1 /*
2 * Copyright (c) 1999, 2024, 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 "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "ci/ciFlatArrayKlass.hpp"
28 #include "ci/ciUtilities.inline.hpp"
29 #include "classfile/vmIntrinsics.hpp"
30 #include "compiler/compileBroker.hpp"
31 #include "compiler/compileLog.hpp"
32 #include "gc/shared/barrierSet.hpp"
33 #include "jfr/support/jfrIntrinsics.hpp"
34 #include "memory/resourceArea.hpp"
35 #include "oops/klass.inline.hpp"
36 #include "oops/objArrayKlass.hpp"
37 #include "opto/addnode.hpp"
38 #include "opto/arraycopynode.hpp"
39 #include "opto/c2compiler.hpp"
40 #include "opto/castnode.hpp"
41 #include "opto/cfgnode.hpp"
42 #include "opto/convertnode.hpp"
43 #include "opto/countbitsnode.hpp"
44 #include "opto/idealKit.hpp"
45 #include "opto/library_call.hpp"
46 #include "opto/mathexactnode.hpp"
47 #include "opto/mulnode.hpp"
48 #include "opto/narrowptrnode.hpp"
49 #include "opto/opaquenode.hpp"
50 #include "opto/parse.hpp"
51 #include "opto/runtime.hpp"
52 #include "opto/rootnode.hpp"
53 #include "opto/subnode.hpp"
54 #include "prims/jvmtiExport.hpp"
55 #include "prims/jvmtiThreadState.hpp"
56 #include "prims/unsafe.hpp"
57 #include "runtime/jniHandles.inline.hpp"
58 #include "runtime/objectMonitor.hpp"
59 #include "runtime/sharedRuntime.hpp"
60 #include "runtime/stubRoutines.hpp"
61 #include "utilities/macros.hpp"
62 #include "utilities/powerOfTwo.hpp"
63
64 //---------------------------make_vm_intrinsic----------------------------
65 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
66 vmIntrinsicID id = m->intrinsic_id();
67 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
68
69 if (!m->is_loaded()) {
70 // Do not attempt to inline unloaded methods.
71 return nullptr;
72 }
73
74 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
75 bool is_available = false;
76
77 {
78 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
79 // the compiler must transition to '_thread_in_vm' state because both
80 // methods access VM-internal data.
81 VM_ENTRY_MARK;
82 methodHandle mh(THREAD, m->get_Method());
83 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
84 if (is_available && is_virtual) {
85 is_available = vmIntrinsics::does_virtual_dispatch(id);
86 }
87 }
88
89 if (is_available) {
90 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
91 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
92 return new LibraryIntrinsic(m, is_virtual,
93 vmIntrinsics::predicates_needed(id),
94 vmIntrinsics::does_virtual_dispatch(id),
95 id);
96 } else {
97 return nullptr;
98 }
99 }
100
101 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
102 LibraryCallKit kit(jvms, this);
103 Compile* C = kit.C;
104 int nodes = C->unique();
105 #ifndef PRODUCT
106 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
107 char buf[1000];
108 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
109 tty->print_cr("Intrinsic %s", str);
110 }
111 #endif
112 ciMethod* callee = kit.callee();
113 const int bci = kit.bci();
114 #ifdef ASSERT
115 Node* ctrl = kit.control();
116 #endif
117 // Try to inline the intrinsic.
118 if (callee->check_intrinsic_candidate() &&
119 kit.try_to_inline(_last_predicate)) {
120 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
121 : "(intrinsic)";
122 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
123 if (C->print_intrinsics() || C->print_inlining()) {
124 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
125 }
126 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
127 if (C->log()) {
128 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
129 vmIntrinsics::name_at(intrinsic_id()),
130 (is_virtual() ? " virtual='1'" : ""),
131 C->unique() - nodes);
132 }
133 // Push the result from the inlined method onto the stack.
134 kit.push_result();
135 C->print_inlining_update(this);
136 return kit.transfer_exceptions_into_jvms();
137 }
138
139 // The intrinsic bailed out
140 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
141 if (jvms->has_method()) {
142 // Not a root compile.
143 const char* msg;
144 if (callee->intrinsic_candidate()) {
145 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
146 } else {
147 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
148 : "failed to inline (intrinsic), method not annotated";
149 }
150 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
151 if (C->print_intrinsics() || C->print_inlining()) {
152 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
153 }
154 } else {
155 // Root compile
156 ResourceMark rm;
157 stringStream msg_stream;
158 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
159 vmIntrinsics::name_at(intrinsic_id()),
160 is_virtual() ? " (virtual)" : "", bci);
161 const char *msg = msg_stream.freeze();
162 log_debug(jit, inlining)("%s", msg);
163 if (C->print_intrinsics() || C->print_inlining()) {
164 tty->print("%s", msg);
165 }
166 }
167 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
168 C->print_inlining_update(this);
169
170 return nullptr;
171 }
172
173 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
174 LibraryCallKit kit(jvms, this);
175 Compile* C = kit.C;
176 int nodes = C->unique();
177 _last_predicate = predicate;
178 #ifndef PRODUCT
179 assert(is_predicated() && predicate < predicates_count(), "sanity");
180 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
181 char buf[1000];
182 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
183 tty->print_cr("Predicate for intrinsic %s", str);
184 }
185 #endif
186 ciMethod* callee = kit.callee();
187 const int bci = kit.bci();
188
189 Node* slow_ctl = kit.try_to_predicate(predicate);
190 if (!kit.failing()) {
191 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
192 : "(intrinsic, predicate)";
193 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
194 if (C->print_intrinsics() || C->print_inlining()) {
195 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
196 }
197 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
198 if (C->log()) {
199 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
200 vmIntrinsics::name_at(intrinsic_id()),
201 (is_virtual() ? " virtual='1'" : ""),
202 C->unique() - nodes);
203 }
204 return slow_ctl; // Could be null if the check folds.
205 }
206
207 // The intrinsic bailed out
208 if (jvms->has_method()) {
209 // Not a root compile.
210 const char* msg = "failed to generate predicate for intrinsic";
211 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
212 if (C->print_intrinsics() || C->print_inlining()) {
213 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
214 }
215 } else {
216 // Root compile
217 ResourceMark rm;
218 stringStream msg_stream;
219 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
220 vmIntrinsics::name_at(intrinsic_id()),
221 is_virtual() ? " (virtual)" : "", bci);
222 const char *msg = msg_stream.freeze();
223 log_debug(jit, inlining)("%s", msg);
224 if (C->print_intrinsics() || C->print_inlining()) {
225 C->print_inlining_stream()->print("%s", msg);
226 }
227 }
228 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
229 return nullptr;
230 }
231
232 bool LibraryCallKit::try_to_inline(int predicate) {
233 // Handle symbolic names for otherwise undistinguished boolean switches:
234 const bool is_store = true;
235 const bool is_compress = true;
236 const bool is_static = true;
237 const bool is_volatile = true;
238
239 if (!jvms()->has_method()) {
240 // Root JVMState has a null method.
241 assert(map()->memory()->Opcode() == Op_Parm, "");
242 // Insert the memory aliasing node
243 set_all_memory(reset_memory());
244 }
245 assert(merged_memory(), "");
246
247 switch (intrinsic_id()) {
248 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
249 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
250 case vmIntrinsics::_getClass: return inline_native_getClass();
251
252 case vmIntrinsics::_ceil:
253 case vmIntrinsics::_floor:
254 case vmIntrinsics::_rint:
255 case vmIntrinsics::_dsin:
256 case vmIntrinsics::_dcos:
257 case vmIntrinsics::_dtan:
258 case vmIntrinsics::_dabs:
259 case vmIntrinsics::_fabs:
260 case vmIntrinsics::_iabs:
261 case vmIntrinsics::_labs:
262 case vmIntrinsics::_datan2:
263 case vmIntrinsics::_dsqrt:
264 case vmIntrinsics::_dsqrt_strict:
265 case vmIntrinsics::_dexp:
266 case vmIntrinsics::_dlog:
267 case vmIntrinsics::_dlog10:
268 case vmIntrinsics::_dpow:
269 case vmIntrinsics::_dcopySign:
270 case vmIntrinsics::_fcopySign:
271 case vmIntrinsics::_dsignum:
272 case vmIntrinsics::_roundF:
273 case vmIntrinsics::_roundD:
274 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
275
276 case vmIntrinsics::_notify:
277 case vmIntrinsics::_notifyAll:
278 return inline_notify(intrinsic_id());
279
280 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
281 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
282 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
283 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
284 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
285 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
286 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
287 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
288 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
289 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
290 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
291 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
292 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
293 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
294
295 case vmIntrinsics::_arraycopy: return inline_arraycopy();
296
297 case vmIntrinsics::_arraySort: return inline_array_sort();
298 case vmIntrinsics::_arrayPartition: return inline_array_partition();
299
300 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
301 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
302 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
303 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
304
305 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
306 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
307 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
308 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
309 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
310 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
311 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
312 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
313
314 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
315
316 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
317
318 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
319 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
320 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
321 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
322
323 case vmIntrinsics::_compressStringC:
324 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
325 case vmIntrinsics::_inflateStringC:
326 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
327
328 case vmIntrinsics::_makePrivateBuffer: return inline_unsafe_make_private_buffer();
329 case vmIntrinsics::_finishPrivateBuffer: return inline_unsafe_finish_private_buffer();
330 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
331 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
332 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
333 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
334 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
335 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
336 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
337 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
338 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
339 case vmIntrinsics::_getValue: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false, true);
340
341 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
342 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
343 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
344 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
345 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
346 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
347 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
348 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
349 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
350 case vmIntrinsics::_putValue: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false, true);
351
352 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
353 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
354 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
355 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
356 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
357 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
358 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
359 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
360 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
361
362 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
363 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
364 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
365 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
366 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
367 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
368 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
369 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
370 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
371
372 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
373 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
374 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
375 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
376
377 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
378 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
379 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
380 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
381
382 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
383 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
384 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
385 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
386 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
387 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
388 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
389 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
390 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
391
392 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
393 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
394 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
395 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
396 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
397 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
398 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
399 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
400 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
401
402 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
403 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
404 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
405 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
406 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
407 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
408 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
409 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
410 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
411
412 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
413 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
414 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
415 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
416 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
417 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
418 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
419 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
420 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
421
422 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
423 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
424 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
425 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
426 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
427
428 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
429 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
430 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
431 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
432 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
433 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
434 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
435 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
436 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
437 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
438 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
439 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
440 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
441 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
442 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
443 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
444 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
445 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
446 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
447 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
448
449 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
450 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
451 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
452 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
453 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
454 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
455 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
456 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
457 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
458 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
459 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
460 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
461 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
462 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
463 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
464
465 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
466 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
467 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
468 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
469
470 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
471 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
472 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
473 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
474 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
475
476 case vmIntrinsics::_loadFence:
477 case vmIntrinsics::_storeFence:
478 case vmIntrinsics::_storeStoreFence:
479 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
480
481 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
482
483 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
484 case vmIntrinsics::_currentThread: return inline_native_currentThread();
485 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
486
487 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
488 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
489
490 #if INCLUDE_JVMTI
491 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
492 "notifyJvmtiStart", true, false);
493 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
494 "notifyJvmtiEnd", false, true);
495 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
496 "notifyJvmtiMount", false, false);
497 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
498 "notifyJvmtiUnmount", false, false);
499 case vmIntrinsics::_notifyJvmtiVThreadHideFrames: return inline_native_notify_jvmti_hide();
500 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
501 #endif
502
503 #ifdef JFR_HAVE_INTRINSICS
504 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
505 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
506 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
507 #endif
508 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
509 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
510 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
511 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
512 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
513 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
514 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
515 case vmIntrinsics::_isFlatArray: return inline_unsafe_isFlatArray();
516 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
517 case vmIntrinsics::_getLength: return inline_native_getLength();
518 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
519 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
520 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
521 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
522 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
523 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
524 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
525
526 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
527 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
528 case vmIntrinsics::_newNullRestrictedArray: return inline_newNullRestrictedArray();
529
530 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
531
532 case vmIntrinsics::_isInstance:
533 case vmIntrinsics::_getModifiers:
534 case vmIntrinsics::_isInterface:
535 case vmIntrinsics::_isArray:
536 case vmIntrinsics::_isPrimitive:
537 case vmIntrinsics::_isHidden:
538 case vmIntrinsics::_getSuperclass:
539 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
540
541 case vmIntrinsics::_floatToRawIntBits:
542 case vmIntrinsics::_floatToIntBits:
543 case vmIntrinsics::_intBitsToFloat:
544 case vmIntrinsics::_doubleToRawLongBits:
545 case vmIntrinsics::_doubleToLongBits:
546 case vmIntrinsics::_longBitsToDouble:
547 case vmIntrinsics::_floatToFloat16:
548 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
549
550 case vmIntrinsics::_floatIsFinite:
551 case vmIntrinsics::_floatIsInfinite:
552 case vmIntrinsics::_doubleIsFinite:
553 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
554
555 case vmIntrinsics::_numberOfLeadingZeros_i:
556 case vmIntrinsics::_numberOfLeadingZeros_l:
557 case vmIntrinsics::_numberOfTrailingZeros_i:
558 case vmIntrinsics::_numberOfTrailingZeros_l:
559 case vmIntrinsics::_bitCount_i:
560 case vmIntrinsics::_bitCount_l:
561 case vmIntrinsics::_reverse_i:
562 case vmIntrinsics::_reverse_l:
563 case vmIntrinsics::_reverseBytes_i:
564 case vmIntrinsics::_reverseBytes_l:
565 case vmIntrinsics::_reverseBytes_s:
566 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
567
568 case vmIntrinsics::_compress_i:
569 case vmIntrinsics::_compress_l:
570 case vmIntrinsics::_expand_i:
571 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
572
573 case vmIntrinsics::_compareUnsigned_i:
574 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
575
576 case vmIntrinsics::_divideUnsigned_i:
577 case vmIntrinsics::_divideUnsigned_l:
578 case vmIntrinsics::_remainderUnsigned_i:
579 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
580
581 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
582
583 case vmIntrinsics::_Reference_get: return inline_reference_get();
584 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
585 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
586
587 case vmIntrinsics::_Class_cast: return inline_Class_cast();
588
589 case vmIntrinsics::_aescrypt_encryptBlock:
590 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
591
592 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
593 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
594 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
595
596 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
597 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
598 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
599
600 case vmIntrinsics::_counterMode_AESCrypt:
601 return inline_counterMode_AESCrypt(intrinsic_id());
602
603 case vmIntrinsics::_galoisCounterMode_AESCrypt:
604 return inline_galoisCounterMode_AESCrypt();
605
606 case vmIntrinsics::_md5_implCompress:
607 case vmIntrinsics::_sha_implCompress:
608 case vmIntrinsics::_sha2_implCompress:
609 case vmIntrinsics::_sha5_implCompress:
610 case vmIntrinsics::_sha3_implCompress:
611 return inline_digestBase_implCompress(intrinsic_id());
612
613 case vmIntrinsics::_digestBase_implCompressMB:
614 return inline_digestBase_implCompressMB(predicate);
615
616 case vmIntrinsics::_multiplyToLen:
617 return inline_multiplyToLen();
618
619 case vmIntrinsics::_squareToLen:
620 return inline_squareToLen();
621
622 case vmIntrinsics::_mulAdd:
623 return inline_mulAdd();
624
625 case vmIntrinsics::_montgomeryMultiply:
626 return inline_montgomeryMultiply();
627 case vmIntrinsics::_montgomerySquare:
628 return inline_montgomerySquare();
629
630 case vmIntrinsics::_bigIntegerRightShiftWorker:
631 return inline_bigIntegerShift(true);
632 case vmIntrinsics::_bigIntegerLeftShiftWorker:
633 return inline_bigIntegerShift(false);
634
635 case vmIntrinsics::_vectorizedMismatch:
636 return inline_vectorizedMismatch();
637
638 case vmIntrinsics::_ghash_processBlocks:
639 return inline_ghash_processBlocks();
640 case vmIntrinsics::_chacha20Block:
641 return inline_chacha20Block();
642 case vmIntrinsics::_base64_encodeBlock:
643 return inline_base64_encodeBlock();
644 case vmIntrinsics::_base64_decodeBlock:
645 return inline_base64_decodeBlock();
646 case vmIntrinsics::_poly1305_processBlocks:
647 return inline_poly1305_processBlocks();
648 case vmIntrinsics::_intpoly_montgomeryMult_P256:
649 return inline_intpoly_montgomeryMult_P256();
650 case vmIntrinsics::_intpoly_assign:
651 return inline_intpoly_assign();
652 case vmIntrinsics::_encodeISOArray:
653 case vmIntrinsics::_encodeByteISOArray:
654 return inline_encodeISOArray(false);
655 case vmIntrinsics::_encodeAsciiArray:
656 return inline_encodeISOArray(true);
657
658 case vmIntrinsics::_updateCRC32:
659 return inline_updateCRC32();
660 case vmIntrinsics::_updateBytesCRC32:
661 return inline_updateBytesCRC32();
662 case vmIntrinsics::_updateByteBufferCRC32:
663 return inline_updateByteBufferCRC32();
664
665 case vmIntrinsics::_updateBytesCRC32C:
666 return inline_updateBytesCRC32C();
667 case vmIntrinsics::_updateDirectByteBufferCRC32C:
668 return inline_updateDirectByteBufferCRC32C();
669
670 case vmIntrinsics::_updateBytesAdler32:
671 return inline_updateBytesAdler32();
672 case vmIntrinsics::_updateByteBufferAdler32:
673 return inline_updateByteBufferAdler32();
674
675 case vmIntrinsics::_profileBoolean:
676 return inline_profileBoolean();
677 case vmIntrinsics::_isCompileConstant:
678 return inline_isCompileConstant();
679
680 case vmIntrinsics::_countPositives:
681 return inline_countPositives();
682
683 case vmIntrinsics::_fmaD:
684 case vmIntrinsics::_fmaF:
685 return inline_fma(intrinsic_id());
686
687 case vmIntrinsics::_isDigit:
688 case vmIntrinsics::_isLowerCase:
689 case vmIntrinsics::_isUpperCase:
690 case vmIntrinsics::_isWhitespace:
691 return inline_character_compare(intrinsic_id());
692
693 case vmIntrinsics::_min:
694 case vmIntrinsics::_max:
695 case vmIntrinsics::_min_strict:
696 case vmIntrinsics::_max_strict:
697 return inline_min_max(intrinsic_id());
698
699 case vmIntrinsics::_maxF:
700 case vmIntrinsics::_minF:
701 case vmIntrinsics::_maxD:
702 case vmIntrinsics::_minD:
703 case vmIntrinsics::_maxF_strict:
704 case vmIntrinsics::_minF_strict:
705 case vmIntrinsics::_maxD_strict:
706 case vmIntrinsics::_minD_strict:
707 return inline_fp_min_max(intrinsic_id());
708
709 case vmIntrinsics::_VectorUnaryOp:
710 return inline_vector_nary_operation(1);
711 case vmIntrinsics::_VectorBinaryOp:
712 return inline_vector_nary_operation(2);
713 case vmIntrinsics::_VectorTernaryOp:
714 return inline_vector_nary_operation(3);
715 case vmIntrinsics::_VectorFromBitsCoerced:
716 return inline_vector_frombits_coerced();
717 case vmIntrinsics::_VectorShuffleIota:
718 return inline_vector_shuffle_iota();
719 case vmIntrinsics::_VectorMaskOp:
720 return inline_vector_mask_operation();
721 case vmIntrinsics::_VectorShuffleToVector:
722 return inline_vector_shuffle_to_vector();
723 case vmIntrinsics::_VectorLoadOp:
724 return inline_vector_mem_operation(/*is_store=*/false);
725 case vmIntrinsics::_VectorLoadMaskedOp:
726 return inline_vector_mem_masked_operation(/*is_store*/false);
727 case vmIntrinsics::_VectorStoreOp:
728 return inline_vector_mem_operation(/*is_store=*/true);
729 case vmIntrinsics::_VectorStoreMaskedOp:
730 return inline_vector_mem_masked_operation(/*is_store=*/true);
731 case vmIntrinsics::_VectorGatherOp:
732 return inline_vector_gather_scatter(/*is_scatter*/ false);
733 case vmIntrinsics::_VectorScatterOp:
734 return inline_vector_gather_scatter(/*is_scatter*/ true);
735 case vmIntrinsics::_VectorReductionCoerced:
736 return inline_vector_reduction();
737 case vmIntrinsics::_VectorTest:
738 return inline_vector_test();
739 case vmIntrinsics::_VectorBlend:
740 return inline_vector_blend();
741 case vmIntrinsics::_VectorRearrange:
742 return inline_vector_rearrange();
743 case vmIntrinsics::_VectorCompare:
744 return inline_vector_compare();
745 case vmIntrinsics::_VectorBroadcastInt:
746 return inline_vector_broadcast_int();
747 case vmIntrinsics::_VectorConvert:
748 return inline_vector_convert();
749 case vmIntrinsics::_VectorInsert:
750 return inline_vector_insert();
751 case vmIntrinsics::_VectorExtract:
752 return inline_vector_extract();
753 case vmIntrinsics::_VectorCompressExpand:
754 return inline_vector_compress_expand();
755 case vmIntrinsics::_IndexVector:
756 return inline_index_vector();
757 case vmIntrinsics::_IndexPartiallyInUpperRange:
758 return inline_index_partially_in_upper_range();
759
760 case vmIntrinsics::_getObjectSize:
761 return inline_getObjectSize();
762
763 case vmIntrinsics::_blackhole:
764 return inline_blackhole();
765
766 default:
767 // If you get here, it may be that someone has added a new intrinsic
768 // to the list in vmIntrinsics.hpp without implementing it here.
