set_predefined_input_for_runtime_call(pow);
1860 pow = _gvn.transform(pow)->as_CallLeafPure();
1861 set_predefined_output_for_runtime_call(pow);
1862 Node* result = _gvn.transform(new ProjNode(pow, TypeFunc::Parms + 0));
1863 record_for_igvn(pow);
1864 set_result(result);
1865 return true;
1866 }
1867
1868 //------------------------------inline_math_native-----------------------------
1869 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1870 switch (id) {
1871 case vmIntrinsics::_dsin:
1872 return StubRoutines::dsin() != nullptr ?
1873 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1874 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1875 case vmIntrinsics::_dcos:
1876 return StubRoutines::dcos() != nullptr ?
1877 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1878 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1879 case vmIntrinsics::_dtan:
1880 return StubRoutines::dtan() != nullptr ?
1881 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1882 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1883 case vmIntrinsics::_dsinh:
1884 return StubRoutines::dsinh() != nullptr ?
1885 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsinh(), "dsinh") : false;
1886 case vmIntrinsics::_dtanh:
1887 return StubRoutines::dtanh() != nullptr ?
1888 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1889 case vmIntrinsics::_dcbrt:
1890 return StubRoutines::dcbrt() != nullptr ?
1891 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcbrt(), "dcbrt") : false;
1892 case vmIntrinsics::_dexp:
1893 return StubRoutines::dexp() != nullptr ?
1894 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1895 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1896 case vmIntrinsics::_dlog:
1897 return StubRoutines::dlog() != nullptr ?
1898 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1899 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1900 case vmIntrinsics::_dlog10:
1901 return StubRoutines::dlog10() != nullptr ?
1902 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1903 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1904
1905 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1906 case vmIntrinsics::_ceil:
1907 case vmIntrinsics::_floor:
1908 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1909
1910 case vmIntrinsics::_dsqrt:
1911 case vmIntrinsics::_dsqrt_strict:
1912 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1913 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1914 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1915 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1916 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1917
1918 case vmIntrinsics::_dpow: return inline_math_pow();
1919 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1920 case vmIntrinsics::_fcopySign: return inline_math(id);
1921 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1922 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1923 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1924
1925 // These intrinsics are not yet correctly implemented
1926 case vmIntrinsics::_datan2:
1927 return false;
1928
1929 default:
1930 fatal_unexpected_iid(id);
1931 return false;
1932 }
1933 }
1934
1935 //----------------------------inline_notify-----------------------------------*
1936 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1937 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1938 address func;
1939 if (id == vmIntrinsics::_notify) {
1940 func = OptoRuntime::monitor_notify_Java();
1941 } else {
1942 func = OptoRuntime::monitor_notifyAll_Java();
1943 }
1944 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1945 make_slow_call_ex(call, env()->Throwable_klass(), false);
1946 return true;
1947 }
1948
1949
1950 //----------------------------inline_min_max-----------------------------------
1951 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1952 Node* a = nullptr;
1953 Node* b = nullptr;
1954 Node* n = nullptr;
1955 switch (id) {
1956 case vmIntrinsics::_min:
1957 case vmIntrinsics::_max:
1958 case vmIntrinsics::_minF:
1959 case vmIntrinsics::_maxF:
1960 case vmIntrinsics::_minF_strict:
1961 case vmIntrinsics::_maxF_strict:
1962 case vmIntrinsics::_min_strict:
1963 case vmIntrinsics::_max_strict:
1964 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
1965 a = argument(0);
1966 b = argument(1);
1967 break;
1968 case vmIntrinsics::_minD:
1969 case vmIntrinsics::_maxD:
1970 case vmIntrinsics::_minD_strict:
1971 case vmIntrinsics::_maxD_strict:
1972 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
1973 a = argument(0);
1974 b = argument(2);
1975 break;
1976 case vmIntrinsics::_minL:
1977 case vmIntrinsics::_maxL:
1978 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
1979 a = argument(0);
1980 b = argument(2);
1981 break;
1982 default:
1983 fatal_unexpected_iid(id);
1984 break;
1985 }
1986
1987 switch (id) {
1988 case vmIntrinsics::_min:
1989 case vmIntrinsics::_min_strict:
1990 n = new MinINode(a, b);
1991 break;
1992 case vmIntrinsics::_max:
1993 case vmIntrinsics::_max_strict:
1994 n = new MaxINode(a, b);
1995 break;
1996 case vmIntrinsics::_minF:
1997 case vmIntrinsics::_minF_strict:
1998 n = new MinFNode(a, b);
1999 break;
2000 case vmIntrinsics::_maxF:
2001 case vmIntrinsics::_maxF_strict:
2002 n = new MaxFNode(a, b);
2003 break;
2004 case vmIntrinsics::_minD:
2005 case vmIntrinsics::_minD_strict:
2006 n = new MinDNode(a, b);
2007 break;
2008 case vmIntrinsics::_maxD:
2009 case vmIntrinsics::_maxD_strict:
2010 n = new MaxDNode(a, b);
2011 break;
2012 case vmIntrinsics::_minL:
2013 n = new MinLNode(_gvn.C, a, b);
2014 break;
2015 case vmIntrinsics::_maxL:
2016 n = new MaxLNode(_gvn.C, a, b);
2017 break;
2018 default:
2019 fatal_unexpected_iid(id);
2020 break;
2021 }
2022
2023 set_result(_gvn.transform(n));
2024 return true;
2025 }
2026
2027 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
2028 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
2029 env()->ArithmeticException_instance())) {
2030 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
2031 // so let's bail out intrinsic rather than risking deopting again.
2032 return false;
2033 }
2034
2035 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2036 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2037 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2038 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2039
2040 {
2041 PreserveJVMState pjvms(this);
2042 PreserveReexecuteState preexecs(this);
2043 jvms()->set_should_reexecute(true);
2044
2045 set_control(slow_path);
2046 set_i_o(i_o());
2047
2048 builtin_throw(Deoptimization::Reason_intrinsic,
2049 env()->ArithmeticException_instance(),
2050 /*allow_too_many_traps*/ false);
2051 }
2052
2053 set_control(fast_path);
2054 set_result(math);
2055 return true;
2056 }
2057
2058 template <typename OverflowOp>
2059 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2060 typedef typename OverflowOp::MathOp MathOp;
2061
2062 MathOp* mathOp = new MathOp(arg1, arg2);
2063 Node* operation = _gvn.transform( mathOp );
2064 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2065 return inline_math_mathExact(operation, ofcheck);
2066 }
2067
2068 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2069 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2070 }
2071
2072 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2073 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2074 }
2075
2076 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2077 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2078 }
2079
2080 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2081 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2082 }
2083
2084 bool LibraryCallKit::inline_math_negateExactI() {
2085 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2086 }
2087
2088 bool LibraryCallKit::inline_math_negateExactL() {
2089 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2090 }
2091
2092 bool LibraryCallKit::inline_math_multiplyExactI() {
2093 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2094 }
2095
2096 bool LibraryCallKit::inline_math_multiplyExactL() {
2097 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2098 }
2099
2100 bool LibraryCallKit::inline_math_multiplyHigh() {
2101 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2102 return true;
2103 }
2104
2105 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2106 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2107 return true;
2108 }
2109
2110 inline int
2111 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2112 const TypePtr* base_type = TypePtr::NULL_PTR;
2113 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2114 if (base_type == nullptr) {
2115 // Unknown type.
2116 return Type::AnyPtr;
2117 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2118 // Since this is a null+long form, we have to switch to a rawptr.
2119 base = _gvn.transform(new CastX2PNode(offset));
2120 offset = MakeConX(0);
2121 return Type::RawPtr;
2122 } else if (base_type->base() == Type::RawPtr) {
2123 return Type::RawPtr;
2124 } else if (base_type->isa_oopptr()) {
2125 // Base is never null => always a heap address.
2126 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2127 return Type::OopPtr;
2128 }
2129 // Offset is small => always a heap address.
2130 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2131 if (offset_type != nullptr &&
2132 base_type->offset() == 0 && // (should always be?)
2133 offset_type->_lo >= 0 &&
2134 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2135 return Type::OopPtr;
2136 } else if (type == T_OBJECT) {
2137 // off heap access to an oop doesn't make any sense. Has to be on
2138 // heap.
2139 return Type::OopPtr;
2140 }
2141 // Otherwise, it might either be oop+off or null+addr.
2142 return Type::AnyPtr;
2143 } else {
2144 // No information:
2145 return Type::AnyPtr;
2146 }
2147 }
2148
2149 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2150 Node* uncasted_base = base;
2151 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2152 if (kind == Type::RawPtr) {
2153 return off_heap_plus_addr(uncasted_base, offset);
2154 } else if (kind == Type::AnyPtr) {
2155 assert(base == uncasted_base, "unexpected base change");
2156 if (can_cast) {
2157 if (!_gvn.type(base)->speculative_maybe_null() &&
2158 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2159 // According to profiling, this access is always on
2160 // heap. Casting the base to not null and thus avoiding membars
2161 // around the access should allow better optimizations
2162 Node* null_ctl = top();
2163 base = null_check_oop(base, &null_ctl, true, true, true);
2164 assert(null_ctl->is_top(), "no null control here");
2165 return basic_plus_adr(base, offset);
2166 } else if (_gvn.type(base)->speculative_always_null() &&
2167 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2168 // According to profiling, this access is always off
2169 // heap.
2170 base = null_assert(base);
2171 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2172 offset = MakeConX(0);
2173 return off_heap_plus_addr(raw_base, offset);
2174 }
2175 }
2176 // We don't know if it's an on heap or off heap access. Fall back
2177 // to raw memory access.
2178 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2179 return off_heap_plus_addr(raw, offset);
2180 } else {
2181 assert(base == uncasted_base, "unexpected base change");
2182 // We know it's an on heap access so base can't be null
2183 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2184 base = must_be_not_null(base, true);
2185 }
2186 return basic_plus_adr(base, offset);
2187 }
2188 }
2189
2190 //--------------------------inline_number_methods-----------------------------
2191 // inline int Integer.numberOfLeadingZeros(int)
2192 // inline int Long.numberOfLeadingZeros(long)
2193 //
2194 // inline int Integer.numberOfTrailingZeros(int)
2195 // inline int Long.numberOfTrailingZeros(long)
2196 //
2197 // inline int Integer.bitCount(int)
2198 // inline int Long.bitCount(long)
2199 //
2200 // inline char Character.reverseBytes(char)
2201 // inline short Short.reverseBytes(short)
2202 // inline int Integer.reverseBytes(int)
2203 // inline long Long.reverseBytes(long)
2204 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2205 Node* arg = argument(0);
2206 Node* n = nullptr;
2207 switch (id) {
2208 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2209 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2210 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2211 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2212 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2213 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2214 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2215 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2216 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2217 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2218 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2219 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2220 default: fatal_unexpected_iid(id); break;
2221 }
2222 set_result(_gvn.transform(n));
2223 return true;
2224 }
2225
2226 //--------------------------inline_bitshuffle_methods-----------------------------
2227 // inline int Integer.compress(int, int)
2228 // inline int Integer.expand(int, int)
2229 // inline long Long.compress(long, long)
2230 // inline long Long.expand(long, long)
2231 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2232 Node* n = nullptr;
2233 switch (id) {
2234 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2235 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2236 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2237 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2238 default: fatal_unexpected_iid(id); break;
2239 }
2240 set_result(_gvn.transform(n));
2241 return true;
2242 }
2243
2244 //--------------------------inline_number_methods-----------------------------
2245 // inline int Integer.compareUnsigned(int, int)
2246 // inline int Long.compareUnsigned(long, long)
2247 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2248 Node* arg1 = argument(0);
2249 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2250 Node* n = nullptr;
2251 switch (id) {
2252 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2253 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2254 default: fatal_unexpected_iid(id); break;
2255 }
2256 set_result(_gvn.transform(n));
2257 return true;
2258 }
2259
2260 //--------------------------inline_unsigned_divmod_methods-----------------------------
2261 // inline int Integer.divideUnsigned(int, int)
2262 // inline int Integer.remainderUnsigned(int, int)
2263 // inline long Long.divideUnsigned(long, long)
2264 // inline long Long.remainderUnsigned(long, long)
2265 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2266 Node* n = nullptr;
2267 switch (id) {
2268 case vmIntrinsics::_divideUnsigned_i: {
2269 zero_check_int(argument(1));
2270 // Compile-time detect of null-exception
2271 if (stopped()) {
2272 return true; // keep the graph constructed so far
2273 }
2274 n = new UDivINode(control(), argument(0), argument(1));
2275 break;
2276 }
2277 case vmIntrinsics::_divideUnsigned_l: {
2278 zero_check_long(argument(2));
2279 // Compile-time detect of null-exception
2280 if (stopped()) {
2281 return true; // keep the graph constructed so far
2282 }
2283 n = new UDivLNode(control(), argument(0), argument(2));
2284 break;
2285 }
2286 case vmIntrinsics::_remainderUnsigned_i: {
2287 zero_check_int(argument(1));
2288 // Compile-time detect of null-exception
2289 if (stopped()) {
2290 return true; // keep the graph constructed so far
2291 }
2292 n = new UModINode(control(), argument(0), argument(1));
2293 break;
2294 }
2295 case vmIntrinsics::_remainderUnsigned_l: {
2296 zero_check_long(argument(2));
2297 // Compile-time detect of null-exception
2298 if (stopped()) {
2299 return true; // keep the graph constructed so far
2300 }
2301 n = new UModLNode(control(), argument(0), argument(2));
2302 break;
2303 }
2304 default: fatal_unexpected_iid(id); break;
2305 }
2306 set_result(_gvn.transform(n));
2307 return true;
2308 }
2309
2310 //----------------------------inline_unsafe_access----------------------------
2311
2312 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2313 // Attempt to infer a sharper value type from the offset and base type.
2314 ciKlass* sharpened_klass = nullptr;
2315 bool null_free = false;
2316
2317 // See if it is an instance field, with an object type.
2318 if (alias_type->field() != nullptr) {
2319 if (alias_type->field()->type()->is_klass()) {
2320 sharpened_klass = alias_type->field()->type()->as_klass();
2321 null_free = alias_type->field()->is_null_free();
2322 }
2323 }
2324
2325 const TypeOopPtr* result = nullptr;
2326 // See if it is a narrow oop array.
2327 if (adr_type->isa_aryptr()) {
2328 if (adr_type->offset() >= refArrayOopDesc::base_offset_in_bytes()) {
2329 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2330 null_free = adr_type->is_aryptr()->is_null_free();
2331 if (elem_type != nullptr && elem_type->is_loaded()) {
2332 // Sharpen the value type.
2333 result = elem_type;
2334 }
2335 }
2336 }
2337
2338 // The sharpened class might be unloaded if there is no class loader
2339 // contraint in place.
2340 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2341 // Sharpen the value type.
2342 result = TypeOopPtr::make_from_klass(sharpened_klass);
2343 if (null_free) {
2344 result = result->join_speculative(TypePtr::NOTNULL)->is_oopptr();
2345 }
2346 }
2347 if (result != nullptr) {
2348 #ifndef PRODUCT
2349 if (C->print_intrinsics() || C->print_inlining()) {
2350 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2351 tty->print(" sharpened value: "); result->dump(); tty->cr();
2352 }
2353 #endif
2354 }
2355 return result;
2356 }
2357
2358 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2359 switch (kind) {
2360 case Relaxed:
2361 return MO_UNORDERED;
2362 case Opaque:
2363 return MO_RELAXED;
2364 case Acquire:
2365 return MO_ACQUIRE;
2366 case Release:
2367 return MO_RELEASE;
2368 case Volatile:
2369 return MO_SEQ_CST;
2370 default:
2371 ShouldNotReachHere();
2372 return 0;
2373 }
2374 }
2375
2376 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2377 if (callee()->is_static()) return false; // caller must have the capability!
2378 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2379 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2380 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2381 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2382
2383 if (is_reference_type(type)) {
2384 decorators |= ON_UNKNOWN_OOP_REF;
2385 }
2386
2387 if (unaligned) {
2388 decorators |= C2_UNALIGNED;
2389 }
2390
2391 #ifndef PRODUCT
2392 {
2393 ResourceMark rm;
2394 // Check the signatures.
2395 ciSignature* sig = callee()->signature();
2396 #ifdef ASSERT
2397 if (!is_store) {
2398 // Object getReference(Object base, int/long offset), etc.
2399 BasicType rtype = sig->return_type()->basic_type();
2400 assert(rtype == type, "getter must return the expected value");
2401 assert(sig->count() == 2, "oop getter has 2 arguments");
2402 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2403 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2404 } else {
2405 // void putReference(Object base, int/long offset, Object x), etc.
2406 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2407 assert(sig->count() == 3, "oop putter has 3 arguments");
2408 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2409 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2410 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2411 assert(vtype == type, "putter must accept the expected value");
2412 }
2413 #endif // ASSERT
2414 }
2415 #endif //PRODUCT
2416
2417 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2418
2419 Node* receiver = argument(0); // type: oop
2420
2421 // Build address expression.
2422 Node* heap_base_oop = top();
2423
2424 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2425 Node* base = argument(1); // type: oop
2426 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2427 Node* offset = argument(2); // type: long
2428 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2429 // to be plain byte offsets, which are also the same as those accepted
2430 // by oopDesc::field_addr.
2431 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2432 "fieldOffset must be byte-scaled");
2433
2434 if (base->is_InlineType()) {
2435 assert(!is_store, "InlineTypeNodes are non-larval value objects");
2436 InlineTypeNode* vt = base->as_InlineType();
2437 if (offset->is_Con()) {
2438 long off = find_long_con(offset, 0);
2439 ciInlineKlass* vk = vt->type()->inline_klass();
2440 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2441 return false;
2442 }
2443
2444 ciField* field = vk->get_non_flat_field_by_offset(off);
2445 if (field != nullptr) {
2446 BasicType bt = type2field[field->type()->basic_type()];
2447 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2448 bt = T_OBJECT;
2449 }
2450 if (bt == type && !field->is_flat()) {
2451 Node* value = vt->field_value_by_offset(off, false);
2452 const Type* value_type = _gvn.type(value);
2453 if (value_type->is_inlinetypeptr()) {
2454 value = InlineTypeNode::make_from_oop(this, value, value_type->inline_klass());
2455 }
2456 set_result(value);
2457 return true;
2458 }
2459 }
2460 }
2461 {
2462 // Re-execute the unsafe access if allocation triggers deoptimization.
2463 PreserveReexecuteState preexecs(this);
2464 jvms()->set_should_reexecute(true);
2465 vt = vt->buffer(this);
2466 }
2467 base = vt->get_oop();
2468 }
2469
2470 // 32-bit machines ignore the high half!
2471 offset = ConvL2X(offset);
2472
2473 // Save state and restore on bailout
2474 SavedState old_state(this);
2475
2476 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2477 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2478
2479 bool is_non_heap_access = (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR);
2480 if (is_non_heap_access) {
2481 if (type != T_OBJECT) {
2482 decorators |= IN_NATIVE; // off-heap primitive access
2483 } else {
2484 return false; // off-heap oop accesses are not supported
2485 }
2486 } else {
2487 heap_base_oop = base; // on-heap or mixed access
2488 }
2489
2490 // Can base be null? Otherwise, always on-heap access.
2491 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2492
2493 assert(!is_non_heap_access || can_access_non_heap, "sanity"); // is_non_heap_access implies can_access_non_heap
2494
2495 if (!can_access_non_heap) {
2496 decorators |= IN_HEAP;
2497 }
2498
2499 Node* val = is_store ? argument(4) : nullptr;
2500
2501 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2502 if (adr_type == TypePtr::NULL_PTR) {
2503 return false; // off-heap access with zero address
2504 }
2505
2506 // Try to categorize the address.
2507 Compile::AliasType* alias_type = C->alias_type(adr_type);
2508 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2509
2510 assert((alias_type->index() == Compile::AliasIdxRaw) ==
2511 (is_non_heap_access || (can_access_non_heap && alias_type->field() == nullptr)), "wrong alias");
2512
2513 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2514 alias_type->adr_type() == TypeAryPtr::RANGE) {
2515 return false; // not supported
2516 }
2517
2518 bool mismatched = false;
2519 BasicType bt = T_ILLEGAL;
2520 ciField* field = nullptr;
2521 if (adr_type->isa_instptr()) {
2522 const TypeInstPtr* instptr = adr_type->is_instptr();
2523 ciInstanceKlass* k = instptr->instance_klass();
2524 int off = instptr->offset();
2525 if (instptr->const_oop() != nullptr &&
2526 k == ciEnv::current()->Class_klass() &&
2527 instptr->offset() >= (k->size_helper() * wordSize)) {
2528 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2529 field = k->get_field_by_offset(off, true);
2530 } else {
2531 field = k->get_non_flat_field_by_offset(off);
2532 }
2533 if (field != nullptr) {
2534 bt = type2field[field->type()->basic_type()];
2535 }
2536 if (bt != alias_type->basic_type()) {
2537 // Type mismatch. Is it an access to a nested flat field?
2538 field = k->get_field_by_offset(off, false);
2539 if (field != nullptr) {
2540 bt = type2field[field->type()->basic_type()];
2541 }
2542 }
2543 assert(bt == alias_type->basic_type(), "should match");
2544 } else {
2545 bt = alias_type->basic_type();
2546 }
2547
2548 if (bt != T_ILLEGAL) {
2549 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2550 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2551 // Alias type doesn't differentiate between byte[] and boolean[]).
2552 // Use address type to get the element type.
2553 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2554 }
2555 if (is_reference_type(bt, true)) {
2556 // accessing an array field with getReference is not a mismatch
2557 bt = T_OBJECT;
2558 }
2559 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2560 // Don't intrinsify mismatched object accesses
2561 return false;
2562 }
2563 mismatched = (bt != type);
2564 } else if (alias_type->adr_type()->isa_oopptr()) {
2565 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2566 }
2567
2568 old_state.discard();
2569 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2570
2571 if (mismatched) {
2572 decorators |= C2_MISMATCHED;
2573 }
2574
2575 // First guess at the value type.
2576 const Type *value_type = Type::get_const_basic_type(type);
2577
2578 // Figure out the memory ordering.
2579 decorators |= mo_decorator_for_access_kind(kind);
2580
2581 if (!is_store) {
2582 if (type == T_OBJECT) {
2583 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2584 if (tjp != nullptr) {
2585 value_type = tjp;
2586 }
2587 } else if (type == T_BOOLEAN) {
2588 if (mismatched || alias_type->index() == Compile::AliasIdxRaw) {
2589 value_type = TypeInt::UBYTE;
2590 }
2591 }
2592 }
2593
2594 receiver = null_check(receiver);
2595 if (stopped()) {
2596 return true;
2597 }
2598 // Heap pointers get a null-check from the interpreter,
2599 // as a courtesy. However, this is not guaranteed by Unsafe,
2600 // and it is not possible to fully distinguish unintended nulls
2601 // from intended ones in this API.
2602
2603 if (!is_store) {
2604 Node* p = nullptr;
2605 // Try to constant fold a load from a constant field
2606
2607 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !field->is_flat() && !mismatched) {
2608 // final or stable field
2609 p = make_constant_from_field(field, heap_base_oop);
2610 }
2611
2612 if (p == nullptr) { // Could not constant fold the load
2613 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2614 const TypeOopPtr* ptr = value_type->make_oopptr();
2615 if (ptr != nullptr && ptr->is_inlinetypeptr()) {
2616 // Load a non-flattened inline type from memory
2617 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass());
2618 }
2619 }
2620 if (type == T_ADDRESS) {
2621 p = gvn().transform(new CastP2XNode(nullptr, p));
2622 p = ConvX2UL(p);
2623 } else if (type == T_BOOLEAN) {
2624 // Truncate boolean values returned by unsafe operations.
2625 p = gvn().transform(new AndINode(p, gvn().intcon(0x1)));
2626 }
2627 // The load node has the control of the preceding MemBarCPUOrder. All
2628 // following nodes will have the control of the MemBarCPUOrder inserted at
2629 // the end of this method. So, pushing the load onto the stack at a later
2630 // point is fine.
2631 set_result(p);
2632 } else {
2633 if (bt == T_ADDRESS) {
2634 // Repackage the long as a pointer.
2635 val = ConvL2X(val);
2636 val = gvn().transform(new CastX2PNode(val));
2637 }
2638 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2639 }
2640
2641 return true;
2642 }
2643
2644 bool LibraryCallKit::inline_unsafe_flat_access(bool is_store, AccessKind kind) {
2645 #ifdef ASSERT
2646 {
2647 ResourceMark rm;
2648 // Check the signatures.
2649 ciSignature* sig = callee()->signature();
2650 assert(sig->type_at(0)->basic_type() == T_OBJECT, "base should be object, but is %s", type2name(sig->type_at(0)->basic_type()));
2651 assert(sig->type_at(1)->basic_type() == T_LONG, "offset should be long, but is %s", type2name(sig->type_at(1)->basic_type()));
2652 assert(sig->type_at(2)->basic_type() == T_INT, "layout kind should be int, but is %s", type2name(sig->type_at(3)->basic_type()));
2653 assert(sig->type_at(3)->basic_type() == T_OBJECT, "value klass should be object, but is %s", type2name(sig->type_at(4)->basic_type()));
2654 if (is_store) {
2655 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value, but returns %s", type2name(sig->return_type()->basic_type()));
2656 assert(sig->count() == 5, "flat putter should have 5 arguments, but has %d", sig->count());
2657 assert(sig->type_at(4)->basic_type() == T_OBJECT, "put value should be object, but is %s", type2name(sig->type_at(5)->basic_type()));
2658 } else {
2659 assert(sig->return_type()->basic_type() == T_OBJECT, "getter must return an object, but returns %s", type2name(sig->return_type()->basic_type()));
2660 assert(sig->count() == 4, "flat getter should have 4 arguments, but has %d", sig->count());
2661 }
2662 }
2663 #endif // ASSERT
2664
2665 assert(kind == Relaxed, "Only plain accesses for now");
2666 if (callee()->is_static()) {
2667 // caller must have the capability!
2668 return false;
2669 }
2670 C->set_has_unsafe_access(true);
2671
2672 const TypeInstPtr* value_klass_node = _gvn.type(argument(5))->isa_instptr();
2673 if (value_klass_node == nullptr || value_klass_node->const_oop() == nullptr) {
2674 // parameter valueType is not a constant
2675 return false;
2676 }
2677 ciType* mirror_type = value_klass_node->const_oop()->as_instance()->java_mirror_type();
2678 if (!mirror_type->is_inlinetype()) {
2679 // Dead code
2680 return false;
2681 }
2682 ciInlineKlass* value_klass = mirror_type->as_inline_klass();
2683
2684 const TypeInt* layout_type = _gvn.type(argument(4))->isa_int();
2685 if (layout_type == nullptr || !layout_type->is_con()) {
2686 // parameter layoutKind is not a constant
2687 return false;
2688 }
2689 assert(layout_type->get_con() >= static_cast<int>(LayoutKind::REFERENCE) &&
2690 layout_type->get_con() < static_cast<int>(LayoutKind::UNKNOWN),
2691 "invalid layoutKind %d", layout_type->get_con());
2692 LayoutKind layout = static_cast<LayoutKind>(layout_type->get_con());
2693 assert(layout == LayoutKind::REFERENCE || LayoutKindHelper::is_flat(layout),
2694 "unexpected layoutKind %d", layout_type->get_con());
2695
2696 null_check(argument(0));
2697 if (stopped()) {
2698 return true;
2699 }
2700
2701 Node* base = must_be_not_null(argument(1), true);
2702 Node* offset = argument(2);
2703 const Type* base_type = _gvn.type(base);
2704
2705 Node* ptr;
2706 bool immutable_memory = false;
2707 DecoratorSet decorators = C2_UNSAFE_ACCESS | IN_HEAP | MO_UNORDERED;
2708 if (base_type->isa_instptr()) {
2709 const TypeLong* offset_type = _gvn.type(offset)->isa_long();
2710 if (offset_type == nullptr || !offset_type->is_con()) {
2711 // Offset into a non-array should be a constant
2712 decorators |= C2_MISMATCHED;
2713 } else {
2714 int offset_con = checked_cast<int>(offset_type->get_con());
2715 ciInstanceKlass* base_klass = base_type->is_instptr()->instance_klass();
2716 ciField* field = base_klass->get_non_flat_field_by_offset(offset_con);
2717 if (field == nullptr) {
2718 assert(!base_klass->is_final(), "non-existence field at offset %d of class %s", offset_con, base_klass->name()->as_utf8());
2719 decorators |= C2_MISMATCHED;
2720 } else {
2721 assert(field->type() == value_klass, "field at offset %d of %s is of type %s, but valueType is %s",
2722 offset_con, base_klass->name()->as_utf8(), field->type()->name(), value_klass->name()->as_utf8());
2723 immutable_memory = field->is_strict() && field->is_final();
2724
2725 if (base->is_InlineType()) {
2726 assert(!is_store, "Cannot store into a non-larval value object");
2727 set_result(base->as_InlineType()->field_value_by_offset(offset_con, false));
2728 return true;
2729 }
2730 }
2731 }
2732
2733 if (base->is_InlineType()) {
2734 assert(!is_store, "Cannot store into a non-larval value object");
2735 base = base->as_InlineType()->buffer(this, true);
2736 }
2737 ptr = basic_plus_adr(base, ConvL2X(offset));
2738 } else if (base_type->isa_aryptr()) {
2739 decorators |= IS_ARRAY;
2740 if (layout == LayoutKind::REFERENCE) {
2741 if (!base_type->is_aryptr()->is_not_flat()) {
2742 const TypeAryPtr* array_type = base_type->is_aryptr()->cast_to_not_flat();
2743 // TODO 8350865 This should be a CheckCastPP, can we add a test?
2744 Node* new_base = _gvn.transform(new CastPPNode(control(), base, array_type, ConstraintCastNode::DependencyType::NonFloatingNarrowing));
2745 replace_in_map(base, new_base);
2746 base = new_base;
2747 }
2748 ptr = basic_plus_adr(base, ConvL2X(offset));
2749 } else {
2750 if (UseArrayFlattening) {
2751 // Flat array must have an exact type
2752 bool is_null_free = !LayoutKindHelper::is_nullable_flat(layout);
2753 bool is_atomic = LayoutKindHelper::is_atomic_flat(layout);
2754 Node* new_base = cast_to_flat_array_exact(base, value_klass, is_null_free, is_atomic);
2755 replace_in_map(base, new_base);
2756 base = new_base;
2757 ptr = basic_plus_adr(base, ConvL2X(offset));
2758 const TypeAryPtr* ptr_type = _gvn.type(ptr)->is_aryptr();
2759 if (ptr_type->field_offset().get() != 0) {
2760 // TODO 8350865 This should be a CheckCastPP, can we add a test?
2761 ptr = _gvn.transform(new CastPPNode(control(), ptr, ptr_type->with_field_offset(0), ConstraintCastNode::DependencyType::NonFloatingNarrowing));
2762 }
2763 } else {
2764 uncommon_trap(Deoptimization::Reason_intrinsic,
2765 Deoptimization::Action_none);
2766 return true;
2767 }
2768 }
2769 } else {
2770 decorators |= C2_MISMATCHED;
2771 ptr = basic_plus_adr(base, ConvL2X(offset));
2772 }
2773
2774 if (is_store) {
2775 Node* value = argument(6);
2776 const Type* value_type = _gvn.type(value);
2777 if (!value_type->is_inlinetypeptr()) {
2778 value_type = Type::get_const_type(value_klass)->filter_speculative(value_type);
2779 Node* new_value = _gvn.transform(new CheckCastPPNode(control(), value, value_type, ConstraintCastNode::DependencyType::NonFloatingNarrowing));
2780 new_value = InlineTypeNode::make_from_oop(this, new_value, value_klass);
2781 replace_in_map(value, new_value);
2782 value = new_value;
2783 }
2784
2785 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());
2786 if (layout == LayoutKind::REFERENCE) {
2787 const TypePtr* ptr_type = (decorators & C2_MISMATCHED) != 0 ? TypeRawPtr::BOTTOM : _gvn.type(ptr)->is_ptr();
2788 access_store_at(base, ptr, ptr_type, value, value_type, T_OBJECT, decorators);
2789 } else {
2790 bool atomic = LayoutKindHelper::is_atomic_flat(layout);
2791 bool null_free = !LayoutKindHelper::is_nullable_flat(layout);
2792 value->as_InlineType()->store_flat(this, base, ptr, atomic, immutable_memory, null_free, decorators);
2793 }
2794
2795 return true;
2796 } else {
2797 decorators |= (C2_CONTROL_DEPENDENT_LOAD | C2_UNKNOWN_CONTROL_LOAD);
2798 InlineTypeNode* result;
2799 if (layout == LayoutKind::REFERENCE) {
2800 const TypePtr* ptr_type = (decorators & C2_MISMATCHED) != 0 ? TypeRawPtr::BOTTOM : _gvn.type(ptr)->is_ptr();
2801 Node* oop = access_load_at(base, ptr, ptr_type, Type::get_const_type(value_klass), T_OBJECT, decorators);
2802 result = InlineTypeNode::make_from_oop(this, oop, value_klass);
2803 } else {
2804 bool atomic = LayoutKindHelper::is_atomic_flat(layout);
2805 bool null_free = !LayoutKindHelper::is_nullable_flat(layout);
2806 result = InlineTypeNode::make_from_flat(this, value_klass, base, ptr, atomic, immutable_memory, null_free, decorators);
2807 }
2808
2809 set_result(result);
2810 return true;
2811 }
2812 }
2813
2814 //----------------------------inline_unsafe_load_store----------------------------
2815 // This method serves a couple of different customers (depending on LoadStoreKind):
2816 //
2817 // LS_cmp_swap:
2818 //
2819 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2820 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2821 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2822 //
2823 // LS_cmp_swap_weak:
2824 //
2825 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2826 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2827 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2828 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2829 //
2830 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2831 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2832 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2833 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2834 //
2835 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2836 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2837 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2838 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2839 //
2840 // LS_cmp_exchange:
2841 //
2842 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2843 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2844 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2845 //
2846 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2847 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2848 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2849 //
2850 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2851 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2852 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2853 //
2854 // LS_get_add:
2855 //
2856 // int getAndAddInt( Object o, long offset, int delta)
2857 // long getAndAddLong(Object o, long offset, long delta)
2858 //
2859 // LS_get_set:
2860 //
2861 // int getAndSet(Object o, long offset, int newValue)
2862 // long getAndSet(Object o, long offset, long newValue)
2863 // Object getAndSet(Object o, long offset, Object newValue)
2864 //
2865 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2866 // This basic scheme here is the same as inline_unsafe_access, but
2867 // differs in enough details that combining them would make the code
2868 // overly confusing. (This is a true fact! I originally combined
2869 // them, but even I was confused by it!) As much code/comments as
2870 // possible are retained from inline_unsafe_access though to make
2871 // the correspondences clearer. - dl
2872
2873 if (callee()->is_static()) return false; // caller must have the capability!
