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 return inline_double_keccak();
630
631 case vmIntrinsics::_digestBase_implCompressMB:
632 return inline_digestBase_implCompressMB(predicate);
633
634 case vmIntrinsics::_multiplyToLen:
635 return inline_multiplyToLen();
636
637 case vmIntrinsics::_squareToLen:
638 return inline_squareToLen();
639
640 case vmIntrinsics::_mulAdd:
641 return inline_mulAdd();
642
643 case vmIntrinsics::_montgomeryMultiply:
644 return inline_montgomeryMultiply();
645 case vmIntrinsics::_montgomerySquare:
646 return inline_montgomerySquare();
647
648 case vmIntrinsics::_bigIntegerRightShiftWorker:
649 return inline_bigIntegerShift(true);
650 case vmIntrinsics::_bigIntegerLeftShiftWorker:
651 return inline_bigIntegerShift(false);
652
653 case vmIntrinsics::_vectorizedMismatch:
654 return inline_vectorizedMismatch();
655
656 case vmIntrinsics::_ghash_processBlocks:
657 return inline_ghash_processBlocks();
658 case vmIntrinsics::_chacha20Block:
659 return inline_chacha20Block();
660 case vmIntrinsics::_kyberNtt:
661 return inline_kyberNtt();
662 case vmIntrinsics::_kyberInverseNtt:
663 return inline_kyberInverseNtt();
664 case vmIntrinsics::_kyberNttMult:
665 return inline_kyberNttMult();
666 case vmIntrinsics::_kyberAddPoly_2:
667 return inline_kyberAddPoly_2();
668 case vmIntrinsics::_kyberAddPoly_3:
669 return inline_kyberAddPoly_3();
670 case vmIntrinsics::_kyber12To16:
671 return inline_kyber12To16();
672 case vmIntrinsics::_kyberBarrettReduce:
673 return inline_kyberBarrettReduce();
674 case vmIntrinsics::_dilithiumAlmostNtt:
675 return inline_dilithiumAlmostNtt();
676 case vmIntrinsics::_dilithiumAlmostInverseNtt:
677 return inline_dilithiumAlmostInverseNtt();
678 case vmIntrinsics::_dilithiumNttMult:
679 return inline_dilithiumNttMult();
680 case vmIntrinsics::_dilithiumMontMulByConstant:
681 return inline_dilithiumMontMulByConstant();
682 case vmIntrinsics::_dilithiumDecomposePoly:
683 return inline_dilithiumDecomposePoly();
684 case vmIntrinsics::_base64_encodeBlock:
685 return inline_base64_encodeBlock();
686 case vmIntrinsics::_base64_decodeBlock:
687 return inline_base64_decodeBlock();
688 case vmIntrinsics::_poly1305_processBlocks:
689 return inline_poly1305_processBlocks();
690 case vmIntrinsics::_intpoly_montgomeryMult_P256:
691 return inline_intpoly_montgomeryMult_P256();
692 case vmIntrinsics::_intpoly_assign:
693 return inline_intpoly_assign();
694 case vmIntrinsics::_encodeISOArray:
695 case vmIntrinsics::_encodeByteISOArray:
696 return inline_encodeISOArray(false);
697 case vmIntrinsics::_encodeAsciiArray:
698 return inline_encodeISOArray(true);
699
700 case vmIntrinsics::_updateCRC32:
701 return inline_updateCRC32();
702 case vmIntrinsics::_updateBytesCRC32:
703 return inline_updateBytesCRC32();
704 case vmIntrinsics::_updateByteBufferCRC32:
705 return inline_updateByteBufferCRC32();
706
707 case vmIntrinsics::_updateBytesCRC32C:
708 return inline_updateBytesCRC32C();
709 case vmIntrinsics::_updateDirectByteBufferCRC32C:
710 return inline_updateDirectByteBufferCRC32C();
711
712 case vmIntrinsics::_updateBytesAdler32:
713 return inline_updateBytesAdler32();
714 case vmIntrinsics::_updateByteBufferAdler32:
715 return inline_updateByteBufferAdler32();
716
717 case vmIntrinsics::_profileBoolean:
718 return inline_profileBoolean();
719 case vmIntrinsics::_isCompileConstant:
720 return inline_isCompileConstant();
721
722 case vmIntrinsics::_countPositives:
723 return inline_countPositives();
724
725 case vmIntrinsics::_fmaD:
726 case vmIntrinsics::_fmaF:
727 return inline_fma(intrinsic_id());
728
729 case vmIntrinsics::_isDigit:
730 case vmIntrinsics::_isLowerCase:
731 case vmIntrinsics::_isUpperCase:
732 case vmIntrinsics::_isWhitespace:
733 return inline_character_compare(intrinsic_id());
734
735 case vmIntrinsics::_min:
736 case vmIntrinsics::_max:
737 case vmIntrinsics::_min_strict:
738 case vmIntrinsics::_max_strict:
739 case vmIntrinsics::_minL:
740 case vmIntrinsics::_maxL:
741 case vmIntrinsics::_minF:
742 case vmIntrinsics::_maxF:
743 case vmIntrinsics::_minD:
744 case vmIntrinsics::_maxD:
745 case vmIntrinsics::_minF_strict:
746 case vmIntrinsics::_maxF_strict:
747 case vmIntrinsics::_minD_strict:
748 case vmIntrinsics::_maxD_strict:
749 return inline_min_max(intrinsic_id());
750
751 case vmIntrinsics::_VectorUnaryOp:
752 return inline_vector_nary_operation(1);
753 case vmIntrinsics::_VectorBinaryOp:
754 return inline_vector_nary_operation(2);
755 case vmIntrinsics::_VectorUnaryLibOp:
756 return inline_vector_call(1);
757 case vmIntrinsics::_VectorBinaryLibOp:
758 return inline_vector_call(2);
759 case vmIntrinsics::_VectorTernaryOp:
760 return inline_vector_nary_operation(3);
761 case vmIntrinsics::_VectorFromBitsCoerced:
762 return inline_vector_frombits_coerced();
763 case vmIntrinsics::_VectorMaskOp:
764 return inline_vector_mask_operation();
765 case vmIntrinsics::_VectorLoadOp:
766 return inline_vector_mem_operation(/*is_store=*/false);
767 case vmIntrinsics::_VectorLoadMaskedOp:
768 return inline_vector_mem_masked_operation(/*is_store*/false);
769 case vmIntrinsics::_VectorStoreOp:
770 return inline_vector_mem_operation(/*is_store=*/true);
771 case vmIntrinsics::_VectorStoreMaskedOp:
772 return inline_vector_mem_masked_operation(/*is_store=*/true);
773 case vmIntrinsics::_VectorGatherOp:
774 return inline_vector_gather_scatter(/*is_scatter*/ false);
775 case vmIntrinsics::_VectorScatterOp:
776 return inline_vector_gather_scatter(/*is_scatter*/ true);
777 case vmIntrinsics::_VectorReductionCoerced:
778 return inline_vector_reduction();
779 case vmIntrinsics::_VectorTest:
780 return inline_vector_test();
781 case vmIntrinsics::_VectorBlend:
782 return inline_vector_blend();
783 case vmIntrinsics::_VectorRearrange:
784 return inline_vector_rearrange();
785 case vmIntrinsics::_VectorSelectFrom:
786 return inline_vector_select_from();
787 case vmIntrinsics::_VectorCompare:
788 return inline_vector_compare();
789 case vmIntrinsics::_VectorBroadcastInt:
790 return inline_vector_broadcast_int();
791 case vmIntrinsics::_VectorConvert:
792 return inline_vector_convert();
793 case vmIntrinsics::_VectorInsert:
794 return inline_vector_insert();
795 case vmIntrinsics::_VectorExtract:
796 return inline_vector_extract();
797 case vmIntrinsics::_VectorCompressExpand:
798 return inline_vector_compress_expand();
799 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
800 return inline_vector_select_from_two_vectors();
801 case vmIntrinsics::_IndexVector:
802 return inline_index_vector();
803 case vmIntrinsics::_IndexPartiallyInUpperRange:
804 return inline_index_partially_in_upper_range();
805
806 case vmIntrinsics::_getObjectSize:
807 return inline_getObjectSize();
808
809 case vmIntrinsics::_blackhole:
810 return inline_blackhole();
811
812 default:
813 // If you get here, it may be that someone has added a new intrinsic
814 // to the list in vmIntrinsics.hpp without implementing it here.
815 #ifndef PRODUCT
816 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
817 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
818 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
819 }
820 #endif
821 return false;
822 }
823 }
824
825 Node* LibraryCallKit::try_to_predicate(int predicate) {
826 if (!jvms()->has_method()) {
827 // Root JVMState has a null method.
828 assert(map()->memory()->Opcode() == Op_Parm, "");
829 // Insert the memory aliasing node
830 set_all_memory(reset_memory());
831 }
832 assert(merged_memory(), "");
833
834 switch (intrinsic_id()) {
835 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
836 return inline_cipherBlockChaining_AESCrypt_predicate(false);
837 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
838 return inline_cipherBlockChaining_AESCrypt_predicate(true);
839 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
840 return inline_electronicCodeBook_AESCrypt_predicate(false);
841 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
842 return inline_electronicCodeBook_AESCrypt_predicate(true);
843 case vmIntrinsics::_counterMode_AESCrypt:
844 return inline_counterMode_AESCrypt_predicate();
845 case vmIntrinsics::_digestBase_implCompressMB:
846 return inline_digestBase_implCompressMB_predicate(predicate);
847 case vmIntrinsics::_galoisCounterMode_AESCrypt:
848 return inline_galoisCounterMode_AESCrypt_predicate();
849
850 default:
851 // If you get here, it may be that someone has added a new intrinsic
852 // to the list in vmIntrinsics.hpp without implementing it here.
853 #ifndef PRODUCT
854 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
855 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
856 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
857 }
858 #endif
859 Node* slow_ctl = control();
860 set_control(top()); // No fast path intrinsic
861 return slow_ctl;
862 }
863 }
864
865 //------------------------------set_result-------------------------------
866 // Helper function for finishing intrinsics.
867 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
868 record_for_igvn(region);
869 set_control(_gvn.transform(region));
870 set_result( _gvn.transform(value));
871 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
872 }
873
874 RegionNode* LibraryCallKit::create_bailout() {
875 RegionNode* bailout = new RegionNode(1);
876 record_for_igvn(bailout);
877 return bailout;
878 }
879
880 bool LibraryCallKit::check_bailout(RegionNode* bailout) {
881 if (bailout->req() > 1) {
882 bailout = _gvn.transform(bailout)->as_Region();
883 Node* frame = _gvn.transform(new ParmNode(C->start(), TypeFunc::FramePtr));
884 Node* halt = _gvn.transform(new HaltNode(bailout, frame, "unexpected guard failure in intrinsic"));
885 C->root()->add_req(halt);
886 }
887 return stopped();
888 }
889
890 //------------------------------generate_guard---------------------------
891 // Helper function for generating guarded fast-slow graph structures.
892 // The given 'test', if true, guards a slow path. If the test fails
893 // then a fast path can be taken. (We generally hope it fails.)
894 // In all cases, GraphKit::control() is updated to the fast path.
895 // The returned value represents the control for the slow path.
896 // The return value is never 'top'; it is either a valid control
897 // or null if it is obvious that the slow path can never be taken.
898 // Also, if region and the slow control are not null, the slow edge
899 // is appended to the region.
900 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
901 if (stopped()) {
902 // Already short circuited.
903 return nullptr;
904 }
905
906 // Build an if node and its projections.
907 // If test is true we take the slow path, which we assume is uncommon.
908 if (_gvn.type(test) == TypeInt::ZERO) {
909 // The slow branch is never taken. No need to build this guard.
910 return nullptr;
911 }
912
913 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
914
915 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
916 if (if_slow == top()) {
917 // The slow branch is never taken. No need to build this guard.
918 return nullptr;
919 }
920
921 if (region != nullptr)
922 region->add_req(if_slow);
923
924 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
925 set_control(if_fast);
926
927 return if_slow;
928 }
929
930 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
931 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
932 }
933 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
934 return generate_guard(test, region, PROB_FAIR);
935 }
936
937 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
938 Node** pos_index, bool with_opaque) {
939 if (stopped())
940 return nullptr; // already stopped
941 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
942 return nullptr; // index is already adequately typed
943 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
944 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
945 if (with_opaque) {
946 bol_lt = _gvn.transform(new OpaqueConstantBoolNode(C, bol_lt, false));
947 }
948 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
949 if (is_neg != nullptr && pos_index != nullptr) {
950 // Emulate effect of Parse::adjust_map_after_if.
951 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
952 (*pos_index) = _gvn.transform(ccast);
953 }
954 return is_neg;
955 }
956
957 // Make sure that 'position' is a valid limit index, in [0..length].
958 // There are two equivalent plans for checking this:
959 // A. (offset + copyLength) unsigned<= arrayLength
960 // B. offset <= (arrayLength - copyLength)
961 // We require that all of the values above, except for the sum and
962 // difference, are already known to be non-negative.
963 // Plan A is robust in the face of overflow, if offset and copyLength
964 // are both hugely positive.
965 //
966 // Plan B is less direct and intuitive, but it does not overflow at
967 // all, since the difference of two non-negatives is always
968 // representable. Whenever Java methods must perform the equivalent
969 // check they generally use Plan B instead of Plan A.
970 // For the moment we use Plan A.
971 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
972 Node* subseq_length,
973 Node* array_length,
974 RegionNode* region,
975 bool with_opaque) {
976 if (stopped())
977 return nullptr; // already stopped
978 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
979 if (zero_offset && subseq_length->eqv_uncast(array_length))
980 return nullptr; // common case of whole-array copy
981 Node* last = subseq_length;
982 if (!zero_offset) // last += offset
983 last = _gvn.transform(new AddINode(last, offset));
984 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
985 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
986 if (with_opaque) {
987 bol_lt = _gvn.transform(new OpaqueConstantBoolNode(C, bol_lt, false));
988 }
989 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
990 return is_over;
991 }
992
993 // Emit range checks for the given String.value byte array
994 void LibraryCallKit::generate_string_range_check(Node* array,
995 Node* offset,
996 Node* count,
997 bool char_count,
998 RegionNode* region) {
999 if (stopped()) {
1000 return; // already stopped
1001 }
1002 if (char_count) {
1003 // Convert char count to byte count
1004 count = _gvn.transform(new LShiftINode(count, intcon(1)));
1005 }
1006 // Offset and count must not be negative
1007 generate_negative_guard(offset, region, nullptr, true);
1008 generate_negative_guard(count, region, nullptr, true);
1009 // Offset + count must not exceed length of array
1010 generate_limit_guard(offset, count, load_array_length(array), region, true);
1011 }
1012
1013 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
1014 bool is_immutable) {
1015 ciKlass* thread_klass = env()->Thread_klass();
1016 const Type* thread_type
1017 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1018
1019 Node* thread = _gvn.transform(new ThreadLocalNode());
1020 Node* p = off_heap_plus_addr(thread, in_bytes(handle_offset));
1021 tls_output = thread;
1022
1023 Node* thread_obj_handle
1024 = (is_immutable
1025 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
1026 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
1027 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
1028 thread_obj_handle = _gvn.transform(thread_obj_handle);
1029
1030 DecoratorSet decorators = IN_NATIVE;
1031 if (is_immutable) {
1032 decorators |= C2_IMMUTABLE_MEMORY;
1033 }
1034 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
1035 }
1036
1037 //--------------------------generate_current_thread--------------------
1038 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1039 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
1040 /*is_immutable*/false);
1041 }
1042
1043 //--------------------------generate_virtual_thread--------------------
1044 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
1045 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
1046 !C->method()->changes_current_thread());
1047 }
1048
1049 //------------------------------make_string_method_node------------------------
1050 // Helper method for String intrinsic functions. This version is called with
1051 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1052 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1053 // containing the lengths of str1 and str2.
1054 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1055 Node* result = nullptr;
1056 switch (opcode) {
1057 case Op_StrIndexOf:
1058 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1059 str1_start, cnt1, str2_start, cnt2, ae);
1060 break;
1061 case Op_StrComp:
1062 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1063 str1_start, cnt1, str2_start, cnt2, ae);
1064 break;
1065 case Op_StrEquals:
1066 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1067 // Use the constant length if there is one because optimized match rule may exist.
1068 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1069 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1070 break;
1071 default:
1072 ShouldNotReachHere();
1073 return nullptr;
1074 }
1075
1076 // All these intrinsics have checks.
1077 C->set_has_split_ifs(true); // Has chance for split-if optimization
1078 clear_upper_avx();
1079
1080 return _gvn.transform(result);
1081 }
1082
1083 //------------------------------inline_string_compareTo------------------------
1084 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1085 Node* arg1 = argument(0);
1086 Node* arg2 = argument(1);
1087
1088 arg1 = must_be_not_null(arg1, true);
1089 arg2 = must_be_not_null(arg2, true);
1090
1091 // Get start addr and length of first argument
1092 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1093 Node* arg1_cnt = load_array_length(arg1);
1094
1095 // Get start addr and length of second argument
1096 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1097 Node* arg2_cnt = load_array_length(arg2);
1098
1099 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1100 set_result(result);
1101 return true;
1102 }
1103
1104 //------------------------------inline_string_equals------------------------
1105 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1106 Node* arg1 = argument(0);
1107 Node* arg2 = argument(1);
1108
1109 // paths (plus control) merge
1110 RegionNode* region = new RegionNode(3);
1111 Node* phi = new PhiNode(region, TypeInt::BOOL);
1112
1113 if (!stopped()) {
1114
1115 arg1 = must_be_not_null(arg1, true);
1116 arg2 = must_be_not_null(arg2, true);
1117
1118 // Get start addr and length of first argument
1119 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1120 Node* arg1_cnt = load_array_length(arg1);
1121
1122 // Get start addr and length of second argument
1123 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1124 Node* arg2_cnt = load_array_length(arg2);
1125
1126 // Check for arg1_cnt != arg2_cnt
1127 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1128 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1129 Node* if_ne = generate_slow_guard(bol, nullptr);
1130 if (if_ne != nullptr) {
1131 phi->init_req(2, intcon(0));
1132 region->init_req(2, if_ne);
1133 }
1134
1135 // Check for count == 0 is done by assembler code for StrEquals.
1136
1137 if (!stopped()) {
1138 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1139 phi->init_req(1, equals);
1140 region->init_req(1, control());
1141 }
1142 }
1143
1144 // post merge
1145 set_control(_gvn.transform(region));
1146 record_for_igvn(region);
1147
1148 set_result(_gvn.transform(phi));
1149 return true;
1150 }
1151
1152 //------------------------------inline_array_equals----------------------------
1153 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1154 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1155 Node* arg1 = argument(0);
1156 Node* arg2 = argument(1);
1157
1158 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1159 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), mtype, arg1, arg2, ae)));
1160 clear_upper_avx();
1161
1162 return true;
1163 }
1164
1165
1166 //------------------------------inline_countPositives------------------------------
1167 // int java.lang.StringCoding#countPositives0(byte[] ba, int off, int len)
1168 bool LibraryCallKit::inline_countPositives() {
1169 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1170 // no receiver since it is static method
1171 Node* ba = argument(0);
1172 Node* offset = argument(1);
1173 Node* len = argument(2);
1174
1175 ba = must_be_not_null(ba, true);
1176 RegionNode* bailout = create_bailout();
1177 generate_string_range_check(ba, offset, len, false, bailout);
1178 if (check_bailout(bailout)) {
1179 return true;
1180 }
1181
1182 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1183 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1184 set_result(_gvn.transform(result));
1185 clear_upper_avx();
1186 return true;
1187 }
1188
1189 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1190 Node* index = argument(0);
1191 Node* length = bt == T_INT ? argument(1) : argument(2);
1192 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1193 return false;
1194 }
1195
1196 // check that length is positive
1197 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1198 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1199
1200 {
1201 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1202 uncommon_trap(Deoptimization::Reason_intrinsic,
1203 Deoptimization::Action_make_not_entrant);
1204 }
1205
1206 if (stopped()) {
1207 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1208 return true;
1209 }
1210
1211 // length is now known positive, add a cast node to make this explicit
1212 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1213 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1214 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1215 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1216 casted_length = _gvn.transform(casted_length);
1217 replace_in_map(length, casted_length);
1218 length = casted_length;
1219
1220 // Use an unsigned comparison for the range check itself
1221 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1222 BoolTest::mask btest = BoolTest::lt;
1223 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1224 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1225 _gvn.set_type(rc, rc->Value(&_gvn));
1226 if (!rc_bool->is_Con()) {
1227 record_for_igvn(rc);
1228 }
1229 set_control(_gvn.transform(new IfTrueNode(rc)));
1230 {
1231 PreserveJVMState pjvms(this);
1232 set_control(_gvn.transform(new IfFalseNode(rc)));
1233 uncommon_trap(Deoptimization::Reason_range_check,
1234 Deoptimization::Action_make_not_entrant);
1235 }
1236
1237 if (stopped()) {
1238 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1239 return true;
1240 }
1241
1242 // index is now known to be >= 0 and < length, cast it
1243 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1244 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1245 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1246 result = _gvn.transform(result);
1247 set_result(result);
1248 replace_in_map(index, result);
1249 return true;
1250 }
1251
1252 //------------------------------inline_string_indexOf------------------------
1253 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1254 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1255 return false;
1256 }
1257 Node* src = argument(0);
1258 Node* tgt = argument(1);
1259
1260 // Make the merge point
1261 RegionNode* result_rgn = new RegionNode(4);
1262 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1263
1264 src = must_be_not_null(src, true);
1265 tgt = must_be_not_null(tgt, true);
1266
1267 // Get start addr and length of source string
1268 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1269 Node* src_count = load_array_length(src);
1270
1271 // Get start addr and length of substring
1272 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1273 Node* tgt_count = load_array_length(tgt);
1274
1275 Node* result = nullptr;
1276 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1277
1278 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1279 // Divide src size by 2 if String is UTF16 encoded
1280 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1281 }
1282 if (ae == StrIntrinsicNode::UU) {
1283 // Divide substring size by 2 if String is UTF16 encoded
1284 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1285 }
1286
1287 if (call_opt_stub) {
1288 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1289 StubRoutines::_string_indexof_array[ae],
1290 "stringIndexOf", TypePtr::BOTTOM, src_start,
1291 src_count, tgt_start, tgt_count);
1292 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1293 } else {
1294 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1295 result_rgn, result_phi, ae);
1296 }
1297 if (result != nullptr) {
1298 result_phi->init_req(3, result);
1299 result_rgn->init_req(3, control());
1300 }
1301 set_control(_gvn.transform(result_rgn));
1302 record_for_igvn(result_rgn);
1303 set_result(_gvn.transform(result_phi));
1304
1305 return true;
1306 }
1307
1308 //-----------------------------inline_string_indexOfI-----------------------
1309 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1310 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1311 return false;
1312 }
1313
1314 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1315 Node* src = argument(0); // byte[]
1316 Node* src_count = argument(1); // char count
1317 Node* tgt = argument(2); // byte[]
1318 Node* tgt_count = argument(3); // char count
1319 Node* from_index = argument(4); // char index
1320
1321 src = must_be_not_null(src, true);
1322 tgt = must_be_not_null(tgt, true);
1323
1324 // Multiply byte array index by 2 if String is UTF16 encoded
1325 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1326 src_count = _gvn.transform(new SubINode(src_count, from_index));
1327 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1328 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1329
1330 // Range checks
1331 RegionNode* bailout = create_bailout();
1332 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL, bailout);
1333 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU, bailout);
1334 if (check_bailout(bailout)) {
1335 return true;
1336 }
1337
1338 RegionNode* region = new RegionNode(5);
1339 Node* phi = new PhiNode(region, TypeInt::INT);
1340 Node* result = nullptr;
1341
1342 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1343
1344 if (call_opt_stub) {
1345 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1346 StubRoutines::_string_indexof_array[ae],
1347 "stringIndexOf", TypePtr::BOTTOM, src_start,
1348 src_count, tgt_start, tgt_count);
1349 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1350 } else {
1351 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1352 region, phi, ae);
1353 }
1354 if (result != nullptr) {
1355 // The result is index relative to from_index if substring was found, -1 otherwise.
1356 // Generate code which will fold into cmove.
1357 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1358 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1359
1360 Node* if_lt = generate_slow_guard(bol, nullptr);
1361 if (if_lt != nullptr) {
1362 // result == -1
1363 phi->init_req(3, result);
1364 region->init_req(3, if_lt);
1365 }
1366 if (!stopped()) {
1367 result = _gvn.transform(new AddINode(result, from_index));
1368 phi->init_req(4, result);
1369 region->init_req(4, control());
1370 }
1371 }
1372
1373 set_control(_gvn.transform(region));
1374 record_for_igvn(region);
1375 set_result(_gvn.transform(phi));
1376 clear_upper_avx();
1377
1378 return true;
1379 }
1380
1381 // Create StrIndexOfNode with fast path checks
1382 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1383 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1384 // Check for substr count > string count
1385 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1386 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1387 Node* if_gt = generate_slow_guard(bol, nullptr);
1388 if (if_gt != nullptr) {
1389 phi->init_req(1, intcon(-1));
1390 region->init_req(1, if_gt);
1391 }
1392 if (!stopped()) {
1393 // Check for substr count == 0
1394 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1395 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1396 Node* if_zero = generate_slow_guard(bol, nullptr);
1397 if (if_zero != nullptr) {
1398 phi->init_req(2, intcon(0));
1399 region->init_req(2, if_zero);
1400 }
1401 }
1402 if (!stopped()) {
1403 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1404 }
1405 return nullptr;
1406 }
1407
1408 //-----------------------------inline_string_indexOfChar-----------------------
1409 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1410 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1411 return false;
1412 }
1413 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1414 return false;
1415 }
1416 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1417 Node* src = argument(0); // byte[]
1418 Node* int_ch = argument(1);
1419 Node* from_index = argument(2);
1420 Node* max = argument(3);
1421
1422 src = must_be_not_null(src, true);
1423
1424 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1425 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1426 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1427
1428 // Range checks
1429 RegionNode* bailout = create_bailout();
1430 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U, bailout);
1431 if (check_bailout(bailout)) {
1432 return true;
1433 }
1434
1435 // Check for int_ch >= 0
1436 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1437 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1438 {
1439 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1440 uncommon_trap(Deoptimization::Reason_intrinsic,
1441 Deoptimization::Action_maybe_recompile);
1442 }
1443 if (stopped()) {
1444 return true;
1445 }
1446
1447 RegionNode* region = new RegionNode(3);
1448 Node* phi = new PhiNode(region, TypeInt::INT);
1449
1450 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1451 C->set_has_split_ifs(true); // Has chance for split-if optimization
1452 _gvn.transform(result);
1453
1454 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1455 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1456
1457 Node* if_lt = generate_slow_guard(bol, nullptr);
1458 if (if_lt != nullptr) {
1459 // result == -1
1460 phi->init_req(2, result);
1461 region->init_req(2, if_lt);
1462 }
1463 if (!stopped()) {
1464 result = _gvn.transform(new AddINode(result, from_index));
1465 phi->init_req(1, result);
1466 region->init_req(1, control());
1467 }
1468 set_control(_gvn.transform(region));
1469 record_for_igvn(region);
1470 set_result(_gvn.transform(phi));
1471 clear_upper_avx();
1472
1473 return true;
1474 }
1475 //---------------------------inline_string_copy---------------------
1476 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1477 // int StringUTF16.compress0(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1478 // int StringUTF16.compress0(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1479 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1480 // void StringLatin1.inflate0(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1481 // void StringLatin1.inflate0(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1482 bool LibraryCallKit::inline_string_copy(bool compress) {
1483 int nargs = 5; // 2 oops, 3 ints
1484 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1485
1486 Node* src = argument(0);
1487 Node* src_offset = argument(1);
1488 Node* dst = argument(2);
1489 Node* dst_offset = argument(3);
1490 Node* length = argument(4);
1491
1492 // Check for allocation before we add nodes that would confuse
1493 // tightly_coupled_allocation()
1494 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1495
1496 // Figure out the size and type of the elements we will be copying.
1497 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1498 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1499 if (src_type == nullptr || dst_type == nullptr) {
1500 return false;
1501 }
1502 BasicType src_elem = src_type->elem()->array_element_basic_type();
1503 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1504 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1505 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1506 "Unsupported array types for inline_string_copy");
1507
1508 src = must_be_not_null(src, true);
1509 dst = must_be_not_null(dst, true);
1510
1511 // Convert char[] offsets to byte[] offsets
1512 bool convert_src = (compress && src_elem == T_BYTE);
1513 bool convert_dst = (!compress && dst_elem == T_BYTE);
1514 if (convert_src) {
1515 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1516 } else if (convert_dst) {
1517 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1518 }
1519
1520 // Range checks
1521 RegionNode* bailout = create_bailout();
1522 generate_string_range_check(src, src_offset, length, convert_src, bailout);
1523 generate_string_range_check(dst, dst_offset, length, convert_dst, bailout);
1524 if (check_bailout(bailout)) {
1525 return true;
1526 }
1527
1528 Node* src_start = array_element_address(src, src_offset, src_elem);
1529 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1530 // 'src_start' points to src array + scaled offset
1531 // 'dst_start' points to dst array + scaled offset
1532 Node* count = nullptr;
1533 if (compress) {
1534 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1535 } else {
1536 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1537 }
1538
1539 if (alloc != nullptr) {
1540 if (alloc->maybe_set_complete(&_gvn)) {
1541 // "You break it, you buy it."
1542 InitializeNode* init = alloc->initialization();
1543 assert(init->is_complete(), "we just did this");
1544 init->set_complete_with_arraycopy();
1545 assert(dst->is_CheckCastPP(), "sanity");
1546 assert(dst->in(0)->in(0) == init, "dest pinned");
1547 }
1548 // Do not let stores that initialize this object be reordered with
1549 // a subsequent store that would make this object accessible by
1550 // other threads.
1551 // Record what AllocateNode this StoreStore protects so that
1552 // escape analysis can go from the MemBarStoreStoreNode to the
1553 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1554 // based on the escape status of the AllocateNode.
1555 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1556 }
1557 if (compress) {
1558 set_result(_gvn.transform(count));
1559 }
1560 clear_upper_avx();
1561
1562 return true;
1563 }
1564
1565 #ifdef _LP64
1566 #define XTOP ,top() /*additional argument*/
1567 #else //_LP64
1568 #define XTOP /*no additional argument*/
1569 #endif //_LP64
1570
1571 //------------------------inline_string_toBytesU--------------------------
1572 // public static byte[] StringUTF16.toBytes0(char[] value, int off, int len)
1573 bool LibraryCallKit::inline_string_toBytesU() {
1574 // Get the arguments.
1575 assert(callee()->signature()->size() == 3, "character array encoder requires 3 arguments");
1576 Node* value = argument(0);
1577 Node* offset = argument(1);
1578 Node* length = argument(2);
1579
1580 Node* newcopy = nullptr;
1581
1582 // Set the original stack and the reexecute bit for the interpreter to reexecute
1583 // the bytecode that invokes StringUTF16.toBytes0() if deoptimization happens.
1584 { PreserveReexecuteState preexecs(this);
1585 jvms()->set_should_reexecute(true);
1586
1587 value = must_be_not_null(value, true);
1588 RegionNode* bailout = create_bailout();
1589 generate_negative_guard(offset, bailout, nullptr, true);
1590 generate_negative_guard(length, bailout, nullptr, true);
1591 generate_limit_guard(offset, length, load_array_length(value), bailout, true);
1592 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1593 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout, true);
1594 if (check_bailout(bailout)) {
1595 return true;
1596 }
1597
1598 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1599 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1600 newcopy = new_array(klass_node, size, 0); // no arguments to push
1601 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1602 guarantee(alloc != nullptr, "created above");
1603
1604 // Calculate starting addresses.
1605 Node* src_start = array_element_address(value, offset, T_CHAR);
1606 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1607
1608 // Check if dst array address is aligned to HeapWordSize
1609 bool aligned = (arrayOopDesc::base_offset_in_bytes(T_BYTE) % HeapWordSize == 0);
1610 // If true, then check if src array address is aligned to HeapWordSize
1611 if (aligned) {
1612 const TypeInt* toffset = gvn().type(offset)->is_int();
1613 aligned = toffset->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) +
1614 toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1615 }
1616
1617 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1618 const char* copyfunc_name = "arraycopy";
1619 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1620 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1621 OptoRuntime::fast_arraycopy_Type(),
1622 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1623 src_start, dst_start, ConvI2X(length) XTOP);
1624 // Do not let reads from the cloned object float above the arraycopy.
1625 if (alloc->maybe_set_complete(&_gvn)) {
1626 // "You break it, you buy it."
1627 InitializeNode* init = alloc->initialization();
1628 assert(init->is_complete(), "we just did this");
1629 init->set_complete_with_arraycopy();
1630 assert(newcopy->is_CheckCastPP(), "sanity");
1631 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1632 }
1633 // Do not let stores that initialize this object be reordered with
1634 // a subsequent store that would make this object accessible by
1635 // other threads.
1636 // Record what AllocateNode this StoreStore protects so that
1637 // escape analysis can go from the MemBarStoreStoreNode to the
1638 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1639 // based on the escape status of the AllocateNode.
1640 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1641 } // original reexecute is set back here
1642
1643 C->set_has_split_ifs(true); // Has chance for split-if optimization
1644 if (!stopped()) {
1645 set_result(newcopy);
1646 }
1647 clear_upper_avx();
1648
1649 return true;
1650 }
1651
1652 //------------------------inline_string_getCharsU--------------------------
1653 // public void StringUTF16.getChars0(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1654 bool LibraryCallKit::inline_string_getCharsU() {
1655 assert(callee()->signature()->size() == 5, "StringUTF16.getChars0() has 5 arguments");
1656 // Get the arguments.
1657 Node* src = argument(0);
1658 Node* src_begin = argument(1);
1659 Node* src_end = argument(2); // exclusive offset (i < src_end)
1660 Node* dst = argument(3);
1661 Node* dst_begin = argument(4);
1662
1663 // Check for allocation before we add nodes that would confuse
1664 // tightly_coupled_allocation()
1665 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1666
1667 // Check if a null path was taken unconditionally.
1668 src = must_be_not_null(src, true);
1669 dst = must_be_not_null(dst, true);
1670 if (stopped()) {
1671 return true;
1672 }
1673
1674 // Get length and convert char[] offset to byte[] offset
1675 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1676 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1677
1678 // Range checks
1679 RegionNode* bailout = create_bailout();
1680 generate_string_range_check(src, src_begin, length, true, bailout);
1681 generate_string_range_check(dst, dst_begin, length, false, bailout);
1682 if (check_bailout(bailout)) {
1683 return true;
1684 }
1685
1686 // Calculate starting addresses.
1687 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1688 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1689
1690 // Check if array addresses are aligned to HeapWordSize
1691 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1692 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1693 bool aligned = tsrc->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_BYTE) + tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1694 tdst->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) + tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1695
1696 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1697 const char* copyfunc_name = "arraycopy";
1698 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1699 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1700 OptoRuntime::fast_arraycopy_Type(),
1701 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1702 src_start, dst_start, ConvI2X(length) XTOP);
1703 // Do not let reads from the cloned object float above the arraycopy.
1704 if (alloc != nullptr) {
1705 if (alloc->maybe_set_complete(&_gvn)) {
1706 // "You break it, you buy it."
1707 InitializeNode* init = alloc->initialization();
1708 assert(init->is_complete(), "we just did this");
1709 init->set_complete_with_arraycopy();
1710 assert(dst->is_CheckCastPP(), "sanity");
1711 assert(dst->in(0)->in(0) == init, "dest pinned");
1712 }
1713 // Do not let stores that initialize this object be reordered with
1714 // a subsequent store that would make this object accessible by
1715 // other threads.
1716 // Record what AllocateNode this StoreStore protects so that
1717 // escape analysis can go from the MemBarStoreStoreNode to the
1718 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1719 // based on the escape status of the AllocateNode.
1720 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1721 } else {
1722 insert_mem_bar(Op_MemBarCPUOrder);
1723 }
1724
1725 C->set_has_split_ifs(true); // Has chance for split-if optimization
1726 return true;
1727 }
1728
1729 //----------------------inline_string_char_access----------------------------
1730 // Store/Load char to/from byte[] array.
1731 // static void StringUTF16.putChar(byte[] val, int index, int c)
1732 // static char StringUTF16.getChar(byte[] val, int index)
1733 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1734 Node* ch;
1735 if (is_store) {
1736 assert(callee()->signature()->size() == 3, "StringUTF16.putChar() has 3 arguments");
1737 ch = argument(2);
1738 } else {
1739 assert(callee()->signature()->size() == 2, "StringUTF16.getChar() has 2 arguments");
1740 ch = nullptr;
1741 }
1742 Node* value = argument(0);
1743 Node* index = argument(1);
1744
1745 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1746 // correctly requires matched array shapes.
1747 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1748 "sanity: byte[] and char[] bases agree");
1749 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1750 "sanity: byte[] and char[] scales agree");
1751
1752 // Bail when getChar over constants is requested: constant folding would
1753 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1754 // Java method would constant fold nicely instead.