769 #ifndef PRODUCT
770 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
771 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
772 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
773 }
774 #endif
775 return false;
776 }
777 }
778
779 Node* LibraryCallKit::try_to_predicate(int predicate) {
780 if (!jvms()->has_method()) {
781 // Root JVMState has a null method.
782 assert(map()->memory()->Opcode() == Op_Parm, "");
783 // Insert the memory aliasing node
784 set_all_memory(reset_memory());
785 }
786 assert(merged_memory(), "");
787
788 switch (intrinsic_id()) {
789 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
790 return inline_cipherBlockChaining_AESCrypt_predicate(false);
791 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
792 return inline_cipherBlockChaining_AESCrypt_predicate(true);
793 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
794 return inline_electronicCodeBook_AESCrypt_predicate(false);
795 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
796 return inline_electronicCodeBook_AESCrypt_predicate(true);
797 case vmIntrinsics::_counterMode_AESCrypt:
798 return inline_counterMode_AESCrypt_predicate();
799 case vmIntrinsics::_digestBase_implCompressMB:
800 return inline_digestBase_implCompressMB_predicate(predicate);
801 case vmIntrinsics::_galoisCounterMode_AESCrypt:
802 return inline_galoisCounterMode_AESCrypt_predicate();
803
804 default:
805 // If you get here, it may be that someone has added a new intrinsic
806 // to the list in vmIntrinsics.hpp without implementing it here.
807 #ifndef PRODUCT
808 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
809 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
810 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
811 }
812 #endif
813 Node* slow_ctl = control();
814 set_control(top()); // No fast path intrinsic
815 return slow_ctl;
816 }
817 }
818
819 //------------------------------set_result-------------------------------
820 // Helper function for finishing intrinsics.
821 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
822 record_for_igvn(region);
823 set_control(_gvn.transform(region));
824 set_result( _gvn.transform(value));
825 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
826 }
827
828 //------------------------------generate_guard---------------------------
829 // Helper function for generating guarded fast-slow graph structures.
830 // The given 'test', if true, guards a slow path. If the test fails
831 // then a fast path can be taken. (We generally hope it fails.)
832 // In all cases, GraphKit::control() is updated to the fast path.
833 // The returned value represents the control for the slow path.
834 // The return value is never 'top'; it is either a valid control
835 // or null if it is obvious that the slow path can never be taken.
836 // Also, if region and the slow control are not null, the slow edge
837 // is appended to the region.
838 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
839 if (stopped()) {
840 // Already short circuited.
841 return nullptr;
842 }
843
844 // Build an if node and its projections.
845 // If test is true we take the slow path, which we assume is uncommon.
846 if (_gvn.type(test) == TypeInt::ZERO) {
847 // The slow branch is never taken. No need to build this guard.
848 return nullptr;
849 }
850
851 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
852
853 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
854 if (if_slow == top()) {
855 // The slow branch is never taken. No need to build this guard.
856 return nullptr;
857 }
858
859 if (region != nullptr)
860 region->add_req(if_slow);
861
862 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
863 set_control(if_fast);
864
865 return if_slow;
866 }
867
868 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
869 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
870 }
871 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
872 return generate_guard(test, region, PROB_FAIR);
873 }
874
875 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
876 Node* *pos_index) {
877 if (stopped())
878 return nullptr; // already stopped
879 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
880 return nullptr; // index is already adequately typed
881 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
882 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
883 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
884 if (is_neg != nullptr && pos_index != nullptr) {
885 // Emulate effect of Parse::adjust_map_after_if.
886 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
887 (*pos_index) = _gvn.transform(ccast);
888 }
889 return is_neg;
890 }
891
892 // Make sure that 'position' is a valid limit index, in [0..length].
893 // There are two equivalent plans for checking this:
894 // A. (offset + copyLength) unsigned<= arrayLength
895 // B. offset <= (arrayLength - copyLength)
896 // We require that all of the values above, except for the sum and
897 // difference, are already known to be non-negative.
898 // Plan A is robust in the face of overflow, if offset and copyLength
899 // are both hugely positive.
900 //
901 // Plan B is less direct and intuitive, but it does not overflow at
902 // all, since the difference of two non-negatives is always
903 // representable. Whenever Java methods must perform the equivalent
904 // check they generally use Plan B instead of Plan A.
905 // For the moment we use Plan A.
906 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
907 Node* subseq_length,
908 Node* array_length,
909 RegionNode* region) {
910 if (stopped())
911 return nullptr; // already stopped
912 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
913 if (zero_offset && subseq_length->eqv_uncast(array_length))
914 return nullptr; // common case of whole-array copy
915 Node* last = subseq_length;
916 if (!zero_offset) // last += offset
917 last = _gvn.transform(new AddINode(last, offset));
918 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
919 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
920 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
921 return is_over;
922 }
923
924 // Emit range checks for the given String.value byte array
925 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
926 if (stopped()) {
927 return; // already stopped
928 }
929 RegionNode* bailout = new RegionNode(1);
930 record_for_igvn(bailout);
931 if (char_count) {
932 // Convert char count to byte count
933 count = _gvn.transform(new LShiftINode(count, intcon(1)));
934 }
935
936 // Offset and count must not be negative
937 generate_negative_guard(offset, bailout);
938 generate_negative_guard(count, bailout);
939 // Offset + count must not exceed length of array
940 generate_limit_guard(offset, count, load_array_length(array), bailout);
941
942 if (bailout->req() > 1) {
943 PreserveJVMState pjvms(this);
944 set_control(_gvn.transform(bailout));
945 uncommon_trap(Deoptimization::Reason_intrinsic,
946 Deoptimization::Action_maybe_recompile);
947 }
948 }
949
950 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
951 bool is_immutable) {
952 ciKlass* thread_klass = env()->Thread_klass();
953 const Type* thread_type
954 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
955
956 Node* thread = _gvn.transform(new ThreadLocalNode());
957 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
958 tls_output = thread;
959
960 Node* thread_obj_handle
961 = (is_immutable
962 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
963 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
964 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
965 thread_obj_handle = _gvn.transform(thread_obj_handle);
966
967 DecoratorSet decorators = IN_NATIVE;
968 if (is_immutable) {
969 decorators |= C2_IMMUTABLE_MEMORY;
970 }
971 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
972 }
973
974 //--------------------------generate_current_thread--------------------
975 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
976 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
977 /*is_immutable*/false);
978 }
979
980 //--------------------------generate_virtual_thread--------------------
981 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
982 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
983 !C->method()->changes_current_thread());
984 }
985
986 //------------------------------make_string_method_node------------------------
987 // Helper method for String intrinsic functions. This version is called with
988 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
989 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
990 // containing the lengths of str1 and str2.
991 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
992 Node* result = nullptr;
993 switch (opcode) {
994 case Op_StrIndexOf:
995 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
996 str1_start, cnt1, str2_start, cnt2, ae);
997 break;
998 case Op_StrComp:
999 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1000 str1_start, cnt1, str2_start, cnt2, ae);
1001 break;
1002 case Op_StrEquals:
1003 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1004 // Use the constant length if there is one because optimized match rule may exist.
1005 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1006 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1007 break;
1008 default:
1009 ShouldNotReachHere();
1010 return nullptr;
1011 }
1012
1013 // All these intrinsics have checks.
1014 C->set_has_split_ifs(true); // Has chance for split-if optimization
1015 clear_upper_avx();
1016
1017 return _gvn.transform(result);
1018 }
1019
1020 //------------------------------inline_string_compareTo------------------------
1021 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1022 Node* arg1 = argument(0);
1023 Node* arg2 = argument(1);
1024
1025 arg1 = must_be_not_null(arg1, true);
1026 arg2 = must_be_not_null(arg2, true);
1027
1028 // Get start addr and length of first argument
1029 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1030 Node* arg1_cnt = load_array_length(arg1);
1031
1032 // Get start addr and length of second argument
1033 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1034 Node* arg2_cnt = load_array_length(arg2);
1035
1036 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1037 set_result(result);
1038 return true;
1039 }
1040
1041 //------------------------------inline_string_equals------------------------
1042 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1043 Node* arg1 = argument(0);
1044 Node* arg2 = argument(1);
1045
1046 // paths (plus control) merge
1047 RegionNode* region = new RegionNode(3);
1048 Node* phi = new PhiNode(region, TypeInt::BOOL);
1049
1050 if (!stopped()) {
1051
1052 arg1 = must_be_not_null(arg1, true);
1053 arg2 = must_be_not_null(arg2, true);
1054
1055 // Get start addr and length of first argument
1056 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1057 Node* arg1_cnt = load_array_length(arg1);
1058
1059 // Get start addr and length of second argument
1060 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1061 Node* arg2_cnt = load_array_length(arg2);
1062
1063 // Check for arg1_cnt != arg2_cnt
1064 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1065 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1066 Node* if_ne = generate_slow_guard(bol, nullptr);
1067 if (if_ne != nullptr) {
1068 phi->init_req(2, intcon(0));
1069 region->init_req(2, if_ne);
1070 }
1071
1072 // Check for count == 0 is done by assembler code for StrEquals.
1073
1074 if (!stopped()) {
1075 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1076 phi->init_req(1, equals);
1077 region->init_req(1, control());
1078 }
1079 }
1080
1081 // post merge
1082 set_control(_gvn.transform(region));
1083 record_for_igvn(region);
1084
1085 set_result(_gvn.transform(phi));
1086 return true;
1087 }
1088
1089 //------------------------------inline_array_equals----------------------------
1090 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1091 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1092 Node* arg1 = argument(0);
1093 Node* arg2 = argument(1);
1094
1095 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1096 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1097 clear_upper_avx();
1098
1099 return true;
1100 }
1101
1102
1103 //------------------------------inline_countPositives------------------------------
1104 bool LibraryCallKit::inline_countPositives() {
1105 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1106 return false;
1107 }
1108
1109 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1110 // no receiver since it is static method
1111 Node* ba = argument(0);
1112 Node* offset = argument(1);
1113 Node* len = argument(2);
1114
1115 ba = must_be_not_null(ba, true);
1116
1117 // Range checks
1118 generate_string_range_check(ba, offset, len, false);
1119 if (stopped()) {
1120 return true;
1121 }
1122 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1123 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1124 set_result(_gvn.transform(result));
1125 clear_upper_avx();
1126 return true;
1127 }
1128
1129 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1130 Node* index = argument(0);
1131 Node* length = bt == T_INT ? argument(1) : argument(2);
1132 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1133 return false;
1134 }
1135
1136 // check that length is positive
1137 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1138 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1139
1140 {
1141 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1142 uncommon_trap(Deoptimization::Reason_intrinsic,
1143 Deoptimization::Action_make_not_entrant);
1144 }
1145
1146 if (stopped()) {
1147 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1148 return true;
1149 }
1150
1151 // length is now known positive, add a cast node to make this explicit
1152 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1153 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1154 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1155 ConstraintCastNode::RegularDependency, bt);
1156 casted_length = _gvn.transform(casted_length);
1157 replace_in_map(length, casted_length);
1158 length = casted_length;
1159
1160 // Use an unsigned comparison for the range check itself
1161 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1162 BoolTest::mask btest = BoolTest::lt;
1163 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1164 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1165 _gvn.set_type(rc, rc->Value(&_gvn));
1166 if (!rc_bool->is_Con()) {
1167 record_for_igvn(rc);
1168 }
1169 set_control(_gvn.transform(new IfTrueNode(rc)));
1170 {
1171 PreserveJVMState pjvms(this);
1172 set_control(_gvn.transform(new IfFalseNode(rc)));
1173 uncommon_trap(Deoptimization::Reason_range_check,
1174 Deoptimization::Action_make_not_entrant);
1175 }
1176
1177 if (stopped()) {
1178 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1179 return true;
1180 }
1181
1182 // index is now known to be >= 0 and < length, cast it
1183 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1184 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1185 ConstraintCastNode::RegularDependency, bt);
1186 result = _gvn.transform(result);
1187 set_result(result);
1188 replace_in_map(index, result);
1189 return true;
1190 }
1191
1192 //------------------------------inline_string_indexOf------------------------
1193 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1194 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1195 return false;
1196 }
1197 Node* src = argument(0);
1198 Node* tgt = argument(1);
1199
1200 // Make the merge point
1201 RegionNode* result_rgn = new RegionNode(4);
1202 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1203
1204 src = must_be_not_null(src, true);
1205 tgt = must_be_not_null(tgt, true);
1206
1207 // Get start addr and length of source string
1208 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1209 Node* src_count = load_array_length(src);
1210
1211 // Get start addr and length of substring
1212 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1213 Node* tgt_count = load_array_length(tgt);
1214
1215 Node* result = nullptr;
1216 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1217
1218 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1219 // Divide src size by 2 if String is UTF16 encoded
1220 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1221 }
1222 if (ae == StrIntrinsicNode::UU) {
1223 // Divide substring size by 2 if String is UTF16 encoded
1224 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1225 }
1226
1227 if (call_opt_stub) {
1228 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1229 StubRoutines::_string_indexof_array[ae],
1230 "stringIndexOf", TypePtr::BOTTOM, src_start,
1231 src_count, tgt_start, tgt_count);
1232 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1233 } else {
1234 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1235 result_rgn, result_phi, ae);
1236 }
1237 if (result != nullptr) {
1238 result_phi->init_req(3, result);
1239 result_rgn->init_req(3, control());
1240 }
1241 set_control(_gvn.transform(result_rgn));
1242 record_for_igvn(result_rgn);
1243 set_result(_gvn.transform(result_phi));
1244
1245 return true;
1246 }
1247
1248 //-----------------------------inline_string_indexOfI-----------------------
1249 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1250 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1251 return false;
1252 }
1253 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1254 return false;
1255 }
1256
1257 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1258 Node* src = argument(0); // byte[]
1259 Node* src_count = argument(1); // char count
1260 Node* tgt = argument(2); // byte[]
1261 Node* tgt_count = argument(3); // char count
1262 Node* from_index = argument(4); // char index
1263
1264 src = must_be_not_null(src, true);
1265 tgt = must_be_not_null(tgt, true);
1266
1267 // Multiply byte array index by 2 if String is UTF16 encoded
1268 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1269 src_count = _gvn.transform(new SubINode(src_count, from_index));
1270 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1271 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1272
1273 // Range checks
1274 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1275 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1276 if (stopped()) {
1277 return true;
1278 }
1279
1280 RegionNode* region = new RegionNode(5);
1281 Node* phi = new PhiNode(region, TypeInt::INT);
1282 Node* result = nullptr;
1283
1284 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1285
1286 if (call_opt_stub) {
1287 assert(arrayOopDesc::base_offset_in_bytes(T_BYTE) >= 16, "Needed for indexOf");
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 region, phi, ae);
1296 }
1297 if (result != nullptr) {
1298 // The result is index relative to from_index if substring was found, -1 otherwise.
1299 // Generate code which will fold into cmove.