2874
2875 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2876 decorators |= mo_decorator_for_access_kind(access_kind);
2877
2878 #ifndef PRODUCT
2879 BasicType rtype;
2880 {
2881 ResourceMark rm;
2882 // Check the signatures.
2883 ciSignature* sig = callee()->signature();
2884 rtype = sig->return_type()->basic_type();
2885 switch(kind) {
2886 case LS_get_add:
2887 case LS_get_set: {
2888 // Check the signatures.
2889 #ifdef ASSERT
2890 assert(rtype == type, "get and set must return the expected type");
2891 assert(sig->count() == 3, "get and set has 3 arguments");
2892 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2893 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2894 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2895 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2896 #endif // ASSERT
2897 break;
2898 }
2899 case LS_cmp_swap:
2900 case LS_cmp_swap_weak: {
2901 // Check the signatures.
2902 #ifdef ASSERT
2903 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2904 assert(sig->count() == 4, "CAS has 4 arguments");
2905 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2906 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2907 #endif // ASSERT
2908 break;
2909 }
2910 case LS_cmp_exchange: {
2911 // Check the signatures.
2912 #ifdef ASSERT
2913 assert(rtype == type, "CAS must return the expected type");
2914 assert(sig->count() == 4, "CAS has 4 arguments");
2915 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2916 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2917 #endif // ASSERT
2918 break;
2919 }
2920 default:
2921 ShouldNotReachHere();
2922 }
2923 }
2924 #endif //PRODUCT
2925
2926 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2927
2928 // Get arguments:
2929 Node* receiver = nullptr;
2930 Node* base = nullptr;
2931 Node* offset = nullptr;
2932 Node* oldval = nullptr;
2933 Node* newval = nullptr;
2934 switch(kind) {
2935 case LS_cmp_swap:
2936 case LS_cmp_swap_weak:
2937 case LS_cmp_exchange: {
2938 const bool two_slot_type = type2size[type] == 2;
2939 receiver = argument(0); // type: oop
2940 base = argument(1); // type: oop
2941 offset = argument(2); // type: long
2942 oldval = argument(4); // type: oop, int, or long
2943 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2944 break;
2945 }
2946 case LS_get_add:
2947 case LS_get_set: {
2948 receiver = argument(0); // type: oop
2949 base = argument(1); // type: oop
2950 offset = argument(2); // type: long
2951 oldval = nullptr;
2952 newval = argument(4); // type: oop, int, or long
2953 break;
2954 }
2955 default:
2956 ShouldNotReachHere();
2957 }
2958
2959 // Build field offset expression.
2960 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2961 // to be plain byte offsets, which are also the same as those accepted
2962 // by oopDesc::field_addr.
2963 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2964 // 32-bit machines ignore the high half of long offsets
2965 offset = ConvL2X(offset);
2966 // Save state and restore on bailout
2967 SavedState old_state(this);
2968 Node* adr = make_unsafe_address(base, offset,type, false);
2969 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2970
2971 Compile::AliasType* alias_type = C->alias_type(adr_type);
2972 BasicType bt = alias_type->basic_type();
2973 if (bt != T_ILLEGAL &&
2974 (is_reference_type(bt) != (type == T_OBJECT))) {
2975 // Don't intrinsify mismatched object accesses.
2976 return false;
2977 }
2978
2979 old_state.discard();
2980
2981 // For CAS, unlike inline_unsafe_access, there seems no point in
2982 // trying to refine types. Just use the coarse types here.
2983 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2984 const Type *value_type = Type::get_const_basic_type(type);
2985
2986 switch (kind) {
2987 case LS_get_set:
2988 case LS_cmp_exchange: {
2989 if (type == T_OBJECT) {
2990 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2991 if (tjp != nullptr) {
2992 value_type = tjp;
2993 }
2994 }
2995 break;
2996 }
2997 case LS_cmp_swap:
2998 case LS_cmp_swap_weak:
2999 case LS_get_add:
3000 break;
3001 default:
3002 ShouldNotReachHere();
3003 }
3004
3005 // Null check receiver.
3006 receiver = null_check(receiver);
3007 if (stopped()) {
3008 return true;
3009 }
3010
3011 int alias_idx = C->get_alias_index(adr_type);
3012
3013 if (is_reference_type(type)) {
3014 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
3015
3016 if (oldval != nullptr && oldval->is_InlineType()) {
3017 // Re-execute the unsafe access if allocation triggers deoptimization.
3018 PreserveReexecuteState preexecs(this);
3019 jvms()->set_should_reexecute(true);
3020 oldval = oldval->as_InlineType()->buffer(this)->get_oop();
3021 }
3022 if (newval != nullptr && newval->is_InlineType()) {
3023 // Re-execute the unsafe access if allocation triggers deoptimization.
3024 PreserveReexecuteState preexecs(this);
3025 jvms()->set_should_reexecute(true);
3026 newval = newval->as_InlineType()->buffer(this)->get_oop();
3027 }
3028
3029 // Transformation of a value which could be null pointer (CastPP #null)
3030 // could be delayed during Parse (for example, in adjust_map_after_if()).
3031 // Execute transformation here to avoid barrier generation in such case.
3032 if (_gvn.type(newval) == TypePtr::NULL_PTR)
3033 newval = _gvn.makecon(TypePtr::NULL_PTR);
3034
3035 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
3036 // Refine the value to a null constant, when it is known to be null
3037 oldval = _gvn.makecon(TypePtr::NULL_PTR);
3038 }
3039 }
3040
3041 Node* result = nullptr;
3042 switch (kind) {
3043 case LS_cmp_exchange: {
3044 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
3045 oldval, newval, value_type, type, decorators);
3046 break;
3047 }
3048 case LS_cmp_swap_weak:
3049 decorators |= C2_WEAK_CMPXCHG;
3050 case LS_cmp_swap: {
3051 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
3052 oldval, newval, value_type, type, decorators);
3053 break;
3054 }
3055 case LS_get_set: {
3056 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
3057 newval, value_type, type, decorators);
3058 break;
3059 }
3060 case LS_get_add: {
3061 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
3062 newval, value_type, type, decorators);
3063 break;
3064 }
3065 default:
3066 ShouldNotReachHere();
3067 }
3068
3069 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3070 set_result(result);
3071 return true;
3072 }
3073
3074 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3075 // Regardless of form, don't allow previous ld/st to move down,
3076 // then issue acquire, release, or volatile mem_bar.
3077 insert_mem_bar(Op_MemBarCPUOrder);
3078 switch(id) {
3079 case vmIntrinsics::_loadFence:
3080 insert_mem_bar(Op_LoadFence);
3081 return true;
3082 case vmIntrinsics::_storeFence:
3083 insert_mem_bar(Op_StoreFence);
3084 return true;
3085 case vmIntrinsics::_storeStoreFence:
3086 insert_mem_bar(Op_StoreStoreFence);
3087 return true;
3088 case vmIntrinsics::_fullFence:
3089 insert_mem_bar(Op_MemBarFull);
3090 return true;
3091 default:
3092 fatal_unexpected_iid(id);
3093 return false;
3094 }
3095 }
3096
3097 // private native int arrayInstanceBaseOffset0(Object[] array);
3098 bool LibraryCallKit::inline_arrayInstanceBaseOffset() {
3099 Node* array = argument(1);
3100 Node* klass_node = load_object_klass(array);
3101
3102 jint layout_con = Klass::_lh_neutral_value;
3103 Node* layout_val = get_layout_helper(klass_node, layout_con);
3104 int layout_is_con = (layout_val == nullptr);
3105
3106 Node* header_size = nullptr;
3107 if (layout_is_con) {
3108 int hsize = Klass::layout_helper_header_size(layout_con);
3109 header_size = intcon(hsize);
3110 } else {
3111 Node* hss = intcon(Klass::_lh_header_size_shift);
3112 Node* hsm = intcon(Klass::_lh_header_size_mask);
3113 header_size = _gvn.transform(new URShiftINode(layout_val, hss));
3114 header_size = _gvn.transform(new AndINode(header_size, hsm));
3115 }
3116 set_result(header_size);
3117 return true;
3118 }
3119
3120 // private native int arrayInstanceIndexScale0(Object[] array);
3121 bool LibraryCallKit::inline_arrayInstanceIndexScale() {
3122 Node* array = argument(1);
3123 Node* klass_node = load_object_klass(array);
3124
3125 jint layout_con = Klass::_lh_neutral_value;
3126 Node* layout_val = get_layout_helper(klass_node, layout_con);
3127 int layout_is_con = (layout_val == nullptr);
3128
3129 Node* element_size = nullptr;
3130 if (layout_is_con) {
3131 int log_element_size = Klass::layout_helper_log2_element_size(layout_con);
3132 int elem_size = 1 << log_element_size;
3133 element_size = intcon(elem_size);
3134 } else {
3135 Node* ess = intcon(Klass::_lh_log2_element_size_shift);
3136 Node* esm = intcon(Klass::_lh_log2_element_size_mask);
3137 Node* log_element_size = _gvn.transform(new URShiftINode(layout_val, ess));
3138 log_element_size = _gvn.transform(new AndINode(log_element_size, esm));
3139 element_size = _gvn.transform(new LShiftINode(intcon(1), log_element_size));
3140 }
3141 set_result(element_size);
3142 return true;
3143 }
3144
3145 // private native int arrayLayout0(Object[] array);
3146 bool LibraryCallKit::inline_arrayLayout() {
3147 RegionNode* region = new RegionNode(2);
3148 Node* phi = new PhiNode(region, TypeInt::POS);
3149
3150 Node* array = argument(1);
3151 Node* klass_node = load_object_klass(array);
3152 generate_refArray_guard(klass_node, region);
3153 if (region->req() == 3) {
3154 phi->add_req(intcon((jint)LayoutKind::REFERENCE));
3155 }
3156
3157 int layout_kind_offset = in_bytes(FlatArrayKlass::layout_kind_offset());
3158 Node* layout_kind_addr = basic_plus_adr(top(), klass_node, layout_kind_offset);
3159 Node* layout_kind = make_load(nullptr, layout_kind_addr, TypeInt::POS, T_INT, MemNode::unordered);
3160
3161 region->init_req(1, control());
3162 phi->init_req(1, layout_kind);
3163
3164 set_control(_gvn.transform(region));
3165 set_result(_gvn.transform(phi));
3166 return true;
3167 }
3168
3169 // private native int[] getFieldMap0(Class <?> c);
3170 // int offset = c._klass._acmp_maps_offset;
3171 // return (int[])c.obj_field(offset);
3172 bool LibraryCallKit::inline_getFieldMap() {
3173 Node* mirror = argument(1);
3174 Node* klass = load_klass_from_mirror(mirror, false, nullptr, 0);
3175
3176 int field_map_offset_offset = in_bytes(InstanceKlass::acmp_maps_offset_offset());
3177 Node* field_map_offset_addr = basic_plus_adr(top(), klass, field_map_offset_offset);
3178 Node* field_map_offset = make_load(nullptr, field_map_offset_addr, TypeInt::INT, T_INT, MemNode::unordered);
3179 field_map_offset = _gvn.transform(ConvI2L(field_map_offset));
3180
3181 Node* map_addr = basic_plus_adr(mirror, field_map_offset);
3182 const TypeAryPtr* val_type = TypeAryPtr::INTS->cast_to_ptr_type(TypePtr::NotNull)->with_offset(0);
3183 Node* map = access_load_at(mirror, map_addr, TypeAryPtr::INTS, val_type, T_ARRAY, IN_HEAP | MO_UNORDERED);
3184
3185 set_result(map);
3186 return true;
3187 }
3188
3189 bool LibraryCallKit::inline_onspinwait() {
3190 insert_mem_bar(Op_OnSpinWait);
3191 return true;
3192 }
3193
3194 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3195 if (!kls->is_Con()) {
3196 return true;
3197 }
3198 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
3199 if (klsptr == nullptr) {
3200 return true;
3201 }
3202 ciInstanceKlass* ik = klsptr->instance_klass();
3203 // don't need a guard for a klass that is already initialized
3204 return !ik->is_initialized();
3205 }
3206
3207 //----------------------------inline_unsafe_writeback0-------------------------
3208 // public native void Unsafe.writeback0(long address)
3209 bool LibraryCallKit::inline_unsafe_writeback0() {
3210 if (!Matcher::has_match_rule(Op_CacheWB)) {
3211 return false;
3212 }
3213 #ifndef PRODUCT
3214 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
3215 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
3216 ciSignature* sig = callee()->signature();
3217 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
3218 #endif
3219 null_check_receiver(); // null-check, then ignore
3220 Node *addr = argument(1);
3221 addr = new CastX2PNode(addr);
3222 addr = _gvn.transform(addr);
3223 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
3224 flush = _gvn.transform(flush);
3225 set_memory(flush, TypeRawPtr::BOTTOM);
3226 return true;
3227 }
3228
3229 //----------------------------inline_unsafe_writeback0-------------------------
3230 // public native void Unsafe.writeback0(long address)
3231 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
3232 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
3233 return false;
3234 }
3235 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
3236 return false;
3237 }
3238 #ifndef PRODUCT
3239 assert(Matcher::has_match_rule(Op_CacheWB),
3240 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
3241 : "found match rule for CacheWBPostSync but not CacheWB"));
3242
3243 #endif
3244 null_check_receiver(); // null-check, then ignore
3245 Node *sync;
3246 if (is_pre) {
3247 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3248 } else {
3249 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3250 }
3251 sync = _gvn.transform(sync);
3252 set_memory(sync, TypeRawPtr::BOTTOM);
3253 return true;
3254 }
3255
3256 //----------------------------inline_unsafe_allocate---------------------------
3257 // public native Object Unsafe.allocateInstance(Class<?> cls);
3258 bool LibraryCallKit::inline_unsafe_allocate() {
3259
3260 #if INCLUDE_JVMTI
3261 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3262 return false;
3263 }
3264 #endif //INCLUDE_JVMTI
3265
3266 if (callee()->is_static()) return false; // caller must have the capability!
3267
3268 null_check_receiver(); // null-check, then ignore
3269 Node* cls = null_check(argument(1));
3270 if (stopped()) return true;
3271
3272 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
3273 kls = null_check(kls);
3274 if (stopped()) return true; // argument was like int.class
3275
3276 #if INCLUDE_JVMTI
3277 // Don't try to access new allocated obj in the intrinsic.
3278 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
3279 // Deoptimize and allocate in interpreter instead.
3280 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
3281 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
3282 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
3283 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
3284 {
3285 BuildCutout unless(this, tst, PROB_MAX);
3286 uncommon_trap(Deoptimization::Reason_intrinsic,
3287 Deoptimization::Action_make_not_entrant);
3288 }
3289 if (stopped()) {
3290 return true;
3291 }
3292 #endif //INCLUDE_JVMTI
3293
3294 Node* test = nullptr;
3295 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3296 // Note: The argument might still be an illegal value like
3297 // Serializable.class or Object[].class. The runtime will handle it.
3298 // But we must make an explicit check for initialization.
3299 Node* insp = off_heap_plus_addr(kls, in_bytes(InstanceKlass::init_state_offset()));
3300 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3301 // can generate code to load it as unsigned byte.
3302 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
3303 Node* bits = intcon(InstanceKlass::fully_initialized);
3304 test = _gvn.transform(new SubINode(inst, bits));
3305 // The 'test' is non-zero if we need to take a slow path.
3306 }
3307 Node* obj = new_instance(kls, test);
3308 set_result(obj);
3309 return true;
3310 }
3311
3312 //------------------------inline_native_time_funcs--------------
3313 // inline code for System.currentTimeMillis() and System.nanoTime()
3314 // these have the same type and signature
3315 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3316 const TypeFunc* tf = OptoRuntime::void_long_Type();
3317 const TypePtr* no_memory_effects = nullptr;
3318 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3319 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3320 #ifdef ASSERT
3321 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3322 assert(value_top == top(), "second value must be top");
3323 #endif
3324 set_result(value);
3325 return true;
3326 }
3327
3328 //--------------------inline_native_vthread_start_transition--------------------
3329 // inline void startTransition(boolean is_mount);
3330 // inline void startFinalTransition();
3331 // Pseudocode of implementation:
3332 //
3333 // java_lang_Thread::set_is_in_vthread_transition(vthread, true);
3334 // carrier->set_is_in_vthread_transition(true);
3335 // OrderAccess::storeload();
3336 // int disable_requests = java_lang_Thread::vthread_transition_disable_count(vthread)
3337 // + global_vthread_transition_disable_count();
3338 // if (disable_requests > 0) {
3339 // slow path: runtime call
3340 // }
3341 bool LibraryCallKit::inline_native_vthread_start_transition(address funcAddr, const char* funcName, bool is_final_transition) {
3342 Node* vt_oop = must_be_not_null(argument(0), true); // VirtualThread this argument
3343 IdealKit ideal(this);
3344
3345 Node* thread = ideal.thread();
3346 Node* jt_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3347 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3348 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3349 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3350 insert_mem_bar(Op_MemBarStoreLoad);
3351 ideal.sync_kit(this);
3352
3353 Node* global_disable_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::global_vthread_transition_disable_count_address()));
3354 Node* global_disable = ideal.load(ideal.ctrl(), global_disable_addr, TypeInt::INT, T_INT, Compile::AliasIdxRaw, true /*require_atomic_access*/);
3355 Node* vt_disable_addr = basic_plus_adr(vt_oop, java_lang_Thread::vthread_transition_disable_count_offset());
3356 const TypePtr* vt_disable_addr_t = _gvn.type(vt_disable_addr)->is_ptr();
3357 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*/);
3358 Node* disabled = _gvn.transform(new AddINode(global_disable, vt_disable));
3359
3360 ideal.if_then(disabled, BoolTest::ne, ideal.ConI(0)); {
3361 sync_kit(ideal);
3362 Node* is_mount = is_final_transition ? ideal.ConI(0) : argument(1);
3363 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3364 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3365 ideal.sync_kit(this);
3366 }
3367 ideal.end_if();
3368
3369 final_sync(ideal);
3370 return true;
3371 }
3372
3373 bool LibraryCallKit::inline_native_vthread_end_transition(address funcAddr, const char* funcName, bool is_first_transition) {
3374 Node* vt_oop = must_be_not_null(argument(0), true); // VirtualThread this argument
3375 IdealKit ideal(this);
3376
3377 Node* _notify_jvmti_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::notify_jvmti_events_address()));
3378 Node* _notify_jvmti = ideal.load(ideal.ctrl(), _notify_jvmti_addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3379
3380 ideal.if_then(_notify_jvmti, BoolTest::eq, ideal.ConI(1)); {
3381 sync_kit(ideal);
3382 Node* is_mount = is_first_transition ? ideal.ConI(1) : argument(1);
3383 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3384 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3385 ideal.sync_kit(this);
3386 } ideal.else_(); {
3387 Node* thread = ideal.thread();
3388 Node* jt_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3389 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3390
3391 sync_kit(ideal);
3392 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3393 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3394 ideal.sync_kit(this);
3395 } ideal.end_if();
3396
3397 final_sync(ideal);
3398 return true;
3399 }
3400
3401 #if INCLUDE_JVMTI
3402
3403 // Always update the is_disable_suspend bit.
3404 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3405 if (!DoJVMTIVirtualThreadTransitions) {
3406 return true;
3407 }
3408 IdealKit ideal(this);
3409
3410 {
3411 // unconditionally update the is_disable_suspend bit in current JavaThread
3412 Node* thread = ideal.thread();
3413 Node* arg = argument(0); // argument for notification
3414 Node* addr = off_heap_plus_addr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3415 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3416
3417 sync_kit(ideal);
3418 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3419 ideal.sync_kit(this);
3420 }
3421 final_sync(ideal);
3422
3423 return true;
3424 }
3425
3426 #endif // INCLUDE_JVMTI
3427
3428 #ifdef JFR_HAVE_INTRINSICS
3429
3430 /**
3431 * if oop->klass != null
3432 * // normal class
3433 * epoch = _epoch_state ? 2 : 1
3434 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3435 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3436 * }
3437 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3438 * else
3439 * // primitive class
3440 * if oop->array_klass != null
3441 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3442 * else
3443 * id = LAST_TYPE_ID + 1 // void class path
3444 * if (!signaled)
3445 * signaled = true
3446 */
3447 bool LibraryCallKit::inline_native_classID() {
3448 Node* cls = argument(0);
3449
3450 IdealKit ideal(this);
3451 #define __ ideal.
3452 IdealVariable result(ideal); __ declarations_done();
3453 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3454 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3455 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3456
3457
3458 __ if_then(kls, BoolTest::ne, null()); {
3459 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3460 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3461
3462 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3463 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3464 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3465 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3466 mask = _gvn.transform(new OrLNode(mask, epoch));
3467 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3468
3469 float unlikely = PROB_UNLIKELY(0.999);
3470 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3471 sync_kit(ideal);
3472 make_runtime_call(RC_LEAF,
3473 OptoRuntime::class_id_load_barrier_Type(),
3474 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3475 "class id load barrier",
3476 TypePtr::BOTTOM,
3477 kls);
3478 ideal.sync_kit(this);
3479 } __ end_if();
3480
3481 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3482 } __ else_(); {
3483 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3484 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3485 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3486 __ if_then(array_kls, BoolTest::ne, null()); {
3487 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3488 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3489 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3490 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3491 } __ else_(); {
3492 // void class case
3493 ideal.set(result, longcon(LAST_TYPE_ID + 1));
3494 } __ end_if();
3495
3496 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3497 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3498 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3499 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3500 } __ end_if();
3501 } __ end_if();
3502
3503 final_sync(ideal);
3504 set_result(ideal.value(result));
3505 #undef __
3506 return true;
3507 }
3508
3509 //------------------------inline_native_jvm_commit------------------
3510 bool LibraryCallKit::inline_native_jvm_commit() {
3511 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3512
3513 // Save input memory and i_o state.
3514 Node* input_memory_state = reset_memory();
3515 set_all_memory(input_memory_state);
3516 Node* input_io_state = i_o();
3517
3518 // TLS.
3519 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3520 // Jfr java buffer.
3521 Node* java_buffer_offset = _gvn.transform(AddPNode::make_off_heap(tls_ptr, MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR))));
3522 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3523 Node* java_buffer_pos_offset = _gvn.transform(AddPNode::make_off_heap(java_buffer, MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET))));
3524
3525 // Load the current value of the notified field in the JfrThreadLocal.
3526 Node* notified_offset = off_heap_plus_addr(tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3527 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3528
3529 // Test for notification.
3530 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3531 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3532 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3533
3534 // True branch, is notified.
3535 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3536 set_control(is_notified);
3537
3538 // Reset notified state.
3539 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3540 Node* notified_reset_memory = reset_memory();
3541
3542 // 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.
3543 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3544 // Convert the machine-word to a long.
3545 Node* current_pos = ConvX2L(current_pos_X);
3546
3547 // False branch, not notified.
3548 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3549 set_control(not_notified);
3550 set_all_memory(input_memory_state);
3551
3552 // Arg is the next position as a long.
3553 Node* arg = argument(0);
3554 // Convert long to machine-word.
3555 Node* next_pos_X = ConvL2X(arg);
3556
3557 // Store the next_position to the underlying jfr java buffer.
3558 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3559
3560 Node* commit_memory = reset_memory();
3561 set_all_memory(commit_memory);
3562
3563 // 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.
3564 Node* java_buffer_flags_offset = _gvn.transform(AddPNode::make_off_heap(java_buffer, MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET))));
3565 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3566 Node* lease_constant = _gvn.intcon(4);
3567
3568 // And flags with lease constant.
3569 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3570
3571 // Branch on lease to conditionalize returning the leased java buffer.
3572 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3573 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3574 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3575
3576 // False branch, not a lease.
3577 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3578
3579 // True branch, is lease.
3580 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3581 set_control(is_lease);
3582
3583 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3584 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3585 OptoRuntime::void_void_Type(),
3586 SharedRuntime::jfr_return_lease(),
3587 "return_lease", TypePtr::BOTTOM);
3588 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3589
3590 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3591 record_for_igvn(lease_compare_rgn);
3592 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3593 record_for_igvn(lease_compare_mem);
3594 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3595 record_for_igvn(lease_compare_io);
3596 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3597 record_for_igvn(lease_result_value);
3598
3599 // Update control and phi nodes.
3600 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3601 lease_compare_rgn->init_req(_false_path, not_lease);
3602
3603 lease_compare_mem->init_req(_true_path, reset_memory());
3604 lease_compare_mem->init_req(_false_path, commit_memory);
3605
3606 lease_compare_io->init_req(_true_path, i_o());
3607 lease_compare_io->init_req(_false_path, input_io_state);
3608
3609 lease_result_value->init_req(_true_path, _gvn.longcon(0)); // if the lease was returned, return 0L.
3610 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3611
3612 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3613 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3614 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3615 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3616
3617 // Update control and phi nodes.
3618 result_rgn->init_req(_true_path, is_notified);
3619 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3620
3621 result_mem->init_req(_true_path, notified_reset_memory);
3622 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3623
3624 result_io->init_req(_true_path, input_io_state);
3625 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3626
3627 result_value->init_req(_true_path, current_pos);
3628 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3629
3630 // Set output state.
3631 set_control(_gvn.transform(result_rgn));
3632 set_all_memory(_gvn.transform(result_mem));
3633 set_i_o(_gvn.transform(result_io));
3634 set_result(result_rgn, result_value);
3635 return true;
3636 }
3637
3638 /*
3639 * The intrinsic is a model of this pseudo-code:
3640 *
3641 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3642 * jobject h_event_writer = tl->java_event_writer();
3643 * if (h_event_writer == nullptr) {
3644 * return nullptr;
3645 * }
3646 * oop threadObj = Thread::threadObj();
3647 * oop vthread = java_lang_Thread::vthread(threadObj);
3648 * traceid tid;
3649 * bool pinVirtualThread;
3650 * bool excluded;
3651 * if (vthread != threadObj) { // i.e. current thread is virtual
3652 * tid = java_lang_Thread::tid(vthread);
3653 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3654 * pinVirtualThread = VMContinuations;
3655 * excluded = vthread_epoch_raw & excluded_mask;
3656 * if (!excluded) {
3657 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3658 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3659 * if (vthread_epoch != current_epoch) {
3660 * write_checkpoint();
3661 * }
3662 * }
3663 * } else {
3664 * tid = java_lang_Thread::tid(threadObj);
3665 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3666 * pinVirtualThread = false;
3667 * excluded = thread_epoch_raw & excluded_mask;
3668 * }
3669 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3670 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3671 * if (tid_in_event_writer != tid) {
3672 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3673 * setField(event_writer, "excluded", excluded);
3674 * setField(event_writer, "threadID", tid);
3675 * }
3676 * return event_writer
3677 */
3678 bool LibraryCallKit::inline_native_getEventWriter() {
3679 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3680
3681 // Save input memory and i_o state.
3682 Node* input_memory_state = reset_memory();
3683 set_all_memory(input_memory_state);
3684 Node* input_io_state = i_o();
3685
3686 // The most significant bit of the u2 is used to denote thread exclusion
3687 Node* excluded_shift = _gvn.intcon(15);
3688 Node* excluded_mask = _gvn.intcon(1 << 15);
3689 // The epoch generation is the range [1-32767]
3690 Node* epoch_mask = _gvn.intcon(32767);
3691
3692 // TLS
3693 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3694
3695 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3696 Node* jobj_ptr = off_heap_plus_addr(tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3697
3698 // Load the eventwriter jobject handle.
3699 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3700
3701 // Null check the jobject handle.
3702 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3703 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3704 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3705
3706 // False path, jobj is null.
3707 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3708
3709 // True path, jobj is not null.
3710 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3711
3712 set_control(jobj_is_not_null);
3713
3714 // Load the threadObj for the CarrierThread.
3715 Node* threadObj = generate_current_thread(tls_ptr);
3716
3717 // Load the vthread.
3718 Node* vthread = generate_virtual_thread(tls_ptr);
3719
3720 // If vthread != threadObj, this is a virtual thread.
3721 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3722 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3723 IfNode* iff_vthread_not_equal_threadObj =
3724 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3725
3726 // False branch, fallback to threadObj.
3727 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3728 set_control(vthread_equal_threadObj);
3729
3730 // Load the tid field from the vthread object.
3731 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3732
3733 // Load the raw epoch value from the threadObj.
3734 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3735 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3736 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3737 TypeInt::CHAR, T_CHAR,
3738 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3739
3740 // Mask off the excluded information from the epoch.
3741 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3742
3743 // True branch, this is a virtual thread.
3744 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3745 set_control(vthread_not_equal_threadObj);
3746
3747 // Load the tid field from the vthread object.
3748 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3749
3750 // Continuation support determines if a virtual thread should be pinned.
3751 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3752 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3753
3754 // Load the raw epoch value from the vthread.
3755 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3756 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3757 TypeInt::CHAR, T_CHAR,
3758 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3759
3760 // Mask off the excluded information from the epoch.
3761 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, excluded_mask));
3762
3763 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3764 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, excluded_mask));
3765 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3766 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3767
3768 // False branch, vthread is excluded, no need to write epoch info.
3769 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3770
3771 // True branch, vthread is included, update epoch info.
3772 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3773 set_control(included);
3774
3775 // Get epoch value.
3776 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, epoch_mask));
3777
3778 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3779 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3780 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3781
3782 // Compare the epoch in the vthread to the current epoch generation.
3783 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3784 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3785 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3786
3787 // False path, epoch is equal, checkpoint information is valid.
3788 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3789
3790 // True path, epoch is not equal, write a checkpoint for the vthread.
3791 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3792
3793 set_control(epoch_is_not_equal);
3794
3795 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3796 // The call also updates the native thread local thread id and the vthread with the current epoch.
3797 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3798 OptoRuntime::jfr_write_checkpoint_Type(),
3799 SharedRuntime::jfr_write_checkpoint(),
3800 "write_checkpoint", TypePtr::BOTTOM);
3801 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3802
3803 // vthread epoch != current epoch
3804 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3805 record_for_igvn(epoch_compare_rgn);
3806 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3807 record_for_igvn(epoch_compare_mem);
3808 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3809 record_for_igvn(epoch_compare_io);
3810
3811 // Update control and phi nodes.
3812 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3813 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3814 epoch_compare_mem->init_req(_true_path, reset_memory());
3815 epoch_compare_mem->init_req(_false_path, input_memory_state);
3816 epoch_compare_io->init_req(_true_path, i_o());
3817 epoch_compare_io->init_req(_false_path, input_io_state);
3818
3819 // excluded != true
3820 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3821 record_for_igvn(exclude_compare_rgn);
3822 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3823 record_for_igvn(exclude_compare_mem);
3824 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3825 record_for_igvn(exclude_compare_io);
3826
3827 // Update control and phi nodes.
3828 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3829 exclude_compare_rgn->init_req(_false_path, excluded);
3830 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3831 exclude_compare_mem->init_req(_false_path, input_memory_state);
3832 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3833 exclude_compare_io->init_req(_false_path, input_io_state);
3834
3835 // vthread != threadObj
3836 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3837 record_for_igvn(vthread_compare_rgn);
3838 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3839 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3840 record_for_igvn(vthread_compare_io);
3841 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3842 record_for_igvn(tid);
3843 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3844 record_for_igvn(exclusion);
3845 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3846 record_for_igvn(pinVirtualThread);
3847
3848 // Update control and phi nodes.
3849 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3850 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3851 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3852 vthread_compare_mem->init_req(_false_path, input_memory_state);
3853 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3854 vthread_compare_io->init_req(_false_path, input_io_state);
3855 tid->init_req(_true_path, vthread_tid);
3856 tid->init_req(_false_path, thread_obj_tid);
3857 exclusion->init_req(_true_path, vthread_is_excluded);
3858 exclusion->init_req(_false_path, threadObj_is_excluded);
3859 pinVirtualThread->init_req(_true_path, continuation_support);
3860 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3861
3862 // Update branch state.
3863 set_control(_gvn.transform(vthread_compare_rgn));
3864 set_all_memory(_gvn.transform(vthread_compare_mem));
3865 set_i_o(_gvn.transform(vthread_compare_io));
3866
3867 // Load the event writer oop by dereferencing the jobject handle.
3868 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3869 assert(klass_EventWriter->is_loaded(), "invariant");
3870 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3871 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3872 const TypeOopPtr* const xtype = aklass->as_exact_instance_type();
3873 Node* jobj_untagged = _gvn.transform(AddPNode::make_off_heap(jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3874 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3875
3876 // Load the current thread id from the event writer object.
3877 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3878 // Get the field offset to, conditionally, store an updated tid value later.
3879 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3880 // Get the field offset to, conditionally, store an updated exclusion value later.
3881 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3882 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3883 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3884
3885 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3886 record_for_igvn(event_writer_tid_compare_rgn);
3887 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3888 record_for_igvn(event_writer_tid_compare_mem);
3889 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3890 record_for_igvn(event_writer_tid_compare_io);
3891
3892 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3893 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3894 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3895 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3896
3897 // False path, tids are the same.
3898 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3899
3900 // True path, tid is not equal, need to update the tid in the event writer.
3901 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3902 record_for_igvn(tid_is_not_equal);
3903
3904 // Store the pin state to the event writer.
3905 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3906
3907 // Store the exclusion state to the event writer.
3908 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3909 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3910
3911 // Store the tid to the event writer.
3912 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3913
3914 // Update control and phi nodes.
3915 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3916 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3917 event_writer_tid_compare_mem->init_req(_true_path, reset_memory());
3918 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3919 event_writer_tid_compare_io->init_req(_true_path, i_o());
3920 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3921
3922 // Result of top level CFG, Memory, IO and Value.
3923 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3924 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3925 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3926 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3927
3928 // Result control.
3929 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3930 result_rgn->init_req(_false_path, jobj_is_null);
3931
3932 // Result memory.
3933 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3934 result_mem->init_req(_false_path, input_memory_state);
3935
3936 // Result IO.
3937 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3938 result_io->init_req(_false_path, input_io_state);
3939
3940 // Result value.
3941 result_value->init_req(_true_path, event_writer); // return event writer oop
3942 result_value->init_req(_false_path, null()); // return null
3943
3944 // Set output state.