1755 if (!is_store && value->is_Con() && index->is_Con()) {
1756 return false;
1757 }
1758
1759 // Save state and restore on bailout
1760 SavedState old_state(this);
1761
1762 value = must_be_not_null(value, true);
1763
1764 Node* adr = array_element_address(value, index, T_CHAR);
1765 if (adr->is_top()) {
1766 return false;
1767 }
1768 old_state.discard();
1769 if (is_store) {
1770 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1771 } else {
1772 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);
1773 set_result(ch);
1774 }
1775 return true;
1776 }
1777
1778
1779 //------------------------------inline_math-----------------------------------
1780 // public static double Math.abs(double)
1781 // public static double Math.sqrt(double)
1782 // public static double Math.log(double)
1783 // public static double Math.log10(double)
1784 // public static double Math.round(double)
1785 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1786 Node* arg = argument(0);
1787 Node* n = nullptr;
1788 switch (id) {
1789 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1790 case vmIntrinsics::_dsqrt:
1791 case vmIntrinsics::_dsqrt_strict:
1792 n = new SqrtDNode(C, control(), arg); break;
1793 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1794 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1795 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1796 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1797 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1798 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1799 default: fatal_unexpected_iid(id); break;
1800 }
1801 set_result(_gvn.transform(n));
1802 return true;
1803 }
1804
1805 //------------------------------inline_math-----------------------------------
1806 // public static float Math.abs(float)
1807 // public static int Math.abs(int)
1808 // public static long Math.abs(long)
1809 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1810 Node* arg = argument(0);
1811 Node* n = nullptr;
1812 switch (id) {
1813 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1814 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1815 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1816 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1817 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1818 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1819 default: fatal_unexpected_iid(id); break;
1820 }
1821 set_result(_gvn.transform(n));
1822 return true;
1823 }
1824
1825 //------------------------------runtime_math-----------------------------
1826 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1827 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1828 "must be (DD)D or (D)D type");
1829
1830 // Inputs
1831 Node* a = argument(0);
1832 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1833
1834 const TypePtr* no_memory_effects = nullptr;
1835 Node* trig = make_runtime_call(RC_LEAF | RC_PURE, call_type, funcAddr, funcName,
1836 no_memory_effects,
1837 a, top(), b, b ? top() : nullptr);
1838 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1839 #ifdef ASSERT
1840 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1841 assert(value_top == top(), "second value must be top");
1842 #endif
1843
1844 set_result(value);
1845 return true;
1846 }
1847
1848 //------------------------------inline_math_pow-----------------------------
1849 bool LibraryCallKit::inline_math_pow() {
1850 Node* base = argument(0);
1851 Node* exp = argument(2);
1852
1853 CallNode* pow = new PowDNode(C, base, exp);
1854 set_predefined_input_for_runtime_call(pow);
1855 pow = _gvn.transform(pow)->as_CallLeafPure();
1856 set_predefined_output_for_runtime_call(pow);
1857 Node* result = _gvn.transform(new ProjNode(pow, TypeFunc::Parms + 0));
1858 record_for_igvn(pow);
1859 set_result(result);
1860 return true;
1861 }
1862
1863 //------------------------------inline_math_native-----------------------------
1864 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1865 switch (id) {
1866 case vmIntrinsics::_dsin:
1867 return StubRoutines::dsin() != nullptr ?
1868 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1869 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1870 case vmIntrinsics::_dcos:
1871 return StubRoutines::dcos() != nullptr ?
1872 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1873 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1874 case vmIntrinsics::_dtan:
1875 return StubRoutines::dtan() != nullptr ?
1876 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1877 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1878 case vmIntrinsics::_dsinh:
1879 return StubRoutines::dsinh() != nullptr ?
1880 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsinh(), "dsinh") : false;
1881 case vmIntrinsics::_dtanh:
1882 return StubRoutines::dtanh() != nullptr ?
1883 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1884 case vmIntrinsics::_dcbrt:
1885 return StubRoutines::dcbrt() != nullptr ?
1886 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcbrt(), "dcbrt") : false;
1887 case vmIntrinsics::_dexp:
1888 return StubRoutines::dexp() != nullptr ?
1889 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1890 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1891 case vmIntrinsics::_dlog:
1892 return StubRoutines::dlog() != nullptr ?
1893 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1894 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1895 case vmIntrinsics::_dlog10:
1896 return StubRoutines::dlog10() != nullptr ?
1897 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1898 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1899
1900 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1901 case vmIntrinsics::_ceil:
1902 case vmIntrinsics::_floor:
1903 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1904
1905 case vmIntrinsics::_dsqrt:
1906 case vmIntrinsics::_dsqrt_strict:
1907 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1908 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1909 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1910 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1911 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1912
1913 case vmIntrinsics::_dpow: return inline_math_pow();
1914 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1915 case vmIntrinsics::_fcopySign: return inline_math(id);
1916 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1917 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1918 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1919
1920 // These intrinsics are not yet correctly implemented
1921 case vmIntrinsics::_datan2:
1922 return false;
1923
1924 default:
1925 fatal_unexpected_iid(id);
1926 return false;
1927 }
1928 }
1929
1930 //----------------------------inline_notify-----------------------------------*
1931 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1932 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1933 address func;
1934 if (id == vmIntrinsics::_notify) {
1935 func = OptoRuntime::monitor_notify_Java();
1936 } else {
1937 func = OptoRuntime::monitor_notifyAll_Java();
1938 }
1939 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1940 make_slow_call_ex(call, env()->Throwable_klass(), false);
1941 return true;
1942 }
1943
1944
1945 //----------------------------inline_min_max-----------------------------------
1946 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1947 Node* a = nullptr;
1948 Node* b = nullptr;
1949 Node* n = nullptr;
1950 switch (id) {
1951 case vmIntrinsics::_min:
1952 case vmIntrinsics::_max:
1953 case vmIntrinsics::_minF:
1954 case vmIntrinsics::_maxF:
1955 case vmIntrinsics::_minF_strict:
1956 case vmIntrinsics::_maxF_strict:
1957 case vmIntrinsics::_min_strict:
1958 case vmIntrinsics::_max_strict:
1959 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
1960 a = argument(0);
1961 b = argument(1);
1962 break;
1963 case vmIntrinsics::_minD:
1964 case vmIntrinsics::_maxD:
1965 case vmIntrinsics::_minD_strict:
1966 case vmIntrinsics::_maxD_strict:
1967 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
1968 a = argument(0);
1969 b = argument(2);
1970 break;
1971 case vmIntrinsics::_minL:
1972 case vmIntrinsics::_maxL:
1973 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
1974 a = argument(0);
1975 b = argument(2);
1976 break;
1977 default:
1978 fatal_unexpected_iid(id);
1979 break;
1980 }
1981
1982 switch (id) {
1983 case vmIntrinsics::_min:
1984 case vmIntrinsics::_min_strict:
1985 n = new MinINode(a, b);
1986 break;
1987 case vmIntrinsics::_max:
1988 case vmIntrinsics::_max_strict:
1989 n = new MaxINode(a, b);
1990 break;
1991 case vmIntrinsics::_minF:
1992 case vmIntrinsics::_minF_strict:
1993 n = new MinFNode(a, b);
1994 break;
1995 case vmIntrinsics::_maxF:
1996 case vmIntrinsics::_maxF_strict:
1997 n = new MaxFNode(a, b);
1998 break;
1999 case vmIntrinsics::_minD:
2000 case vmIntrinsics::_minD_strict:
2001 n = new MinDNode(a, b);
2002 break;
2003 case vmIntrinsics::_maxD:
2004 case vmIntrinsics::_maxD_strict:
2005 n = new MaxDNode(a, b);
2006 break;
2007 case vmIntrinsics::_minL:
2008 n = new MinLNode(_gvn.C, a, b);
2009 break;
2010 case vmIntrinsics::_maxL:
2011 n = new MaxLNode(_gvn.C, a, b);
2012 break;
2013 default:
2014 fatal_unexpected_iid(id);
2015 break;
2016 }
2017
2018 set_result(_gvn.transform(n));
2019 return true;
2020 }
2021
2022 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
2023 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
2024 env()->ArithmeticException_instance())) {
2025 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
2026 // so let's bail out intrinsic rather than risking deopting again.
2027 return false;
2028 }
2029
2030 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2031 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2032 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2033 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2034
2035 {
2036 PreserveJVMState pjvms(this);
2037 PreserveReexecuteState preexecs(this);
2038 jvms()->set_should_reexecute(true);
2039
2040 set_control(slow_path);
2041 set_i_o(i_o());
2042
2043 builtin_throw(Deoptimization::Reason_intrinsic,
2044 env()->ArithmeticException_instance(),
2045 /*allow_too_many_traps*/ false);
2046 }
2047
2048 set_control(fast_path);
2049 set_result(math);
2050 return true;
2051 }
2052
2053 template <typename OverflowOp>
2054 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2055 typedef typename OverflowOp::MathOp MathOp;
2056
2057 MathOp* mathOp = new MathOp(arg1, arg2);
2058 Node* operation = _gvn.transform( mathOp );
2059 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2060 return inline_math_mathExact(operation, ofcheck);
2061 }
2062
2063 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2064 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2065 }
2066
2067 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2068 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2069 }
2070
2071 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2072 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2073 }
2074
2075 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2076 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2077 }
2078
2079 bool LibraryCallKit::inline_math_negateExactI() {
2080 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2081 }
2082
2083 bool LibraryCallKit::inline_math_negateExactL() {
2084 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2085 }
2086
2087 bool LibraryCallKit::inline_math_multiplyExactI() {
2088 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2089 }
2090
2091 bool LibraryCallKit::inline_math_multiplyExactL() {
2092 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2093 }
2094
2095 bool LibraryCallKit::inline_math_multiplyHigh() {
2096 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2097 return true;
2098 }
2099
2100 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2101 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2102 return true;
2103 }
2104
2105 inline int
2106 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2107 const TypePtr* base_type = TypePtr::NULL_PTR;
2108 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2109 if (base_type == nullptr) {
2110 // Unknown type.
2111 return Type::AnyPtr;
2112 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2113 // Since this is a null+long form, we have to switch to a rawptr.
2114 base = _gvn.transform(new CastX2PNode(offset));
2115 offset = MakeConX(0);
2116 return Type::RawPtr;
2117 } else if (base_type->base() == Type::RawPtr) {
2118 return Type::RawPtr;
2119 } else if (base_type->isa_oopptr()) {
2120 // Base is never null => always a heap address.
2121 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2122 return Type::OopPtr;
2123 }
2124 // Offset is small => always a heap address.
2125 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2126 if (offset_type != nullptr &&
2127 base_type->offset() == 0 && // (should always be?)
2128 offset_type->_lo >= 0 &&
2129 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2130 return Type::OopPtr;
2131 } else if (type == T_OBJECT) {
2132 // off heap access to an oop doesn't make any sense. Has to be on
2133 // heap.
2134 return Type::OopPtr;
2135 }
2136 // Otherwise, it might either be oop+off or null+addr.
2137 return Type::AnyPtr;
2138 } else {
2139 // No information:
2140 return Type::AnyPtr;
2141 }
2142 }
2143
2144 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2145 Node* uncasted_base = base;
2146 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2147 if (kind == Type::RawPtr) {
2148 return off_heap_plus_addr(uncasted_base, offset);
2149 } else if (kind == Type::AnyPtr) {
2150 assert(base == uncasted_base, "unexpected base change");
2151 if (can_cast) {
2152 if (!_gvn.type(base)->speculative_maybe_null() &&
2153 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2154 // According to profiling, this access is always on
2155 // heap. Casting the base to not null and thus avoiding membars
2156 // around the access should allow better optimizations
2157 Node* null_ctl = top();
2158 base = null_check_oop(base, &null_ctl, true, true, true);
2159 assert(null_ctl->is_top(), "no null control here");
2160 return basic_plus_adr(base, offset);
2161 } else if (_gvn.type(base)->speculative_always_null() &&
2162 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2163 // According to profiling, this access is always off
2164 // heap.
2165 base = null_assert(base);
2166 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2167 offset = MakeConX(0);
2168 return off_heap_plus_addr(raw_base, offset);
2169 }
2170 }
2171 // We don't know if it's an on heap or off heap access. Fall back
2172 // to raw memory access.
2173 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2174 return off_heap_plus_addr(raw, offset);
2175 } else {
2176 assert(base == uncasted_base, "unexpected base change");
2177 // We know it's an on heap access so base can't be null
2178 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2179 base = must_be_not_null(base, true);
2180 }
2181 return basic_plus_adr(base, offset);
2182 }
2183 }
2184
2185 //--------------------------inline_number_methods-----------------------------
2186 // inline int Integer.numberOfLeadingZeros(int)
2187 // inline int Long.numberOfLeadingZeros(long)
2188 //
2189 // inline int Integer.numberOfTrailingZeros(int)
2190 // inline int Long.numberOfTrailingZeros(long)
2191 //
2192 // inline int Integer.bitCount(int)
2193 // inline int Long.bitCount(long)
2194 //
2195 // inline char Character.reverseBytes(char)
2196 // inline short Short.reverseBytes(short)
2197 // inline int Integer.reverseBytes(int)
2198 // inline long Long.reverseBytes(long)
2199 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2200 Node* arg = argument(0);
2201 Node* n = nullptr;
2202 switch (id) {
2203 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2204 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2205 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2206 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2207 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2208 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2209 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2210 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2211 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2212 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2213 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2214 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2215 default: fatal_unexpected_iid(id); break;
2216 }
2217 set_result(_gvn.transform(n));
2218 return true;
2219 }
2220
2221 //--------------------------inline_bitshuffle_methods-----------------------------
2222 // inline int Integer.compress(int, int)
2223 // inline int Integer.expand(int, int)
2224 // inline long Long.compress(long, long)
2225 // inline long Long.expand(long, long)
2226 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2227 Node* n = nullptr;
2228 switch (id) {
2229 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2230 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2231 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2232 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2233 default: fatal_unexpected_iid(id); break;
2234 }
2235 set_result(_gvn.transform(n));
2236 return true;
2237 }
2238
2239 //--------------------------inline_number_methods-----------------------------
2240 // inline int Integer.compareUnsigned(int, int)
2241 // inline int Long.compareUnsigned(long, long)
2242 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2243 Node* arg1 = argument(0);
2244 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2245 Node* n = nullptr;
2246 switch (id) {
2247 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2248 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2249 default: fatal_unexpected_iid(id); break;
2250 }
2251 set_result(_gvn.transform(n));
2252 return true;
2253 }
2254
2255 //--------------------------inline_unsigned_divmod_methods-----------------------------
2256 // inline int Integer.divideUnsigned(int, int)
2257 // inline int Integer.remainderUnsigned(int, int)
2258 // inline long Long.divideUnsigned(long, long)
2259 // inline long Long.remainderUnsigned(long, long)
2260 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2261 Node* n = nullptr;
2262 switch (id) {
2263 case vmIntrinsics::_divideUnsigned_i: {
2264 zero_check_int(argument(1));
2265 // Compile-time detect of null-exception
2266 if (stopped()) {
2267 return true; // keep the graph constructed so far
2268 }
2269 n = new UDivINode(control(), argument(0), argument(1));
2270 break;
2271 }
2272 case vmIntrinsics::_divideUnsigned_l: {
2273 zero_check_long(argument(2));
2274 // Compile-time detect of null-exception
2275 if (stopped()) {
2276 return true; // keep the graph constructed so far
2277 }
2278 n = new UDivLNode(control(), argument(0), argument(2));
2279 break;
2280 }
2281 case vmIntrinsics::_remainderUnsigned_i: {
2282 zero_check_int(argument(1));
2283 // Compile-time detect of null-exception
2284 if (stopped()) {
2285 return true; // keep the graph constructed so far
2286 }
2287 n = new UModINode(control(), argument(0), argument(1));
2288 break;
2289 }
2290 case vmIntrinsics::_remainderUnsigned_l: {
2291 zero_check_long(argument(2));
2292 // Compile-time detect of null-exception
2293 if (stopped()) {
2294 return true; // keep the graph constructed so far
2295 }
2296 n = new UModLNode(control(), argument(0), argument(2));
2297 break;
2298 }
2299 default: fatal_unexpected_iid(id); break;
2300 }
2301 set_result(_gvn.transform(n));
2302 return true;
2303 }
2304
2305 //----------------------------inline_unsafe_access----------------------------
2306
2307 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2308 // Attempt to infer a sharper value type from the offset and base type.
2309 ciKlass* sharpened_klass = nullptr;
2310 bool null_free = false;
2311
2312 // See if it is an instance field, with an object type.
2313 if (alias_type->field() != nullptr) {
2314 if (alias_type->field()->type()->is_klass()) {
2315 sharpened_klass = alias_type->field()->type()->as_klass();
2316 null_free = alias_type->field()->is_null_free();
2317 }
2318 }
2319
2320 const TypeOopPtr* result = nullptr;
2321 // See if it is a narrow oop array.
2322 if (adr_type->isa_aryptr()) {
2323 if (adr_type->offset() >= refArrayOopDesc::base_offset_in_bytes()) {
2324 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2325 null_free = adr_type->is_aryptr()->is_null_free();
2326 if (elem_type != nullptr && elem_type->is_loaded()) {
2327 // Sharpen the value type.
2328 result = elem_type;
2329 }
2330 }
2331 }
2332
2333 // The sharpened class might be unloaded if there is no class loader
2334 // contraint in place.
2335 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2336 // Sharpen the value type.
2337 result = TypeOopPtr::make_from_klass(sharpened_klass);
2338 if (null_free) {
2339 result = result->join_speculative(TypePtr::NOTNULL)->is_oopptr();
2340 }
2341 }
2342 if (result != nullptr) {
2343 #ifndef PRODUCT
2344 if (C->print_intrinsics() || C->print_inlining()) {
2345 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2346 tty->print(" sharpened value: "); result->dump(); tty->cr();
2347 }
2348 #endif
2349 }
2350 return result;
2351 }
2352
2353 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2354 switch (kind) {
2355 case Relaxed:
2356 return MO_UNORDERED;
2357 case Opaque:
2358 return MO_RELAXED;
2359 case Acquire:
2360 return MO_ACQUIRE;
2361 case Release:
2362 return MO_RELEASE;
2363 case Volatile:
2364 return MO_SEQ_CST;
2365 default:
2366 ShouldNotReachHere();
2367 return 0;
2368 }
2369 }
2370
2371 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2372 if (callee()->is_static()) return false; // caller must have the capability!
2373 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2374 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2375 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2376 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2377
2378 if (is_reference_type(type)) {
2379 decorators |= ON_UNKNOWN_OOP_REF;
2380 }
2381
2382 if (unaligned) {
2383 decorators |= C2_UNALIGNED;
2384 }
2385
2386 #ifndef PRODUCT
2387 {
2388 ResourceMark rm;
2389 // Check the signatures.
2390 ciSignature* sig = callee()->signature();
2391 #ifdef ASSERT
2392 if (!is_store) {
2393 // Object getReference(Object base, int/long offset), etc.
2394 BasicType rtype = sig->return_type()->basic_type();
2395 assert(rtype == type, "getter must return the expected value");
2396 assert(sig->count() == 2, "oop getter has 2 arguments");
2397 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2398 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2399 } else {
2400 // void putReference(Object base, int/long offset, Object x), etc.
2401 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2402 assert(sig->count() == 3, "oop putter has 3 arguments");
2403 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2404 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2405 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2406 assert(vtype == type, "putter must accept the expected value");
2407 }
2408 #endif // ASSERT
2409 }
2410 #endif //PRODUCT
2411
2412 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2413
2414 Node* receiver = argument(0); // type: oop
2415
2416 // Build address expression.
2417 Node* heap_base_oop = top();
2418
2419 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2420 Node* base = argument(1); // type: oop
2421 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2422 Node* offset = argument(2); // type: long
2423 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2424 // to be plain byte offsets, which are also the same as those accepted
2425 // by oopDesc::field_addr.
2426 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2427 "fieldOffset must be byte-scaled");
2428
2429 if (base->is_InlineType()) {
2430 assert(!is_store, "InlineTypeNodes are non-larval value objects");
2431 InlineTypeNode* vt = base->as_InlineType();
2432 if (offset->is_Con()) {
2433 long off = find_long_con(offset, 0);
2434 ciInlineKlass* vk = vt->type()->inline_klass();
2435 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2436 return false;
2437 }
2438
2439 ciField* field = vk->get_non_flat_field_by_offset(off);
2440 if (field != nullptr) {
2441 BasicType bt = type2field[field->type()->basic_type()];
2442 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2443 bt = T_OBJECT;
2444 }
2445 if (bt == type && !field->is_flat()) {
2446 Node* value = vt->field_value_by_offset(off, false);
2447 const Type* value_type = _gvn.type(value);
2448 if (value_type->is_inlinetypeptr()) {
2449 value = InlineTypeNode::make_from_oop(this, value, value_type->inline_klass());
2450 }
2451 set_result(value);
2452 return true;
2453 }
2454 }
2455 }
2456 {
2457 // Re-execute the unsafe access if allocation triggers deoptimization.
2458 PreserveReexecuteState preexecs(this);
2459 jvms()->set_should_reexecute(true);
2460 vt = vt->buffer(this);
2461 }
2462 base = vt->get_oop();
2463 }
2464
2465 // 32-bit machines ignore the high half!
2466 offset = ConvL2X(offset);
2467
2468 // Save state and restore on bailout
2469 SavedState old_state(this);
2470
2471 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2472 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2473
2474 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2475 if (type != T_OBJECT) {
2476 decorators |= IN_NATIVE; // off-heap primitive access
2477 } else {
2478 return false; // off-heap oop accesses are not supported
2479 }
2480 } else {
2481 heap_base_oop = base; // on-heap or mixed access
2482 }
2483
2484 // Can base be null? Otherwise, always on-heap access.
2485 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2486
2487 if (!can_access_non_heap) {
2488 decorators |= IN_HEAP;
2489 }
2490
2491 Node* val = is_store ? argument(4) : nullptr;
2492
2493 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2494 if (adr_type == TypePtr::NULL_PTR) {
2495 return false; // off-heap access with zero address
2496 }
2497
2498 // Try to categorize the address.
2499 Compile::AliasType* alias_type = C->alias_type(adr_type);
2500 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2501
2502 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2503 alias_type->adr_type() == TypeAryPtr::RANGE) {
2504 return false; // not supported
2505 }
2506
2507 bool mismatched = false;
2508 BasicType bt = T_ILLEGAL;
2509 ciField* field = nullptr;
2510 if (adr_type->isa_instptr()) {
2511 const TypeInstPtr* instptr = adr_type->is_instptr();
2512 ciInstanceKlass* k = instptr->instance_klass();
2513 int off = instptr->offset();
2514 if (instptr->const_oop() != nullptr &&
2515 k == ciEnv::current()->Class_klass() &&
2516 instptr->offset() >= (k->size_helper() * wordSize)) {
2517 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2518 field = k->get_field_by_offset(off, true);
2519 } else {
2520 field = k->get_non_flat_field_by_offset(off);
2521 }
2522 if (field != nullptr) {
2523 bt = type2field[field->type()->basic_type()];
2524 }
2525 if (bt != alias_type->basic_type()) {
2526 // Type mismatch. Is it an access to a nested flat field?
2527 field = k->get_field_by_offset(off, false);
2528 if (field != nullptr) {
2529 bt = type2field[field->type()->basic_type()];
2530 }
2531 }
2532 assert(bt == alias_type->basic_type(), "should match");
2533 } else {
2534 bt = alias_type->basic_type();
2535 }
2536
2537 if (bt != T_ILLEGAL) {
2538 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2539 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2540 // Alias type doesn't differentiate between byte[] and boolean[]).
2541 // Use address type to get the element type.
2542 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2543 }
2544 if (is_reference_type(bt, true)) {
2545 // accessing an array field with getReference is not a mismatch
2546 bt = T_OBJECT;
2547 }
2548 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2549 // Don't intrinsify mismatched object accesses
2550 return false;
2551 }
2552 mismatched = (bt != type);
2553 } else if (alias_type->adr_type()->isa_oopptr()) {
2554 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2555 }
2556
2557 old_state.discard();
2558 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2559
2560 if (mismatched) {
2561 decorators |= C2_MISMATCHED;
2562 }
2563
2564 // First guess at the value type.
2565 const Type *value_type = Type::get_const_basic_type(type);
2566
2567 // Figure out the memory ordering.
2568 decorators |= mo_decorator_for_access_kind(kind);
2569
2570 if (!is_store) {
2571 if (type == T_OBJECT) {
2572 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2573 if (tjp != nullptr) {
2574 value_type = tjp;
2575 }
2576 }
2577 }
2578
2579 receiver = null_check(receiver);
2580 if (stopped()) {
2581 return true;
2582 }
2583 // Heap pointers get a null-check from the interpreter,
2584 // as a courtesy. However, this is not guaranteed by Unsafe,
2585 // and it is not possible to fully distinguish unintended nulls
2586 // from intended ones in this API.
2587
2588 if (!is_store) {
2589 Node* p = nullptr;
2590 // Try to constant fold a load from a constant field
2591
2592 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !field->is_flat() && !mismatched) {
2593 // final or stable field
2594 p = make_constant_from_field(field, heap_base_oop);
2595 }
2596
2597 if (p == nullptr) { // Could not constant fold the load
2598 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2599 const TypeOopPtr* ptr = value_type->make_oopptr();
2600 if (ptr != nullptr && ptr->is_inlinetypeptr()) {
2601 // Load a non-flattened inline type from memory
2602 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass());
2603 }
2604 // Normalize the value returned by getBoolean in the following cases
2605 if (type == T_BOOLEAN &&
2606 (mismatched ||
2607 heap_base_oop == top() || // - heap_base_oop is null or
2608 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2609 // and the unsafe access is made to large offset
2610 // (i.e., larger than the maximum offset necessary for any
2611 // field access)
2612 ) {
2613 IdealKit ideal = IdealKit(this);
2614 #define __ ideal.
2615 IdealVariable normalized_result(ideal);
2616 __ declarations_done();
2617 __ set(normalized_result, p);
2618 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2619 __ set(normalized_result, ideal.ConI(1));
2620 ideal.end_if();
2621 final_sync(ideal);
2622 p = __ value(normalized_result);
2623 #undef __
2624 }
2625 }
2626 if (type == T_ADDRESS) {
2627 p = gvn().transform(new CastP2XNode(nullptr, p));
2628 p = ConvX2UL(p);
2629 }
2630 // The load node has the control of the preceding MemBarCPUOrder. All
2631 // following nodes will have the control of the MemBarCPUOrder inserted at
2632 // the end of this method. So, pushing the load onto the stack at a later
2633 // point is fine.
2634 set_result(p);
2635 } else {
2636 if (bt == T_ADDRESS) {
2637 // Repackage the long as a pointer.
2638 val = ConvL2X(val);
2639 val = gvn().transform(new CastX2PNode(val));
2640 }
2641 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2642 }
2643
2644 return true;
2645 }
2646
2647 bool LibraryCallKit::inline_unsafe_flat_access(bool is_store, AccessKind kind) {
2648 #ifdef ASSERT
2649 {
2650 ResourceMark rm;
2651 // Check the signatures.
2652 ciSignature* sig = callee()->signature();
2653 assert(sig->type_at(0)->basic_type() == T_OBJECT, "base should be object, but is %s", type2name(sig->type_at(0)->basic_type()));
2654 assert(sig->type_at(1)->basic_type() == T_LONG, "offset should be long, but is %s", type2name(sig->type_at(1)->basic_type()));
2655 assert(sig->type_at(2)->basic_type() == T_INT, "layout kind should be int, but is %s", type2name(sig->type_at(3)->basic_type()));
2656 assert(sig->type_at(3)->basic_type() == T_OBJECT, "value klass should be object, but is %s", type2name(sig->type_at(4)->basic_type()));
2657 if (is_store) {
2658 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value, but returns %s", type2name(sig->return_type()->basic_type()));
2659 assert(sig->count() == 5, "flat putter should have 5 arguments, but has %d", sig->count());
2660 assert(sig->type_at(4)->basic_type() == T_OBJECT, "put value should be object, but is %s", type2name(sig->type_at(5)->basic_type()));
2661 } else {
2662 assert(sig->return_type()->basic_type() == T_OBJECT, "getter must return an object, but returns %s", type2name(sig->return_type()->basic_type()));
2663 assert(sig->count() == 4, "flat getter should have 4 arguments, but has %d", sig->count());
2664 }
2665 }
2666 #endif // ASSERT
2667
2668 assert(kind == Relaxed, "Only plain accesses for now");
2669 if (callee()->is_static()) {
2670 // caller must have the capability!
2671 return false;
2672 }
2673 C->set_has_unsafe_access(true);
2674
2675 const TypeInstPtr* value_klass_node = _gvn.type(argument(5))->isa_instptr();
2676 if (value_klass_node == nullptr || value_klass_node->const_oop() == nullptr) {
2677 // parameter valueType is not a constant
2678 return false;
2679 }
2680 ciType* mirror_type = value_klass_node->const_oop()->as_instance()->java_mirror_type();
2681 if (!mirror_type->is_inlinetype()) {
2682 // Dead code
2683 return false;
2684 }
2685 ciInlineKlass* value_klass = mirror_type->as_inline_klass();
2686
2687 const TypeInt* layout_type = _gvn.type(argument(4))->isa_int();
2688 if (layout_type == nullptr || !layout_type->is_con()) {
2689 // parameter layoutKind is not a constant
2690 return false;
2691 }
2692 assert(layout_type->get_con() >= static_cast<int>(LayoutKind::REFERENCE) &&
2693 layout_type->get_con() < static_cast<int>(LayoutKind::UNKNOWN),
2694 "invalid layoutKind %d", layout_type->get_con());
2695 LayoutKind layout = static_cast<LayoutKind>(layout_type->get_con());
2696 assert(layout == LayoutKind::REFERENCE || layout == LayoutKind::NULL_FREE_NON_ATOMIC_FLAT ||
2697 layout == LayoutKind::NULL_FREE_ATOMIC_FLAT || layout == LayoutKind::NULLABLE_ATOMIC_FLAT,
2698 "unexpected layoutKind %d", layout_type->get_con());
2699
2700 null_check(argument(0));
2701 if (stopped()) {
2702 return true;
2703 }
2704
2705 Node* base = must_be_not_null(argument(1), true);
2706 Node* offset = argument(2);
2707 const Type* base_type = _gvn.type(base);
2708
2709 Node* ptr;
2710 bool immutable_memory = false;
2711 DecoratorSet decorators = C2_UNSAFE_ACCESS | IN_HEAP | MO_UNORDERED;
2712 if (base_type->isa_instptr()) {
2713 const TypeLong* offset_type = _gvn.type(offset)->isa_long();
2714 if (offset_type == nullptr || !offset_type->is_con()) {
2715 // Offset into a non-array should be a constant
2716 decorators |= C2_MISMATCHED;
2717 } else {
2718 int offset_con = checked_cast<int>(offset_type->get_con());
2719 ciInstanceKlass* base_klass = base_type->is_instptr()->instance_klass();
2720 ciField* field = base_klass->get_non_flat_field_by_offset(offset_con);
2721 if (field == nullptr) {
2722 assert(!base_klass->is_final(), "non-existence field at offset %d of class %s", offset_con, base_klass->name()->as_utf8());
2723 decorators |= C2_MISMATCHED;
2724 } else {
2725 assert(field->type() == value_klass, "field at offset %d of %s is of type %s, but valueType is %s",
2726 offset_con, base_klass->name()->as_utf8(), field->type()->name(), value_klass->name()->as_utf8());
2727 immutable_memory = field->is_strict() && field->is_final();
2728
2729 if (base->is_InlineType()) {
2730 assert(!is_store, "Cannot store into a non-larval value object");
2731 set_result(base->as_InlineType()->field_value_by_offset(offset_con, false));
2732 return true;
2733 }
2734 }
2735 }
2736
2737 if (base->is_InlineType()) {
2738 assert(!is_store, "Cannot store into a non-larval value object");
2739 base = base->as_InlineType()->buffer(this, true);
2740 }
2741 ptr = basic_plus_adr(base, ConvL2X(offset));
2742 } else if (base_type->isa_aryptr()) {
2743 decorators |= IS_ARRAY;
2744 if (layout == LayoutKind::REFERENCE) {
2745 if (!base_type->is_aryptr()->is_not_flat()) {
2746 const TypeAryPtr* array_type = base_type->is_aryptr()->cast_to_not_flat();
2747 // TODO 8350865 This should be a CheckCastPP, can we add a test?
2748 Node* new_base = _gvn.transform(new CastPPNode(control(), base, array_type, ConstraintCastNode::DependencyType::NonFloatingNarrowing));
2749 replace_in_map(base, new_base);
2750 base = new_base;
2751 }
2752 ptr = basic_plus_adr(base, ConvL2X(offset));
2753 } else {
2754 if (UseArrayFlattening) {
2755 // Flat array must have an exact type
2756 bool is_null_free = !LayoutKindHelper::is_nullable_flat(layout);
2757 bool is_atomic = LayoutKindHelper::is_atomic_flat(layout);
2758 Node* new_base = cast_to_flat_array_exact(base, value_klass, is_null_free, is_atomic);
2759 replace_in_map(base, new_base);
2760 base = new_base;
2761 ptr = basic_plus_adr(base, ConvL2X(offset));
2762 const TypeAryPtr* ptr_type = _gvn.type(ptr)->is_aryptr();
2763 if (ptr_type->field_offset().get() != 0) {
2764 // TODO 8350865 This should be a CheckCastPP, can we add a test?
2765 ptr = _gvn.transform(new CastPPNode(control(), ptr, ptr_type->with_field_offset(0), ConstraintCastNode::DependencyType::NonFloatingNarrowing));
2766 }
2767 } else {
2768 uncommon_trap(Deoptimization::Reason_intrinsic,
2769 Deoptimization::Action_none);
2770 return true;
2771 }
2772 }
2773 } else {
2774 decorators |= C2_MISMATCHED;
2775 ptr = basic_plus_adr(base, ConvL2X(offset));
2776 }
2777
2778 if (is_store) {
2779 Node* value = argument(6);
2780 const Type* value_type = _gvn.type(value);
2781 if (!value_type->is_inlinetypeptr()) {
2782 value_type = Type::get_const_type(value_klass)->filter_speculative(value_type);
2783 Node* new_value = _gvn.transform(new CheckCastPPNode(control(), value, value_type, ConstraintCastNode::DependencyType::NonFloatingNarrowing));
2784 new_value = InlineTypeNode::make_from_oop(this, new_value, value_klass);
2785 replace_in_map(value, new_value);
2786 value = new_value;
2787 }
2788
2789 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());
2790 if (layout == LayoutKind::REFERENCE) {
2791 const TypePtr* ptr_type = (decorators & C2_MISMATCHED) != 0 ? TypeRawPtr::BOTTOM : _gvn.type(ptr)->is_ptr();
2792 access_store_at(base, ptr, ptr_type, value, value_type, T_OBJECT, decorators);
2793 } else {
2794 bool atomic = LayoutKindHelper::is_atomic_flat(layout);
2795 bool null_free = !LayoutKindHelper::is_nullable_flat(layout);
2796 value->as_InlineType()->store_flat(this, base, ptr, atomic, immutable_memory, null_free, decorators);
2797 }
2798
2799 return true;
2800 } else {
2801 decorators |= (C2_CONTROL_DEPENDENT_LOAD | C2_UNKNOWN_CONTROL_LOAD);
2802 InlineTypeNode* result;
2803 if (layout == LayoutKind::REFERENCE) {
2804 const TypePtr* ptr_type = (decorators & C2_MISMATCHED) != 0 ? TypeRawPtr::BOTTOM : _gvn.type(ptr)->is_ptr();
2805 Node* oop = access_load_at(base, ptr, ptr_type, Type::get_const_type(value_klass), T_OBJECT, decorators);
2806 result = InlineTypeNode::make_from_oop(this, oop, value_klass);
2807 } else {
2808 bool atomic = LayoutKindHelper::is_atomic_flat(layout);
2809 bool null_free = !LayoutKindHelper::is_nullable_flat(layout);
2810 result = InlineTypeNode::make_from_flat(this, value_klass, base, ptr, atomic, immutable_memory, null_free, decorators);
2811 }
2812
2813 set_result(result);
2814 return true;
2815 }
2816 }
2817
2818 //----------------------------inline_unsafe_load_store----------------------------
2819 // This method serves a couple of different customers (depending on LoadStoreKind):
2820 //
2821 // LS_cmp_swap:
2822 //
2823 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2824 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2825 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2826 //
2827 // LS_cmp_swap_weak:
2828 //
2829 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2830 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2831 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2832 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2833 //
2834 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2835 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2836 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2837 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2838 //
2839 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2840 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2841 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2842 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2843 //
2844 // LS_cmp_exchange:
2845 //
2846 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2847 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2848 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2849 //
2850 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2851 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2852 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2853 //
2854 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2855 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2856 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2857 //
2858 // LS_get_add:
2859 //
2860 // int getAndAddInt( Object o, long offset, int delta)
2861 // long getAndAddLong(Object o, long offset, long delta)
2862 //
2863 // LS_get_set:
2864 //
2865 // int getAndSet(Object o, long offset, int newValue)
2866 // long getAndSet(Object o, long offset, long newValue)
2867 // Object getAndSet(Object o, long offset, Object newValue)
2868 //
2869 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2870 // This basic scheme here is the same as inline_unsafe_access, but
2871 // differs in enough details that combining them would make the code
2872 // overly confusing. (This is a true fact! I originally combined
2873 // them, but even I was confused by it!) As much code/comments as
2874 // possible are retained from inline_unsafe_access though to make
2875 // the correspondences clearer. - dl
2876
2877 if (callee()->is_static()) return false; // caller must have the capability!
2878
2879 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2880 decorators |= mo_decorator_for_access_kind(access_kind);
2881
2882 #ifndef PRODUCT
2883 BasicType rtype;
2884 {
2885 ResourceMark rm;
2886 // Check the signatures.
2887 ciSignature* sig = callee()->signature();
2888 rtype = sig->return_type()->basic_type();
2889 switch(kind) {
2890 case LS_get_add:
2891 case LS_get_set: {
2892 // Check the signatures.
2893 #ifdef ASSERT
2894 assert(rtype == type, "get and set must return the expected type");
2895 assert(sig->count() == 3, "get and set has 3 arguments");
2896 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2897 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2898 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2899 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2900 #endif // ASSERT
2901 break;
2902 }
2903 case LS_cmp_swap:
2904 case LS_cmp_swap_weak: {
2905 // Check the signatures.
2906 #ifdef ASSERT
2907 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2908 assert(sig->count() == 4, "CAS has 4 arguments");
2909 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2910 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2911 #endif // ASSERT
2912 break;
2913 }
2914 case LS_cmp_exchange: {
2915 // Check the signatures.
2916 #ifdef ASSERT
2917 assert(rtype == type, "CAS must return the expected type");
2918 assert(sig->count() == 4, "CAS has 4 arguments");
2919 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2920 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2921 #endif // ASSERT
2922 break;
2923 }
2924 default:
2925 ShouldNotReachHere();
2926 }
2927 }
2928 #endif //PRODUCT
2929
2930 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2931
2932 // Get arguments:
2933 Node* receiver = nullptr;
2934 Node* base = nullptr;
2935 Node* offset = nullptr;
2936 Node* oldval = nullptr;
2937 Node* newval = nullptr;
2938 switch(kind) {
2939 case LS_cmp_swap:
2940 case LS_cmp_swap_weak:
2941 case LS_cmp_exchange: {
2942 const bool two_slot_type = type2size[type] == 2;
2943 receiver = argument(0); // type: oop
2944 base = argument(1); // type: oop
2945 offset = argument(2); // type: long
2946 oldval = argument(4); // type: oop, int, or long
2947 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2948 break;
2949 }
2950 case LS_get_add:
2951 case LS_get_set: {
2952 receiver = argument(0); // type: oop
2953 base = argument(1); // type: oop
2954 offset = argument(2); // type: long
2955 oldval = nullptr;
2956 newval = argument(4); // type: oop, int, or long
2957 break;
2958 }
2959 default:
2960 ShouldNotReachHere();
2961 }
2962
2963 // Build field offset expression.