1300 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1301 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1302
1303 Node* if_lt = generate_slow_guard(bol, nullptr);
1304 if (if_lt != nullptr) {
1305 // result == -1
1306 phi->init_req(3, result);
1307 region->init_req(3, if_lt);
1308 }
1309 if (!stopped()) {
1310 result = _gvn.transform(new AddINode(result, from_index));
1311 phi->init_req(4, result);
1312 region->init_req(4, control());
1313 }
1314 }
1315
1316 set_control(_gvn.transform(region));
1317 record_for_igvn(region);
1318 set_result(_gvn.transform(phi));
1319 clear_upper_avx();
1320
1321 return true;
1322 }
1323
1324 // Create StrIndexOfNode with fast path checks
1325 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1326 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1327 // Check for substr count > string count
1328 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1329 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1330 Node* if_gt = generate_slow_guard(bol, nullptr);
1331 if (if_gt != nullptr) {
1332 phi->init_req(1, intcon(-1));
1333 region->init_req(1, if_gt);
1334 }
1335 if (!stopped()) {
1336 // Check for substr count == 0
1337 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1338 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1339 Node* if_zero = generate_slow_guard(bol, nullptr);
1340 if (if_zero != nullptr) {
1341 phi->init_req(2, intcon(0));
1342 region->init_req(2, if_zero);
1343 }
1344 }
1345 if (!stopped()) {
1346 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1347 }
1348 return nullptr;
1349 }
1350
1351 //-----------------------------inline_string_indexOfChar-----------------------
1352 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1353 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1354 return false;
1355 }
1356 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1357 return false;
1358 }
1359 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1360 Node* src = argument(0); // byte[]
1361 Node* int_ch = argument(1);
1362 Node* from_index = argument(2);
1363 Node* max = argument(3);
1364
1365 src = must_be_not_null(src, true);
1366
1367 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1368 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1369 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1370
1371 // Range checks
1372 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1373
1374 // Check for int_ch >= 0
1375 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1376 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1377 {
1378 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1379 uncommon_trap(Deoptimization::Reason_intrinsic,
1380 Deoptimization::Action_maybe_recompile);
1381 }
1382 if (stopped()) {
1383 return true;
1384 }
1385
1386 RegionNode* region = new RegionNode(3);
1387 Node* phi = new PhiNode(region, TypeInt::INT);
1388
1389 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1390 C->set_has_split_ifs(true); // Has chance for split-if optimization
1391 _gvn.transform(result);
1392
1393 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1394 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1395
1396 Node* if_lt = generate_slow_guard(bol, nullptr);
1397 if (if_lt != nullptr) {
1398 // result == -1
1399 phi->init_req(2, result);
1400 region->init_req(2, if_lt);
1401 }
1402 if (!stopped()) {
1403 result = _gvn.transform(new AddINode(result, from_index));
1404 phi->init_req(1, result);
1405 region->init_req(1, control());
1406 }
1407 set_control(_gvn.transform(region));
1408 record_for_igvn(region);
1409 set_result(_gvn.transform(phi));
1410 clear_upper_avx();
1411
1412 return true;
1413 }
1414 //---------------------------inline_string_copy---------------------
1415 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1416 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1417 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1418 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1419 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1420 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1421 bool LibraryCallKit::inline_string_copy(bool compress) {
1422 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1423 return false;
1424 }
1425 int nargs = 5; // 2 oops, 3 ints
1426 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1427
1428 Node* src = argument(0);
1429 Node* src_offset = argument(1);
1430 Node* dst = argument(2);
1431 Node* dst_offset = argument(3);
1432 Node* length = argument(4);
1433
1434 // Check for allocation before we add nodes that would confuse
1435 // tightly_coupled_allocation()
1436 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1437
1438 // Figure out the size and type of the elements we will be copying.
1439 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1440 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1441 if (src_type == nullptr || dst_type == nullptr) {
1442 return false;
1443 }
1444 BasicType src_elem = src_type->elem()->array_element_basic_type();
1445 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1446 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1447 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1448 "Unsupported array types for inline_string_copy");
1449
1450 src = must_be_not_null(src, true);
1451 dst = must_be_not_null(dst, true);
1452
1453 // Convert char[] offsets to byte[] offsets
1454 bool convert_src = (compress && src_elem == T_BYTE);
1455 bool convert_dst = (!compress && dst_elem == T_BYTE);
1456 if (convert_src) {
1457 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1458 } else if (convert_dst) {
1459 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1460 }
1461
1462 // Range checks
1463 generate_string_range_check(src, src_offset, length, convert_src);
1464 generate_string_range_check(dst, dst_offset, length, convert_dst);
1465 if (stopped()) {
1466 return true;
1467 }
1468
1469 Node* src_start = array_element_address(src, src_offset, src_elem);
1470 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1471 // 'src_start' points to src array + scaled offset
1472 // 'dst_start' points to dst array + scaled offset
1473 Node* count = nullptr;
1474 if (compress) {
1475 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1476 } else {
1477 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1478 }
1479
1480 if (alloc != nullptr) {
1481 if (alloc->maybe_set_complete(&_gvn)) {
1482 // "You break it, you buy it."
1483 InitializeNode* init = alloc->initialization();
1484 assert(init->is_complete(), "we just did this");
1485 init->set_complete_with_arraycopy();
1486 assert(dst->is_CheckCastPP(), "sanity");
1487 assert(dst->in(0)->in(0) == init, "dest pinned");
1488 }
1489 // Do not let stores that initialize this object be reordered with
1490 // a subsequent store that would make this object accessible by
1491 // other threads.
1492 // Record what AllocateNode this StoreStore protects so that
1493 // escape analysis can go from the MemBarStoreStoreNode to the
1494 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1495 // based on the escape status of the AllocateNode.
1496 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1497 }
1498 if (compress) {
1499 set_result(_gvn.transform(count));
1500 }
1501 clear_upper_avx();
1502
1503 return true;
1504 }
1505
1506 #ifdef _LP64
1507 #define XTOP ,top() /*additional argument*/
1508 #else //_LP64
1509 #define XTOP /*no additional argument*/
1510 #endif //_LP64
1511
1512 //------------------------inline_string_toBytesU--------------------------
1513 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1514 bool LibraryCallKit::inline_string_toBytesU() {
1515 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1516 return false;
1517 }
1518 // Get the arguments.
1519 Node* value = argument(0);
1520 Node* offset = argument(1);
1521 Node* length = argument(2);
1522
1523 Node* newcopy = nullptr;
1524
1525 // Set the original stack and the reexecute bit for the interpreter to reexecute
1526 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1527 { PreserveReexecuteState preexecs(this);
1528 jvms()->set_should_reexecute(true);
1529
1530 // Check if a null path was taken unconditionally.
1531 value = null_check(value);
1532
1533 RegionNode* bailout = new RegionNode(1);
1534 record_for_igvn(bailout);
1535
1536 // Range checks
1537 generate_negative_guard(offset, bailout);
1538 generate_negative_guard(length, bailout);
1539 generate_limit_guard(offset, length, load_array_length(value), bailout);
1540 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1541 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1542
1543 if (bailout->req() > 1) {
1544 PreserveJVMState pjvms(this);
1545 set_control(_gvn.transform(bailout));
1546 uncommon_trap(Deoptimization::Reason_intrinsic,
1547 Deoptimization::Action_maybe_recompile);
1548 }
1549 if (stopped()) {
1550 return true;
1551 }
1552
1553 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1554 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1555 newcopy = new_array(klass_node, size, 0); // no arguments to push
1556 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1557 guarantee(alloc != nullptr, "created above");
1558
1559 // Calculate starting addresses.
1560 Node* src_start = array_element_address(value, offset, T_CHAR);
1561 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1562
1563 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1564 const TypeInt* toffset = gvn().type(offset)->is_int();
1565 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1566
1567 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1568 const char* copyfunc_name = "arraycopy";
1569 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1570 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1571 OptoRuntime::fast_arraycopy_Type(),
1572 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1573 src_start, dst_start, ConvI2X(length) XTOP);
1574 // Do not let reads from the cloned object float above the arraycopy.
1575 if (alloc->maybe_set_complete(&_gvn)) {
1576 // "You break it, you buy it."
1577 InitializeNode* init = alloc->initialization();
1578 assert(init->is_complete(), "we just did this");
1579 init->set_complete_with_arraycopy();
1580 assert(newcopy->is_CheckCastPP(), "sanity");
1581 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1582 }
1583 // Do not let stores that initialize this object be reordered with
1584 // a subsequent store that would make this object accessible by
1585 // other threads.
1586 // Record what AllocateNode this StoreStore protects so that
1587 // escape analysis can go from the MemBarStoreStoreNode to the
1588 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1589 // based on the escape status of the AllocateNode.
1590 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1591 } // original reexecute is set back here
1592
1593 C->set_has_split_ifs(true); // Has chance for split-if optimization
1594 if (!stopped()) {
1595 set_result(newcopy);
1596 }
1597 clear_upper_avx();
1598
1599 return true;
1600 }
1601
1602 //------------------------inline_string_getCharsU--------------------------
1603 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1604 bool LibraryCallKit::inline_string_getCharsU() {
1605 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1606 return false;
1607 }
1608
1609 // Get the arguments.
1610 Node* src = argument(0);
1611 Node* src_begin = argument(1);
1612 Node* src_end = argument(2); // exclusive offset (i < src_end)
1613 Node* dst = argument(3);
1614 Node* dst_begin = argument(4);
1615
1616 // Check for allocation before we add nodes that would confuse
1617 // tightly_coupled_allocation()
1618 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1619
1620 // Check if a null path was taken unconditionally.
1621 src = null_check(src);
1622 dst = null_check(dst);
1623 if (stopped()) {
1624 return true;
1625 }
1626
1627 // Get length and convert char[] offset to byte[] offset
1628 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1629 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1630
1631 // Range checks
1632 generate_string_range_check(src, src_begin, length, true);
1633 generate_string_range_check(dst, dst_begin, length, false);
1634 if (stopped()) {
1635 return true;
1636 }
1637
1638 if (!stopped()) {
1639 // Calculate starting addresses.
1640 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1641 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1642
1643 // Check if array addresses are aligned to HeapWordSize
1644 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1645 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1646 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1647 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1648
1649 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1650 const char* copyfunc_name = "arraycopy";
1651 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1652 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1653 OptoRuntime::fast_arraycopy_Type(),
1654 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1655 src_start, dst_start, ConvI2X(length) XTOP);
1656 // Do not let reads from the cloned object float above the arraycopy.
1657 if (alloc != nullptr) {
1658 if (alloc->maybe_set_complete(&_gvn)) {
1659 // "You break it, you buy it."
1660 InitializeNode* init = alloc->initialization();
1661 assert(init->is_complete(), "we just did this");
1662 init->set_complete_with_arraycopy();
1663 assert(dst->is_CheckCastPP(), "sanity");
1664 assert(dst->in(0)->in(0) == init, "dest pinned");
1665 }
1666 // Do not let stores that initialize this object be reordered with
1667 // a subsequent store that would make this object accessible by
1668 // other threads.
1669 // Record what AllocateNode this StoreStore protects so that
1670 // escape analysis can go from the MemBarStoreStoreNode to the
1671 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1672 // based on the escape status of the AllocateNode.
1673 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1674 } else {
1675 insert_mem_bar(Op_MemBarCPUOrder);
1676 }
1677 }
1678
1679 C->set_has_split_ifs(true); // Has chance for split-if optimization
1680 return true;
1681 }
1682
1683 //----------------------inline_string_char_access----------------------------
1684 // Store/Load char to/from byte[] array.
1685 // static void StringUTF16.putChar(byte[] val, int index, int c)
1686 // static char StringUTF16.getChar(byte[] val, int index)
1687 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1688 Node* value = argument(0);
1689 Node* index = argument(1);
1690 Node* ch = is_store ? argument(2) : nullptr;
1691
1692 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1693 // correctly requires matched array shapes.
1694 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1695 "sanity: byte[] and char[] bases agree");
1696 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1697 "sanity: byte[] and char[] scales agree");
1698
1699 // Bail when getChar over constants is requested: constant folding would
1700 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1701 // Java method would constant fold nicely instead.
1702 if (!is_store && value->is_Con() && index->is_Con()) {
1703 return false;
1704 }
1705
1706 // Save state and restore on bailout
1707 uint old_sp = sp();
1708 SafePointNode* old_map = clone_map();
1709
1710 value = must_be_not_null(value, true);
1711
1712 Node* adr = array_element_address(value, index, T_CHAR);
1713 if (adr->is_top()) {
1714 set_map(old_map);
1715 set_sp(old_sp);
1716 return false;
1717 }
1718 destruct_map_clone(old_map);
1719 if (is_store) {
1720 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1721 } else {
1722 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);
1723 set_result(ch);
1724 }
1725 return true;
1726 }
1727
1728 //--------------------------round_double_node--------------------------------
1729 // Round a double node if necessary.
1730 Node* LibraryCallKit::round_double_node(Node* n) {
1731 if (Matcher::strict_fp_requires_explicit_rounding) {
1732 #ifdef IA32
1733 if (UseSSE < 2) {
1734 n = _gvn.transform(new RoundDoubleNode(nullptr, n));
1735 }
1736 #else
1737 Unimplemented();
1738 #endif // IA32
1739 }
1740 return n;
1741 }
1742
1743 //------------------------------inline_math-----------------------------------
1744 // public static double Math.abs(double)
1745 // public static double Math.sqrt(double)
1746 // public static double Math.log(double)
1747 // public static double Math.log10(double)
1748 // public static double Math.round(double)
1749 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1750 Node* arg = round_double_node(argument(0));
1751 Node* n = nullptr;
1752 switch (id) {
1753 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1754 case vmIntrinsics::_dsqrt:
1755 case vmIntrinsics::_dsqrt_strict:
1756 n = new SqrtDNode(C, control(), arg); break;
1757 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1758 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1759 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1760 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1761 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1762 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1763 default: fatal_unexpected_iid(id); break;
1764 }
1765 set_result(_gvn.transform(n));
1766 return true;
1767 }
1768
1769 //------------------------------inline_math-----------------------------------
1770 // public static float Math.abs(float)
1771 // public static int Math.abs(int)
1772 // public static long Math.abs(long)
1773 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1774 Node* arg = argument(0);
1775 Node* n = nullptr;
1776 switch (id) {
1777 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1778 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1779 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1780 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1781 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1782 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1783 default: fatal_unexpected_iid(id); break;
1784 }
1785 set_result(_gvn.transform(n));
1786 return true;
1787 }
1788
1789 //------------------------------runtime_math-----------------------------
1790 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1791 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1792 "must be (DD)D or (D)D type");
1793
1794 // Inputs
1795 Node* a = round_double_node(argument(0));
1796 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : nullptr;
1797
1798 const TypePtr* no_memory_effects = nullptr;
1799 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1800 no_memory_effects,
1801 a, top(), b, b ? top() : nullptr);
1802 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1803 #ifdef ASSERT
1804 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1805 assert(value_top == top(), "second value must be top");
1806 #endif
1807
1808 set_result(value);
1809 return true;
1810 }
1811
1812 //------------------------------inline_math_pow-----------------------------
1813 bool LibraryCallKit::inline_math_pow() {
1814 Node* exp = round_double_node(argument(2));
1815 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1816 if (d != nullptr) {
1817 if (d->getd() == 2.0) {
1818 // Special case: pow(x, 2.0) => x * x
1819 Node* base = round_double_node(argument(0));
1820 set_result(_gvn.transform(new MulDNode(base, base)));
1821 return true;
1822 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1823 // Special case: pow(x, 0.5) => sqrt(x)
1824 Node* base = round_double_node(argument(0));
1825 Node* zero = _gvn.zerocon(T_DOUBLE);
1826
1827 RegionNode* region = new RegionNode(3);
1828 Node* phi = new PhiNode(region, Type::DOUBLE);
1829
1830 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1831 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1832 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1833 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1834 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1835
1836 Node* if_pow = generate_slow_guard(test, nullptr);
1837 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1838 phi->init_req(1, value_sqrt);
1839 region->init_req(1, control());
1840
1841 if (if_pow != nullptr) {
1842 set_control(if_pow);
1843 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1844 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1845 const TypePtr* no_memory_effects = nullptr;
1846 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1847 no_memory_effects, base, top(), exp, top());
1848 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1849 #ifdef ASSERT
1850 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1851 assert(value_top == top(), "second value must be top");
1852 #endif
1853 phi->init_req(2, value_pow);
1854 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1855 }
1856
1857 C->set_has_split_ifs(true); // Has chance for split-if optimization
1858 set_control(_gvn.transform(region));
1859 record_for_igvn(region);
1860 set_result(_gvn.transform(phi));
1861
1862 return true;
1863 }
1864 }
1865
1866 return StubRoutines::dpow() != nullptr ?
1867 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1868 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1869 }
1870
1871 //------------------------------inline_math_native-----------------------------
1872 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1873 switch (id) {
1874 case vmIntrinsics::_dsin:
1875 return StubRoutines::dsin() != nullptr ?
1876 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1877 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1878 case vmIntrinsics::_dcos:
1879 return StubRoutines::dcos() != nullptr ?
1880 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1881 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1882 case vmIntrinsics::_dtan:
1883 return StubRoutines::dtan() != nullptr ?
1884 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1885 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1886 case vmIntrinsics::_dexp:
1887 return StubRoutines::dexp() != nullptr ?
1888 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1889 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1890 case vmIntrinsics::_dlog:
1891 return StubRoutines::dlog() != nullptr ?
1892 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1893 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1894 case vmIntrinsics::_dlog10:
1895 return StubRoutines::dlog10() != nullptr ?
1896 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1897 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1898
1899 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1900 case vmIntrinsics::_ceil:
1901 case vmIntrinsics::_floor:
1902 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1903
1904 case vmIntrinsics::_dsqrt:
1905 case vmIntrinsics::_dsqrt_strict:
1906 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1907 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1908 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1909 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1910 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1911
1912 case vmIntrinsics::_dpow: return inline_math_pow();
1913 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1914 case vmIntrinsics::_fcopySign: return inline_math(id);
1915 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1916 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1917 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1918
1919 // These intrinsics are not yet correctly implemented
1920 case vmIntrinsics::_datan2:
1921 return false;
1922
1923 default:
1924 fatal_unexpected_iid(id);
1925 return false;
1926 }
1927 }
1928
1929 //----------------------------inline_notify-----------------------------------*
1930 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1931 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1932 address func;
1933 if (id == vmIntrinsics::_notify) {
1934 func = OptoRuntime::monitor_notify_Java();
1935 } else {
1936 func = OptoRuntime::monitor_notifyAll_Java();
1937 }
1938 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1939 make_slow_call_ex(call, env()->Throwable_klass(), false);
1940 return true;
1941 }
1942
1943
1944 //----------------------------inline_min_max-----------------------------------
1945 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1946 set_result(generate_min_max(id, argument(0), argument(1)));
1947 return true;
1948 }
1949
1950 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1951 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1952 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1953 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1954 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1955
1956 {
1957 PreserveJVMState pjvms(this);
1958 PreserveReexecuteState preexecs(this);
1959 jvms()->set_should_reexecute(true);
1960
1961 set_control(slow_path);
1962 set_i_o(i_o());
1963
1964 uncommon_trap(Deoptimization::Reason_intrinsic,
1965 Deoptimization::Action_none);
1966 }
1967
1968 set_control(fast_path);
1969 set_result(math);
1970 }
1971
1972 template <typename OverflowOp>
1973 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1974 typedef typename OverflowOp::MathOp MathOp;
1975
1976 MathOp* mathOp = new MathOp(arg1, arg2);
1977 Node* operation = _gvn.transform( mathOp );
1978 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1979 inline_math_mathExact(operation, ofcheck);
1980 return true;
1981 }
1982
1983 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1984 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1985 }
1986
1987 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1988 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1989 }
1990
1991 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1992 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1993 }
1994
1995 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1996 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1997 }
1998
1999 bool LibraryCallKit::inline_math_negateExactI() {
2000 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2001 }
2002
2003 bool LibraryCallKit::inline_math_negateExactL() {
2004 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2005 }
2006
2007 bool LibraryCallKit::inline_math_multiplyExactI() {
2008 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2009 }
2010
2011 bool LibraryCallKit::inline_math_multiplyExactL() {
2012 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2013 }
2014
2015 bool LibraryCallKit::inline_math_multiplyHigh() {
2016 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2017 return true;
2018 }
2019
2020 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2021 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2022 return true;
2023 }
2024
2025 Node*
2026 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
2027 Node* result_val = nullptr;
2028 switch (id) {
2029 case vmIntrinsics::_min:
2030 case vmIntrinsics::_min_strict:
2031 result_val = _gvn.transform(new MinINode(x0, y0));
2032 break;
2033 case vmIntrinsics::_max:
2034 case vmIntrinsics::_max_strict:
2035 result_val = _gvn.transform(new MaxINode(x0, y0));
2036 break;
2037 default:
2038 fatal_unexpected_iid(id);
2039 break;
2040 }
2041 return result_val;
2042 }
2043
2044 inline int
2045 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2046 const TypePtr* base_type = TypePtr::NULL_PTR;
2047 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2048 if (base_type == nullptr) {
2049 // Unknown type.