3945 set_control(_gvn.transform(result_rgn));
3946 set_all_memory(_gvn.transform(result_mem));
3947 set_i_o(_gvn.transform(result_io));
3948 set_result(result_rgn, result_value);
3949 return true;
3950 }
3951
3952 /*
3953 * The intrinsic is a model of this pseudo-code:
3954 *
3955 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3956 * if (carrierThread != thread) { // is virtual thread
3957 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3958 * bool excluded = vthread_epoch_raw & excluded_mask;
3959 * AtomicAccess::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3960 * AtomicAccess::store(&tl->_contextual_thread_excluded, is_excluded);
3961 * if (!excluded) {
3962 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3963 * AtomicAccess::store(&tl->_vthread_epoch, vthread_epoch);
3964 * }
3965 * AtomicAccess::release_store(&tl->_vthread, true);
3966 * return;
3967 * }
3968 * AtomicAccess::release_store(&tl->_vthread, false);
3969 */
3970 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3971 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3972
3973 Node* input_memory_state = reset_memory();
3974 set_all_memory(input_memory_state);
3975
3976 // The most significant bit of the u2 is used to denote thread exclusion
3977 Node* excluded_mask = _gvn.intcon(1 << 15);
3978 // The epoch generation is the range [1-32767]
3979 Node* epoch_mask = _gvn.intcon(32767);
3980
3981 Node* const carrierThread = generate_current_thread(jt);
3982 // If thread != carrierThread, this is a virtual thread.
3983 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3984 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3985 IfNode* iff_thread_not_equal_carrierThread =
3986 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3987
3988 Node* vthread_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3989
3990 // False branch, is carrierThread.
3991 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3992 // Store release
3993 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3994
3995 set_all_memory(input_memory_state);
3996
3997 // True branch, is virtual thread.
3998 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3999 set_control(thread_not_equal_carrierThread);
4000
4001 // Load the raw epoch value from the vthread.
4002 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
4003 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
4004 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
4005
4006 // Mask off the excluded information from the epoch.
4007 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, excluded_mask));
4008
4009 // Load the tid field from the thread.
4010 Node* tid = load_field_from_object(thread, "tid", "J");
4011
4012 // Store the vthread tid to the jfr thread local.
4013 Node* thread_id_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
4014 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
4015
4016 // Branch is_excluded to conditionalize updating the epoch .
4017 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, excluded_mask));
4018 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
4019 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
4020
4021 // True branch, vthread is excluded, no need to write epoch info.
4022 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
4023 set_control(excluded);
4024 Node* vthread_is_excluded = _gvn.intcon(1);
4025
4026 // False branch, vthread is included, update epoch info.
4027 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
4028 set_control(included);
4029 Node* vthread_is_included = _gvn.intcon(0);
4030
4031 // Get epoch value.
4032 Node* epoch = _gvn.transform(new AndINode(epoch_raw, epoch_mask));
4033
4034 // Store the vthread epoch to the jfr thread local.
4035 Node* vthread_epoch_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
4036 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
4037
4038 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
4039 record_for_igvn(excluded_rgn);
4040 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
4041 record_for_igvn(excluded_mem);
4042 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
4043 record_for_igvn(exclusion);
4044
4045 // Merge the excluded control and memory.
4046 excluded_rgn->init_req(_true_path, excluded);
4047 excluded_rgn->init_req(_false_path, included);
4048 excluded_mem->init_req(_true_path, tid_memory);
4049 excluded_mem->init_req(_false_path, included_memory);
4050 exclusion->init_req(_true_path, vthread_is_excluded);
4051 exclusion->init_req(_false_path, vthread_is_included);
4052
4053 // Set intermediate state.
4054 set_control(_gvn.transform(excluded_rgn));
4055 set_all_memory(excluded_mem);
4056
4057 // Store the vthread exclusion state to the jfr thread local.
4058 Node* thread_local_excluded_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
4059 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
4060
4061 // Store release
4062 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
4063
4064 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
4065 record_for_igvn(thread_compare_rgn);
4066 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
4067 record_for_igvn(thread_compare_mem);
4068 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
4069 record_for_igvn(vthread);
4070
4071 // Merge the thread_compare control and memory.
4072 thread_compare_rgn->init_req(_true_path, control());
4073 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
4074 thread_compare_mem->init_req(_true_path, vthread_true_memory);
4075 thread_compare_mem->init_req(_false_path, vthread_false_memory);
4076
4077 // Set output state.
4078 set_control(_gvn.transform(thread_compare_rgn));
4079 set_all_memory(_gvn.transform(thread_compare_mem));
4080 }
4081
4082 //------------------------inline_native_try_update_epoch------------------
4083 //
4084 // The generated code is a function of the argument type.
4085 //
4086 bool LibraryCallKit::inline_native_try_update_epoch() {
4087 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
4088
4089 // Save input memory.
4090 Node* input_memory_state = reset_memory();
4091 set_all_memory(input_memory_state);
4092
4093 // Argument is an oop whose class has an injected instance field,
4094 // called 'jfr_epoch' of type T_INT, used for holding a jfr epoch value.
4095 Node* oop = argument(0);
4096 const TypeInstPtr* tinst = _gvn.type(oop)->isa_instptr();
4097 assert(tinst != nullptr, "oop is null");
4098 assert(tinst->is_loaded(), "klass is not loaded");
4099 ciInstanceKlass* const ik = tinst->instance_klass();
4100
4101 ciField* const field = ik->get_injected_instance_field_by_name(ciSymbol::make("jfr_epoch"),
4102 ciSymbol::make("I"));
4103
4104 assert(field != nullptr, "field 'jfr_epoch' of type I not injected in klass %s", ik->name()->as_utf8());
4105
4106 const int jfr_epoch_field_offset = field->offset_in_bytes();
4107 Node* oop_epoch_field_offset = basic_plus_adr(oop, jfr_epoch_field_offset);
4108 const TypePtr* adr_type = _gvn.type(oop_epoch_field_offset)->isa_ptr();
4109 const int alias_idx = C->get_alias_index(adr_type);
4110 BasicType bt = field->layout_type();
4111 const Type * oop_epoch_field_type = Type::get_const_basic_type(bt);
4112
4113 // Load the epoch value from the oop.
4114 Node* oop_epoch = access_load_at(oop,
4115 oop_epoch_field_offset,
4116 adr_type, oop_epoch_field_type,
4117 bt, IN_HEAP | MO_UNORDERED);
4118
4119 // Load the current JFR epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
4120 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
4121 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
4122
4123 // Compare the epoch in the oop against the current JFR epoch generation.
4124 Node* const epochs_cmp = _gvn.transform(new CmpINode(current_epoch_generation, oop_epoch));
4125 Node* epochs_equal_test = _gvn.transform(new BoolNode(epochs_cmp, BoolTest::eq));
4126 IfNode* iff_epochs_equal = create_and_map_if(control(), epochs_equal_test, PROB_LIKELY(0.999), COUNT_UNKNOWN);
4127
4128 // True path.
4129 Node* epochs_are_equal = _gvn.transform(new IfTrueNode(iff_epochs_equal));
4130
4131 // False path.
4132 Node* epochs_are_not_equal = _gvn.transform(new IfFalseNode(iff_epochs_equal));
4133
4134 set_control(_gvn.transform(epochs_are_not_equal));
4135
4136 // Attempt to cas the current JFR epoch generation into the oop epoch field.
4137 DecoratorSet decorators = IN_HEAP;
4138 decorators |= mo_decorator_for_access_kind(Volatile);
4139
4140 Node* result = access_atomic_cmpxchg_val_at(oop,
4141 oop_epoch_field_offset,
4142 adr_type, alias_idx,
4143 oop_epoch, // expected value
4144 current_epoch_generation, // new value
4145 oop_epoch_field_type,
4146 bt,
4147 decorators);
4148
4149 // Compare the result of the cas operation to the expected value.
4150 Node* const cas_cmp_to_expected_value = _gvn.transform(new CmpINode(result, oop_epoch));
4151 Node* cas_operation_test = _gvn.transform(new BoolNode(cas_cmp_to_expected_value, BoolTest::eq));
4152 IfNode* iff_cas_success = create_and_map_if(control(), cas_operation_test, PROB_LIKELY(0.999), COUNT_UNKNOWN);
4153
4154 // True path.
4155 Node* cas_success = _gvn.transform(new IfTrueNode(iff_cas_success));
4156
4157 // False path.
4158 Node* cas_failure = _gvn.transform(new IfFalseNode(iff_cas_success));
4159
4160 // Cas result region and phi nodes.
4161 RegionNode* cas_operation_rgn = new RegionNode(PATH_LIMIT);
4162 record_for_igvn(cas_operation_rgn);
4163 PhiNode* cas_operation_mem = new PhiNode(cas_operation_rgn, Type::MEMORY, TypePtr::BOTTOM);
4164 record_for_igvn(cas_operation_mem);
4165 PhiNode* cas_result = new PhiNode(cas_operation_rgn, TypeInt::BOOL);
4166 record_for_igvn(cas_result);
4167
4168 cas_operation_rgn->init_req(_true_path, _gvn.transform(cas_success));
4169 cas_operation_rgn->init_req(_false_path, _gvn.transform(cas_failure));
4170 cas_operation_mem->init_req(_true_path, reset_memory());
4171 cas_operation_mem->init_req(_false_path, input_memory_state);
4172 cas_result->init_req(_true_path, _gvn.intcon(1));
4173 cas_result->init_req(_false_path, _gvn.intcon(0));
4174
4175 // Epoch compare region and phi nodes.
4176 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
4177 record_for_igvn(epoch_compare_rgn);
4178 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
4179 record_for_igvn(epoch_compare_mem);
4180 PhiNode* result_value = new PhiNode(epoch_compare_rgn, TypeInt::BOOL);
4181 record_for_igvn(result_value);
4182
4183 epoch_compare_rgn->init_req(_true_path, _gvn.transform(epochs_are_equal));
4184 epoch_compare_rgn->init_req(_false_path, _gvn.transform(cas_operation_rgn));
4185 epoch_compare_mem->init_req(_true_path, _gvn.transform(input_memory_state));
4186 epoch_compare_mem->init_req(_false_path, _gvn.transform(cas_operation_mem));
4187 result_value->init_req(_true_path, _gvn.intcon(0));
4188 result_value->init_req(_false_path, _gvn.transform(cas_result));
4189
4190 // Set output state.
4191 set_result(epoch_compare_rgn, result_value);
4192 set_all_memory(_gvn.transform(epoch_compare_mem));
4193
4194 return true;
4195 }
4196
4197 #endif // JFR_HAVE_INTRINSICS
4198
4199 //------------------------inline_native_currentCarrierThread------------------
4200 bool LibraryCallKit::inline_native_currentCarrierThread() {
4201 Node* junk = nullptr;
4202 set_result(generate_current_thread(junk));
4203 return true;
4204 }
4205
4206 //------------------------inline_native_currentThread------------------
4207 bool LibraryCallKit::inline_native_currentThread() {
4208 Node* junk = nullptr;
4209 set_result(generate_virtual_thread(junk));
4210 return true;
4211 }
4212
4213 //------------------------inline_native_setVthread------------------
4214 bool LibraryCallKit::inline_native_setCurrentThread() {
4215 assert(C->method()->changes_current_thread(),
4216 "method changes current Thread but is not annotated ChangesCurrentThread");
4217 Node* arr = argument(1);
4218 Node* thread = _gvn.transform(new ThreadLocalNode());
4219 Node* p = off_heap_plus_addr(thread, in_bytes(JavaThread::vthread_offset()));
4220 Node* thread_obj_handle
4221 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
4222 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
4223 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
4224
4225 // Change the _monitor_owner_id of the JavaThread
4226 Node* tid = load_field_from_object(arr, "tid", "J");
4227 Node* monitor_owner_id_offset = off_heap_plus_addr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
4228 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
4229
4230 JFR_ONLY(extend_setCurrentThread(thread, arr);)
4231 return true;
4232 }
4233
4234 const Type* LibraryCallKit::scopedValueCache_type() {
4235 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
4236 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
4237 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS, /* stable= */ false, /* flat= */ false, /* not_flat= */ true, /* not_null_free= */ true, true);
4238
4239 // Because we create the scopedValue cache lazily we have to make the
4240 // type of the result BotPTR.
4241 bool xk = etype->klass_is_exact();
4242 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, TypeAryPtr::Offset(0));
4243 return objects_type;
4244 }
4245
4246 Node* LibraryCallKit::scopedValueCache_helper() {
4247 Node* thread = _gvn.transform(new ThreadLocalNode());
4248 Node* p = off_heap_plus_addr(thread, in_bytes(JavaThread::scopedValueCache_offset()));
4249 // We cannot use immutable_memory() because we might flip onto a
4250 // different carrier thread, at which point we'll need to use that
4251 // carrier thread's cache.
4252 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
4253 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
4254 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
4255 }
4256
4257 //------------------------inline_native_scopedValueCache------------------
4258 bool LibraryCallKit::inline_native_scopedValueCache() {
4259 Node* cache_obj_handle = scopedValueCache_helper();
4260 const Type* objects_type = scopedValueCache_type();
4261 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
4262
4263 return true;
4264 }
4265
4266 //------------------------inline_native_setScopedValueCache------------------
4267 bool LibraryCallKit::inline_native_setScopedValueCache() {
4268 Node* arr = argument(0);
4269 Node* cache_obj_handle = scopedValueCache_helper();
4270 const Type* objects_type = scopedValueCache_type();
4271
4272 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
4273 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
4274
4275 return true;
4276 }
4277
4278 //------------------------inline_native_Continuation_pin and unpin-----------
4279
4280 // Shared implementation routine for both pin and unpin.
4281 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
4282 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
4283
4284 // Save input memory.
4285 Node* input_memory_state = reset_memory();
4286 set_all_memory(input_memory_state);
4287
4288 // TLS
4289 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
4290 Node* last_continuation_offset = off_heap_plus_addr(tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
4291 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
4292
4293 // Null check the last continuation object.
4294 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
4295 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
4296 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
4297
4298 // False path, last continuation is null.
4299 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
4300
4301 // True path, last continuation is not null.
4302 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
4303
4304 set_control(continuation_is_not_null);
4305
4306 // Load the pin count from the last continuation.
4307 Node* pin_count_offset = off_heap_plus_addr(last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
4308 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
4309
4310 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
4311 Node* pin_count_rhs;
4312 if (unpin) {
4313 pin_count_rhs = _gvn.intcon(0);
4314 } else {
4315 pin_count_rhs = _gvn.intcon(UINT32_MAX);
4316 }
4317 Node* pin_count_cmp = _gvn.transform(new CmpUNode(pin_count, pin_count_rhs));
4318 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
4319 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
4320
4321 // True branch, pin count over/underflow.
4322 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
4323 {
4324 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
4325 // which will throw IllegalStateException for pin count over/underflow.
4326 // No memory changed so far - we can use memory create by reset_memory()
4327 // at the beginning of this intrinsic. No need to call reset_memory() again.
4328 PreserveJVMState pjvms(this);
4329 set_control(pin_count_over_underflow);
4330 uncommon_trap(Deoptimization::Reason_intrinsic,
4331 Deoptimization::Action_none);
4332 assert(stopped(), "invariant");
4333 }
4334
4335 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
4336 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
4337 set_control(valid_pin_count);
4338
4339 Node* next_pin_count;
4340 if (unpin) {
4341 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
4342 } else {
4343 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
4344 }
4345
4346 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
4347
4348 // Result of top level CFG and Memory.
4349 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
4350 record_for_igvn(result_rgn);
4351 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
4352 record_for_igvn(result_mem);
4353
4354 result_rgn->init_req(_true_path, valid_pin_count);
4355 result_rgn->init_req(_false_path, continuation_is_null);
4356 result_mem->init_req(_true_path, reset_memory());
4357 result_mem->init_req(_false_path, input_memory_state);
4358
4359 // Set output state.
4360 set_control(_gvn.transform(result_rgn));
4361 set_all_memory(_gvn.transform(result_mem));
4362
4363 return true;
4364 }
4365
4366 //---------------------------load_mirror_from_klass----------------------------
4367 // Given a klass oop, load its java mirror (a java.lang.Class oop).
4368 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
4369 Node* p = off_heap_plus_addr(klass, in_bytes(Klass::java_mirror_offset()));
4370 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4371 // mirror = ((OopHandle)mirror)->resolve();
4372 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
4373 }
4374
4375 //-----------------------load_klass_from_mirror_common-------------------------
4376 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
4377 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
4378 // and branch to the given path on the region.
4379 // If never_see_null, take an uncommon trap on null, so we can optimistically
4380 // compile for the non-null case.
4381 // If the region is null, force never_see_null = true.
4382 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
4383 bool never_see_null,
4384 RegionNode* region,
4385 int null_path,
4386 int offset) {
4387 if (region == nullptr) never_see_null = true;
4388 Node* p = basic_plus_adr(mirror, offset);
4389 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4390 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
4391 Node* null_ctl = top();
4392 kls = null_check_oop(kls, &null_ctl, never_see_null);
4393 if (region != nullptr) {
4394 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
4395 region->init_req(null_path, null_ctl);
4396 } else {
4397 assert(null_ctl == top(), "no loose ends");
4398 }
4399 return kls;
4400 }
4401
4402 //--------------------(inline_native_Class_query helpers)---------------------
4403 // Use this for JVM_ACC_INTERFACE.
4404 // Fall through if (mods & mask) == bits, take the guard otherwise.
4405 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
4406 ByteSize offset, const Type* type, BasicType bt) {
4407 // Branch around if the given klass has the given modifier bit set.
4408 // Like generate_guard, adds a new path onto the region.
4409 Node* modp = off_heap_plus_addr(kls, in_bytes(offset));
4410 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
4411 Node* mask = intcon(modifier_mask);
4412 Node* bits = intcon(modifier_bits);
4413 Node* mbit = _gvn.transform(new AndINode(mods, mask));
4414 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
4415 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
4416 return generate_fair_guard(bol, region);
4417 }
4418
4419 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
4420 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
4421 InstanceKlass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
4422 }
4423
4424 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
4425 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
4426 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
4427 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
4428 }
4429
4430 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
4431 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
4432 }
4433
4434 //-------------------------inline_native_Class_query-------------------
4435 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
4436 const Type* return_type = TypeInt::BOOL;
4437 Node* prim_return_value = top(); // what happens if it's a primitive class?
4438 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4439 bool expect_prim = false; // most of these guys expect to work on refs
4440
4441 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
4442
4443 Node* mirror = argument(0);
4444 Node* obj = top();
4445
4446 switch (id) {
4447 case vmIntrinsics::_isInstance:
4448 // nothing is an instance of a primitive type
4449 prim_return_value = intcon(0);
4450 obj = argument(1);
4451 break;
4452 case vmIntrinsics::_isHidden:
4453 prim_return_value = intcon(0);
4454 break;
4455 case vmIntrinsics::_getSuperclass:
4456 prim_return_value = null();
4457 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
4458 break;
4459 default:
4460 fatal_unexpected_iid(id);
4461 break;
4462 }
4463
4464 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4465 if (mirror_con == nullptr) return false; // cannot happen?
4466
4467 #ifndef PRODUCT
4468 if (C->print_intrinsics() || C->print_inlining()) {
4469 ciType* k = mirror_con->java_mirror_type();
4470 if (k) {
4471 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4472 k->print_name();
4473 tty->cr();
4474 }
4475 }
4476 #endif
4477
4478 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4479 RegionNode* region = new RegionNode(PATH_LIMIT);
4480 record_for_igvn(region);
4481 PhiNode* phi = new PhiNode(region, return_type);
4482
4483 // The mirror will never be null of Reflection.getClassAccessFlags, however
4484 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4485 // if it is. See bug 4774291.
4486
4487 // For Reflection.getClassAccessFlags(), the null check occurs in
4488 // the wrong place; see inline_unsafe_access(), above, for a similar
4489 // situation.
4490 mirror = null_check(mirror);
4491 // If mirror or obj is dead, only null-path is taken.
4492 if (stopped()) return true;
4493
4494 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4495
4496 // Now load the mirror's klass metaobject, and null-check it.
4497 // Side-effects region with the control path if the klass is null.
4498 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4499 // If kls is null, we have a primitive mirror.
4500 phi->init_req(_prim_path, prim_return_value);
4501 if (stopped()) { set_result(region, phi); return true; }
4502 bool safe_for_replace = (region->in(_prim_path) == top());
4503
4504 Node* p; // handy temp
4505 Node* null_ctl;
4506
4507 // Now that we have the non-null klass, we can perform the real query.
4508 // For constant classes, the query will constant-fold in LoadNode::Value.
4509 Node* query_value = top();
4510 switch (id) {
4511 case vmIntrinsics::_isInstance:
4512 // nothing is an instance of a primitive type
4513 query_value = gen_instanceof(obj, kls, safe_for_replace);
4514 break;
4515
4516 case vmIntrinsics::_isHidden:
4517 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4518 if (generate_hidden_class_guard(kls, region) != nullptr)
4519 // A guard was added. If the guard is taken, it was an hidden class.
4520 phi->add_req(intcon(1));
4521 // If we fall through, it's a plain class.
4522 query_value = intcon(0);
4523 break;
4524
4525
4526 case vmIntrinsics::_getSuperclass:
4527 // The rules here are somewhat unfortunate, but we can still do better
4528 // with random logic than with a JNI call.
4529 // Interfaces store null or Object as _super, but must report null.
4530 // Arrays store an intermediate super as _super, but must report Object.
4531 // Other types can report the actual _super.
4532 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4533 if (generate_array_guard(kls, region) != nullptr) {
4534 // A guard was added. If the guard is taken, it was an array.
4535 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4536 }
4537 // Check for interface after array since this checks AccessFlags offset into InstanceKlass.
4538 // In other words, we are accessing subtype-specific information, so we need to determine the subtype first.
4539 if (generate_interface_guard(kls, region) != nullptr) {
4540 // A guard was added. If the guard is taken, it was an interface.
4541 phi->add_req(null());
4542 }
4543 // If we fall through, it's a plain class. Get its _super.
4544 if (!stopped()) {
4545 p = basic_plus_adr(top(), kls, in_bytes(Klass::super_offset()));
4546 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4547 null_ctl = top();
4548 kls = null_check_oop(kls, &null_ctl);
4549 if (null_ctl != top()) {
4550 // If the guard is taken, Object.superClass is null (both klass and mirror).
4551 region->add_req(null_ctl);
4552 phi ->add_req(null());
4553 }
4554 if (!stopped()) {
4555 query_value = load_mirror_from_klass(kls);
4556 }
4557 }
4558 break;
4559
4560 default:
4561 fatal_unexpected_iid(id);
4562 break;
4563 }
4564
4565 // Fall-through is the normal case of a query to a real class.
4566 phi->init_req(1, query_value);
4567 region->init_req(1, control());
4568
4569 C->set_has_split_ifs(true); // Has chance for split-if optimization
4570 set_result(region, phi);
4571 return true;
4572 }
4573
4574
4575 //-------------------------inline_Class_cast-------------------
4576 bool LibraryCallKit::inline_Class_cast() {
4577 Node* mirror = argument(0); // Class
4578 Node* obj = argument(1);
4579 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4580 if (mirror_con == nullptr) {
4581 return false; // dead path (mirror->is_top()).
4582 }
4583 if (obj == nullptr || obj->is_top()) {
4584 return false; // dead path
4585 }
4586 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4587
4588 // First, see if Class.cast() can be folded statically.
4589 // java_mirror_type() returns non-null for compile-time Class constants.
4590 ciType* tm = mirror_con->java_mirror_type();
4591 if (tm != nullptr && tm->is_klass() &&
4592 tp != nullptr) {
4593 if (!tp->is_loaded()) {
4594 // Don't use intrinsic when class is not loaded.
4595 return false;
4596 } else {
4597 const TypeKlassPtr* tklass = TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces);
4598 int static_res = C->static_subtype_check(tklass, tp->as_klass_type());
4599 if (static_res == Compile::SSC_always_true) {
4600 // isInstance() is true - fold the code.
4601 set_result(obj);
4602 return true;
4603 } else if (static_res == Compile::SSC_always_false) {
4604 // Don't use intrinsic, have to throw ClassCastException.
4605 // If the reference is null, the non-intrinsic bytecode will
4606 // be optimized appropriately.
4607 return false;
4608 }
4609 }
4610 }
4611
4612 // Bailout intrinsic and do normal inlining if exception path is frequent.
4613 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4614 return false;
4615 }
4616
4617 // Generate dynamic checks.
4618 // Class.cast() is java implementation of _checkcast bytecode.
4619 // Do checkcast (Parse::do_checkcast()) optimizations here.
4620
4621 mirror = null_check(mirror);
4622 // If mirror is dead, only null-path is taken.
4623 if (stopped()) {
4624 return true;
4625 }
4626
4627 // Not-subtype or the mirror's klass ptr is nullptr (in case it is a primitive).
4628 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
4629 RegionNode* region = new RegionNode(PATH_LIMIT);
4630 record_for_igvn(region);
4631
4632 // Now load the mirror's klass metaobject, and null-check it.
4633 // If kls is null, we have a primitive mirror and
4634 // nothing is an instance of a primitive type.
4635 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4636
4637 Node* res = top();
4638 Node* io = i_o();
4639 Node* mem = merged_memory();
4640 SafePointNode* new_cast_failure_map = nullptr;
4641
4642 if (!stopped()) {
4643
4644 Node* bad_type_ctrl = top();
4645 // Do checkcast optimizations.
4646 res = gen_checkcast(obj, kls, &bad_type_ctrl, &new_cast_failure_map);
4647 region->init_req(_bad_type_path, bad_type_ctrl);
4648 }
4649 if (region->in(_prim_path) != top() ||
4650 region->in(_bad_type_path) != top() ||
4651 region->in(_npe_path) != top()) {
4652 // Let Interpreter throw ClassCastException.
4653 PreserveJVMState pjvms(this);
4654 if (new_cast_failure_map != nullptr) {
4655 // The current map on the success path could have been modified. Use the dedicated failure path map.
4656 set_map(new_cast_failure_map);
4657 }
4658 set_control(_gvn.transform(region));
4659 // Set IO and memory because gen_checkcast may override them when buffering inline types
4660 set_i_o(io);
4661 set_all_memory(mem);
4662 uncommon_trap(Deoptimization::Reason_intrinsic,
4663 Deoptimization::Action_maybe_recompile);
4664 }
4665 if (!stopped()) {
4666 set_result(res);
4667 }
4668 return true;
4669 }
4670
4671
4672 //--------------------------inline_native_subtype_check------------------------
4673 // This intrinsic takes the JNI calls out of the heart of
4674 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4675 bool LibraryCallKit::inline_native_subtype_check() {
4676 // Pull both arguments off the stack.
4677 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4678 args[0] = argument(0);
4679 args[1] = argument(1);
4680 Node* klasses[2]; // corresponding Klasses: superk, subk
4681 klasses[0] = klasses[1] = top();
4682
4683 enum {
4684 // A full decision tree on {superc is prim, subc is prim}:
4685 _prim_0_path = 1, // {P,N} => false
4686 // {P,P} & superc!=subc => false
4687 _prim_same_path, // {P,P} & superc==subc => true
4688 _prim_1_path, // {N,P} => false
4689 _ref_subtype_path, // {N,N} & subtype check wins => true
4690 _both_ref_path, // {N,N} & subtype check loses => false
4691 PATH_LIMIT
4692 };
4693
4694 RegionNode* region = new RegionNode(PATH_LIMIT);
4695 RegionNode* prim_region = new RegionNode(2);
4696 Node* phi = new PhiNode(region, TypeInt::BOOL);
4697 record_for_igvn(region);
4698 record_for_igvn(prim_region);
4699
4700 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4701 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4702 int class_klass_offset = java_lang_Class::klass_offset();
4703
4704 // First null-check both mirrors and load each mirror's klass metaobject.
4705 int which_arg;
4706 for (which_arg = 0; which_arg <= 1; which_arg++) {
4707 Node* arg = args[which_arg];
4708 arg = null_check(arg);
4709 if (stopped()) break;
4710 args[which_arg] = arg;
4711
4712 Node* p = basic_plus_adr(arg, class_klass_offset);
4713 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4714 klasses[which_arg] = _gvn.transform(kls);
4715 }
4716
4717 // Having loaded both klasses, test each for null.
4718 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4719 for (which_arg = 0; which_arg <= 1; which_arg++) {
4720 Node* kls = klasses[which_arg];
4721 Node* null_ctl = top();
4722 kls = null_check_oop(kls, &null_ctl, never_see_null);
4723 if (which_arg == 0) {
4724 prim_region->init_req(1, null_ctl);
4725 } else {
4726 region->init_req(_prim_1_path, null_ctl);
4727 }
4728 if (stopped()) break;
4729 klasses[which_arg] = kls;
4730 }
4731
4732 if (!stopped()) {
4733 // now we have two reference types, in klasses[0..1]
4734 Node* subk = klasses[1]; // the argument to isAssignableFrom
4735 Node* superk = klasses[0]; // the receiver
4736 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4737 region->set_req(_ref_subtype_path, control());
4738 }
4739
4740 // If both operands are primitive (both klasses null), then
4741 // we must return true when they are identical primitives.
4742 // It is convenient to test this after the first null klass check.
4743 // This path is also used if superc is a value mirror.
4744 set_control(_gvn.transform(prim_region));
4745 if (!stopped()) {
4746 // Since superc is primitive, make a guard for the superc==subc case.
4747 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4748 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4749 generate_fair_guard(bol_eq, region);
4750 if (region->req() == PATH_LIMIT+1) {
4751 // A guard was added. If the added guard is taken, superc==subc.
4752 region->swap_edges(PATH_LIMIT, _prim_same_path);
4753 region->del_req(PATH_LIMIT);
4754 }
4755 region->set_req(_prim_0_path, control()); // Not equal after all.
4756 }
4757
4758 // these are the only paths that produce 'true':
4759 phi->set_req(_prim_same_path, intcon(1));
4760 phi->set_req(_ref_subtype_path, intcon(1));
4761
4762 // pull together the cases:
4763 assert(region->req() == PATH_LIMIT, "sane region");
4764 for (uint i = 1; i < region->req(); i++) {
4765 Node* ctl = region->in(i);
4766 if (ctl == nullptr || ctl == top()) {
4767 region->set_req(i, top());
4768 phi ->set_req(i, top());
4769 } else if (phi->in(i) == nullptr) {
4770 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4771 }
4772 }
4773
4774 set_control(_gvn.transform(region));
4775 set_result(_gvn.transform(phi));
4776 return true;
4777 }
4778
4779 //---------------------generate_array_guard_common------------------------
4780 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind, Node** obj) {
4781
4782 if (stopped()) {
4783 return nullptr;
4784 }
4785
4786 // Like generate_guard, adds a new path onto the region.
4787 jint layout_con = 0;
4788 Node* layout_val = get_layout_helper(kls, layout_con);
4789 if (layout_val == nullptr) {
4790 bool query = 0;
4791 switch(kind) {
4792 case RefArray: query = Klass::layout_helper_is_refArray(layout_con); break;
4793 case NonRefArray: query = !Klass::layout_helper_is_refArray(layout_con); break;
4794 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
4795 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
4796 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
4797 default:
4798 ShouldNotReachHere();
4799 }
4800 if (!query) {
4801 return nullptr; // never a branch
4802 } else { // always a branch
4803 Node* always_branch = control();
4804 if (region != nullptr)
4805 region->add_req(always_branch);
4806 set_control(top());
4807 return always_branch;
4808 }
4809 }
4810 unsigned int value = 0;
4811 BoolTest::mask btest = BoolTest::illegal;
4812 switch(kind) {
4813 case RefArray:
4814 case NonRefArray: {
4815 value = Klass::_lh_array_tag_ref_value;
4816 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4817 btest = (kind == RefArray) ? BoolTest::eq : BoolTest::ne;
4818 break;
4819 }
4820 case TypeArray: {
4821 value = Klass::_lh_array_tag_type_value;
4822 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4823 btest = BoolTest::eq;
4824 break;
4825 }
4826 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
4827 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
4828 default:
4829 ShouldNotReachHere();
4830 }
4831 // Now test the correct condition.
4832 jint nval = (jint)value;
4833 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4834 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4835 Node* ctrl = generate_fair_guard(bol, region);
4836 Node* is_array_ctrl = kind == NonArray ? control() : ctrl;
4837 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4838 // Keep track of the fact that 'obj' is an array to prevent
4839 // array specific accesses from floating above the guard.
4840 *obj = _gvn.transform(new CheckCastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4841 }
4842 return ctrl;
4843 }
4844
4845 // public static native Object[] ValueClass::newNullRestrictedAtomicArray(Class<?> componentType, int length, Object initVal);
4846 // public static native Object[] ValueClass::newNullRestrictedNonAtomicArray(Class<?> componentType, int length, Object initVal);
4847 // public static native Object[] ValueClass::newNullableAtomicArray(Class<?> componentType, int length);
4848 bool LibraryCallKit::inline_newArray(bool null_free, bool atomic) {
4849 assert(null_free || atomic, "nullable implies atomic");
4850 Node* componentType = argument(0);
4851 Node* length = argument(1);
4852 Node* init_val = null_free ? argument(2) : nullptr;
4853
4854 const TypeInstPtr* tp = _gvn.type(componentType)->isa_instptr();
4855 if (tp != nullptr) {
4856 ciInstanceKlass* ik = tp->instance_klass();
4857 if (ik == C->env()->Class_klass()) {
4858 ciType* t = tp->java_mirror_type();
4859 if (t != nullptr && t->is_inlinetype()) {
4860
4861 ciArrayKlass* array_klass = ciArrayKlass::make(t, null_free, atomic, true);
4862 assert(array_klass->is_elem_null_free() == null_free, "inconsistency");
4863
4864 // TODO 8350865 ZGC needs card marks on initializing oop stores
4865 if ((UseZGC || UseShenandoahGC) && null_free && !array_klass->is_flat_array_klass()) {
4866 return false;
4867 }
4868
4869 if (array_klass->is_loaded() && array_klass->element_klass()->as_inline_klass()->is_initialized()) {
4870 const TypeAryKlassPtr* array_klass_type = TypeAryKlassPtr::make(array_klass, Type::trust_interfaces);
4871 if (null_free) {
4872 if (init_val->is_InlineType()) {
4873 if (array_klass_type->is_flat() && init_val->as_InlineType()->is_all_zero(&gvn(), /* flat */ true)) {
4874 // Zeroing is enough because the init value is the all-zero value
4875 init_val = nullptr;
4876 } else {
4877 init_val = init_val->as_InlineType()->buffer(this);
4878 }
4879 }
4880 if (init_val != nullptr) {
4881 #ifdef ASSERT
4882 init_val = null_check(init_val);
4883 Node* wrong_type_ctl = gen_subtype_check(init_val, makecon(TypeKlassPtr::make(array_klass->element_klass())));
4884 {
4885 PreserveJVMState pjvms(this);
4886 set_control(wrong_type_ctl);
4887 halt(control(), frameptr(), "incompatible type for initVal in newArray");
4888 stop_and_kill_map();
4889 }
4890 #endif
4891 init_val = _gvn.transform(new CheckCastPPNode(control(), init_val, TypeOopPtr::make_from_klass(array_klass->element_klass()), ConstraintCastNode::DependencyType::NonFloatingNarrowing));
4892 }
4893 }
4894 Node* obj = new_array(makecon(array_klass_type), length, 0, nullptr, false, init_val);
4895 const TypeAryPtr* arytype = gvn().type(obj)->is_aryptr();
4896 assert(arytype->is_null_free() == null_free, "inconsistency");
4897 assert(arytype->is_not_null_free() == !null_free, "inconsistency");
4898 set_result(obj);
4899 return true;
4900 }
4901 }
4902 }
4903 }
4904 return false;
4905 }
4906
4907 // public static native boolean ValueClass::isFlatArray(Object array);
4908 // public static native boolean ValueClass::isNullRestrictedArray(Object array);
4909 // public static native boolean ValueClass::isAtomicArray(Object array);
4910 bool LibraryCallKit::inline_getArrayProperties(ArrayPropertiesCheck check) {
4911 Node* array = argument(0);
4912
4913 Node* bol;
4914 switch(check) {
4915 case IsFlat:
4916 bol = flat_array_test(load_object_klass(array));
4917 break;
4918 case IsNullRestricted:
4919 bol = null_free_array_test(array);
4920 break;
4921 case IsAtomic: {
4922 // See conditions in JVM_IsAtomicArray
4923 // 1. If not flat, then atomic, or else...