2964 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2965 // to be plain byte offsets, which are also the same as those accepted
2966 // by oopDesc::field_addr.
2967 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2968 // 32-bit machines ignore the high half of long offsets
2969 offset = ConvL2X(offset);
2970 // Save state and restore on bailout
2971 SavedState old_state(this);
2972 Node* adr = make_unsafe_address(base, offset,type, false);
2973 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2974
2975 Compile::AliasType* alias_type = C->alias_type(adr_type);
2976 BasicType bt = alias_type->basic_type();
2977 if (bt != T_ILLEGAL &&
2978 (is_reference_type(bt) != (type == T_OBJECT))) {
2979 // Don't intrinsify mismatched object accesses.
2980 return false;
2981 }
2982
2983 old_state.discard();
2984
2985 // For CAS, unlike inline_unsafe_access, there seems no point in
2986 // trying to refine types. Just use the coarse types here.
2987 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2988 const Type *value_type = Type::get_const_basic_type(type);
2989
2990 switch (kind) {
2991 case LS_get_set:
2992 case LS_cmp_exchange: {
2993 if (type == T_OBJECT) {
2994 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2995 if (tjp != nullptr) {
2996 value_type = tjp;
2997 }
2998 }
2999 break;
3000 }
3001 case LS_cmp_swap:
3002 case LS_cmp_swap_weak:
3003 case LS_get_add:
3004 break;
3005 default:
3006 ShouldNotReachHere();
3007 }
3008
3009 // Null check receiver.
3010 receiver = null_check(receiver);
3011 if (stopped()) {
3012 return true;
3013 }
3014
3015 int alias_idx = C->get_alias_index(adr_type);
3016
3017 if (is_reference_type(type)) {
3018 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
3019
3020 if (oldval != nullptr && oldval->is_InlineType()) {
3021 // Re-execute the unsafe access if allocation triggers deoptimization.
3022 PreserveReexecuteState preexecs(this);
3023 jvms()->set_should_reexecute(true);
3024 oldval = oldval->as_InlineType()->buffer(this)->get_oop();
3025 }
3026 if (newval != nullptr && newval->is_InlineType()) {
3027 // Re-execute the unsafe access if allocation triggers deoptimization.
3028 PreserveReexecuteState preexecs(this);
3029 jvms()->set_should_reexecute(true);
3030 newval = newval->as_InlineType()->buffer(this)->get_oop();
3031 }
3032
3033 // Transformation of a value which could be null pointer (CastPP #null)
3034 // could be delayed during Parse (for example, in adjust_map_after_if()).
3035 // Execute transformation here to avoid barrier generation in such case.
3036 if (_gvn.type(newval) == TypePtr::NULL_PTR)
3037 newval = _gvn.makecon(TypePtr::NULL_PTR);
3038
3039 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
3040 // Refine the value to a null constant, when it is known to be null
3041 oldval = _gvn.makecon(TypePtr::NULL_PTR);
3042 }
3043 }
3044
3045 Node* result = nullptr;
3046 switch (kind) {
3047 case LS_cmp_exchange: {
3048 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
3049 oldval, newval, value_type, type, decorators);
3050 break;
3051 }
3052 case LS_cmp_swap_weak:
3053 decorators |= C2_WEAK_CMPXCHG;
3054 case LS_cmp_swap: {
3055 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
3056 oldval, newval, value_type, type, decorators);
3057 break;
3058 }
3059 case LS_get_set: {
3060 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
3061 newval, value_type, type, decorators);
3062 break;
3063 }
3064 case LS_get_add: {
3065 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
3066 newval, value_type, type, decorators);
3067 break;
3068 }
3069 default:
3070 ShouldNotReachHere();
3071 }
3072
3073 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3074 set_result(result);
3075 return true;
3076 }
3077
3078 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3079 // Regardless of form, don't allow previous ld/st to move down,
3080 // then issue acquire, release, or volatile mem_bar.
3081 insert_mem_bar(Op_MemBarCPUOrder);
3082 switch(id) {
3083 case vmIntrinsics::_loadFence:
3084 insert_mem_bar(Op_LoadFence);
3085 return true;
3086 case vmIntrinsics::_storeFence:
3087 insert_mem_bar(Op_StoreFence);
3088 return true;
3089 case vmIntrinsics::_storeStoreFence:
3090 insert_mem_bar(Op_StoreStoreFence);
3091 return true;
3092 case vmIntrinsics::_fullFence:
3093 insert_mem_bar(Op_MemBarFull);
3094 return true;
3095 default:
3096 fatal_unexpected_iid(id);
3097 return false;
3098 }
3099 }
3100
3101 // private native int arrayInstanceBaseOffset0(Object[] array);
3102 bool LibraryCallKit::inline_arrayInstanceBaseOffset() {
3103 Node* array = argument(1);
3104 Node* klass_node = load_object_klass(array);
3105
3106 jint layout_con = Klass::_lh_neutral_value;
3107 Node* layout_val = get_layout_helper(klass_node, layout_con);
3108 int layout_is_con = (layout_val == nullptr);
3109
3110 Node* header_size = nullptr;
3111 if (layout_is_con) {
3112 int hsize = Klass::layout_helper_header_size(layout_con);
3113 header_size = intcon(hsize);
3114 } else {
3115 Node* hss = intcon(Klass::_lh_header_size_shift);
3116 Node* hsm = intcon(Klass::_lh_header_size_mask);
3117 header_size = _gvn.transform(new URShiftINode(layout_val, hss));
3118 header_size = _gvn.transform(new AndINode(header_size, hsm));
3119 }
3120 set_result(header_size);
3121 return true;
3122 }
3123
3124 // private native int arrayInstanceIndexScale0(Object[] array);
3125 bool LibraryCallKit::inline_arrayInstanceIndexScale() {
3126 Node* array = argument(1);
3127 Node* klass_node = load_object_klass(array);
3128
3129 jint layout_con = Klass::_lh_neutral_value;
3130 Node* layout_val = get_layout_helper(klass_node, layout_con);
3131 int layout_is_con = (layout_val == nullptr);
3132
3133 Node* element_size = nullptr;
3134 if (layout_is_con) {
3135 int log_element_size = Klass::layout_helper_log2_element_size(layout_con);
3136 int elem_size = 1 << log_element_size;
3137 element_size = intcon(elem_size);
3138 } else {
3139 Node* ess = intcon(Klass::_lh_log2_element_size_shift);
3140 Node* esm = intcon(Klass::_lh_log2_element_size_mask);
3141 Node* log_element_size = _gvn.transform(new URShiftINode(layout_val, ess));
3142 log_element_size = _gvn.transform(new AndINode(log_element_size, esm));
3143 element_size = _gvn.transform(new LShiftINode(intcon(1), log_element_size));
3144 }
3145 set_result(element_size);
3146 return true;
3147 }
3148
3149 // private native int arrayLayout0(Object[] array);
3150 bool LibraryCallKit::inline_arrayLayout() {
3151 RegionNode* region = new RegionNode(2);
3152 Node* phi = new PhiNode(region, TypeInt::POS);
3153
3154 Node* array = argument(1);
3155 Node* klass_node = load_object_klass(array);
3156 generate_refArray_guard(klass_node, region);
3157 if (region->req() == 3) {
3158 phi->add_req(intcon((jint)LayoutKind::REFERENCE));
3159 }
3160
3161 int layout_kind_offset = in_bytes(FlatArrayKlass::layout_kind_offset());
3162 Node* layout_kind_addr = basic_plus_adr(top(), klass_node, layout_kind_offset);
3163 Node* layout_kind = make_load(nullptr, layout_kind_addr, TypeInt::POS, T_INT, MemNode::unordered);
3164
3165 region->init_req(1, control());
3166 phi->init_req(1, layout_kind);
3167
3168 set_control(_gvn.transform(region));
3169 set_result(_gvn.transform(phi));
3170 return true;
3171 }
3172
3173 // private native int[] getFieldMap0(Class <?> c);
3174 // int offset = c._klass._acmp_maps_offset;
3175 // return (int[])c.obj_field(offset);
3176 bool LibraryCallKit::inline_getFieldMap() {
3177 Node* mirror = argument(1);
3178 Node* klass = load_klass_from_mirror(mirror, false, nullptr, 0);
3179
3180 int field_map_offset_offset = in_bytes(InstanceKlass::acmp_maps_offset_offset());
3181 Node* field_map_offset_addr = basic_plus_adr(top(), klass, field_map_offset_offset);
3182 Node* field_map_offset = make_load(nullptr, field_map_offset_addr, TypeInt::INT, T_INT, MemNode::unordered);
3183 field_map_offset = _gvn.transform(ConvI2L(field_map_offset));
3184
3185 Node* map_addr = basic_plus_adr(mirror, field_map_offset);
3186 const TypeAryPtr* val_type = TypeAryPtr::INTS->cast_to_ptr_type(TypePtr::NotNull)->with_offset(0);
3187 Node* map = access_load_at(mirror, map_addr, TypeAryPtr::INTS, val_type, T_ARRAY, IN_HEAP | MO_UNORDERED);
3188
3189 set_result(map);
3190 return true;
3191 }
3192
3193 bool LibraryCallKit::inline_onspinwait() {
3194 insert_mem_bar(Op_OnSpinWait);
3195 return true;
3196 }
3197
3198 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3199 if (!kls->is_Con()) {
3200 return true;
3201 }
3202 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
3203 if (klsptr == nullptr) {
3204 return true;
3205 }
3206 ciInstanceKlass* ik = klsptr->instance_klass();
3207 // don't need a guard for a klass that is already initialized
3208 return !ik->is_initialized();
3209 }
3210
3211 //----------------------------inline_unsafe_writeback0-------------------------
3212 // public native void Unsafe.writeback0(long address)
3213 bool LibraryCallKit::inline_unsafe_writeback0() {
3214 if (!Matcher::has_match_rule(Op_CacheWB)) {
3215 return false;
3216 }
3217 #ifndef PRODUCT
3218 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
3219 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
3220 ciSignature* sig = callee()->signature();
3221 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
3222 #endif
3223 null_check_receiver(); // null-check, then ignore
3224 Node *addr = argument(1);
3225 addr = new CastX2PNode(addr);
3226 addr = _gvn.transform(addr);
3227 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
3228 flush = _gvn.transform(flush);
3229 set_memory(flush, TypeRawPtr::BOTTOM);
3230 return true;
3231 }
3232
3233 //----------------------------inline_unsafe_writeback0-------------------------
3234 // public native void Unsafe.writeback0(long address)
3235 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
3236 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
3237 return false;
3238 }
3239 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
3240 return false;
3241 }
3242 #ifndef PRODUCT
3243 assert(Matcher::has_match_rule(Op_CacheWB),
3244 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
3245 : "found match rule for CacheWBPostSync but not CacheWB"));
3246
3247 #endif
3248 null_check_receiver(); // null-check, then ignore
3249 Node *sync;
3250 if (is_pre) {
3251 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3252 } else {
3253 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3254 }
3255 sync = _gvn.transform(sync);
3256 set_memory(sync, TypeRawPtr::BOTTOM);
3257 return true;
3258 }
3259
3260 //----------------------------inline_unsafe_allocate---------------------------
3261 // public native Object Unsafe.allocateInstance(Class<?> cls);
3262 bool LibraryCallKit::inline_unsafe_allocate() {
3263
3264 #if INCLUDE_JVMTI
3265 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3266 return false;
3267 }
3268 #endif //INCLUDE_JVMTI
3269
3270 if (callee()->is_static()) return false; // caller must have the capability!
3271
3272 null_check_receiver(); // null-check, then ignore
3273 Node* cls = null_check(argument(1));
3274 if (stopped()) return true;
3275
3276 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
3277 kls = null_check(kls);
3278 if (stopped()) return true; // argument was like int.class
3279
3280 #if INCLUDE_JVMTI
3281 // Don't try to access new allocated obj in the intrinsic.
3282 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
3283 // Deoptimize and allocate in interpreter instead.
3284 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
3285 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
3286 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
3287 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
3288 {
3289 BuildCutout unless(this, tst, PROB_MAX);
3290 uncommon_trap(Deoptimization::Reason_intrinsic,
3291 Deoptimization::Action_make_not_entrant);
3292 }
3293 if (stopped()) {
3294 return true;
3295 }
3296 #endif //INCLUDE_JVMTI
3297
3298 Node* test = nullptr;
3299 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3300 // Note: The argument might still be an illegal value like
3301 // Serializable.class or Object[].class. The runtime will handle it.
3302 // But we must make an explicit check for initialization.
3303 Node* insp = off_heap_plus_addr(kls, in_bytes(InstanceKlass::init_state_offset()));
3304 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3305 // can generate code to load it as unsigned byte.
3306 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
3307 Node* bits = intcon(InstanceKlass::fully_initialized);
3308 test = _gvn.transform(new SubINode(inst, bits));
3309 // The 'test' is non-zero if we need to take a slow path.
3310 }
3311 Node* obj = new_instance(kls, test);
3312 set_result(obj);
3313 return true;
3314 }
3315
3316 //------------------------inline_native_time_funcs--------------
3317 // inline code for System.currentTimeMillis() and System.nanoTime()
3318 // these have the same type and signature
3319 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3320 const TypeFunc* tf = OptoRuntime::void_long_Type();
3321 const TypePtr* no_memory_effects = nullptr;
3322 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3323 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3324 #ifdef ASSERT
3325 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3326 assert(value_top == top(), "second value must be top");
3327 #endif
3328 set_result(value);
3329 return true;
3330 }
3331
3332 //--------------------inline_native_vthread_start_transition--------------------
3333 // inline void startTransition(boolean is_mount);
3334 // inline void startFinalTransition();
3335 // Pseudocode of implementation:
3336 //
3337 // java_lang_Thread::set_is_in_vthread_transition(vthread, true);
3338 // carrier->set_is_in_vthread_transition(true);
3339 // OrderAccess::storeload();
3340 // int disable_requests = java_lang_Thread::vthread_transition_disable_count(vthread)
3341 // + global_vthread_transition_disable_count();
3342 // if (disable_requests > 0) {
3343 // slow path: runtime call
3344 // }
3345 bool LibraryCallKit::inline_native_vthread_start_transition(address funcAddr, const char* funcName, bool is_final_transition) {
3346 Node* vt_oop = must_be_not_null(argument(0), true); // VirtualThread this argument
3347 IdealKit ideal(this);
3348
3349 Node* thread = ideal.thread();
3350 Node* jt_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3351 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3352 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3353 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3354 insert_mem_bar(Op_MemBarStoreLoad);
3355 ideal.sync_kit(this);
3356
3357 Node* global_disable_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::global_vthread_transition_disable_count_address()));
3358 Node* global_disable = ideal.load(ideal.ctrl(), global_disable_addr, TypeInt::INT, T_INT, Compile::AliasIdxRaw, true /*require_atomic_access*/);
3359 Node* vt_disable_addr = basic_plus_adr(vt_oop, java_lang_Thread::vthread_transition_disable_count_offset());
3360 const TypePtr* vt_disable_addr_t = _gvn.type(vt_disable_addr)->is_ptr();
3361 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*/);
3362 Node* disabled = _gvn.transform(new AddINode(global_disable, vt_disable));
3363
3364 ideal.if_then(disabled, BoolTest::ne, ideal.ConI(0)); {
3365 sync_kit(ideal);
3366 Node* is_mount = is_final_transition ? ideal.ConI(0) : argument(1);
3367 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3368 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3369 ideal.sync_kit(this);
3370 }
3371 ideal.end_if();
3372
3373 final_sync(ideal);
3374 return true;
3375 }
3376
3377 bool LibraryCallKit::inline_native_vthread_end_transition(address funcAddr, const char* funcName, bool is_first_transition) {
3378 Node* vt_oop = must_be_not_null(argument(0), true); // VirtualThread this argument
3379 IdealKit ideal(this);
3380
3381 Node* _notify_jvmti_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::notify_jvmti_events_address()));
3382 Node* _notify_jvmti = ideal.load(ideal.ctrl(), _notify_jvmti_addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3383
3384 ideal.if_then(_notify_jvmti, BoolTest::eq, ideal.ConI(1)); {
3385 sync_kit(ideal);
3386 Node* is_mount = is_first_transition ? ideal.ConI(1) : argument(1);
3387 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3388 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3389 ideal.sync_kit(this);
3390 } ideal.else_(); {
3391 Node* thread = ideal.thread();
3392 Node* jt_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3393 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3394
3395 sync_kit(ideal);
3396 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3397 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3398 ideal.sync_kit(this);
3399 } ideal.end_if();
3400
3401 final_sync(ideal);
3402 return true;
3403 }
3404
3405 #if INCLUDE_JVMTI
3406
3407 // Always update the is_disable_suspend bit.
3408 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3409 if (!DoJVMTIVirtualThreadTransitions) {
3410 return true;
3411 }
3412 IdealKit ideal(this);
3413
3414 {
3415 // unconditionally update the is_disable_suspend bit in current JavaThread
3416 Node* thread = ideal.thread();
3417 Node* arg = argument(0); // argument for notification
3418 Node* addr = off_heap_plus_addr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3419 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3420
3421 sync_kit(ideal);
3422 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3423 ideal.sync_kit(this);
3424 }
3425 final_sync(ideal);
3426
3427 return true;
3428 }
3429
3430 #endif // INCLUDE_JVMTI
3431
3432 #ifdef JFR_HAVE_INTRINSICS
3433
3434 /**
3435 * if oop->klass != null
3436 * // normal class
3437 * epoch = _epoch_state ? 2 : 1
3438 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3439 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3440 * }
3441 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3442 * else
3443 * // primitive class
3444 * if oop->array_klass != null
3445 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3446 * else
3447 * id = LAST_TYPE_ID + 1 // void class path
3448 * if (!signaled)
3449 * signaled = true
3450 */
3451 bool LibraryCallKit::inline_native_classID() {
3452 Node* cls = argument(0);
3453
3454 IdealKit ideal(this);
3455 #define __ ideal.
3456 IdealVariable result(ideal); __ declarations_done();
3457 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3458 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3459 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3460
3461
3462 __ if_then(kls, BoolTest::ne, null()); {
3463 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3464 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3465
3466 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3467 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3468 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3469 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3470 mask = _gvn.transform(new OrLNode(mask, epoch));
3471 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3472
3473 float unlikely = PROB_UNLIKELY(0.999);
3474 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3475 sync_kit(ideal);
3476 make_runtime_call(RC_LEAF,
3477 OptoRuntime::class_id_load_barrier_Type(),
3478 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3479 "class id load barrier",
3480 TypePtr::BOTTOM,
3481 kls);
3482 ideal.sync_kit(this);
3483 } __ end_if();
3484
3485 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3486 } __ else_(); {
3487 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3488 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3489 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3490 __ if_then(array_kls, BoolTest::ne, null()); {
3491 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3492 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3493 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3494 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3495 } __ else_(); {
3496 // void class case
3497 ideal.set(result, longcon(LAST_TYPE_ID + 1));
3498 } __ end_if();
3499
3500 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3501 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3502 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3503 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3504 } __ end_if();
3505 } __ end_if();
3506
3507 final_sync(ideal);
3508 set_result(ideal.value(result));
3509 #undef __
3510 return true;
3511 }
3512
3513 //------------------------inline_native_jvm_commit------------------
3514 bool LibraryCallKit::inline_native_jvm_commit() {
3515 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3516
3517 // Save input memory and i_o state.
3518 Node* input_memory_state = reset_memory();
3519 set_all_memory(input_memory_state);
3520 Node* input_io_state = i_o();
3521
3522 // TLS.
3523 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3524 // Jfr java buffer.
3525 Node* java_buffer_offset = _gvn.transform(AddPNode::make_off_heap(tls_ptr, MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR))));
3526 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3527 Node* java_buffer_pos_offset = _gvn.transform(AddPNode::make_off_heap(java_buffer, MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET))));
3528
3529 // Load the current value of the notified field in the JfrThreadLocal.
3530 Node* notified_offset = off_heap_plus_addr(tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3531 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3532
3533 // Test for notification.
3534 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3535 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3536 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3537
3538 // True branch, is notified.
3539 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3540 set_control(is_notified);
3541
3542 // Reset notified state.
3543 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3544 Node* notified_reset_memory = reset_memory();
3545
3546 // 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.
3547 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3548 // Convert the machine-word to a long.
3549 Node* current_pos = ConvX2L(current_pos_X);
3550
3551 // False branch, not notified.
3552 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3553 set_control(not_notified);
3554 set_all_memory(input_memory_state);
3555
3556 // Arg is the next position as a long.
3557 Node* arg = argument(0);
3558 // Convert long to machine-word.
3559 Node* next_pos_X = ConvL2X(arg);
3560
3561 // Store the next_position to the underlying jfr java buffer.
3562 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3563
3564 Node* commit_memory = reset_memory();
3565 set_all_memory(commit_memory);
3566
3567 // 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.
3568 Node* java_buffer_flags_offset = _gvn.transform(AddPNode::make_off_heap(java_buffer, MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET))));
3569 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3570 Node* lease_constant = _gvn.intcon(4);
3571
3572 // And flags with lease constant.
3573 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3574
3575 // Branch on lease to conditionalize returning the leased java buffer.
3576 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3577 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3578 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3579
3580 // False branch, not a lease.
3581 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3582
3583 // True branch, is lease.
3584 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3585 set_control(is_lease);
3586
3587 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3588 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3589 OptoRuntime::void_void_Type(),
3590 SharedRuntime::jfr_return_lease(),
3591 "return_lease", TypePtr::BOTTOM);
3592 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3593
3594 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3595 record_for_igvn(lease_compare_rgn);
3596 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3597 record_for_igvn(lease_compare_mem);
3598 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3599 record_for_igvn(lease_compare_io);
3600 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3601 record_for_igvn(lease_result_value);
3602
3603 // Update control and phi nodes.
3604 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3605 lease_compare_rgn->init_req(_false_path, not_lease);
3606
3607 lease_compare_mem->init_req(_true_path, reset_memory());
3608 lease_compare_mem->init_req(_false_path, commit_memory);
3609
3610 lease_compare_io->init_req(_true_path, i_o());
3611 lease_compare_io->init_req(_false_path, input_io_state);
3612
3613 lease_result_value->init_req(_true_path, _gvn.longcon(0)); // if the lease was returned, return 0L.
3614 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3615
3616 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3617 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3618 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3619 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3620
3621 // Update control and phi nodes.
3622 result_rgn->init_req(_true_path, is_notified);
3623 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3624
3625 result_mem->init_req(_true_path, notified_reset_memory);
3626 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3627
3628 result_io->init_req(_true_path, input_io_state);
3629 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3630
3631 result_value->init_req(_true_path, current_pos);
3632 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3633
3634 // Set output state.
3635 set_control(_gvn.transform(result_rgn));
3636 set_all_memory(_gvn.transform(result_mem));
3637 set_i_o(_gvn.transform(result_io));
3638 set_result(result_rgn, result_value);
3639 return true;
3640 }
3641
3642 /*
3643 * The intrinsic is a model of this pseudo-code:
3644 *
3645 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3646 * jobject h_event_writer = tl->java_event_writer();
3647 * if (h_event_writer == nullptr) {
3648 * return nullptr;
3649 * }
3650 * oop threadObj = Thread::threadObj();
3651 * oop vthread = java_lang_Thread::vthread(threadObj);
3652 * traceid tid;
3653 * bool pinVirtualThread;
3654 * bool excluded;
3655 * if (vthread != threadObj) { // i.e. current thread is virtual
3656 * tid = java_lang_Thread::tid(vthread);
3657 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3658 * pinVirtualThread = VMContinuations;
3659 * excluded = vthread_epoch_raw & excluded_mask;
3660 * if (!excluded) {
3661 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3662 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3663 * if (vthread_epoch != current_epoch) {
3664 * write_checkpoint();
3665 * }
3666 * }
3667 * } else {
3668 * tid = java_lang_Thread::tid(threadObj);
3669 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3670 * pinVirtualThread = false;
3671 * excluded = thread_epoch_raw & excluded_mask;
3672 * }
3673 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3674 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3675 * if (tid_in_event_writer != tid) {
3676 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3677 * setField(event_writer, "excluded", excluded);
3678 * setField(event_writer, "threadID", tid);
3679 * }
3680 * return event_writer
3681 */
3682 bool LibraryCallKit::inline_native_getEventWriter() {
3683 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3684
3685 // Save input memory and i_o state.
3686 Node* input_memory_state = reset_memory();
3687 set_all_memory(input_memory_state);
3688 Node* input_io_state = i_o();
3689
3690 // The most significant bit of the u2 is used to denote thread exclusion
3691 Node* excluded_shift = _gvn.intcon(15);
3692 Node* excluded_mask = _gvn.intcon(1 << 15);
3693 // The epoch generation is the range [1-32767]
3694 Node* epoch_mask = _gvn.intcon(32767);
3695
3696 // TLS
3697 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3698
3699 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3700 Node* jobj_ptr = off_heap_plus_addr(tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3701
3702 // Load the eventwriter jobject handle.
3703 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3704
3705 // Null check the jobject handle.
3706 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3707 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3708 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3709
3710 // False path, jobj is null.
3711 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3712
3713 // True path, jobj is not null.
3714 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3715
3716 set_control(jobj_is_not_null);
3717
3718 // Load the threadObj for the CarrierThread.
3719 Node* threadObj = generate_current_thread(tls_ptr);
3720
3721 // Load the vthread.
3722 Node* vthread = generate_virtual_thread(tls_ptr);
3723
3724 // If vthread != threadObj, this is a virtual thread.
3725 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3726 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3727 IfNode* iff_vthread_not_equal_threadObj =
3728 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3729
3730 // False branch, fallback to threadObj.
3731 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3732 set_control(vthread_equal_threadObj);
3733
3734 // Load the tid field from the vthread object.
3735 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3736
3737 // Load the raw epoch value from the threadObj.
3738 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3739 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3740 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3741 TypeInt::CHAR, T_CHAR,
3742 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3743
3744 // Mask off the excluded information from the epoch.
3745 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3746
3747 // True branch, this is a virtual thread.
3748 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3749 set_control(vthread_not_equal_threadObj);
3750
3751 // Load the tid field from the vthread object.
3752 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3753
3754 // Continuation support determines if a virtual thread should be pinned.
3755 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3756 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3757
3758 // Load the raw epoch value from the vthread.
3759 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3760 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3761 TypeInt::CHAR, T_CHAR,
3762 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3763
3764 // Mask off the excluded information from the epoch.
3765 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, excluded_mask));
3766
3767 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3768 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, excluded_mask));
3769 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3770 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3771
3772 // False branch, vthread is excluded, no need to write epoch info.
3773 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3774
3775 // True branch, vthread is included, update epoch info.
3776 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3777 set_control(included);
3778
3779 // Get epoch value.
3780 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, epoch_mask));
3781
3782 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3783 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3784 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3785
3786 // Compare the epoch in the vthread to the current epoch generation.
3787 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3788 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3789 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3790
3791 // False path, epoch is equal, checkpoint information is valid.
3792 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3793
3794 // True path, epoch is not equal, write a checkpoint for the vthread.
3795 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3796
3797 set_control(epoch_is_not_equal);
3798
3799 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3800 // The call also updates the native thread local thread id and the vthread with the current epoch.
3801 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3802 OptoRuntime::jfr_write_checkpoint_Type(),
3803 SharedRuntime::jfr_write_checkpoint(),
3804 "write_checkpoint", TypePtr::BOTTOM);
3805 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3806
3807 // vthread epoch != current epoch
3808 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3809 record_for_igvn(epoch_compare_rgn);
3810 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3811 record_for_igvn(epoch_compare_mem);
3812 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3813 record_for_igvn(epoch_compare_io);
3814
3815 // Update control and phi nodes.
3816 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3817 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3818 epoch_compare_mem->init_req(_true_path, reset_memory());
3819 epoch_compare_mem->init_req(_false_path, input_memory_state);
3820 epoch_compare_io->init_req(_true_path, i_o());
3821 epoch_compare_io->init_req(_false_path, input_io_state);
3822
3823 // excluded != true
3824 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3825 record_for_igvn(exclude_compare_rgn);
3826 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3827 record_for_igvn(exclude_compare_mem);
3828 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3829 record_for_igvn(exclude_compare_io);
3830
3831 // Update control and phi nodes.
3832 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3833 exclude_compare_rgn->init_req(_false_path, excluded);
3834 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3835 exclude_compare_mem->init_req(_false_path, input_memory_state);
3836 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3837 exclude_compare_io->init_req(_false_path, input_io_state);
3838
3839 // vthread != threadObj
3840 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3841 record_for_igvn(vthread_compare_rgn);
3842 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3843 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3844 record_for_igvn(vthread_compare_io);
3845 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3846 record_for_igvn(tid);
3847 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3848 record_for_igvn(exclusion);
3849 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3850 record_for_igvn(pinVirtualThread);
3851
3852 // Update control and phi nodes.
3853 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3854 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3855 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3856 vthread_compare_mem->init_req(_false_path, input_memory_state);
3857 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3858 vthread_compare_io->init_req(_false_path, input_io_state);
3859 tid->init_req(_true_path, vthread_tid);
3860 tid->init_req(_false_path, thread_obj_tid);
3861 exclusion->init_req(_true_path, vthread_is_excluded);
3862 exclusion->init_req(_false_path, threadObj_is_excluded);
3863 pinVirtualThread->init_req(_true_path, continuation_support);
3864 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3865
3866 // Update branch state.
3867 set_control(_gvn.transform(vthread_compare_rgn));
3868 set_all_memory(_gvn.transform(vthread_compare_mem));
3869 set_i_o(_gvn.transform(vthread_compare_io));
3870
3871 // Load the event writer oop by dereferencing the jobject handle.
3872 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3873 assert(klass_EventWriter->is_loaded(), "invariant");
3874 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3875 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3876 const TypeOopPtr* const xtype = aklass->as_instance_type();
3877 Node* jobj_untagged = _gvn.transform(AddPNode::make_off_heap(jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3878 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3879
3880 // Load the current thread id from the event writer object.
3881 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3882 // Get the field offset to, conditionally, store an updated tid value later.
3883 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3884 // Get the field offset to, conditionally, store an updated exclusion value later.
3885 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3886 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3887 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3888
3889 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3890 record_for_igvn(event_writer_tid_compare_rgn);
3891 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3892 record_for_igvn(event_writer_tid_compare_mem);
3893 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3894 record_for_igvn(event_writer_tid_compare_io);
3895
3896 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3897 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3898 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3899 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3900
3901 // False path, tids are the same.
3902 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3903
3904 // True path, tid is not equal, need to update the tid in the event writer.
3905 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3906 record_for_igvn(tid_is_not_equal);
3907
3908 // Store the pin state to the event writer.
3909 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3910
3911 // Store the exclusion state to the event writer.
3912 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3913 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3914
3915 // Store the tid to the event writer.
3916 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3917
3918 // Update control and phi nodes.
3919 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3920 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3921 event_writer_tid_compare_mem->init_req(_true_path, reset_memory());
3922 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3923 event_writer_tid_compare_io->init_req(_true_path, i_o());
3924 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3925
3926 // Result of top level CFG, Memory, IO and Value.
3927 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3928 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3929 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3930 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3931
3932 // Result control.
3933 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3934 result_rgn->init_req(_false_path, jobj_is_null);
3935
3936 // Result memory.
3937 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3938 result_mem->init_req(_false_path, input_memory_state);
3939
3940 // Result IO.
3941 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3942 result_io->init_req(_false_path, input_io_state);
3943
3944 // Result value.
3945 result_value->init_req(_true_path, event_writer); // return event writer oop
3946 result_value->init_req(_false_path, null()); // return null
3947
3948 // Set output state.
3949 set_control(_gvn.transform(result_rgn));
3950 set_all_memory(_gvn.transform(result_mem));
3951 set_i_o(_gvn.transform(result_io));
3952 set_result(result_rgn, result_value);
3953 return true;
3954 }
3955
3956 /*
3957 * The intrinsic is a model of this pseudo-code:
3958 *
3959 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3960 * if (carrierThread != thread) { // is virtual thread
3961 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3962 * bool excluded = vthread_epoch_raw & excluded_mask;
3963 * AtomicAccess::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3964 * AtomicAccess::store(&tl->_contextual_thread_excluded, is_excluded);
3965 * if (!excluded) {
3966 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3967 * AtomicAccess::store(&tl->_vthread_epoch, vthread_epoch);
3968 * }
3969 * AtomicAccess::release_store(&tl->_vthread, true);
3970 * return;
3971 * }
3972 * AtomicAccess::release_store(&tl->_vthread, false);
3973 */
3974 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3975 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3976
3977 Node* input_memory_state = reset_memory();
3978 set_all_memory(input_memory_state);
3979
3980 // The most significant bit of the u2 is used to denote thread exclusion
3981 Node* excluded_mask = _gvn.intcon(1 << 15);
3982 // The epoch generation is the range [1-32767]
3983 Node* epoch_mask = _gvn.intcon(32767);
3984
3985 Node* const carrierThread = generate_current_thread(jt);
3986 // If thread != carrierThread, this is a virtual thread.
3987 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3988 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3989 IfNode* iff_thread_not_equal_carrierThread =
3990 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3991
3992 Node* vthread_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3993
3994 // False branch, is carrierThread.
3995 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3996 // Store release
3997 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3998
3999 set_all_memory(input_memory_state);
4000
4001 // True branch, is virtual thread.
4002 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
4003 set_control(thread_not_equal_carrierThread);
4004
4005 // Load the raw epoch value from the vthread.
4006 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
4007 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
4008 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
4009
4010 // Mask off the excluded information from the epoch.
4011 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, excluded_mask));
4012
4013 // Load the tid field from the thread.
4014 Node* tid = load_field_from_object(thread, "tid", "J");
4015
4016 // Store the vthread tid to the jfr thread local.
4017 Node* thread_id_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
4018 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
4019
4020 // Branch is_excluded to conditionalize updating the epoch .
4021 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, excluded_mask));
4022 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
4023 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
4024
4025 // True branch, vthread is excluded, no need to write epoch info.
4026 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
4027 set_control(excluded);
4028 Node* vthread_is_excluded = _gvn.intcon(1);
4029
4030 // False branch, vthread is included, update epoch info.
4031 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
4032 set_control(included);
4033 Node* vthread_is_included = _gvn.intcon(0);
4034
4035 // Get epoch value.
4036 Node* epoch = _gvn.transform(new AndINode(epoch_raw, epoch_mask));
4037
4038 // Store the vthread epoch to the jfr thread local.
4039 Node* vthread_epoch_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
4040 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
4041
4042 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
4043 record_for_igvn(excluded_rgn);
4044 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
4045 record_for_igvn(excluded_mem);
4046 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
4047 record_for_igvn(exclusion);
4048
4049 // Merge the excluded control and memory.
4050 excluded_rgn->init_req(_true_path, excluded);
4051 excluded_rgn->init_req(_false_path, included);
4052 excluded_mem->init_req(_true_path, tid_memory);
4053 excluded_mem->init_req(_false_path, included_memory);
4054 exclusion->init_req(_true_path, vthread_is_excluded);
4055 exclusion->init_req(_false_path, vthread_is_included);
4056
4057 // Set intermediate state.
4058 set_control(_gvn.transform(excluded_rgn));
4059 set_all_memory(excluded_mem);
4060
4061 // Store the vthread exclusion state to the jfr thread local.
4062 Node* thread_local_excluded_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
4063 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
4064
4065 // Store release
4066 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
4067
4068 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
4069 record_for_igvn(thread_compare_rgn);
4070 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
4071 record_for_igvn(thread_compare_mem);
4072 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
4073 record_for_igvn(vthread);
4074
4075 // Merge the thread_compare control and memory.
4076 thread_compare_rgn->init_req(_true_path, control());
4077 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
4078 thread_compare_mem->init_req(_true_path, vthread_true_memory);
4079 thread_compare_mem->init_req(_false_path, vthread_false_memory);
4080
4081 // Set output state.
4082 set_control(_gvn.transform(thread_compare_rgn));
4083 set_all_memory(_gvn.transform(thread_compare_mem));
4084 }
4085
4086 //------------------------inline_native_try_update_epoch------------------
4087 //
4088 // The generated code is a function of the argument type.
4089 //
4090 bool LibraryCallKit::inline_native_try_update_epoch() {
4091 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
4092
4093 // Save input memory.
4094 Node* input_memory_state = reset_memory();
4095 set_all_memory(input_memory_state);
4096
4097 // Argument is an oop whose class has an injected instance field,
4098 // called 'jfr_epoch' of type T_INT, used for holding a jfr epoch value.
4099 Node* oop = argument(0);
4100 const TypeInstPtr* tinst = _gvn.type(oop)->isa_instptr();
4101 assert(tinst != nullptr, "oop is null");
4102 assert(tinst->is_loaded(), "klass is not loaded");
4103 ciInstanceKlass* const ik = tinst->instance_klass();
4104
4105 ciField* const field = ik->get_injected_instance_field_by_name(ciSymbol::make("jfr_epoch"),
4106 ciSymbol::make("I"));
4107
4108 assert(field != nullptr, "field 'jfr_epoch' of type I not injected in klass %s", ik->name()->as_utf8());
4109
4110 const int jfr_epoch_field_offset = field->offset_in_bytes();
4111 Node* oop_epoch_field_offset = basic_plus_adr(oop, jfr_epoch_field_offset);
4112 const TypePtr* adr_type = _gvn.type(oop_epoch_field_offset)->isa_ptr();
4113 const int alias_idx = C->get_alias_index(adr_type);
4114 BasicType bt = field->layout_type();
4115 const Type * oop_epoch_field_type = Type::get_const_basic_type(bt);
4116
4117 // Load the epoch value from the oop.
4118 Node* oop_epoch = access_load_at(oop,
4119 oop_epoch_field_offset,
4120 adr_type, oop_epoch_field_type,
4121 bt, IN_HEAP | MO_UNORDERED);
4122
4123 // Load the current JFR epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
4124 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
4125 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
4126
4127 // Compare the epoch in the oop against the current JFR epoch generation.
4128 Node* const epochs_cmp = _gvn.transform(new CmpINode(current_epoch_generation, oop_epoch));
4129 Node* epochs_equal_test = _gvn.transform(new BoolNode(epochs_cmp, BoolTest::eq));
4130 IfNode* iff_epochs_equal = create_and_map_if(control(), epochs_equal_test, PROB_LIKELY(0.999), COUNT_UNKNOWN);
4131
4132 // True path.
4133 Node* epochs_are_equal = _gvn.transform(new IfTrueNode(iff_epochs_equal));
4134
4135 // False path.
4136 Node* epochs_are_not_equal = _gvn.transform(new IfFalseNode(iff_epochs_equal));
4137
4138 set_control(_gvn.transform(epochs_are_not_equal));
4139
4140 // Attempt to cas the current JFR epoch generation into the oop epoch field.