2050 return Type::AnyPtr;
2051 } else if (base_type == TypePtr::NULL_PTR) {
2052 // Since this is a null+long form, we have to switch to a rawptr.
2053 base = _gvn.transform(new CastX2PNode(offset));
2054 offset = MakeConX(0);
2055 return Type::RawPtr;
2056 } else if (base_type->base() == Type::RawPtr) {
2057 return Type::RawPtr;
2058 } else if (base_type->isa_oopptr()) {
2059 // Base is never null => always a heap address.
2060 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2061 return Type::OopPtr;
2062 }
2063 // Offset is small => always a heap address.
2064 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2065 if (offset_type != nullptr &&
2066 base_type->offset() == 0 && // (should always be?)
2067 offset_type->_lo >= 0 &&
2068 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2069 return Type::OopPtr;
2070 } else if (type == T_OBJECT) {
2071 // off heap access to an oop doesn't make any sense. Has to be on
2072 // heap.
2073 return Type::OopPtr;
2074 }
2075 // Otherwise, it might either be oop+off or null+addr.
2076 return Type::AnyPtr;
2077 } else {
2078 // No information:
2079 return Type::AnyPtr;
2080 }
2081 }
2082
2083 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2084 Node* uncasted_base = base;
2085 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2086 if (kind == Type::RawPtr) {
2087 return basic_plus_adr(top(), uncasted_base, offset);
2088 } else if (kind == Type::AnyPtr) {
2089 assert(base == uncasted_base, "unexpected base change");
2090 if (can_cast) {
2091 if (!_gvn.type(base)->speculative_maybe_null() &&
2092 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2093 // According to profiling, this access is always on
2094 // heap. Casting the base to not null and thus avoiding membars
2095 // around the access should allow better optimizations
2096 Node* null_ctl = top();
2097 base = null_check_oop(base, &null_ctl, true, true, true);
2098 assert(null_ctl->is_top(), "no null control here");
2099 return basic_plus_adr(base, offset);
2100 } else if (_gvn.type(base)->speculative_always_null() &&
2101 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2102 // According to profiling, this access is always off
2103 // heap.
2104 base = null_assert(base);
2105 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2106 offset = MakeConX(0);
2107 return basic_plus_adr(top(), raw_base, offset);
2108 }
2109 }
2110 // We don't know if it's an on heap or off heap access. Fall back
2111 // to raw memory access.
2112 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2113 return basic_plus_adr(top(), raw, offset);
2114 } else {
2115 assert(base == uncasted_base, "unexpected base change");
2116 // We know it's an on heap access so base can't be null
2117 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2118 base = must_be_not_null(base, true);
2119 }
2120 return basic_plus_adr(base, offset);
2121 }
2122 }
2123
2124 //--------------------------inline_number_methods-----------------------------
2125 // inline int Integer.numberOfLeadingZeros(int)
2126 // inline int Long.numberOfLeadingZeros(long)
2127 //
2128 // inline int Integer.numberOfTrailingZeros(int)
2129 // inline int Long.numberOfTrailingZeros(long)
2130 //
2131 // inline int Integer.bitCount(int)
2132 // inline int Long.bitCount(long)
2133 //
2134 // inline char Character.reverseBytes(char)
2135 // inline short Short.reverseBytes(short)
2136 // inline int Integer.reverseBytes(int)
2137 // inline long Long.reverseBytes(long)
2138 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2139 Node* arg = argument(0);
2140 Node* n = nullptr;
2141 switch (id) {
2142 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2143 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2144 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2145 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2146 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2147 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2148 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2149 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2150 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2151 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2152 case vmIntrinsics::_reverse_i: n = new ReverseINode(0, arg); break;
2153 case vmIntrinsics::_reverse_l: n = new ReverseLNode(0, arg); break;
2154 default: fatal_unexpected_iid(id); break;
2155 }
2156 set_result(_gvn.transform(n));
2157 return true;
2158 }
2159
2160 //--------------------------inline_bitshuffle_methods-----------------------------
2161 // inline int Integer.compress(int, int)
2162 // inline int Integer.expand(int, int)
2163 // inline long Long.compress(long, long)
2164 // inline long Long.expand(long, long)
2165 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2166 Node* n = nullptr;
2167 switch (id) {
2168 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2169 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2170 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2171 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2172 default: fatal_unexpected_iid(id); break;
2173 }
2174 set_result(_gvn.transform(n));
2175 return true;
2176 }
2177
2178 //--------------------------inline_number_methods-----------------------------
2179 // inline int Integer.compareUnsigned(int, int)
2180 // inline int Long.compareUnsigned(long, long)
2181 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2182 Node* arg1 = argument(0);
2183 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2184 Node* n = nullptr;
2185 switch (id) {
2186 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2187 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2188 default: fatal_unexpected_iid(id); break;
2189 }
2190 set_result(_gvn.transform(n));
2191 return true;
2192 }
2193
2194 //--------------------------inline_unsigned_divmod_methods-----------------------------
2195 // inline int Integer.divideUnsigned(int, int)
2196 // inline int Integer.remainderUnsigned(int, int)
2197 // inline long Long.divideUnsigned(long, long)
2198 // inline long Long.remainderUnsigned(long, long)
2199 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2200 Node* n = nullptr;
2201 switch (id) {
2202 case vmIntrinsics::_divideUnsigned_i: {
2203 zero_check_int(argument(1));
2204 // Compile-time detect of null-exception
2205 if (stopped()) {
2206 return true; // keep the graph constructed so far
2207 }
2208 n = new UDivINode(control(), argument(0), argument(1));
2209 break;
2210 }
2211 case vmIntrinsics::_divideUnsigned_l: {
2212 zero_check_long(argument(2));
2213 // Compile-time detect of null-exception
2214 if (stopped()) {
2215 return true; // keep the graph constructed so far
2216 }
2217 n = new UDivLNode(control(), argument(0), argument(2));
2218 break;
2219 }
2220 case vmIntrinsics::_remainderUnsigned_i: {
2221 zero_check_int(argument(1));
2222 // Compile-time detect of null-exception
2223 if (stopped()) {
2224 return true; // keep the graph constructed so far
2225 }
2226 n = new UModINode(control(), argument(0), argument(1));
2227 break;
2228 }
2229 case vmIntrinsics::_remainderUnsigned_l: {
2230 zero_check_long(argument(2));
2231 // Compile-time detect of null-exception
2232 if (stopped()) {
2233 return true; // keep the graph constructed so far
2234 }
2235 n = new UModLNode(control(), argument(0), argument(2));
2236 break;
2237 }
2238 default: fatal_unexpected_iid(id); break;
2239 }
2240 set_result(_gvn.transform(n));
2241 return true;
2242 }
2243
2244 //----------------------------inline_unsafe_access----------------------------
2245
2246 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2247 // Attempt to infer a sharper value type from the offset and base type.
2248 ciKlass* sharpened_klass = nullptr;
2249 bool null_free = false;
2250
2251 // See if it is an instance field, with an object type.
2252 if (alias_type->field() != nullptr) {
2253 if (alias_type->field()->type()->is_klass()) {
2254 sharpened_klass = alias_type->field()->type()->as_klass();
2255 null_free = alias_type->field()->is_null_free();
2256 }
2257 }
2258
2259 const TypeOopPtr* result = nullptr;
2260 // See if it is a narrow oop array.
2261 if (adr_type->isa_aryptr()) {
2262 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2263 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2264 null_free = adr_type->is_aryptr()->is_null_free();
2265 if (elem_type != nullptr && elem_type->is_loaded()) {
2266 // Sharpen the value type.
2267 result = elem_type;
2268 }
2269 }
2270 }
2271
2272 // The sharpened class might be unloaded if there is no class loader
2273 // contraint in place.
2274 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2275 // Sharpen the value type.
2276 result = TypeOopPtr::make_from_klass(sharpened_klass);
2277 if (null_free) {
2278 result = result->join_speculative(TypePtr::NOTNULL)->is_oopptr();
2279 }
2280 }
2281 if (result != nullptr) {
2282 #ifndef PRODUCT
2283 if (C->print_intrinsics() || C->print_inlining()) {
2284 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2285 tty->print(" sharpened value: "); result->dump(); tty->cr();
2286 }
2287 #endif
2288 }
2289 return result;
2290 }
2291
2292 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2293 switch (kind) {
2294 case Relaxed:
2295 return MO_UNORDERED;
2296 case Opaque:
2297 return MO_RELAXED;
2298 case Acquire:
2299 return MO_ACQUIRE;
2300 case Release:
2301 return MO_RELEASE;
2302 case Volatile:
2303 return MO_SEQ_CST;
2304 default:
2305 ShouldNotReachHere();
2306 return 0;
2307 }
2308 }
2309
2310 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned, const bool is_flat) {
2311 if (callee()->is_static()) return false; // caller must have the capability!
2312 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2313 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2314 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2315 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2316
2317 if (is_reference_type(type)) {
2318 decorators |= ON_UNKNOWN_OOP_REF;
2319 }
2320
2321 if (unaligned) {
2322 decorators |= C2_UNALIGNED;
2323 }
2324
2325 #ifndef PRODUCT
2326 {
2327 ResourceMark rm;
2328 // Check the signatures.
2329 ciSignature* sig = callee()->signature();
2330 #ifdef ASSERT
2331 if (!is_store) {
2332 // Object getReference(Object base, int/long offset), etc.
2333 BasicType rtype = sig->return_type()->basic_type();
2334 assert(rtype == type, "getter must return the expected value");
2335 assert(sig->count() == 2 || (is_flat && sig->count() == 3), "oop getter has 2 or 3 arguments");
2336 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2337 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2338 } else {
2339 // void putReference(Object base, int/long offset, Object x), etc.
2340 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2341 assert(sig->count() == 3 || (is_flat && sig->count() == 4), "oop putter has 3 arguments");
2342 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2343 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2344 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2345 assert(vtype == type, "putter must accept the expected value");
2346 }
2347 #endif // ASSERT
2348 }
2349 #endif //PRODUCT
2350
2351 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2352
2353 Node* receiver = argument(0); // type: oop
2354
2355 // Build address expression.
2356 Node* heap_base_oop = top();
2357
2358 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2359 Node* base = argument(1); // type: oop
2360 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2361 Node* offset = argument(2); // type: long
2362 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2363 // to be plain byte offsets, which are also the same as those accepted
2364 // by oopDesc::field_addr.
2365 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2366 "fieldOffset must be byte-scaled");
2367
2368 ciInlineKlass* inline_klass = nullptr;
2369 if (is_flat) {
2370 const TypeInstPtr* cls = _gvn.type(argument(4))->isa_instptr();
2371 if (cls == nullptr || cls->const_oop() == nullptr) {
2372 return false;
2373 }
2374 ciType* mirror_type = cls->const_oop()->as_instance()->java_mirror_type();
2375 if (!mirror_type->is_inlinetype()) {
2376 return false;
2377 }
2378 inline_klass = mirror_type->as_inline_klass();
2379 }
2380
2381 if (base->is_InlineType()) {
2382 InlineTypeNode* vt = base->as_InlineType();
2383 if (is_store) {
2384 if (!vt->is_allocated(&_gvn)) {
2385 return false;
2386 }
2387 base = vt->get_oop();
2388 } else {
2389 if (offset->is_Con()) {
2390 long off = find_long_con(offset, 0);
2391 ciInlineKlass* vk = vt->type()->inline_klass();
2392 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2393 return false;
2394 }
2395
2396 ciField* field = vk->get_non_flat_field_by_offset(off);
2397 if (field != nullptr) {
2398 BasicType bt = type2field[field->type()->basic_type()];
2399 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2400 bt = T_OBJECT;
2401 }
2402 if (bt == type && (!field->is_flat() || field->type() == inline_klass)) {
2403 Node* value = vt->field_value_by_offset(off, false);
2404 if (value->is_InlineType()) {
2405 value = value->as_InlineType()->adjust_scalarization_depth(this);
2406 }
2407 set_result(value);
2408 return true;
2409 }
2410 }
2411 }
2412 {
2413 // Re-execute the unsafe access if allocation triggers deoptimization.
2414 PreserveReexecuteState preexecs(this);
2415 jvms()->set_should_reexecute(true);
2416 vt = vt->buffer(this);
2417 }
2418 base = vt->get_oop();
2419 }
2420 }
2421
2422 // 32-bit machines ignore the high half!
2423 offset = ConvL2X(offset);
2424
2425 // Save state and restore on bailout
2426 uint old_sp = sp();
2427 SafePointNode* old_map = clone_map();
2428
2429 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2430
2431 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) {
2432 if (type != T_OBJECT && (inline_klass == nullptr || !inline_klass->has_object_fields())) {
2433 decorators |= IN_NATIVE; // off-heap primitive access
2434 } else {
2435 set_map(old_map);
2436 set_sp(old_sp);
2437 return false; // off-heap oop accesses are not supported
2438 }
2439 } else {
2440 heap_base_oop = base; // on-heap or mixed access
2441 }
2442
2443 // Can base be null? Otherwise, always on-heap access.
2444 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2445
2446 if (!can_access_non_heap) {
2447 decorators |= IN_HEAP;
2448 }
2449
2450 Node* val = is_store ? argument(4 + (is_flat ? 1 : 0)) : nullptr;
2451
2452 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2453 if (adr_type == TypePtr::NULL_PTR) {
2454 set_map(old_map);
2455 set_sp(old_sp);
2456 return false; // off-heap access with zero address
2457 }
2458
2459 // Try to categorize the address.
2460 Compile::AliasType* alias_type = C->alias_type(adr_type);
2461 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2462
2463 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2464 alias_type->adr_type() == TypeAryPtr::RANGE) {
2465 set_map(old_map);
2466 set_sp(old_sp);
2467 return false; // not supported
2468 }
2469
2470 bool mismatched = false;
2471 BasicType bt = T_ILLEGAL;
2472 ciField* field = nullptr;
2473 if (adr_type->isa_instptr()) {
2474 const TypeInstPtr* instptr = adr_type->is_instptr();
2475 ciInstanceKlass* k = instptr->instance_klass();
2476 int off = instptr->offset();
2477 if (instptr->const_oop() != nullptr &&
2478 k == ciEnv::current()->Class_klass() &&
2479 instptr->offset() >= (k->size_helper() * wordSize)) {
2480 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2481 field = k->get_field_by_offset(off, true);
2482 } else {
2483 field = k->get_non_flat_field_by_offset(off);
2484 }
2485 if (field != nullptr) {
2486 bt = type2field[field->type()->basic_type()];
2487 }
2488 assert(bt == alias_type->basic_type() || is_flat, "should match");
2489 } else {
2490 bt = alias_type->basic_type();
2491 }
2492
2493 if (bt != T_ILLEGAL) {
2494 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2495 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2496 // Alias type doesn't differentiate between byte[] and boolean[]).
2497 // Use address type to get the element type.
2498 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2499 }
2500 if (is_reference_type(bt, true)) {
2501 // accessing an array field with getReference is not a mismatch
2502 bt = T_OBJECT;
2503 }
2504 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2505 // Don't intrinsify mismatched object accesses
2506 set_map(old_map);
2507 set_sp(old_sp);
2508 return false;
2509 }
2510 mismatched = (bt != type);
2511 } else if (alias_type->adr_type()->isa_oopptr()) {
2512 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2513 }
2514
2515 if (is_flat) {
2516 if (adr_type->isa_instptr()) {
2517 if (field == nullptr || field->type() != inline_klass) {
2518 mismatched = true;
2519 }
2520 } else if (adr_type->isa_aryptr()) {
2521 const Type* elem = adr_type->is_aryptr()->elem();
2522 if (!adr_type->is_flat() || elem->inline_klass() != inline_klass) {
2523 mismatched = true;
2524 }
2525 } else {
2526 mismatched = true;
2527 }
2528 if (is_store) {
2529 const Type* val_t = _gvn.type(val);
2530 if (!val_t->is_inlinetypeptr() || val_t->inline_klass() != inline_klass) {
2531 set_map(old_map);
2532 set_sp(old_sp);
2533 return false;
2534 }
2535 }
2536 }
2537
2538 destruct_map_clone(old_map);
2539 assert(!mismatched || is_flat || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2540
2541 if (mismatched) {
2542 decorators |= C2_MISMATCHED;
2543 }
2544
2545 // First guess at the value type.
2546 const Type *value_type = Type::get_const_basic_type(type);
2547
2548 // Figure out the memory ordering.
2549 decorators |= mo_decorator_for_access_kind(kind);
2550
2551 if (!is_store) {
2552 if (type == T_OBJECT && !is_flat) {
2553 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2554 if (tjp != nullptr) {
2555 value_type = tjp;
2556 }
2557 }
2558 }
2559
2560 receiver = null_check(receiver);
2561 if (stopped()) {
2562 return true;
2563 }
2564 // Heap pointers get a null-check from the interpreter,
2565 // as a courtesy. However, this is not guaranteed by Unsafe,
2566 // and it is not possible to fully distinguish unintended nulls
2567 // from intended ones in this API.