4924 RegionNode* atomic_region = new RegionNode(1);
4925 RegionNode* non_atomic_region = new RegionNode(1);
4926 Node* array_klass = load_object_klass(array);
4927 Node* is_flat_bol = flat_array_test(array_klass);
4928 IfNode* iff_is_flat = create_and_xform_if(control(), is_flat_bol, PROB_FAIR, COUNT_UNKNOWN);
4929 atomic_region->add_req(_gvn.transform(new IfFalseNode(iff_is_flat)));
4930 set_control(_gvn.transform(new IfTrueNode(iff_is_flat)));
4931
4932 // 2. ...if the layout is atomic, then atomic, or else...
4933 Node* layout_kind = atomic_layout_array_test_and_get_layout_kind(array, atomic_region);
4934
4935 // 3. ...if the element type is naturally atomic and null-free OR empty and nullable, then atomic, or else...
4936 int element_klass_offset = in_bytes(ObjArrayKlass::element_klass_offset());
4937 Node* array_element_klass_addr = off_heap_plus_addr(array_klass, element_klass_offset);
4938 Node* array_element_klass = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), array_element_klass_addr, _gvn.type(array_klass)->is_klassptr()));
4939 int klass_flags_offset = in_bytes(InstanceKlass::misc_flags_offset() + InstanceKlassFlags::flags_offset());
4940 Node* array_element_klass_flags_addr = off_heap_plus_addr(array_element_klass, klass_flags_offset);
4941 Node* array_element_klass_flags = make_load(control(), array_element_klass_flags_addr, TypeInt::INT, T_INT, MemNode::unordered);
4942
4943 // Here, layout can only be non-atomic, otherwise atomic_layout_array_test_and_get_layout_kind already decides the array to be atomic.
4944 Node* is_null_free_cmp = _gvn.transform(new CmpINode(layout_kind, intcon(static_cast<jint>(LayoutKind::NULL_FREE_NON_ATOMIC_FLAT))));
4945 Node* is_null_free_bol = _gvn.transform(new BoolNode(is_null_free_cmp, BoolTest::eq));
4946 IfNode* iff_is_null_free_bol = create_and_xform_if(control(), is_null_free_bol, PROB_FAIR, COUNT_UNKNOWN);
4947 Node* is_null_free_ctl = _gvn.transform(new IfTrueNode(iff_is_null_free_bol));
4948 Node* is_nullable_ctl = _gvn.transform(new IfFalseNode(iff_is_null_free_bol));
4949
4950 Node* is_naturally_atomic_flag = _gvn.transform(new AndINode(array_element_klass_flags, intcon(InstanceKlassFlags::_misc_is_naturally_atomic)));
4951 Node* is_naturally_atomic_cmp = _gvn.transform(new CmpINode(is_naturally_atomic_flag, intcon(0)));
4952 Node* is_naturally_atomic_bol = _gvn.transform(new BoolNode(is_naturally_atomic_cmp, BoolTest::ne));
4953 IfNode* iff_is_naturally_atomic = create_and_xform_if(is_null_free_ctl, is_naturally_atomic_bol, PROB_FAIR, COUNT_UNKNOWN);
4954 Node* is_naturally_atomic_ctl = _gvn.transform(new IfTrueNode(iff_is_naturally_atomic));
4955 Node* is_not_naturally_atomic_ctl = _gvn.transform(new IfFalseNode(iff_is_naturally_atomic));
4956 atomic_region->add_req(is_naturally_atomic_ctl);
4957 non_atomic_region->add_req(is_not_naturally_atomic_ctl);
4958
4959 Node* is_empty_inline_type_flag = _gvn.transform(new AndINode(array_element_klass_flags, intcon(InstanceKlassFlags::_misc_is_empty_inline_type)));
4960 Node* is_empty_inline_type_cmp = _gvn.transform(new CmpINode(is_empty_inline_type_flag, intcon(0)));
4961 Node* is_empty_inline_type_bol = _gvn.transform(new BoolNode(is_empty_inline_type_cmp, BoolTest::ne));
4962 IfNode* iff_is_empty_inline_type = create_and_xform_if(is_nullable_ctl, is_empty_inline_type_bol, PROB_FAIR, COUNT_UNKNOWN);
4963 Node* is_empty_inline_type_ctl = _gvn.transform(new IfTrueNode(iff_is_empty_inline_type));
4964 Node* is_nonempty_inline_type_ctl = _gvn.transform(new IfFalseNode(iff_is_empty_inline_type));
4965 atomic_region->add_req(is_empty_inline_type_ctl);
4966 non_atomic_region->add_req(is_nonempty_inline_type_ctl);
4967
4968 // ...non-atomic, but we tried everything.
4969 RegionNode* decision = new RegionNode(3);
4970 decision->set_req(1, _gvn.transform(atomic_region));
4971 decision->set_req(2, _gvn.transform(non_atomic_region));
4972 PhiNode* result = PhiNode::make(decision, intcon(1), TypeInt::BOOL);
4973 result->set_req(2, intcon(0));
4974 set_control(_gvn.transform(decision));
4975 set_result(_gvn.transform(result));
4976 return true;
4977 }
4978 default:
4979 ShouldNotReachHere();
4980 }
4981
4982 Node* res = gvn().transform(new CMoveINode(bol, intcon(0), intcon(1), TypeInt::BOOL));
4983 set_result(res);
4984 return true;
4985 }
4986
4987 // Load the default refined array klass from an ObjArrayKlass. This relies on the first entry in the
4988 // '_next_refined_array_klass' linked list being the default (see ObjArrayKlass::klass_with_properties).
4989 Node* LibraryCallKit::load_default_refined_array_klass(Node* klass_node, bool type_array_guard) {
4990 RegionNode* region = new RegionNode(2);
4991 Node* phi = new PhiNode(region, TypeInstKlassPtr::OBJECT_OR_NULL);
4992
4993 if (type_array_guard) {
4994 generate_typeArray_guard(klass_node, region);
4995 if (region->req() == 3) {
4996 phi->add_req(klass_node);
4997 }
4998 }
4999 Node* adr_refined_klass = basic_plus_adr(top(), klass_node, in_bytes(ObjArrayKlass::next_refined_array_klass_offset()));
5000 Node* refined_klass = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), adr_refined_klass, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
5001
5002 // Can be null if not initialized yet, just deopt
5003 Node* null_ctl = top();
5004 refined_klass = null_check_oop(refined_klass, &null_ctl, /* never_see_null= */ true);
5005
5006 region->init_req(1, control());
5007 phi->init_req(1, refined_klass);
5008
5009 set_control(_gvn.transform(region));
5010 return _gvn.transform(phi);
5011 }
5012
5013 // Load the non-refined array klass from an ObjArrayKlass.
5014 Node* LibraryCallKit::load_non_refined_array_klass(Node* klass_node) {
5015 const TypeAryKlassPtr* ary_klass_ptr = _gvn.type(klass_node)->isa_aryklassptr();
5016 if (ary_klass_ptr != nullptr && ary_klass_ptr->klass_is_exact()) {
5017 return _gvn.makecon(ary_klass_ptr->cast_to_refined_array_klass_ptr(false));
5018 }
5019
5020 RegionNode* region = new RegionNode(2);
5021 Node* phi = new PhiNode(region, TypeInstKlassPtr::OBJECT);
5022
5023 generate_typeArray_guard(klass_node, region);
5024 if (region->req() == 3) {
5025 phi->add_req(klass_node);
5026 }
5027 Node* super_adr = basic_plus_adr(top(), klass_node, in_bytes(Klass::super_offset()));
5028 Node* super_klass = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), super_adr, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT));
5029
5030 region->init_req(1, control());
5031 phi->init_req(1, super_klass);
5032
5033 set_control(_gvn.transform(region));
5034 return _gvn.transform(phi);
5035 }
5036
5037 //-----------------------inline_native_newArray--------------------------
5038 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
5039 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
5040 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
5041 Node* mirror;
5042 Node* count_val;
5043 if (uninitialized) {
5044 null_check_receiver();
5045 mirror = argument(1);
5046 count_val = argument(2);
5047 } else {
5048 mirror = argument(0);
5049 count_val = argument(1);
5050 }
5051
5052 mirror = null_check(mirror);
5053 // If mirror or obj is dead, only null-path is taken.
5054 if (stopped()) return true;
5055
5056 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
5057 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5058 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5059 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
5060 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5061
5062 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
5063 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
5064 result_reg, _slow_path);
5065 Node* normal_ctl = control();
5066 Node* no_array_ctl = result_reg->in(_slow_path);
5067
5068 // Generate code for the slow case. We make a call to newArray().
5069 set_control(no_array_ctl);
5070 if (!stopped()) {
5071 // Either the input type is void.class, or else the
5072 // array klass has not yet been cached. Either the
5073 // ensuing call will throw an exception, or else it
5074 // will cache the array klass for next time.
5075 PreserveJVMState pjvms(this);
5076 CallJavaNode* slow_call = nullptr;
5077 if (uninitialized) {
5078 // Generate optimized virtual call (holder class 'Unsafe' is final)
5079 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
5080 } else {
5081 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
5082 }
5083 Node* slow_result = set_results_for_java_call(slow_call);
5084 // this->control() comes from set_results_for_java_call
5085 result_reg->set_req(_slow_path, control());
5086 result_val->set_req(_slow_path, slow_result);
5087 result_io ->set_req(_slow_path, i_o());
5088 result_mem->set_req(_slow_path, reset_memory());
5089 }
5090
5091 set_control(normal_ctl);
5092 if (!stopped()) {
5093 // Normal case: The array type has been cached in the java.lang.Class.
5094 // The following call works fine even if the array type is polymorphic.
5095 // It could be a dynamic mix of int[], boolean[], Object[], etc.
5096
5097 klass_node = load_default_refined_array_klass(klass_node);
5098
5099 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
5100 result_reg->init_req(_normal_path, control());
5101 result_val->init_req(_normal_path, obj);
5102 result_io ->init_req(_normal_path, i_o());
5103 result_mem->init_req(_normal_path, reset_memory());
5104
5105 if (uninitialized) {
5106 // Mark the allocation so that zeroing is skipped
5107 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
5108 alloc->maybe_set_complete(&_gvn);
5109 }
5110 }
5111
5112 // Return the combined state.
5113 set_i_o( _gvn.transform(result_io) );
5114 set_all_memory( _gvn.transform(result_mem));
5115
5116 C->set_has_split_ifs(true); // Has chance for split-if optimization
5117 set_result(result_reg, result_val);
5118 return true;
5119 }
5120
5121 //----------------------inline_native_getLength--------------------------
5122 // public static native int java.lang.reflect.Array.getLength(Object array);
5123 bool LibraryCallKit::inline_native_getLength() {
5124 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
5125
5126 Node* array = null_check(argument(0));
5127 // If array is dead, only null-path is taken.
5128 if (stopped()) return true;
5129
5130 // Deoptimize if it is a non-array.
5131 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
5132
5133 if (non_array != nullptr) {
5134 PreserveJVMState pjvms(this);
5135 set_control(non_array);
5136 uncommon_trap(Deoptimization::Reason_intrinsic,
5137 Deoptimization::Action_maybe_recompile);
5138 }
5139
5140 // If control is dead, only non-array-path is taken.
5141 if (stopped()) return true;
5142
5143 // The works fine even if the array type is polymorphic.
5144 // It could be a dynamic mix of int[], boolean[], Object[], etc.
5145 Node* result = load_array_length(array);
5146
5147 C->set_has_split_ifs(true); // Has chance for split-if optimization
5148 set_result(result);
5149 return true;
5150 }
5151
5152 //------------------------inline_array_copyOf----------------------------
5153 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
5154 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
5155 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
5156 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
5157
5158 // Get the arguments.
5159 Node* original = argument(0);
5160 Node* start = is_copyOfRange? argument(1): intcon(0);
5161 Node* end = is_copyOfRange? argument(2): argument(1);
5162 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
5163
5164 Node* newcopy = nullptr;
5165
5166 // Set the original stack and the reexecute bit for the interpreter to reexecute
5167 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
5168 { PreserveReexecuteState preexecs(this);
5169 jvms()->set_should_reexecute(true);
5170
5171 array_type_mirror = null_check(array_type_mirror);
5172 original = null_check(original);
5173
5174 // Check if a null path was taken unconditionally.
5175 if (stopped()) return true;
5176
5177 Node* orig_length = load_array_length(original);
5178
5179 RegionNode* bailout = new RegionNode(2);
5180 record_for_igvn(bailout);
5181
5182 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, bailout, 1);
5183 if (stopped()) {
5184 // Arrays.copyOf() uses a generic Class parameter which is erased to the raw type Class. This also allows
5185 // passing in primitive class mirrors like int.class which do not have corresponding Klass* pointers.
5186 // In these cases, klass_node will be top. Emit a trap to throw in the interpreter in this case.
5187 bail_out_from_array_copyOf(bailout);
5188 return true;
5189 }
5190
5191 klass_node = null_check(klass_node);
5192
5193 const TypeAryPtr* src_t = _gvn.type(original)->is_aryptr();
5194 const TypeKlassPtr* dest_klass_t = _gvn.type(klass_node)->is_klassptr()->is_klassptr();
5195
5196 Node* success_proj;
5197 if (should_bail_out_on_non_ref_arrays(src_t, dest_klass_t)) {
5198 success_proj = generate_non_refArray_guard(klass_node, bailout);
5199 } else {
5200 success_proj = generate_typeArray_guard(klass_node, bailout);
5201 }
5202
5203 Node* refined_klass_node = load_default_refined_array_klass(klass_node, /* type_array_guard= */ false);
5204
5205 if (success_proj != nullptr) {
5206 // Improve the klass node's type from the new optimistic assumption:
5207 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
5208 bool not_flat = !UseArrayFlattening;
5209 bool not_null_free = !Arguments::is_valhalla_enabled();
5210 const Type* akls = TypeAryKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0), Type::trust_interfaces, not_flat, not_null_free, false, false, not_flat, true);
5211 Node* cast = new CastPPNode(control(), refined_klass_node, akls);
5212 refined_klass_node = _gvn.transform(cast);
5213 }
5214
5215 // Bail out if either start or end is negative.
5216 generate_negative_guard(start, bailout, &start);
5217 generate_negative_guard(end, bailout, &end);
5218
5219 Node* length = end;
5220 if (_gvn.type(start) != TypeInt::ZERO) {
5221 length = _gvn.transform(new SubINode(end, start));
5222 }
5223
5224 // Bail out if length is negative (i.e., if start > end).
5225 // Without this the new_array would throw
5226 // NegativeArraySizeException but IllegalArgumentException is what
5227 // should be thrown
5228 generate_negative_guard(length, bailout, &length);
5229
5230 // Handle inline type arrays
5231 // TODO 8251971 This is too strong
5232 generate_fair_guard(flat_array_test(load_object_klass(original)), bailout);
5233 generate_fair_guard(flat_array_test(refined_klass_node), bailout);
5234 generate_fair_guard(null_free_array_test(original), bailout);
5235
5236 // Bail out if start is larger than the original length
5237 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
5238 generate_negative_guard(orig_tail, bailout, &orig_tail);
5239
5240 if (bailout->req() > 1) {
5241 bail_out_from_array_copyOf(bailout);
5242 }
5243
5244 if (!stopped()) {
5245 // How many elements will we copy from the original?
5246 // The answer is MinI(orig_tail, length).
5247 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
5248
5249 // Generate a direct call to the right arraycopy function(s).
5250 // We know the copy is disjoint but we might not know if the
5251 // oop stores need checking.
5252 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
5253 // This will fail a store-check if x contains any non-nulls.
5254
5255 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
5256 // loads/stores but it is legal only if we're sure the
5257 // Arrays.copyOf would succeed. So we need all input arguments
5258 // to the copyOf to be validated, including that the copy to the
5259 // new array won't trigger an ArrayStoreException. That subtype
5260 // check can be optimized if we know something on the type of
5261 // the input array from type speculation.
5262 if (_gvn.type(klass_node)->singleton()) {
5263 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
5264 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
5265
5266 int test = C->static_subtype_check(superk, subk);
5267 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
5268 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
5269 if (t_original->speculative_type() != nullptr) {
5270 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
5271 }
5272 }
5273 }
5274
5275 bool validated = false;
5276 // Reason_class_check rather than Reason_intrinsic because we
5277 // want to intrinsify even if this traps.
5278 if (!too_many_traps(Deoptimization::Reason_class_check)) {
5279 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
5280
5281 if (not_subtype_ctrl != top()) {
5282 PreserveJVMState pjvms(this);
5283 set_control(not_subtype_ctrl);
5284 uncommon_trap(Deoptimization::Reason_class_check,
5285 Deoptimization::Action_make_not_entrant);
5286 assert(stopped(), "Should be stopped");
5287 }
5288 validated = true;
5289 }
5290
5291 if (!stopped()) {
5292 newcopy = new_array(refined_klass_node, length, 0); // no arguments to push
5293
5294 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
5295 load_object_klass(original), klass_node);
5296 if (!is_copyOfRange) {
5297 ac->set_copyof(validated);
5298 } else {
5299 ac->set_copyofrange(validated);
5300 }
5301 Node* n = _gvn.transform(ac);
5302 if (n == ac) {
5303 ac->connect_outputs(this);
5304 } else {
5305 assert(validated, "shouldn't transform if all arguments not validated");
5306 set_all_memory(n);
5307 }
5308 }
5309 }
5310 } // original reexecute is set back here
5311
5312 C->set_has_split_ifs(true); // Has chance for split-if optimization
5313 if (!stopped()) {
5314 set_result(newcopy);
5315 }
5316 return true;
5317 }
5318
5319 void LibraryCallKit::bail_out_from_array_copyOf(RegionNode* bailout_region) {
5320 PreserveJVMState pjvms(this);
5321 set_control(_gvn.transform(bailout_region));
5322 uncommon_trap(Deoptimization::Reason_intrinsic,
5323 Deoptimization::Action_maybe_recompile);
5324 }
5325
5326 bool LibraryCallKit::should_bail_out_on_non_ref_arrays(const TypeAryPtr* src_type, const TypeKlassPtr* dest_klass_type) {
5327 const TypeAryKlassPtr* dest_ary_klass_type = dest_klass_type->isa_aryklassptr();
5328 if (dest_ary_klass_type == nullptr) {
5329 // Dest klass is not known to be an array class. There are multiple cases:
5330 // - Primitive class mirror: We already bailed out before.
5331 // - Instance class mirror: We should bail out.
5332 // - java.lang.Object (possible due to type erasure): Could be anything including primitive or instance class mirror
5333 // or also flat arrays. Bail out.
5334 return true;
5335 }
5336
5337 if (UseArrayFlattening) {
5338 // The remaining checks revolve around array flatness. Without array flatness, we don't need the stronger non-ref
5339 // runtime check excluding flat arrays.
5340 return false;
5341 }
5342
5343 // We now know that src and dest are proper array pointers.
5344 const bool src_maybe_flat = !src_type->is_not_flat();
5345 const bool dest_maybe_flat = !dest_ary_klass_type->is_not_flat();
5346
5347 // We could have abstract flat value class arrays whose layout we don't know. Bail out.
5348 const bool can_src_be_abstract_flat_value_class_array = src_maybe_flat && !src_type->elem()->is_inlinetypeptr();
5349 const bool can_dest_be_abstract_flat_value_class_array = dest_maybe_flat &&
5350 !dest_ary_klass_type->elem()->is_instklassptr()->instance_klass()->is_inlinetype();
5351 if (can_src_be_abstract_flat_value_class_array || can_dest_be_abstract_flat_value_class_array) {
5352 return true;
5353 }
5354
5355 // Value class array may have object field that would require a write barrier. Conservatively bail out.
5356 // TODO 8251971: Optimize for the case when flat src/dst are later found to not contain
5357 // oops (i.e., move this check to the macro expansion phase).
5358 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5359 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing)) {
5360 // No barriers required.
5361 return false;
5362 }
5363
5364 const bool can_src_be_flat_with_oops = src_maybe_flat && src_type->elem()->inline_klass()->contains_oops();
5365 const bool can_dest_be_flat_with_oops = dest_maybe_flat && dest_ary_klass_type->elem()->is_instklassptr()->instance_klass()->as_inline_klass()->contains_oops();
5366 if (can_src_be_flat_with_oops || can_dest_be_flat_with_oops) {
5367 return true;
5368 }
5369
5370 // Can handle remaining flat arrays.
5371 return false;
5372 }
5373
5374 //----------------------generate_virtual_guard---------------------------
5375 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
5376 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
5377 RegionNode* slow_region) {
5378 ciMethod* method = callee();
5379 int vtable_index = method->vtable_index();
5380 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5381 "bad index %d", vtable_index);
5382 // Get the Method* out of the appropriate vtable entry.
5383 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
5384 vtable_index*vtableEntry::size_in_bytes() +
5385 in_bytes(vtableEntry::method_offset());
5386 Node* entry_addr = off_heap_plus_addr(obj_klass, entry_offset);
5387 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
5388
5389 // Compare the target method with the expected method (e.g., Object.hashCode).
5390 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
5391
5392 Node* native_call = makecon(native_call_addr);
5393 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
5394 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
5395
5396 return generate_slow_guard(test_native, slow_region);
5397 }
5398
5399 //-----------------------generate_method_call----------------------------
5400 // Use generate_method_call to make a slow-call to the real
5401 // method if the fast path fails. An alternative would be to
5402 // use a stub like OptoRuntime::slow_arraycopy_Java.
5403 // This only works for expanding the current library call,
5404 // not another intrinsic. (E.g., don't use this for making an
5405 // arraycopy call inside of the copyOf intrinsic.)
5406 CallJavaNode*
5407 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
5408 // When compiling the intrinsic method itself, do not use this technique.
5409 guarantee(callee() != C->method(), "cannot make slow-call to self");
5410
5411 ciMethod* method = callee();
5412 // ensure the JVMS we have will be correct for this call
5413 guarantee(method_id == method->intrinsic_id(), "must match");
5414
5415 const TypeFunc* tf = TypeFunc::make(method);
5416 if (res_not_null) {
5417 assert(tf->return_type() == T_OBJECT, "");
5418 const TypeTuple* range = tf->range_cc();
5419 const Type** fields = TypeTuple::fields(range->cnt());
5420 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
5421 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
5422 tf = TypeFunc::make(tf->domain_cc(), new_range);
5423 }
5424 CallJavaNode* slow_call;
5425 if (is_static) {
5426 assert(!is_virtual, "");
5427 slow_call = new CallStaticJavaNode(C, tf,
5428 SharedRuntime::get_resolve_static_call_stub(), method);
5429 } else if (is_virtual) {
5430 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5431 int vtable_index = Method::invalid_vtable_index;
5432 if (UseInlineCaches) {
5433 // Suppress the vtable call
5434 } else {
5435 // hashCode and clone are not a miranda methods,
5436 // so the vtable index is fixed.
5437 // No need to use the linkResolver to get it.
5438 vtable_index = method->vtable_index();
5439 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5440 "bad index %d", vtable_index);
5441 }
5442 slow_call = new CallDynamicJavaNode(tf,
5443 SharedRuntime::get_resolve_virtual_call_stub(),
5444 method, vtable_index);
5445 } else { // neither virtual nor static: opt_virtual
5446 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5447 slow_call = new CallStaticJavaNode(C, tf,
5448 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
5449 slow_call->set_optimized_virtual(true);
5450 }
5451 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
5452 // To be able to issue a direct call (optimized virtual or virtual)
5453 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
5454 // about the method being invoked should be attached to the call site to
5455 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
5456 slow_call->set_override_symbolic_info(true);
5457 }
5458 set_arguments_for_java_call(slow_call);
5459 set_edges_for_java_call(slow_call);
5460 return slow_call;
5461 }
5462
5463
5464 /**
5465 * Build special case code for calls to hashCode on an object. This call may
5466 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
5467 * slightly different code.
5468 */
5469 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
5470 assert(is_static == callee()->is_static(), "correct intrinsic selection");
5471 assert(!(is_virtual && is_static), "either virtual, special, or static");
5472
5473 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
5474
5475 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5476 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
5477 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
5478 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5479 Node* obj = argument(0);
5480
5481 // Don't intrinsify hashcode on inline types for now.
5482 // The "is locked" runtime check also subsumes the inline type check (as inline types cannot be locked) and goes to the slow path.
5483 if (gvn().type(obj)->is_inlinetypeptr()) {
5484 return false;
5485 }
5486
5487 if (!is_static) {
5488 // Check for hashing null object
5489 obj = null_check_receiver();
5490 if (stopped()) return true; // unconditionally null
5491 result_reg->init_req(_null_path, top());
5492 result_val->init_req(_null_path, top());
5493 } else {
5494 // Do a null check, and return zero if null.
5495 // System.identityHashCode(null) == 0
5496 Node* null_ctl = top();
5497 obj = null_check_oop(obj, &null_ctl);
5498 result_reg->init_req(_null_path, null_ctl);
5499 result_val->init_req(_null_path, _gvn.intcon(0));
5500 }
5501
5502 // Unconditionally null? Then return right away.
5503 if (stopped()) {
5504 set_control( result_reg->in(_null_path));
5505 if (!stopped())
5506 set_result(result_val->in(_null_path));
5507 return true;
5508 }
5509
5510 // We only go to the fast case code if we pass a number of guards. The
5511 // paths which do not pass are accumulated in the slow_region.
5512 RegionNode* slow_region = new RegionNode(1);
5513 record_for_igvn(slow_region);
5514
5515 // If this is a virtual call, we generate a funny guard. We pull out
5516 // the vtable entry corresponding to hashCode() from the target object.
5517 // If the target method which we are calling happens to be the native
5518 // Object hashCode() method, we pass the guard. We do not need this
5519 // guard for non-virtual calls -- the caller is known to be the native
5520 // Object hashCode().
5521 if (is_virtual) {
5522 // After null check, get the object's klass.
5523 Node* obj_klass = load_object_klass(obj);
5524 generate_virtual_guard(obj_klass, slow_region);
5525 }
5526
5527 // Get the header out of the object, use LoadMarkNode when available
5528 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
5529 // The control of the load must be null. Otherwise, the load can move before
5530 // the null check after castPP removal.
5531 Node* no_ctrl = nullptr;
5532 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
5533
5534 if (!UseObjectMonitorTable) {
5535 // Test the header to see if it is safe to read w.r.t. locking.
5536 // We cannot use the inline type mask as this may check bits that are overridden
5537 // by an object monitor's pointer when inflating locking.
5538 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
5539 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
5540 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
5541 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
5542 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
5543
5544 generate_slow_guard(test_monitor, slow_region);
5545 }
5546
5547 // Get the hash value and check to see that it has been properly assigned.
5548 // We depend on hash_mask being at most 32 bits and avoid the use of
5549 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
5550 // vm: see markWord.hpp.
5551 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
5552 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
5553 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
5554 // This hack lets the hash bits live anywhere in the mark object now, as long
5555 // as the shift drops the relevant bits into the low 32 bits. Note that
5556 // Java spec says that HashCode is an int so there's no point in capturing
5557 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
5558 hshifted_header = ConvX2I(hshifted_header);
5559 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
5560
5561 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
5562 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
5563 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
5564
5565 generate_slow_guard(test_assigned, slow_region);
5566
5567 Node* init_mem = reset_memory();
5568 // fill in the rest of the null path:
5569 result_io ->init_req(_null_path, i_o());
5570 result_mem->init_req(_null_path, init_mem);
5571
5572 result_val->init_req(_fast_path, hash_val);
5573 result_reg->init_req(_fast_path, control());
5574 result_io ->init_req(_fast_path, i_o());
5575 result_mem->init_req(_fast_path, init_mem);
5576
5577 // Generate code for the slow case. We make a call to hashCode().
5578 set_control(_gvn.transform(slow_region));
5579 if (!stopped()) {
5580 // No need for PreserveJVMState, because we're using up the present state.
5581 set_all_memory(init_mem);
5582 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
5583 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
5584 Node* slow_result = set_results_for_java_call(slow_call);
5585 // this->control() comes from set_results_for_java_call
5586 result_reg->init_req(_slow_path, control());
5587 result_val->init_req(_slow_path, slow_result);
5588 result_io ->set_req(_slow_path, i_o());
5589 result_mem ->set_req(_slow_path, reset_memory());
5590 }
5591
5592 // Return the combined state.
5593 set_i_o( _gvn.transform(result_io) );
5594 set_all_memory( _gvn.transform(result_mem));
5595
5596 set_result(result_reg, result_val);
5597 return true;
5598 }
5599
5600 //---------------------------inline_native_getClass----------------------------
5601 // public final native Class<?> java.lang.Object.getClass();
5602 //
5603 // Build special case code for calls to getClass on an object.
5604 bool LibraryCallKit::inline_native_getClass() {
5605 Node* obj = argument(0);
5606 if (obj->is_InlineType()) {
5607 const Type* t = _gvn.type(obj);
5608 if (t->maybe_null()) {
5609 null_check(obj);
5610 }
5611 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror())));
5612 return true;
5613 }
5614 obj = null_check_receiver();
5615 if (stopped()) return true;
5616 set_result(load_mirror_from_klass(load_object_klass(obj)));
5617 return true;
5618 }
5619
5620 //-----------------inline_native_Reflection_getCallerClass---------------------
5621 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
5622 //
5623 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
5624 //
5625 // NOTE: This code must perform the same logic as JVM_GetCallerClass
5626 // in that it must skip particular security frames and checks for
5627 // caller sensitive methods.
5628 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
5629 #ifndef PRODUCT
5630 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5631 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
5632 }
5633 #endif
5634
5635 if (!jvms()->has_method()) {
5636 #ifndef PRODUCT
5637 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5638 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
5639 }
5640 #endif
5641 return false;
5642 }
5643
5644 // Walk back up the JVM state to find the caller at the required
5645 // depth.
5646 JVMState* caller_jvms = jvms();
5647
5648 // Cf. JVM_GetCallerClass
5649 // NOTE: Start the loop at depth 1 because the current JVM state does
5650 // not include the Reflection.getCallerClass() frame.
5651 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
5652 ciMethod* m = caller_jvms->method();
5653 switch (n) {
5654 case 0:
5655 fatal("current JVM state does not include the Reflection.getCallerClass frame");
5656 break;
5657 case 1:
5658 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
5659 if (!m->caller_sensitive()) {
5660 #ifndef PRODUCT
5661 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5662 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
5663 }
5664 #endif
5665 return false; // bail-out; let JVM_GetCallerClass do the work
5666 }
5667 break;
5668 default:
5669 if (!m->is_ignored_by_security_stack_walk()) {
5670 // We have reached the desired frame; return the holder class.
5671 // Acquire method holder as java.lang.Class and push as constant.
5672 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
5673 ciInstance* caller_mirror = caller_klass->java_mirror();
5674 set_result(makecon(TypeInstPtr::make(caller_mirror)));
5675
5676 #ifndef PRODUCT
5677 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5678 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());
5679 tty->print_cr(" JVM state at this point:");
5680 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5681 ciMethod* m = jvms()->of_depth(i)->method();
5682 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5683 }
5684 }
5685 #endif
5686 return true;
5687 }
5688 break;
5689 }
5690 }
5691
5692 #ifndef PRODUCT
5693 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5694 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
5695 tty->print_cr(" JVM state at this point:");
5696 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5697 ciMethod* m = jvms()->of_depth(i)->method();
5698 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5699 }
5700 }
5701 #endif
5702
5703 return false; // bail-out; let JVM_GetCallerClass do the work
5704 }
5705
5706 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
5707 Node* arg = argument(0);
5708 Node* result = nullptr;
5709
5710 switch (id) {
5711 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
5712 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
5713 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5714 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5715 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5716 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5717
5718 case vmIntrinsics::_doubleToLongBits: {
5719 // two paths (plus control) merge in a wood
5720 RegionNode *r = new RegionNode(3);
5721 Node *phi = new PhiNode(r, TypeLong::LONG);
5722
5723 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5724 // Build the boolean node
5725 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5726
5727 // Branch either way.
5728 // NaN case is less traveled, which makes all the difference.
5729 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5730 Node *opt_isnan = _gvn.transform(ifisnan);
5731 assert( opt_isnan->is_If(), "Expect an IfNode");
5732 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5733 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5734
5735 set_control(iftrue);
5736
5737 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5738 Node *slow_result = longcon(nan_bits); // return NaN
5739 phi->init_req(1, _gvn.transform( slow_result ));
5740 r->init_req(1, iftrue);
5741
5742 // Else fall through
5743 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5744 set_control(iffalse);
5745
5746 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5747 r->init_req(2, iffalse);
5748
5749 // Post merge
5750 set_control(_gvn.transform(r));
5751 record_for_igvn(r);
5752
5753 C->set_has_split_ifs(true); // Has chance for split-if optimization
5754 result = phi;
5755 assert(result->bottom_type()->isa_long(), "must be");
5756 break;
5757 }
5758
5759 case vmIntrinsics::_floatToIntBits: {
5760 // two paths (plus control) merge in a wood
5761 RegionNode *r = new RegionNode(3);
5762 Node *phi = new PhiNode(r, TypeInt::INT);
5763
5764 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5765 // Build the boolean node
5766 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5767
5768 // Branch either way.