4141 DecoratorSet decorators = IN_HEAP;
4142 decorators |= mo_decorator_for_access_kind(Volatile);
4143
4144 Node* result = access_atomic_cmpxchg_val_at(oop,
4145 oop_epoch_field_offset,
4146 adr_type, alias_idx,
4147 oop_epoch, // expected value
4148 current_epoch_generation, // new value
4149 oop_epoch_field_type,
4150 bt,
4151 decorators);
4152
4153 // Compare the result of the cas operation to the expected value.
4154 Node* const cas_cmp_to_expected_value = _gvn.transform(new CmpINode(result, oop_epoch));
4155 Node* cas_operation_test = _gvn.transform(new BoolNode(cas_cmp_to_expected_value, BoolTest::eq));
4156 IfNode* iff_cas_success = create_and_map_if(control(), cas_operation_test, PROB_LIKELY(0.999), COUNT_UNKNOWN);
4157
4158 // True path.
4159 Node* cas_success = _gvn.transform(new IfTrueNode(iff_cas_success));
4160
4161 // False path.
4162 Node* cas_failure = _gvn.transform(new IfFalseNode(iff_cas_success));
4163
4164 // Cas result region and phi nodes.
4165 RegionNode* cas_operation_rgn = new RegionNode(PATH_LIMIT);
4166 record_for_igvn(cas_operation_rgn);
4167 PhiNode* cas_operation_mem = new PhiNode(cas_operation_rgn, Type::MEMORY, TypePtr::BOTTOM);
4168 record_for_igvn(cas_operation_mem);
4169 PhiNode* cas_result = new PhiNode(cas_operation_rgn, TypeInt::BOOL);
4170 record_for_igvn(cas_result);
4171
4172 cas_operation_rgn->init_req(_true_path, _gvn.transform(cas_success));
4173 cas_operation_rgn->init_req(_false_path, _gvn.transform(cas_failure));
4174 cas_operation_mem->init_req(_true_path, reset_memory());
4175 cas_operation_mem->init_req(_false_path, input_memory_state);
4176 cas_result->init_req(_true_path, _gvn.intcon(1));
4177 cas_result->init_req(_false_path, _gvn.intcon(0));
4178
4179 // Epoch compare region and phi nodes.
4180 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
4181 record_for_igvn(epoch_compare_rgn);
4182 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
4183 record_for_igvn(epoch_compare_mem);
4184 PhiNode* result_value = new PhiNode(epoch_compare_rgn, TypeInt::BOOL);
4185 record_for_igvn(result_value);
4186
4187 epoch_compare_rgn->init_req(_true_path, _gvn.transform(epochs_are_equal));
4188 epoch_compare_rgn->init_req(_false_path, _gvn.transform(cas_operation_rgn));
4189 epoch_compare_mem->init_req(_true_path, _gvn.transform(input_memory_state));
4190 epoch_compare_mem->init_req(_false_path, _gvn.transform(cas_operation_mem));
4191 result_value->init_req(_true_path, _gvn.intcon(0));
4192 result_value->init_req(_false_path, _gvn.transform(cas_result));
4193
4194 // Set output state.
4195 set_result(epoch_compare_rgn, result_value);
4196 set_all_memory(_gvn.transform(epoch_compare_mem));
4197
4198 return true;
4199 }
4200
4201 #endif // JFR_HAVE_INTRINSICS
4202
4203 //------------------------inline_native_currentCarrierThread------------------
4204 bool LibraryCallKit::inline_native_currentCarrierThread() {
4205 Node* junk = nullptr;
4206 set_result(generate_current_thread(junk));
4207 return true;
4208 }
4209
4210 //------------------------inline_native_currentThread------------------
4211 bool LibraryCallKit::inline_native_currentThread() {
4212 Node* junk = nullptr;
4213 set_result(generate_virtual_thread(junk));
4214 return true;
4215 }
4216
4217 //------------------------inline_native_setVthread------------------
4218 bool LibraryCallKit::inline_native_setCurrentThread() {
4219 assert(C->method()->changes_current_thread(),
4220 "method changes current Thread but is not annotated ChangesCurrentThread");
4221 Node* arr = argument(1);
4222 Node* thread = _gvn.transform(new ThreadLocalNode());
4223 Node* p = off_heap_plus_addr(thread, in_bytes(JavaThread::vthread_offset()));
4224 Node* thread_obj_handle
4225 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
4226 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
4227 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
4228
4229 // Change the _monitor_owner_id of the JavaThread
4230 Node* tid = load_field_from_object(arr, "tid", "J");
4231 Node* monitor_owner_id_offset = off_heap_plus_addr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
4232 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
4233
4234 JFR_ONLY(extend_setCurrentThread(thread, arr);)
4235 return true;
4236 }
4237
4238 const Type* LibraryCallKit::scopedValueCache_type() {
4239 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
4240 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
4241 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS, /* stable= */ false, /* flat= */ false, /* not_flat= */ true, /* not_null_free= */ true, true);
4242
4243 // Because we create the scopedValue cache lazily we have to make the
4244 // type of the result BotPTR.
4245 bool xk = etype->klass_is_exact();
4246 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, TypeAryPtr::Offset(0));
4247 return objects_type;
4248 }
4249
4250 Node* LibraryCallKit::scopedValueCache_helper() {
4251 Node* thread = _gvn.transform(new ThreadLocalNode());
4252 Node* p = off_heap_plus_addr(thread, in_bytes(JavaThread::scopedValueCache_offset()));
4253 // We cannot use immutable_memory() because we might flip onto a
4254 // different carrier thread, at which point we'll need to use that
4255 // carrier thread's cache.
4256 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
4257 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
4258 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
4259 }
4260
4261 //------------------------inline_native_scopedValueCache------------------
4262 bool LibraryCallKit::inline_native_scopedValueCache() {
4263 Node* cache_obj_handle = scopedValueCache_helper();
4264 const Type* objects_type = scopedValueCache_type();
4265 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
4266
4267 return true;
4268 }
4269
4270 //------------------------inline_native_setScopedValueCache------------------
4271 bool LibraryCallKit::inline_native_setScopedValueCache() {
4272 Node* arr = argument(0);
4273 Node* cache_obj_handle = scopedValueCache_helper();
4274 const Type* objects_type = scopedValueCache_type();
4275
4276 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
4277 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
4278
4279 return true;
4280 }
4281
4282 //------------------------inline_native_Continuation_pin and unpin-----------
4283
4284 // Shared implementation routine for both pin and unpin.
4285 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
4286 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
4287
4288 // Save input memory.
4289 Node* input_memory_state = reset_memory();
4290 set_all_memory(input_memory_state);
4291
4292 // TLS
4293 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
4294 Node* last_continuation_offset = off_heap_plus_addr(tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
4295 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
4296
4297 // Null check the last continuation object.
4298 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
4299 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
4300 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
4301
4302 // False path, last continuation is null.
4303 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
4304
4305 // True path, last continuation is not null.
4306 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
4307
4308 set_control(continuation_is_not_null);
4309
4310 // Load the pin count from the last continuation.
4311 Node* pin_count_offset = off_heap_plus_addr(last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
4312 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
4313
4314 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
4315 Node* pin_count_rhs;
4316 if (unpin) {
4317 pin_count_rhs = _gvn.intcon(0);
4318 } else {
4319 pin_count_rhs = _gvn.intcon(UINT32_MAX);
4320 }
4321 Node* pin_count_cmp = _gvn.transform(new CmpUNode(pin_count, pin_count_rhs));
4322 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
4323 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
4324
4325 // True branch, pin count over/underflow.
4326 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
4327 {
4328 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
4329 // which will throw IllegalStateException for pin count over/underflow.
4330 // No memory changed so far - we can use memory create by reset_memory()
4331 // at the beginning of this intrinsic. No need to call reset_memory() again.
4332 PreserveJVMState pjvms(this);
4333 set_control(pin_count_over_underflow);
4334 uncommon_trap(Deoptimization::Reason_intrinsic,
4335 Deoptimization::Action_none);
4336 assert(stopped(), "invariant");
4337 }
4338
4339 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
4340 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
4341 set_control(valid_pin_count);
4342
4343 Node* next_pin_count;
4344 if (unpin) {
4345 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
4346 } else {
4347 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
4348 }
4349
4350 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
4351
4352 // Result of top level CFG and Memory.
4353 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
4354 record_for_igvn(result_rgn);
4355 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
4356 record_for_igvn(result_mem);
4357
4358 result_rgn->init_req(_true_path, valid_pin_count);
4359 result_rgn->init_req(_false_path, continuation_is_null);
4360 result_mem->init_req(_true_path, reset_memory());
4361 result_mem->init_req(_false_path, input_memory_state);
4362
4363 // Set output state.
4364 set_control(_gvn.transform(result_rgn));
4365 set_all_memory(_gvn.transform(result_mem));
4366
4367 return true;
4368 }
4369
4370 //---------------------------load_mirror_from_klass----------------------------
4371 // Given a klass oop, load its java mirror (a java.lang.Class oop).
4372 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
4373 Node* p = off_heap_plus_addr(klass, in_bytes(Klass::java_mirror_offset()));
4374 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4375 // mirror = ((OopHandle)mirror)->resolve();
4376 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
4377 }
4378
4379 //-----------------------load_klass_from_mirror_common-------------------------
4380 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
4381 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
4382 // and branch to the given path on the region.
4383 // If never_see_null, take an uncommon trap on null, so we can optimistically
4384 // compile for the non-null case.
4385 // If the region is null, force never_see_null = true.
4386 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
4387 bool never_see_null,
4388 RegionNode* region,
4389 int null_path,
4390 int offset) {
4391 if (region == nullptr) never_see_null = true;
4392 Node* p = basic_plus_adr(mirror, offset);
4393 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4394 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
4395 Node* null_ctl = top();
4396 kls = null_check_oop(kls, &null_ctl, never_see_null);
4397 if (region != nullptr) {
4398 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
4399 region->init_req(null_path, null_ctl);
4400 } else {
4401 assert(null_ctl == top(), "no loose ends");
4402 }
4403 return kls;
4404 }
4405
4406 //--------------------(inline_native_Class_query helpers)---------------------
4407 // Use this for JVM_ACC_INTERFACE.
4408 // Fall through if (mods & mask) == bits, take the guard otherwise.
4409 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
4410 ByteSize offset, const Type* type, BasicType bt) {
4411 // Branch around if the given klass has the given modifier bit set.
4412 // Like generate_guard, adds a new path onto the region.
4413 Node* modp = off_heap_plus_addr(kls, in_bytes(offset));
4414 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
4415 Node* mask = intcon(modifier_mask);
4416 Node* bits = intcon(modifier_bits);
4417 Node* mbit = _gvn.transform(new AndINode(mods, mask));
4418 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
4419 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
4420 return generate_fair_guard(bol, region);
4421 }
4422
4423 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
4424 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
4425 InstanceKlass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
4426 }
4427
4428 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
4429 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
4430 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
4431 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
4432 }
4433
4434 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
4435 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
4436 }
4437
4438 //-------------------------inline_native_Class_query-------------------
4439 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
4440 const Type* return_type = TypeInt::BOOL;
4441 Node* prim_return_value = top(); // what happens if it's a primitive class?
4442 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4443 bool expect_prim = false; // most of these guys expect to work on refs
4444
4445 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
4446
4447 Node* mirror = argument(0);
4448 Node* obj = top();
4449
4450 switch (id) {
4451 case vmIntrinsics::_isInstance:
4452 // nothing is an instance of a primitive type
4453 prim_return_value = intcon(0);
4454 obj = argument(1);
4455 break;
4456 case vmIntrinsics::_isHidden:
4457 prim_return_value = intcon(0);
4458 break;
4459 case vmIntrinsics::_getSuperclass:
4460 prim_return_value = null();
4461 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
4462 break;
4463 default:
4464 fatal_unexpected_iid(id);
4465 break;
4466 }
4467
4468 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4469 if (mirror_con == nullptr) return false; // cannot happen?
4470
4471 #ifndef PRODUCT
4472 if (C->print_intrinsics() || C->print_inlining()) {
4473 ciType* k = mirror_con->java_mirror_type();
4474 if (k) {
4475 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4476 k->print_name();
4477 tty->cr();
4478 }
4479 }
4480 #endif
4481
4482 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4483 RegionNode* region = new RegionNode(PATH_LIMIT);
4484 record_for_igvn(region);
4485 PhiNode* phi = new PhiNode(region, return_type);
4486
4487 // The mirror will never be null of Reflection.getClassAccessFlags, however
4488 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4489 // if it is. See bug 4774291.
4490
4491 // For Reflection.getClassAccessFlags(), the null check occurs in
4492 // the wrong place; see inline_unsafe_access(), above, for a similar
4493 // situation.
4494 mirror = null_check(mirror);
4495 // If mirror or obj is dead, only null-path is taken.
4496 if (stopped()) return true;
4497
4498 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4499
4500 // Now load the mirror's klass metaobject, and null-check it.
4501 // Side-effects region with the control path if the klass is null.
4502 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4503 // If kls is null, we have a primitive mirror.
4504 phi->init_req(_prim_path, prim_return_value);
4505 if (stopped()) { set_result(region, phi); return true; }
4506 bool safe_for_replace = (region->in(_prim_path) == top());
4507
4508 Node* p; // handy temp
4509 Node* null_ctl;
4510
4511 // Now that we have the non-null klass, we can perform the real query.
4512 // For constant classes, the query will constant-fold in LoadNode::Value.
4513 Node* query_value = top();
4514 switch (id) {
4515 case vmIntrinsics::_isInstance:
4516 // nothing is an instance of a primitive type
4517 query_value = gen_instanceof(obj, kls, safe_for_replace);
4518 break;
4519
4520 case vmIntrinsics::_isHidden:
4521 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4522 if (generate_hidden_class_guard(kls, region) != nullptr)
4523 // A guard was added. If the guard is taken, it was an hidden class.
4524 phi->add_req(intcon(1));
4525 // If we fall through, it's a plain class.
4526 query_value = intcon(0);
4527 break;
4528
4529
4530 case vmIntrinsics::_getSuperclass:
4531 // The rules here are somewhat unfortunate, but we can still do better
4532 // with random logic than with a JNI call.
4533 // Interfaces store null or Object as _super, but must report null.
4534 // Arrays store an intermediate super as _super, but must report Object.
4535 // Other types can report the actual _super.
4536 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4537 if (generate_array_guard(kls, region) != nullptr) {
4538 // A guard was added. If the guard is taken, it was an array.
4539 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4540 }
4541 // Check for interface after array since this checks AccessFlags offset into InstanceKlass.
4542 // In other words, we are accessing subtype-specific information, so we need to determine the subtype first.
4543 if (generate_interface_guard(kls, region) != nullptr) {
4544 // A guard was added. If the guard is taken, it was an interface.
4545 phi->add_req(null());
4546 }
4547 // If we fall through, it's a plain class. Get its _super.
4548 if (!stopped()) {
4549 p = basic_plus_adr(top(), kls, in_bytes(Klass::super_offset()));
4550 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4551 null_ctl = top();
4552 kls = null_check_oop(kls, &null_ctl);
4553 if (null_ctl != top()) {
4554 // If the guard is taken, Object.superClass is null (both klass and mirror).
4555 region->add_req(null_ctl);
4556 phi ->add_req(null());
4557 }
4558 if (!stopped()) {
4559 query_value = load_mirror_from_klass(kls);
4560 }
4561 }
4562 break;
4563
4564 default:
4565 fatal_unexpected_iid(id);
4566 break;
4567 }
4568
4569 // Fall-through is the normal case of a query to a real class.
4570 phi->init_req(1, query_value);
4571 region->init_req(1, control());
4572
4573 C->set_has_split_ifs(true); // Has chance for split-if optimization
4574 set_result(region, phi);
4575 return true;
4576 }
4577
4578
4579 //-------------------------inline_Class_cast-------------------
4580 bool LibraryCallKit::inline_Class_cast() {
4581 Node* mirror = argument(0); // Class
4582 Node* obj = argument(1);
4583 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4584 if (mirror_con == nullptr) {
4585 return false; // dead path (mirror->is_top()).
4586 }
4587 if (obj == nullptr || obj->is_top()) {
4588 return false; // dead path
4589 }
4590 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4591
4592 // First, see if Class.cast() can be folded statically.
4593 // java_mirror_type() returns non-null for compile-time Class constants.
4594 ciType* tm = mirror_con->java_mirror_type();
4595 if (tm != nullptr && tm->is_klass() &&
4596 tp != nullptr) {
4597 if (!tp->is_loaded()) {
4598 // Don't use intrinsic when class is not loaded.
4599 return false;
4600 } else {
4601 const TypeKlassPtr* tklass = TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces);
4602 int static_res = C->static_subtype_check(tklass, tp->as_klass_type());
4603 if (static_res == Compile::SSC_always_true) {
4604 // isInstance() is true - fold the code.
4605 set_result(obj);
4606 return true;
4607 } else if (static_res == Compile::SSC_always_false) {
4608 // Don't use intrinsic, have to throw ClassCastException.
4609 // If the reference is null, the non-intrinsic bytecode will
4610 // be optimized appropriately.
4611 return false;
4612 }
4613 }
4614 }
4615
4616 // Bailout intrinsic and do normal inlining if exception path is frequent.
4617 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4618 return false;
4619 }
4620
4621 // Generate dynamic checks.
4622 // Class.cast() is java implementation of _checkcast bytecode.
4623 // Do checkcast (Parse::do_checkcast()) optimizations here.
4624
4625 mirror = null_check(mirror);
4626 // If mirror is dead, only null-path is taken.
4627 if (stopped()) {
4628 return true;
4629 }
4630
4631 // Not-subtype or the mirror's klass ptr is nullptr (in case it is a primitive).
4632 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
4633 RegionNode* region = new RegionNode(PATH_LIMIT);
4634 record_for_igvn(region);
4635
4636 // Now load the mirror's klass metaobject, and null-check it.
4637 // If kls is null, we have a primitive mirror and
4638 // nothing is an instance of a primitive type.
4639 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4640
4641 Node* res = top();
4642 Node* io = i_o();
4643 Node* mem = merged_memory();
4644 SafePointNode* new_cast_failure_map = nullptr;
4645
4646 if (!stopped()) {
4647
4648 Node* bad_type_ctrl = top();
4649 // Do checkcast optimizations.
4650 res = gen_checkcast(obj, kls, &bad_type_ctrl, &new_cast_failure_map);
4651 region->init_req(_bad_type_path, bad_type_ctrl);
4652 }
4653 if (region->in(_prim_path) != top() ||
4654 region->in(_bad_type_path) != top() ||
4655 region->in(_npe_path) != top()) {
4656 // Let Interpreter throw ClassCastException.
4657 PreserveJVMState pjvms(this);
4658 if (new_cast_failure_map != nullptr) {
4659 // The current map on the success path could have been modified. Use the dedicated failure path map.
4660 set_map(new_cast_failure_map);
4661 }
4662 set_control(_gvn.transform(region));
4663 // Set IO and memory because gen_checkcast may override them when buffering inline types
4664 set_i_o(io);
4665 set_all_memory(mem);
4666 uncommon_trap(Deoptimization::Reason_intrinsic,
4667 Deoptimization::Action_maybe_recompile);
4668 }
4669 if (!stopped()) {
4670 set_result(res);
4671 }
4672 return true;
4673 }
4674
4675
4676 //--------------------------inline_native_subtype_check------------------------
4677 // This intrinsic takes the JNI calls out of the heart of
4678 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4679 bool LibraryCallKit::inline_native_subtype_check() {
4680 // Pull both arguments off the stack.
4681 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4682 args[0] = argument(0);
4683 args[1] = argument(1);
4684 Node* klasses[2]; // corresponding Klasses: superk, subk
4685 klasses[0] = klasses[1] = top();
4686
4687 enum {
4688 // A full decision tree on {superc is prim, subc is prim}:
4689 _prim_0_path = 1, // {P,N} => false
4690 // {P,P} & superc!=subc => false
4691 _prim_same_path, // {P,P} & superc==subc => true
4692 _prim_1_path, // {N,P} => false
4693 _ref_subtype_path, // {N,N} & subtype check wins => true
4694 _both_ref_path, // {N,N} & subtype check loses => false
4695 PATH_LIMIT
4696 };
4697
4698 RegionNode* region = new RegionNode(PATH_LIMIT);
4699 RegionNode* prim_region = new RegionNode(2);
4700 Node* phi = new PhiNode(region, TypeInt::BOOL);
4701 record_for_igvn(region);
4702 record_for_igvn(prim_region);
4703
4704 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4705 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4706 int class_klass_offset = java_lang_Class::klass_offset();
4707
4708 // First null-check both mirrors and load each mirror's klass metaobject.
4709 int which_arg;
4710 for (which_arg = 0; which_arg <= 1; which_arg++) {
4711 Node* arg = args[which_arg];
4712 arg = null_check(arg);
4713 if (stopped()) break;
4714 args[which_arg] = arg;
4715
4716 Node* p = basic_plus_adr(arg, class_klass_offset);
4717 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4718 klasses[which_arg] = _gvn.transform(kls);
4719 }
4720
4721 // Having loaded both klasses, test each for null.
4722 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4723 for (which_arg = 0; which_arg <= 1; which_arg++) {
4724 Node* kls = klasses[which_arg];
4725 Node* null_ctl = top();
4726 kls = null_check_oop(kls, &null_ctl, never_see_null);
4727 if (which_arg == 0) {
4728 prim_region->init_req(1, null_ctl);
4729 } else {
4730 region->init_req(_prim_1_path, null_ctl);
4731 }
4732 if (stopped()) break;
4733 klasses[which_arg] = kls;
4734 }
4735
4736 if (!stopped()) {
4737 // now we have two reference types, in klasses[0..1]
4738 Node* subk = klasses[1]; // the argument to isAssignableFrom
4739 Node* superk = klasses[0]; // the receiver
4740 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4741 region->set_req(_ref_subtype_path, control());
4742 }
4743
4744 // If both operands are primitive (both klasses null), then
4745 // we must return true when they are identical primitives.
4746 // It is convenient to test this after the first null klass check.
4747 // This path is also used if superc is a value mirror.
4748 set_control(_gvn.transform(prim_region));
4749 if (!stopped()) {
4750 // Since superc is primitive, make a guard for the superc==subc case.
4751 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4752 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4753 generate_fair_guard(bol_eq, region);
4754 if (region->req() == PATH_LIMIT+1) {
4755 // A guard was added. If the added guard is taken, superc==subc.
4756 region->swap_edges(PATH_LIMIT, _prim_same_path);
4757 region->del_req(PATH_LIMIT);
4758 }
4759 region->set_req(_prim_0_path, control()); // Not equal after all.
4760 }
4761
4762 // these are the only paths that produce 'true':
4763 phi->set_req(_prim_same_path, intcon(1));
4764 phi->set_req(_ref_subtype_path, intcon(1));
4765
4766 // pull together the cases:
4767 assert(region->req() == PATH_LIMIT, "sane region");
4768 for (uint i = 1; i < region->req(); i++) {
4769 Node* ctl = region->in(i);
4770 if (ctl == nullptr || ctl == top()) {
4771 region->set_req(i, top());
4772 phi ->set_req(i, top());
4773 } else if (phi->in(i) == nullptr) {
4774 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4775 }
4776 }
4777
4778 set_control(_gvn.transform(region));
4779 set_result(_gvn.transform(phi));
4780 return true;
4781 }
4782
4783 //---------------------generate_array_guard_common------------------------
4784 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind, Node** obj) {
4785
4786 if (stopped()) {
4787 return nullptr;
4788 }
4789
4790 // Like generate_guard, adds a new path onto the region.
4791 jint layout_con = 0;
4792 Node* layout_val = get_layout_helper(kls, layout_con);
4793 if (layout_val == nullptr) {
4794 bool query = 0;
4795 switch(kind) {
4796 case RefArray: query = Klass::layout_helper_is_refArray(layout_con); break;
4797 case NonRefArray: query = !Klass::layout_helper_is_refArray(layout_con); break;
4798 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
4799 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
4800 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
4801 default:
4802 ShouldNotReachHere();
4803 }
4804 if (!query) {
4805 return nullptr; // never a branch
4806 } else { // always a branch
4807 Node* always_branch = control();
4808 if (region != nullptr)
4809 region->add_req(always_branch);
4810 set_control(top());
4811 return always_branch;
4812 }
4813 }
4814 unsigned int value = 0;
4815 BoolTest::mask btest = BoolTest::illegal;
4816 switch(kind) {
4817 case RefArray:
4818 case NonRefArray: {
4819 value = Klass::_lh_array_tag_ref_value;
4820 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4821 btest = (kind == RefArray) ? BoolTest::eq : BoolTest::ne;
4822 break;
4823 }
4824 case TypeArray: {
4825 value = Klass::_lh_array_tag_type_value;
4826 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4827 btest = BoolTest::eq;
4828 break;
4829 }
4830 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
4831 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
4832 default:
4833 ShouldNotReachHere();
4834 }
4835 // Now test the correct condition.
4836 jint nval = (jint)value;
4837 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4838 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4839 Node* ctrl = generate_fair_guard(bol, region);
4840 Node* is_array_ctrl = kind == NonArray ? control() : ctrl;
4841 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4842 // Keep track of the fact that 'obj' is an array to prevent
4843 // array specific accesses from floating above the guard.
4844 *obj = _gvn.transform(new CheckCastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4845 }
4846 return ctrl;
4847 }
4848
4849 // public static native Object[] ValueClass::newNullRestrictedAtomicArray(Class<?> componentType, int length, Object initVal);
4850 // public static native Object[] ValueClass::newNullRestrictedNonAtomicArray(Class<?> componentType, int length, Object initVal);
4851 // public static native Object[] ValueClass::newNullableAtomicArray(Class<?> componentType, int length);
4852 bool LibraryCallKit::inline_newArray(bool null_free, bool atomic) {
4853 assert(null_free || atomic, "nullable implies atomic");
4854 Node* componentType = argument(0);
4855 Node* length = argument(1);
4856 Node* init_val = null_free ? argument(2) : nullptr;
4857
4858 const TypeInstPtr* tp = _gvn.type(componentType)->isa_instptr();
4859 if (tp != nullptr) {
4860 ciInstanceKlass* ik = tp->instance_klass();
4861 if (ik == C->env()->Class_klass()) {
4862 ciType* t = tp->java_mirror_type();
4863 if (t != nullptr && t->is_inlinetype()) {
4864
4865 ciArrayKlass* array_klass = ciArrayKlass::make(t, null_free, atomic, true);
4866 assert(array_klass->is_elem_null_free() == null_free, "inconsistency");
4867
4868 // TOOD 8350865 ZGC needs card marks on initializing oop stores
4869 if (UseZGC && null_free && !array_klass->is_flat_array_klass()) {
4870 return false;
4871 }
4872
4873 if (array_klass->is_loaded() && array_klass->element_klass()->as_inline_klass()->is_initialized()) {
4874 const TypeAryKlassPtr* array_klass_type = TypeAryKlassPtr::make(array_klass, Type::trust_interfaces);
4875 if (null_free) {
4876 if (init_val->is_InlineType()) {
4877 if (array_klass_type->is_flat() && init_val->as_InlineType()->is_all_zero(&gvn(), /* flat */ true)) {
4878 // Zeroing is enough because the init value is the all-zero value
4879 init_val = nullptr;
4880 } else {
4881 init_val = init_val->as_InlineType()->buffer(this);
4882 }
4883 }
4884 if (init_val != nullptr) {
4885 #ifdef ASSERT
4886 init_val = null_check(init_val);
4887 Node* wrong_type_ctl = gen_subtype_check(init_val, makecon(TypeKlassPtr::make(array_klass->element_klass())));
4888 {
4889 PreserveJVMState pjvms(this);
4890 set_control(wrong_type_ctl);
4891 halt(control(), frameptr(), "incompatible type for initVal in newArray");
4892 stop_and_kill_map();
4893 }
4894 #endif
4895 init_val = _gvn.transform(new CheckCastPPNode(control(), init_val, TypeOopPtr::make_from_klass(array_klass->element_klass()), ConstraintCastNode::DependencyType::NonFloatingNarrowing));
4896 }
4897 }
4898 Node* obj = new_array(makecon(array_klass_type), length, 0, nullptr, false, init_val);
4899 const TypeAryPtr* arytype = gvn().type(obj)->is_aryptr();
4900 assert(arytype->is_null_free() == null_free, "inconsistency");
4901 assert(arytype->is_not_null_free() == !null_free, "inconsistency");
4902 set_result(obj);
4903 return true;
4904 }
4905 }
4906 }
4907 }
4908 return false;
4909 }
4910
4911 // public static native boolean ValueClass::isFlatArray(Object array);
4912 // public static native boolean ValueClass::isNullRestrictedArray(Object array);
4913 // public static native boolean ValueClass::isAtomicArray(Object array);
4914 bool LibraryCallKit::inline_getArrayProperties(ArrayPropertiesCheck check) {
4915 Node* array = argument(0);
4916
4917 Node* bol;
4918 switch(check) {
4919 case IsFlat:
4920 bol = flat_array_test(load_object_klass(array));
4921 break;
4922 case IsNullRestricted:
4923 bol = null_free_array_test(array);
4924 break;
4925 case IsAtomic: {
4926 // See conditions in JVM_IsAtomicArray
4927 // 1. If not flat, then atomic, or else...
4928 RegionNode* atomic_region = new RegionNode(1);
4929 RegionNode* non_atomic_region = new RegionNode(1);
4930 Node* array_klass = load_object_klass(array);
4931 Node* is_flat_bol = flat_array_test(array_klass);
4932 IfNode* iff_is_flat = create_and_xform_if(control(), is_flat_bol, PROB_FAIR, COUNT_UNKNOWN);
4933 atomic_region->add_req(_gvn.transform(new IfFalseNode(iff_is_flat)));
4934 set_control(_gvn.transform(new IfTrueNode(iff_is_flat)));
4935
4936 // 2. ...if the layout is atomic, then atomic, or else...
4937 Node* layout_kind = atomic_layout_array_test_and_get_layout_kind(array, atomic_region);
4938
4939 // 3. ...if the element type is naturally atomic and null-free OR empty and nullable, then atomic, or else...
4940 int element_klass_offset = in_bytes(ObjArrayKlass::element_klass_offset());
4941 Node* array_element_klass_addr = off_heap_plus_addr(array_klass, element_klass_offset);
4942 Node* array_element_klass = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), array_element_klass_addr, _gvn.type(array_klass)->is_klassptr()));
4943 int klass_flags_offset = in_bytes(InstanceKlass::misc_flags_offset() + InstanceKlassFlags::flags_offset());
4944 Node* array_element_klass_flags_addr = off_heap_plus_addr(array_element_klass, klass_flags_offset);
4945 Node* array_element_klass_flags = make_load(control(), array_element_klass_flags_addr, TypeInt::INT, T_INT, MemNode::unordered);
4946
4947 // Here, layout can only be non-atomic, otherwise atomic_layout_array_test_and_get_layout_kind already decides the array to be atomic.
4948 Node* is_null_free_cmp = _gvn.transform(new CmpINode(layout_kind, intcon(static_cast<jint>(LayoutKind::NULL_FREE_NON_ATOMIC_FLAT))));
4949 Node* is_null_free_bol = _gvn.transform(new BoolNode(is_null_free_cmp, BoolTest::eq));
4950 IfNode* iff_is_null_free_bol = create_and_xform_if(control(), is_null_free_bol, PROB_FAIR, COUNT_UNKNOWN);
4951 Node* is_null_free_ctl = _gvn.transform(new IfTrueNode(iff_is_null_free_bol));
4952 Node* is_nullable_ctl = _gvn.transform(new IfFalseNode(iff_is_null_free_bol));
4953
4954 Node* is_naturally_atomic_flag = _gvn.transform(new AndINode(array_element_klass_flags, intcon(InstanceKlassFlags::_misc_is_naturally_atomic)));
4955 Node* is_naturally_atomic_cmp = _gvn.transform(new CmpINode(is_naturally_atomic_flag, intcon(0)));
4956 Node* is_naturally_atomic_bol = _gvn.transform(new BoolNode(is_naturally_atomic_cmp, BoolTest::ne));
4957 IfNode* iff_is_naturally_atomic = create_and_xform_if(is_null_free_ctl, is_naturally_atomic_bol, PROB_FAIR, COUNT_UNKNOWN);
4958 Node* is_naturally_atomic_ctl = _gvn.transform(new IfTrueNode(iff_is_naturally_atomic));
4959 Node* is_not_naturally_atomic_ctl = _gvn.transform(new IfFalseNode(iff_is_naturally_atomic));
4960 atomic_region->add_req(is_naturally_atomic_ctl);
4961 non_atomic_region->add_req(is_not_naturally_atomic_ctl);
4962
4963 Node* is_empty_inline_type_flag = _gvn.transform(new AndINode(array_element_klass_flags, intcon(InstanceKlassFlags::_misc_is_empty_inline_type)));
4964 Node* is_empty_inline_type_cmp = _gvn.transform(new CmpINode(is_empty_inline_type_flag, intcon(0)));
4965 Node* is_empty_inline_type_bol = _gvn.transform(new BoolNode(is_empty_inline_type_cmp, BoolTest::ne));
4966 IfNode* iff_is_empty_inline_type = create_and_xform_if(is_nullable_ctl, is_empty_inline_type_bol, PROB_FAIR, COUNT_UNKNOWN);
4967 Node* is_empty_inline_type_ctl = _gvn.transform(new IfTrueNode(iff_is_empty_inline_type));
4968 Node* is_nonempty_inline_type_ctl = _gvn.transform(new IfFalseNode(iff_is_empty_inline_type));
4969 atomic_region->add_req(is_empty_inline_type_ctl);
4970 non_atomic_region->add_req(is_nonempty_inline_type_ctl);
4971
4972 // ...non-atomic, but we tried everything.
4973 RegionNode* decision = new RegionNode(3);
4974 decision->set_req(1, _gvn.transform(atomic_region));
4975 decision->set_req(2, _gvn.transform(non_atomic_region));
4976 PhiNode* result = PhiNode::make(decision, intcon(1), TypeInt::BOOL);
4977 result->set_req(2, intcon(0));
4978 set_control(_gvn.transform(decision));
4979 set_result(_gvn.transform(result));
4980 return true;
4981 }
4982 default:
4983 ShouldNotReachHere();
4984 }
4985
4986 Node* res = gvn().transform(new CMoveINode(bol, intcon(0), intcon(1), TypeInt::BOOL));
4987 set_result(res);
4988 return true;
4989 }
4990
4991 // Load the default refined array klass from an ObjArrayKlass. This relies on the first entry in the
4992 // '_next_refined_array_klass' linked list being the default (see ObjArrayKlass::klass_with_properties).
4993 Node* LibraryCallKit::load_default_refined_array_klass(Node* klass_node, bool type_array_guard) {
4994 RegionNode* region = new RegionNode(2);
4995 Node* phi = new PhiNode(region, TypeInstKlassPtr::OBJECT_OR_NULL);
4996
4997 if (type_array_guard) {
4998 generate_typeArray_guard(klass_node, region);
4999 if (region->req() == 3) {
5000 phi->add_req(klass_node);
5001 }
5002 }
5003 Node* adr_refined_klass = basic_plus_adr(top(), klass_node, in_bytes(ObjArrayKlass::next_refined_array_klass_offset()));
5004 Node* refined_klass = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), adr_refined_klass, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
5005
5006 // Can be null if not initialized yet, just deopt
5007 Node* null_ctl = top();
5008 refined_klass = null_check_oop(refined_klass, &null_ctl, /* never_see_null= */ true);
5009
5010 region->init_req(1, control());
5011 phi->init_req(1, refined_klass);
5012
5013 set_control(_gvn.transform(region));
5014 return _gvn.transform(phi);
5015 }
5016
5017 // Load the non-refined array klass from an ObjArrayKlass.
5018 Node* LibraryCallKit::load_non_refined_array_klass(Node* klass_node) {
5019 const TypeAryKlassPtr* ary_klass_ptr = _gvn.type(klass_node)->isa_aryklassptr();
5020 if (ary_klass_ptr != nullptr && ary_klass_ptr->klass_is_exact()) {
5021 return _gvn.makecon(ary_klass_ptr->cast_to_refined_array_klass_ptr(false));
5022 }
5023
5024 RegionNode* region = new RegionNode(2);
5025 Node* phi = new PhiNode(region, TypeInstKlassPtr::OBJECT);
5026
5027 generate_typeArray_guard(klass_node, region);
5028 if (region->req() == 3) {
5029 phi->add_req(klass_node);
5030 }
5031 Node* super_adr = basic_plus_adr(top(), klass_node, in_bytes(Klass::super_offset()));
5032 Node* super_klass = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), super_adr, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT));
5033
5034 region->init_req(1, control());
5035 phi->init_req(1, super_klass);
5036
5037 set_control(_gvn.transform(region));
5038 return _gvn.transform(phi);
5039 }
5040
5041 //-----------------------inline_native_newArray--------------------------
5042 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
5043 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
5044 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
5045 Node* mirror;
5046 Node* count_val;
5047 if (uninitialized) {
5048 null_check_receiver();
5049 mirror = argument(1);
5050 count_val = argument(2);
5051 } else {
5052 mirror = argument(0);
5053 count_val = argument(1);
5054 }
5055
5056 mirror = null_check(mirror);
5057 // If mirror or obj is dead, only null-path is taken.
5058 if (stopped()) return true;
5059
5060 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
5061 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5062 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5063 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
5064 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5065
5066 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
5067 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
5068 result_reg, _slow_path);
5069 Node* normal_ctl = control();
5070 Node* no_array_ctl = result_reg->in(_slow_path);
5071
5072 // Generate code for the slow case. We make a call to newArray().
5073 set_control(no_array_ctl);
5074 if (!stopped()) {
5075 // Either the input type is void.class, or else the
5076 // array klass has not yet been cached. Either the
5077 // ensuing call will throw an exception, or else it
5078 // will cache the array klass for next time.
5079 PreserveJVMState pjvms(this);
5080 CallJavaNode* slow_call = nullptr;
5081 if (uninitialized) {
5082 // Generate optimized virtual call (holder class 'Unsafe' is final)
5083 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
5084 } else {
5085 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
5086 }
5087 Node* slow_result = set_results_for_java_call(slow_call);
5088 // this->control() comes from set_results_for_java_call
5089 result_reg->set_req(_slow_path, control());
5090 result_val->set_req(_slow_path, slow_result);
5091 result_io ->set_req(_slow_path, i_o());
5092 result_mem->set_req(_slow_path, reset_memory());
5093 }
5094
5095 set_control(normal_ctl);
5096 if (!stopped()) {
5097 // Normal case: The array type has been cached in the java.lang.Class.