2568
2569 if (!is_store) {
2570 Node* p = nullptr;
2571 // Try to constant fold a load from a constant field
2572
2573 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !field->is_flat() && !mismatched) {
2574 // final or stable field
2575 p = make_constant_from_field(field, heap_base_oop);
2576 }
2577
2578 if (p == nullptr) { // Could not constant fold the load
2579 if (is_flat) {
2580 if (adr_type->isa_instptr() && !mismatched) {
2581 ciInstanceKlass* holder = adr_type->is_instptr()->instance_klass();
2582 int offset = adr_type->is_instptr()->offset();
2583 p = InlineTypeNode::make_from_flat(this, inline_klass, base, base, holder, offset, decorators);
2584 } else {
2585 p = InlineTypeNode::make_from_flat(this, inline_klass, base, adr, nullptr, 0, decorators);
2586 }
2587 } else {
2588 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2589 const TypeOopPtr* ptr = value_type->make_oopptr();
2590 if (ptr != nullptr && ptr->is_inlinetypeptr()) {
2591 // Load a non-flattened inline type from memory
2592 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass(), !ptr->maybe_null());
2593 }
2594 }
2595 // Normalize the value returned by getBoolean in the following cases
2596 if (type == T_BOOLEAN &&
2597 (mismatched ||
2598 heap_base_oop == top() || // - heap_base_oop is null or
2599 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2600 // and the unsafe access is made to large offset
2601 // (i.e., larger than the maximum offset necessary for any
2602 // field access)
2603 ) {
2604 IdealKit ideal = IdealKit(this);
2605 #define __ ideal.
2606 IdealVariable normalized_result(ideal);
2607 __ declarations_done();
2608 __ set(normalized_result, p);
2609 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2610 __ set(normalized_result, ideal.ConI(1));
2611 ideal.end_if();
2612 final_sync(ideal);
2613 p = __ value(normalized_result);
2614 #undef __
2615 }
2616 }
2617 if (type == T_ADDRESS) {
2618 p = gvn().transform(new CastP2XNode(nullptr, p));
2619 p = ConvX2UL(p);
2620 }
2621 // The load node has the control of the preceding MemBarCPUOrder. All
2622 // following nodes will have the control of the MemBarCPUOrder inserted at
2623 // the end of this method. So, pushing the load onto the stack at a later
2624 // point is fine.
2625 set_result(p);
2626 } else {
2627 if (bt == T_ADDRESS) {
2628 // Repackage the long as a pointer.
2629 val = ConvL2X(val);
2630 val = gvn().transform(new CastX2PNode(val));
2631 }
2632 if (is_flat) {
2633 if (adr_type->isa_instptr() && !mismatched) {
2634 ciInstanceKlass* holder = adr_type->is_instptr()->instance_klass();
2635 int offset = adr_type->is_instptr()->offset();
2636 val->as_InlineType()->store_flat(this, base, base, holder, offset, decorators);
2637 } else {
2638 val->as_InlineType()->store_flat(this, base, adr, nullptr, 0, decorators);
2639 }
2640 } else {
2641 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2642 }
2643 }
2644
2645 if (argument(1)->is_InlineType() && is_store) {
2646 InlineTypeNode* value = InlineTypeNode::make_from_oop(this, base, _gvn.type(argument(1))->inline_klass());
2647 value = value->make_larval(this, false);
2648 replace_in_map(argument(1), value);
2649 }
2650
2651 return true;
2652 }
2653
2654 bool LibraryCallKit::inline_unsafe_make_private_buffer() {
2655 Node* receiver = argument(0);
2656 Node* value = argument(1);
2657 if (!value->is_InlineType()) {
2658 return false;
2659 }
2660
2661 receiver = null_check(receiver);
2662 if (stopped()) {
2663 return true;
2664 }
2665
2666 set_result(value->as_InlineType()->make_larval(this, true));
2667 return true;
2668 }
2669
2670 bool LibraryCallKit::inline_unsafe_finish_private_buffer() {
2671 Node* receiver = argument(0);
2672 Node* buffer = argument(1);
2673 if (!buffer->is_InlineType()) {
2674 return false;
2675 }
2676 InlineTypeNode* vt = buffer->as_InlineType();
2677 if (!vt->is_allocated(&_gvn)) {
2678 return false;
2679 }
2680 // TODO 8239003 Why is this needed?
2681 if (AllocateNode::Ideal_allocation(vt->get_oop()) == nullptr) {
2682 return false;
2683 }
2684
2685 receiver = null_check(receiver);
2686 if (stopped()) {
2687 return true;
2688 }
2689
2690 set_result(vt->finish_larval(this));
2691 return true;
2692 }
2693
2694 //----------------------------inline_unsafe_load_store----------------------------
2695 // This method serves a couple of different customers (depending on LoadStoreKind):
2696 //
2697 // LS_cmp_swap:
2698 //
2699 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2700 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2701 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2702 //
2703 // LS_cmp_swap_weak:
2704 //
2705 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2706 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2707 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2708 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2709 //
2710 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2711 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2712 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2713 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2714 //
2715 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2716 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2717 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2718 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2719 //
2720 // LS_cmp_exchange:
2721 //
2722 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2723 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2724 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2725 //
2726 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2727 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2728 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2729 //
2730 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2731 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2732 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2733 //
2734 // LS_get_add:
2735 //
2736 // int getAndAddInt( Object o, long offset, int delta)
2737 // long getAndAddLong(Object o, long offset, long delta)
2738 //
2739 // LS_get_set:
2740 //
2741 // int getAndSet(Object o, long offset, int newValue)
2742 // long getAndSet(Object o, long offset, long newValue)
2743 // Object getAndSet(Object o, long offset, Object newValue)
2744 //
2745 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2746 // This basic scheme here is the same as inline_unsafe_access, but
2747 // differs in enough details that combining them would make the code
2748 // overly confusing. (This is a true fact! I originally combined
2749 // them, but even I was confused by it!) As much code/comments as
2750 // possible are retained from inline_unsafe_access though to make
2751 // the correspondences clearer. - dl
2752
2753 if (callee()->is_static()) return false; // caller must have the capability!
2754
2755 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2756 decorators |= mo_decorator_for_access_kind(access_kind);
2757
2758 #ifndef PRODUCT
2759 BasicType rtype;
2760 {
2761 ResourceMark rm;
2762 // Check the signatures.
2763 ciSignature* sig = callee()->signature();
2764 rtype = sig->return_type()->basic_type();
2765 switch(kind) {
2766 case LS_get_add:
2767 case LS_get_set: {
2768 // Check the signatures.
2769 #ifdef ASSERT
2770 assert(rtype == type, "get and set must return the expected type");
2771 assert(sig->count() == 3, "get and set has 3 arguments");
2772 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2773 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2774 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2775 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2776 #endif // ASSERT
2777 break;
2778 }
2779 case LS_cmp_swap:
2780 case LS_cmp_swap_weak: {
2781 // Check the signatures.
2782 #ifdef ASSERT
2783 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2784 assert(sig->count() == 4, "CAS has 4 arguments");
2785 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2786 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2787 #endif // ASSERT
2788 break;
2789 }
2790 case LS_cmp_exchange: {
2791 // Check the signatures.
2792 #ifdef ASSERT
2793 assert(rtype == type, "CAS must return the expected type");
2794 assert(sig->count() == 4, "CAS has 4 arguments");
2795 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2796 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2797 #endif // ASSERT
2798 break;
2799 }
2800 default:
2801 ShouldNotReachHere();
2802 }
2803 }
2804 #endif //PRODUCT
2805
2806 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2807
2808 // Get arguments:
2809 Node* receiver = nullptr;
2810 Node* base = nullptr;
2811 Node* offset = nullptr;
2812 Node* oldval = nullptr;
2813 Node* newval = nullptr;
2814 switch(kind) {
2815 case LS_cmp_swap:
2816 case LS_cmp_swap_weak:
2817 case LS_cmp_exchange: {
2818 const bool two_slot_type = type2size[type] == 2;
2819 receiver = argument(0); // type: oop
2820 base = argument(1); // type: oop
2821 offset = argument(2); // type: long
2822 oldval = argument(4); // type: oop, int, or long
2823 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2824 break;
2825 }
2826 case LS_get_add:
2827 case LS_get_set: {
2828 receiver = argument(0); // type: oop
2829 base = argument(1); // type: oop
2830 offset = argument(2); // type: long
2831 oldval = nullptr;
2832 newval = argument(4); // type: oop, int, or long
2833 break;
2834 }
2835 default:
2836 ShouldNotReachHere();
2837 }
2838
2839 // Build field offset expression.
2840 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2841 // to be plain byte offsets, which are also the same as those accepted
2842 // by oopDesc::field_addr.
2843 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2844 // 32-bit machines ignore the high half of long offsets
2845 offset = ConvL2X(offset);
2846 // Save state and restore on bailout
2847 uint old_sp = sp();
2848 SafePointNode* old_map = clone_map();
2849 Node* adr = make_unsafe_address(base, offset,type, false);
2850 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2851
2852 Compile::AliasType* alias_type = C->alias_type(adr_type);
2853 BasicType bt = alias_type->basic_type();
2854 if (bt != T_ILLEGAL &&
2855 (is_reference_type(bt) != (type == T_OBJECT))) {
2856 // Don't intrinsify mismatched object accesses.
2857 set_map(old_map);
2858 set_sp(old_sp);
2859 return false;
2860 }
2861
2862 destruct_map_clone(old_map);
2863
2864 // For CAS, unlike inline_unsafe_access, there seems no point in
2865 // trying to refine types. Just use the coarse types here.
2866 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2867 const Type *value_type = Type::get_const_basic_type(type);
2868
2869 switch (kind) {
2870 case LS_get_set:
2871 case LS_cmp_exchange: {
2872 if (type == T_OBJECT) {
2873 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2874 if (tjp != nullptr) {
2875 value_type = tjp;
2876 }
2877 }
2878 break;
2879 }
2880 case LS_cmp_swap:
2881 case LS_cmp_swap_weak:
2882 case LS_get_add:
2883 break;
2884 default:
2885 ShouldNotReachHere();
2886 }
2887
2888 // Null check receiver.
2889 receiver = null_check(receiver);
2890 if (stopped()) {
2891 return true;
2892 }
2893
2894 int alias_idx = C->get_alias_index(adr_type);
2895
2896 if (is_reference_type(type)) {
2897 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2898
2899 if (oldval != nullptr && oldval->is_InlineType()) {
2900 // Re-execute the unsafe access if allocation triggers deoptimization.
2901 PreserveReexecuteState preexecs(this);
2902 jvms()->set_should_reexecute(true);
2903 oldval = oldval->as_InlineType()->buffer(this)->get_oop();
2904 }
2905 if (newval != nullptr && newval->is_InlineType()) {
2906 // Re-execute the unsafe access if allocation triggers deoptimization.
2907 PreserveReexecuteState preexecs(this);
2908 jvms()->set_should_reexecute(true);
2909 newval = newval->as_InlineType()->buffer(this)->get_oop();
2910 }
2911
2912 // Transformation of a value which could be null pointer (CastPP #null)
2913 // could be delayed during Parse (for example, in adjust_map_after_if()).
2914 // Execute transformation here to avoid barrier generation in such case.
2915 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2916 newval = _gvn.makecon(TypePtr::NULL_PTR);
2917
2918 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2919 // Refine the value to a null constant, when it is known to be null
2920 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2921 }
2922 }
2923
2924 Node* result = nullptr;
2925 switch (kind) {
2926 case LS_cmp_exchange: {
2927 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2928 oldval, newval, value_type, type, decorators);
2929 break;
2930 }
2931 case LS_cmp_swap_weak:
2932 decorators |= C2_WEAK_CMPXCHG;
2933 case LS_cmp_swap: {
2934 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2935 oldval, newval, value_type, type, decorators);
2936 break;
2937 }
2938 case LS_get_set: {
2939 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2940 newval, value_type, type, decorators);
2941 break;
2942 }
2943 case LS_get_add: {
2944 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2945 newval, value_type, type, decorators);
2946 break;
2947 }
2948 default:
2949 ShouldNotReachHere();
2950 }
2951
2952 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2953 set_result(result);
2954 return true;
2955 }
2956
2957 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2958 // Regardless of form, don't allow previous ld/st to move down,
2959 // then issue acquire, release, or volatile mem_bar.
2960 insert_mem_bar(Op_MemBarCPUOrder);
2961 switch(id) {
2962 case vmIntrinsics::_loadFence:
2963 insert_mem_bar(Op_LoadFence);
2964 return true;
2965 case vmIntrinsics::_storeFence:
2966 insert_mem_bar(Op_StoreFence);
2967 return true;
2968 case vmIntrinsics::_storeStoreFence:
2969 insert_mem_bar(Op_StoreStoreFence);
2970 return true;
2971 case vmIntrinsics::_fullFence:
2972 insert_mem_bar(Op_MemBarVolatile);
2973 return true;
2974 default:
2975 fatal_unexpected_iid(id);
2976 return false;
2977 }
2978 }
2979
2980 bool LibraryCallKit::inline_onspinwait() {
2981 insert_mem_bar(Op_OnSpinWait);
2982 return true;
2983 }
2984
2985 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2986 if (!kls->is_Con()) {
2987 return true;
2988 }
2989 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2990 if (klsptr == nullptr) {
2991 return true;
2992 }
2993 ciInstanceKlass* ik = klsptr->instance_klass();
2994 // don't need a guard for a klass that is already initialized
2995 return !ik->is_initialized();
2996 }
2997
2998 //----------------------------inline_unsafe_writeback0-------------------------
2999 // public native void Unsafe.writeback0(long address)
3000 bool LibraryCallKit::inline_unsafe_writeback0() {
3001 if (!Matcher::has_match_rule(Op_CacheWB)) {
3002 return false;
3003 }
3004 #ifndef PRODUCT
3005 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
3006 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
3007 ciSignature* sig = callee()->signature();
3008 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
3009 #endif
3010 null_check_receiver(); // null-check, then ignore
3011 Node *addr = argument(1);
3012 addr = new CastX2PNode(addr);
3013 addr = _gvn.transform(addr);
3014 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
3015 flush = _gvn.transform(flush);
3016 set_memory(flush, TypeRawPtr::BOTTOM);
3017 return true;
3018 }
3019
3020 //----------------------------inline_unsafe_writeback0-------------------------
3021 // public native void Unsafe.writeback0(long address)
3022 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
3023 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
3024 return false;
3025 }
3026 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
3027 return false;
3028 }
3029 #ifndef PRODUCT
3030 assert(Matcher::has_match_rule(Op_CacheWB),
3031 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
3032 : "found match rule for CacheWBPostSync but not CacheWB"));
3033
3034 #endif
3035 null_check_receiver(); // null-check, then ignore
3036 Node *sync;
3037 if (is_pre) {
3038 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3039 } else {
3040 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3041 }
3042 sync = _gvn.transform(sync);
3043 set_memory(sync, TypeRawPtr::BOTTOM);
3044 return true;
3045 }
3046
3047 //----------------------------inline_unsafe_allocate---------------------------
3048 // public native Object Unsafe.allocateInstance(Class<?> cls);
3049 bool LibraryCallKit::inline_unsafe_allocate() {
3050
3051 #if INCLUDE_JVMTI
3052 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3053 return false;
3054 }
3055 #endif //INCLUDE_JVMTI
3056
3057 if (callee()->is_static()) return false; // caller must have the capability!
3058
3059 null_check_receiver(); // null-check, then ignore
3060 Node* cls = null_check(argument(1));
3061 if (stopped()) return true;
3062
3063 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
3064 kls = null_check(kls);
3065 if (stopped()) return true; // argument was like int.class
3066
3067 #if INCLUDE_JVMTI
3068 // Don't try to access new allocated obj in the intrinsic.
3069 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
3070 // Deoptimize and allocate in interpreter instead.
3071 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
3072 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
3073 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
3074 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
3075 {
3076 BuildCutout unless(this, tst, PROB_MAX);
3077 uncommon_trap(Deoptimization::Reason_intrinsic,
3078 Deoptimization::Action_make_not_entrant);
3079 }
3080 if (stopped()) {
3081 return true;
3082 }
3083 #endif //INCLUDE_JVMTI
3084
3085 Node* test = nullptr;
3086 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3087 // Note: The argument might still be an illegal value like
3088 // Serializable.class or Object[].class. The runtime will handle it.
3089 // But we must make an explicit check for initialization.
3090 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3091 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3092 // can generate code to load it as unsigned byte.
3093 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
3094 Node* bits = intcon(InstanceKlass::fully_initialized);
3095 test = _gvn.transform(new SubINode(inst, bits));
3096 // The 'test' is non-zero if we need to take a slow path.
3097 }
3098 Node* obj = nullptr;
3099 const TypeInstKlassPtr* tkls = _gvn.type(kls)->isa_instklassptr();
3100 if (tkls != nullptr && tkls->instance_klass()->is_inlinetype()) {
3101 obj = InlineTypeNode::make_default(_gvn, tkls->instance_klass()->as_inline_klass())->buffer(this);
3102 } else {
3103 obj = new_instance(kls, test);
3104 }
3105 set_result(obj);
3106 return true;
3107 }
3108
3109 //------------------------inline_native_time_funcs--------------
3110 // inline code for System.currentTimeMillis() and System.nanoTime()
3111 // these have the same type and signature
3112 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3113 const TypeFunc* tf = OptoRuntime::void_long_Type();
3114 const TypePtr* no_memory_effects = nullptr;
3115 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3116 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3117 #ifdef ASSERT
3118 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3119 assert(value_top == top(), "second value must be top");
3120 #endif
3121 set_result(value);
3122 return true;
3123 }
3124
3125
3126 #if INCLUDE_JVMTI
3127
3128 // When notifications are disabled then just update the VTMS transition bit and return.
3129 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
3130 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
3131 if (!DoJVMTIVirtualThreadTransitions) {
3132 return true;
3133 }
3134 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3135 IdealKit ideal(this);
3136
3137 Node* ONE = ideal.ConI(1);
3138 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
3139 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
3140 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3141
3142 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
3143 sync_kit(ideal);
3144 // if notifyJvmti enabled then make a call to the given SharedRuntime function
3145 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
3146 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
3147 ideal.sync_kit(this);
3148 } ideal.else_(); {
3149 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
3150 Node* thread = ideal.thread();
3151 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
3152 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
3153 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3154
3155 sync_kit(ideal);
3156 access_store_at(nullptr, jt_addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3157 access_store_at(nullptr, vt_addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3158
3159 ideal.sync_kit(this);
3160 } ideal.end_if();
3161 final_sync(ideal);
3162
3163 return true;
3164 }
3165
3166 // Always update the temporary VTMS transition bit.