5769 // NaN case is less traveled, which makes all the difference.
5770 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5771 Node *opt_isnan = _gvn.transform(ifisnan);
5772 assert( opt_isnan->is_If(), "Expect an IfNode");
5773 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5774 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5775
5776 set_control(iftrue);
5777
5778 static const jint nan_bits = 0x7fc00000;
5779 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5780 phi->init_req(1, _gvn.transform( slow_result ));
5781 r->init_req(1, iftrue);
5782
5783 // Else fall through
5784 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5785 set_control(iffalse);
5786
5787 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5788 r->init_req(2, iffalse);
5789
5790 // Post merge
5791 set_control(_gvn.transform(r));
5792 record_for_igvn(r);
5793
5794 C->set_has_split_ifs(true); // Has chance for split-if optimization
5795 result = phi;
5796 assert(result->bottom_type()->isa_int(), "must be");
5797 break;
5798 }
5799
5800 default:
5801 fatal_unexpected_iid(id);
5802 break;
5803 }
5804 set_result(_gvn.transform(result));
5805 return true;
5806 }
5807
5808 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5809 Node* arg = argument(0);
5810 Node* result = nullptr;
5811
5812 switch (id) {
5813 case vmIntrinsics::_floatIsInfinite:
5814 result = new IsInfiniteFNode(arg);
5815 break;
5816 case vmIntrinsics::_floatIsFinite:
5817 result = new IsFiniteFNode(arg);
5818 break;
5819 case vmIntrinsics::_doubleIsInfinite:
5820 result = new IsInfiniteDNode(arg);
5821 break;
5822 case vmIntrinsics::_doubleIsFinite:
5823 result = new IsFiniteDNode(arg);
5824 break;
5825 default:
5826 fatal_unexpected_iid(id);
5827 break;
5828 }
5829 set_result(_gvn.transform(result));
5830 return true;
5831 }
5832
5833 //----------------------inline_unsafe_copyMemory-------------------------
5834 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5835
5836 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5837 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5838 const Type* base_t = gvn.type(base);
5839
5840 bool in_native = (base_t == TypePtr::NULL_PTR);
5841 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5842 bool is_mixed = !in_heap && !in_native;
5843
5844 if (is_mixed) {
5845 return true; // mixed accesses can touch both on-heap and off-heap memory
5846 }
5847 if (in_heap) {
5848 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5849 if (!is_prim_array) {
5850 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5851 // there's not enough type information available to determine proper memory slice for it.
5852 return true;
5853 }
5854 }
5855 return false;
5856 }
5857
5858 bool LibraryCallKit::inline_unsafe_copyMemory() {
5859 if (callee()->is_static()) return false; // caller must have the capability!
5860 null_check_receiver(); // null-check receiver
5861 if (stopped()) return true;
5862
5863 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5864
5865 Node* src_base = argument(1); // type: oop
5866 Node* src_off = ConvL2X(argument(2)); // type: long
5867 Node* dst_base = argument(4); // type: oop
5868 Node* dst_off = ConvL2X(argument(5)); // type: long
5869 Node* size = ConvL2X(argument(7)); // type: long
5870
5871 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5872 "fieldOffset must be byte-scaled");
5873
5874 Node* src_addr = make_unsafe_address(src_base, src_off);
5875 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5876
5877 Node* thread = _gvn.transform(new ThreadLocalNode());
5878 Node* doing_unsafe_access_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5879 BasicType doing_unsafe_access_bt = T_BYTE;
5880 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5881
5882 // update volatile field
5883 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5884
5885 int flags = RC_LEAF | RC_NO_FP;
5886
5887 const TypePtr* dst_type = TypePtr::BOTTOM;
5888
5889 // Adjust memory effects of the runtime call based on input values.
5890 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5891 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5892 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5893
5894 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5895 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5896 flags |= RC_NARROW_MEM; // narrow in memory
5897 }
5898 }
5899
5900 // Call it. Note that the length argument is not scaled.
5901 make_runtime_call(flags,
5902 OptoRuntime::fast_arraycopy_Type(),
5903 StubRoutines::unsafe_arraycopy(),
5904 "unsafe_arraycopy",
5905 dst_type,
5906 src_addr, dst_addr, size XTOP);
5907
5908 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5909
5910 return true;
5911 }
5912
5913 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5914 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5915 bool LibraryCallKit::inline_unsafe_setMemory() {
5916 if (callee()->is_static()) return false; // caller must have the capability!
5917 null_check_receiver(); // null-check receiver
5918 if (stopped()) return true;
5919
5920 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5921
5922 Node* dst_base = argument(1); // type: oop
5923 Node* dst_off = ConvL2X(argument(2)); // type: long
5924 Node* size = ConvL2X(argument(4)); // type: long
5925 Node* byte = argument(6); // type: byte
5926
5927 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5928 "fieldOffset must be byte-scaled");
5929
5930 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5931
5932 Node* thread = _gvn.transform(new ThreadLocalNode());
5933 Node* doing_unsafe_access_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5934 BasicType doing_unsafe_access_bt = T_BYTE;
5935 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5936
5937 // update volatile field
5938 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5939
5940 int flags = RC_LEAF | RC_NO_FP;
5941
5942 const TypePtr* dst_type = TypePtr::BOTTOM;
5943
5944 // Adjust memory effects of the runtime call based on input values.
5945 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5946 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5947
5948 flags |= RC_NARROW_MEM; // narrow in memory
5949 }
5950
5951 // Call it. Note that the length argument is not scaled.
5952 make_runtime_call(flags,
5953 OptoRuntime::unsafe_setmemory_Type(),
5954 StubRoutines::unsafe_setmemory(),
5955 "unsafe_setmemory",
5956 dst_type,
5957 dst_addr, size XTOP, byte);
5958
5959 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5960
5961 return true;
5962 }
5963
5964 #undef XTOP
5965
5966 //------------------------clone_coping-----------------------------------
5967 // Helper function for inline_native_clone.
5968 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5969 assert(obj_size != nullptr, "");
5970 Node* raw_obj = alloc_obj->in(1);
5971 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5972
5973 AllocateNode* alloc = nullptr;
5974 if (ReduceBulkZeroing &&
5975 // If we are implementing an array clone without knowing its source type
5976 // (can happen when compiling the array-guarded branch of a reflective
5977 // Object.clone() invocation), initialize the array within the allocation.
5978 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5979 // to a runtime clone call that assumes fully initialized source arrays.
5980 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5981 // We will be completely responsible for initializing this object -
5982 // mark Initialize node as complete.
5983 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5984 // The object was just allocated - there should be no any stores!
5985 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5986 // Mark as complete_with_arraycopy so that on AllocateNode
5987 // expansion, we know this AllocateNode is initialized by an array
5988 // copy and a StoreStore barrier exists after the array copy.
5989 alloc->initialization()->set_complete_with_arraycopy();
5990 }
5991
5992 Node* size = _gvn.transform(obj_size);
5993 access_clone(obj, alloc_obj, size, is_array);
5994
5995 // Do not let reads from the cloned object float above the arraycopy.
5996 if (alloc != nullptr) {
5997 // Do not let stores that initialize this object be reordered with
5998 // a subsequent store that would make this object accessible by
5999 // other threads.
6000 // Record what AllocateNode this StoreStore protects so that
6001 // escape analysis can go from the MemBarStoreStoreNode to the
6002 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
6003 // based on the escape status of the AllocateNode.
6004 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
6005 } else {
6006 insert_mem_bar(Op_MemBarCPUOrder);
6007 }
6008 }
6009
6010 //------------------------inline_native_clone----------------------------
6011 // protected native Object java.lang.Object.clone();
6012 //
6013 // Here are the simple edge cases:
6014 // null receiver => normal trap
6015 // virtual and clone was overridden => slow path to out-of-line clone
6016 // not cloneable or finalizer => slow path to out-of-line Object.clone
6017 //
6018 // The general case has two steps, allocation and copying.
6019 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
6020 //
6021 // Copying also has two cases, oop arrays and everything else.
6022 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
6023 // Everything else uses the tight inline loop supplied by CopyArrayNode.
6024 //
6025 // These steps fold up nicely if and when the cloned object's klass
6026 // can be sharply typed as an object array, a type array, or an instance.
6027 //
6028 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
6029 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
6030 return false;
6031 }
6032
6033 PhiNode* result_val;
6034
6035 // Set the reexecute bit for the interpreter to reexecute
6036 // the bytecode that invokes Object.clone if deoptimization happens.
6037 { PreserveReexecuteState preexecs(this);
6038 jvms()->set_should_reexecute(true);
6039
6040 Node* obj = argument(0);
6041 obj = null_check_receiver();
6042 if (stopped()) return true;
6043
6044 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
6045 if (obj_type->is_inlinetypeptr()) {
6046 // If the object to clone is an inline type, we can simply return it (i.e. a nop) since inline types have
6047 // no identity. But we first need to check whether the value class is actually implementing the Cloneable
6048 // interface. If not, we trap.
6049 if (obj_type->inline_klass()->is_cloneable()) {
6050 set_result(obj);
6051 } else {
6052 uncommon_trap(Deoptimization::Reason_intrinsic,
6053 Deoptimization::Action_maybe_recompile);
6054 }
6055 return true;
6056 }
6057
6058 // If we are going to clone an instance, we need its exact type to
6059 // know the number and types of fields to convert the clone to
6060 // loads/stores. Maybe a speculative type can help us.
6061 if (!obj_type->klass_is_exact() &&
6062 obj_type->speculative_type() != nullptr &&
6063 obj_type->speculative_type()->is_instance_klass() &&
6064 !obj_type->speculative_type()->is_inlinetype()) {
6065 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
6066 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
6067 !spec_ik->has_injected_fields()) {
6068 if (!obj_type->isa_instptr() ||
6069 obj_type->is_instptr()->instance_klass()->has_subklass()) {
6070 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
6071 }
6072 }
6073 }
6074
6075 // Conservatively insert a memory barrier on all memory slices.
6076 // Do not let writes into the original float below the clone.
6077 insert_mem_bar(Op_MemBarCPUOrder);
6078
6079 // paths into result_reg:
6080 enum {
6081 _slow_path = 1, // out-of-line call to clone method (virtual or not)
6082 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
6083 _array_path, // plain array allocation, plus arrayof_long_arraycopy
6084 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
6085 PATH_LIMIT
6086 };
6087 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
6088 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
6089 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
6090 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
6091 record_for_igvn(result_reg);
6092
6093 Node* obj_klass = load_object_klass(obj);
6094 // We only go to the fast case code if we pass a number of guards.
6095 // The paths which do not pass are accumulated in the slow_region.
6096 RegionNode* slow_region = new RegionNode(1);
6097 record_for_igvn(slow_region);
6098
6099 Node* array_obj = obj;
6100 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
6101 if (array_ctl != nullptr) {
6102 // It's an array.
6103 PreserveJVMState pjvms(this);
6104 set_control(array_ctl);
6105
6106 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
6107 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr();
6108 if (UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) &&
6109 obj_type->can_be_inline_array() &&
6110 (ary_ptr == nullptr || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) {
6111 // Flat inline type array may have object field that would require a
6112 // write barrier. Conservatively, go to slow path.
6113 generate_fair_guard(flat_array_test(obj_klass), slow_region);
6114 }
6115
6116 if (!stopped()) {
6117 Node* obj_length = load_array_length(array_obj);
6118 Node* array_size = nullptr; // Size of the array without object alignment padding.
6119 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
6120
6121 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
6122 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
6123 // If it is an oop array, it requires very special treatment,
6124 // because gc barriers are required when accessing the array.
6125 Node* is_obja = generate_refArray_guard(obj_klass, (RegionNode*)nullptr);
6126 if (is_obja != nullptr) {
6127 PreserveJVMState pjvms2(this);
6128 set_control(is_obja);
6129 // Generate a direct call to the right arraycopy function(s).
6130 // Clones are always tightly coupled.
6131 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
6132 ac->set_clone_oop_array();
6133 Node* n = _gvn.transform(ac);
6134 assert(n == ac, "cannot disappear");
6135 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
6136
6137 result_reg->init_req(_objArray_path, control());
6138 result_val->init_req(_objArray_path, alloc_obj);
6139 result_i_o ->set_req(_objArray_path, i_o());
6140 result_mem ->set_req(_objArray_path, reset_memory());
6141 }
6142 }
6143 // Otherwise, there are no barriers to worry about.
6144 // (We can dispense with card marks if we know the allocation
6145 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
6146 // causes the non-eden paths to take compensating steps to
6147 // simulate a fresh allocation, so that no further
6148 // card marks are required in compiled code to initialize
6149 // the object.)
6150
6151 if (!stopped()) {
6152 copy_to_clone(obj, alloc_obj, array_size, true);
6153
6154 // Present the results of the copy.
6155 result_reg->init_req(_array_path, control());
6156 result_val->init_req(_array_path, alloc_obj);
6157 result_i_o ->set_req(_array_path, i_o());
6158 result_mem ->set_req(_array_path, reset_memory());
6159 }
6160 }
6161 }
6162
6163 if (!stopped()) {
6164 // It's an instance (we did array above). Make the slow-path tests.
6165 // If this is a virtual call, we generate a funny guard. We grab
6166 // the vtable entry corresponding to clone() from the target object.
6167 // If the target method which we are calling happens to be the
6168 // Object clone() method, we pass the guard. We do not need this
6169 // guard for non-virtual calls; the caller is known to be the native
6170 // Object clone().
6171 if (is_virtual) {
6172 generate_virtual_guard(obj_klass, slow_region);
6173 }
6174
6175 // The object must be easily cloneable and must not have a finalizer.
6176 // Both of these conditions may be checked in a single test.
6177 // We could optimize the test further, but we don't care.
6178 generate_misc_flags_guard(obj_klass,
6179 // Test both conditions:
6180 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
6181 // Must be cloneable but not finalizer:
6182 KlassFlags::_misc_is_cloneable_fast,
6183 slow_region);
6184 }
6185
6186 if (!stopped()) {
6187 // It's an instance, and it passed the slow-path tests.
6188 PreserveJVMState pjvms(this);
6189 Node* obj_size = nullptr; // Total object size, including object alignment padding.
6190 // Need to deoptimize on exception from allocation since Object.clone intrinsic
6191 // is reexecuted if deoptimization occurs and there could be problems when merging
6192 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
6193 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
6194
6195 copy_to_clone(obj, alloc_obj, obj_size, false);
6196
6197 // Present the results of the slow call.
6198 result_reg->init_req(_instance_path, control());
6199 result_val->init_req(_instance_path, alloc_obj);
6200 result_i_o ->set_req(_instance_path, i_o());
6201 result_mem ->set_req(_instance_path, reset_memory());
6202 }
6203
6204 // Generate code for the slow case. We make a call to clone().
6205 set_control(_gvn.transform(slow_region));
6206 if (!stopped()) {
6207 PreserveJVMState pjvms(this);
6208 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
6209 // We need to deoptimize on exception (see comment above)
6210 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
6211 // this->control() comes from set_results_for_java_call
6212 result_reg->init_req(_slow_path, control());
6213 result_val->init_req(_slow_path, slow_result);
6214 result_i_o ->set_req(_slow_path, i_o());
6215 result_mem ->set_req(_slow_path, reset_memory());
6216 }
6217
6218 // Return the combined state.
6219 set_control( _gvn.transform(result_reg));
6220 set_i_o( _gvn.transform(result_i_o));
6221 set_all_memory( _gvn.transform(result_mem));
6222 } // original reexecute is set back here
6223
6224 set_result(_gvn.transform(result_val));
6225 return true;
6226 }
6227
6228 // If we have a tightly coupled allocation, the arraycopy may take care
6229 // of the array initialization. If one of the guards we insert between
6230 // the allocation and the arraycopy causes a deoptimization, an
6231 // uninitialized array will escape the compiled method. To prevent that
6232 // we set the JVM state for uncommon traps between the allocation and
6233 // the arraycopy to the state before the allocation so, in case of
6234 // deoptimization, we'll reexecute the allocation and the
6235 // initialization.
6236 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
6237 if (alloc != nullptr) {
6238 ciMethod* trap_method = alloc->jvms()->method();
6239 int trap_bci = alloc->jvms()->bci();
6240
6241 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6242 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
6243 // Make sure there's no store between the allocation and the
6244 // arraycopy otherwise visible side effects could be rexecuted
6245 // in case of deoptimization and cause incorrect execution.
6246 bool no_interfering_store = true;
6247 Node* mem = alloc->in(TypeFunc::Memory);
6248 if (mem->is_MergeMem()) {
6249 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
6250 Node* n = mms.memory();
6251 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
6252 assert(n->is_Store(), "what else?");
6253 no_interfering_store = false;
6254 break;
6255 }
6256 }
6257 } else {
6258 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
6259 Node* n = mms.memory();
6260 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
6261 assert(n->is_Store(), "what else?");
6262 no_interfering_store = false;
6263 break;
6264 }
6265 }
6266 }
6267
6268 if (no_interfering_store) {
6269 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6270
6271 JVMState* saved_jvms = jvms();
6272 saved_reexecute_sp = _reexecute_sp;
6273
6274 set_jvms(sfpt->jvms());
6275 _reexecute_sp = jvms()->sp();
6276
6277 return saved_jvms;
6278 }
6279 }
6280 }
6281 return nullptr;
6282 }
6283
6284 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
6285 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
6286 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
6287 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
6288 uint size = alloc->req();
6289 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
6290 old_jvms->set_map(sfpt);
6291 for (uint i = 0; i < size; i++) {
6292 sfpt->init_req(i, alloc->in(i));
6293 }
6294 int adjustment = 1;
6295 const TypeAryKlassPtr* ary_klass_ptr = alloc->in(AllocateNode::KlassNode)->bottom_type()->is_aryklassptr();
6296 if (ary_klass_ptr->is_null_free()) {
6297 // A null-free, tightly coupled array allocation can only come from LibraryCallKit::inline_newArray which
6298 // also requires the componentType and initVal on stack for re-execution.
6299 // Re-create and push the componentType.
6300 ciArrayKlass* klass = ary_klass_ptr->exact_klass()->as_array_klass();
6301 ciInstance* instance = klass->component_mirror_instance();
6302 const TypeInstPtr* t_instance = TypeInstPtr::make(instance);
6303 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), makecon(t_instance));
6304 adjustment++;
6305 }
6306 // re-push array length for deoptimization
6307 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment - 1, alloc->in(AllocateNode::ALength));
6308 if (ary_klass_ptr->is_null_free()) {
6309 // Re-create and push the initVal.
6310 Node* init_val = alloc->in(AllocateNode::InitValue);
6311 if (init_val == nullptr) {
6312 init_val = InlineTypeNode::make_all_zero(_gvn, ary_klass_ptr->elem()->is_instklassptr()->instance_klass()->as_inline_klass());
6313 } else if (UseCompressedOops) {
6314 init_val = _gvn.transform(new DecodeNNode(init_val, init_val->bottom_type()->make_ptr()));
6315 }
6316 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment, init_val);
6317 adjustment++;
6318 }
6319 old_jvms->set_sp(old_jvms->sp() + adjustment);
6320 old_jvms->set_monoff(old_jvms->monoff() + adjustment);
6321 old_jvms->set_scloff(old_jvms->scloff() + adjustment);
6322 old_jvms->set_endoff(old_jvms->endoff() + adjustment);
6323 old_jvms->set_should_reexecute(true);
6324
6325 sfpt->set_i_o(map()->i_o());
6326 sfpt->set_memory(map()->memory());
6327 sfpt->set_control(map()->control());
6328 return sfpt;
6329 }
6330
6331 // In case of a deoptimization, we restart execution at the
6332 // allocation, allocating a new array. We would leave an uninitialized
6333 // array in the heap that GCs wouldn't expect. Move the allocation
6334 // after the traps so we don't allocate the array if we
6335 // deoptimize. This is possible because tightly_coupled_allocation()
6336 // guarantees there's no observer of the allocated array at this point
6337 // and the control flow is simple enough.
6338 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
6339 int saved_reexecute_sp, uint new_idx) {
6340 if (saved_jvms_before_guards != nullptr && !stopped()) {
6341 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
6342
6343 assert(alloc != nullptr, "only with a tightly coupled allocation");
6344 // restore JVM state to the state at the arraycopy
6345 saved_jvms_before_guards->map()->set_control(map()->control());
6346 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
6347 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
6348 // If we've improved the types of some nodes (null check) while
6349 // emitting the guards, propagate them to the current state
6350 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
6351 set_jvms(saved_jvms_before_guards);
6352 _reexecute_sp = saved_reexecute_sp;
6353
6354 // Remove the allocation from above the guards
6355 CallProjections* callprojs = alloc->extract_projections(true);
6356 InitializeNode* init = alloc->initialization();
6357 Node* alloc_mem = alloc->in(TypeFunc::Memory);
6358 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
6359 init->replace_mem_projs_by(alloc_mem, C);
6360
6361 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
6362 // the allocation (i.e. is only valid if the allocation succeeds):
6363 // 1) replace CastIINode with AllocateArrayNode's length here
6364 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
6365 //
6366 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
6367 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
6368 Node* init_control = init->proj_out(TypeFunc::Control);
6369 Node* alloc_length = alloc->Ideal_length();
6370 #ifdef ASSERT
6371 Node* prev_cast = nullptr;
6372 #endif
6373 for (uint i = 0; i < init_control->outcnt(); i++) {
6374 Node* init_out = init_control->raw_out(i);
6375 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
6376 #ifdef ASSERT
6377 if (prev_cast == nullptr) {
6378 prev_cast = init_out;
6379 } else {
6380 if (prev_cast->cmp(*init_out) == false) {
6381 prev_cast->dump();
6382 init_out->dump();
6383 assert(false, "not equal CastIINode");
6384 }
6385 }
6386 #endif
6387 C->gvn_replace_by(init_out, alloc_length);
6388 }
6389 }
6390 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
6391
6392 // move the allocation here (after the guards)
6393 _gvn.hash_delete(alloc);
6394 alloc->set_req(TypeFunc::Control, control());
6395 alloc->set_req(TypeFunc::I_O, i_o());
6396 Node *mem = reset_memory();
6397 set_all_memory(mem);
6398 alloc->set_req(TypeFunc::Memory, mem);
6399 set_control(init->proj_out_or_null(TypeFunc::Control));
6400 set_i_o(callprojs->fallthrough_ioproj);
6401
6402 // Update memory as done in GraphKit::set_output_for_allocation()
6403 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
6404 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_exact_instance_type();
6405 if (ary_type->isa_aryptr() && length_type != nullptr) {
6406 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
6407 }
6408 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
6409 int elemidx = C->get_alias_index(telemref);
6410 // Need to properly move every memory projection for the Initialize
6411 #ifdef ASSERT
6412 int mark_idx = C->get_alias_index(ary_type->add_offset(oopDesc::mark_offset_in_bytes()));
6413 int klass_idx = C->get_alias_index(ary_type->add_offset(oopDesc::klass_offset_in_bytes()));
6414 #endif
6415 auto move_proj = [&](ProjNode* proj) {
6416 int alias_idx = C->get_alias_index(proj->adr_type());
6417 assert(alias_idx == Compile::AliasIdxRaw ||
6418 alias_idx == elemidx ||
6419 alias_idx == mark_idx ||
6420 alias_idx == klass_idx, "should be raw memory or array element type");
6421 set_memory(proj, alias_idx);
6422 };
6423 init->for_each_proj(move_proj, TypeFunc::Memory);
6424
6425 Node* allocx = _gvn.transform(alloc);
6426 assert(allocx == alloc, "where has the allocation gone?");
6427 assert(dest->is_CheckCastPP(), "not an allocation result?");
6428
6429 _gvn.hash_delete(dest);
6430 dest->set_req(0, control());
6431 Node* destx = _gvn.transform(dest);
6432 assert(destx == dest, "where has the allocation result gone?");
6433
6434 array_ideal_length(alloc, ary_type, true);
6435 }
6436 }
6437
6438 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
6439 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
6440 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
6441 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
6442 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
6443 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
6444 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
6445 JVMState* saved_jvms_before_guards) {
6446 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
6447 // There is at least one unrelated uncommon trap which needs to be replaced.
6448 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6449
6450 JVMState* saved_jvms = jvms();
6451 const int saved_reexecute_sp = _reexecute_sp;
6452 set_jvms(sfpt->jvms());
6453 _reexecute_sp = jvms()->sp();
6454
6455 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
6456
6457 // Restore state
6458 set_jvms(saved_jvms);
6459 _reexecute_sp = saved_reexecute_sp;
6460 }
6461 }
6462
6463 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
6464 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
6465 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
6466 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
6467 while (if_proj->is_IfProj()) {
6468 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
6469 if (uncommon_trap != nullptr) {
6470 create_new_uncommon_trap(uncommon_trap);
6471 }
6472 assert(if_proj->in(0)->is_If(), "must be If");
6473 if_proj = if_proj->in(0)->in(0);
6474 }
6475 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
6476 "must have reached control projection of init node");
6477 }
6478
6479 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
6480 const int trap_request = uncommon_trap_call->uncommon_trap_request();
6481 assert(trap_request != 0, "no valid UCT trap request");
6482 PreserveJVMState pjvms(this);
6483 set_control(uncommon_trap_call->in(0));
6484 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
6485 Deoptimization::trap_request_action(trap_request));
6486 assert(stopped(), "Should be stopped");
6487 _gvn.hash_delete(uncommon_trap_call);
6488 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
6489 }
6490
6491 // Common checks for array sorting intrinsics arguments.
6492 // Returns `true` if checks passed.
6493 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
6494 // check address of the class
6495 if (elementType == nullptr || elementType->is_top()) {
6496 return false; // dead path
6497 }
6498 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
6499 if (elem_klass == nullptr) {
6500 return false; // dead path
6501 }
6502 // java_mirror_type() returns non-null for compile-time Class constants only
6503 ciType* elem_type = elem_klass->java_mirror_type();
6504 if (elem_type == nullptr) {
6505 return false;
6506 }
6507 bt = elem_type->basic_type();
6508 // Disable the intrinsic if the CPU does not support SIMD sort
6509 if (!Matcher::supports_simd_sort(bt)) {
6510 return false;
6511 }
6512 // check address of the array
6513 if (obj == nullptr || obj->is_top()) {
6514 return false; // dead path
6515 }
6516 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
6517 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
6518 return false; // failed input validation
6519 }
6520 return true;
6521 }
6522
6523 //------------------------------inline_array_partition-----------------------
6524 bool LibraryCallKit::inline_array_partition() {
6525 address stubAddr = StubRoutines::select_array_partition_function();
6526 if (stubAddr == nullptr) {
6527 return false; // Intrinsic's stub is not implemented on this platform
6528 }
6529 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
6530
6531 // no receiver because it is a static method
6532 Node* elementType = argument(0);
6533 Node* obj = argument(1);
6534 Node* offset = argument(2); // long
6535 Node* fromIndex = argument(4);
6536 Node* toIndex = argument(5);
6537 Node* indexPivot1 = argument(6);
6538 Node* indexPivot2 = argument(7);
6539 // PartitionOperation: argument(8) is ignored
6540
6541 Node* pivotIndices = nullptr;
6542 BasicType bt = T_ILLEGAL;
6543
6544 if (!check_array_sort_arguments(elementType, obj, bt)) {
6545 return false;
6546 }
6547 null_check(obj);
6548 // If obj is dead, only null-path is taken.
6549 if (stopped()) {
6550 return true;
6551 }
6552 // Set the original stack and the reexecute bit for the interpreter to reexecute
6553 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
6554 { PreserveReexecuteState preexecs(this);
6555 jvms()->set_should_reexecute(true);
6556
6557 Node* obj_adr = make_unsafe_address(obj, offset);
6558
6559 // create the pivotIndices array of type int and size = 2
6560 Node* size = intcon(2);
6561 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
6562 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
6563 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
6564 guarantee(alloc != nullptr, "created above");
6565 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
6566
6567 // pass the basic type enum to the stub
6568 Node* elemType = intcon(bt);
6569
6570 // Call the stub
6571 const char *stubName = "array_partition_stub";
6572 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
6573 stubAddr, stubName, TypePtr::BOTTOM,
6574 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
6575 indexPivot1, indexPivot2);
6576
6577 } // original reexecute is set back here
6578
6579 if (!stopped()) {
6580 set_result(pivotIndices);
6581 }
6582
6583 return true;
6584 }
6585
6586
6587 //------------------------------inline_array_sort-----------------------
6588 bool LibraryCallKit::inline_array_sort() {
6589 address stubAddr = StubRoutines::select_arraysort_function();
6590 if (stubAddr == nullptr) {
6591 return false; // Intrinsic's stub is not implemented on this platform
6592 }
6593 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
6594
6595 // no receiver because it is a static method
6596 Node* elementType = argument(0);
6597 Node* obj = argument(1);
6598 Node* offset = argument(2); // long
6599 Node* fromIndex = argument(4);
6600 Node* toIndex = argument(5);
6601 // SortOperation: argument(6) is ignored
6602
6603 BasicType bt = T_ILLEGAL;
6604
6605 if (!check_array_sort_arguments(elementType, obj, bt)) {
6606 return false;
6607 }
6608 null_check(obj);
6609 // If obj is dead, only null-path is taken.
6610 if (stopped()) {
6611 return true;
6612 }
6613 Node* obj_adr = make_unsafe_address(obj, offset);
6614
6615 // pass the basic type enum to the stub
6616 Node* elemType = intcon(bt);
6617
6618 // Call the stub.
6619 const char *stubName = "arraysort_stub";
6620 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
6621 stubAddr, stubName, TypePtr::BOTTOM,
6622 obj_adr, elemType, fromIndex, toIndex);
6623
6624 return true;
6625 }
6626
6627
6628 //------------------------------inline_arraycopy-----------------------
6629 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
6630 // Object dest, int destPos,
6631 // int length);
6632 bool LibraryCallKit::inline_arraycopy() {
6633 // Get the arguments.
6634 Node* src = argument(0); // type: oop
6635 Node* src_offset = argument(1); // type: int
6636 Node* dest = argument(2); // type: oop
6637 Node* dest_offset = argument(3); // type: int
6638 Node* length = argument(4); // type: int
6639
6640 uint new_idx = C->unique();
6641
6642 // Check for allocation before we add nodes that would confuse
6643 // tightly_coupled_allocation()
6644 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
6645
6646 int saved_reexecute_sp = -1;
6647 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
6648 // See arraycopy_restore_alloc_state() comment
6649 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
6650 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
6651 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
6652 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
6653
6654 // The following tests must be performed
6655 // (1) src and dest are arrays.
6656 // (2) src and dest arrays must have elements of the same BasicType
6657 // (3) src and dest must not be null.
6658 // (4) src_offset must not be negative.
6659 // (5) dest_offset must not be negative.
6660 // (6) length must not be negative.
6661 // (7) src_offset + length must not exceed length of src.
6662 // (8) dest_offset + length must not exceed length of dest.
6663 // (9) each element of an oop array must be assignable
6664
6665 // (3) src and dest must not be null.
6666 // always do this here because we need the JVM state for uncommon traps
6667 Node* null_ctl = top();
6668 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
6669 assert(null_ctl->is_top(), "no null control here");
6670 dest = null_check(dest, T_ARRAY);
6671
6672 if (!can_emit_guards) {
6673 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
6674 // guards but the arraycopy node could still take advantage of a
6675 // tightly allocated allocation. tightly_coupled_allocation() is
6676 // called again to make sure it takes the null check above into
6677 // account: the null check is mandatory and if it caused an
6678 // uncommon trap to be emitted then the allocation can't be
6679 // considered tightly coupled in this context.
6680 alloc = tightly_coupled_allocation(dest);
6681 }
6682
6683 bool validated = false;
6684
6685 const Type* src_type = _gvn.type(src);
6686 const Type* dest_type = _gvn.type(dest);
6687 const TypeAryPtr* top_src = src_type->isa_aryptr();
6688 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6689
6690 // Do we have the type of src?
6691 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6692 // Do we have the type of dest?
6693 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6694 // Is the type for src from speculation?
6695 bool src_spec = false;
6696 // Is the type for dest from speculation?
6697 bool dest_spec = false;
6698
6699 if ((!has_src || !has_dest) && can_emit_guards) {
6700 // We don't have sufficient type information, let's see if
6701 // speculative types can help. We need to have types for both src
6702 // and dest so that it pays off.
6703
6704 // Do we already have or could we have type information for src
6705 bool could_have_src = has_src;
6706 // Do we already have or could we have type information for dest
6707 bool could_have_dest = has_dest;
6708
6709 ciKlass* src_k = nullptr;
6710 if (!has_src) {
6711 src_k = src_type->speculative_type_not_null();
6712 if (src_k != nullptr && src_k->is_array_klass()) {
6713 could_have_src = true;
6714 }
6715 }
6716
6717 ciKlass* dest_k = nullptr;
6718 if (!has_dest) {
6719 dest_k = dest_type->speculative_type_not_null();
6720 if (dest_k != nullptr && dest_k->is_array_klass()) {
6721 could_have_dest = true;
6722 }
6723 }
6724
6725 if (could_have_src && could_have_dest) {
6726 // This is going to pay off so emit the required guards
6727 if (!has_src) {
6728 src = maybe_cast_profiled_obj(src, src_k, true);
6729 src_type = _gvn.type(src);
6730 top_src = src_type->isa_aryptr();
6731 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6732 src_spec = true;
6733 }
6734 if (!has_dest) {
6735 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6736 dest_type = _gvn.type(dest);
6737 top_dest = dest_type->isa_aryptr();
6738 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6739 dest_spec = true;
6740 }
6741 }
6742 }
6743
6744 if (has_src && has_dest && can_emit_guards) {
6745 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
6746 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
6747 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
6748 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
6749
6750 if (src_elem == dest_elem && top_src->is_flat() == top_dest->is_flat() && src_elem == T_OBJECT) {
6751 // If both arrays are object arrays then having the exact types
6752 // for both will remove the need for a subtype check at runtime
6753 // before the call and may make it possible to pick a faster copy
6754 // routine (without a subtype check on every element)
6755 // Do we have the exact type of src?
6756 bool could_have_src = src_spec;
6757 // Do we have the exact type of dest?