5098 // The following call works fine even if the array type is polymorphic.
5099 // It could be a dynamic mix of int[], boolean[], Object[], etc.
5100
5101 klass_node = load_default_refined_array_klass(klass_node);
5102
5103 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
5104 result_reg->init_req(_normal_path, control());
5105 result_val->init_req(_normal_path, obj);
5106 result_io ->init_req(_normal_path, i_o());
5107 result_mem->init_req(_normal_path, reset_memory());
5108
5109 if (uninitialized) {
5110 // Mark the allocation so that zeroing is skipped
5111 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
5112 alloc->maybe_set_complete(&_gvn);
5113 }
5114 }
5115
5116 // Return the combined state.
5117 set_i_o( _gvn.transform(result_io) );
5118 set_all_memory( _gvn.transform(result_mem));
5119
5120 C->set_has_split_ifs(true); // Has chance for split-if optimization
5121 set_result(result_reg, result_val);
5122 return true;
5123 }
5124
5125 //----------------------inline_native_getLength--------------------------
5126 // public static native int java.lang.reflect.Array.getLength(Object array);
5127 bool LibraryCallKit::inline_native_getLength() {
5128 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
5129
5130 Node* array = null_check(argument(0));
5131 // If array is dead, only null-path is taken.
5132 if (stopped()) return true;
5133
5134 // Deoptimize if it is a non-array.
5135 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
5136
5137 if (non_array != nullptr) {
5138 PreserveJVMState pjvms(this);
5139 set_control(non_array);
5140 uncommon_trap(Deoptimization::Reason_intrinsic,
5141 Deoptimization::Action_maybe_recompile);
5142 }
5143
5144 // If control is dead, only non-array-path is taken.
5145 if (stopped()) return true;
5146
5147 // The works fine even if the array type is polymorphic.
5148 // It could be a dynamic mix of int[], boolean[], Object[], etc.
5149 Node* result = load_array_length(array);
5150
5151 C->set_has_split_ifs(true); // Has chance for split-if optimization
5152 set_result(result);
5153 return true;
5154 }
5155
5156 //------------------------inline_array_copyOf----------------------------
5157 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
5158 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
5159 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
5160 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
5161
5162 // Get the arguments.
5163 Node* original = argument(0);
5164 Node* start = is_copyOfRange? argument(1): intcon(0);
5165 Node* end = is_copyOfRange? argument(2): argument(1);
5166 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
5167
5168 Node* newcopy = nullptr;
5169
5170 // Set the original stack and the reexecute bit for the interpreter to reexecute
5171 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
5172 { PreserveReexecuteState preexecs(this);
5173 jvms()->set_should_reexecute(true);
5174
5175 array_type_mirror = null_check(array_type_mirror);
5176 original = null_check(original);
5177
5178 // Check if a null path was taken unconditionally.
5179 if (stopped()) return true;
5180
5181 Node* orig_length = load_array_length(original);
5182
5183 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
5184 klass_node = null_check(klass_node);
5185
5186 RegionNode* bailout = new RegionNode(1);
5187 record_for_igvn(bailout);
5188
5189 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
5190 // Bail out if that is so.
5191 // Inline type array may have object field that would require a
5192 // write barrier. Conservatively, go to slow path.
5193 // TODO 8251971: Optimize for the case when flat src/dst are later found
5194 // to not contain oops (i.e., move this check to the macro expansion phase).
5195 // TODO 8382226: Revisit for flat abstract value class arrays
5196 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5197 const TypeAryPtr* orig_t = _gvn.type(original)->isa_aryptr();
5198 const TypeKlassPtr* tklass = _gvn.type(klass_node)->is_klassptr();
5199 bool exclude_flat = UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing) &&
5200 // Can src array be flat and contain oops?
5201 (orig_t == nullptr || (!orig_t->is_not_flat() && (!orig_t->is_flat() || orig_t->elem()->inline_klass()->contains_oops()))) &&
5202 // Can dest array be flat and contain oops?
5203 tklass->can_be_inline_array() && (!tklass->is_flat() || tklass->is_aryklassptr()->elem()->is_instklassptr()->instance_klass()->as_inline_klass()->contains_oops());
5204 Node* not_objArray = exclude_flat ? generate_non_refArray_guard(klass_node, bailout) : generate_typeArray_guard(klass_node, bailout);
5205
5206 Node* refined_klass_node = load_default_refined_array_klass(klass_node, /* type_array_guard= */ false);
5207
5208 if (not_objArray != nullptr) {
5209 // Improve the klass node's type from the new optimistic assumption:
5210 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
5211 bool not_flat = !UseArrayFlattening;
5212 bool not_null_free = !Arguments::is_valhalla_enabled();
5213 const Type* akls = TypeAryKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0), Type::trust_interfaces, not_flat, not_null_free, false, false, not_flat, true);
5214 Node* cast = new CastPPNode(control(), refined_klass_node, akls);
5215 refined_klass_node = _gvn.transform(cast);
5216 }
5217
5218 // Bail out if either start or end is negative.
5219 generate_negative_guard(start, bailout, &start);
5220 generate_negative_guard(end, bailout, &end);
5221
5222 Node* length = end;
5223 if (_gvn.type(start) != TypeInt::ZERO) {
5224 length = _gvn.transform(new SubINode(end, start));
5225 }
5226
5227 // Bail out if length is negative (i.e., if start > end).
5228 // Without this the new_array would throw
5229 // NegativeArraySizeException but IllegalArgumentException is what
5230 // should be thrown
5231 generate_negative_guard(length, bailout, &length);
5232
5233 // Handle inline type arrays
5234 // TODO 8251971 This is too strong
5235 generate_fair_guard(flat_array_test(load_object_klass(original)), bailout);
5236 generate_fair_guard(flat_array_test(refined_klass_node), bailout);
5237 generate_fair_guard(null_free_array_test(original), bailout);
5238
5239 // Bail out if start is larger than the original length
5240 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
5241 generate_negative_guard(orig_tail, bailout, &orig_tail);
5242
5243 if (bailout->req() > 1) {
5244 PreserveJVMState pjvms(this);
5245 set_control(_gvn.transform(bailout));
5246 uncommon_trap(Deoptimization::Reason_intrinsic,
5247 Deoptimization::Action_maybe_recompile);
5248 }
5249
5250 if (!stopped()) {
5251 // How many elements will we copy from the original?
5252 // The answer is MinI(orig_tail, length).
5253 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
5254
5255 // Generate a direct call to the right arraycopy function(s).
5256 // We know the copy is disjoint but we might not know if the
5257 // oop stores need checking.
5258 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
5259 // This will fail a store-check if x contains any non-nulls.
5260
5261 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
5262 // loads/stores but it is legal only if we're sure the
5263 // Arrays.copyOf would succeed. So we need all input arguments
5264 // to the copyOf to be validated, including that the copy to the
5265 // new array won't trigger an ArrayStoreException. That subtype
5266 // check can be optimized if we know something on the type of
5267 // the input array from type speculation.
5268 if (_gvn.type(klass_node)->singleton()) {
5269 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
5270 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
5271
5272 int test = C->static_subtype_check(superk, subk);
5273 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
5274 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
5275 if (t_original->speculative_type() != nullptr) {
5276 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
5277 }
5278 }
5279 }
5280
5281 bool validated = false;
5282 // Reason_class_check rather than Reason_intrinsic because we
5283 // want to intrinsify even if this traps.
5284 if (!too_many_traps(Deoptimization::Reason_class_check)) {
5285 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
5286
5287 if (not_subtype_ctrl != top()) {
5288 PreserveJVMState pjvms(this);
5289 set_control(not_subtype_ctrl);
5290 uncommon_trap(Deoptimization::Reason_class_check,
5291 Deoptimization::Action_make_not_entrant);
5292 assert(stopped(), "Should be stopped");
5293 }
5294 validated = true;
5295 }
5296
5297 if (!stopped()) {
5298 newcopy = new_array(refined_klass_node, length, 0); // no arguments to push
5299
5300 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
5301 load_object_klass(original), klass_node);
5302 if (!is_copyOfRange) {
5303 ac->set_copyof(validated);
5304 } else {
5305 ac->set_copyofrange(validated);
5306 }
5307 Node* n = _gvn.transform(ac);
5308 if (n == ac) {
5309 ac->connect_outputs(this);
5310 } else {
5311 assert(validated, "shouldn't transform if all arguments not validated");
5312 set_all_memory(n);
5313 }
5314 }
5315 }
5316 } // original reexecute is set back here
5317
5318 C->set_has_split_ifs(true); // Has chance for split-if optimization
5319 if (!stopped()) {
5320 set_result(newcopy);
5321 }
5322 return true;
5323 }
5324
5325
5326 //----------------------generate_virtual_guard---------------------------
5327 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
5328 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
5329 RegionNode* slow_region) {
5330 ciMethod* method = callee();
5331 int vtable_index = method->vtable_index();
5332 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5333 "bad index %d", vtable_index);
5334 // Get the Method* out of the appropriate vtable entry.
5335 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
5336 vtable_index*vtableEntry::size_in_bytes() +
5337 in_bytes(vtableEntry::method_offset());
5338 Node* entry_addr = off_heap_plus_addr(obj_klass, entry_offset);
5339 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
5340
5341 // Compare the target method with the expected method (e.g., Object.hashCode).
5342 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
5343
5344 Node* native_call = makecon(native_call_addr);
5345 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
5346 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
5347
5348 return generate_slow_guard(test_native, slow_region);
5349 }
5350
5351 //-----------------------generate_method_call----------------------------
5352 // Use generate_method_call to make a slow-call to the real
5353 // method if the fast path fails. An alternative would be to
5354 // use a stub like OptoRuntime::slow_arraycopy_Java.
5355 // This only works for expanding the current library call,
5356 // not another intrinsic. (E.g., don't use this for making an
5357 // arraycopy call inside of the copyOf intrinsic.)
5358 CallJavaNode*
5359 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
5360 // When compiling the intrinsic method itself, do not use this technique.
5361 guarantee(callee() != C->method(), "cannot make slow-call to self");
5362
5363 ciMethod* method = callee();
5364 // ensure the JVMS we have will be correct for this call
5365 guarantee(method_id == method->intrinsic_id(), "must match");
5366
5367 const TypeFunc* tf = TypeFunc::make(method);
5368 if (res_not_null) {
5369 assert(tf->return_type() == T_OBJECT, "");
5370 const TypeTuple* range = tf->range_cc();
5371 const Type** fields = TypeTuple::fields(range->cnt());
5372 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
5373 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
5374 tf = TypeFunc::make(tf->domain_cc(), new_range);
5375 }
5376 CallJavaNode* slow_call;
5377 if (is_static) {
5378 assert(!is_virtual, "");
5379 slow_call = new CallStaticJavaNode(C, tf,
5380 SharedRuntime::get_resolve_static_call_stub(), method);
5381 } else if (is_virtual) {
5382 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5383 int vtable_index = Method::invalid_vtable_index;
5384 if (UseInlineCaches) {
5385 // Suppress the vtable call
5386 } else {
5387 // hashCode and clone are not a miranda methods,
5388 // so the vtable index is fixed.
5389 // No need to use the linkResolver to get it.
5390 vtable_index = method->vtable_index();
5391 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5392 "bad index %d", vtable_index);
5393 }
5394 slow_call = new CallDynamicJavaNode(tf,
5395 SharedRuntime::get_resolve_virtual_call_stub(),
5396 method, vtable_index);
5397 } else { // neither virtual nor static: opt_virtual
5398 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5399 slow_call = new CallStaticJavaNode(C, tf,
5400 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
5401 slow_call->set_optimized_virtual(true);
5402 }
5403 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
5404 // To be able to issue a direct call (optimized virtual or virtual)
5405 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
5406 // about the method being invoked should be attached to the call site to
5407 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
5408 slow_call->set_override_symbolic_info(true);
5409 }
5410 set_arguments_for_java_call(slow_call);
5411 set_edges_for_java_call(slow_call);
5412 return slow_call;
5413 }
5414
5415
5416 /**
5417 * Build special case code for calls to hashCode on an object. This call may
5418 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
5419 * slightly different code.
5420 */
5421 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
5422 assert(is_static == callee()->is_static(), "correct intrinsic selection");
5423 assert(!(is_virtual && is_static), "either virtual, special, or static");
5424
5425 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
5426
5427 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5428 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
5429 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
5430 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5431 Node* obj = argument(0);
5432
5433 // Don't intrinsify hashcode on inline types for now.
5434 // The "is locked" runtime check also subsumes the inline type check (as inline types cannot be locked) and goes to the slow path.
5435 if (gvn().type(obj)->is_inlinetypeptr()) {
5436 return false;
5437 }
5438
5439 if (!is_static) {
5440 // Check for hashing null object
5441 obj = null_check_receiver();
5442 if (stopped()) return true; // unconditionally null
5443 result_reg->init_req(_null_path, top());
5444 result_val->init_req(_null_path, top());
5445 } else {
5446 // Do a null check, and return zero if null.
5447 // System.identityHashCode(null) == 0
5448 Node* null_ctl = top();
5449 obj = null_check_oop(obj, &null_ctl);
5450 result_reg->init_req(_null_path, null_ctl);
5451 result_val->init_req(_null_path, _gvn.intcon(0));
5452 }
5453
5454 // Unconditionally null? Then return right away.
5455 if (stopped()) {
5456 set_control( result_reg->in(_null_path));
5457 if (!stopped())
5458 set_result(result_val->in(_null_path));
5459 return true;
5460 }
5461
5462 // We only go to the fast case code if we pass a number of guards. The
5463 // paths which do not pass are accumulated in the slow_region.
5464 RegionNode* slow_region = new RegionNode(1);
5465 record_for_igvn(slow_region);
5466
5467 // If this is a virtual call, we generate a funny guard. We pull out
5468 // the vtable entry corresponding to hashCode() from the target object.
5469 // If the target method which we are calling happens to be the native
5470 // Object hashCode() method, we pass the guard. We do not need this
5471 // guard for non-virtual calls -- the caller is known to be the native
5472 // Object hashCode().
5473 if (is_virtual) {
5474 // After null check, get the object's klass.
5475 Node* obj_klass = load_object_klass(obj);
5476 generate_virtual_guard(obj_klass, slow_region);
5477 }
5478
5479 // Get the header out of the object, use LoadMarkNode when available
5480 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
5481 // The control of the load must be null. Otherwise, the load can move before
5482 // the null check after castPP removal.
5483 Node* no_ctrl = nullptr;
5484 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
5485
5486 if (!UseObjectMonitorTable) {
5487 // Test the header to see if it is safe to read w.r.t. locking.
5488 // We cannot use the inline type mask as this may check bits that are overridden
5489 // by an object monitor's pointer when inflating locking.
5490 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
5491 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
5492 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
5493 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
5494 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
5495
5496 generate_slow_guard(test_monitor, slow_region);
5497 }
5498
5499 // Get the hash value and check to see that it has been properly assigned.
5500 // We depend on hash_mask being at most 32 bits and avoid the use of
5501 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
5502 // vm: see markWord.hpp.
5503 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
5504 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
5505 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
5506 // This hack lets the hash bits live anywhere in the mark object now, as long
5507 // as the shift drops the relevant bits into the low 32 bits. Note that
5508 // Java spec says that HashCode is an int so there's no point in capturing
5509 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
5510 hshifted_header = ConvX2I(hshifted_header);
5511 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
5512
5513 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
5514 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
5515 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
5516
5517 generate_slow_guard(test_assigned, slow_region);
5518
5519 Node* init_mem = reset_memory();
5520 // fill in the rest of the null path:
5521 result_io ->init_req(_null_path, i_o());
5522 result_mem->init_req(_null_path, init_mem);
5523
5524 result_val->init_req(_fast_path, hash_val);
5525 result_reg->init_req(_fast_path, control());
5526 result_io ->init_req(_fast_path, i_o());
5527 result_mem->init_req(_fast_path, init_mem);
5528
5529 // Generate code for the slow case. We make a call to hashCode().
5530 set_control(_gvn.transform(slow_region));
5531 if (!stopped()) {
5532 // No need for PreserveJVMState, because we're using up the present state.
5533 set_all_memory(init_mem);
5534 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
5535 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
5536 Node* slow_result = set_results_for_java_call(slow_call);
5537 // this->control() comes from set_results_for_java_call
5538 result_reg->init_req(_slow_path, control());
5539 result_val->init_req(_slow_path, slow_result);
5540 result_io ->set_req(_slow_path, i_o());
5541 result_mem ->set_req(_slow_path, reset_memory());
5542 }
5543
5544 // Return the combined state.
5545 set_i_o( _gvn.transform(result_io) );
5546 set_all_memory( _gvn.transform(result_mem));
5547
5548 set_result(result_reg, result_val);
5549 return true;
5550 }
5551
5552 //---------------------------inline_native_getClass----------------------------
5553 // public final native Class<?> java.lang.Object.getClass();
5554 //
5555 // Build special case code for calls to getClass on an object.
5556 bool LibraryCallKit::inline_native_getClass() {
5557 Node* obj = argument(0);
5558 if (obj->is_InlineType()) {
5559 const Type* t = _gvn.type(obj);
5560 if (t->maybe_null()) {
5561 null_check(obj);
5562 }
5563 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror())));
5564 return true;
5565 }
5566 obj = null_check_receiver();
5567 if (stopped()) return true;
5568 set_result(load_mirror_from_klass(load_object_klass(obj)));
5569 return true;
5570 }
5571
5572 //-----------------inline_native_Reflection_getCallerClass---------------------
5573 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
5574 //
5575 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
5576 //
5577 // NOTE: This code must perform the same logic as JVM_GetCallerClass
5578 // in that it must skip particular security frames and checks for
5579 // caller sensitive methods.
5580 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
5581 #ifndef PRODUCT
5582 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5583 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
5584 }
5585 #endif
5586
5587 if (!jvms()->has_method()) {
5588 #ifndef PRODUCT
5589 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5590 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
5591 }
5592 #endif
5593 return false;
5594 }
5595
5596 // Walk back up the JVM state to find the caller at the required
5597 // depth.
5598 JVMState* caller_jvms = jvms();
5599
5600 // Cf. JVM_GetCallerClass
5601 // NOTE: Start the loop at depth 1 because the current JVM state does
5602 // not include the Reflection.getCallerClass() frame.
5603 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
5604 ciMethod* m = caller_jvms->method();
5605 switch (n) {
5606 case 0:
5607 fatal("current JVM state does not include the Reflection.getCallerClass frame");
5608 break;
5609 case 1:
5610 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
5611 if (!m->caller_sensitive()) {
5612 #ifndef PRODUCT
5613 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5614 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
5615 }
5616 #endif
5617 return false; // bail-out; let JVM_GetCallerClass do the work
5618 }
5619 break;
5620 default:
5621 if (!m->is_ignored_by_security_stack_walk()) {
5622 // We have reached the desired frame; return the holder class.
5623 // Acquire method holder as java.lang.Class and push as constant.
5624 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
5625 ciInstance* caller_mirror = caller_klass->java_mirror();
5626 set_result(makecon(TypeInstPtr::make(caller_mirror)));
5627
5628 #ifndef PRODUCT
5629 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5630 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());
5631 tty->print_cr(" JVM state at this point:");
5632 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5633 ciMethod* m = jvms()->of_depth(i)->method();
5634 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5635 }
5636 }
5637 #endif
5638 return true;
5639 }
5640 break;
5641 }
5642 }
5643
5644 #ifndef PRODUCT
5645 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5646 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
5647 tty->print_cr(" JVM state at this point:");
5648 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5649 ciMethod* m = jvms()->of_depth(i)->method();
5650 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5651 }
5652 }
5653 #endif
5654
5655 return false; // bail-out; let JVM_GetCallerClass do the work
5656 }
5657
5658 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
5659 Node* arg = argument(0);
5660 Node* result = nullptr;
5661
5662 switch (id) {
5663 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
5664 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
5665 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5666 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5667 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5668 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5669
5670 case vmIntrinsics::_doubleToLongBits: {
5671 // two paths (plus control) merge in a wood
5672 RegionNode *r = new RegionNode(3);
5673 Node *phi = new PhiNode(r, TypeLong::LONG);
5674
5675 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5676 // Build the boolean node
5677 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5678
5679 // Branch either way.
5680 // NaN case is less traveled, which makes all the difference.
5681 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5682 Node *opt_isnan = _gvn.transform(ifisnan);
5683 assert( opt_isnan->is_If(), "Expect an IfNode");
5684 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5685 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5686
5687 set_control(iftrue);
5688
5689 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5690 Node *slow_result = longcon(nan_bits); // return NaN
5691 phi->init_req(1, _gvn.transform( slow_result ));
5692 r->init_req(1, iftrue);
5693
5694 // Else fall through
5695 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5696 set_control(iffalse);
5697
5698 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5699 r->init_req(2, iffalse);
5700
5701 // Post merge
5702 set_control(_gvn.transform(r));
5703 record_for_igvn(r);
5704
5705 C->set_has_split_ifs(true); // Has chance for split-if optimization
5706 result = phi;
5707 assert(result->bottom_type()->isa_long(), "must be");
5708 break;
5709 }
5710
5711 case vmIntrinsics::_floatToIntBits: {
5712 // two paths (plus control) merge in a wood
5713 RegionNode *r = new RegionNode(3);
5714 Node *phi = new PhiNode(r, TypeInt::INT);
5715
5716 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5717 // Build the boolean node
5718 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5719
5720 // Branch either way.
5721 // NaN case is less traveled, which makes all the difference.
5722 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5723 Node *opt_isnan = _gvn.transform(ifisnan);
5724 assert( opt_isnan->is_If(), "Expect an IfNode");
5725 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5726 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5727
5728 set_control(iftrue);
5729
5730 static const jint nan_bits = 0x7fc00000;
5731 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5732 phi->init_req(1, _gvn.transform( slow_result ));
5733 r->init_req(1, iftrue);
5734
5735 // Else fall through
5736 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5737 set_control(iffalse);
5738
5739 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5740 r->init_req(2, iffalse);
5741
5742 // Post merge
5743 set_control(_gvn.transform(r));
5744 record_for_igvn(r);
5745
5746 C->set_has_split_ifs(true); // Has chance for split-if optimization
5747 result = phi;
5748 assert(result->bottom_type()->isa_int(), "must be");
5749 break;
5750 }
5751
5752 default:
5753 fatal_unexpected_iid(id);
5754 break;
5755 }
5756 set_result(_gvn.transform(result));
5757 return true;
5758 }
5759
5760 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5761 Node* arg = argument(0);
5762 Node* result = nullptr;
5763
5764 switch (id) {
5765 case vmIntrinsics::_floatIsInfinite:
5766 result = new IsInfiniteFNode(arg);
5767 break;
5768 case vmIntrinsics::_floatIsFinite:
5769 result = new IsFiniteFNode(arg);
5770 break;
5771 case vmIntrinsics::_doubleIsInfinite:
5772 result = new IsInfiniteDNode(arg);
5773 break;
5774 case vmIntrinsics::_doubleIsFinite:
5775 result = new IsFiniteDNode(arg);
5776 break;
5777 default:
5778 fatal_unexpected_iid(id);
5779 break;
5780 }
5781 set_result(_gvn.transform(result));
5782 return true;
5783 }
5784
5785 //----------------------inline_unsafe_copyMemory-------------------------
5786 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5787
5788 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5789 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5790 const Type* base_t = gvn.type(base);
5791
5792 bool in_native = (base_t == TypePtr::NULL_PTR);
5793 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5794 bool is_mixed = !in_heap && !in_native;
5795
5796 if (is_mixed) {
5797 return true; // mixed accesses can touch both on-heap and off-heap memory
5798 }
5799 if (in_heap) {
5800 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5801 if (!is_prim_array) {
5802 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5803 // there's not enough type information available to determine proper memory slice for it.
5804 return true;
5805 }
5806 }
5807 return false;
5808 }
5809
5810 bool LibraryCallKit::inline_unsafe_copyMemory() {
5811 if (callee()->is_static()) return false; // caller must have the capability!
5812 null_check_receiver(); // null-check receiver
5813 if (stopped()) return true;
5814
5815 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5816
5817 Node* src_base = argument(1); // type: oop
5818 Node* src_off = ConvL2X(argument(2)); // type: long
5819 Node* dst_base = argument(4); // type: oop
5820 Node* dst_off = ConvL2X(argument(5)); // type: long
5821 Node* size = ConvL2X(argument(7)); // type: long
5822
5823 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5824 "fieldOffset must be byte-scaled");
5825
5826 Node* src_addr = make_unsafe_address(src_base, src_off);
5827 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5828
5829 Node* thread = _gvn.transform(new ThreadLocalNode());
5830 Node* doing_unsafe_access_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5831 BasicType doing_unsafe_access_bt = T_BYTE;
5832 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5833
5834 // update volatile field
5835 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5836
5837 int flags = RC_LEAF | RC_NO_FP;
5838
5839 const TypePtr* dst_type = TypePtr::BOTTOM;
5840
5841 // Adjust memory effects of the runtime call based on input values.
5842 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5843 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5844 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5845
5846 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5847 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5848 flags |= RC_NARROW_MEM; // narrow in memory
5849 }
5850 }
5851
5852 // Call it. Note that the length argument is not scaled.
5853 make_runtime_call(flags,
5854 OptoRuntime::fast_arraycopy_Type(),
5855 StubRoutines::unsafe_arraycopy(),
5856 "unsafe_arraycopy",
5857 dst_type,
5858 src_addr, dst_addr, size XTOP);
5859
5860 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5861
5862 return true;
5863 }
5864
5865 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5866 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5867 bool LibraryCallKit::inline_unsafe_setMemory() {
5868 if (callee()->is_static()) return false; // caller must have the capability!
5869 null_check_receiver(); // null-check receiver
5870 if (stopped()) return true;
5871
5872 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5873
5874 Node* dst_base = argument(1); // type: oop
5875 Node* dst_off = ConvL2X(argument(2)); // type: long
5876 Node* size = ConvL2X(argument(4)); // type: long
5877 Node* byte = argument(6); // type: byte
5878
5879 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5880 "fieldOffset must be byte-scaled");
5881
5882 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5883
5884 Node* thread = _gvn.transform(new ThreadLocalNode());
5885 Node* doing_unsafe_access_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5886 BasicType doing_unsafe_access_bt = T_BYTE;
5887 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5888
5889 // update volatile field
5890 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5891
5892 int flags = RC_LEAF | RC_NO_FP;
5893
5894 const TypePtr* dst_type = TypePtr::BOTTOM;
5895
5896 // Adjust memory effects of the runtime call based on input values.
5897 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5898 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5899
5900 flags |= RC_NARROW_MEM; // narrow in memory
5901 }
5902
5903 // Call it. Note that the length argument is not scaled.
5904 make_runtime_call(flags,
5905 OptoRuntime::unsafe_setmemory_Type(),
5906 StubRoutines::unsafe_setmemory(),
5907 "unsafe_setmemory",
5908 dst_type,
5909 dst_addr, size XTOP, byte);
5910
5911 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5912
5913 return true;
5914 }
5915
5916 #undef XTOP
5917
5918 //------------------------clone_coping-----------------------------------
5919 // Helper function for inline_native_clone.
5920 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5921 assert(obj_size != nullptr, "");
5922 Node* raw_obj = alloc_obj->in(1);
5923 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5924
5925 AllocateNode* alloc = nullptr;
5926 if (ReduceBulkZeroing &&
5927 // If we are implementing an array clone without knowing its source type
5928 // (can happen when compiling the array-guarded branch of a reflective
5929 // Object.clone() invocation), initialize the array within the allocation.
5930 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5931 // to a runtime clone call that assumes fully initialized source arrays.
5932 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5933 // We will be completely responsible for initializing this object -
5934 // mark Initialize node as complete.
5935 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5936 // The object was just allocated - there should be no any stores!
5937 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5938 // Mark as complete_with_arraycopy so that on AllocateNode
5939 // expansion, we know this AllocateNode is initialized by an array
5940 // copy and a StoreStore barrier exists after the array copy.
5941 alloc->initialization()->set_complete_with_arraycopy();
5942 }
5943
5944 Node* size = _gvn.transform(obj_size);
5945 access_clone(obj, alloc_obj, size, is_array);
5946
5947 // Do not let reads from the cloned object float above the arraycopy.
5948 if (alloc != nullptr) {
5949 // Do not let stores that initialize this object be reordered with
5950 // a subsequent store that would make this object accessible by
5951 // other threads.
5952 // Record what AllocateNode this StoreStore protects so that
5953 // escape analysis can go from the MemBarStoreStoreNode to the
5954 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5955 // based on the escape status of the AllocateNode.
5956 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5957 } else {
5958 insert_mem_bar(Op_MemBarCPUOrder);
5959 }
5960 }
5961
5962 //------------------------inline_native_clone----------------------------
5963 // protected native Object java.lang.Object.clone();
5964 //
5965 // Here are the simple edge cases:
5966 // null receiver => normal trap
5967 // virtual and clone was overridden => slow path to out-of-line clone
5968 // not cloneable or finalizer => slow path to out-of-line Object.clone
5969 //
5970 // The general case has two steps, allocation and copying.
5971 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5972 //
5973 // Copying also has two cases, oop arrays and everything else.
5974 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5975 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5976 //
5977 // These steps fold up nicely if and when the cloned object's klass
5978 // can be sharply typed as an object array, a type array, or an instance.
5979 //
5980 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5981 PhiNode* result_val;
5982
5983 // Set the reexecute bit for the interpreter to reexecute
5984 // the bytecode that invokes Object.clone if deoptimization happens.
5985 { PreserveReexecuteState preexecs(this);
5986 jvms()->set_should_reexecute(true);
5987
5988 Node* obj = argument(0);
5989 obj = null_check_receiver();
5990 if (stopped()) return true;
5991
5992 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5993 if (obj_type->is_inlinetypeptr()) {
5994 // If the object to clone is an inline type, we can simply return it (i.e. a nop) since inline types have
5995 // no identity.
5996 set_result(obj);
5997 return true;
5998 }
5999
6000 // If we are going to clone an instance, we need its exact type to
6001 // know the number and types of fields to convert the clone to
6002 // loads/stores. Maybe a speculative type can help us.
6003 if (!obj_type->klass_is_exact() &&
6004 obj_type->speculative_type() != nullptr &&
6005 obj_type->speculative_type()->is_instance_klass() &&
6006 !obj_type->speculative_type()->is_inlinetype()) {
6007 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
6008 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
6009 !spec_ik->has_injected_fields()) {
6010 if (!obj_type->isa_instptr() ||
6011 obj_type->is_instptr()->instance_klass()->has_subklass()) {
6012 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
6013 }
6014 }
6015 }
6016
6017 // Conservatively insert a memory barrier on all memory slices.
6018 // Do not let writes into the original float below the clone.
6019 insert_mem_bar(Op_MemBarCPUOrder);
6020
6021 // paths into result_reg:
6022 enum {
6023 _slow_path = 1, // out-of-line call to clone method (virtual or not)
6024 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
6025 _array_path, // plain array allocation, plus arrayof_long_arraycopy
6026 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
6027 PATH_LIMIT
6028 };
6029 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
6030 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
6031 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
6032 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
6033 record_for_igvn(result_reg);
6034
6035 Node* obj_klass = load_object_klass(obj);
6036 // We only go to the fast case code if we pass a number of guards.
6037 // The paths which do not pass are accumulated in the slow_region.
6038 RegionNode* slow_region = new RegionNode(1);
6039 record_for_igvn(slow_region);
6040
6041 Node* array_obj = obj;
6042 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
6043 if (array_ctl != nullptr) {
6044 // It's an array.
6045 PreserveJVMState pjvms(this);
6046 set_control(array_ctl);
6047
6048 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
6049 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr();
6050 if (UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) &&
6051 obj_type->can_be_inline_array() &&
6052 (ary_ptr == nullptr || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) {
6053 // Flat inline type array may have object field that would require a
6054 // write barrier. Conservatively, go to slow path.
6055 generate_fair_guard(flat_array_test(obj_klass), slow_region);
6056 }
6057
6058 if (!stopped()) {
6059 Node* obj_length = load_array_length(array_obj);
6060 Node* array_size = nullptr; // Size of the array without object alignment padding.
6061 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
6062
6063 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
6064 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
6065 // If it is an oop array, it requires very special treatment,
6066 // because gc barriers are required when accessing the array.
6067 Node* is_obja = generate_refArray_guard(obj_klass, (RegionNode*)nullptr);
6068 if (is_obja != nullptr) {
6069 PreserveJVMState pjvms2(this);
6070 set_control(is_obja);
6071 // Generate a direct call to the right arraycopy function(s).
6072 // Clones are always tightly coupled.
6073 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
6074 ac->set_clone_oop_array();
6075 Node* n = _gvn.transform(ac);
6076 assert(n == ac, "cannot disappear");
6077 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
6078
6079 result_reg->init_req(_objArray_path, control());
6080 result_val->init_req(_objArray_path, alloc_obj);
6081 result_i_o ->set_req(_objArray_path, i_o());
6082 result_mem ->set_req(_objArray_path, reset_memory());
6083 }
6084 }
6085 // Otherwise, there are no barriers to worry about.
6086 // (We can dispense with card marks if we know the allocation
6087 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
6088 // causes the non-eden paths to take compensating steps to
6089 // simulate a fresh allocation, so that no further
6090 // card marks are required in compiled code to initialize
6091 // the object.)
6092
6093 if (!stopped()) {
6094 copy_to_clone(obj, alloc_obj, array_size, true);
6095
6096 // Present the results of the copy.
6097 result_reg->init_req(_array_path, control());
6098 result_val->init_req(_array_path, alloc_obj);
6099 result_i_o ->set_req(_array_path, i_o());
6100 result_mem ->set_req(_array_path, reset_memory());
6101 }
6102 }
6103 }
6104
6105 if (!stopped()) {
6106 // It's an instance (we did array above). Make the slow-path tests.
6107 // If this is a virtual call, we generate a funny guard. We grab
6108 // the vtable entry corresponding to clone() from the target object.
6109 // If the target method which we are calling happens to be the
6110 // Object clone() method, we pass the guard. We do not need this
6111 // guard for non-virtual calls; the caller is known to be the native
6112 // Object clone().
6113 if (is_virtual) {
6114 generate_virtual_guard(obj_klass, slow_region);
6115 }
6116
6117 // The object must be easily cloneable and must not have a finalizer.
6118 // Both of these conditions may be checked in a single test.
6119 // We could optimize the test further, but we don't care.
6120 generate_misc_flags_guard(obj_klass,
6121 // Test both conditions:
6122 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
6123 // Must be cloneable but not finalizer:
6124 KlassFlags::_misc_is_cloneable_fast,
6125 slow_region);
6126 }
6127
6128 if (!stopped()) {
6129 // It's an instance, and it passed the slow-path tests.
6130 PreserveJVMState pjvms(this);
6131 Node* obj_size = nullptr; // Total object size, including object alignment padding.
6132 // Need to deoptimize on exception from allocation since Object.clone intrinsic
6133 // is reexecuted if deoptimization occurs and there could be problems when merging
6134 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
6135 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
6136
6137 copy_to_clone(obj, alloc_obj, obj_size, false);
6138
6139 // Present the results of the slow call.
6140 result_reg->init_req(_instance_path, control());
6141 result_val->init_req(_instance_path, alloc_obj);
6142 result_i_o ->set_req(_instance_path, i_o());
6143 result_mem ->set_req(_instance_path, reset_memory());
6144 }
6145
6146 // Generate code for the slow case. We make a call to clone().
6147 set_control(_gvn.transform(slow_region));
6148 if (!stopped()) {
6149 PreserveJVMState pjvms(this);
6150 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
6151 // We need to deoptimize on exception (see comment above)
6152 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
6153 // this->control() comes from set_results_for_java_call
6154 result_reg->init_req(_slow_path, control());
6155 result_val->init_req(_slow_path, slow_result);
6156 result_i_o ->set_req(_slow_path, i_o());
6157 result_mem ->set_req(_slow_path, reset_memory());
6158 }
6159
6160 // Return the combined state.
6161 set_control( _gvn.transform(result_reg));
6162 set_i_o( _gvn.transform(result_i_o));
6163 set_all_memory( _gvn.transform(result_mem));
6164 } // original reexecute is set back here
6165
6166 set_result(_gvn.transform(result_val));
6167 return true;
6168 }
6169
6170 // If we have a tightly coupled allocation, the arraycopy may take care
6171 // of the array initialization. If one of the guards we insert between
6172 // the allocation and the arraycopy causes a deoptimization, an
6173 // uninitialized array will escape the compiled method. To prevent that
6174 // we set the JVM state for uncommon traps between the allocation and
6175 // the arraycopy to the state before the allocation so, in case of
6176 // deoptimization, we'll reexecute the allocation and the
6177 // initialization.
6178 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
6179 if (alloc != nullptr) {
6180 ciMethod* trap_method = alloc->jvms()->method();
6181 int trap_bci = alloc->jvms()->bci();
6182
6183 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6184 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
6185 // Make sure there's no store between the allocation and the
6186 // arraycopy otherwise visible side effects could be rexecuted
6187 // in case of deoptimization and cause incorrect execution.
6188 bool no_interfering_store = true;
6189 Node* mem = alloc->in(TypeFunc::Memory);
6190 if (mem->is_MergeMem()) {
6191 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
6192 Node* n = mms.memory();
6193 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
6194 assert(n->is_Store(), "what else?");
6195 no_interfering_store = false;
6196 break;
6197 }
6198 }
6199 } else {
6200 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
6201 Node* n = mms.memory();
6202 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
6203 assert(n->is_Store(), "what else?");
6204 no_interfering_store = false;
6205 break;
6206 }
6207 }
6208 }
6209
6210 if (no_interfering_store) {
6211 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6212
6213 JVMState* saved_jvms = jvms();
6214 saved_reexecute_sp = _reexecute_sp;
6215
6216 set_jvms(sfpt->jvms());
6217 _reexecute_sp = jvms()->sp();
6218
6219 return saved_jvms;
6220 }
6221 }
6222 }
6223 return nullptr;
6224 }
6225
6226 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
6227 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
6228 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
6229 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
6230 uint size = alloc->req();
6231 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
6232 old_jvms->set_map(sfpt);
6233 for (uint i = 0; i < size; i++) {
6234 sfpt->init_req(i, alloc->in(i));
6235 }
6236 int adjustment = 1;
6237 const TypeAryKlassPtr* ary_klass_ptr = alloc->in(AllocateNode::KlassNode)->bottom_type()->is_aryklassptr();
6238 if (ary_klass_ptr->is_null_free()) {
6239 // A null-free, tightly coupled array allocation can only come from LibraryCallKit::inline_newArray which
6240 // also requires the componentType and initVal on stack for re-execution.