3167 bool LibraryCallKit::inline_native_notify_jvmti_hide() {
3168 if (!DoJVMTIVirtualThreadTransitions) {
3169 return true;
3170 }
3171 IdealKit ideal(this);
3172
3173 {
3174 // unconditionally update the temporary VTMS transition bit in current JavaThread
3175 Node* thread = ideal.thread();
3176 Node* hide = _gvn.transform(argument(0)); // hide argument for temporary VTMS transition notification
3177 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_tmp_VTMS_transition_offset()));
3178 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3179
3180 sync_kit(ideal);
3181 access_store_at(nullptr, addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3182 ideal.sync_kit(this);
3183 }
3184 final_sync(ideal);
3185
3186 return true;
3187 }
3188
3189 // Always update the is_disable_suspend bit.
3190 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3191 if (!DoJVMTIVirtualThreadTransitions) {
3192 return true;
3193 }
3194 IdealKit ideal(this);
3195
3196 {
3197 // unconditionally update the is_disable_suspend bit in current JavaThread
3198 Node* thread = ideal.thread();
3199 Node* arg = _gvn.transform(argument(0)); // argument for notification
3200 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3201 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3202
3203 sync_kit(ideal);
3204 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3205 ideal.sync_kit(this);
3206 }
3207 final_sync(ideal);
3208
3209 return true;
3210 }
3211
3212 #endif // INCLUDE_JVMTI
3213
3214 #ifdef JFR_HAVE_INTRINSICS
3215
3216 /**
3217 * if oop->klass != null
3218 * // normal class
3219 * epoch = _epoch_state ? 2 : 1
3220 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3221 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3222 * }
3223 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3224 * else
3225 * // primitive class
3226 * if oop->array_klass != null
3227 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3228 * else
3229 * id = LAST_TYPE_ID + 1 // void class path
3230 * if (!signaled)
3231 * signaled = true
3232 */
3233 bool LibraryCallKit::inline_native_classID() {
3234 Node* cls = argument(0);
3235
3236 IdealKit ideal(this);
3237 #define __ ideal.
3238 IdealVariable result(ideal); __ declarations_done();
3239 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3240 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3241 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3242
3243
3244 __ if_then(kls, BoolTest::ne, null()); {
3245 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3246 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3247
3248 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3249 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3250 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3251 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3252 mask = _gvn.transform(new OrLNode(mask, epoch));
3253 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3254
3255 float unlikely = PROB_UNLIKELY(0.999);
3256 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3257 sync_kit(ideal);
3258 make_runtime_call(RC_LEAF,
3259 OptoRuntime::class_id_load_barrier_Type(),
3260 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3261 "class id load barrier",
3262 TypePtr::BOTTOM,
3263 kls);
3264 ideal.sync_kit(this);
3265 } __ end_if();
3266
3267 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3268 } __ else_(); {
3269 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3270 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3271 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3272 __ if_then(array_kls, BoolTest::ne, null()); {
3273 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3274 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3275 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3276 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3277 } __ else_(); {
3278 // void class case
3279 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3280 } __ end_if();
3281
3282 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3283 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3284 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3285 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3286 } __ end_if();
3287 } __ end_if();
3288
3289 final_sync(ideal);
3290 set_result(ideal.value(result));
3291 #undef __
3292 return true;
3293 }
3294
3295 //------------------------inline_native_jvm_commit------------------
3296 bool LibraryCallKit::inline_native_jvm_commit() {
3297 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3298
3299 // Save input memory and i_o state.
3300 Node* input_memory_state = reset_memory();
3301 set_all_memory(input_memory_state);
3302 Node* input_io_state = i_o();
3303
3304 // TLS.
3305 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3306 // Jfr java buffer.
3307 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3308 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3309 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3310
3311 // Load the current value of the notified field in the JfrThreadLocal.
3312 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3313 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3314
3315 // Test for notification.
3316 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3317 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3318 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3319
3320 // True branch, is notified.
3321 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3322 set_control(is_notified);
3323
3324 // Reset notified state.
3325 Node* notified_reset_memory = store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered);
3326
3327 // 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.
3328 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3329 // Convert the machine-word to a long.
3330 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3331
3332 // False branch, not notified.
3333 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3334 set_control(not_notified);
3335 set_all_memory(input_memory_state);
3336
3337 // Arg is the next position as a long.
3338 Node* arg = argument(0);
3339 // Convert long to machine-word.
3340 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3341
3342 // Store the next_position to the underlying jfr java buffer.
3343 Node* commit_memory;
3344 #ifdef _LP64
3345 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_LONG, Compile::AliasIdxRaw, MemNode::release);
3346 #else
3347 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_INT, Compile::AliasIdxRaw, MemNode::release);
3348 #endif
3349
3350 // 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.
3351 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3352 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3353 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3354
3355 // And flags with lease constant.
3356 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3357
3358 // Branch on lease to conditionalize returning the leased java buffer.
3359 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3360 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3361 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3362
3363 // False branch, not a lease.
3364 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3365
3366 // True branch, is lease.
3367 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3368 set_control(is_lease);
3369
3370 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3371 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3372 OptoRuntime::void_void_Type(),
3373 StubRoutines::jfr_return_lease(),
3374 "return_lease", TypePtr::BOTTOM);
3375 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3376
3377 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3378 record_for_igvn(lease_compare_rgn);
3379 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3380 record_for_igvn(lease_compare_mem);
3381 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3382 record_for_igvn(lease_compare_io);
3383 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3384 record_for_igvn(lease_result_value);
3385
3386 // Update control and phi nodes.
3387 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3388 lease_compare_rgn->init_req(_false_path, not_lease);
3389
3390 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3391 lease_compare_mem->init_req(_false_path, commit_memory);
3392
3393 lease_compare_io->init_req(_true_path, i_o());
3394 lease_compare_io->init_req(_false_path, input_io_state);
3395
3396 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3397 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3398
3399 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3400 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3401 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3402 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3403
3404 // Update control and phi nodes.
3405 result_rgn->init_req(_true_path, is_notified);
3406 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3407
3408 result_mem->init_req(_true_path, notified_reset_memory);
3409 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3410
3411 result_io->init_req(_true_path, input_io_state);
3412 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3413
3414 result_value->init_req(_true_path, current_pos);
3415 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3416
3417 // Set output state.
3418 set_control(_gvn.transform(result_rgn));
3419 set_all_memory(_gvn.transform(result_mem));
3420 set_i_o(_gvn.transform(result_io));
3421 set_result(result_rgn, result_value);
3422 return true;
3423 }
3424
3425 /*
3426 * The intrinsic is a model of this pseudo-code:
3427 *
3428 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3429 * jobject h_event_writer = tl->java_event_writer();
3430 * if (h_event_writer == nullptr) {
3431 * return nullptr;
3432 * }
3433 * oop threadObj = Thread::threadObj();
3434 * oop vthread = java_lang_Thread::vthread(threadObj);
3435 * traceid tid;
3436 * bool excluded;
3437 * if (vthread != threadObj) { // i.e. current thread is virtual
3438 * tid = java_lang_Thread::tid(vthread);
3439 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3440 * excluded = vthread_epoch_raw & excluded_mask;
3441 * if (!excluded) {
3442 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3443 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3444 * if (vthread_epoch != current_epoch) {
3445 * write_checkpoint();
3446 * }
3447 * }
3448 * } else {
3449 * tid = java_lang_Thread::tid(threadObj);
3450 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3451 * excluded = thread_epoch_raw & excluded_mask;
3452 * }
3453 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3454 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3455 * if (tid_in_event_writer != tid) {
3456 * setField(event_writer, "threadID", tid);
3457 * setField(event_writer, "excluded", excluded);
3458 * }
3459 * return event_writer
3460 */
3461 bool LibraryCallKit::inline_native_getEventWriter() {
3462 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3463
3464 // Save input memory and i_o state.
3465 Node* input_memory_state = reset_memory();
3466 set_all_memory(input_memory_state);
3467 Node* input_io_state = i_o();
3468
3469 Node* excluded_mask = _gvn.intcon(32768);
3470 Node* epoch_mask = _gvn.intcon(32767);
3471
3472 // TLS
3473 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3474
3475 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3476 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3477
3478 // Load the eventwriter jobject handle.
3479 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3480
3481 // Null check the jobject handle.
3482 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3483 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3484 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3485
3486 // False path, jobj is null.
3487 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3488
3489 // True path, jobj is not null.
3490 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3491
3492 set_control(jobj_is_not_null);
3493
3494 // Load the threadObj for the CarrierThread.
3495 Node* threadObj = generate_current_thread(tls_ptr);
3496
3497 // Load the vthread.
3498 Node* vthread = generate_virtual_thread(tls_ptr);
3499
3500 // If vthread != threadObj, this is a virtual thread.
3501 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3502 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3503 IfNode* iff_vthread_not_equal_threadObj =
3504 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3505
3506 // False branch, fallback to threadObj.
3507 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3508 set_control(vthread_equal_threadObj);
3509
3510 // Load the tid field from the vthread object.
3511 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3512
3513 // Load the raw epoch value from the threadObj.
3514 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3515 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3516 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3517
3518 // Mask off the excluded information from the epoch.
3519 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3520
3521 // True branch, this is a virtual thread.
3522 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3523 set_control(vthread_not_equal_threadObj);
3524
3525 // Load the tid field from the vthread object.
3526 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3527
3528 // Load the raw epoch value from the vthread.
3529 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3530 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3531 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3532
3533 // Mask off the excluded information from the epoch.
3534 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3535
3536 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3537 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3538 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3539 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3540
3541 // False branch, vthread is excluded, no need to write epoch info.
3542 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3543
3544 // True branch, vthread is included, update epoch info.
3545 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3546 set_control(included);
3547
3548 // Get epoch value.
3549 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3550
3551 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3552 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3553 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3554
3555 // Compare the epoch in the vthread to the current epoch generation.
3556 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3557 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3558 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3559
3560 // False path, epoch is equal, checkpoint information is valid.
3561 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3562
3563 // True path, epoch is not equal, write a checkpoint for the vthread.
3564 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3565
3566 set_control(epoch_is_not_equal);
3567
3568 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3569 // The call also updates the native thread local thread id and the vthread with the current epoch.
3570 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3571 OptoRuntime::jfr_write_checkpoint_Type(),
3572 StubRoutines::jfr_write_checkpoint(),
3573 "write_checkpoint", TypePtr::BOTTOM);
3574 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3575
3576 // vthread epoch != current epoch
3577 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3578 record_for_igvn(epoch_compare_rgn);
3579 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3580 record_for_igvn(epoch_compare_mem);
3581 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3582 record_for_igvn(epoch_compare_io);
3583
3584 // Update control and phi nodes.
3585 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3586 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3587 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3588 epoch_compare_mem->init_req(_false_path, input_memory_state);
3589 epoch_compare_io->init_req(_true_path, i_o());
3590 epoch_compare_io->init_req(_false_path, input_io_state);
3591
3592 // excluded != true
3593 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3594 record_for_igvn(exclude_compare_rgn);
3595 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3596 record_for_igvn(exclude_compare_mem);
3597 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3598 record_for_igvn(exclude_compare_io);
3599
3600 // Update control and phi nodes.
3601 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3602 exclude_compare_rgn->init_req(_false_path, excluded);
3603 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3604 exclude_compare_mem->init_req(_false_path, input_memory_state);
3605 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3606 exclude_compare_io->init_req(_false_path, input_io_state);
3607
3608 // vthread != threadObj
3609 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3610 record_for_igvn(vthread_compare_rgn);
3611 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3612 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3613 record_for_igvn(vthread_compare_io);
3614 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3615 record_for_igvn(tid);
3616 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3617 record_for_igvn(exclusion);
3618
3619 // Update control and phi nodes.
3620 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3621 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3622 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3623 vthread_compare_mem->init_req(_false_path, input_memory_state);
3624 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3625 vthread_compare_io->init_req(_false_path, input_io_state);
3626 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3627 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3628 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3629 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3630
3631 // Update branch state.
3632 set_control(_gvn.transform(vthread_compare_rgn));
3633 set_all_memory(_gvn.transform(vthread_compare_mem));
3634 set_i_o(_gvn.transform(vthread_compare_io));
3635
3636 // Load the event writer oop by dereferencing the jobject handle.
3637 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3638 assert(klass_EventWriter->is_loaded(), "invariant");
3639 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3640 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3641 const TypeOopPtr* const xtype = aklass->as_instance_type();
3642 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3643 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3644
3645 // Load the current thread id from the event writer object.
3646 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3647 // Get the field offset to, conditionally, store an updated tid value later.
3648 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3649 const TypePtr* event_writer_tid_field_type = _gvn.type(event_writer_tid_field)->isa_ptr();
3650 // Get the field offset to, conditionally, store an updated exclusion value later.
3651 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3652 const TypePtr* event_writer_excluded_field_type = _gvn.type(event_writer_excluded_field)->isa_ptr();
3653
3654 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3655 record_for_igvn(event_writer_tid_compare_rgn);
3656 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3657 record_for_igvn(event_writer_tid_compare_mem);
3658 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3659 record_for_igvn(event_writer_tid_compare_io);
3660
3661 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3662 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3663 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3664 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3665
3666 // False path, tids are the same.
3667 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3668
3669 // True path, tid is not equal, need to update the tid in the event writer.
3670 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3671 record_for_igvn(tid_is_not_equal);
3672
3673 // Store the exclusion state to the event writer.
3674 store_to_memory(tid_is_not_equal, event_writer_excluded_field, _gvn.transform(exclusion), T_BOOLEAN, event_writer_excluded_field_type, MemNode::unordered);
3675
3676 // Store the tid to the event writer.
3677 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, event_writer_tid_field_type, MemNode::unordered);
3678
3679 // Update control and phi nodes.
3680 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3681 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3682 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3683 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3684 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3685 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3686
3687 // Result of top level CFG, Memory, IO and Value.
3688 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3689 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3690 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3691 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3692
3693 // Result control.
3694 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3695 result_rgn->init_req(_false_path, jobj_is_null);
3696
3697 // Result memory.
3698 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3699 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3700
3701 // Result IO.
3702 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3703 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3704
3705 // Result value.
3706 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3707 result_value->init_req(_false_path, null()); // return null
3708
3709 // Set output state.
3710 set_control(_gvn.transform(result_rgn));
3711 set_all_memory(_gvn.transform(result_mem));
3712 set_i_o(_gvn.transform(result_io));
3713 set_result(result_rgn, result_value);
3714 return true;
3715 }
3716
3717 /*
3718 * The intrinsic is a model of this pseudo-code:
3719 *
3720 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3721 * if (carrierThread != thread) { // is virtual thread
3722 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3723 * bool excluded = vthread_epoch_raw & excluded_mask;
3724 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3725 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3726 * if (!excluded) {
3727 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3728 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3729 * }
3730 * Atomic::release_store(&tl->_vthread, true);
3731 * return;
3732 * }
3733 * Atomic::release_store(&tl->_vthread, false);
3734 */
3735 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3736 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3737
3738 Node* input_memory_state = reset_memory();
3739 set_all_memory(input_memory_state);
3740
3741 Node* excluded_mask = _gvn.intcon(32768);
3742 Node* epoch_mask = _gvn.intcon(32767);
3743
3744 Node* const carrierThread = generate_current_thread(jt);
3745 // If thread != carrierThread, this is a virtual thread.
3746 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3747 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3748 IfNode* iff_thread_not_equal_carrierThread =
3749 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3750
3751 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3752
3753 // False branch, is carrierThread.
3754 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3755 // Store release
3756 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3757
3758 set_all_memory(input_memory_state);
3759
3760 // True branch, is virtual thread.
3761 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3762 set_control(thread_not_equal_carrierThread);
3763
3764 // Load the raw epoch value from the vthread.
3765 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3766 Node* epoch_raw = access_load_at(thread, epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3767 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3768
3769 // Mask off the excluded information from the epoch.
3770 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3771
3772 // Load the tid field from the thread.
3773 Node* tid = load_field_from_object(thread, "tid", "J");
3774
3775 // Store the vthread tid to the jfr thread local.
3776 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3777 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, Compile::AliasIdxRaw, MemNode::unordered, true);
3778
3779 // Branch is_excluded to conditionalize updating the epoch .
3780 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3781 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3782 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3783
3784 // True branch, vthread is excluded, no need to write epoch info.
3785 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3786 set_control(excluded);
3787 Node* vthread_is_excluded = _gvn.intcon(1);
3788
3789 // False branch, vthread is included, update epoch info.
3790 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3791 set_control(included);
3792 Node* vthread_is_included = _gvn.intcon(0);
3793
3794 // Get epoch value.
3795 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3796
3797 // Store the vthread epoch to the jfr thread local.
3798 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3799 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, Compile::AliasIdxRaw, MemNode::unordered, true);
3800
3801 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3802 record_for_igvn(excluded_rgn);
3803 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3804 record_for_igvn(excluded_mem);
3805 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3806 record_for_igvn(exclusion);
3807
3808 // Merge the excluded control and memory.
3809 excluded_rgn->init_req(_true_path, excluded);
3810 excluded_rgn->init_req(_false_path, included);
3811 excluded_mem->init_req(_true_path, tid_memory);
3812 excluded_mem->init_req(_false_path, included_memory);
3813 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3814 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3815
3816 // Set intermediate state.
3817 set_control(_gvn.transform(excluded_rgn));
3818 set_all_memory(excluded_mem);
3819
3820 // Store the vthread exclusion state to the jfr thread local.
3821 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3822 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered, true);
3823
3824 // Store release
3825 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3826
3827 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3828 record_for_igvn(thread_compare_rgn);
3829 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3830 record_for_igvn(thread_compare_mem);
3831 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3832 record_for_igvn(vthread);
3833
3834 // Merge the thread_compare control and memory.
3835 thread_compare_rgn->init_req(_true_path, control());
3836 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3837 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3838 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3839
3840 // Set output state.
3841 set_control(_gvn.transform(thread_compare_rgn));
3842 set_all_memory(_gvn.transform(thread_compare_mem));
3843 }
3844
3845 #endif // JFR_HAVE_INTRINSICS
3846
3847 //------------------------inline_native_currentCarrierThread------------------
3848 bool LibraryCallKit::inline_native_currentCarrierThread() {
3849 Node* junk = nullptr;
3850 set_result(generate_current_thread(junk));
3851 return true;
3852 }
3853
3854 //------------------------inline_native_currentThread------------------
3855 bool LibraryCallKit::inline_native_currentThread() {
3856 Node* junk = nullptr;
3857 set_result(generate_virtual_thread(junk));
3858 return true;
3859 }
3860
3861 //------------------------inline_native_setVthread------------------
3862 bool LibraryCallKit::inline_native_setCurrentThread() {
3863 assert(C->method()->changes_current_thread(),
3864 "method changes current Thread but is not annotated ChangesCurrentThread");
3865 Node* arr = argument(1);
3866 Node* thread = _gvn.transform(new ThreadLocalNode());
3867 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3868 Node* thread_obj_handle
3869 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3870 thread_obj_handle = _gvn.transform(thread_obj_handle);
3871 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3872 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3873 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3874 return true;
3875 }
3876
3877 const Type* LibraryCallKit::scopedValueCache_type() {
3878 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3879 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3880 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS, /* stable= */ false, /* flat= */ false, /* not_flat= */ true, /* not_null_free= */ true);
3881
3882 // Because we create the scopedValue cache lazily we have to make the
3883 // type of the result BotPTR.