6758 bool could_have_dest = dest_spec;
6759 ciKlass* src_k = nullptr;
6760 ciKlass* dest_k = nullptr;
6761 if (!src_spec) {
6762 src_k = src_type->speculative_type_not_null();
6763 if (src_k != nullptr && src_k->is_array_klass()) {
6764 could_have_src = true;
6765 }
6766 }
6767 if (!dest_spec) {
6768 dest_k = dest_type->speculative_type_not_null();
6769 if (dest_k != nullptr && dest_k->is_array_klass()) {
6770 could_have_dest = true;
6771 }
6772 }
6773 if (could_have_src && could_have_dest) {
6774 // If we can have both exact types, emit the missing guards
6775 if (could_have_src && !src_spec) {
6776 src = maybe_cast_profiled_obj(src, src_k, true);
6777 src_type = _gvn.type(src);
6778 top_src = src_type->isa_aryptr();
6779 }
6780 if (could_have_dest && !dest_spec) {
6781 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6782 dest_type = _gvn.type(dest);
6783 top_dest = dest_type->isa_aryptr();
6784 }
6785 }
6786 }
6787 }
6788
6789 ciMethod* trap_method = method();
6790 int trap_bci = bci();
6791 if (saved_jvms_before_guards != nullptr) {
6792 trap_method = alloc->jvms()->method();
6793 trap_bci = alloc->jvms()->bci();
6794 }
6795
6796 bool negative_length_guard_generated = false;
6797
6798 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6799 can_emit_guards && !src->is_top() && !dest->is_top()) {
6800 // validate arguments: enables transformation the ArrayCopyNode
6801 validated = true;
6802
6803 RegionNode* slow_region = new RegionNode(1);
6804 record_for_igvn(slow_region);
6805
6806 // (1) src and dest are arrays.
6807 generate_non_array_guard(load_object_klass(src), slow_region, &src);
6808 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
6809
6810 // (2) src and dest arrays must have elements of the same BasicType
6811 // done at macro expansion or at Ideal transformation time
6812
6813 // (4) src_offset must not be negative.
6814 generate_negative_guard(src_offset, slow_region);
6815
6816 // (5) dest_offset must not be negative.
6817 generate_negative_guard(dest_offset, slow_region);
6818
6819 // (7) src_offset + length must not exceed length of src.
6820 generate_limit_guard(src_offset, length,
6821 load_array_length(src),
6822 slow_region);
6823
6824 // (8) dest_offset + length must not exceed length of dest.
6825 generate_limit_guard(dest_offset, length,
6826 load_array_length(dest),
6827 slow_region);
6828
6829 // (6) length must not be negative.
6830 // This is also checked in generate_arraycopy() during macro expansion, but
6831 // we also have to check it here for the case where the ArrayCopyNode will
6832 // be eliminated by Escape Analysis.
6833 if (EliminateAllocations) {
6834 generate_negative_guard(length, slow_region);
6835 negative_length_guard_generated = true;
6836 }
6837
6838 // (9) each element of an oop array must be assignable
6839 Node* dest_klass = load_object_klass(dest);
6840 Node* refined_dest_klass = dest_klass;
6841 if (src != dest) {
6842 dest_klass = load_non_refined_array_klass(refined_dest_klass);
6843 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6844 slow_region->add_req(not_subtype_ctrl);
6845 }
6846
6847 // TODO 8251971 Improve this. What about atomicity? Make sure this is always folded for type arrays.
6848 // If destination is null-restricted, source must be null-restricted as well: src_null_restricted || !dst_null_restricted
6849 Node* src_klass = load_object_klass(src);
6850 Node* adr_prop_src = basic_plus_adr(top(), src_klass, in_bytes(ArrayKlass::properties_offset()));
6851 Node* prop_src = _gvn.transform(LoadNode::make(_gvn, control(), immutable_memory(), adr_prop_src,
6852 _gvn.type(adr_prop_src)->is_ptr(), TypeInt::INT, T_INT,
6853 MemNode::unordered));
6854 Node* adr_prop_dest = basic_plus_adr(top(), refined_dest_klass, in_bytes(ArrayKlass::properties_offset()));
6855 Node* prop_dest = _gvn.transform(LoadNode::make(_gvn, control(), immutable_memory(), adr_prop_dest,
6856 _gvn.type(adr_prop_dest)->is_ptr(), TypeInt::INT, T_INT,
6857 MemNode::unordered));
6858
6859 const ArrayProperties props_null_restricted = ArrayProperties::Default().with_null_restricted();
6860 jint props_value = (jint)props_null_restricted.value();
6861
6862 prop_dest = _gvn.transform(new XorINode(prop_dest, intcon(props_value)));
6863 prop_src = _gvn.transform(new OrINode(prop_dest, prop_src));
6864 prop_src = _gvn.transform(new AndINode(prop_src, intcon(props_value)));
6865
6866 Node* chk = _gvn.transform(new CmpINode(prop_src, intcon(props_value)));
6867 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::ne));
6868 generate_fair_guard(tst, slow_region);
6869
6870 // TODO 8251971 This is too strong
6871 generate_fair_guard(flat_array_test(src), slow_region);
6872 generate_fair_guard(flat_array_test(dest), slow_region);
6873
6874 {
6875 PreserveJVMState pjvms(this);
6876 set_control(_gvn.transform(slow_region));
6877 uncommon_trap(Deoptimization::Reason_intrinsic,
6878 Deoptimization::Action_make_not_entrant);
6879 assert(stopped(), "Should be stopped");
6880 }
6881
6882 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->isa_klassptr();
6883 if (dest_klass_t == nullptr) {
6884 // refined_dest_klass may not be an array, which leads to dest_klass being top. This means we
6885 // are in a dead path.
6886 uncommon_trap(Deoptimization::Reason_intrinsic,
6887 Deoptimization::Action_make_not_entrant);
6888 return true;
6889 }
6890
6891 const Type* toop = dest_klass_t->as_subtype_instance_type();
6892 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6893 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6894 }
6895
6896 if (stopped()) {
6897 return true;
6898 }
6899
6900 Node* dest_klass = load_object_klass(dest);
6901 dest_klass = load_non_refined_array_klass(dest_klass);
6902
6903 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6904 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6905 // so the compiler has a chance to eliminate them: during macro expansion,
6906 // we have to set their control (CastPP nodes are eliminated).
6907 load_object_klass(src), dest_klass,
6908 load_array_length(src), load_array_length(dest));
6909
6910 ac->set_arraycopy(validated);
6911
6912 Node* n = _gvn.transform(ac);
6913 if (n == ac) {
6914 ac->connect_outputs(this);
6915 } else {
6916 assert(validated, "shouldn't transform if all arguments not validated");
6917 set_all_memory(n);
6918 }
6919 clear_upper_avx();
6920
6921
6922 return true;
6923 }
6924
6925
6926 // Helper function which determines if an arraycopy immediately follows
6927 // an allocation, with no intervening tests or other escapes for the object.
6928 AllocateArrayNode*
6929 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6930 if (stopped()) return nullptr; // no fast path
6931 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6932
6933 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6934 if (alloc == nullptr) return nullptr;
6935
6936 Node* rawmem = memory(Compile::AliasIdxRaw);
6937 // Is the allocation's memory state untouched?
6938 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6939 // Bail out if there have been raw-memory effects since the allocation.
6940 // (Example: There might have been a call or safepoint.)
6941 return nullptr;
6942 }
6943 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6944 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6945 return nullptr;
6946 }
6947
6948 // There must be no unexpected observers of this allocation.
6949 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6950 Node* obs = ptr->fast_out(i);
6951 if (obs != this->map()) {
6952 return nullptr;
6953 }
6954 }
6955
6956 // This arraycopy must unconditionally follow the allocation of the ptr.
6957 Node* alloc_ctl = ptr->in(0);
6958 Node* ctl = control();
6959 while (ctl != alloc_ctl) {
6960 // There may be guards which feed into the slow_region.
6961 // Any other control flow means that we might not get a chance
6962 // to finish initializing the allocated object.
6963 // Various low-level checks bottom out in uncommon traps. These
6964 // are considered safe since we've already checked above that
6965 // there is no unexpected observer of this allocation.
6966 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6967 assert(ctl->in(0)->is_If(), "must be If");
6968 ctl = ctl->in(0)->in(0);
6969 } else {
6970 return nullptr;
6971 }
6972 }
6973
6974 // If we get this far, we have an allocation which immediately
6975 // precedes the arraycopy, and we can take over zeroing the new object.
6976 // The arraycopy will finish the initialization, and provide
6977 // a new control state to which we will anchor the destination pointer.
6978
6979 return alloc;
6980 }
6981
6982 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6983 if (node->is_IfProj()) {
6984 IfProjNode* other_proj = node->as_IfProj()->other_if_proj();
6985 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6986 Node* obs = other_proj->fast_out(j);
6987 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6988 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6989 return obs->as_CallStaticJava();
6990 }
6991 }
6992 }
6993 return nullptr;
6994 }
6995
6996 //-------------inline_encodeISOArray-----------------------------------
6997 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6998 // int java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6999 // int java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
7000 // encode char[] to byte[] in ISO_8859_1 or ASCII
7001 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
7002 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
7003 // no receiver since it is static method
7004 Node *src = argument(0);
7005 Node *src_offset = argument(1);
7006 Node *dst = argument(2);
7007 Node *dst_offset = argument(3);
7008 Node *length = argument(4);
7009
7010 // Cast source & target arrays to not-null
7011 src = must_be_not_null(src, true);
7012 dst = must_be_not_null(dst, true);
7013 if (stopped()) {
7014 return true;
7015 }
7016
7017 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7018 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
7019 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
7020 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
7021 // failed array check
7022 return false;
7023 }
7024
7025 // Figure out the size and type of the elements we will be copying.
7026 BasicType src_elem = src_type->elem()->array_element_basic_type();
7027 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
7028 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
7029 return false;
7030 }
7031
7032 // Check source & target bounds
7033 RegionNode* bailout = create_bailout();
7034 generate_string_range_check(src, src_offset, length, src_elem == T_BYTE, bailout);
7035 generate_string_range_check(dst, dst_offset, length, false, bailout);
7036 if (check_bailout(bailout)) {
7037 return true;
7038 }
7039
7040 Node* src_start = array_element_address(src, src_offset, T_CHAR);
7041 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
7042 // 'src_start' points to src array + scaled offset
7043 // 'dst_start' points to dst array + scaled offset
7044
7045 // See GraphKit::compress_string
7046 const TypePtr* adr_type;
7047 Node* mem = capture_memory(adr_type, src_type, dst_type);
7048 Node* enc = new EncodeISOArrayNode(control(), mem, adr_type, src_start, dst_start, length, ascii);
7049 enc = _gvn.transform(enc);
7050 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
7051 memory_effect(res_mem, src_type, dst_type);
7052
7053 set_result(enc);
7054 clear_upper_avx();
7055
7056 return true;
7057 }
7058
7059 //-------------inline_multiplyToLen-----------------------------------
7060 bool LibraryCallKit::inline_multiplyToLen() {
7061 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
7062
7063 address stubAddr = StubRoutines::multiplyToLen();
7064 if (stubAddr == nullptr) {
7065 return false; // Intrinsic's stub is not implemented on this platform
7066 }
7067 const char* stubName = "multiplyToLen";
7068
7069 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
7070
7071 // no receiver because it is a static method
7072 Node* x = argument(0);
7073 Node* xlen = argument(1);
7074 Node* y = argument(2);
7075 Node* ylen = argument(3);
7076 Node* z = argument(4);
7077
7078 x = must_be_not_null(x, true);
7079 y = must_be_not_null(y, true);
7080
7081 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
7082 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
7083 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
7084 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
7085 // failed array check
7086 return false;
7087 }
7088
7089 BasicType x_elem = x_type->elem()->array_element_basic_type();
7090 BasicType y_elem = y_type->elem()->array_element_basic_type();
7091 if (x_elem != T_INT || y_elem != T_INT) {
7092 return false;
7093 }
7094
7095 Node* x_start = array_element_address(x, intcon(0), x_elem);
7096 Node* y_start = array_element_address(y, intcon(0), y_elem);
7097 // 'x_start' points to x array + scaled xlen
7098 // 'y_start' points to y array + scaled ylen
7099
7100 Node* z_start = array_element_address(z, intcon(0), T_INT);
7101
7102 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
7103 OptoRuntime::multiplyToLen_Type(),
7104 stubAddr, stubName, TypePtr::BOTTOM,
7105 x_start, xlen, y_start, ylen, z_start);
7106
7107 C->set_has_split_ifs(true); // Has chance for split-if optimization
7108 set_result(z);
7109 return true;
7110 }
7111
7112 //-------------inline_squareToLen------------------------------------
7113 bool LibraryCallKit::inline_squareToLen() {
7114 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
7115
7116 address stubAddr = StubRoutines::squareToLen();
7117 if (stubAddr == nullptr) {
7118 return false; // Intrinsic's stub is not implemented on this platform
7119 }
7120 const char* stubName = "squareToLen";
7121
7122 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
7123
7124 Node* x = argument(0);
7125 Node* len = argument(1);
7126 Node* z = argument(2);
7127 Node* zlen = argument(3);
7128
7129 x = must_be_not_null(x, true);
7130 z = must_be_not_null(z, true);
7131
7132 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
7133 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
7134 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
7135 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
7136 // failed array check
7137 return false;
7138 }
7139
7140 BasicType x_elem = x_type->elem()->array_element_basic_type();
7141 BasicType z_elem = z_type->elem()->array_element_basic_type();
7142 if (x_elem != T_INT || z_elem != T_INT) {
7143 return false;
7144 }
7145
7146
7147 Node* x_start = array_element_address(x, intcon(0), x_elem);
7148 Node* z_start = array_element_address(z, intcon(0), z_elem);
7149
7150 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
7151 OptoRuntime::squareToLen_Type(),
7152 stubAddr, stubName, TypePtr::BOTTOM,
7153 x_start, len, z_start, zlen);
7154
7155 set_result(z);
7156 return true;
7157 }
7158
7159 //-------------inline_mulAdd------------------------------------------
7160 bool LibraryCallKit::inline_mulAdd() {
7161 assert(UseMulAddIntrinsic, "not implemented on this platform");
7162
7163 address stubAddr = StubRoutines::mulAdd();
7164 if (stubAddr == nullptr) {
7165 return false; // Intrinsic's stub is not implemented on this platform
7166 }
7167 const char* stubName = "mulAdd";
7168
7169 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
7170
7171 Node* out = argument(0);
7172 Node* in = argument(1);
7173 Node* offset = argument(2);
7174 Node* len = argument(3);
7175 Node* k = argument(4);
7176
7177 in = must_be_not_null(in, true);
7178 out = must_be_not_null(out, true);
7179
7180 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
7181 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
7182 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
7183 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
7184 // failed array check
7185 return false;
7186 }
7187
7188 BasicType out_elem = out_type->elem()->array_element_basic_type();
7189 BasicType in_elem = in_type->elem()->array_element_basic_type();
7190 if (out_elem != T_INT || in_elem != T_INT) {
7191 return false;
7192 }
7193
7194 Node* outlen = load_array_length(out);
7195 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
7196 Node* out_start = array_element_address(out, intcon(0), out_elem);
7197 Node* in_start = array_element_address(in, intcon(0), in_elem);
7198
7199 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
7200 OptoRuntime::mulAdd_Type(),
7201 stubAddr, stubName, TypePtr::BOTTOM,
7202 out_start,in_start, new_offset, len, k);
7203 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7204 set_result(result);
7205 return true;
7206 }
7207
7208 //-------------inline_montgomeryMultiply-----------------------------------
7209 bool LibraryCallKit::inline_montgomeryMultiply() {
7210 address stubAddr = StubRoutines::montgomeryMultiply();
7211 if (stubAddr == nullptr) {
7212 return false; // Intrinsic's stub is not implemented on this platform
7213 }
7214
7215 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
7216 const char* stubName = "montgomery_multiply";
7217
7218 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
7219
7220 Node* a = argument(0);
7221 Node* b = argument(1);
7222 Node* n = argument(2);
7223 Node* len = argument(3);
7224 Node* inv = argument(4);
7225 Node* m = argument(6);
7226
7227 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
7228 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
7229 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
7230 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
7231 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
7232 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
7233 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
7234 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
7235 // failed array check
7236 return false;
7237 }
7238
7239 BasicType a_elem = a_type->elem()->array_element_basic_type();
7240 BasicType b_elem = b_type->elem()->array_element_basic_type();
7241 BasicType n_elem = n_type->elem()->array_element_basic_type();
7242 BasicType m_elem = m_type->elem()->array_element_basic_type();
7243 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
7244 return false;
7245 }
7246
7247 // Make the call
7248 {
7249 Node* a_start = array_element_address(a, intcon(0), a_elem);
7250 Node* b_start = array_element_address(b, intcon(0), b_elem);
7251 Node* n_start = array_element_address(n, intcon(0), n_elem);
7252 Node* m_start = array_element_address(m, intcon(0), m_elem);
7253
7254 Node* call = make_runtime_call(RC_LEAF,
7255 OptoRuntime::montgomeryMultiply_Type(),
7256 stubAddr, stubName, TypePtr::BOTTOM,
7257 a_start, b_start, n_start, len, inv, top(),
7258 m_start);
7259 set_result(m);
7260 }
7261
7262 return true;
7263 }
7264
7265 bool LibraryCallKit::inline_montgomerySquare() {
7266 address stubAddr = StubRoutines::montgomerySquare();
7267 if (stubAddr == nullptr) {
7268 return false; // Intrinsic's stub is not implemented on this platform
7269 }
7270
7271 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
7272 const char* stubName = "montgomery_square";
7273
7274 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
7275
7276 Node* a = argument(0);
7277 Node* n = argument(1);
7278 Node* len = argument(2);
7279 Node* inv = argument(3);
7280 Node* m = argument(5);
7281
7282 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
7283 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
7284 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
7285 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
7286 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
7287 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
7288 // failed array check
7289 return false;
7290 }
7291
7292 BasicType a_elem = a_type->elem()->array_element_basic_type();
7293 BasicType n_elem = n_type->elem()->array_element_basic_type();
7294 BasicType m_elem = m_type->elem()->array_element_basic_type();
7295 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
7296 return false;
7297 }
7298
7299 // Make the call
7300 {
7301 Node* a_start = array_element_address(a, intcon(0), a_elem);
7302 Node* n_start = array_element_address(n, intcon(0), n_elem);
7303 Node* m_start = array_element_address(m, intcon(0), m_elem);
7304
7305 Node* call = make_runtime_call(RC_LEAF,
7306 OptoRuntime::montgomerySquare_Type(),
7307 stubAddr, stubName, TypePtr::BOTTOM,
7308 a_start, n_start, len, inv, top(),
7309 m_start);
7310 set_result(m);
7311 }
7312
7313 return true;
7314 }
7315
7316 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
7317 address stubAddr = nullptr;
7318 const char* stubName = nullptr;
7319
7320 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
7321 if (stubAddr == nullptr) {
7322 return false; // Intrinsic's stub is not implemented on this platform
7323 }
7324
7325 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
7326
7327 assert(callee()->signature()->size() == 5, "expected 5 arguments");
7328
7329 Node* newArr = argument(0);
7330 Node* oldArr = argument(1);
7331 Node* newIdx = argument(2);
7332 Node* shiftCount = argument(3);
7333 Node* numIter = argument(4);
7334
7335 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
7336 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
7337 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
7338 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
7339 return false;
7340 }
7341
7342 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
7343 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
7344 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
7345 return false;
7346 }
7347
7348 // Make the call
7349 {
7350 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
7351 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
7352
7353 Node* call = make_runtime_call(RC_LEAF,
7354 OptoRuntime::bigIntegerShift_Type(),
7355 stubAddr,
7356 stubName,
7357 TypePtr::BOTTOM,
7358 newArr_start,
7359 oldArr_start,
7360 newIdx,
7361 shiftCount,
7362 numIter);
7363 }
7364
7365 return true;
7366 }
7367
7368 //-------------inline_vectorizedMismatch------------------------------
7369 bool LibraryCallKit::inline_vectorizedMismatch() {
7370 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
7371
7372 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
7373 Node* obja = argument(0); // Object
7374 Node* aoffset = argument(1); // long
7375 Node* objb = argument(3); // Object
7376 Node* boffset = argument(4); // long
7377 Node* length = argument(6); // int
7378 Node* scale = argument(7); // int
7379
7380 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
7381 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
7382 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
7383 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
7384 scale == top()) {
7385 return false; // failed input validation
7386 }
7387
7388 Node* obja_adr = make_unsafe_address(obja, aoffset);
7389 Node* objb_adr = make_unsafe_address(objb, boffset);
7390
7391 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
7392 //
7393 // inline_limit = ArrayOperationPartialInlineSize / element_size;
7394 // if (length <= inline_limit) {
7395 // inline_path:
7396 // vmask = VectorMaskGen length
7397 // vload1 = LoadVectorMasked obja, vmask
7398 // vload2 = LoadVectorMasked objb, vmask
7399 // result1 = VectorCmpMasked vload1, vload2, vmask
7400 // } else {
7401 // call_stub_path:
7402 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
7403 // }
7404 // exit_block:
7405 // return Phi(result1, result2);
7406 //
7407 enum { inline_path = 1, // input is small enough to process it all at once
7408 stub_path = 2, // input is too large; call into the VM
7409 PATH_LIMIT = 3
7410 };
7411
7412 Node* exit_block = new RegionNode(PATH_LIMIT);
7413 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
7414 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
7415
7416 Node* call_stub_path = control();
7417
7418 BasicType elem_bt = T_ILLEGAL;
7419
7420 const TypeInt* scale_t = _gvn.type(scale)->is_int();
7421 if (scale_t->is_con()) {
7422 switch (scale_t->get_con()) {
7423 case 0: elem_bt = T_BYTE; break;
7424 case 1: elem_bt = T_SHORT; break;
7425 case 2: elem_bt = T_INT; break;
7426 case 3: elem_bt = T_LONG; break;
7427
7428 default: elem_bt = T_ILLEGAL; break; // not supported
7429 }
7430 }
7431
7432 int inline_limit = 0;
7433 bool do_partial_inline = false;
7434
7435 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
7436 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
7437 do_partial_inline = inline_limit >= 16;
7438 }
7439
7440 if (do_partial_inline) {
7441 assert(elem_bt != T_ILLEGAL, "sanity");
7442
7443 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
7444 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
7445 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
7446
7447 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
7448 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
7449 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
7450
7451 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
7452
7453 if (!stopped()) {
7454 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
7455
7456 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
7457 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
7458 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
7459 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
7460
7461 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
7462 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
7463 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
7464 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
7465
7466 exit_block->init_req(inline_path, control());
7467 memory_phi->init_req(inline_path, map()->memory());
7468 result_phi->init_req(inline_path, result);
7469
7470 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
7471 clear_upper_avx();
7472 }
7473 }
7474 }
7475
7476 if (call_stub_path != nullptr) {
7477 set_control(call_stub_path);
7478
7479 Node* call = make_runtime_call(RC_LEAF,
7480 OptoRuntime::vectorizedMismatch_Type(),
7481 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
7482 obja_adr, objb_adr, length, scale);
7483
7484 exit_block->init_req(stub_path, control());
7485 memory_phi->init_req(stub_path, map()->memory());
7486 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
7487 }
7488
7489 exit_block = _gvn.transform(exit_block);
7490 memory_phi = _gvn.transform(memory_phi);
7491 result_phi = _gvn.transform(result_phi);
7492
7493 record_for_igvn(exit_block);
7494 record_for_igvn(memory_phi);
7495 record_for_igvn(result_phi);
7496
7497 set_control(exit_block);
7498 set_all_memory(memory_phi);
7499 set_result(result_phi);
7500
7501 return true;
7502 }
7503
7504 //------------------------------inline_vectorizedHashcode----------------------------
7505 bool LibraryCallKit::inline_vectorizedHashCode() {
7506 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
7507
7508 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
7509 Node* array = argument(0);
7510 Node* offset = argument(1);
7511 Node* length = argument(2);
7512 Node* initialValue = argument(3);
7513 Node* basic_type = argument(4);
7514
7515 if (basic_type == top()) {
7516 return false; // failed input validation
7517 }
7518
7519 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
7520 if (!basic_type_t->is_con()) {
7521 return false; // Only intrinsify if mode argument is constant
7522 }
7523
7524 array = must_be_not_null(array, true);
7525
7526 BasicType bt = (BasicType)basic_type_t->get_con();
7527
7528 // Resolve address of first element
7529 Node* array_start = array_element_address(array, offset, bt);
7530
7531 const TypeAryPtr* in_adr_type = TypeAryPtr::get_array_body_type(bt);
7532 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(in_adr_type), in_adr_type,
7533 array_start, length, initialValue, basic_type)));
7534 clear_upper_avx();
7535
7536 return true;
7537 }
7538
7539 /**
7540 * Calculate CRC32 for byte.
7541 * int java.util.zip.CRC32.update(int crc, int b)
7542 */
7543 bool LibraryCallKit::inline_updateCRC32() {
7544 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7545 assert(callee()->signature()->size() == 2, "update has 2 parameters");
7546 // no receiver since it is static method
7547 Node* crc = argument(0); // type: int
7548 Node* b = argument(1); // type: int
7549
7550 /*
7551 * int c = ~ crc;
7552 * b = timesXtoThe32[(b ^ c) & 0xFF];
7553 * b = b ^ (c >>> 8);
7554 * crc = ~b;
7555 */
7556
7557 Node* M1 = intcon(-1);
7558 crc = _gvn.transform(new XorINode(crc, M1));
7559 Node* result = _gvn.transform(new XorINode(crc, b));
7560 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
7561
7562 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
7563 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
7564 Node* adr = off_heap_plus_addr(base, ConvI2X(offset));
7565 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
7566
7567 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
7568 result = _gvn.transform(new XorINode(crc, result));
7569 result = _gvn.transform(new XorINode(result, M1));
7570 set_result(result);
7571 return true;
7572 }
7573
7574 /**
7575 * Calculate CRC32 for byte[] array.
7576 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
7577 */
7578 bool LibraryCallKit::inline_updateBytesCRC32() {
7579 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7580 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7581 // no receiver since it is static method
7582 Node* crc = argument(0); // type: int
7583 Node* src = argument(1); // type: oop
7584 Node* offset = argument(2); // type: int
7585 Node* length = argument(3); // type: int
7586
7587 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7588 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7589 // failed array check
7590 return false;
7591 }
7592
7593 // Figure out the size and type of the elements we will be copying.
7594 BasicType src_elem = src_type->elem()->array_element_basic_type();
7595 if (src_elem != T_BYTE) {
7596 return false;
7597 }
7598
7599 // 'src_start' points to src array + scaled offset
7600 src = must_be_not_null(src, true);
7601 Node* src_start = array_element_address(src, offset, src_elem);
7602
7603 // We assume that range check is done by caller.
7604 // TODO: generate range check (offset+length < src.length) in debug VM.
7605
7606 // Call the stub.
7607 address stubAddr = StubRoutines::updateBytesCRC32();
7608 const char *stubName = "updateBytesCRC32";
7609
7610 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7611 stubAddr, stubName, TypePtr::BOTTOM,
7612 crc, src_start, length);
7613 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7614 set_result(result);
7615 return true;
7616 }
7617
7618 /**
7619 * Calculate CRC32 for ByteBuffer.
7620 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
7621 */
7622 bool LibraryCallKit::inline_updateByteBufferCRC32() {
7623 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7624 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7625 // no receiver since it is static method
7626 Node* crc = argument(0); // type: int
7627 Node* src = argument(1); // type: long
7628 Node* offset = argument(3); // type: int
7629 Node* length = argument(4); // type: int
7630
7631 src = ConvL2X(src); // adjust Java long to machine word
7632 Node* base = _gvn.transform(new CastX2PNode(src));
7633 offset = ConvI2X(offset);
7634
7635 // 'src_start' points to src array + scaled offset
7636 Node* src_start = off_heap_plus_addr(base, offset);
7637
7638 // Call the stub.
7639 address stubAddr = StubRoutines::updateBytesCRC32();
7640 const char *stubName = "updateBytesCRC32";
7641
7642 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7643 stubAddr, stubName, TypePtr::BOTTOM,
7644 crc, src_start, length);
7645 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7646 set_result(result);
7647 return true;
7648 }
7649
7650 //------------------------------get_table_from_crc32c_class-----------------------
7651 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
7652 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
7653 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
7654
7655 return table;
7656 }
7657
7658 //------------------------------inline_updateBytesCRC32C-----------------------
7659 //
7660 // Calculate CRC32C for byte[] array.
7661 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
7662 //
7663 bool LibraryCallKit::inline_updateBytesCRC32C() {
7664 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7665 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7666 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7667 // no receiver since it is a static method
7668 Node* crc = argument(0); // type: int
7669 Node* src = argument(1); // type: oop
7670 Node* offset = argument(2); // type: int
7671 Node* end = argument(3); // type: int
7672
7673 Node* length = _gvn.transform(new SubINode(end, offset));
7674
7675 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7676 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7677 // failed array check
7678 return false;
7679 }
7680
7681 // Figure out the size and type of the elements we will be copying.
7682 BasicType src_elem = src_type->elem()->array_element_basic_type();
7683 if (src_elem != T_BYTE) {
7684 return false;
7685 }
7686
7687 // 'src_start' points to src array + scaled offset
7688 src = must_be_not_null(src, true);
7689 Node* src_start = array_element_address(src, offset, src_elem);
7690
7691 // static final int[] byteTable in class CRC32C
7692 Node* table = get_table_from_crc32c_class(callee()->holder());
7693 table = must_be_not_null(table, true);
7694 Node* table_start = array_element_address(table, intcon(0), T_INT);
7695
7696 // We assume that range check is done by caller.
7697 // TODO: generate range check (offset+length < src.length) in debug VM.
7698
7699 // Call the stub.
7700 address stubAddr = StubRoutines::updateBytesCRC32C();
7701 const char *stubName = "updateBytesCRC32C";
7702
7703 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7704 stubAddr, stubName, TypePtr::BOTTOM,
7705 crc, src_start, length, table_start);
7706 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7707 set_result(result);
7708 return true;
7709 }
7710
7711 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
7712 //
7713 // Calculate CRC32C for DirectByteBuffer.
7714 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
7715 //
7716 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
7717 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7718 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
7719 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7720 // no receiver since it is a static method
7721 Node* crc = argument(0); // type: int
7722 Node* src = argument(1); // type: long
7723 Node* offset = argument(3); // type: int
7724 Node* end = argument(4); // type: int
7725
7726 Node* length = _gvn.transform(new SubINode(end, offset));
7727
7728 src = ConvL2X(src); // adjust Java long to machine word
7729 Node* base = _gvn.transform(new CastX2PNode(src));
7730 offset = ConvI2X(offset);
7731
7732 // 'src_start' points to src array + scaled offset
7733 Node* src_start = off_heap_plus_addr(base, offset);
7734
7735 // static final int[] byteTable in class CRC32C
7736 Node* table = get_table_from_crc32c_class(callee()->holder());
7737 table = must_be_not_null(table, true);
7738 Node* table_start = array_element_address(table, intcon(0), T_INT);
7739
7740 // Call the stub.
7741 address stubAddr = StubRoutines::updateBytesCRC32C();
7742 const char *stubName = "updateBytesCRC32C";
7743
7744 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7745 stubAddr, stubName, TypePtr::BOTTOM,
7746 crc, src_start, length, table_start);
7747 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7748 set_result(result);
7749 return true;
7750 }
7751
7752 //------------------------------inline_updateBytesAdler32----------------------
7753 //
7754 // Calculate Adler32 checksum for byte[] array.
7755 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
7756 //
7757 bool LibraryCallKit::inline_updateBytesAdler32() {
7758 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7759 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7760 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7761 // no receiver since it is static method
7762 Node* crc = argument(0); // type: int
7763 Node* src = argument(1); // type: oop
7764 Node* offset = argument(2); // type: int
7765 Node* length = argument(3); // type: int
7766
7767 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7768 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7769 // failed array check
7770 return false;
7771 }
7772
7773 // Figure out the size and type of the elements we will be copying.
7774 BasicType src_elem = src_type->elem()->array_element_basic_type();
7775 if (src_elem != T_BYTE) {
7776 return false;
7777 }
7778
7779 // 'src_start' points to src array + scaled offset
7780 Node* src_start = array_element_address(src, offset, src_elem);
7781
7782 // We assume that range check is done by caller.
7783 // TODO: generate range check (offset+length < src.length) in debug VM.
7784
7785 // Call the stub.
7786 address stubAddr = StubRoutines::updateBytesAdler32();
7787 const char *stubName = "updateBytesAdler32";
7788
7789 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7790 stubAddr, stubName, TypePtr::BOTTOM,
7791 crc, src_start, length);
7792 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7793 set_result(result);
7794 return true;
7795 }
7796
7797 //------------------------------inline_updateByteBufferAdler32---------------
7798 //
7799 // Calculate Adler32 checksum for DirectByteBuffer.
7800 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
7801 //
7802 bool LibraryCallKit::inline_updateByteBufferAdler32() {
7803 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7804 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7805 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7806 // no receiver since it is static method
7807 Node* crc = argument(0); // type: int
7808 Node* src = argument(1); // type: long
7809 Node* offset = argument(3); // type: int
7810 Node* length = argument(4); // type: int
7811
7812 src = ConvL2X(src); // adjust Java long to machine word
7813 Node* base = _gvn.transform(new CastX2PNode(src));
7814 offset = ConvI2X(offset);
7815
7816 // 'src_start' points to src array + scaled offset
7817 Node* src_start = off_heap_plus_addr(base, offset);
7818
7819 // Call the stub.
7820 address stubAddr = StubRoutines::updateBytesAdler32();
7821 const char *stubName = "updateBytesAdler32";
7822
7823 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7824 stubAddr, stubName, TypePtr::BOTTOM,
7825 crc, src_start, length);
7826
7827 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7828 set_result(result);
7829 return true;
7830 }
7831
7832 //----------------------------inline_reference_get0----------------------------
7833 // public T java.lang.ref.Reference.get();
7834 bool LibraryCallKit::inline_reference_get0() {
7835 const int referent_offset = java_lang_ref_Reference::referent_offset();
7836
7837 // Get the argument:
7838 Node* reference_obj = null_check_receiver();
7839 if (stopped()) return true;
7840
7841 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7842 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7843 decorators, /*is_static*/ false,
7844 env()->Reference_klass());
7845 if (result == nullptr) return false;
7846
7847 // Add memory barrier to prevent commoning reads from this field
7848 // across safepoint since GC can change its value.
7849 insert_mem_bar(Op_MemBarCPUOrder);
7850
7851 set_result(result);
7852 return true;
7853 }
7854
7855 //----------------------------inline_reference_refersTo0----------------------------
7856 // bool java.lang.ref.Reference.refersTo0();
7857 // bool java.lang.ref.PhantomReference.refersTo0();
7858 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7859 // Get arguments:
7860 Node* reference_obj = null_check_receiver();
7861 Node* other_obj = argument(1);
7862 if (stopped()) return true;
7863
7864 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7865 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7866 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7867 decorators, /*is_static*/ false,
7868 env()->Reference_klass());
7869 if (referent == nullptr) return false;
7870
7871 // Add memory barrier to prevent commoning reads from this field
7872 // across safepoint since GC can change its value.