6241 // Re-create and push the componentType.
6242 ciArrayKlass* klass = ary_klass_ptr->exact_klass()->as_array_klass();
6243 ciInstance* instance = klass->component_mirror_instance();
6244 const TypeInstPtr* t_instance = TypeInstPtr::make(instance);
6245 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), makecon(t_instance));
6246 adjustment++;
6247 }
6248 // re-push array length for deoptimization
6249 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment - 1, alloc->in(AllocateNode::ALength));
6250 if (ary_klass_ptr->is_null_free()) {
6251 // Re-create and push the initVal.
6252 Node* init_val = alloc->in(AllocateNode::InitValue);
6253 if (init_val == nullptr) {
6254 init_val = InlineTypeNode::make_all_zero(_gvn, ary_klass_ptr->elem()->is_instklassptr()->instance_klass()->as_inline_klass());
6255 } else if (UseCompressedOops) {
6256 init_val = _gvn.transform(new DecodeNNode(init_val, init_val->bottom_type()->make_ptr()));
6257 }
6258 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment, init_val);
6259 adjustment++;
6260 }
6261 old_jvms->set_sp(old_jvms->sp() + adjustment);
6262 old_jvms->set_monoff(old_jvms->monoff() + adjustment);
6263 old_jvms->set_scloff(old_jvms->scloff() + adjustment);
6264 old_jvms->set_endoff(old_jvms->endoff() + adjustment);
6265 old_jvms->set_should_reexecute(true);
6266
6267 sfpt->set_i_o(map()->i_o());
6268 sfpt->set_memory(map()->memory());
6269 sfpt->set_control(map()->control());
6270 return sfpt;
6271 }
6272
6273 // In case of a deoptimization, we restart execution at the
6274 // allocation, allocating a new array. We would leave an uninitialized
6275 // array in the heap that GCs wouldn't expect. Move the allocation
6276 // after the traps so we don't allocate the array if we
6277 // deoptimize. This is possible because tightly_coupled_allocation()
6278 // guarantees there's no observer of the allocated array at this point
6279 // and the control flow is simple enough.
6280 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
6281 int saved_reexecute_sp, uint new_idx) {
6282 if (saved_jvms_before_guards != nullptr && !stopped()) {
6283 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
6284
6285 assert(alloc != nullptr, "only with a tightly coupled allocation");
6286 // restore JVM state to the state at the arraycopy
6287 saved_jvms_before_guards->map()->set_control(map()->control());
6288 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
6289 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
6290 // If we've improved the types of some nodes (null check) while
6291 // emitting the guards, propagate them to the current state
6292 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
6293 set_jvms(saved_jvms_before_guards);
6294 _reexecute_sp = saved_reexecute_sp;
6295
6296 // Remove the allocation from above the guards
6297 CallProjections* callprojs = alloc->extract_projections(true);
6298 InitializeNode* init = alloc->initialization();
6299 Node* alloc_mem = alloc->in(TypeFunc::Memory);
6300 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
6301 init->replace_mem_projs_by(alloc_mem, C);
6302
6303 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
6304 // the allocation (i.e. is only valid if the allocation succeeds):
6305 // 1) replace CastIINode with AllocateArrayNode's length here
6306 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
6307 //
6308 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
6309 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
6310 Node* init_control = init->proj_out(TypeFunc::Control);
6311 Node* alloc_length = alloc->Ideal_length();
6312 #ifdef ASSERT
6313 Node* prev_cast = nullptr;
6314 #endif
6315 for (uint i = 0; i < init_control->outcnt(); i++) {
6316 Node* init_out = init_control->raw_out(i);
6317 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
6318 #ifdef ASSERT
6319 if (prev_cast == nullptr) {
6320 prev_cast = init_out;
6321 } else {
6322 if (prev_cast->cmp(*init_out) == false) {
6323 prev_cast->dump();
6324 init_out->dump();
6325 assert(false, "not equal CastIINode");
6326 }
6327 }
6328 #endif
6329 C->gvn_replace_by(init_out, alloc_length);
6330 }
6331 }
6332 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
6333
6334 // move the allocation here (after the guards)
6335 _gvn.hash_delete(alloc);
6336 alloc->set_req(TypeFunc::Control, control());
6337 alloc->set_req(TypeFunc::I_O, i_o());
6338 Node *mem = reset_memory();
6339 set_all_memory(mem);
6340 alloc->set_req(TypeFunc::Memory, mem);
6341 set_control(init->proj_out_or_null(TypeFunc::Control));
6342 set_i_o(callprojs->fallthrough_ioproj);
6343
6344 // Update memory as done in GraphKit::set_output_for_allocation()
6345 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
6346 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
6347 if (ary_type->isa_aryptr() && length_type != nullptr) {
6348 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
6349 }
6350 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
6351 int elemidx = C->get_alias_index(telemref);
6352 // Need to properly move every memory projection for the Initialize
6353 #ifdef ASSERT
6354 int mark_idx = C->get_alias_index(ary_type->add_offset(oopDesc::mark_offset_in_bytes()));
6355 int klass_idx = C->get_alias_index(ary_type->add_offset(oopDesc::klass_offset_in_bytes()));
6356 #endif
6357 auto move_proj = [&](ProjNode* proj) {
6358 int alias_idx = C->get_alias_index(proj->adr_type());
6359 assert(alias_idx == Compile::AliasIdxRaw ||
6360 alias_idx == elemidx ||
6361 alias_idx == mark_idx ||
6362 alias_idx == klass_idx, "should be raw memory or array element type");
6363 set_memory(proj, alias_idx);
6364 };
6365 init->for_each_proj(move_proj, TypeFunc::Memory);
6366
6367 Node* allocx = _gvn.transform(alloc);
6368 assert(allocx == alloc, "where has the allocation gone?");
6369 assert(dest->is_CheckCastPP(), "not an allocation result?");
6370
6371 _gvn.hash_delete(dest);
6372 dest->set_req(0, control());
6373 Node* destx = _gvn.transform(dest);
6374 assert(destx == dest, "where has the allocation result gone?");
6375
6376 array_ideal_length(alloc, ary_type, true);
6377 }
6378 }
6379
6380 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
6381 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
6382 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
6383 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
6384 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
6385 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
6386 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
6387 JVMState* saved_jvms_before_guards) {
6388 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
6389 // There is at least one unrelated uncommon trap which needs to be replaced.
6390 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6391
6392 JVMState* saved_jvms = jvms();
6393 const int saved_reexecute_sp = _reexecute_sp;
6394 set_jvms(sfpt->jvms());
6395 _reexecute_sp = jvms()->sp();
6396
6397 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
6398
6399 // Restore state
6400 set_jvms(saved_jvms);
6401 _reexecute_sp = saved_reexecute_sp;
6402 }
6403 }
6404
6405 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
6406 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
6407 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
6408 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
6409 while (if_proj->is_IfProj()) {
6410 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
6411 if (uncommon_trap != nullptr) {
6412 create_new_uncommon_trap(uncommon_trap);
6413 }
6414 assert(if_proj->in(0)->is_If(), "must be If");
6415 if_proj = if_proj->in(0)->in(0);
6416 }
6417 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
6418 "must have reached control projection of init node");
6419 }
6420
6421 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
6422 const int trap_request = uncommon_trap_call->uncommon_trap_request();
6423 assert(trap_request != 0, "no valid UCT trap request");
6424 PreserveJVMState pjvms(this);
6425 set_control(uncommon_trap_call->in(0));
6426 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
6427 Deoptimization::trap_request_action(trap_request));
6428 assert(stopped(), "Should be stopped");
6429 _gvn.hash_delete(uncommon_trap_call);
6430 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
6431 }
6432
6433 // Common checks for array sorting intrinsics arguments.
6434 // Returns `true` if checks passed.
6435 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
6436 // check address of the class
6437 if (elementType == nullptr || elementType->is_top()) {
6438 return false; // dead path
6439 }
6440 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
6441 if (elem_klass == nullptr) {
6442 return false; // dead path
6443 }
6444 // java_mirror_type() returns non-null for compile-time Class constants only
6445 ciType* elem_type = elem_klass->java_mirror_type();
6446 if (elem_type == nullptr) {
6447 return false;
6448 }
6449 bt = elem_type->basic_type();
6450 // Disable the intrinsic if the CPU does not support SIMD sort
6451 if (!Matcher::supports_simd_sort(bt)) {
6452 return false;
6453 }
6454 // check address of the array
6455 if (obj == nullptr || obj->is_top()) {
6456 return false; // dead path
6457 }
6458 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
6459 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
6460 return false; // failed input validation
6461 }
6462 return true;
6463 }
6464
6465 //------------------------------inline_array_partition-----------------------
6466 bool LibraryCallKit::inline_array_partition() {
6467 address stubAddr = StubRoutines::select_array_partition_function();
6468 if (stubAddr == nullptr) {
6469 return false; // Intrinsic's stub is not implemented on this platform
6470 }
6471 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
6472
6473 // no receiver because it is a static method
6474 Node* elementType = argument(0);
6475 Node* obj = argument(1);
6476 Node* offset = argument(2); // long
6477 Node* fromIndex = argument(4);
6478 Node* toIndex = argument(5);
6479 Node* indexPivot1 = argument(6);
6480 Node* indexPivot2 = argument(7);
6481 // PartitionOperation: argument(8) is ignored
6482
6483 Node* pivotIndices = nullptr;
6484 BasicType bt = T_ILLEGAL;
6485
6486 if (!check_array_sort_arguments(elementType, obj, bt)) {
6487 return false;
6488 }
6489 null_check(obj);
6490 // If obj is dead, only null-path is taken.
6491 if (stopped()) {
6492 return true;
6493 }
6494 // Set the original stack and the reexecute bit for the interpreter to reexecute
6495 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
6496 { PreserveReexecuteState preexecs(this);
6497 jvms()->set_should_reexecute(true);
6498
6499 Node* obj_adr = make_unsafe_address(obj, offset);
6500
6501 // create the pivotIndices array of type int and size = 2
6502 Node* size = intcon(2);
6503 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
6504 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
6505 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
6506 guarantee(alloc != nullptr, "created above");
6507 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
6508
6509 // pass the basic type enum to the stub
6510 Node* elemType = intcon(bt);
6511
6512 // Call the stub
6513 const char *stubName = "array_partition_stub";
6514 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
6515 stubAddr, stubName, TypePtr::BOTTOM,
6516 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
6517 indexPivot1, indexPivot2);
6518
6519 } // original reexecute is set back here
6520
6521 if (!stopped()) {
6522 set_result(pivotIndices);
6523 }
6524
6525 return true;
6526 }
6527
6528
6529 //------------------------------inline_array_sort-----------------------
6530 bool LibraryCallKit::inline_array_sort() {
6531 address stubAddr = StubRoutines::select_arraysort_function();
6532 if (stubAddr == nullptr) {
6533 return false; // Intrinsic's stub is not implemented on this platform
6534 }
6535 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
6536
6537 // no receiver because it is a static method
6538 Node* elementType = argument(0);
6539 Node* obj = argument(1);
6540 Node* offset = argument(2); // long
6541 Node* fromIndex = argument(4);
6542 Node* toIndex = argument(5);
6543 // SortOperation: argument(6) is ignored
6544
6545 BasicType bt = T_ILLEGAL;
6546
6547 if (!check_array_sort_arguments(elementType, obj, bt)) {
6548 return false;
6549 }
6550 null_check(obj);
6551 // If obj is dead, only null-path is taken.
6552 if (stopped()) {
6553 return true;
6554 }
6555 Node* obj_adr = make_unsafe_address(obj, offset);
6556
6557 // pass the basic type enum to the stub
6558 Node* elemType = intcon(bt);
6559
6560 // Call the stub.
6561 const char *stubName = "arraysort_stub";
6562 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
6563 stubAddr, stubName, TypePtr::BOTTOM,
6564 obj_adr, elemType, fromIndex, toIndex);
6565
6566 return true;
6567 }
6568
6569
6570 //------------------------------inline_arraycopy-----------------------
6571 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
6572 // Object dest, int destPos,
6573 // int length);
6574 bool LibraryCallKit::inline_arraycopy() {
6575 // Get the arguments.
6576 Node* src = argument(0); // type: oop
6577 Node* src_offset = argument(1); // type: int
6578 Node* dest = argument(2); // type: oop
6579 Node* dest_offset = argument(3); // type: int
6580 Node* length = argument(4); // type: int
6581
6582 uint new_idx = C->unique();
6583
6584 // Check for allocation before we add nodes that would confuse
6585 // tightly_coupled_allocation()
6586 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
6587
6588 int saved_reexecute_sp = -1;
6589 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
6590 // See arraycopy_restore_alloc_state() comment
6591 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
6592 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
6593 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
6594 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
6595
6596 // The following tests must be performed
6597 // (1) src and dest are arrays.
6598 // (2) src and dest arrays must have elements of the same BasicType
6599 // (3) src and dest must not be null.
6600 // (4) src_offset must not be negative.
6601 // (5) dest_offset must not be negative.
6602 // (6) length must not be negative.
6603 // (7) src_offset + length must not exceed length of src.
6604 // (8) dest_offset + length must not exceed length of dest.
6605 // (9) each element of an oop array must be assignable
6606
6607 // (3) src and dest must not be null.
6608 // always do this here because we need the JVM state for uncommon traps
6609 Node* null_ctl = top();
6610 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
6611 assert(null_ctl->is_top(), "no null control here");
6612 dest = null_check(dest, T_ARRAY);
6613
6614 if (!can_emit_guards) {
6615 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
6616 // guards but the arraycopy node could still take advantage of a
6617 // tightly allocated allocation. tightly_coupled_allocation() is
6618 // called again to make sure it takes the null check above into
6619 // account: the null check is mandatory and if it caused an
6620 // uncommon trap to be emitted then the allocation can't be
6621 // considered tightly coupled in this context.
6622 alloc = tightly_coupled_allocation(dest);
6623 }
6624
6625 bool validated = false;
6626
6627 const Type* src_type = _gvn.type(src);
6628 const Type* dest_type = _gvn.type(dest);
6629 const TypeAryPtr* top_src = src_type->isa_aryptr();
6630 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6631
6632 // Do we have the type of src?
6633 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6634 // Do we have the type of dest?
6635 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6636 // Is the type for src from speculation?
6637 bool src_spec = false;
6638 // Is the type for dest from speculation?
6639 bool dest_spec = false;
6640
6641 if ((!has_src || !has_dest) && can_emit_guards) {
6642 // We don't have sufficient type information, let's see if
6643 // speculative types can help. We need to have types for both src
6644 // and dest so that it pays off.
6645
6646 // Do we already have or could we have type information for src
6647 bool could_have_src = has_src;
6648 // Do we already have or could we have type information for dest
6649 bool could_have_dest = has_dest;
6650
6651 ciKlass* src_k = nullptr;
6652 if (!has_src) {
6653 src_k = src_type->speculative_type_not_null();
6654 if (src_k != nullptr && src_k->is_array_klass()) {
6655 could_have_src = true;
6656 }
6657 }
6658
6659 ciKlass* dest_k = nullptr;
6660 if (!has_dest) {
6661 dest_k = dest_type->speculative_type_not_null();
6662 if (dest_k != nullptr && dest_k->is_array_klass()) {
6663 could_have_dest = true;
6664 }
6665 }
6666
6667 if (could_have_src && could_have_dest) {
6668 // This is going to pay off so emit the required guards
6669 if (!has_src) {
6670 src = maybe_cast_profiled_obj(src, src_k, true);
6671 src_type = _gvn.type(src);
6672 top_src = src_type->isa_aryptr();
6673 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6674 src_spec = true;
6675 }
6676 if (!has_dest) {
6677 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6678 dest_type = _gvn.type(dest);
6679 top_dest = dest_type->isa_aryptr();
6680 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6681 dest_spec = true;
6682 }
6683 }
6684 }
6685
6686 if (has_src && has_dest && can_emit_guards) {
6687 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
6688 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
6689 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
6690 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
6691
6692 if (src_elem == dest_elem && top_src->is_flat() == top_dest->is_flat() && src_elem == T_OBJECT) {
6693 // If both arrays are object arrays then having the exact types
6694 // for both will remove the need for a subtype check at runtime
6695 // before the call and may make it possible to pick a faster copy
6696 // routine (without a subtype check on every element)
6697 // Do we have the exact type of src?
6698 bool could_have_src = src_spec;
6699 // Do we have the exact type of dest?
6700 bool could_have_dest = dest_spec;
6701 ciKlass* src_k = nullptr;
6702 ciKlass* dest_k = nullptr;
6703 if (!src_spec) {
6704 src_k = src_type->speculative_type_not_null();
6705 if (src_k != nullptr && src_k->is_array_klass()) {
6706 could_have_src = true;
6707 }
6708 }
6709 if (!dest_spec) {
6710 dest_k = dest_type->speculative_type_not_null();
6711 if (dest_k != nullptr && dest_k->is_array_klass()) {
6712 could_have_dest = true;
6713 }
6714 }
6715 if (could_have_src && could_have_dest) {
6716 // If we can have both exact types, emit the missing guards
6717 if (could_have_src && !src_spec) {
6718 src = maybe_cast_profiled_obj(src, src_k, true);
6719 src_type = _gvn.type(src);
6720 top_src = src_type->isa_aryptr();
6721 }
6722 if (could_have_dest && !dest_spec) {
6723 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6724 dest_type = _gvn.type(dest);
6725 top_dest = dest_type->isa_aryptr();
6726 }
6727 }
6728 }
6729 }
6730
6731 ciMethod* trap_method = method();
6732 int trap_bci = bci();
6733 if (saved_jvms_before_guards != nullptr) {
6734 trap_method = alloc->jvms()->method();
6735 trap_bci = alloc->jvms()->bci();
6736 }
6737
6738 bool negative_length_guard_generated = false;
6739
6740 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6741 can_emit_guards && !src->is_top() && !dest->is_top()) {
6742 // validate arguments: enables transformation the ArrayCopyNode
6743 validated = true;
6744
6745 RegionNode* slow_region = new RegionNode(1);
6746 record_for_igvn(slow_region);
6747
6748 // (1) src and dest are arrays.
6749 generate_non_array_guard(load_object_klass(src), slow_region, &src);
6750 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
6751
6752 // (2) src and dest arrays must have elements of the same BasicType
6753 // done at macro expansion or at Ideal transformation time
6754
6755 // (4) src_offset must not be negative.
6756 generate_negative_guard(src_offset, slow_region);
6757
6758 // (5) dest_offset must not be negative.
6759 generate_negative_guard(dest_offset, slow_region);
6760
6761 // (7) src_offset + length must not exceed length of src.
6762 generate_limit_guard(src_offset, length,
6763 load_array_length(src),
6764 slow_region);
6765
6766 // (8) dest_offset + length must not exceed length of dest.
6767 generate_limit_guard(dest_offset, length,
6768 load_array_length(dest),
6769 slow_region);
6770
6771 // (6) length must not be negative.
6772 // This is also checked in generate_arraycopy() during macro expansion, but
6773 // we also have to check it here for the case where the ArrayCopyNode will
6774 // be eliminated by Escape Analysis.
6775 if (EliminateAllocations) {
6776 generate_negative_guard(length, slow_region);
6777 negative_length_guard_generated = true;
6778 }
6779
6780 // (9) each element of an oop array must be assignable
6781 Node* dest_klass = load_object_klass(dest);
6782 Node* refined_dest_klass = dest_klass;
6783 if (src != dest) {
6784 dest_klass = load_non_refined_array_klass(refined_dest_klass);
6785 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6786 slow_region->add_req(not_subtype_ctrl);
6787 }
6788
6789 // TODO 8251971 Improve this. What about atomicity? Make sure this is always folded for type arrays.
6790 // If destination is null-restricted, source must be null-restricted as well: src_null_restricted || !dst_null_restricted
6791 Node* src_klass = load_object_klass(src);
6792 Node* adr_prop_src = basic_plus_adr(top(), src_klass, in_bytes(ArrayKlass::properties_offset()));
6793 Node* prop_src = _gvn.transform(LoadNode::make(_gvn, control(), immutable_memory(), adr_prop_src,
6794 _gvn.type(adr_prop_src)->is_ptr(), TypeInt::INT, T_INT,
6795 MemNode::unordered));
6796 Node* adr_prop_dest = basic_plus_adr(top(), refined_dest_klass, in_bytes(ArrayKlass::properties_offset()));
6797 Node* prop_dest = _gvn.transform(LoadNode::make(_gvn, control(), immutable_memory(), adr_prop_dest,
6798 _gvn.type(adr_prop_dest)->is_ptr(), TypeInt::INT, T_INT,
6799 MemNode::unordered));
6800
6801 const ArrayProperties props_null_restricted = ArrayProperties::Default().with_null_restricted();
6802 jint props_value = (jint)props_null_restricted.value();
6803
6804 prop_dest = _gvn.transform(new XorINode(prop_dest, intcon(props_value)));
6805 prop_src = _gvn.transform(new OrINode(prop_dest, prop_src));
6806 prop_src = _gvn.transform(new AndINode(prop_src, intcon(props_value)));
6807
6808 Node* chk = _gvn.transform(new CmpINode(prop_src, intcon(props_value)));
6809 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::ne));
6810 generate_fair_guard(tst, slow_region);
6811
6812 // TODO 8251971 This is too strong
6813 generate_fair_guard(flat_array_test(src), slow_region);
6814 generate_fair_guard(flat_array_test(dest), slow_region);
6815
6816 {
6817 PreserveJVMState pjvms(this);
6818 set_control(_gvn.transform(slow_region));
6819 uncommon_trap(Deoptimization::Reason_intrinsic,
6820 Deoptimization::Action_make_not_entrant);
6821 assert(stopped(), "Should be stopped");
6822 }
6823
6824 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->isa_klassptr();
6825 if (dest_klass_t == nullptr) {
6826 // refined_dest_klass may not be an array, which leads to dest_klass being top. This means we
6827 // are in a dead path.
6828 uncommon_trap(Deoptimization::Reason_intrinsic,
6829 Deoptimization::Action_make_not_entrant);
6830 return true;
6831 }
6832
6833 const Type* toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6834 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6835 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6836 }
6837
6838 if (stopped()) {
6839 return true;
6840 }
6841
6842 Node* dest_klass = load_object_klass(dest);
6843 dest_klass = load_non_refined_array_klass(dest_klass);
6844
6845 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6846 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6847 // so the compiler has a chance to eliminate them: during macro expansion,
6848 // we have to set their control (CastPP nodes are eliminated).
6849 load_object_klass(src), dest_klass,
6850 load_array_length(src), load_array_length(dest));
6851
6852 ac->set_arraycopy(validated);
6853
6854 Node* n = _gvn.transform(ac);
6855 if (n == ac) {
6856 ac->connect_outputs(this);
6857 } else {
6858 assert(validated, "shouldn't transform if all arguments not validated");
6859 set_all_memory(n);
6860 }
6861 clear_upper_avx();
6862
6863
6864 return true;
6865 }
6866
6867
6868 // Helper function which determines if an arraycopy immediately follows
6869 // an allocation, with no intervening tests or other escapes for the object.
6870 AllocateArrayNode*
6871 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6872 if (stopped()) return nullptr; // no fast path
6873 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6874
6875 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6876 if (alloc == nullptr) return nullptr;
6877
6878 Node* rawmem = memory(Compile::AliasIdxRaw);
6879 // Is the allocation's memory state untouched?
6880 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6881 // Bail out if there have been raw-memory effects since the allocation.
6882 // (Example: There might have been a call or safepoint.)
6883 return nullptr;
6884 }
6885 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6886 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6887 return nullptr;
6888 }
6889
6890 // There must be no unexpected observers of this allocation.
6891 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6892 Node* obs = ptr->fast_out(i);
6893 if (obs != this->map()) {
6894 return nullptr;
6895 }
6896 }
6897
6898 // This arraycopy must unconditionally follow the allocation of the ptr.
6899 Node* alloc_ctl = ptr->in(0);
6900 Node* ctl = control();
6901 while (ctl != alloc_ctl) {
6902 // There may be guards which feed into the slow_region.
6903 // Any other control flow means that we might not get a chance
6904 // to finish initializing the allocated object.
6905 // Various low-level checks bottom out in uncommon traps. These
6906 // are considered safe since we've already checked above that
6907 // there is no unexpected observer of this allocation.
6908 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6909 assert(ctl->in(0)->is_If(), "must be If");
6910 ctl = ctl->in(0)->in(0);
6911 } else {
6912 return nullptr;
6913 }
6914 }
6915
6916 // If we get this far, we have an allocation which immediately
6917 // precedes the arraycopy, and we can take over zeroing the new object.
6918 // The arraycopy will finish the initialization, and provide
6919 // a new control state to which we will anchor the destination pointer.
6920
6921 return alloc;
6922 }
6923
6924 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6925 if (node->is_IfProj()) {
6926 IfProjNode* other_proj = node->as_IfProj()->other_if_proj();
6927 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6928 Node* obs = other_proj->fast_out(j);
6929 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6930 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6931 return obs->as_CallStaticJava();
6932 }
6933 }
6934 }
6935 return nullptr;
6936 }
6937
6938 //-------------inline_encodeISOArray-----------------------------------
6939 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6940 // int java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6941 // int java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6942 // encode char[] to byte[] in ISO_8859_1 or ASCII
6943 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6944 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6945 // no receiver since it is static method
6946 Node *src = argument(0);
6947 Node *src_offset = argument(1);
6948 Node *dst = argument(2);
6949 Node *dst_offset = argument(3);
6950 Node *length = argument(4);
6951
6952 // Cast source & target arrays to not-null
6953 src = must_be_not_null(src, true);
6954 dst = must_be_not_null(dst, true);
6955 if (stopped()) {
6956 return true;
6957 }
6958
6959 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6960 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6961 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6962 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6963 // failed array check
6964 return false;
6965 }
6966
6967 // Figure out the size and type of the elements we will be copying.
6968 BasicType src_elem = src_type->elem()->array_element_basic_type();
6969 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6970 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6971 return false;
6972 }
6973
6974 // Check source & target bounds
6975 RegionNode* bailout = create_bailout();
6976 generate_string_range_check(src, src_offset, length, src_elem == T_BYTE, bailout);
6977 generate_string_range_check(dst, dst_offset, length, false, bailout);
6978 if (check_bailout(bailout)) {
6979 return true;
6980 }
6981
6982 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6983 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6984 // 'src_start' points to src array + scaled offset
6985 // 'dst_start' points to dst array + scaled offset
6986
6987 // See GraphKit::compress_string
6988 const TypePtr* adr_type;
6989 Node* mem = capture_memory(adr_type, src_type, dst_type);
6990 Node* enc = new EncodeISOArrayNode(control(), mem, adr_type, src_start, dst_start, length, ascii);
6991 enc = _gvn.transform(enc);
6992 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6993 memory_effect(res_mem, src_type, dst_type);
6994
6995 set_result(enc);
6996 clear_upper_avx();
6997
6998 return true;
6999 }
7000
7001 //-------------inline_multiplyToLen-----------------------------------
7002 bool LibraryCallKit::inline_multiplyToLen() {
7003 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
7004
7005 address stubAddr = StubRoutines::multiplyToLen();
7006 if (stubAddr == nullptr) {
7007 return false; // Intrinsic's stub is not implemented on this platform
7008 }
7009 const char* stubName = "multiplyToLen";
7010
7011 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
7012
7013 // no receiver because it is a static method
7014 Node* x = argument(0);
7015 Node* xlen = argument(1);
7016 Node* y = argument(2);
7017 Node* ylen = argument(3);
7018 Node* z = argument(4);
7019
7020 x = must_be_not_null(x, true);
7021 y = must_be_not_null(y, true);
7022
7023 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
7024 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
7025 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
7026 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
7027 // failed array check
7028 return false;
7029 }
7030
7031 BasicType x_elem = x_type->elem()->array_element_basic_type();
7032 BasicType y_elem = y_type->elem()->array_element_basic_type();
7033 if (x_elem != T_INT || y_elem != T_INT) {
7034 return false;
7035 }
7036
7037 Node* x_start = array_element_address(x, intcon(0), x_elem);
7038 Node* y_start = array_element_address(y, intcon(0), y_elem);
7039 // 'x_start' points to x array + scaled xlen
7040 // 'y_start' points to y array + scaled ylen
7041
7042 Node* z_start = array_element_address(z, intcon(0), T_INT);
7043
7044 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
7045 OptoRuntime::multiplyToLen_Type(),
7046 stubAddr, stubName, TypePtr::BOTTOM,
7047 x_start, xlen, y_start, ylen, z_start);
7048
7049 C->set_has_split_ifs(true); // Has chance for split-if optimization
7050 set_result(z);
7051 return true;
7052 }
7053
7054 //-------------inline_squareToLen------------------------------------
7055 bool LibraryCallKit::inline_squareToLen() {
7056 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
7057
7058 address stubAddr = StubRoutines::squareToLen();
7059 if (stubAddr == nullptr) {
7060 return false; // Intrinsic's stub is not implemented on this platform
7061 }
7062 const char* stubName = "squareToLen";
7063
7064 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
7065
7066 Node* x = argument(0);
7067 Node* len = argument(1);
7068 Node* z = argument(2);
7069 Node* zlen = argument(3);
7070
7071 x = must_be_not_null(x, true);
7072 z = must_be_not_null(z, true);
7073
7074 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
7075 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
7076 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
7077 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
7078 // failed array check
7079 return false;
7080 }
7081
7082 BasicType x_elem = x_type->elem()->array_element_basic_type();
7083 BasicType z_elem = z_type->elem()->array_element_basic_type();
7084 if (x_elem != T_INT || z_elem != T_INT) {
7085 return false;
7086 }
7087
7088
7089 Node* x_start = array_element_address(x, intcon(0), x_elem);
7090 Node* z_start = array_element_address(z, intcon(0), z_elem);
7091
7092 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
7093 OptoRuntime::squareToLen_Type(),
7094 stubAddr, stubName, TypePtr::BOTTOM,
7095 x_start, len, z_start, zlen);
7096
7097 set_result(z);
7098 return true;
7099 }
7100
7101 //-------------inline_mulAdd------------------------------------------
7102 bool LibraryCallKit::inline_mulAdd() {
7103 assert(UseMulAddIntrinsic, "not implemented on this platform");
7104
7105 address stubAddr = StubRoutines::mulAdd();
7106 if (stubAddr == nullptr) {
7107 return false; // Intrinsic's stub is not implemented on this platform
7108 }
7109 const char* stubName = "mulAdd";
7110
7111 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
7112
7113 Node* out = argument(0);
7114 Node* in = argument(1);
7115 Node* offset = argument(2);
7116 Node* len = argument(3);
7117 Node* k = argument(4);
7118
7119 in = must_be_not_null(in, true);
7120 out = must_be_not_null(out, true);
7121
7122 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
7123 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
7124 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
7125 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
7126 // failed array check
7127 return false;
7128 }
7129
7130 BasicType out_elem = out_type->elem()->array_element_basic_type();
7131 BasicType in_elem = in_type->elem()->array_element_basic_type();
7132 if (out_elem != T_INT || in_elem != T_INT) {
7133 return false;
7134 }
7135
7136 Node* outlen = load_array_length(out);
7137 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
7138 Node* out_start = array_element_address(out, intcon(0), out_elem);
7139 Node* in_start = array_element_address(in, intcon(0), in_elem);
7140
7141 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
7142 OptoRuntime::mulAdd_Type(),
7143 stubAddr, stubName, TypePtr::BOTTOM,
7144 out_start,in_start, new_offset, len, k);
7145 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7146 set_result(result);
7147 return true;
7148 }
7149
7150 //-------------inline_montgomeryMultiply-----------------------------------
7151 bool LibraryCallKit::inline_montgomeryMultiply() {
7152 address stubAddr = StubRoutines::montgomeryMultiply();
7153 if (stubAddr == nullptr) {
7154 return false; // Intrinsic's stub is not implemented on this platform
7155 }
7156
7157 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
7158 const char* stubName = "montgomery_multiply";
7159
7160 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
7161
7162 Node* a = argument(0);
7163 Node* b = argument(1);
7164 Node* n = argument(2);
7165 Node* len = argument(3);
7166 Node* inv = argument(4);
7167 Node* m = argument(6);
7168
7169 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
7170 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
7171 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
7172 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
7173 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
7174 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
7175 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
7176 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
7177 // failed array check
7178 return false;
7179 }
7180
7181 BasicType a_elem = a_type->elem()->array_element_basic_type();
7182 BasicType b_elem = b_type->elem()->array_element_basic_type();
7183 BasicType n_elem = n_type->elem()->array_element_basic_type();
7184 BasicType m_elem = m_type->elem()->array_element_basic_type();
7185 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
7186 return false;
7187 }
7188
7189 // Make the call
7190 {
7191 Node* a_start = array_element_address(a, intcon(0), a_elem);
7192 Node* b_start = array_element_address(b, intcon(0), b_elem);
7193 Node* n_start = array_element_address(n, intcon(0), n_elem);
7194 Node* m_start = array_element_address(m, intcon(0), m_elem);
7195
7196 Node* call = make_runtime_call(RC_LEAF,
7197 OptoRuntime::montgomeryMultiply_Type(),
7198 stubAddr, stubName, TypePtr::BOTTOM,
7199 a_start, b_start, n_start, len, inv, top(),
7200 m_start);
7201 set_result(m);
7202 }
7203
7204 return true;
7205 }
7206
7207 bool LibraryCallKit::inline_montgomerySquare() {
7208 address stubAddr = StubRoutines::montgomerySquare();
7209 if (stubAddr == nullptr) {
7210 return false; // Intrinsic's stub is not implemented on this platform
7211 }
7212
7213 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
7214 const char* stubName = "montgomery_square";
7215
7216 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
7217
7218 Node* a = argument(0);
7219 Node* n = argument(1);
7220 Node* len = argument(2);
7221 Node* inv = argument(3);
7222 Node* m = argument(5);
7223
7224 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
7225 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
7226 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
7227 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
7228 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
7229 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
7230 // failed array check
7231 return false;
7232 }
7233
7234 BasicType a_elem = a_type->elem()->array_element_basic_type();
7235 BasicType n_elem = n_type->elem()->array_element_basic_type();
7236 BasicType m_elem = m_type->elem()->array_element_basic_type();
7237 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
7238 return false;
7239 }
7240
7241 // Make the call
7242 {
7243 Node* a_start = array_element_address(a, intcon(0), a_elem);
7244 Node* n_start = array_element_address(n, intcon(0), n_elem);
7245 Node* m_start = array_element_address(m, intcon(0), m_elem);
7246
7247 Node* call = make_runtime_call(RC_LEAF,
7248 OptoRuntime::montgomerySquare_Type(),
7249 stubAddr, stubName, TypePtr::BOTTOM,
7250 a_start, n_start, len, inv, top(),
7251 m_start);
7252 set_result(m);
7253 }
7254
7255 return true;
7256 }
7257
7258 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
7259 address stubAddr = nullptr;
7260 const char* stubName = nullptr;
7261
7262 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
7263 if (stubAddr == nullptr) {
7264 return false; // Intrinsic's stub is not implemented on this platform
7265 }
7266
7267 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
7268
7269 assert(callee()->signature()->size() == 5, "expected 5 arguments");
7270
7271 Node* newArr = argument(0);
7272 Node* oldArr = argument(1);
7273 Node* newIdx = argument(2);
7274 Node* shiftCount = argument(3);
7275 Node* numIter = argument(4);
7276
7277 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
7278 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
7279 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
7280 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
7281 return false;
7282 }
7283
7284 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
7285 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
7286 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
7287 return false;
7288 }
7289
7290 // Make the call
7291 {
7292 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
7293 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
7294
7295 Node* call = make_runtime_call(RC_LEAF,
7296 OptoRuntime::bigIntegerShift_Type(),
7297 stubAddr,
7298 stubName,
7299 TypePtr::BOTTOM,
7300 newArr_start,
7301 oldArr_start,
7302 newIdx,
7303 shiftCount,
7304 numIter);
7305 }
7306
7307 return true;
7308 }
7309
7310 //-------------inline_vectorizedMismatch------------------------------
7311 bool LibraryCallKit::inline_vectorizedMismatch() {
7312 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
7313
7314 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
7315 Node* obja = argument(0); // Object
7316 Node* aoffset = argument(1); // long
7317 Node* objb = argument(3); // Object
7318 Node* boffset = argument(4); // long
7319 Node* length = argument(6); // int
7320 Node* scale = argument(7); // int
7321
7322 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
7323 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
7324 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
7325 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
7326 scale == top()) {
7327 return false; // failed input validation
7328 }
7329
7330 Node* obja_adr = make_unsafe_address(obja, aoffset);
7331 Node* objb_adr = make_unsafe_address(objb, boffset);
7332
7333 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
7334 //
7335 // inline_limit = ArrayOperationPartialInlineSize / element_size;
7336 // if (length <= inline_limit) {
7337 // inline_path:
7338 // vmask = VectorMaskGen length
7339 // vload1 = LoadVectorMasked obja, vmask
7340 // vload2 = LoadVectorMasked objb, vmask
7341 // result1 = VectorCmpMasked vload1, vload2, vmask
7342 // } else {
7343 // call_stub_path:
7344 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
7345 // }
7346 // exit_block:
7347 // return Phi(result1, result2);
7348 //
7349 enum { inline_path = 1, // input is small enough to process it all at once
7350 stub_path = 2, // input is too large; call into the VM
7351 PATH_LIMIT = 3
7352 };
7353
7354 Node* exit_block = new RegionNode(PATH_LIMIT);
7355 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
7356 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
7357
7358 Node* call_stub_path = control();
7359
7360 BasicType elem_bt = T_ILLEGAL;
7361
7362 const TypeInt* scale_t = _gvn.type(scale)->is_int();
7363 if (scale_t->is_con()) {
7364 switch (scale_t->get_con()) {
7365 case 0: elem_bt = T_BYTE; break;
7366 case 1: elem_bt = T_SHORT; break;
7367 case 2: elem_bt = T_INT; break;
7368 case 3: elem_bt = T_LONG; break;
7369
7370 default: elem_bt = T_ILLEGAL; break; // not supported
7371 }
7372 }
7373
7374 int inline_limit = 0;
7375 bool do_partial_inline = false;
7376
7377 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
7378 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
7379 do_partial_inline = inline_limit >= 16;
7380 }
7381
7382 if (do_partial_inline) {
7383 assert(elem_bt != T_ILLEGAL, "sanity");
7384
7385 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
7386 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
7387 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
7388
7389 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
7390 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
7391 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
7392
7393 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
7394
7395 if (!stopped()) {
7396 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
7397
7398 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
7399 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
7400 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
7401 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
7402
7403 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
7404 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
7405 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
7406 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
7407
7408 exit_block->init_req(inline_path, control());
7409 memory_phi->init_req(inline_path, map()->memory());
7410 result_phi->init_req(inline_path, result);
7411
7412 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
7413 clear_upper_avx();
7414 }
7415 }
7416 }
7417
7418 if (call_stub_path != nullptr) {
7419 set_control(call_stub_path);
7420
7421 Node* call = make_runtime_call(RC_LEAF,
7422 OptoRuntime::vectorizedMismatch_Type(),
7423 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
7424 obja_adr, objb_adr, length, scale);
7425
7426 exit_block->init_req(stub_path, control());
7427 memory_phi->init_req(stub_path, map()->memory());
7428 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
7429 }
7430
7431 exit_block = _gvn.transform(exit_block);
7432 memory_phi = _gvn.transform(memory_phi);
7433 result_phi = _gvn.transform(result_phi);
7434
7435 record_for_igvn(exit_block);
7436 record_for_igvn(memory_phi);
7437 record_for_igvn(result_phi);
7438
7439 set_control(exit_block);
7440 set_all_memory(memory_phi);
7441 set_result(result_phi);
7442
7443 return true;
7444 }
7445
7446 //------------------------------inline_vectorizedHashcode----------------------------
7447 bool LibraryCallKit::inline_vectorizedHashCode() {
7448 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
7449
7450 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
7451 Node* array = argument(0);
7452 Node* offset = argument(1);
7453 Node* length = argument(2);
7454 Node* initialValue = argument(3);
7455 Node* basic_type = argument(4);
7456
7457 if (basic_type == top()) {
7458 return false; // failed input validation
7459 }
7460
7461 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
7462 if (!basic_type_t->is_con()) {
7463 return false; // Only intrinsify if mode argument is constant
7464 }
7465
7466 array = must_be_not_null(array, true);
7467
7468 BasicType bt = (BasicType)basic_type_t->get_con();
7469
7470 // Resolve address of first element
7471 Node* array_start = array_element_address(array, offset, bt);
7472
7473 const TypeAryPtr* in_adr_type = TypeAryPtr::get_array_body_type(bt);
7474 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(in_adr_type), in_adr_type,
7475 array_start, length, initialValue, basic_type)));
7476 clear_upper_avx();
7477
7478 return true;
7479 }
7480
7481 /**
7482 * Calculate CRC32 for byte.