3884 bool xk = etype->klass_is_exact();
3885 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, TypeAryPtr::Offset(0));
3886 return objects_type;
3887 }
3888
3889 Node* LibraryCallKit::scopedValueCache_helper() {
3890 Node* thread = _gvn.transform(new ThreadLocalNode());
3891 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3892 // We cannot use immutable_memory() because we might flip onto a
3893 // different carrier thread, at which point we'll need to use that
3894 // carrier thread's cache.
3895 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3896 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3897 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3898 }
3899
3900 //------------------------inline_native_scopedValueCache------------------
3901 bool LibraryCallKit::inline_native_scopedValueCache() {
3902 Node* cache_obj_handle = scopedValueCache_helper();
3903 const Type* objects_type = scopedValueCache_type();
3904 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3905
3906 return true;
3907 }
3908
3909 //------------------------inline_native_setScopedValueCache------------------
3910 bool LibraryCallKit::inline_native_setScopedValueCache() {
3911 Node* arr = argument(0);
3912 Node* cache_obj_handle = scopedValueCache_helper();
3913 const Type* objects_type = scopedValueCache_type();
3914
3915 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3916 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3917
3918 return true;
3919 }
3920
3921 //-----------------------load_klass_from_mirror_common-------------------------
3922 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3923 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3924 // and branch to the given path on the region.
3925 // If never_see_null, take an uncommon trap on null, so we can optimistically
3926 // compile for the non-null case.
3927 // If the region is null, force never_see_null = true.
3928 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3929 bool never_see_null,
3930 RegionNode* region,
3931 int null_path,
3932 int offset) {
3933 if (region == nullptr) never_see_null = true;
3934 Node* p = basic_plus_adr(mirror, offset);
3935 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3936 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3937 Node* null_ctl = top();
3938 kls = null_check_oop(kls, &null_ctl, never_see_null);
3939 if (region != nullptr) {
3940 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3941 region->init_req(null_path, null_ctl);
3942 } else {
3943 assert(null_ctl == top(), "no loose ends");
3944 }
3945 return kls;
3946 }
3947
3948 //--------------------(inline_native_Class_query helpers)---------------------
3949 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3950 // Fall through if (mods & mask) == bits, take the guard otherwise.
3951 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3952 // Branch around if the given klass has the given modifier bit set.
3953 // Like generate_guard, adds a new path onto the region.
3954 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3955 Node* mods = make_load(nullptr, modp, TypeInt::INT, T_INT, MemNode::unordered);
3956 Node* mask = intcon(modifier_mask);
3957 Node* bits = intcon(modifier_bits);
3958 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3959 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3960 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3961 return generate_fair_guard(bol, region);
3962 }
3963
3964 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3965 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3966 }
3967 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3968 return generate_access_flags_guard(kls, JVM_ACC_IS_HIDDEN_CLASS, 0, region);
3969 }
3970
3971 //-------------------------inline_native_Class_query-------------------
3972 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3973 const Type* return_type = TypeInt::BOOL;
3974 Node* prim_return_value = top(); // what happens if it's a primitive class?
3975 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3976 bool expect_prim = false; // most of these guys expect to work on refs
3977
3978 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3979
3980 Node* mirror = argument(0);
3981 Node* obj = top();
3982
3983 switch (id) {
3984 case vmIntrinsics::_isInstance:
3985 // nothing is an instance of a primitive type
3986 prim_return_value = intcon(0);
3987 obj = argument(1);
3988 break;
3989 case vmIntrinsics::_getModifiers:
3990 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3991 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3992 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3993 break;
3994 case vmIntrinsics::_isInterface:
3995 prim_return_value = intcon(0);
3996 break;
3997 case vmIntrinsics::_isArray:
3998 prim_return_value = intcon(0);
3999 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
4000 break;
4001 case vmIntrinsics::_isPrimitive:
4002 prim_return_value = intcon(1);
4003 expect_prim = true; // obviously
4004 break;
4005 case vmIntrinsics::_isHidden:
4006 prim_return_value = intcon(0);
4007 break;
4008 case vmIntrinsics::_getSuperclass:
4009 prim_return_value = null();
4010 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
4011 break;
4012 case vmIntrinsics::_getClassAccessFlags:
4013 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
4014 return_type = TypeInt::INT; // not bool! 6297094
4015 break;
4016 default:
4017 fatal_unexpected_iid(id);
4018 break;
4019 }
4020
4021 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4022 if (mirror_con == nullptr) return false; // cannot happen?
4023
4024 #ifndef PRODUCT
4025 if (C->print_intrinsics() || C->print_inlining()) {
4026 ciType* k = mirror_con->java_mirror_type();
4027 if (k) {
4028 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4029 k->print_name();
4030 tty->cr();
4031 }
4032 }
4033 #endif
4034
4035 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4036 RegionNode* region = new RegionNode(PATH_LIMIT);
4037 record_for_igvn(region);
4038 PhiNode* phi = new PhiNode(region, return_type);
4039
4040 // The mirror will never be null of Reflection.getClassAccessFlags, however
4041 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4042 // if it is. See bug 4774291.
4043
4044 // For Reflection.getClassAccessFlags(), the null check occurs in
4045 // the wrong place; see inline_unsafe_access(), above, for a similar
4046 // situation.
4047 mirror = null_check(mirror);
4048 // If mirror or obj is dead, only null-path is taken.
4049 if (stopped()) return true;
4050
4051 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4052
4053 // Now load the mirror's klass metaobject, and null-check it.
4054 // Side-effects region with the control path if the klass is null.
4055 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4056 // If kls is null, we have a primitive mirror.
4057 phi->init_req(_prim_path, prim_return_value);
4058 if (stopped()) { set_result(region, phi); return true; }
4059 bool safe_for_replace = (region->in(_prim_path) == top());
4060
4061 Node* p; // handy temp
4062 Node* null_ctl;
4063
4064 // Now that we have the non-null klass, we can perform the real query.
4065 // For constant classes, the query will constant-fold in LoadNode::Value.
4066 Node* query_value = top();
4067 switch (id) {
4068 case vmIntrinsics::_isInstance:
4069 // nothing is an instance of a primitive type
4070 query_value = gen_instanceof(obj, kls, safe_for_replace);
4071 break;
4072
4073 case vmIntrinsics::_getModifiers:
4074 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
4075 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
4076 break;
4077
4078 case vmIntrinsics::_isInterface:
4079 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4080 if (generate_interface_guard(kls, region) != nullptr)
4081 // A guard was added. If the guard is taken, it was an interface.
4082 phi->add_req(intcon(1));
4083 // If we fall through, it's a plain class.
4084 query_value = intcon(0);
4085 break;
4086
4087 case vmIntrinsics::_isArray:
4088 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
4089 if (generate_array_guard(kls, region) != nullptr)
4090 // A guard was added. If the guard is taken, it was an array.
4091 phi->add_req(intcon(1));
4092 // If we fall through, it's a plain class.
4093 query_value = intcon(0);
4094 break;
4095
4096 case vmIntrinsics::_isPrimitive:
4097 query_value = intcon(0); // "normal" path produces false
4098 break;
4099
4100 case vmIntrinsics::_isHidden:
4101 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4102 if (generate_hidden_class_guard(kls, region) != nullptr)
4103 // A guard was added. If the guard is taken, it was an hidden class.
4104 phi->add_req(intcon(1));
4105 // If we fall through, it's a plain class.
4106 query_value = intcon(0);
4107 break;
4108
4109
4110 case vmIntrinsics::_getSuperclass:
4111 // The rules here are somewhat unfortunate, but we can still do better
4112 // with random logic than with a JNI call.
4113 // Interfaces store null or Object as _super, but must report null.
4114 // Arrays store an intermediate super as _super, but must report Object.
4115 // Other types can report the actual _super.
4116 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4117 if (generate_interface_guard(kls, region) != nullptr)
4118 // A guard was added. If the guard is taken, it was an interface.
4119 phi->add_req(null());
4120 if (generate_array_guard(kls, region) != nullptr)
4121 // A guard was added. If the guard is taken, it was an array.
4122 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4123 // If we fall through, it's a plain class. Get its _super.
4124 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4125 kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4126 null_ctl = top();
4127 kls = null_check_oop(kls, &null_ctl);
4128 if (null_ctl != top()) {
4129 // If the guard is taken, Object.superClass is null (both klass and mirror).
4130 region->add_req(null_ctl);
4131 phi ->add_req(null());
4132 }
4133 if (!stopped()) {
4134 query_value = load_mirror_from_klass(kls);
4135 }
4136 break;
4137
4138 case vmIntrinsics::_getClassAccessFlags:
4139 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
4140 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
4141 break;
4142
4143 default:
4144 fatal_unexpected_iid(id);
4145 break;
4146 }
4147
4148 // Fall-through is the normal case of a query to a real class.
4149 phi->init_req(1, query_value);
4150 region->init_req(1, control());
4151
4152 C->set_has_split_ifs(true); // Has chance for split-if optimization
4153 set_result(region, phi);
4154 return true;
4155 }
4156
4157
4158 //-------------------------inline_Class_cast-------------------
4159 bool LibraryCallKit::inline_Class_cast() {
4160 Node* mirror = argument(0); // Class
4161 Node* obj = argument(1);
4162 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4163 if (mirror_con == nullptr) {
4164 return false; // dead path (mirror->is_top()).
4165 }
4166 if (obj == nullptr || obj->is_top()) {
4167 return false; // dead path
4168 }
4169 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4170
4171 // First, see if Class.cast() can be folded statically.
4172 // java_mirror_type() returns non-null for compile-time Class constants.
4173 bool is_null_free_array = false;
4174 ciType* tm = mirror_con->java_mirror_type(&is_null_free_array);
4175 if (tm != nullptr && tm->is_klass() &&
4176 tp != nullptr) {
4177 if (!tp->is_loaded()) {
4178 // Don't use intrinsic when class is not loaded.
4179 return false;
4180 } else {
4181 const TypeKlassPtr* tklass = TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces);
4182 if (is_null_free_array) {
4183 tklass = tklass->is_aryklassptr()->cast_to_null_free();
4184 }
4185 int static_res = C->static_subtype_check(tklass, tp->as_klass_type());
4186 if (static_res == Compile::SSC_always_true) {
4187 // isInstance() is true - fold the code.
4188 set_result(obj);
4189 return true;
4190 } else if (static_res == Compile::SSC_always_false) {
4191 // Don't use intrinsic, have to throw ClassCastException.
4192 // If the reference is null, the non-intrinsic bytecode will
4193 // be optimized appropriately.
4194 return false;
4195 }
4196 }
4197 }
4198
4199 // Bailout intrinsic and do normal inlining if exception path is frequent.
4200 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4201 return false;
4202 }
4203
4204 // Generate dynamic checks.
4205 // Class.cast() is java implementation of _checkcast bytecode.
4206 // Do checkcast (Parse::do_checkcast()) optimizations here.
4207
4208 mirror = null_check(mirror);
4209 // If mirror is dead, only null-path is taken.
4210 if (stopped()) {
4211 return true;
4212 }
4213
4214 // Not-subtype or the mirror's klass ptr is nullptr (in case it is a primitive).
4215 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
4216 RegionNode* region = new RegionNode(PATH_LIMIT);
4217 record_for_igvn(region);
4218
4219 // Now load the mirror's klass metaobject, and null-check it.
4220 // If kls is null, we have a primitive mirror and
4221 // nothing is an instance of a primitive type.
4222 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4223
4224 Node* res = top();
4225 Node* io = i_o();
4226 Node* mem = merged_memory();
4227 if (!stopped()) {
4228
4229 Node* bad_type_ctrl = top();
4230 // Do checkcast optimizations.
4231 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4232 region->init_req(_bad_type_path, bad_type_ctrl);
4233 }
4234 if (region->in(_prim_path) != top() ||
4235 region->in(_bad_type_path) != top() ||
4236 region->in(_npe_path) != top()) {
4237 // Let Interpreter throw ClassCastException.
4238 PreserveJVMState pjvms(this);
4239 set_control(_gvn.transform(region));
4240 // Set IO and memory because gen_checkcast may override them when buffering inline types
4241 set_i_o(io);
4242 set_all_memory(mem);
4243 uncommon_trap(Deoptimization::Reason_intrinsic,
4244 Deoptimization::Action_maybe_recompile);
4245 }
4246 if (!stopped()) {
4247 set_result(res);
4248 }
4249 return true;
4250 }
4251
4252
4253 //--------------------------inline_native_subtype_check------------------------
4254 // This intrinsic takes the JNI calls out of the heart of
4255 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4256 bool LibraryCallKit::inline_native_subtype_check() {
4257 // Pull both arguments off the stack.
4258 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4259 args[0] = argument(0);
4260 args[1] = argument(1);
4261 Node* klasses[2]; // corresponding Klasses: superk, subk
4262 klasses[0] = klasses[1] = top();
4263
4264 enum {
4265 // A full decision tree on {superc is prim, subc is prim}:
4266 _prim_0_path = 1, // {P,N} => false
4267 // {P,P} & superc!=subc => false
4268 _prim_same_path, // {P,P} & superc==subc => true
4269 _prim_1_path, // {N,P} => false
4270 _ref_subtype_path, // {N,N} & subtype check wins => true
4271 _both_ref_path, // {N,N} & subtype check loses => false
4272 PATH_LIMIT
4273 };
4274
4275 RegionNode* region = new RegionNode(PATH_LIMIT);
4276 RegionNode* prim_region = new RegionNode(2);
4277 Node* phi = new PhiNode(region, TypeInt::BOOL);
4278 record_for_igvn(region);
4279 record_for_igvn(prim_region);
4280
4281 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4282 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4283 int class_klass_offset = java_lang_Class::klass_offset();
4284
4285 // First null-check both mirrors and load each mirror's klass metaobject.
4286 int which_arg;
4287 for (which_arg = 0; which_arg <= 1; which_arg++) {
4288 Node* arg = args[which_arg];
4289 arg = null_check(arg);
4290 if (stopped()) break;
4291 args[which_arg] = arg;
4292
4293 Node* p = basic_plus_adr(arg, class_klass_offset);
4294 Node* kls = LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, adr_type, kls_type);
4295 klasses[which_arg] = _gvn.transform(kls);
4296 }
4297
4298 // Having loaded both klasses, test each for null.
4299 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4300 for (which_arg = 0; which_arg <= 1; which_arg++) {
4301 Node* kls = klasses[which_arg];
4302 Node* null_ctl = top();
4303 kls = null_check_oop(kls, &null_ctl, never_see_null);
4304 if (which_arg == 0) {
4305 prim_region->init_req(1, null_ctl);
4306 } else {
4307 region->init_req(_prim_1_path, null_ctl);
4308 }
4309 if (stopped()) break;
4310 klasses[which_arg] = kls;
4311 }
4312
4313 if (!stopped()) {
4314 // now we have two reference types, in klasses[0..1]
4315 Node* subk = klasses[1]; // the argument to isAssignableFrom
4316 Node* superk = klasses[0]; // the receiver
4317 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4318 region->set_req(_ref_subtype_path, control());
4319 }
4320
4321 // If both operands are primitive (both klasses null), then
4322 // we must return true when they are identical primitives.
4323 // It is convenient to test this after the first null klass check.
4324 // This path is also used if superc is a value mirror.
4325 set_control(_gvn.transform(prim_region));
4326 if (!stopped()) {
4327 // Since superc is primitive, make a guard for the superc==subc case.
4328 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4329 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4330 generate_fair_guard(bol_eq, region);
4331 if (region->req() == PATH_LIMIT+1) {
4332 // A guard was added. If the added guard is taken, superc==subc.
4333 region->swap_edges(PATH_LIMIT, _prim_same_path);
4334 region->del_req(PATH_LIMIT);
4335 }
4336 region->set_req(_prim_0_path, control()); // Not equal after all.
4337 }
4338
4339 // these are the only paths that produce 'true':
4340 phi->set_req(_prim_same_path, intcon(1));
4341 phi->set_req(_ref_subtype_path, intcon(1));
4342
4343 // pull together the cases:
4344 assert(region->req() == PATH_LIMIT, "sane region");
4345 for (uint i = 1; i < region->req(); i++) {
4346 Node* ctl = region->in(i);
4347 if (ctl == nullptr || ctl == top()) {
4348 region->set_req(i, top());
4349 phi ->set_req(i, top());
4350 } else if (phi->in(i) == nullptr) {
4351 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4352 }
4353 }
4354
4355 set_control(_gvn.transform(region));
4356 set_result(_gvn.transform(phi));
4357 return true;
4358 }
4359
4360 //---------------------generate_array_guard_common------------------------
4361 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind) {
4362
4363 if (stopped()) {
4364 return nullptr;
4365 }
4366
4367 // Like generate_guard, adds a new path onto the region.
4368 jint layout_con = 0;
4369 Node* layout_val = get_layout_helper(kls, layout_con);
4370 if (layout_val == nullptr) {
4371 bool query = 0;
4372 switch(kind) {
4373 case ObjectArray: query = Klass::layout_helper_is_objArray(layout_con); break;
4374 case NonObjectArray: query = !Klass::layout_helper_is_objArray(layout_con); break;
4375 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
4376 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
4377 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
4378 default:
4379 ShouldNotReachHere();
4380 }
4381 if (!query) {
4382 return nullptr; // never a branch
4383 } else { // always a branch
4384 Node* always_branch = control();
4385 if (region != nullptr)
4386 region->add_req(always_branch);
4387 set_control(top());
4388 return always_branch;
4389 }
4390 }
4391 unsigned int value = 0;
4392 BoolTest::mask btest = BoolTest::illegal;
4393 switch(kind) {
4394 case ObjectArray:
4395 case NonObjectArray: {
4396 value = Klass::_lh_array_tag_obj_value;
4397 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4398 btest = (kind == ObjectArray) ? BoolTest::eq : BoolTest::ne;
4399 break;
4400 }
4401 case TypeArray: {
4402 value = Klass::_lh_array_tag_type_value;
4403 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4404 btest = BoolTest::eq;
4405 break;
4406 }
4407 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
4408 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
4409 default:
4410 ShouldNotReachHere();
4411 }
4412 // Now test the correct condition.