7873 insert_mem_bar(Op_MemBarCPUOrder);
7874
7875 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7876 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7877 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7878
7879 RegionNode* region = new RegionNode(3);
7880 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7881
7882 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7883 region->init_req(1, if_true);
7884 phi->init_req(1, intcon(1));
7885
7886 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7887 region->init_req(2, if_false);
7888 phi->init_req(2, intcon(0));
7889
7890 set_control(_gvn.transform(region));
7891 record_for_igvn(region);
7892 set_result(_gvn.transform(phi));
7893 return true;
7894 }
7895
7896 //----------------------------inline_reference_clear0----------------------------
7897 // void java.lang.ref.Reference.clear0();
7898 // void java.lang.ref.PhantomReference.clear0();
7899 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7900 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7901
7902 // Get arguments
7903 Node* reference_obj = null_check_receiver();
7904 if (stopped()) return true;
7905
7906 // Common access parameters
7907 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7908 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7909 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7910 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7911 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7912
7913 Node* referent = access_load_at(reference_obj,
7914 referent_field_addr,
7915 referent_field_addr_type,
7916 val_type,
7917 T_OBJECT,
7918 decorators);
7919
7920 IdealKit ideal(this);
7921 #define __ ideal.
7922 __ if_then(referent, BoolTest::ne, null());
7923 sync_kit(ideal);
7924 access_store_at(reference_obj,
7925 referent_field_addr,
7926 referent_field_addr_type,
7927 null(),
7928 val_type,
7929 T_OBJECT,
7930 decorators);
7931 __ sync_kit(this);
7932 __ end_if();
7933 final_sync(ideal);
7934 #undef __
7935
7936 return true;
7937 }
7938
7939 //-----------------------inline_reference_reachabilityFence-----------------
7940 // bool java.lang.ref.Reference.reachabilityFence();
7941 bool LibraryCallKit::inline_reference_reachabilityFence() {
7942 Node* referent = argument(0);
7943 insert_reachability_fence(referent);
7944 return true;
7945 }
7946
7947 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7948 DecoratorSet decorators, bool is_static,
7949 ciInstanceKlass* fromKls) {
7950 if (fromKls == nullptr) {
7951 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7952 assert(tinst != nullptr, "obj is null");
7953 assert(tinst->is_loaded(), "obj is not loaded");
7954 fromKls = tinst->instance_klass();
7955 }
7956 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7957 ciSymbol::make(fieldTypeString),
7958 is_static);
7959
7960 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7961 if (field == nullptr) return (Node *) nullptr;
7962
7963 if (is_static) {
7964 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7965 fromObj = makecon(tip);
7966 }
7967
7968 // Next code copied from Parse::do_get_xxx():
7969
7970 // Compute address and memory type.
7971 int offset = field->offset_in_bytes();
7972 bool is_vol = field->is_volatile();
7973 ciType* field_klass = field->type();
7974 assert(field_klass->is_loaded(), "should be loaded");
7975 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7976 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7977 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7978 "slice of address and input slice don't match");
7979 BasicType bt = field->layout_type();
7980
7981 // Build the resultant type of the load
7982 const Type *type;
7983 if (bt == T_OBJECT) {
7984 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7985 } else {
7986 type = Type::get_const_basic_type(bt);
7987 }
7988
7989 if (is_vol) {
7990 decorators |= MO_SEQ_CST;
7991 }
7992
7993 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7994 }
7995
7996 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7997 bool is_exact /* true */, bool is_static /* false */,
7998 ciInstanceKlass * fromKls /* nullptr */) {
7999 if (fromKls == nullptr) {
8000 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
8001 assert(tinst != nullptr, "obj is null");
8002 assert(tinst->is_loaded(), "obj is not loaded");
8003 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
8004 fromKls = tinst->instance_klass();
8005 }
8006 else {
8007 assert(is_static, "only for static field access");
8008 }
8009 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
8010 ciSymbol::make(fieldTypeString),
8011 is_static);
8012
8013 assert(field != nullptr, "undefined field");
8014 assert(!field->is_volatile(), "not defined for volatile fields");
8015
8016 if (is_static) {
8017 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
8018 fromObj = makecon(tip);
8019 }
8020
8021 // Next code copied from Parse::do_get_xxx():
8022
8023 // Compute address and memory type.
8024 int offset = field->offset_in_bytes();
8025 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
8026
8027 return adr;
8028 }
8029
8030 //------------------------------inline_aescrypt_Block-----------------------
8031 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
8032 address stubAddr = nullptr;
8033 const char *stubName;
8034 bool is_decrypt = false;
8035 assert(UseAES, "need AES instruction support");
8036
8037 switch(id) {
8038 case vmIntrinsics::_aescrypt_encryptBlock:
8039 stubAddr = StubRoutines::aescrypt_encryptBlock();
8040 stubName = "aescrypt_encryptBlock";
8041 break;
8042 case vmIntrinsics::_aescrypt_decryptBlock:
8043 stubAddr = StubRoutines::aescrypt_decryptBlock();
8044 stubName = "aescrypt_decryptBlock";
8045 is_decrypt = true;
8046 break;
8047 default:
8048 break;
8049 }
8050 if (stubAddr == nullptr) return false;
8051
8052 Node* aescrypt_object = argument(0);
8053 Node* src = argument(1);
8054 Node* src_offset = argument(2);
8055 Node* dest = argument(3);
8056 Node* dest_offset = argument(4);
8057
8058 src = must_be_not_null(src, true);
8059 dest = must_be_not_null(dest, true);
8060
8061 // (1) src and dest are arrays.
8062 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8063 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8064 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8065 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8066
8067 // for the quick and dirty code we will skip all the checks.
8068 // we are just trying to get the call to be generated.
8069 Node* src_start = src;
8070 Node* dest_start = dest;
8071 if (src_offset != nullptr || dest_offset != nullptr) {
8072 assert(src_offset != nullptr && dest_offset != nullptr, "");
8073 src_start = array_element_address(src, src_offset, T_BYTE);
8074 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8075 }
8076
8077 // now need to get the start of its expanded key array
8078 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8079 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
8080 if (k_start == nullptr) return false;
8081
8082 // Call the stub.
8083 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
8084 stubAddr, stubName, TypePtr::BOTTOM,
8085 src_start, dest_start, k_start);
8086
8087 return true;
8088 }
8089
8090 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
8091 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
8092 address stubAddr = nullptr;
8093 const char *stubName = nullptr;
8094 bool is_decrypt = false;
8095 assert(UseAES, "need AES instruction support");
8096
8097 switch(id) {
8098 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
8099 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
8100 stubName = "cipherBlockChaining_encryptAESCrypt";
8101 break;
8102 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
8103 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
8104 stubName = "cipherBlockChaining_decryptAESCrypt";
8105 is_decrypt = true;
8106 break;
8107 default:
8108 break;
8109 }
8110 if (stubAddr == nullptr) return false;
8111
8112 Node* cipherBlockChaining_object = argument(0);
8113 Node* src = argument(1);
8114 Node* src_offset = argument(2);
8115 Node* len = argument(3);
8116 Node* dest = argument(4);
8117 Node* dest_offset = argument(5);
8118
8119 src = must_be_not_null(src, false);
8120 dest = must_be_not_null(dest, false);
8121
8122 // (1) src and dest are arrays.
8123 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8124 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8125 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8126 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8127
8128 // checks are the responsibility of the caller
8129 Node* src_start = src;
8130 Node* dest_start = dest;
8131 if (src_offset != nullptr || dest_offset != nullptr) {
8132 assert(src_offset != nullptr && dest_offset != nullptr, "");
8133 src_start = array_element_address(src, src_offset, T_BYTE);
8134 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8135 }
8136
8137 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8138 // (because of the predicated logic executed earlier).
8139 // so we cast it here safely.
8140 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8141
8142 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8143 if (embeddedCipherObj == nullptr) return false;
8144
8145 // cast it to what we know it will be at runtime
8146 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
8147 assert(tinst != nullptr, "CBC obj is null");
8148 assert(tinst->is_loaded(), "CBC obj is not loaded");
8149 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8150 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8151
8152 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8153 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8154 const TypeOopPtr* xtype = aklass->as_exact_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8155 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8156 aescrypt_object = _gvn.transform(aescrypt_object);
8157
8158 // we need to get the start of the aescrypt_object's expanded key array
8159 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
8160 if (k_start == nullptr) return false;
8161
8162 // similarly, get the start address of the r vector
8163 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
8164 if (objRvec == nullptr) return false;
8165 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
8166
8167 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8168 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8169 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
8170 stubAddr, stubName, TypePtr::BOTTOM,
8171 src_start, dest_start, k_start, r_start, len);
8172
8173 // return cipher length (int)
8174 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
8175 set_result(retvalue);
8176 return true;
8177 }
8178
8179 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
8180 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
8181 address stubAddr = nullptr;
8182 const char *stubName = nullptr;
8183 bool is_decrypt = false;
8184 assert(UseAES, "need AES instruction support");
8185
8186 switch (id) {
8187 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
8188 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
8189 stubName = "electronicCodeBook_encryptAESCrypt";
8190 break;
8191 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
8192 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
8193 stubName = "electronicCodeBook_decryptAESCrypt";
8194 is_decrypt = true;
8195 break;
8196 default:
8197 break;
8198 }
8199
8200 if (stubAddr == nullptr) return false;
8201
8202 Node* electronicCodeBook_object = argument(0);
8203 Node* src = argument(1);
8204 Node* src_offset = argument(2);
8205 Node* len = argument(3);
8206 Node* dest = argument(4);
8207 Node* dest_offset = argument(5);
8208
8209 // (1) src and dest are arrays.
8210 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8211 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8212 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8213 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8214
8215 // checks are the responsibility of the caller
8216 Node* src_start = src;
8217 Node* dest_start = dest;
8218 if (src_offset != nullptr || dest_offset != nullptr) {
8219 assert(src_offset != nullptr && dest_offset != nullptr, "");
8220 src_start = array_element_address(src, src_offset, T_BYTE);
8221 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8222 }
8223
8224 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8225 // (because of the predicated logic executed earlier).
8226 // so we cast it here safely.
8227 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8228
8229 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8230 if (embeddedCipherObj == nullptr) return false;
8231
8232 // cast it to what we know it will be at runtime
8233 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
8234 assert(tinst != nullptr, "ECB obj is null");
8235 assert(tinst->is_loaded(), "ECB obj is not loaded");
8236 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8237 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8238
8239 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8240 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8241 const TypeOopPtr* xtype = aklass->as_exact_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8242 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8243 aescrypt_object = _gvn.transform(aescrypt_object);
8244
8245 // we need to get the start of the aescrypt_object's expanded key array
8246 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
8247 if (k_start == nullptr) return false;
8248
8249 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8250 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
8251 OptoRuntime::electronicCodeBook_aescrypt_Type(),
8252 stubAddr, stubName, TypePtr::BOTTOM,
8253 src_start, dest_start, k_start, len);
8254
8255 // return cipher length (int)
8256 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
8257 set_result(retvalue);
8258 return true;
8259 }
8260
8261 //------------------------------inline_counterMode_AESCrypt-----------------------
8262 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
8263 assert(UseAES, "need AES instruction support");
8264 if (!UseAESCTRIntrinsics) return false;
8265
8266 address stubAddr = nullptr;
8267 const char *stubName = nullptr;
8268 if (id == vmIntrinsics::_counterMode_AESCrypt) {
8269 stubAddr = StubRoutines::counterMode_AESCrypt();
8270 stubName = "counterMode_AESCrypt";
8271 }
8272 if (stubAddr == nullptr) return false;
8273
8274 Node* counterMode_object = argument(0);
8275 Node* src = argument(1);
8276 Node* src_offset = argument(2);
8277 Node* len = argument(3);
8278 Node* dest = argument(4);
8279 Node* dest_offset = argument(5);
8280
8281 // (1) src and dest are arrays.
8282 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8283 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8284 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8285 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8286
8287 // checks are the responsibility of the caller
8288 Node* src_start = src;
8289 Node* dest_start = dest;
8290 if (src_offset != nullptr || dest_offset != nullptr) {
8291 assert(src_offset != nullptr && dest_offset != nullptr, "");
8292 src_start = array_element_address(src, src_offset, T_BYTE);
8293 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8294 }
8295
8296 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8297 // (because of the predicated logic executed earlier).
8298 // so we cast it here safely.
8299 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8300 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8301 if (embeddedCipherObj == nullptr) return false;
8302 // cast it to what we know it will be at runtime
8303 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
8304 assert(tinst != nullptr, "CTR obj is null");
8305 assert(tinst->is_loaded(), "CTR obj is not loaded");
8306 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8307 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8308 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8309 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8310 const TypeOopPtr* xtype = aklass->as_exact_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8311 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8312 aescrypt_object = _gvn.transform(aescrypt_object);
8313 // we need to get the start of the aescrypt_object's expanded key array
8314 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
8315 if (k_start == nullptr) return false;
8316 // similarly, get the start address of the r vector
8317 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
8318 if (obj_counter == nullptr) return false;
8319 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
8320
8321 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
8322 if (saved_encCounter == nullptr) return false;
8323 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
8324 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
8325
8326 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8327 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8328 OptoRuntime::counterMode_aescrypt_Type(),
8329 stubAddr, stubName, TypePtr::BOTTOM,
8330 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
8331
8332 // return cipher length (int)
8333 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
8334 set_result(retvalue);
8335 return true;
8336 }
8337
8338 //------------------------------get_key_start_from_aescrypt_object-----------------------
8339 Node* LibraryCallKit::get_key_start_from_aescrypt_object(Node* aescrypt_object, bool is_decrypt) {
8340 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
8341 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
8342 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
8343 // The following platform specific stubs of encryption and decryption use the same round keys.
8344 #if defined(PPC64) || defined(S390) || defined(RISCV64)
8345 bool use_decryption_key = false;
8346 #else
8347 bool use_decryption_key = is_decrypt;
8348 #endif
8349 Node* objAESCryptKey = load_field_from_object(aescrypt_object, use_decryption_key ? "sessionKd" : "sessionKe", "[I");
8350 assert(objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AES_Crypt");
8351 if (objAESCryptKey == nullptr) return (Node *) nullptr;
8352
8353 // now have the array, need to get the start address of the selected key array
8354 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
8355 return k_start;
8356 }
8357
8358 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
8359 // Return node representing slow path of predicate check.
8360 // the pseudo code we want to emulate with this predicate is:
8361 // for encryption:
8362 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8363 // for decryption:
8364 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8365 // note cipher==plain is more conservative than the original java code but that's OK
8366 //
8367 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
8368 // The receiver was checked for null already.
8369 Node* objCBC = argument(0);
8370
8371 Node* src = argument(1);
8372 Node* dest = argument(4);
8373
8374 // Load embeddedCipher field of CipherBlockChaining object.
8375 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8376
8377 // get AESCrypt klass for instanceOf check
8378 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8379 // will have same classloader as CipherBlockChaining object
8380 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
8381 assert(tinst != nullptr, "CBCobj is null");
8382 assert(tinst->is_loaded(), "CBCobj is not loaded");
8383
8384 // we want to do an instanceof comparison against the AESCrypt class
8385 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8386 if (!klass_AESCrypt->is_loaded()) {
8387 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8388 Node* ctrl = control();
8389 set_control(top()); // no regular fast path
8390 return ctrl;
8391 }
8392
8393 src = must_be_not_null(src, true);
8394 dest = must_be_not_null(dest, true);
8395
8396 // Resolve oops to stable for CmpP below.
8397 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8398
8399 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8400 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8401 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8402
8403 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8404
8405 // for encryption, we are done
8406 if (!decrypting)
8407 return instof_false; // even if it is null
8408
8409 // for decryption, we need to add a further check to avoid
8410 // taking the intrinsic path when cipher and plain are the same
8411 // see the original java code for why.
8412 RegionNode* region = new RegionNode(3);
8413 region->init_req(1, instof_false);
8414
8415 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8416 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8417 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8418 region->init_req(2, src_dest_conjoint);
8419
8420 record_for_igvn(region);
8421 return _gvn.transform(region);
8422 }
8423
8424 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
8425 // Return node representing slow path of predicate check.
8426 // the pseudo code we want to emulate with this predicate is:
8427 // for encryption:
8428 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8429 // for decryption:
8430 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8431 // note cipher==plain is more conservative than the original java code but that's OK
8432 //
8433 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
8434 // The receiver was checked for null already.
8435 Node* objECB = argument(0);
8436
8437 // Load embeddedCipher field of ElectronicCodeBook object.
8438 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8439
8440 // get AESCrypt klass for instanceOf check
8441 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8442 // will have same classloader as ElectronicCodeBook object
8443 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
8444 assert(tinst != nullptr, "ECBobj is null");
8445 assert(tinst->is_loaded(), "ECBobj is not loaded");
8446
8447 // we want to do an instanceof comparison against the AESCrypt class
8448 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8449 if (!klass_AESCrypt->is_loaded()) {
8450 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8451 Node* ctrl = control();
8452 set_control(top()); // no regular fast path
8453 return ctrl;
8454 }
8455 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8456
8457 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8458 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8459 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8460
8461 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8462
8463 // for encryption, we are done
8464 if (!decrypting)
8465 return instof_false; // even if it is null
8466
8467 // for decryption, we need to add a further check to avoid
8468 // taking the intrinsic path when cipher and plain are the same
8469 // see the original java code for why.
8470 RegionNode* region = new RegionNode(3);
8471 region->init_req(1, instof_false);
8472 Node* src = argument(1);
8473 Node* dest = argument(4);
8474 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8475 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8476 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8477 region->init_req(2, src_dest_conjoint);
8478
8479 record_for_igvn(region);
8480 return _gvn.transform(region);
8481 }
8482
8483 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
8484 // Return node representing slow path of predicate check.
8485 // the pseudo code we want to emulate with this predicate is:
8486 // for encryption:
8487 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8488 // for decryption:
8489 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8490 // note cipher==plain is more conservative than the original java code but that's OK
8491 //
8492
8493 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
8494 // The receiver was checked for null already.
8495 Node* objCTR = argument(0);
8496
8497 // Load embeddedCipher field of CipherBlockChaining object.
8498 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8499
8500 // get AESCrypt klass for instanceOf check
8501 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8502 // will have same classloader as CipherBlockChaining object
8503 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
8504 assert(tinst != nullptr, "CTRobj is null");
8505 assert(tinst->is_loaded(), "CTRobj is not loaded");
8506
8507 // we want to do an instanceof comparison against the AESCrypt class
8508 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8509 if (!klass_AESCrypt->is_loaded()) {
8510 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8511 Node* ctrl = control();
8512 set_control(top()); // no regular fast path
8513 return ctrl;
8514 }
8515
8516 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8517 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8518 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8519 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8520 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8521
8522 return instof_false; // even if it is null
8523 }
8524
8525 //------------------------------inline_ghash_processBlocks
8526 bool LibraryCallKit::inline_ghash_processBlocks() {
8527 address stubAddr;
8528 const char *stubName;
8529 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
8530
8531 stubAddr = StubRoutines::ghash_processBlocks();
8532 stubName = "ghash_processBlocks";
8533
8534 Node* data = argument(0);
8535 Node* offset = argument(1);
8536 Node* len = argument(2);
8537 Node* state = argument(3);
8538 Node* subkeyH = argument(4);
8539
8540 state = must_be_not_null(state, true);
8541 subkeyH = must_be_not_null(subkeyH, true);
8542 data = must_be_not_null(data, true);
8543
8544 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8545 assert(state_start, "state is null");
8546 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
8547 assert(subkeyH_start, "subkeyH is null");
8548 Node* data_start = array_element_address(data, offset, T_BYTE);
8549 assert(data_start, "data is null");
8550
8551 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
8552 OptoRuntime::ghash_processBlocks_Type(),
8553 stubAddr, stubName, TypePtr::BOTTOM,
8554 state_start, subkeyH_start, data_start, len);
8555 return true;
8556 }
8557
8558 //------------------------------inline_chacha20Block
8559 bool LibraryCallKit::inline_chacha20Block() {
8560 address stubAddr;
8561 const char *stubName;
8562 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
8563
8564 stubAddr = StubRoutines::chacha20Block();
8565 stubName = "chacha20Block";
8566
8567 Node* state = argument(0);
8568 Node* result = argument(1);
8569
8570 state = must_be_not_null(state, true);
8571 result = must_be_not_null(result, true);
8572
8573 Node* state_start = array_element_address(state, intcon(0), T_INT);
8574 assert(state_start, "state is null");
8575 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
8576 assert(result_start, "result is null");
8577
8578 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
8579 OptoRuntime::chacha20Block_Type(),
8580 stubAddr, stubName, TypePtr::BOTTOM,
8581 state_start, result_start);
8582 // return key stream length (int)
8583 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
8584 set_result(retvalue);
8585 return true;
8586 }
8587
8588 //------------------------------inline_kyberNtt
8589 bool LibraryCallKit::inline_kyberNtt() {
8590 address stubAddr;
8591 const char *stubName;
8592 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8593 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
8594
8595 stubAddr = StubRoutines::kyberNtt();
8596 stubName = "kyberNtt";
8597 if (!stubAddr) return false;
8598
8599 Node* coeffs = argument(0);
8600 Node* ntt_zetas = argument(1);
8601
8602 coeffs = must_be_not_null(coeffs, true);
8603 ntt_zetas = must_be_not_null(ntt_zetas, true);
8604
8605 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8606 assert(coeffs_start, "coeffs is null");
8607 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
8608 assert(ntt_zetas_start, "ntt_zetas is null");
8609 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8610 OptoRuntime::kyberNtt_Type(),
8611 stubAddr, stubName, TypePtr::BOTTOM,
8612 coeffs_start, ntt_zetas_start);
8613 // return an int
8614 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
8615 set_result(retvalue);
8616 return true;
8617 }
8618
8619 //------------------------------inline_kyberInverseNtt
8620 bool LibraryCallKit::inline_kyberInverseNtt() {
8621 address stubAddr;
8622 const char *stubName;
8623 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8624 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
8625
8626 stubAddr = StubRoutines::kyberInverseNtt();
8627 stubName = "kyberInverseNtt";
8628 if (!stubAddr) return false;
8629
8630 Node* coeffs = argument(0);
8631 Node* zetas = argument(1);
8632
8633 coeffs = must_be_not_null(coeffs, true);
8634 zetas = must_be_not_null(zetas, true);
8635
8636 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8637 assert(coeffs_start, "coeffs is null");
8638 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8639 assert(zetas_start, "inverseNtt_zetas is null");
8640 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8641 OptoRuntime::kyberInverseNtt_Type(),
8642 stubAddr, stubName, TypePtr::BOTTOM,
8643 coeffs_start, zetas_start);
8644
8645 // return an int
8646 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
8647 set_result(retvalue);
8648 return true;
8649 }
8650
8651 //------------------------------inline_kyberNttMult
8652 bool LibraryCallKit::inline_kyberNttMult() {
8653 address stubAddr;
8654 const char *stubName;
8655 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8656 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
8657
8658 stubAddr = StubRoutines::kyberNttMult();
8659 stubName = "kyberNttMult";
8660 if (!stubAddr) return false;
8661
8662 Node* result = argument(0);
8663 Node* ntta = argument(1);
8664 Node* nttb = argument(2);
8665 Node* zetas = argument(3);
8666
8667 result = must_be_not_null(result, true);
8668 ntta = must_be_not_null(ntta, true);
8669 nttb = must_be_not_null(nttb, true);
8670 zetas = must_be_not_null(zetas, true);
8671
8672 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8673 assert(result_start, "result is null");
8674 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
8675 assert(ntta_start, "ntta is null");
8676 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
8677 assert(nttb_start, "nttb is null");
8678 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8679 assert(zetas_start, "nttMult_zetas is null");
8680 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8681 OptoRuntime::kyberNttMult_Type(),
8682 stubAddr, stubName, TypePtr::BOTTOM,
8683 result_start, ntta_start, nttb_start,
8684 zetas_start);
8685
8686 // return an int
8687 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
8688 set_result(retvalue);
8689
8690 return true;
8691 }
8692
8693 //------------------------------inline_kyberAddPoly_2
8694 bool LibraryCallKit::inline_kyberAddPoly_2() {
8695 address stubAddr;
8696 const char *stubName;
8697 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8698 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
8699
8700 stubAddr = StubRoutines::kyberAddPoly_2();
8701 stubName = "kyberAddPoly_2";
8702 if (!stubAddr) return false;
8703
8704 Node* result = argument(0);
8705 Node* a = argument(1);
8706 Node* b = argument(2);
8707
8708 result = must_be_not_null(result, true);
8709 a = must_be_not_null(a, true);
8710 b = must_be_not_null(b, true);
8711
8712 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8713 assert(result_start, "result is null");
8714 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8715 assert(a_start, "a is null");
8716 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8717 assert(b_start, "b is null");
8718 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
8719 OptoRuntime::kyberAddPoly_2_Type(),
8720 stubAddr, stubName, TypePtr::BOTTOM,
8721 result_start, a_start, b_start);
8722 // return an int
8723 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
8724 set_result(retvalue);
8725 return true;
8726 }
8727
8728 //------------------------------inline_kyberAddPoly_3
8729 bool LibraryCallKit::inline_kyberAddPoly_3() {
8730 address stubAddr;
8731 const char *stubName;
8732 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8733 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
8734
8735 stubAddr = StubRoutines::kyberAddPoly_3();
8736 stubName = "kyberAddPoly_3";
8737 if (!stubAddr) return false;
8738
8739 Node* result = argument(0);
8740 Node* a = argument(1);
8741 Node* b = argument(2);
8742 Node* c = argument(3);
8743
8744 result = must_be_not_null(result, true);
8745 a = must_be_not_null(a, true);
8746 b = must_be_not_null(b, true);
8747 c = must_be_not_null(c, true);
8748
8749 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8750 assert(result_start, "result is null");
8751 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8752 assert(a_start, "a is null");
8753 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8754 assert(b_start, "b is null");
8755 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
8756 assert(c_start, "c is null");
8757 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
8758 OptoRuntime::kyberAddPoly_3_Type(),
8759 stubAddr, stubName, TypePtr::BOTTOM,
8760 result_start, a_start, b_start, c_start);
8761 // return an int
8762 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
8763 set_result(retvalue);
8764 return true;
8765 }
8766
8767 //------------------------------inline_kyber12To16
8768 bool LibraryCallKit::inline_kyber12To16() {
8769 address stubAddr;
8770 const char *stubName;
8771 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8772 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
8773
8774 stubAddr = StubRoutines::kyber12To16();
8775 stubName = "kyber12To16";
8776 if (!stubAddr) return false;
8777
8778 Node* condensed = argument(0);
8779 Node* condensedOffs = argument(1);
8780 Node* parsed = argument(2);
8781 Node* parsedLength = argument(3);
8782
8783 condensed = must_be_not_null(condensed, true);
8784 parsed = must_be_not_null(parsed, true);
8785
8786 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
8787 assert(condensed_start, "condensed is null");
8788 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
8789 assert(parsed_start, "parsed is null");
8790 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
8791 OptoRuntime::kyber12To16_Type(),
8792 stubAddr, stubName, TypePtr::BOTTOM,
8793 condensed_start, condensedOffs, parsed_start, parsedLength);
8794 // return an int
8795 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
8796 set_result(retvalue);
8797 return true;
8798
8799 }
8800
8801 //------------------------------inline_kyberBarrettReduce
8802 bool LibraryCallKit::inline_kyberBarrettReduce() {
8803 address stubAddr;
8804 const char *stubName;
8805 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8806 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
8807
8808 stubAddr = StubRoutines::kyberBarrettReduce();
8809 stubName = "kyberBarrettReduce";
8810 if (!stubAddr) return false;
8811
8812 Node* coeffs = argument(0);
8813
8814 coeffs = must_be_not_null(coeffs, true);
8815
8816 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8817 assert(coeffs_start, "coeffs is null");
8818 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
8819 OptoRuntime::kyberBarrettReduce_Type(),
8820 stubAddr, stubName, TypePtr::BOTTOM,
8821 coeffs_start);
8822 // return an int
8823 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
8824 set_result(retvalue);
8825 return true;
8826 }
8827
8828 //------------------------------inline_dilithiumAlmostNtt
8829 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
8830 address stubAddr;
8831 const char *stubName;
8832 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8833 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
8834
8835 stubAddr = StubRoutines::dilithiumAlmostNtt();
8836 stubName = "dilithiumAlmostNtt";
8837 if (!stubAddr) return false;
8838
8839 Node* coeffs = argument(0);
8840 Node* ntt_zetas = argument(1);
8841
8842 coeffs = must_be_not_null(coeffs, true);
8843 ntt_zetas = must_be_not_null(ntt_zetas, true);
8844
8845 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8846 assert(coeffs_start, "coeffs is null");
8847 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
8848 assert(ntt_zetas_start, "ntt_zetas is null");
8849 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8850 OptoRuntime::dilithiumAlmostNtt_Type(),
8851 stubAddr, stubName, TypePtr::BOTTOM,
8852 coeffs_start, ntt_zetas_start);
8853 // return an int
8854 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
8855 set_result(retvalue);
8856 return true;
8857 }
8858
8859 //------------------------------inline_dilithiumAlmostInverseNtt
8860 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
8861 address stubAddr;
8862 const char *stubName;
8863 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8864 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
8865
8866 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
8867 stubName = "dilithiumAlmostInverseNtt";
8868 if (!stubAddr) return false;
8869
8870 Node* coeffs = argument(0);
8871 Node* zetas = argument(1);
8872
8873 coeffs = must_be_not_null(coeffs, true);
8874 zetas = must_be_not_null(zetas, true);
8875
8876 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8877 assert(coeffs_start, "coeffs is null");
8878 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
8879 assert(zetas_start, "inverseNtt_zetas is null");
8880 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8881 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
8882 stubAddr, stubName, TypePtr::BOTTOM,
8883 coeffs_start, zetas_start);
8884 // return an int
8885 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
8886 set_result(retvalue);
8887 return true;
8888 }
8889
8890 //------------------------------inline_dilithiumNttMult
8891 bool LibraryCallKit::inline_dilithiumNttMult() {
8892 address stubAddr;
8893 const char *stubName;
8894 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8895 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
8896
8897 stubAddr = StubRoutines::dilithiumNttMult();
8898 stubName = "dilithiumNttMult";
8899 if (!stubAddr) return false;
8900
8901 Node* result = argument(0);
8902 Node* ntta = argument(1);
8903 Node* nttb = argument(2);
8904 Node* zetas = argument(3);
8905
8906 result = must_be_not_null(result, true);
8907 ntta = must_be_not_null(ntta, true);
8908 nttb = must_be_not_null(nttb, true);
8909 zetas = must_be_not_null(zetas, true);
8910
8911 Node* result_start = array_element_address(result, intcon(0), T_INT);
8912 assert(result_start, "result is null");
8913 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
8914 assert(ntta_start, "ntta is null");
8915 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
8916 assert(nttb_start, "nttb is null");
8917 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8918 OptoRuntime::dilithiumNttMult_Type(),
8919 stubAddr, stubName, TypePtr::BOTTOM,
8920 result_start, ntta_start, nttb_start);
8921
8922 // return an int
8923 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8924 set_result(retvalue);
8925
8926 return true;
8927 }
8928
8929 //------------------------------inline_dilithiumMontMulByConstant
8930 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8931 address stubAddr;
8932 const char *stubName;
8933 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8934 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8935
8936 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8937 stubName = "dilithiumMontMulByConstant";
8938 if (!stubAddr) return false;
8939
8940 Node* coeffs = argument(0);
8941 Node* constant = argument(1);
8942
8943 coeffs = must_be_not_null(coeffs, true);
8944
8945 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8946 assert(coeffs_start, "coeffs is null");
8947 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8948 OptoRuntime::dilithiumMontMulByConstant_Type(),
8949 stubAddr, stubName, TypePtr::BOTTOM,
8950 coeffs_start, constant);
8951
8952 // return an int
8953 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8954 set_result(retvalue);
8955 return true;
8956 }
8957
8958
8959 //------------------------------inline_dilithiumDecomposePoly
8960 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8961 address stubAddr;
8962 const char *stubName;
8963 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8964 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8965
8966 stubAddr = StubRoutines::dilithiumDecomposePoly();
8967 stubName = "dilithiumDecomposePoly";
8968 if (!stubAddr) return false;
8969
8970 Node* input = argument(0);
8971 Node* lowPart = argument(1);
8972 Node* highPart = argument(2);
8973 Node* twoGamma2 = argument(3);
8974 Node* multiplier = argument(4);
8975
8976 input = must_be_not_null(input, true);
8977 lowPart = must_be_not_null(lowPart, true);
8978 highPart = must_be_not_null(highPart, true);
8979
8980 Node* input_start = array_element_address(input, intcon(0), T_INT);
8981 assert(input_start, "input is null");
8982 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8983 assert(lowPart_start, "lowPart is null");
8984 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8985 assert(highPart_start, "highPart is null");
8986
8987 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8988 OptoRuntime::dilithiumDecomposePoly_Type(),
8989 stubAddr, stubName, TypePtr::BOTTOM,
8990 input_start, lowPart_start, highPart_start,
8991 twoGamma2, multiplier);
8992
8993 // return an int
8994 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8995 set_result(retvalue);
8996 return true;
8997 }
8998
8999 bool LibraryCallKit::inline_base64_encodeBlock() {
9000 address stubAddr;
9001 const char *stubName;
9002 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
9003 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
9004 stubAddr = StubRoutines::base64_encodeBlock();
9005 stubName = "encodeBlock";
9006
9007 if (!stubAddr) return false;
9008 Node* base64obj = argument(0);
9009 Node* src = argument(1);
9010 Node* offset = argument(2);
9011 Node* len = argument(3);
9012 Node* dest = argument(4);
9013 Node* dp = argument(5);
9014 Node* isURL = argument(6);
9015
9016 src = must_be_not_null(src, true);
9017 dest = must_be_not_null(dest, true);
9018
9019 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
9020 assert(src_start, "source array is null");
9021 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
9022 assert(dest_start, "destination array is null");
9023
9024 Node* base64 = make_runtime_call(RC_LEAF,
9025 OptoRuntime::base64_encodeBlock_Type(),
9026 stubAddr, stubName, TypePtr::BOTTOM,
9027 src_start, offset, len, dest_start, dp, isURL);
9028 return true;
9029 }
9030
9031 bool LibraryCallKit::inline_base64_decodeBlock() {
9032 address stubAddr;
9033 const char *stubName;
9034 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
9035 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
9036 stubAddr = StubRoutines::base64_decodeBlock();
9037 stubName = "decodeBlock";
9038
9039 if (!stubAddr) return false;
9040 Node* base64obj = argument(0);
9041 Node* src = argument(1);
9042 Node* src_offset = argument(2);
9043 Node* len = argument(3);
9044 Node* dest = argument(4);
9045 Node* dest_offset = argument(5);
9046 Node* isURL = argument(6);
9047 Node* isMIME = argument(7);
9048
9049 src = must_be_not_null(src, true);
9050 dest = must_be_not_null(dest, true);
9051
9052 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
9053 assert(src_start, "source array is null");
9054 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
9055 assert(dest_start, "destination array is null");
9056
9057 Node* call = make_runtime_call(RC_LEAF,
9058 OptoRuntime::base64_decodeBlock_Type(),
9059 stubAddr, stubName, TypePtr::BOTTOM,
9060 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
9061 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
9062 set_result(result);
9063 return true;
9064 }
9065
9066 bool LibraryCallKit::inline_poly1305_processBlocks() {
9067 address stubAddr;
9068 const char *stubName;
9069 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
9070 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
9071 stubAddr = StubRoutines::poly1305_processBlocks();
9072 stubName = "poly1305_processBlocks";
9073
9074 if (!stubAddr) return false;
9075 null_check_receiver(); // null-check receiver
9076 if (stopped()) return true;
9077
9078 Node* input = argument(1);
9079 Node* input_offset = argument(2);
9080 Node* len = argument(3);
9081 Node* alimbs = argument(4);
9082 Node* rlimbs = argument(5);
9083
9084 input = must_be_not_null(input, true);
9085 alimbs = must_be_not_null(alimbs, true);
9086 rlimbs = must_be_not_null(rlimbs, true);
9087
9088 Node* input_start = array_element_address(input, input_offset, T_BYTE);
9089 assert(input_start, "input array is null");
9090 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
9091 assert(acc_start, "acc array is null");
9092 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
9093 assert(r_start, "r array is null");
9094
9095 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
9096 OptoRuntime::poly1305_processBlocks_Type(),
9097 stubAddr, stubName, TypePtr::BOTTOM,
9098 input_start, len, acc_start, r_start);
9099 return true;
9100 }
9101
9102 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
9103 address stubAddr;
9104 const char *stubName;
9105 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
9106 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
9107 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
9108 stubName = "intpoly_montgomeryMult_P256";
9109
9110 if (!stubAddr) return false;
9111 null_check_receiver(); // null-check receiver
9112 if (stopped()) return true;
9113
9114 Node* a = argument(1);
9115 Node* b = argument(2);
9116 Node* r = argument(3);
9117
9118 a = must_be_not_null(a, true);
9119 b = must_be_not_null(b, true);
9120 r = must_be_not_null(r, true);
9121
9122 Node* a_start = array_element_address(a, intcon(0), T_LONG);
9123 assert(a_start, "a array is null");
9124 Node* b_start = array_element_address(b, intcon(0), T_LONG);
9125 assert(b_start, "b array is null");
9126 Node* r_start = array_element_address(r, intcon(0), T_LONG);
9127 assert(r_start, "r array is null");
9128
9129 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
9130 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
9131 stubAddr, stubName, TypePtr::BOTTOM,
9132 a_start, b_start, r_start);
9133 return true;
9134 }
9135
9136 bool LibraryCallKit::inline_intpoly_assign() {
9137 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
9138 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
9139 const char *stubName = "intpoly_assign";
9140 address stubAddr = StubRoutines::intpoly_assign();
9141 if (!stubAddr) return false;
9142
9143 Node* set = argument(0);
9144 Node* a = argument(1);
9145 Node* b = argument(2);
9146 Node* arr_length = load_array_length(a);
9147
9148 a = must_be_not_null(a, true);
9149 b = must_be_not_null(b, true);
9150
9151 Node* a_start = array_element_address(a, intcon(0), T_LONG);
9152 assert(a_start, "a array is null");
9153 Node* b_start = array_element_address(b, intcon(0), T_LONG);
9154 assert(b_start, "b array is null");
9155
9156 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
9157 OptoRuntime::intpoly_assign_Type(),
9158 stubAddr, stubName, TypePtr::BOTTOM,
9159 set, a_start, b_start, arr_length);
9160 return true;
9161 }
9162
9163 bool LibraryCallKit::inline_intpoly_mult_25519() {
9164 address stubAddr;
9165 const char *stubName;
9166 assert(UseIntPoly25519Intrinsics, "need intpoly25519 intrinsics support");
9167 assert(callee()->signature()->size() == 3, "intpoly_mult_25519 has %d parameters", callee()->signature()->size());
9168 stubAddr = StubRoutines::intpoly_mult_25519();
9169 stubName = "intpoly_mult_25519";
9170
9171 if (!stubAddr) return false;
9172 null_check_receiver(); // null-check receiver
9173 if (stopped()) return true;
9174
9175 Node* a = argument(1);
9176 Node* b = argument(2);
9177 Node* r = argument(3);
9178
9179 a = must_be_not_null(a, true);
9180 b = must_be_not_null(b, true);
9181 r = must_be_not_null(r, true);
9182
9183 Node* a_start = array_element_address(a, intcon(0), T_LONG);
9184 assert(a_start, "a array is null");
9185 Node* b_start = array_element_address(b, intcon(0), T_LONG);
9186 assert(b_start, "b array is null");
9187 Node* r_start = array_element_address(r, intcon(0), T_LONG);
9188 assert(r_start, "r array is null");
9189
9190 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
9191 OptoRuntime::intpoly_mult_25519_Type(),
9192 stubAddr, stubName, TypePtr::BOTTOM,
9193 a_start, b_start, r_start);
9194 return true;
9195 }
9196
9197 bool LibraryCallKit::inline_intpoly_square_25519() {
9198 address stubAddr;
9199 const char *stubName;
9200 assert(UseIntPoly25519Intrinsics, "need intpoly25519 intrinsics support");
9201 assert(callee()->signature()->size() == 2, "intpoly_mult_25519 has %d parameters", callee()->signature()->size());
9202 stubAddr = StubRoutines::intpoly_square_25519();
9203 stubName = "intpoly_square_25519";
9204
9205 if (!stubAddr) return false;
9206 null_check_receiver(); // null-check receiver
9207 if (stopped()) return true;
9208
9209 Node* a = argument(1);
9210 Node* r = argument(2);
9211
9212 a = must_be_not_null(a, true);
9213 r = must_be_not_null(r, true);
9214
9215 Node* a_start = array_element_address(a, intcon(0), T_LONG);
9216 assert(a_start, "a array is null");
9217 Node* r_start = array_element_address(r, intcon(0), T_LONG);
9218 assert(r_start, "r array is null");
9219
9220 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
9221 OptoRuntime::intpoly_square_25519_Type(),
9222 stubAddr, stubName, TypePtr::BOTTOM,
9223 a_start, r_start);
9224 return true;
9225 }
9226
9227 //------------------------------inline_digestBase_implCompress-----------------------
9228 //
9229 // Calculate MD5 for single-block byte[] array.