7483 * int java.util.zip.CRC32.update(int crc, int b)
7484 */
7485 bool LibraryCallKit::inline_updateCRC32() {
7486 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7487 assert(callee()->signature()->size() == 2, "update has 2 parameters");
7488 // no receiver since it is static method
7489 Node* crc = argument(0); // type: int
7490 Node* b = argument(1); // type: int
7491
7492 /*
7493 * int c = ~ crc;
7494 * b = timesXtoThe32[(b ^ c) & 0xFF];
7495 * b = b ^ (c >>> 8);
7496 * crc = ~b;
7497 */
7498
7499 Node* M1 = intcon(-1);
7500 crc = _gvn.transform(new XorINode(crc, M1));
7501 Node* result = _gvn.transform(new XorINode(crc, b));
7502 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
7503
7504 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
7505 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
7506 Node* adr = off_heap_plus_addr(base, ConvI2X(offset));
7507 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
7508
7509 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
7510 result = _gvn.transform(new XorINode(crc, result));
7511 result = _gvn.transform(new XorINode(result, M1));
7512 set_result(result);
7513 return true;
7514 }
7515
7516 /**
7517 * Calculate CRC32 for byte[] array.
7518 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
7519 */
7520 bool LibraryCallKit::inline_updateBytesCRC32() {
7521 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7522 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7523 // no receiver since it is static method
7524 Node* crc = argument(0); // type: int
7525 Node* src = argument(1); // type: oop
7526 Node* offset = argument(2); // type: int
7527 Node* length = argument(3); // type: int
7528
7529 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7530 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7531 // failed array check
7532 return false;
7533 }
7534
7535 // Figure out the size and type of the elements we will be copying.
7536 BasicType src_elem = src_type->elem()->array_element_basic_type();
7537 if (src_elem != T_BYTE) {
7538 return false;
7539 }
7540
7541 // 'src_start' points to src array + scaled offset
7542 src = must_be_not_null(src, true);
7543 Node* src_start = array_element_address(src, offset, src_elem);
7544
7545 // We assume that range check is done by caller.
7546 // TODO: generate range check (offset+length < src.length) in debug VM.
7547
7548 // Call the stub.
7549 address stubAddr = StubRoutines::updateBytesCRC32();
7550 const char *stubName = "updateBytesCRC32";
7551
7552 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7553 stubAddr, stubName, TypePtr::BOTTOM,
7554 crc, src_start, length);
7555 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7556 set_result(result);
7557 return true;
7558 }
7559
7560 /**
7561 * Calculate CRC32 for ByteBuffer.
7562 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
7563 */
7564 bool LibraryCallKit::inline_updateByteBufferCRC32() {
7565 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7566 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7567 // no receiver since it is static method
7568 Node* crc = argument(0); // type: int
7569 Node* src = argument(1); // type: long
7570 Node* offset = argument(3); // type: int
7571 Node* length = argument(4); // type: int
7572
7573 src = ConvL2X(src); // adjust Java long to machine word
7574 Node* base = _gvn.transform(new CastX2PNode(src));
7575 offset = ConvI2X(offset);
7576
7577 // 'src_start' points to src array + scaled offset
7578 Node* src_start = off_heap_plus_addr(base, offset);
7579
7580 // Call the stub.
7581 address stubAddr = StubRoutines::updateBytesCRC32();
7582 const char *stubName = "updateBytesCRC32";
7583
7584 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7585 stubAddr, stubName, TypePtr::BOTTOM,
7586 crc, src_start, length);
7587 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7588 set_result(result);
7589 return true;
7590 }
7591
7592 //------------------------------get_table_from_crc32c_class-----------------------
7593 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
7594 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
7595 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
7596
7597 return table;
7598 }
7599
7600 //------------------------------inline_updateBytesCRC32C-----------------------
7601 //
7602 // Calculate CRC32C for byte[] array.
7603 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
7604 //
7605 bool LibraryCallKit::inline_updateBytesCRC32C() {
7606 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7607 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7608 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7609 // no receiver since it is a static method
7610 Node* crc = argument(0); // type: int
7611 Node* src = argument(1); // type: oop
7612 Node* offset = argument(2); // type: int
7613 Node* end = argument(3); // type: int
7614
7615 Node* length = _gvn.transform(new SubINode(end, offset));
7616
7617 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7618 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7619 // failed array check
7620 return false;
7621 }
7622
7623 // Figure out the size and type of the elements we will be copying.
7624 BasicType src_elem = src_type->elem()->array_element_basic_type();
7625 if (src_elem != T_BYTE) {
7626 return false;
7627 }
7628
7629 // 'src_start' points to src array + scaled offset
7630 src = must_be_not_null(src, true);
7631 Node* src_start = array_element_address(src, offset, src_elem);
7632
7633 // static final int[] byteTable in class CRC32C
7634 Node* table = get_table_from_crc32c_class(callee()->holder());
7635 table = must_be_not_null(table, true);
7636 Node* table_start = array_element_address(table, intcon(0), T_INT);
7637
7638 // We assume that range check is done by caller.
7639 // TODO: generate range check (offset+length < src.length) in debug VM.
7640
7641 // Call the stub.
7642 address stubAddr = StubRoutines::updateBytesCRC32C();
7643 const char *stubName = "updateBytesCRC32C";
7644
7645 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7646 stubAddr, stubName, TypePtr::BOTTOM,
7647 crc, src_start, length, table_start);
7648 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7649 set_result(result);
7650 return true;
7651 }
7652
7653 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
7654 //
7655 // Calculate CRC32C for DirectByteBuffer.
7656 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
7657 //
7658 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
7659 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7660 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
7661 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7662 // no receiver since it is a static method
7663 Node* crc = argument(0); // type: int
7664 Node* src = argument(1); // type: long
7665 Node* offset = argument(3); // type: int
7666 Node* end = argument(4); // type: int
7667
7668 Node* length = _gvn.transform(new SubINode(end, offset));
7669
7670 src = ConvL2X(src); // adjust Java long to machine word
7671 Node* base = _gvn.transform(new CastX2PNode(src));
7672 offset = ConvI2X(offset);
7673
7674 // 'src_start' points to src array + scaled offset
7675 Node* src_start = off_heap_plus_addr(base, offset);
7676
7677 // static final int[] byteTable in class CRC32C
7678 Node* table = get_table_from_crc32c_class(callee()->holder());
7679 table = must_be_not_null(table, true);
7680 Node* table_start = array_element_address(table, intcon(0), T_INT);
7681
7682 // Call the stub.
7683 address stubAddr = StubRoutines::updateBytesCRC32C();
7684 const char *stubName = "updateBytesCRC32C";
7685
7686 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7687 stubAddr, stubName, TypePtr::BOTTOM,
7688 crc, src_start, length, table_start);
7689 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7690 set_result(result);
7691 return true;
7692 }
7693
7694 //------------------------------inline_updateBytesAdler32----------------------
7695 //
7696 // Calculate Adler32 checksum for byte[] array.
7697 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
7698 //
7699 bool LibraryCallKit::inline_updateBytesAdler32() {
7700 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7701 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7702 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7703 // no receiver since it is static method
7704 Node* crc = argument(0); // type: int
7705 Node* src = argument(1); // type: oop
7706 Node* offset = argument(2); // type: int
7707 Node* length = argument(3); // type: int
7708
7709 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7710 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7711 // failed array check
7712 return false;
7713 }
7714
7715 // Figure out the size and type of the elements we will be copying.
7716 BasicType src_elem = src_type->elem()->array_element_basic_type();
7717 if (src_elem != T_BYTE) {
7718 return false;
7719 }
7720
7721 // 'src_start' points to src array + scaled offset
7722 Node* src_start = array_element_address(src, offset, src_elem);
7723
7724 // We assume that range check is done by caller.
7725 // TODO: generate range check (offset+length < src.length) in debug VM.
7726
7727 // Call the stub.
7728 address stubAddr = StubRoutines::updateBytesAdler32();
7729 const char *stubName = "updateBytesAdler32";
7730
7731 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7732 stubAddr, stubName, TypePtr::BOTTOM,
7733 crc, src_start, length);
7734 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7735 set_result(result);
7736 return true;
7737 }
7738
7739 //------------------------------inline_updateByteBufferAdler32---------------
7740 //
7741 // Calculate Adler32 checksum for DirectByteBuffer.
7742 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
7743 //
7744 bool LibraryCallKit::inline_updateByteBufferAdler32() {
7745 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7746 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7747 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7748 // no receiver since it is static method
7749 Node* crc = argument(0); // type: int
7750 Node* src = argument(1); // type: long
7751 Node* offset = argument(3); // type: int
7752 Node* length = argument(4); // type: int
7753
7754 src = ConvL2X(src); // adjust Java long to machine word
7755 Node* base = _gvn.transform(new CastX2PNode(src));
7756 offset = ConvI2X(offset);
7757
7758 // 'src_start' points to src array + scaled offset
7759 Node* src_start = off_heap_plus_addr(base, offset);
7760
7761 // Call the stub.
7762 address stubAddr = StubRoutines::updateBytesAdler32();
7763 const char *stubName = "updateBytesAdler32";
7764
7765 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7766 stubAddr, stubName, TypePtr::BOTTOM,
7767 crc, src_start, length);
7768
7769 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7770 set_result(result);
7771 return true;
7772 }
7773
7774 //----------------------------inline_reference_get0----------------------------
7775 // public T java.lang.ref.Reference.get();
7776 bool LibraryCallKit::inline_reference_get0() {
7777 const int referent_offset = java_lang_ref_Reference::referent_offset();
7778
7779 // Get the argument:
7780 Node* reference_obj = null_check_receiver();
7781 if (stopped()) return true;
7782
7783 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7784 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7785 decorators, /*is_static*/ false,
7786 env()->Reference_klass());
7787 if (result == nullptr) return false;
7788
7789 // Add memory barrier to prevent commoning reads from this field
7790 // across safepoint since GC can change its value.
7791 insert_mem_bar(Op_MemBarCPUOrder);
7792
7793 set_result(result);
7794 return true;
7795 }
7796
7797 //----------------------------inline_reference_refersTo0----------------------------
7798 // bool java.lang.ref.Reference.refersTo0();
7799 // bool java.lang.ref.PhantomReference.refersTo0();
7800 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7801 // Get arguments:
7802 Node* reference_obj = null_check_receiver();
7803 Node* other_obj = argument(1);
7804 if (stopped()) return true;
7805
7806 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7807 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7808 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7809 decorators, /*is_static*/ false,
7810 env()->Reference_klass());
7811 if (referent == nullptr) return false;
7812
7813 // Add memory barrier to prevent commoning reads from this field
7814 // across safepoint since GC can change its value.
7815 insert_mem_bar(Op_MemBarCPUOrder);
7816
7817 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7818 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7819 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7820
7821 RegionNode* region = new RegionNode(3);
7822 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7823
7824 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7825 region->init_req(1, if_true);
7826 phi->init_req(1, intcon(1));
7827
7828 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7829 region->init_req(2, if_false);
7830 phi->init_req(2, intcon(0));
7831
7832 set_control(_gvn.transform(region));
7833 record_for_igvn(region);
7834 set_result(_gvn.transform(phi));
7835 return true;
7836 }
7837
7838 //----------------------------inline_reference_clear0----------------------------
7839 // void java.lang.ref.Reference.clear0();
7840 // void java.lang.ref.PhantomReference.clear0();
7841 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7842 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7843
7844 // Get arguments
7845 Node* reference_obj = null_check_receiver();
7846 if (stopped()) return true;
7847
7848 // Common access parameters
7849 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7850 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7851 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7852 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7853 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7854
7855 Node* referent = access_load_at(reference_obj,
7856 referent_field_addr,
7857 referent_field_addr_type,
7858 val_type,
7859 T_OBJECT,
7860 decorators);
7861
7862 IdealKit ideal(this);
7863 #define __ ideal.
7864 __ if_then(referent, BoolTest::ne, null());
7865 sync_kit(ideal);
7866 access_store_at(reference_obj,
7867 referent_field_addr,
7868 referent_field_addr_type,
7869 null(),
7870 val_type,
7871 T_OBJECT,
7872 decorators);
7873 __ sync_kit(this);
7874 __ end_if();
7875 final_sync(ideal);
7876 #undef __
7877
7878 return true;
7879 }
7880
7881 //-----------------------inline_reference_reachabilityFence-----------------
7882 // bool java.lang.ref.Reference.reachabilityFence();
7883 bool LibraryCallKit::inline_reference_reachabilityFence() {
7884 Node* referent = argument(0);
7885 insert_reachability_fence(referent);
7886 return true;
7887 }
7888
7889 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7890 DecoratorSet decorators, bool is_static,
7891 ciInstanceKlass* fromKls) {
7892 if (fromKls == nullptr) {
7893 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7894 assert(tinst != nullptr, "obj is null");
7895 assert(tinst->is_loaded(), "obj is not loaded");
7896 fromKls = tinst->instance_klass();
7897 }
7898 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7899 ciSymbol::make(fieldTypeString),
7900 is_static);
7901
7902 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7903 if (field == nullptr) return (Node *) nullptr;
7904
7905 if (is_static) {
7906 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7907 fromObj = makecon(tip);
7908 }
7909
7910 // Next code copied from Parse::do_get_xxx():
7911
7912 // Compute address and memory type.
7913 int offset = field->offset_in_bytes();
7914 bool is_vol = field->is_volatile();
7915 ciType* field_klass = field->type();
7916 assert(field_klass->is_loaded(), "should be loaded");
7917 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7918 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7919 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7920 "slice of address and input slice don't match");
7921 BasicType bt = field->layout_type();
7922
7923 // Build the resultant type of the load
7924 const Type *type;
7925 if (bt == T_OBJECT) {
7926 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7927 } else {
7928 type = Type::get_const_basic_type(bt);
7929 }
7930
7931 if (is_vol) {
7932 decorators |= MO_SEQ_CST;
7933 }
7934
7935 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7936 }
7937
7938 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7939 bool is_exact /* true */, bool is_static /* false */,
7940 ciInstanceKlass * fromKls /* nullptr */) {
7941 if (fromKls == nullptr) {
7942 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7943 assert(tinst != nullptr, "obj is null");
7944 assert(tinst->is_loaded(), "obj is not loaded");
7945 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7946 fromKls = tinst->instance_klass();
7947 }
7948 else {
7949 assert(is_static, "only for static field access");
7950 }
7951 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7952 ciSymbol::make(fieldTypeString),
7953 is_static);
7954
7955 assert(field != nullptr, "undefined field");
7956 assert(!field->is_volatile(), "not defined for volatile fields");
7957
7958 if (is_static) {
7959 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7960 fromObj = makecon(tip);
7961 }
7962
7963 // Next code copied from Parse::do_get_xxx():
7964
7965 // Compute address and memory type.
7966 int offset = field->offset_in_bytes();
7967 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7968
7969 return adr;
7970 }
7971
7972 //------------------------------inline_aescrypt_Block-----------------------
7973 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7974 address stubAddr = nullptr;
7975 const char *stubName;
7976 bool is_decrypt = false;
7977 assert(UseAES, "need AES instruction support");
7978
7979 switch(id) {
7980 case vmIntrinsics::_aescrypt_encryptBlock:
7981 stubAddr = StubRoutines::aescrypt_encryptBlock();
7982 stubName = "aescrypt_encryptBlock";
7983 break;
7984 case vmIntrinsics::_aescrypt_decryptBlock:
7985 stubAddr = StubRoutines::aescrypt_decryptBlock();
7986 stubName = "aescrypt_decryptBlock";
7987 is_decrypt = true;
7988 break;
7989 default:
7990 break;
7991 }
7992 if (stubAddr == nullptr) return false;
7993
7994 Node* aescrypt_object = argument(0);
7995 Node* src = argument(1);
7996 Node* src_offset = argument(2);
7997 Node* dest = argument(3);
7998 Node* dest_offset = argument(4);
7999
8000 src = must_be_not_null(src, true);
8001 dest = must_be_not_null(dest, true);
8002
8003 // (1) src and dest are arrays.
8004 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8005 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8006 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8007 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8008
8009 // for the quick and dirty code we will skip all the checks.
8010 // we are just trying to get the call to be generated.
8011 Node* src_start = src;
8012 Node* dest_start = dest;
8013 if (src_offset != nullptr || dest_offset != nullptr) {
8014 assert(src_offset != nullptr && dest_offset != nullptr, "");
8015 src_start = array_element_address(src, src_offset, T_BYTE);
8016 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8017 }
8018
8019 // now need to get the start of its expanded key array
8020 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8021 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
8022 if (k_start == nullptr) return false;
8023
8024 // Call the stub.
8025 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
8026 stubAddr, stubName, TypePtr::BOTTOM,
8027 src_start, dest_start, k_start);
8028
8029 return true;
8030 }
8031
8032 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
8033 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
8034 address stubAddr = nullptr;
8035 const char *stubName = nullptr;
8036 bool is_decrypt = false;
8037 assert(UseAES, "need AES instruction support");
8038
8039 switch(id) {
8040 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
8041 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
8042 stubName = "cipherBlockChaining_encryptAESCrypt";
8043 break;
8044 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
8045 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
8046 stubName = "cipherBlockChaining_decryptAESCrypt";
8047 is_decrypt = true;
8048 break;
8049 default:
8050 break;
8051 }
8052 if (stubAddr == nullptr) return false;
8053
8054 Node* cipherBlockChaining_object = argument(0);
8055 Node* src = argument(1);
8056 Node* src_offset = argument(2);
8057 Node* len = argument(3);
8058 Node* dest = argument(4);
8059 Node* dest_offset = argument(5);
8060
8061 src = must_be_not_null(src, false);
8062 dest = must_be_not_null(dest, false);
8063
8064 // (1) src and dest are arrays.
8065 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8066 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8067 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8068 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8069
8070 // checks are the responsibility of the caller
8071 Node* src_start = src;
8072 Node* dest_start = dest;
8073 if (src_offset != nullptr || dest_offset != nullptr) {
8074 assert(src_offset != nullptr && dest_offset != nullptr, "");
8075 src_start = array_element_address(src, src_offset, T_BYTE);
8076 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8077 }
8078
8079 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8080 // (because of the predicated logic executed earlier).
8081 // so we cast it here safely.
8082 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8083
8084 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8085 if (embeddedCipherObj == nullptr) return false;
8086
8087 // cast it to what we know it will be at runtime
8088 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
8089 assert(tinst != nullptr, "CBC obj is null");
8090 assert(tinst->is_loaded(), "CBC obj is not loaded");
8091 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8092 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8093
8094 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8095 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8096 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8097 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8098 aescrypt_object = _gvn.transform(aescrypt_object);
8099
8100 // we need to get the start of the aescrypt_object's expanded key array
8101 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
8102 if (k_start == nullptr) return false;
8103
8104 // similarly, get the start address of the r vector
8105 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
8106 if (objRvec == nullptr) return false;
8107 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
8108
8109 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8110 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8111 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
8112 stubAddr, stubName, TypePtr::BOTTOM,
8113 src_start, dest_start, k_start, r_start, len);
8114
8115 // return cipher length (int)
8116 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
8117 set_result(retvalue);
8118 return true;
8119 }
8120
8121 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
8122 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
8123 address stubAddr = nullptr;
8124 const char *stubName = nullptr;
8125 bool is_decrypt = false;
8126 assert(UseAES, "need AES instruction support");
8127
8128 switch (id) {
8129 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
8130 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
8131 stubName = "electronicCodeBook_encryptAESCrypt";
8132 break;
8133 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
8134 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
8135 stubName = "electronicCodeBook_decryptAESCrypt";
8136 is_decrypt = true;
8137 break;
8138 default:
8139 break;
8140 }
8141
8142 if (stubAddr == nullptr) return false;
8143
8144 Node* electronicCodeBook_object = argument(0);
8145 Node* src = argument(1);
8146 Node* src_offset = argument(2);
8147 Node* len = argument(3);
8148 Node* dest = argument(4);
8149 Node* dest_offset = argument(5);
8150
8151 // (1) src and dest are arrays.
8152 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8153 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8154 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8155 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8156
8157 // checks are the responsibility of the caller
8158 Node* src_start = src;
8159 Node* dest_start = dest;
8160 if (src_offset != nullptr || dest_offset != nullptr) {
8161 assert(src_offset != nullptr && dest_offset != nullptr, "");
8162 src_start = array_element_address(src, src_offset, T_BYTE);
8163 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8164 }
8165
8166 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8167 // (because of the predicated logic executed earlier).
8168 // so we cast it here safely.
8169 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8170
8171 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8172 if (embeddedCipherObj == nullptr) return false;
8173
8174 // cast it to what we know it will be at runtime
8175 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
8176 assert(tinst != nullptr, "ECB obj is null");
8177 assert(tinst->is_loaded(), "ECB obj is not loaded");
8178 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8179 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8180
8181 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8182 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8183 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8184 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8185 aescrypt_object = _gvn.transform(aescrypt_object);
8186
8187 // we need to get the start of the aescrypt_object's expanded key array
8188 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
8189 if (k_start == nullptr) return false;
8190
8191 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8192 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
8193 OptoRuntime::electronicCodeBook_aescrypt_Type(),
8194 stubAddr, stubName, TypePtr::BOTTOM,
8195 src_start, dest_start, k_start, len);
8196
8197 // return cipher length (int)
8198 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
8199 set_result(retvalue);
8200 return true;
8201 }
8202
8203 //------------------------------inline_counterMode_AESCrypt-----------------------
8204 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
8205 assert(UseAES, "need AES instruction support");
8206 if (!UseAESCTRIntrinsics) return false;
8207
8208 address stubAddr = nullptr;
8209 const char *stubName = nullptr;
8210 if (id == vmIntrinsics::_counterMode_AESCrypt) {
8211 stubAddr = StubRoutines::counterMode_AESCrypt();
8212 stubName = "counterMode_AESCrypt";
8213 }
8214 if (stubAddr == nullptr) return false;
8215
8216 Node* counterMode_object = argument(0);
8217 Node* src = argument(1);
8218 Node* src_offset = argument(2);
8219 Node* len = argument(3);
8220 Node* dest = argument(4);
8221 Node* dest_offset = argument(5);
8222
8223 // (1) src and dest are arrays.
8224 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8225 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8226 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8227 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8228
8229 // checks are the responsibility of the caller
8230 Node* src_start = src;
8231 Node* dest_start = dest;
8232 if (src_offset != nullptr || dest_offset != nullptr) {
8233 assert(src_offset != nullptr && dest_offset != nullptr, "");
8234 src_start = array_element_address(src, src_offset, T_BYTE);
8235 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8236 }
8237
8238 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8239 // (because of the predicated logic executed earlier).
8240 // so we cast it here safely.
8241 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8242 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8243 if (embeddedCipherObj == nullptr) return false;
8244 // cast it to what we know it will be at runtime
8245 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
8246 assert(tinst != nullptr, "CTR obj is null");
8247 assert(tinst->is_loaded(), "CTR obj is not loaded");
8248 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8249 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8250 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8251 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8252 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8253 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8254 aescrypt_object = _gvn.transform(aescrypt_object);
8255 // we need to get the start of the aescrypt_object's expanded key array
8256 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
8257 if (k_start == nullptr) return false;
8258 // similarly, get the start address of the r vector
8259 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
8260 if (obj_counter == nullptr) return false;
8261 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
8262
8263 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
8264 if (saved_encCounter == nullptr) return false;
8265 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
8266 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
8267
8268 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8269 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8270 OptoRuntime::counterMode_aescrypt_Type(),
8271 stubAddr, stubName, TypePtr::BOTTOM,
8272 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
8273
8274 // return cipher length (int)
8275 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
8276 set_result(retvalue);
8277 return true;
8278 }
8279
8280 //------------------------------get_key_start_from_aescrypt_object-----------------------
8281 Node* LibraryCallKit::get_key_start_from_aescrypt_object(Node* aescrypt_object, bool is_decrypt) {
8282 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
8283 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
8284 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
8285 // The following platform specific stubs of encryption and decryption use the same round keys.
8286 #if defined(PPC64) || defined(S390) || defined(RISCV64)
8287 bool use_decryption_key = false;
8288 #else
8289 bool use_decryption_key = is_decrypt;
8290 #endif
8291 Node* objAESCryptKey = load_field_from_object(aescrypt_object, use_decryption_key ? "sessionKd" : "sessionKe", "[I");
8292 assert(objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AES_Crypt");
8293 if (objAESCryptKey == nullptr) return (Node *) nullptr;
8294
8295 // now have the array, need to get the start address of the selected key array
8296 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
8297 return k_start;
8298 }
8299
8300 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
8301 // Return node representing slow path of predicate check.
8302 // the pseudo code we want to emulate with this predicate is:
8303 // for encryption:
8304 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8305 // for decryption:
8306 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8307 // note cipher==plain is more conservative than the original java code but that's OK
8308 //
8309 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
8310 // The receiver was checked for null already.
8311 Node* objCBC = argument(0);
8312
8313 Node* src = argument(1);
8314 Node* dest = argument(4);
8315
8316 // Load embeddedCipher field of CipherBlockChaining object.
8317 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8318
8319 // get AESCrypt klass for instanceOf check
8320 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8321 // will have same classloader as CipherBlockChaining object
8322 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
8323 assert(tinst != nullptr, "CBCobj is null");
8324 assert(tinst->is_loaded(), "CBCobj is not loaded");
8325
8326 // we want to do an instanceof comparison against the AESCrypt class
8327 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8328 if (!klass_AESCrypt->is_loaded()) {
8329 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8330 Node* ctrl = control();
8331 set_control(top()); // no regular fast path
8332 return ctrl;
8333 }
8334
8335 src = must_be_not_null(src, true);
8336 dest = must_be_not_null(dest, true);
8337
8338 // Resolve oops to stable for CmpP below.
8339 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8340
8341 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8342 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8343 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8344
8345 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8346
8347 // for encryption, we are done
8348 if (!decrypting)
8349 return instof_false; // even if it is null
8350
8351 // for decryption, we need to add a further check to avoid
8352 // taking the intrinsic path when cipher and plain are the same
8353 // see the original java code for why.
8354 RegionNode* region = new RegionNode(3);
8355 region->init_req(1, instof_false);
8356
8357 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8358 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8359 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8360 region->init_req(2, src_dest_conjoint);
8361
8362 record_for_igvn(region);
8363 return _gvn.transform(region);
8364 }
8365
8366 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
8367 // Return node representing slow path of predicate check.
8368 // the pseudo code we want to emulate with this predicate is:
8369 // for encryption:
8370 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8371 // for decryption:
8372 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8373 // note cipher==plain is more conservative than the original java code but that's OK
8374 //
8375 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
8376 // The receiver was checked for null already.
8377 Node* objECB = argument(0);
8378
8379 // Load embeddedCipher field of ElectronicCodeBook object.
8380 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8381
8382 // get AESCrypt klass for instanceOf check
8383 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8384 // will have same classloader as ElectronicCodeBook object
8385 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
8386 assert(tinst != nullptr, "ECBobj is null");
8387 assert(tinst->is_loaded(), "ECBobj is not loaded");
8388
8389 // we want to do an instanceof comparison against the AESCrypt class
8390 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8391 if (!klass_AESCrypt->is_loaded()) {
8392 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8393 Node* ctrl = control();
8394 set_control(top()); // no regular fast path
8395 return ctrl;
8396 }
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 Node* src = argument(1);
8415 Node* dest = argument(4);
8416 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8417 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8418 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8419 region->init_req(2, src_dest_conjoint);
8420
8421 record_for_igvn(region);
8422 return _gvn.transform(region);
8423 }
8424
8425 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
8426 // Return node representing slow path of predicate check.
8427 // the pseudo code we want to emulate with this predicate is:
8428 // for encryption:
8429 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8430 // for decryption:
8431 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8432 // note cipher==plain is more conservative than the original java code but that's OK
8433 //
8434
8435 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
8436 // The receiver was checked for null already.
8437 Node* objCTR = argument(0);
8438
8439 // Load embeddedCipher field of CipherBlockChaining object.
8440 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8441
8442 // get AESCrypt klass for instanceOf check
8443 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8444 // will have same classloader as CipherBlockChaining object
8445 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
8446 assert(tinst != nullptr, "CTRobj is null");
8447 assert(tinst->is_loaded(), "CTRobj is not loaded");
8448
8449 // we want to do an instanceof comparison against the AESCrypt class
8450 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8451 if (!klass_AESCrypt->is_loaded()) {
8452 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8453 Node* ctrl = control();
8454 set_control(top()); // no regular fast path
8455 return ctrl;
8456 }
8457
8458 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8459 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8460 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8461 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8462 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8463
8464 return instof_false; // even if it is null
8465 }
8466
8467 //------------------------------inline_ghash_processBlocks
8468 bool LibraryCallKit::inline_ghash_processBlocks() {
8469 address stubAddr;
8470 const char *stubName;
8471 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
8472
8473 stubAddr = StubRoutines::ghash_processBlocks();
8474 stubName = "ghash_processBlocks";
8475
8476 Node* data = argument(0);
8477 Node* offset = argument(1);
8478 Node* len = argument(2);
8479 Node* state = argument(3);
8480 Node* subkeyH = argument(4);
8481
8482 state = must_be_not_null(state, true);
8483 subkeyH = must_be_not_null(subkeyH, true);
8484 data = must_be_not_null(data, true);
8485
8486 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8487 assert(state_start, "state is null");
8488 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
8489 assert(subkeyH_start, "subkeyH is null");
8490 Node* data_start = array_element_address(data, offset, T_BYTE);
8491 assert(data_start, "data is null");
8492
8493 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
8494 OptoRuntime::ghash_processBlocks_Type(),
8495 stubAddr, stubName, TypePtr::BOTTOM,
8496 state_start, subkeyH_start, data_start, len);
8497 return true;
8498 }
8499
8500 //------------------------------inline_chacha20Block
8501 bool LibraryCallKit::inline_chacha20Block() {
8502 address stubAddr;
8503 const char *stubName;
8504 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
8505
8506 stubAddr = StubRoutines::chacha20Block();
8507 stubName = "chacha20Block";
8508
8509 Node* state = argument(0);
8510 Node* result = argument(1);
8511
8512 state = must_be_not_null(state, true);
8513 result = must_be_not_null(result, true);
8514
8515 Node* state_start = array_element_address(state, intcon(0), T_INT);
8516 assert(state_start, "state is null");
8517 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
8518 assert(result_start, "result is null");
8519
8520 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
8521 OptoRuntime::chacha20Block_Type(),
8522 stubAddr, stubName, TypePtr::BOTTOM,
8523 state_start, result_start);
8524 // return key stream length (int)
8525 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
8526 set_result(retvalue);
8527 return true;
8528 }
8529
8530 //------------------------------inline_kyberNtt
8531 bool LibraryCallKit::inline_kyberNtt() {
8532 address stubAddr;
8533 const char *stubName;
8534 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8535 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
8536
8537 stubAddr = StubRoutines::kyberNtt();
8538 stubName = "kyberNtt";
8539 if (!stubAddr) return false;
8540
8541 Node* coeffs = argument(0);
8542 Node* ntt_zetas = argument(1);
8543
8544 coeffs = must_be_not_null(coeffs, true);
8545 ntt_zetas = must_be_not_null(ntt_zetas, true);
8546
8547 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8548 assert(coeffs_start, "coeffs is null");
8549 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
8550 assert(ntt_zetas_start, "ntt_zetas is null");
8551 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8552 OptoRuntime::kyberNtt_Type(),
8553 stubAddr, stubName, TypePtr::BOTTOM,
8554 coeffs_start, ntt_zetas_start);
8555 // return an int
8556 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
8557 set_result(retvalue);
8558 return true;
8559 }
8560
8561 //------------------------------inline_kyberInverseNtt
8562 bool LibraryCallKit::inline_kyberInverseNtt() {
8563 address stubAddr;
8564 const char *stubName;
8565 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8566 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
8567
8568 stubAddr = StubRoutines::kyberInverseNtt();
8569 stubName = "kyberInverseNtt";
8570 if (!stubAddr) return false;
8571
8572 Node* coeffs = argument(0);
8573 Node* zetas = argument(1);
8574
8575 coeffs = must_be_not_null(coeffs, true);
8576 zetas = must_be_not_null(zetas, true);
8577
8578 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8579 assert(coeffs_start, "coeffs is null");
8580 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8581 assert(zetas_start, "inverseNtt_zetas is null");
8582 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8583 OptoRuntime::kyberInverseNtt_Type(),
8584 stubAddr, stubName, TypePtr::BOTTOM,
8585 coeffs_start, zetas_start);
8586
8587 // return an int
8588 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
8589 set_result(retvalue);
8590 return true;
8591 }
8592
8593 //------------------------------inline_kyberNttMult
8594 bool LibraryCallKit::inline_kyberNttMult() {
8595 address stubAddr;
8596 const char *stubName;
8597 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8598 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
8599
8600 stubAddr = StubRoutines::kyberNttMult();
8601 stubName = "kyberNttMult";
8602 if (!stubAddr) return false;
8603
8604 Node* result = argument(0);
8605 Node* ntta = argument(1);
8606 Node* nttb = argument(2);
8607 Node* zetas = argument(3);
8608
8609 result = must_be_not_null(result, true);
8610 ntta = must_be_not_null(ntta, true);
8611 nttb = must_be_not_null(nttb, true);
8612 zetas = must_be_not_null(zetas, true);
8613
8614 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8615 assert(result_start, "result is null");
8616 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
8617 assert(ntta_start, "ntta is null");
8618 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
8619 assert(nttb_start, "nttb is null");
8620 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8621 assert(zetas_start, "nttMult_zetas is null");
8622 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8623 OptoRuntime::kyberNttMult_Type(),
8624 stubAddr, stubName, TypePtr::BOTTOM,
8625 result_start, ntta_start, nttb_start,
8626 zetas_start);
8627
8628 // return an int
8629 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
8630 set_result(retvalue);
8631
8632 return true;
8633 }
8634
8635 //------------------------------inline_kyberAddPoly_2
8636 bool LibraryCallKit::inline_kyberAddPoly_2() {
8637 address stubAddr;
8638 const char *stubName;
8639 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8640 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
8641
8642 stubAddr = StubRoutines::kyberAddPoly_2();
8643 stubName = "kyberAddPoly_2";
8644 if (!stubAddr) return false;
8645
8646 Node* result = argument(0);
8647 Node* a = argument(1);
8648 Node* b = argument(2);
8649
8650 result = must_be_not_null(result, true);
8651 a = must_be_not_null(a, true);
8652 b = must_be_not_null(b, true);
8653
8654 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8655 assert(result_start, "result is null");
8656 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8657 assert(a_start, "a is null");
8658 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8659 assert(b_start, "b is null");
8660 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
8661 OptoRuntime::kyberAddPoly_2_Type(),
8662 stubAddr, stubName, TypePtr::BOTTOM,
8663 result_start, a_start, b_start);
8664 // return an int
8665 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
8666 set_result(retvalue);
8667 return true;
8668 }
8669
8670 //------------------------------inline_kyberAddPoly_3
8671 bool LibraryCallKit::inline_kyberAddPoly_3() {
8672 address stubAddr;
8673 const char *stubName;
8674 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8675 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
8676
8677 stubAddr = StubRoutines::kyberAddPoly_3();
8678 stubName = "kyberAddPoly_3";
8679 if (!stubAddr) return false;
8680
8681 Node* result = argument(0);
8682 Node* a = argument(1);
8683 Node* b = argument(2);
8684 Node* c = argument(3);
8685
8686 result = must_be_not_null(result, true);
8687 a = must_be_not_null(a, true);
8688 b = must_be_not_null(b, true);
8689 c = must_be_not_null(c, true);
8690
8691 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8692 assert(result_start, "result is null");
8693 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8694 assert(a_start, "a is null");
8695 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8696 assert(b_start, "b is null");
8697 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
8698 assert(c_start, "c is null");
8699 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
8700 OptoRuntime::kyberAddPoly_3_Type(),
8701 stubAddr, stubName, TypePtr::BOTTOM,
8702 result_start, a_start, b_start, c_start);
8703 // return an int
8704 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
8705 set_result(retvalue);
8706 return true;
8707 }
8708
8709 //------------------------------inline_kyber12To16
8710 bool LibraryCallKit::inline_kyber12To16() {
8711 address stubAddr;
8712 const char *stubName;
8713 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8714 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
8715
8716 stubAddr = StubRoutines::kyber12To16();
8717 stubName = "kyber12To16";
8718 if (!stubAddr) return false;
8719
8720 Node* condensed = argument(0);
8721 Node* condensedOffs = argument(1);
8722 Node* parsed = argument(2);
8723 Node* parsedLength = argument(3);
8724
8725 condensed = must_be_not_null(condensed, true);
8726 parsed = must_be_not_null(parsed, true);
8727
8728 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
8729 assert(condensed_start, "condensed is null");
8730 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
8731 assert(parsed_start, "parsed is null");
8732 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
8733 OptoRuntime::kyber12To16_Type(),
8734 stubAddr, stubName, TypePtr::BOTTOM,
8735 condensed_start, condensedOffs, parsed_start, parsedLength);
8736 // return an int
8737 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
8738 set_result(retvalue);
8739 return true;
8740
8741 }
8742
8743 //------------------------------inline_kyberBarrettReduce
8744 bool LibraryCallKit::inline_kyberBarrettReduce() {
8745 address stubAddr;
8746 const char *stubName;
8747 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8748 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
8749
8750 stubAddr = StubRoutines::kyberBarrettReduce();
8751 stubName = "kyberBarrettReduce";
8752 if (!stubAddr) return false;
8753
8754 Node* coeffs = argument(0);
8755
8756 coeffs = must_be_not_null(coeffs, true);
8757
8758 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8759 assert(coeffs_start, "coeffs is null");
8760 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
8761 OptoRuntime::kyberBarrettReduce_Type(),
8762 stubAddr, stubName, TypePtr::BOTTOM,
8763 coeffs_start);
8764 // return an int
8765 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
8766 set_result(retvalue);
8767 return true;
8768 }
8769
8770 //------------------------------inline_dilithiumAlmostNtt
8771 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
8772 address stubAddr;
8773 const char *stubName;
8774 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8775 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
8776
8777 stubAddr = StubRoutines::dilithiumAlmostNtt();
8778 stubName = "dilithiumAlmostNtt";
8779 if (!stubAddr) return false;
8780
8781 Node* coeffs = argument(0);
8782 Node* ntt_zetas = argument(1);
8783
8784 coeffs = must_be_not_null(coeffs, true);
8785 ntt_zetas = must_be_not_null(ntt_zetas, true);
8786
8787 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8788 assert(coeffs_start, "coeffs is null");
8789 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
8790 assert(ntt_zetas_start, "ntt_zetas is null");
8791 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8792 OptoRuntime::dilithiumAlmostNtt_Type(),
8793 stubAddr, stubName, TypePtr::BOTTOM,
8794 coeffs_start, ntt_zetas_start);
8795 // return an int
8796 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
8797 set_result(retvalue);
8798 return true;
8799 }
8800
8801 //------------------------------inline_dilithiumAlmostInverseNtt
8802 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
8803 address stubAddr;
8804 const char *stubName;
8805 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8806 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
8807
8808 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
8809 stubName = "dilithiumAlmostInverseNtt";
8810 if (!stubAddr) return false;
8811
8812 Node* coeffs = argument(0);
8813 Node* zetas = argument(1);
8814
8815 coeffs = must_be_not_null(coeffs, true);
8816 zetas = must_be_not_null(zetas, true);
8817
8818 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8819 assert(coeffs_start, "coeffs is null");
8820 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
8821 assert(zetas_start, "inverseNtt_zetas is null");
8822 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8823 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
8824 stubAddr, stubName, TypePtr::BOTTOM,
8825 coeffs_start, zetas_start);
8826 // return an int
8827 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
8828 set_result(retvalue);
8829 return true;
8830 }
8831
8832 //------------------------------inline_dilithiumNttMult
8833 bool LibraryCallKit::inline_dilithiumNttMult() {
8834 address stubAddr;
8835 const char *stubName;
8836 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8837 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
8838
8839 stubAddr = StubRoutines::dilithiumNttMult();
8840 stubName = "dilithiumNttMult";
8841 if (!stubAddr) return false;
8842
8843 Node* result = argument(0);
8844 Node* ntta = argument(1);
8845 Node* nttb = argument(2);
8846 Node* zetas = argument(3);
8847
8848 result = must_be_not_null(result, true);
8849 ntta = must_be_not_null(ntta, true);
8850 nttb = must_be_not_null(nttb, true);
8851 zetas = must_be_not_null(zetas, true);
8852
8853 Node* result_start = array_element_address(result, intcon(0), T_INT);
8854 assert(result_start, "result is null");
8855 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
8856 assert(ntta_start, "ntta is null");
8857 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
8858 assert(nttb_start, "nttb is null");
8859 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8860 OptoRuntime::dilithiumNttMult_Type(),
8861 stubAddr, stubName, TypePtr::BOTTOM,
8862 result_start, ntta_start, nttb_start);
8863
8864 // return an int
8865 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8866 set_result(retvalue);
8867
8868 return true;
8869 }
8870
8871 //------------------------------inline_dilithiumMontMulByConstant
8872 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8873 address stubAddr;
8874 const char *stubName;
8875 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8876 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8877
8878 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8879 stubName = "dilithiumMontMulByConstant";
8880 if (!stubAddr) return false;
8881
8882 Node* coeffs = argument(0);
8883 Node* constant = argument(1);
8884
8885 coeffs = must_be_not_null(coeffs, true);
8886
8887 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8888 assert(coeffs_start, "coeffs is null");
8889 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8890 OptoRuntime::dilithiumMontMulByConstant_Type(),
8891 stubAddr, stubName, TypePtr::BOTTOM,
8892 coeffs_start, constant);
8893
8894 // return an int
8895 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8896 set_result(retvalue);
8897 return true;
8898 }
8899
8900
8901 //------------------------------inline_dilithiumDecomposePoly
8902 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8903 address stubAddr;
8904 const char *stubName;
8905 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8906 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8907
8908 stubAddr = StubRoutines::dilithiumDecomposePoly();
8909 stubName = "dilithiumDecomposePoly";
8910 if (!stubAddr) return false;
8911
8912 Node* input = argument(0);
8913 Node* lowPart = argument(1);
8914 Node* highPart = argument(2);
8915 Node* twoGamma2 = argument(3);
8916 Node* multiplier = argument(4);
8917
8918 input = must_be_not_null(input, true);
8919 lowPart = must_be_not_null(lowPart, true);
8920 highPart = must_be_not_null(highPart, true);
8921
8922 Node* input_start = array_element_address(input, intcon(0), T_INT);
8923 assert(input_start, "input is null");
8924 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8925 assert(lowPart_start, "lowPart is null");
8926 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8927 assert(highPart_start, "highPart is null");
8928
8929 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8930 OptoRuntime::dilithiumDecomposePoly_Type(),
8931 stubAddr, stubName, TypePtr::BOTTOM,
8932 input_start, lowPart_start, highPart_start,
8933 twoGamma2, multiplier);
8934
8935 // return an int
8936 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8937 set_result(retvalue);
8938 return true;
8939 }
8940
8941 bool LibraryCallKit::inline_base64_encodeBlock() {
8942 address stubAddr;
8943 const char *stubName;
8944 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8945 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8946 stubAddr = StubRoutines::base64_encodeBlock();
8947 stubName = "encodeBlock";
8948
8949 if (!stubAddr) return false;
8950 Node* base64obj = argument(0);
8951 Node* src = argument(1);
8952 Node* offset = argument(2);
8953 Node* len = argument(3);
8954 Node* dest = argument(4);
8955 Node* dp = argument(5);
8956 Node* isURL = argument(6);
8957
8958 src = must_be_not_null(src, true);
8959 dest = must_be_not_null(dest, true);
8960
8961 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8962 assert(src_start, "source array is null");
8963 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8964 assert(dest_start, "destination array is null");
8965
8966 Node* base64 = make_runtime_call(RC_LEAF,
8967 OptoRuntime::base64_encodeBlock_Type(),
8968 stubAddr, stubName, TypePtr::BOTTOM,
8969 src_start, offset, len, dest_start, dp, isURL);
8970 return true;
8971 }
8972
8973 bool LibraryCallKit::inline_base64_decodeBlock() {
8974 address stubAddr;
8975 const char *stubName;
8976 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8977 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8978 stubAddr = StubRoutines::base64_decodeBlock();
8979 stubName = "decodeBlock";
8980
8981 if (!stubAddr) return false;
8982 Node* base64obj = argument(0);
8983 Node* src = argument(1);
8984 Node* src_offset = argument(2);
8985 Node* len = argument(3);
8986 Node* dest = argument(4);
8987 Node* dest_offset = argument(5);
8988 Node* isURL = argument(6);
8989 Node* isMIME = argument(7);
8990
8991 src = must_be_not_null(src, true);
8992 dest = must_be_not_null(dest, true);
8993
8994 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8995 assert(src_start, "source array is null");
8996 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8997 assert(dest_start, "destination array is null");
8998
8999 Node* call = make_runtime_call(RC_LEAF,
9000 OptoRuntime::base64_decodeBlock_Type(),
9001 stubAddr, stubName, TypePtr::BOTTOM,
9002 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
9003 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
9004 set_result(result);
9005 return true;
9006 }
9007
9008 bool LibraryCallKit::inline_poly1305_processBlocks() {
9009 address stubAddr;
9010 const char *stubName;
9011 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
9012 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
9013 stubAddr = StubRoutines::poly1305_processBlocks();
9014 stubName = "poly1305_processBlocks";
9015
9016 if (!stubAddr) return false;
9017 null_check_receiver(); // null-check receiver
9018 if (stopped()) return true;
9019
9020 Node* input = argument(1);
9021 Node* input_offset = argument(2);
9022 Node* len = argument(3);
9023 Node* alimbs = argument(4);
9024 Node* rlimbs = argument(5);
9025
9026 input = must_be_not_null(input, true);
9027 alimbs = must_be_not_null(alimbs, true);
9028 rlimbs = must_be_not_null(rlimbs, true);
9029
9030 Node* input_start = array_element_address(input, input_offset, T_BYTE);
9031 assert(input_start, "input array is null");
9032 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
9033 assert(acc_start, "acc array is null");
9034 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
9035 assert(r_start, "r array is null");
9036
9037 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
9038 OptoRuntime::poly1305_processBlocks_Type(),
9039 stubAddr, stubName, TypePtr::BOTTOM,
9040 input_start, len, acc_start, r_start);
9041 return true;
9042 }
9043
9044 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
9045 address stubAddr;
9046 const char *stubName;
9047 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
9048 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
9049 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
9050 stubName = "intpoly_montgomeryMult_P256";
9051
9052 if (!stubAddr) return false;
9053 null_check_receiver(); // null-check receiver
9054 if (stopped()) return true;
9055
9056 Node* a = argument(1);
9057 Node* b = argument(2);
9058 Node* r = argument(3);
9059
9060 a = must_be_not_null(a, true);
9061 b = must_be_not_null(b, true);
9062 r = must_be_not_null(r, true);
9063
9064 Node* a_start = array_element_address(a, intcon(0), T_LONG);
9065 assert(a_start, "a array is null");
9066 Node* b_start = array_element_address(b, intcon(0), T_LONG);
9067 assert(b_start, "b array is null");
9068 Node* r_start = array_element_address(r, intcon(0), T_LONG);
9069 assert(r_start, "r array is null");
9070
9071 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
9072 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
9073 stubAddr, stubName, TypePtr::BOTTOM,
9074 a_start, b_start, r_start);
9075 return true;
9076 }
9077
9078 bool LibraryCallKit::inline_intpoly_assign() {
9079 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
9080 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
9081 const char *stubName = "intpoly_assign";
9082 address stubAddr = StubRoutines::intpoly_assign();
9083 if (!stubAddr) return false;
9084
9085 Node* set = argument(0);
9086 Node* a = argument(1);
9087 Node* b = argument(2);
9088 Node* arr_length = load_array_length(a);
9089
9090 a = must_be_not_null(a, true);
9091 b = must_be_not_null(b, true);
9092
9093 Node* a_start = array_element_address(a, intcon(0), T_LONG);
9094 assert(a_start, "a array is null");
9095 Node* b_start = array_element_address(b, intcon(0), T_LONG);
9096 assert(b_start, "b array is null");
9097
9098 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
9099 OptoRuntime::intpoly_assign_Type(),
9100 stubAddr, stubName, TypePtr::BOTTOM,
9101 set, a_start, b_start, arr_length);
9102 return true;
9103 }
9104
9105 //------------------------------inline_digestBase_implCompress-----------------------
9106 //
9107 // Calculate MD5 for single-block byte[] array.
9108 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
9109 //
9110 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
9111 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
9112 //
9113 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
9114 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
9115 //
9116 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
9117 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
9118 //
9119 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
9120 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
9121 //
9122 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
9123 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
9124
9125 Node* digestBase_obj = argument(0);
9126 Node* src = argument(1); // type oop
9127 Node* ofs = argument(2); // type int
9128
9129 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
9130 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
9131 // failed array check
9132 return false;
9133 }
9134 // Figure out the size and type of the elements we will be copying.
9135 BasicType src_elem = src_type->elem()->array_element_basic_type();
9136 if (src_elem != T_BYTE) {
9137 return false;
9138 }
9139 // 'src_start' points to src array + offset
9140 src = must_be_not_null(src, true);
9141 Node* src_start = array_element_address(src, ofs, src_elem);
9142 Node* state = nullptr;
9143 Node* block_size = nullptr;
9144 address stubAddr;
9145 const char *stubName;
9146
9147 switch(id) {
9148 case vmIntrinsics::_md5_implCompress:
9149 assert(UseMD5Intrinsics, "need MD5 instruction support");
9150 state = get_state_from_digest_object(digestBase_obj, T_INT);
9151 stubAddr = StubRoutines::md5_implCompress();
9152 stubName = "md5_implCompress";
9153 break;
9154 case vmIntrinsics::_sha_implCompress:
9155 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
9156 state = get_state_from_digest_object(digestBase_obj, T_INT);
9157 stubAddr = StubRoutines::sha1_implCompress();
9158 stubName = "sha1_implCompress";
9159 break;
9160 case vmIntrinsics::_sha2_implCompress:
9161 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
9162 state = get_state_from_digest_object(digestBase_obj, T_INT);
9163 stubAddr = StubRoutines::sha256_implCompress();
9164 stubName = "sha256_implCompress";
9165 break;
9166 case vmIntrinsics::_sha5_implCompress:
9167 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
9168 state = get_state_from_digest_object(digestBase_obj, T_LONG);
9169 stubAddr = StubRoutines::sha512_implCompress();
9170 stubName = "sha512_implCompress";
9171 break;
9172 case vmIntrinsics::_sha3_implCompress:
9173 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
9174 state = get_state_from_digest_object(digestBase_obj, T_LONG);
9175 stubAddr = StubRoutines::sha3_implCompress();
9176 stubName = "sha3_implCompress";
9177 block_size = get_block_size_from_digest_object(digestBase_obj);
9178 if (block_size == nullptr) return false;
9179 break;
9180 default:
9181 fatal_unexpected_iid(id);
9182 return false;
9183 }
9184 if (state == nullptr) return false;
9185
9186 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
9187 if (stubAddr == nullptr) return false;
9188
9189 // Call the stub.
9190 Node* call;
9191 if (block_size == nullptr) {
9192 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
9193 stubAddr, stubName, TypePtr::BOTTOM,
9194 src_start, state);
9195 } else {
9196 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
9197 stubAddr, stubName, TypePtr::BOTTOM,
9198 src_start, state, block_size);
9199 }
9200
9201 return true;
9202 }
9203
9204 //------------------------------inline_double_keccak
9205 bool LibraryCallKit::inline_double_keccak() {
9206 address stubAddr;
9207 const char *stubName;
9208 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
9209 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
9210
9211 stubAddr = StubRoutines::double_keccak();
9212 stubName = "double_keccak";
9213 if (!stubAddr) return false;
9214
9215 Node* status0 = argument(0);
9216 Node* status1 = argument(1);
9217
9218 status0 = must_be_not_null(status0, true);
9219 status1 = must_be_not_null(status1, true);
9220
9221 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
9222 assert(status0_start, "status0 is null");
9223 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
9224 assert(status1_start, "status1 is null");
9225 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
9226 OptoRuntime::double_keccak_Type(),
9227 stubAddr, stubName, TypePtr::BOTTOM,
9228 status0_start, status1_start);
9229 // return an int
9230 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
9231 set_result(retvalue);
9232 return true;
9233 }
9234
9235
9236 //------------------------------inline_digestBase_implCompressMB-----------------------
9237 //
9238 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
9239 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
9240 //
9241 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
9242 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9243 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9244 assert((uint)predicate < 5, "sanity");
9245 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
9246
9247 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
9248 Node* src = argument(1); // byte[] array
9249 Node* ofs = argument(2); // type int
9250 Node* limit = argument(3); // type int
9251
9252 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
9253 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
9254 // failed array check
9255 return false;
9256 }
9257 // Figure out the size and type of the elements we will be copying.
9258 BasicType src_elem = src_type->elem()->array_element_basic_type();
9259 if (src_elem != T_BYTE) {
9260 return false;
9261 }
9262 // 'src_start' points to src array + offset
9263 src = must_be_not_null(src, false);
9264 Node* src_start = array_element_address(src, ofs, src_elem);
9265
9266 const char* klass_digestBase_name = nullptr;
9267 const char* stub_name = nullptr;
9268 address stub_addr = nullptr;
9269 BasicType elem_type = T_INT;
9270
9271 switch (predicate) {
9272 case 0:
9273 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
9274 klass_digestBase_name = "sun/security/provider/MD5";
9275 stub_name = "md5_implCompressMB";
9276 stub_addr = StubRoutines::md5_implCompressMB();
9277 }
9278 break;
9279 case 1:
9280 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
9281 klass_digestBase_name = "sun/security/provider/SHA";
9282 stub_name = "sha1_implCompressMB";
9283 stub_addr = StubRoutines::sha1_implCompressMB();
9284 }
9285 break;
9286 case 2:
9287 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
9288 klass_digestBase_name = "sun/security/provider/SHA2";
9289 stub_name = "sha256_implCompressMB";
9290 stub_addr = StubRoutines::sha256_implCompressMB();
9291 }
9292 break;
9293 case 3:
9294 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
9295 klass_digestBase_name = "sun/security/provider/SHA5";
9296 stub_name = "sha512_implCompressMB";
9297 stub_addr = StubRoutines::sha512_implCompressMB();
9298 elem_type = T_LONG;
9299 }
9300 break;
9301 case 4:
9302 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
9303 klass_digestBase_name = "sun/security/provider/SHA3";
9304 stub_name = "sha3_implCompressMB";
9305 stub_addr = StubRoutines::sha3_implCompressMB();
9306 elem_type = T_LONG;
9307 }
9308 break;
9309 default:
9310 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
9311 }
9312 if (klass_digestBase_name != nullptr) {
9313 assert(stub_addr != nullptr, "Stub is generated");
9314 if (stub_addr == nullptr) return false;
9315
9316 // get DigestBase klass to lookup for SHA klass
9317 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
9318 assert(tinst != nullptr, "digestBase_obj is not instance???");
9319 assert(tinst->is_loaded(), "DigestBase is not loaded");
9320
9321 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
9322 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
9323 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
9324 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
9325 }
9326 return false;
9327 }
9328
9329 //------------------------------inline_digestBase_implCompressMB-----------------------
9330 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
9331 BasicType elem_type, address stubAddr, const char *stubName,
9332 Node* src_start, Node* ofs, Node* limit) {
9333 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
9334 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
9335 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
9336 digest_obj = _gvn.transform(digest_obj);
9337
9338 Node* state = get_state_from_digest_object(digest_obj, elem_type);
9339 if (state == nullptr) return false;
9340
9341 Node* block_size = nullptr;
9342 if (strcmp("sha3_implCompressMB", stubName) == 0) {
9343 block_size = get_block_size_from_digest_object(digest_obj);
9344 if (block_size == nullptr) return false;
9345 }
9346
9347 // Call the stub.
9348 Node* call;
9349 if (block_size == nullptr) {
9350 call = make_runtime_call(RC_LEAF|RC_NO_FP,
9351 OptoRuntime::digestBase_implCompressMB_Type(false),
9352 stubAddr, stubName, TypePtr::BOTTOM,
9353 src_start, state, ofs, limit);
9354 } else {
9355 call = make_runtime_call(RC_LEAF|RC_NO_FP,
9356 OptoRuntime::digestBase_implCompressMB_Type(true),
9357 stubAddr, stubName, TypePtr::BOTTOM,
9358 src_start, state, block_size, ofs, limit);
9359 }
9360
9361 // return ofs (int)
9362 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
9363 set_result(result);
9364
9365 return true;
9366 }
9367
9368 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
9369 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
9370 assert(UseAES, "need AES instruction support");
9371 address stubAddr = nullptr;
9372 const char *stubName = nullptr;
9373 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
9374 stubName = "galoisCounterMode_AESCrypt";
9375
9376 if (stubAddr == nullptr) return false;
9377
9378 Node* in = argument(0);
9379 Node* inOfs = argument(1);
9380 Node* len = argument(2);
9381 Node* ct = argument(3);
9382 Node* ctOfs = argument(4);
9383 Node* out = argument(5);
9384 Node* outOfs = argument(6);
9385 Node* gctr_object = argument(7);
9386 Node* ghash_object = argument(8);
9387
9388 // (1) in, ct and out are arrays.
9389 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
9390 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
9391 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
9392 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
9393 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
9394 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
9395
9396 // checks are the responsibility of the caller
9397 Node* in_start = in;
9398 Node* ct_start = ct;
9399 Node* out_start = out;
9400 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
9401 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
9402 in_start = array_element_address(in, inOfs, T_BYTE);
9403 ct_start = array_element_address(ct, ctOfs, T_BYTE);
9404 out_start = array_element_address(out, outOfs, T_BYTE);
9405 }
9406
9407 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
9408 // (because of the predicated logic executed earlier).
9409 // so we cast it here safely.
9410 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
9411 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9412 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
9413 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
9414 Node* state = load_field_from_object(ghash_object, "state", "[J");
9415
9416 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
9417 return false;
9418 }
9419 // cast it to what we know it will be at runtime
9420 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
9421 assert(tinst != nullptr, "GCTR obj is null");
9422 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9423 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
9424 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
9425 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9426 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
9427 const TypeOopPtr* xtype = aklass->as_instance_type();
9428 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
9429 aescrypt_object = _gvn.transform(aescrypt_object);
9430 // we need to get the start of the aescrypt_object's expanded key array
9431 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
9432 if (k_start == nullptr) return false;
9433 // similarly, get the start address of the r vector
9434 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
9435 Node* state_start = array_element_address(state, intcon(0), T_LONG);
9436 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
9437
9438
9439 // Call the stub, passing params
9440 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
9441 OptoRuntime::galoisCounterMode_aescrypt_Type(),
9442 stubAddr, stubName, TypePtr::BOTTOM,
9443 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
9444
9445 // return cipher length (int)
9446 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
9447 set_result(retvalue);
9448
9449 return true;
9450 }
9451
9452 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
9453 // Return node representing slow path of predicate check.
9454 // the pseudo code we want to emulate with this predicate is:
9455 // for encryption:
9456 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
9457 // for decryption:
9458 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
9459 // note cipher==plain is more conservative than the original java code but that's OK
9460 //
9461
9462 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
9463 // The receiver was checked for null already.
9464 Node* objGCTR = argument(7);
9465 // Load embeddedCipher field of GCTR object.
9466 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9467 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
9468
9469 // get AESCrypt klass for instanceOf check
9470 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
9471 // will have same classloader as CipherBlockChaining object
9472 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
9473 assert(tinst != nullptr, "GCTR obj is null");
9474 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9475
9476 // we want to do an instanceof comparison against the AESCrypt class
9477 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
9478 if (!klass_AESCrypt->is_loaded()) {
9479 // if AESCrypt is not even loaded, we never take the intrinsic fast path
9480 Node* ctrl = control();
9481 set_control(top()); // no regular fast path
9482 return ctrl;
9483 }
9484
9485 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9486 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
9487 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9488 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9489 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9490
9491 return instof_false; // even if it is null
9492 }
9493
9494 //------------------------------get_state_from_digest_object-----------------------
9495 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
9496 const char* state_type;
9497 switch (elem_type) {
9498 case T_BYTE: state_type = "[B"; break;
9499 case T_INT: state_type = "[I"; break;
9500 case T_LONG: state_type = "[J"; break;
9501 default: ShouldNotReachHere();
9502 }
9503 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
9504 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
9505 if (digest_state == nullptr) return (Node *) nullptr;
9506
9507 // now have the array, need to get the start address of the state array
9508 Node* state = array_element_address(digest_state, intcon(0), elem_type);
9509 return state;
9510 }
9511
9512 //------------------------------get_block_size_from_sha3_object----------------------------------
9513 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
9514 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
9515 assert (block_size != nullptr, "sanity");
9516 return block_size;
9517 }
9518
9519 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
9520 // Return node representing slow path of predicate check.
9521 // the pseudo code we want to emulate with this predicate is:
9522 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
9523 //
9524 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
9525 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9526 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9527 assert((uint)predicate < 5, "sanity");
9528
9529 // The receiver was checked for null already.
9530 Node* digestBaseObj = argument(0);
9531
9532 // get DigestBase klass for instanceOf check
9533 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
9534 assert(tinst != nullptr, "digestBaseObj is null");
9535 assert(tinst->is_loaded(), "DigestBase is not loaded");
9536
9537 const char* klass_name = nullptr;
9538 switch (predicate) {
9539 case 0:
9540 if (UseMD5Intrinsics) {
9541 // we want to do an instanceof comparison against the MD5 class
9542 klass_name = "sun/security/provider/MD5";
9543 }
9544 break;
9545 case 1:
9546 if (UseSHA1Intrinsics) {
9547 // we want to do an instanceof comparison against the SHA class
9548 klass_name = "sun/security/provider/SHA";
9549 }
9550 break;
9551 case 2:
9552 if (UseSHA256Intrinsics) {
9553 // we want to do an instanceof comparison against the SHA2 class
9554 klass_name = "sun/security/provider/SHA2";
9555 }
9556 break;
9557 case 3:
9558 if (UseSHA512Intrinsics) {
9559 // we want to do an instanceof comparison against the SHA5 class
9560 klass_name = "sun/security/provider/SHA5";
9561 }
9562 break;
9563 case 4:
9564 if (UseSHA3Intrinsics) {
9565 // we want to do an instanceof comparison against the SHA3 class
9566 klass_name = "sun/security/provider/SHA3";
9567 }
9568 break;
9569 default:
9570 fatal("unknown SHA intrinsic predicate: %d", predicate);
9571 }
9572
9573 ciKlass* klass = nullptr;
9574 if (klass_name != nullptr) {
9575 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
9576 }
9577 if ((klass == nullptr) || !klass->is_loaded()) {
9578 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
9579 Node* ctrl = control();
9580 set_control(top()); // no intrinsic path
9581 return ctrl;
9582 }
9583 ciInstanceKlass* instklass = klass->as_instance_klass();
9584
9585 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
9586 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9587 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9588 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9589
9590 return instof_false; // even if it is null
9591 }
9592
9593 //-------------inline_fma-----------------------------------
9594 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
9595 Node *a = nullptr;
9596 Node *b = nullptr;
9597 Node *c = nullptr;
9598 Node* result = nullptr;
9599 switch (id) {
9600 case vmIntrinsics::_fmaD:
9601 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
9602 // no receiver since it is static method
9603 a = argument(0);
9604 b = argument(2);
9605 c = argument(4);
9606 result = _gvn.transform(new FmaDNode(a, b, c));
9607 break;
9608 case vmIntrinsics::_fmaF:
9609 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
9610 a = argument(0);
9611 b = argument(1);
9612 c = argument(2);
9613 result = _gvn.transform(new FmaFNode(a, b, c));
9614 break;
9615 default:
9616 fatal_unexpected_iid(id); break;
9617 }
9618 set_result(result);
9619 return true;
9620 }
9621
9622 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
9623 // argument(0) is receiver
9624 Node* codePoint = argument(1);
9625 Node* n = nullptr;
9626
9627 switch (id) {
9628 case vmIntrinsics::_isDigit :
9629 n = new DigitNode(control(), codePoint);
9630 break;
9631 case vmIntrinsics::_isLowerCase :
9632 n = new LowerCaseNode(control(), codePoint);
9633 break;
9634 case vmIntrinsics::_isUpperCase :
9635 n = new UpperCaseNode(control(), codePoint);
9636 break;
9637 case vmIntrinsics::_isWhitespace :
9638 n = new WhitespaceNode(control(), codePoint);
9639 break;
9640 default:
9641 fatal_unexpected_iid(id);
9642 }
9643
9644 set_result(_gvn.transform(n));
9645 return true;
9646 }
9647
9648 bool LibraryCallKit::inline_profileBoolean() {
9649 Node* counts = argument(1);
9650 const TypeAryPtr* ary = nullptr;
9651 ciArray* aobj = nullptr;
9652 if (counts->is_Con()
9653 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
9654 && (aobj = ary->const_oop()->as_array()) != nullptr
9655 && (aobj->length() == 2)) {
9656 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
9657 jint false_cnt = aobj->element_value(0).as_int();
9658 jint true_cnt = aobj->element_value(1).as_int();
9659
9660 if (C->log() != nullptr) {
9661 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
9662 false_cnt, true_cnt);
9663 }
9664
9665 if (false_cnt + true_cnt == 0) {
9666 // According to profile, never executed.
9667 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9668 Deoptimization::Action_reinterpret);
9669 return true;
9670 }
9671
9672 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
9673 // is a number of each value occurrences.
9674 Node* result = argument(0);
9675 if (false_cnt == 0 || true_cnt == 0) {
9676 // According to profile, one value has been never seen.
9677 int expected_val = (false_cnt == 0) ? 1 : 0;
9678
9679 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
9680 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
9681
9682 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
9683 Node* fast_path = _gvn.transform(new IfTrueNode(check));
9684 Node* slow_path = _gvn.transform(new IfFalseNode(check));
9685
9686 { // Slow path: uncommon trap for never seen value and then reexecute
9687 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
9688 // the value has been seen at least once.
9689 PreserveJVMState pjvms(this);
9690 PreserveReexecuteState preexecs(this);
9691 jvms()->set_should_reexecute(true);
9692
9693 set_control(slow_path);
9694 set_i_o(i_o());
9695
9696 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9697 Deoptimization::Action_reinterpret);
9698 }
9699 // The guard for never seen value enables sharpening of the result and
9700 // returning a constant. It allows to eliminate branches on the same value
9701 // later on.
9702 set_control(fast_path);
9703 result = intcon(expected_val);
9704 }
9705 // Stop profiling.
9706 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
9707 // By replacing method body with profile data (represented as ProfileBooleanNode
9708 // on IR level) we effectively disable profiling.
9709 // It enables full speed execution once optimized code is generated.
9710 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
9711 C->record_for_igvn(profile);
9712 set_result(profile);
9713 return true;
9714 } else {
9715 // Continue profiling.
9716 // Profile data isn't available at the moment. So, execute method's bytecode version.
9717 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
9718 // is compiled and counters aren't available since corresponding MethodHandle
9719 // isn't a compile-time constant.
9720 return false;
9721 }
9722 }
9723
9724 bool LibraryCallKit::inline_isCompileConstant() {
9725 Node* n = argument(0);
9726 set_result(n->is_Con() ? intcon(1) : intcon(0));
9727 return true;
9728 }
9729
9730 //------------------------------- inline_getObjectSize --------------------------------------
9731 //
9732 // Calculate the runtime size of the object/array.
9733 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
9734 //
9735 bool LibraryCallKit::inline_getObjectSize() {
9736 Node* obj = argument(3);
9737 Node* klass_node = load_object_klass(obj);
9738
9739 jint layout_con = Klass::_lh_neutral_value;
9740 Node* layout_val = get_layout_helper(klass_node, layout_con);
9741 int layout_is_con = (layout_val == nullptr);
9742
9743 if (layout_is_con) {
9744 // Layout helper is constant, can figure out things at compile time.
9745
9746 if (Klass::layout_helper_is_instance(layout_con)) {
9747 // Instance case: layout_con contains the size itself.
9748 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
9749 set_result(size);
9750 } else {
9751 // Array case: size is round(header + element_size*arraylength).
9752 // Since arraylength is different for every array instance, we have to
9753 // compute the whole thing at runtime.
9754
9755 Node* arr_length = load_array_length(obj);
9756
9757 int round_mask = MinObjAlignmentInBytes - 1;
9758 int hsize = Klass::layout_helper_header_size(layout_con);
9759 int eshift = Klass::layout_helper_log2_element_size(layout_con);
9760
9761 if ((round_mask & ~right_n_bits(eshift)) == 0) {
9762 round_mask = 0; // strength-reduce it if it goes away completely
9763 }
9764 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
9765 Node* header_size = intcon(hsize + round_mask);
9766
9767 Node* lengthx = ConvI2X(arr_length);
9768 Node* headerx = ConvI2X(header_size);
9769
9770 Node* abody = lengthx;
9771 if (eshift != 0) {
9772 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
9773 }
9774 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
9775 if (round_mask != 0) {
9776 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
9777 }
9778 size = ConvX2L(size);
9779 set_result(size);
9780 }
9781 } else {
9782 // Layout helper is not constant, need to test for array-ness at runtime.
9783
9784 enum { _instance_path = 1, _array_path, PATH_LIMIT };
9785 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
9786 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
9787 record_for_igvn(result_reg);
9788
9789 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
9790 if (array_ctl != nullptr) {
9791 // Array case: size is round(header + element_size*arraylength).
9792 // Since arraylength is different for every array instance, we have to
9793 // compute the whole thing at runtime.
9794
9795 PreserveJVMState pjvms(this);
9796 set_control(array_ctl);
9797 Node* arr_length = load_array_length(obj);
9798
9799 int round_mask = MinObjAlignmentInBytes - 1;
9800 Node* mask = intcon(round_mask);
9801
9802 Node* hss = intcon(Klass::_lh_header_size_shift);
9803 Node* hsm = intcon(Klass::_lh_header_size_mask);
9804 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
9805 header_size = _gvn.transform(new AndINode(header_size, hsm));
9806 header_size = _gvn.transform(new AddINode(header_size, mask));
9807
9808 // There is no need to mask or shift this value.
9809 // The semantics of LShiftINode include an implicit mask to 0x1F.
9810 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
9811 Node* elem_shift = layout_val;
9812
9813 Node* lengthx = ConvI2X(arr_length);
9814 Node* headerx = ConvI2X(header_size);
9815
9816 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
9817 Node* size = _gvn.transform(new AddXNode(headerx, abody));
9818 if (round_mask != 0) {
9819 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
9820 }
9821 size = ConvX2L(size);
9822
9823 result_reg->init_req(_array_path, control());
9824 result_val->init_req(_array_path, size);
9825 }
9826
9827 if (!stopped()) {
9828 // Instance case: the layout helper gives us instance size almost directly,
9829 // but we need to mask out the _lh_instance_slow_path_bit.
9830 Node* size = ConvI2X(layout_val);
9831 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
9832 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
9833 size = _gvn.transform(new AndXNode(size, mask));
9834 size = ConvX2L(size);
9835
9836 result_reg->init_req(_instance_path, control());
9837 result_val->init_req(_instance_path, size);
9838 }
9839
9840 set_result(result_reg, result_val);
9841 }
9842
9843 return true;
9844 }
9845
9846 //------------------------------- inline_blackhole --------------------------------------
9847 //
9848 // Make sure all arguments to this node are alive.
9849 // This matches methods that were requested to be blackholed through compile commands.
9850 //
9851 bool LibraryCallKit::inline_blackhole() {
9852 assert(callee()->is_static(), "Should have been checked before: only static methods here");
9853 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
9854 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
9855
9856 // Blackhole node pinches only the control, not memory. This allows
9857 // the blackhole to be pinned in the loop that computes blackholed
9858 // values, but have no other side effects, like breaking the optimizations
9859 // across the blackhole.
9860
9861 Node* bh = _gvn.transform(new BlackholeNode(control()));
9862 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
9863
9864 // Bind call arguments as blackhole arguments to keep them alive
9865 uint nargs = callee()->arg_size();
9866 for (uint i = 0; i < nargs; i++) {
9867 bh->add_req(argument(i));
9868 }
9869
9870 return true;
9871 }
9872
9873 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9874 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9875 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9876 return nullptr; // box klass is not Float16
9877 }
9878
9879 // Null check; get notnull casted pointer
9880 Node* null_ctl = top();
9881 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9882 // If not_null_box is dead, only null-path is taken
9883 if (stopped()) {
9884 set_control(null_ctl);
9885 return nullptr;
9886 }
9887 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9888 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9889 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9890 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9891 }
9892
9893 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9894 PreserveReexecuteState preexecs(this);
9895 jvms()->set_should_reexecute(true);
9896
9897 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9898 Node* klass_node = makecon(klass_type);
9899 Node* box = new_instance(klass_node);
9900
9901 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9902 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9903
9904 Node* field_store = _gvn.transform(access_store_at(box,
9905 value_field,
9906 value_adr_type,
9907 value,
9908 TypeInt::SHORT,
9909 T_SHORT,
9910 IN_HEAP));
9911 set_memory(field_store, value_adr_type);
9912 return box;
9913 }
9914
9915 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9916 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9917 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9918 return false;
9919 }
9920
9921 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9922 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9923 return false;
9924 }
9925
9926 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9927 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9928 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9929 ciSymbols::short_signature(),
9930 false);
9931 assert(field != nullptr, "");
9932
9933 // Transformed nodes
9934 Node* fld1 = nullptr;
9935 Node* fld2 = nullptr;
9936 Node* fld3 = nullptr;
9937 switch(num_args) {
9938 case 3:
9939 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9940 if (fld3 == nullptr) {
9941 return false;
9942 }
9943 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9944 // fall-through
9945 case 2:
9946 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9947 if (fld2 == nullptr) {
9948 return false;
9949 }
9950 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9951 // fall-through
9952 case 1:
9953 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9954 if (fld1 == nullptr) {
9955 return false;
9956 }
9957 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9958 break;
9959 default: fatal("Unsupported number of arguments %d", num_args);
9960 }
9961
9962 Node* result = nullptr;
9963 switch (id) {
9964 // Unary operations
9965 case vmIntrinsics::_sqrt_float16:
9966 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9967 break;
9968 // Ternary operations
9969 case vmIntrinsics::_fma_float16:
9970 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9971 break;
9972 default:
9973 fatal_unexpected_iid(id);
9974 break;
9975 }
9976 result = _gvn.transform(new ReinterpretHF2SNode(result));
9977 set_result(box_fp16_value(float16_box_type, field, result));
9978 return true;
9979 }
9980