9230 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
9231 //
9232 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
9233 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
9234 //
9235 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
9236 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
9237 //
9238 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
9239 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
9240 //
9241 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
9242 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
9243 //
9244 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
9245 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
9246
9247 Node* digestBase_obj = argument(0);
9248 Node* src = argument(1); // type oop
9249 Node* ofs = argument(2); // type int
9250
9251 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
9252 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
9253 // failed array check
9254 return false;
9255 }
9256 // Figure out the size and type of the elements we will be copying.
9257 BasicType src_elem = src_type->elem()->array_element_basic_type();
9258 if (src_elem != T_BYTE) {
9259 return false;
9260 }
9261 // 'src_start' points to src array + offset
9262 src = must_be_not_null(src, true);
9263 Node* src_start = array_element_address(src, ofs, src_elem);
9264 Node* state = nullptr;
9265 Node* block_size = nullptr;
9266 address stubAddr;
9267 const char *stubName;
9268
9269 switch(id) {
9270 case vmIntrinsics::_md5_implCompress:
9271 assert(UseMD5Intrinsics, "need MD5 instruction support");
9272 state = get_state_from_digest_object(digestBase_obj, T_INT);
9273 stubAddr = StubRoutines::md5_implCompress();
9274 stubName = "md5_implCompress";
9275 break;
9276 case vmIntrinsics::_sha_implCompress:
9277 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
9278 state = get_state_from_digest_object(digestBase_obj, T_INT);
9279 stubAddr = StubRoutines::sha1_implCompress();
9280 stubName = "sha1_implCompress";
9281 break;
9282 case vmIntrinsics::_sha2_implCompress:
9283 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
9284 state = get_state_from_digest_object(digestBase_obj, T_INT);
9285 stubAddr = StubRoutines::sha256_implCompress();
9286 stubName = "sha256_implCompress";
9287 break;
9288 case vmIntrinsics::_sha5_implCompress:
9289 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
9290 state = get_state_from_digest_object(digestBase_obj, T_LONG);
9291 stubAddr = StubRoutines::sha512_implCompress();
9292 stubName = "sha512_implCompress";
9293 break;
9294 case vmIntrinsics::_sha3_implCompress:
9295 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
9296 state = get_state_from_digest_object(digestBase_obj, T_LONG);
9297 stubAddr = StubRoutines::sha3_implCompress();
9298 stubName = "sha3_implCompress";
9299 block_size = get_block_size_from_digest_object(digestBase_obj);
9300 if (block_size == nullptr) return false;
9301 break;
9302 default:
9303 fatal_unexpected_iid(id);
9304 return false;
9305 }
9306 if (state == nullptr) return false;
9307
9308 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
9309 if (stubAddr == nullptr) return false;
9310
9311 // Call the stub.
9312 Node* call;
9313 if (block_size == nullptr) {
9314 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
9315 stubAddr, stubName, TypePtr::BOTTOM,
9316 src_start, state);
9317 } else {
9318 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
9319 stubAddr, stubName, TypePtr::BOTTOM,
9320 src_start, state, block_size);
9321 }
9322
9323 return true;
9324 }
9325
9326 //------------------------------inline_keccak
9327 bool LibraryCallKit::inline_keccak(vmIntrinsics::ID id) {
9328 address stubAddr = nullptr;
9329 const char *stubName;
9330 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
9331 assert((id == vmIntrinsics::_double_keccak && callee()->signature()->size() == 2) ||
9332 (id == vmIntrinsics::_quad_keccak && callee()->signature()->size() == 4),
9333 "double_keccak wrong number of parameters");
9334
9335 int parmCnt = 0;
9336 switch (id) {
9337 case vmIntrinsics::_double_keccak:
9338 stubAddr = StubRoutines::double_keccak();
9339 stubName = "double_keccak";
9340 parmCnt = 2;
9341 break;
9342 case vmIntrinsics::_quad_keccak:
9343 stubAddr = StubRoutines::quad_keccak();
9344 stubName = "quad_keccak";
9345 parmCnt = 4;
9346 break;
9347 default:
9348 ShouldNotReachHere();
9349 }
9350
9351 if (!stubAddr) return false;
9352
9353 Node* state[4];
9354 for (int i = 0; i<parmCnt; i++) {
9355 state[i] = must_be_not_null(argument(i), true);
9356 state[i] = array_element_address(state[i], intcon(0), T_LONG);
9357 assert(state[i], "state[%d] is null", i);
9358 }
9359
9360 Node* keccak;
9361 switch (id) {
9362 case vmIntrinsics::_double_keccak:
9363 keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
9364 OptoRuntime::double_keccak_Type(),
9365 stubAddr, stubName, TypePtr::BOTTOM,
9366 state[0], state[1]);
9367 break;
9368 case vmIntrinsics::_quad_keccak:
9369 keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
9370 OptoRuntime::quad_keccak_Type(),
9371 stubAddr, stubName, TypePtr::BOTTOM,
9372 state[0], state[1], state[2], state[3]);
9373 break;
9374 default:
9375 ShouldNotReachHere();
9376 }
9377
9378 // return an int
9379 Node* retvalue = _gvn.transform(new ProjNode(keccak, TypeFunc::Parms));
9380 set_result(retvalue);
9381 return true;
9382 }
9383
9384
9385 //------------------------------inline_digestBase_implCompressMB-----------------------
9386 //
9387 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
9388 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
9389 //
9390 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
9391 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9392 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9393 assert((uint)predicate < 5, "sanity");
9394 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
9395
9396 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
9397 Node* src = argument(1); // byte[] array
9398 Node* ofs = argument(2); // type int
9399 Node* limit = argument(3); // type int
9400
9401 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
9402 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
9403 // failed array check
9404 return false;
9405 }
9406 // Figure out the size and type of the elements we will be copying.
9407 BasicType src_elem = src_type->elem()->array_element_basic_type();
9408 if (src_elem != T_BYTE) {
9409 return false;
9410 }
9411 // 'src_start' points to src array + offset
9412 src = must_be_not_null(src, false);
9413 Node* src_start = array_element_address(src, ofs, src_elem);
9414
9415 const char* klass_digestBase_name = nullptr;
9416 const char* stub_name = nullptr;
9417 address stub_addr = nullptr;
9418 BasicType elem_type = T_INT;
9419
9420 switch (predicate) {
9421 case 0:
9422 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
9423 klass_digestBase_name = "sun/security/provider/MD5";
9424 stub_name = "md5_implCompressMB";
9425 stub_addr = StubRoutines::md5_implCompressMB();
9426 }
9427 break;
9428 case 1:
9429 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
9430 klass_digestBase_name = "sun/security/provider/SHA";
9431 stub_name = "sha1_implCompressMB";
9432 stub_addr = StubRoutines::sha1_implCompressMB();
9433 }
9434 break;
9435 case 2:
9436 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
9437 klass_digestBase_name = "sun/security/provider/SHA2";
9438 stub_name = "sha256_implCompressMB";
9439 stub_addr = StubRoutines::sha256_implCompressMB();
9440 }
9441 break;
9442 case 3:
9443 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
9444 klass_digestBase_name = "sun/security/provider/SHA5";
9445 stub_name = "sha512_implCompressMB";
9446 stub_addr = StubRoutines::sha512_implCompressMB();
9447 elem_type = T_LONG;
9448 }
9449 break;
9450 case 4:
9451 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
9452 klass_digestBase_name = "sun/security/provider/SHA3";
9453 stub_name = "sha3_implCompressMB";
9454 stub_addr = StubRoutines::sha3_implCompressMB();
9455 elem_type = T_LONG;
9456 }
9457 break;
9458 default:
9459 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
9460 }
9461 if (klass_digestBase_name != nullptr) {
9462 assert(stub_addr != nullptr, "Stub is generated");
9463 if (stub_addr == nullptr) return false;
9464
9465 // get DigestBase klass to lookup for SHA klass
9466 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
9467 assert(tinst != nullptr, "digestBase_obj is not instance???");
9468 assert(tinst->is_loaded(), "DigestBase is not loaded");
9469
9470 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
9471 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
9472 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
9473 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
9474 }
9475 return false;
9476 }
9477
9478 //------------------------------inline_digestBase_implCompressMB-----------------------
9479 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
9480 BasicType elem_type, address stubAddr, const char *stubName,
9481 Node* src_start, Node* ofs, Node* limit) {
9482 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
9483 const TypeOopPtr* xtype = aklass->as_subtype_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
9484 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
9485 digest_obj = _gvn.transform(digest_obj);
9486
9487 Node* state = get_state_from_digest_object(digest_obj, elem_type);
9488 if (state == nullptr) return false;
9489
9490 Node* block_size = nullptr;
9491 if (strcmp("sha3_implCompressMB", stubName) == 0) {
9492 block_size = get_block_size_from_digest_object(digest_obj);
9493 if (block_size == nullptr) return false;
9494 }
9495
9496 // Call the stub.
9497 Node* call;
9498 if (block_size == nullptr) {
9499 call = make_runtime_call(RC_LEAF|RC_NO_FP,
9500 OptoRuntime::digestBase_implCompressMB_Type(false),
9501 stubAddr, stubName, TypePtr::BOTTOM,
9502 src_start, state, ofs, limit);
9503 } else {
9504 call = make_runtime_call(RC_LEAF|RC_NO_FP,
9505 OptoRuntime::digestBase_implCompressMB_Type(true),
9506 stubAddr, stubName, TypePtr::BOTTOM,
9507 src_start, state, block_size, ofs, limit);
9508 }
9509
9510 // return ofs (int)
9511 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
9512 set_result(result);
9513
9514 return true;
9515 }
9516
9517 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
9518 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
9519 assert(UseAES, "need AES instruction support");
9520 address stubAddr = nullptr;
9521 const char *stubName = nullptr;
9522 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
9523 stubName = "galoisCounterMode_AESCrypt";
9524
9525 if (stubAddr == nullptr) return false;
9526
9527 Node* in = argument(0);
9528 Node* inOfs = argument(1);
9529 Node* len = argument(2);
9530 Node* ct = argument(3);
9531 Node* ctOfs = argument(4);
9532 Node* out = argument(5);
9533 Node* outOfs = argument(6);
9534 Node* gctr_object = argument(7);
9535 Node* ghash_object = argument(8);
9536
9537 // (1) in, ct and out are arrays.
9538 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
9539 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
9540 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
9541 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
9542 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
9543 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
9544
9545 // checks are the responsibility of the caller
9546 Node* in_start = in;
9547 Node* ct_start = ct;
9548 Node* out_start = out;
9549 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
9550 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
9551 in_start = array_element_address(in, inOfs, T_BYTE);
9552 ct_start = array_element_address(ct, ctOfs, T_BYTE);
9553 out_start = array_element_address(out, outOfs, T_BYTE);
9554 }
9555
9556 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
9557 // (because of the predicated logic executed earlier).
9558 // so we cast it here safely.
9559 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
9560 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9561 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
9562 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
9563 Node* state = load_field_from_object(ghash_object, "state", "[J");
9564
9565 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
9566 return false;
9567 }
9568 // cast it to what we know it will be at runtime
9569 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
9570 assert(tinst != nullptr, "GCTR obj is null");
9571 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9572 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
9573 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
9574 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9575 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
9576 const TypeOopPtr* xtype = aklass->as_exact_instance_type();
9577 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
9578 aescrypt_object = _gvn.transform(aescrypt_object);
9579 // we need to get the start of the aescrypt_object's expanded key array
9580 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
9581 if (k_start == nullptr) return false;
9582 // similarly, get the start address of the r vector
9583 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
9584 Node* state_start = array_element_address(state, intcon(0), T_LONG);
9585 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
9586
9587
9588 // Call the stub, passing params
9589 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
9590 OptoRuntime::galoisCounterMode_aescrypt_Type(),
9591 stubAddr, stubName, TypePtr::BOTTOM,
9592 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
9593
9594 // return cipher length (int)
9595 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
9596 set_result(retvalue);
9597
9598 return true;
9599 }
9600
9601 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
9602 // Return node representing slow path of predicate check.
9603 // the pseudo code we want to emulate with this predicate is:
9604 // for encryption:
9605 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
9606 // for decryption:
9607 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
9608 // note cipher==plain is more conservative than the original java code but that's OK
9609 //
9610
9611 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
9612 // The receiver was checked for null already.
9613 Node* objGCTR = argument(7);
9614 // Load embeddedCipher field of GCTR object.
9615 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9616 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
9617
9618 // get AESCrypt klass for instanceOf check
9619 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
9620 // will have same classloader as CipherBlockChaining object
9621 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
9622 assert(tinst != nullptr, "GCTR obj is null");
9623 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9624
9625 // we want to do an instanceof comparison against the AESCrypt class
9626 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
9627 if (!klass_AESCrypt->is_loaded()) {
9628 // if AESCrypt is not even loaded, we never take the intrinsic fast path
9629 Node* ctrl = control();
9630 set_control(top()); // no regular fast path
9631 return ctrl;
9632 }
9633
9634 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9635 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
9636 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9637 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9638 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9639
9640 return instof_false; // even if it is null
9641 }
9642
9643 //------------------------------get_state_from_digest_object-----------------------
9644 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
9645 const char* state_type;
9646 switch (elem_type) {
9647 case T_BYTE: state_type = "[B"; break;
9648 case T_INT: state_type = "[I"; break;
9649 case T_LONG: state_type = "[J"; break;
9650 default: ShouldNotReachHere();
9651 }
9652 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
9653 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
9654 if (digest_state == nullptr) return (Node *) nullptr;
9655
9656 // now have the array, need to get the start address of the state array
9657 Node* state = array_element_address(digest_state, intcon(0), elem_type);
9658 return state;
9659 }
9660
9661 //------------------------------get_block_size_from_sha3_object----------------------------------
9662 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
9663 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
9664 assert (block_size != nullptr, "sanity");
9665 return block_size;
9666 }
9667
9668 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
9669 // Return node representing slow path of predicate check.
9670 // the pseudo code we want to emulate with this predicate is:
9671 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
9672 //
9673 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
9674 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9675 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9676 assert((uint)predicate < 5, "sanity");
9677
9678 // The receiver was checked for null already.
9679 Node* digestBaseObj = argument(0);
9680
9681 // get DigestBase klass for instanceOf check
9682 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
9683 assert(tinst != nullptr, "digestBaseObj is null");
9684 assert(tinst->is_loaded(), "DigestBase is not loaded");
9685
9686 const char* klass_name = nullptr;
9687 switch (predicate) {
9688 case 0:
9689 if (UseMD5Intrinsics) {
9690 // we want to do an instanceof comparison against the MD5 class
9691 klass_name = "sun/security/provider/MD5";
9692 }
9693 break;
9694 case 1:
9695 if (UseSHA1Intrinsics) {
9696 // we want to do an instanceof comparison against the SHA class
9697 klass_name = "sun/security/provider/SHA";
9698 }
9699 break;
9700 case 2:
9701 if (UseSHA256Intrinsics) {
9702 // we want to do an instanceof comparison against the SHA2 class
9703 klass_name = "sun/security/provider/SHA2";
9704 }
9705 break;
9706 case 3:
9707 if (UseSHA512Intrinsics) {
9708 // we want to do an instanceof comparison against the SHA5 class
9709 klass_name = "sun/security/provider/SHA5";
9710 }
9711 break;
9712 case 4:
9713 if (UseSHA3Intrinsics) {
9714 // we want to do an instanceof comparison against the SHA3 class
9715 klass_name = "sun/security/provider/SHA3";
9716 }
9717 break;
9718 default:
9719 fatal("unknown SHA intrinsic predicate: %d", predicate);
9720 }
9721
9722 ciKlass* klass = nullptr;
9723 if (klass_name != nullptr) {
9724 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
9725 }
9726 if ((klass == nullptr) || !klass->is_loaded()) {
9727 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
9728 Node* ctrl = control();
9729 set_control(top()); // no intrinsic path
9730 return ctrl;
9731 }
9732 ciInstanceKlass* instklass = klass->as_instance_klass();
9733
9734 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
9735 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9736 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9737 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9738
9739 return instof_false; // even if it is null
9740 }
9741
9742 //-------------inline_fma-----------------------------------
9743 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
9744 Node *a = nullptr;
9745 Node *b = nullptr;
9746 Node *c = nullptr;
9747 Node* result = nullptr;
9748 switch (id) {
9749 case vmIntrinsics::_fmaD:
9750 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
9751 // no receiver since it is static method
9752 a = argument(0);
9753 b = argument(2);
9754 c = argument(4);
9755 result = _gvn.transform(new FmaDNode(a, b, c));
9756 break;
9757 case vmIntrinsics::_fmaF:
9758 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
9759 a = argument(0);
9760 b = argument(1);
9761 c = argument(2);
9762 result = _gvn.transform(new FmaFNode(a, b, c));
9763 break;
9764 default:
9765 fatal_unexpected_iid(id); break;
9766 }
9767 set_result(result);
9768 return true;
9769 }
9770
9771 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
9772 // argument(0) is receiver
9773 Node* codePoint = argument(1);
9774 Node* n = nullptr;
9775
9776 switch (id) {
9777 case vmIntrinsics::_isDigit :
9778 n = new DigitNode(control(), codePoint);
9779 break;
9780 case vmIntrinsics::_isLowerCase :
9781 n = new LowerCaseNode(control(), codePoint);
9782 break;
9783 case vmIntrinsics::_isUpperCase :
9784 n = new UpperCaseNode(control(), codePoint);
9785 break;
9786 case vmIntrinsics::_isWhitespace :
9787 n = new WhitespaceNode(control(), codePoint);
9788 break;
9789 default:
9790 fatal_unexpected_iid(id);
9791 }
9792
9793 set_result(_gvn.transform(n));
9794 return true;
9795 }
9796
9797 bool LibraryCallKit::inline_profileBoolean() {
9798 Node* counts = argument(1);
9799 const TypeAryPtr* ary = nullptr;
9800 ciArray* aobj = nullptr;
9801 if (counts->is_Con()
9802 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
9803 && (aobj = ary->const_oop()->as_array()) != nullptr
9804 && (aobj->length() == 2)) {
9805 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
9806 jint false_cnt = aobj->element_value(0).as_int();
9807 jint true_cnt = aobj->element_value(1).as_int();
9808
9809 if (C->log() != nullptr) {
9810 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
9811 false_cnt, true_cnt);
9812 }
9813
9814 if (false_cnt + true_cnt == 0) {
9815 // According to profile, never executed.
9816 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9817 Deoptimization::Action_reinterpret);
9818 return true;
9819 }
9820
9821 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
9822 // is a number of each value occurrences.
9823 Node* result = argument(0);
9824 if (false_cnt == 0 || true_cnt == 0) {
9825 // According to profile, one value has been never seen.
9826 int expected_val = (false_cnt == 0) ? 1 : 0;
9827
9828 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
9829 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
9830
9831 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
9832 Node* fast_path = _gvn.transform(new IfTrueNode(check));
9833 Node* slow_path = _gvn.transform(new IfFalseNode(check));
9834
9835 { // Slow path: uncommon trap for never seen value and then reexecute
9836 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
9837 // the value has been seen at least once.
9838 PreserveJVMState pjvms(this);
9839 PreserveReexecuteState preexecs(this);
9840 jvms()->set_should_reexecute(true);
9841
9842 set_control(slow_path);
9843 set_i_o(i_o());
9844
9845 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9846 Deoptimization::Action_reinterpret);
9847 }
9848 // The guard for never seen value enables sharpening of the result and
9849 // returning a constant. It allows to eliminate branches on the same value
9850 // later on.
9851 set_control(fast_path);
9852 result = intcon(expected_val);
9853 }
9854 // Stop profiling.
9855 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
9856 // By replacing method body with profile data (represented as ProfileBooleanNode
9857 // on IR level) we effectively disable profiling.
9858 // It enables full speed execution once optimized code is generated.
9859 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
9860 C->record_for_igvn(profile);
9861 set_result(profile);
9862 return true;
9863 } else {
9864 // Continue profiling.
9865 // Profile data isn't available at the moment. So, execute method's bytecode version.
9866 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
9867 // is compiled and counters aren't available since corresponding MethodHandle
9868 // isn't a compile-time constant.
9869 return false;
9870 }
9871 }
9872
9873 bool LibraryCallKit::inline_isCompileConstant() {
9874 Node* n = argument(0);
9875 set_result(n->is_Con() ? intcon(1) : intcon(0));
9876 return true;
9877 }
9878
9879 //------------------------------- inline_getObjectSize --------------------------------------
9880 //
9881 // Calculate the runtime size of the object/array.
9882 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
9883 //
9884 bool LibraryCallKit::inline_getObjectSize() {
9885 Node* obj = argument(3);
9886 Node* klass_node = load_object_klass(obj);
9887
9888 jint layout_con = Klass::_lh_neutral_value;
9889 Node* layout_val = get_layout_helper(klass_node, layout_con);
9890 int layout_is_con = (layout_val == nullptr);
9891
9892 if (layout_is_con) {
9893 // Layout helper is constant, can figure out things at compile time.
9894
9895 if (Klass::layout_helper_is_instance(layout_con)) {
9896 // Instance case: layout_con contains the size itself.
9897 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
9898 set_result(size);
9899 } else {
9900 // Array case: size is round(header + element_size*arraylength).
9901 // Since arraylength is different for every array instance, we have to
9902 // compute the whole thing at runtime.
9903
9904 Node* arr_length = load_array_length(obj);
9905
9906 int round_mask = MinObjAlignmentInBytes - 1;
9907 int hsize = Klass::layout_helper_header_size(layout_con);
9908 int eshift = Klass::layout_helper_log2_element_size(layout_con);
9909
9910 if ((round_mask & ~right_n_bits(eshift)) == 0) {
9911 round_mask = 0; // strength-reduce it if it goes away completely
9912 }
9913 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
9914 Node* header_size = intcon(hsize + round_mask);
9915
9916 Node* lengthx = ConvI2X(arr_length);
9917 Node* headerx = ConvI2X(header_size);
9918
9919 Node* abody = lengthx;
9920 if (eshift != 0) {
9921 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
9922 }
9923 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
9924 if (round_mask != 0) {
9925 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
9926 }
9927 size = ConvX2L(size);
9928 set_result(size);
9929 }
9930 } else {
9931 // Layout helper is not constant, need to test for array-ness at runtime.
9932
9933 enum { _instance_path = 1, _array_path, PATH_LIMIT };
9934 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
9935 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
9936 record_for_igvn(result_reg);
9937
9938 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
9939 if (array_ctl != nullptr) {
9940 // Array case: size is round(header + element_size*arraylength).
9941 // Since arraylength is different for every array instance, we have to
9942 // compute the whole thing at runtime.
9943
9944 PreserveJVMState pjvms(this);
9945 set_control(array_ctl);
9946 Node* arr_length = load_array_length(obj);
9947
9948 int round_mask = MinObjAlignmentInBytes - 1;
9949 Node* mask = intcon(round_mask);
9950
9951 Node* hss = intcon(Klass::_lh_header_size_shift);
9952 Node* hsm = intcon(Klass::_lh_header_size_mask);
9953 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
9954 header_size = _gvn.transform(new AndINode(header_size, hsm));
9955 header_size = _gvn.transform(new AddINode(header_size, mask));
9956
9957 // There is no need to mask or shift this value.
9958 // The semantics of LShiftINode include an implicit mask to 0x1F.
9959 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
9960 Node* elem_shift = layout_val;
9961
9962 Node* lengthx = ConvI2X(arr_length);
9963 Node* headerx = ConvI2X(header_size);
9964
9965 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
9966 Node* size = _gvn.transform(new AddXNode(headerx, abody));
9967 if (round_mask != 0) {
9968 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
9969 }
9970 size = ConvX2L(size);
9971
9972 result_reg->init_req(_array_path, control());
9973 result_val->init_req(_array_path, size);
9974 }
9975
9976 if (!stopped()) {
9977 // Instance case: the layout helper gives us instance size almost directly,
9978 // but we need to mask out the _lh_instance_slow_path_bit.
9979 Node* size = ConvI2X(layout_val);
9980 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
9981 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
9982 size = _gvn.transform(new AndXNode(size, mask));
9983 size = ConvX2L(size);
9984
9985 result_reg->init_req(_instance_path, control());
9986 result_val->init_req(_instance_path, size);
9987 }
9988
9989 set_result(result_reg, result_val);
9990 }
9991
9992 return true;
9993 }
9994
9995 //------------------------------- inline_blackhole --------------------------------------
9996 //
9997 // Make sure all arguments to this node are alive.
9998 // This matches methods that were requested to be blackholed through compile commands.
9999 //
10000 bool LibraryCallKit::inline_blackhole() {
10001 assert(callee()->is_static(), "Should have been checked before: only static methods here");
10002 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
10003 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
10004
10005 // Blackhole node pinches only the control, not memory. This allows
10006 // the blackhole to be pinned in the loop that computes blackholed
10007 // values, but have no other side effects, like breaking the optimizations
10008 // across the blackhole.
10009
10010 Node* bh = _gvn.transform(new BlackholeNode(control()));
10011 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
10012
10013 // Bind call arguments as blackhole arguments to keep them alive
10014 uint nargs = callee()->arg_size();
10015 for (uint i = 0; i < nargs; i++) {
10016 bh->add_req(argument(i));
10017 }
10018
10019 return true;
10020 }
10021
10022 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
10023 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
10024 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
10025 return nullptr; // box klass is not Float16
10026 }
10027
10028 // Null check; get notnull casted pointer
10029 Node* null_ctl = top();
10030 Node* not_null_box = null_check_oop(box, &null_ctl, true);
10031 // If not_null_box is dead, only null-path is taken
10032 if (stopped()) {
10033 set_control(null_ctl);
10034 return nullptr;
10035 }
10036 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
10037 const TypePtr* adr_type = C->alias_type(field)->adr_type();
10038 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
10039 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
10040 }
10041
10042 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
10043 PreserveReexecuteState preexecs(this);
10044 jvms()->set_should_reexecute(true);
10045
10046 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
10047 Node* klass_node = makecon(klass_type);
10048 Node* box = new_instance(klass_node);
10049
10050 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
10051 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
10052
10053 Node* field_store = _gvn.transform(access_store_at(box,
10054 value_field,
10055 value_adr_type,
10056 value,
10057 TypeInt::SHORT,
10058 T_SHORT,
10059 IN_HEAP));
10060 set_memory(field_store, value_adr_type);
10061 return box;
10062 }
10063
10064 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
10065 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
10066 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
10067 return false;
10068 }
10069
10070 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
10071 if (box_type == nullptr || box_type->const_oop() == nullptr) {
10072 return false;
10073 }
10074
10075 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
10076 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
10077 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
10078 ciSymbols::short_signature(),
10079 false);
10080 assert(field != nullptr, "");
10081
10082 // Transformed nodes
10083 Node* fld1 = nullptr;
10084 Node* fld2 = nullptr;
10085 Node* fld3 = nullptr;
10086 switch(num_args) {
10087 case 3:
10088 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
10089 if (fld3 == nullptr) {
10090 return false;
10091 }
10092 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
10093 // fall-through
10094 case 2:
10095 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
10096 if (fld2 == nullptr) {
10097 return false;
10098 }
10099 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
10100 // fall-through
10101 case 1:
10102 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
10103 if (fld1 == nullptr) {
10104 return false;
10105 }
10106 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
10107 break;
10108 default: fatal("Unsupported number of arguments %d", num_args);
10109 }
10110
10111 Node* result = nullptr;
10112 switch (id) {
10113 // Unary operations
10114 case vmIntrinsics::_sqrt_float16:
10115 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
10116 break;
10117 // Ternary operations
10118 case vmIntrinsics::_fma_float16:
10119 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
10120 break;
10121 default:
10122 fatal_unexpected_iid(id);
10123 break;
10124 }
10125 result = _gvn.transform(new ReinterpretHF2SNode(result));
10126 set_result(box_fp16_value(float16_box_type, field, result));
10127 return true;
10128 }
10129