4413 jint nval = (jint)value;
4414 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4415 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4416 return generate_fair_guard(bol, region);
4417 }
4418
4419 //-----------------------inline_newNullRestrictedArray--------------------------
4420 // public static native Object[] newNullRestrictedArray(Class<?> componentType, int length);
4421 bool LibraryCallKit::inline_newNullRestrictedArray() {
4422 Node* componentType = argument(0);
4423 Node* length = argument(1);
4424
4425 const TypeInstPtr* tp = _gvn.type(componentType)->isa_instptr();
4426 if (tp != nullptr) {
4427 ciInstanceKlass* ik = tp->instance_klass();
4428 if (ik == C->env()->Class_klass()) {
4429 ciType* t = tp->java_mirror_type();
4430 if (t != nullptr && t->is_inlinetype()) {
4431 ciArrayKlass* array_klass = ciArrayKlass::make(t, true);
4432 if (array_klass->is_loaded() && array_klass->element_klass()->as_inline_klass()->is_initialized()) {
4433 const TypeAryKlassPtr* array_klass_type = TypeKlassPtr::make(array_klass, Type::trust_interfaces)->is_aryklassptr();
4434 array_klass_type = array_klass_type->cast_to_null_free();
4435 Node* obj = new_array(makecon(array_klass_type), length, 0, nullptr, false); // no arguments to push
4436 set_result(obj);
4437 assert(gvn().type(obj)->is_aryptr()->is_null_free(), "must be null-free");
4438 return true;
4439 }
4440 }
4441 }
4442 }
4443 return false;
4444 }
4445
4446 //-----------------------inline_native_newArray--------------------------
4447 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
4448 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4449 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4450 Node* mirror;
4451 Node* count_val;
4452 if (uninitialized) {
4453 null_check_receiver();
4454 mirror = argument(1);
4455 count_val = argument(2);
4456 } else {
4457 mirror = argument(0);
4458 count_val = argument(1);
4459 }
4460
4461 mirror = null_check(mirror);
4462 // If mirror or obj is dead, only null-path is taken.
4463 if (stopped()) return true;
4464
4465 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4466 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4467 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4468 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4469 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4470
4471 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4472 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4473 result_reg, _slow_path);
4474 Node* normal_ctl = control();
4475 Node* no_array_ctl = result_reg->in(_slow_path);
4476
4477 // Generate code for the slow case. We make a call to newArray().
4478 set_control(no_array_ctl);
4479 if (!stopped()) {
4480 // Either the input type is void.class, or else the
4481 // array klass has not yet been cached. Either the
4482 // ensuing call will throw an exception, or else it
4483 // will cache the array klass for next time.
4484 PreserveJVMState pjvms(this);
4485 CallJavaNode* slow_call = nullptr;
4486 if (uninitialized) {
4487 // Generate optimized virtual call (holder class 'Unsafe' is final)
4488 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4489 } else {
4490 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4491 }
4492 Node* slow_result = set_results_for_java_call(slow_call);
4493 // this->control() comes from set_results_for_java_call
4494 result_reg->set_req(_slow_path, control());
4495 result_val->set_req(_slow_path, slow_result);
4496 result_io ->set_req(_slow_path, i_o());
4497 result_mem->set_req(_slow_path, reset_memory());
4498 }
4499
4500 set_control(normal_ctl);
4501 if (!stopped()) {
4502 // Normal case: The array type has been cached in the java.lang.Class.
4503 // The following call works fine even if the array type is polymorphic.
4504 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4505 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4506 result_reg->init_req(_normal_path, control());
4507 result_val->init_req(_normal_path, obj);
4508 result_io ->init_req(_normal_path, i_o());
4509 result_mem->init_req(_normal_path, reset_memory());
4510
4511 if (uninitialized) {
4512 // Mark the allocation so that zeroing is skipped
4513 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4514 alloc->maybe_set_complete(&_gvn);
4515 }
4516 }
4517
4518 // Return the combined state.
4519 set_i_o( _gvn.transform(result_io) );
4520 set_all_memory( _gvn.transform(result_mem));
4521
4522 C->set_has_split_ifs(true); // Has chance for split-if optimization
4523 set_result(result_reg, result_val);
4524 return true;
4525 }
4526
4527 //----------------------inline_native_getLength--------------------------
4528 // public static native int java.lang.reflect.Array.getLength(Object array);
4529 bool LibraryCallKit::inline_native_getLength() {
4530 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4531
4532 Node* array = null_check(argument(0));
4533 // If array is dead, only null-path is taken.
4534 if (stopped()) return true;
4535
4536 // Deoptimize if it is a non-array.
4537 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr);
4538
4539 if (non_array != nullptr) {
4540 PreserveJVMState pjvms(this);
4541 set_control(non_array);
4542 uncommon_trap(Deoptimization::Reason_intrinsic,
4543 Deoptimization::Action_maybe_recompile);
4544 }
4545
4546 // If control is dead, only non-array-path is taken.
4547 if (stopped()) return true;
4548
4549 // The works fine even if the array type is polymorphic.
4550 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4551 Node* result = load_array_length(array);
4552
4553 C->set_has_split_ifs(true); // Has chance for split-if optimization
4554 set_result(result);
4555 return true;
4556 }
4557
4558 //------------------------inline_array_copyOf----------------------------
4559 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4560 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4561 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4562 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4563
4564 // Get the arguments.
4565 Node* original = argument(0);
4566 Node* start = is_copyOfRange? argument(1): intcon(0);
4567 Node* end = is_copyOfRange? argument(2): argument(1);
4568 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4569
4570 Node* newcopy = nullptr;
4571
4572 // Set the original stack and the reexecute bit for the interpreter to reexecute
4573 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4574 { PreserveReexecuteState preexecs(this);
4575 jvms()->set_should_reexecute(true);
4576
4577 array_type_mirror = null_check(array_type_mirror);
4578 original = null_check(original);
4579
4580 // Check if a null path was taken unconditionally.
4581 if (stopped()) return true;
4582
4583 Node* orig_length = load_array_length(original);
4584
4585 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4586 klass_node = null_check(klass_node);
4587
4588 RegionNode* bailout = new RegionNode(1);
4589 record_for_igvn(bailout);
4590
4591 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4592 // Bail out if that is so.
4593 // Inline type array may have object field that would require a
4594 // write barrier. Conservatively, go to slow path.
4595 // TODO 8251971: Optimize for the case when flat src/dst are later found
4596 // to not contain oops (i.e., move this check to the macro expansion phase).
4597 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4598 const TypeAryPtr* orig_t = _gvn.type(original)->isa_aryptr();
4599 const TypeKlassPtr* tklass = _gvn.type(klass_node)->is_klassptr();
4600 bool exclude_flat = UseFlatArray && bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing) &&
4601 // Can src array be flat and contain oops?
4602 (orig_t == nullptr || (!orig_t->is_not_flat() && (!orig_t->is_flat() || orig_t->elem()->inline_klass()->contains_oops()))) &&
4603 // Can dest array be flat and contain oops?
4604 tklass->can_be_inline_array() && (!tklass->is_flat() || tklass->is_aryklassptr()->elem()->is_instklassptr()->instance_klass()->as_inline_klass()->contains_oops());
4605 Node* not_objArray = exclude_flat ? generate_non_objArray_guard(klass_node, bailout) : generate_typeArray_guard(klass_node, bailout);
4606 if (not_objArray != nullptr) {
4607 // Improve the klass node's type from the new optimistic assumption:
4608 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4609 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0));
4610 Node* cast = new CastPPNode(control(), klass_node, akls);
4611 klass_node = _gvn.transform(cast);
4612 }
4613
4614 // Bail out if either start or end is negative.
4615 generate_negative_guard(start, bailout, &start);
4616 generate_negative_guard(end, bailout, &end);
4617
4618 Node* length = end;
4619 if (_gvn.type(start) != TypeInt::ZERO) {
4620 length = _gvn.transform(new SubINode(end, start));
4621 }
4622
4623 // Bail out if length is negative (i.e., if start > end).
4624 // Without this the new_array would throw
4625 // NegativeArraySizeException but IllegalArgumentException is what
4626 // should be thrown
4627 generate_negative_guard(length, bailout, &length);
4628
4629 // Handle inline type arrays
4630 bool can_validate = !too_many_traps(Deoptimization::Reason_class_check);
4631 if (!stopped()) {
4632 // TODO JDK-8329224
4633 if (!orig_t->is_null_free()) {
4634 // Not statically known to be null free, add a check
4635 generate_fair_guard(null_free_array_test(original), bailout);
4636 }
4637 orig_t = _gvn.type(original)->isa_aryptr();
4638 if (orig_t != nullptr && orig_t->is_flat()) {
4639 // Src is flat, check that dest is flat as well
4640 if (exclude_flat) {
4641 // Dest can't be flat, bail out
4642 bailout->add_req(control());
4643 set_control(top());
4644 } else {
4645 generate_fair_guard(flat_array_test(klass_node, /* flat = */ false), bailout);
4646 }
4647 } else if (UseFlatArray && (orig_t == nullptr || !orig_t->is_not_flat()) &&
4648 // If dest is flat, src must be flat as well (guaranteed by src <: dest check if validated).
4649 ((!tklass->is_flat() && tklass->can_be_inline_array()) || !can_validate)) {
4650 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
4651 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
4652 generate_fair_guard(flat_array_test(load_object_klass(original)), bailout);
4653 if (orig_t != nullptr) {
4654 orig_t = orig_t->cast_to_not_flat();
4655 original = _gvn.transform(new CheckCastPPNode(control(), original, orig_t));
4656 }
4657 }
4658 if (!can_validate) {
4659 // No validation. The subtype check emitted at macro expansion time will not go to the slow
4660 // path but call checkcast_arraycopy which can not handle flat/null-free inline type arrays.
4661 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat/null-free.
4662 generate_fair_guard(flat_array_test(klass_node), bailout);
4663 generate_fair_guard(null_free_array_test(original), bailout);
4664 }
4665 }
4666
4667 // Bail out if start is larger than the original length
4668 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4669 generate_negative_guard(orig_tail, bailout, &orig_tail);
4670
4671 if (bailout->req() > 1) {
4672 PreserveJVMState pjvms(this);
4673 set_control(_gvn.transform(bailout));
4674 uncommon_trap(Deoptimization::Reason_intrinsic,
4675 Deoptimization::Action_maybe_recompile);
4676 }
4677
4678 if (!stopped()) {
4679 // How many elements will we copy from the original?
4680 // The answer is MinI(orig_tail, length).
4681 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4682
4683 // Generate a direct call to the right arraycopy function(s).
4684 // We know the copy is disjoint but we might not know if the
4685 // oop stores need checking.
4686 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4687 // This will fail a store-check if x contains any non-nulls.
4688
4689 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4690 // loads/stores but it is legal only if we're sure the
4691 // Arrays.copyOf would succeed. So we need all input arguments
4692 // to the copyOf to be validated, including that the copy to the
4693 // new array won't trigger an ArrayStoreException. That subtype
4694 // check can be optimized if we know something on the type of
4695 // the input array from type speculation.
4696 if (_gvn.type(klass_node)->singleton()) {
4697 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4698 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4699
4700 int test = C->static_subtype_check(superk, subk);
4701 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4702 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4703 if (t_original->speculative_type() != nullptr) {
4704 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4705 }
4706 }
4707 }
4708
4709 bool validated = false;
4710 // Reason_class_check rather than Reason_intrinsic because we
4711 // want to intrinsify even if this traps.
4712 if (can_validate) {
4713 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4714
4715 if (not_subtype_ctrl != top()) {
4716 PreserveJVMState pjvms(this);
4717 set_control(not_subtype_ctrl);
4718 uncommon_trap(Deoptimization::Reason_class_check,
4719 Deoptimization::Action_make_not_entrant);
4720 assert(stopped(), "Should be stopped");
4721 }
4722 validated = true;
4723 }
4724
4725 if (!stopped()) {
4726 newcopy = new_array(klass_node, length, 0); // no arguments to push
4727
4728 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4729 load_object_klass(original), klass_node);
4730 if (!is_copyOfRange) {
4731 ac->set_copyof(validated);
4732 } else {
4733 ac->set_copyofrange(validated);
4734 }
4735 Node* n = _gvn.transform(ac);
4736 if (n == ac) {
4737 ac->connect_outputs(this);
4738 } else {
4739 assert(validated, "shouldn't transform if all arguments not validated");
4740 set_all_memory(n);
4741 }
4742 }
4743 }
4744 } // original reexecute is set back here
4745
4746 C->set_has_split_ifs(true); // Has chance for split-if optimization
4747 if (!stopped()) {
4748 set_result(newcopy);
4749 }
4750 return true;
4751 }
4752
4753
4754 //----------------------generate_virtual_guard---------------------------
4755 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4756 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4757 RegionNode* slow_region) {
4758 ciMethod* method = callee();
4759 int vtable_index = method->vtable_index();
4760 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4761 "bad index %d", vtable_index);
4762 // Get the Method* out of the appropriate vtable entry.
4763 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4764 vtable_index*vtableEntry::size_in_bytes() +
4765 in_bytes(vtableEntry::method_offset());
4766 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4767 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4768
4769 // Compare the target method with the expected method (e.g., Object.hashCode).
4770 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4771
4772 Node* native_call = makecon(native_call_addr);
4773 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4774 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4775
4776 return generate_slow_guard(test_native, slow_region);
4777 }
4778
4779 //-----------------------generate_method_call----------------------------
4780 // Use generate_method_call to make a slow-call to the real
4781 // method if the fast path fails. An alternative would be to
4782 // use a stub like OptoRuntime::slow_arraycopy_Java.
4783 // This only works for expanding the current library call,
4784 // not another intrinsic. (E.g., don't use this for making an
4785 // arraycopy call inside of the copyOf intrinsic.)
4786 CallJavaNode*
4787 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4788 // When compiling the intrinsic method itself, do not use this technique.
4789 guarantee(callee() != C->method(), "cannot make slow-call to self");
4790
4791 ciMethod* method = callee();
4792 // ensure the JVMS we have will be correct for this call
4793 guarantee(method_id == method->intrinsic_id(), "must match");
4794
4795 const TypeFunc* tf = TypeFunc::make(method);
4796 if (res_not_null) {
4797 assert(tf->return_type() == T_OBJECT, "");
4798 const TypeTuple* range = tf->range_cc();
4799 const Type** fields = TypeTuple::fields(range->cnt());
4800 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4801 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4802 tf = TypeFunc::make(tf->domain_cc(), new_range);
4803 }
4804 CallJavaNode* slow_call;
4805 if (is_static) {
4806 assert(!is_virtual, "");
4807 slow_call = new CallStaticJavaNode(C, tf,
4808 SharedRuntime::get_resolve_static_call_stub(), method);
4809 } else if (is_virtual) {
4810 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4811 int vtable_index = Method::invalid_vtable_index;
4812 if (UseInlineCaches) {
4813 // Suppress the vtable call
4814 } else {
4815 // hashCode and clone are not a miranda methods,
4816 // so the vtable index is fixed.
4817 // No need to use the linkResolver to get it.
4818 vtable_index = method->vtable_index();
4819 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4820 "bad index %d", vtable_index);
4821 }
4822 slow_call = new CallDynamicJavaNode(tf,
4823 SharedRuntime::get_resolve_virtual_call_stub(),
4824 method, vtable_index);
4825 } else { // neither virtual nor static: opt_virtual
4826 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4827 slow_call = new CallStaticJavaNode(C, tf,
4828 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4829 slow_call->set_optimized_virtual(true);
4830 }
4831 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4832 // To be able to issue a direct call (optimized virtual or virtual)
4833 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4834 // about the method being invoked should be attached to the call site to
4835 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4836 slow_call->set_override_symbolic_info(true);
4837 }
4838 set_arguments_for_java_call(slow_call);
4839 set_edges_for_java_call(slow_call);
4840 return slow_call;
4841 }
4842
4843
4844 /**
4845 * Build special case code for calls to hashCode on an object. This call may
4846 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4847 * slightly different code.
4848 */
4849 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4850 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4851 assert(!(is_virtual && is_static), "either virtual, special, or static");
4852
4853 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4854
4855 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4856 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4857 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4858 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4859 Node* obj = argument(0);
4860
4861 // Don't intrinsify hashcode on inline types for now.
4862 // The "is locked" runtime check below also serves as inline type check and goes to the slow path.
4863 if (gvn().type(obj)->is_inlinetypeptr()) {
4864 return false;
4865 }
4866
4867 if (!is_static) {
4868 // Check for hashing null object
4869 obj = null_check_receiver();
4870 if (stopped()) return true; // unconditionally null
4871 result_reg->init_req(_null_path, top());
4872 result_val->init_req(_null_path, top());
4873 } else {
4874 // Do a null check, and return zero if null.
4875 // System.identityHashCode(null) == 0
4876 Node* null_ctl = top();
4877 obj = null_check_oop(obj, &null_ctl);
4878 result_reg->init_req(_null_path, null_ctl);
4879 result_val->init_req(_null_path, _gvn.intcon(0));
4880 }
4881
4882 // Unconditionally null? Then return right away.
4883 if (stopped()) {
4884 set_control( result_reg->in(_null_path));
4885 if (!stopped())
4886 set_result(result_val->in(_null_path));
4887 return true;
4888 }
4889
4890 // We only go to the fast case code if we pass a number of guards. The
4891 // paths which do not pass are accumulated in the slow_region.
4892 RegionNode* slow_region = new RegionNode(1);
4893 record_for_igvn(slow_region);
4894
4895 // If this is a virtual call, we generate a funny guard. We pull out
4896 // the vtable entry corresponding to hashCode() from the target object.
4897 // If the target method which we are calling happens to be the native
4898 // Object hashCode() method, we pass the guard. We do not need this
4899 // guard for non-virtual calls -- the caller is known to be the native
4900 // Object hashCode().
4901 if (is_virtual) {
4902 // After null check, get the object's klass.
4903 Node* obj_klass = load_object_klass(obj);
4904 generate_virtual_guard(obj_klass, slow_region);
4905 }
4906
4907 // Get the header out of the object, use LoadMarkNode when available
4908 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4909 // The control of the load must be null. Otherwise, the load can move before
4910 // the null check after castPP removal.
4911 Node* no_ctrl = nullptr;
4912 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4913
4914 // Test the header to see if it is safe to read w.r.t. locking.
4915 // This also serves as guard against inline types
4916 Node *lock_mask = _gvn.MakeConX(markWord::inline_type_mask_in_place);
4917 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4918 if (LockingMode == LM_LIGHTWEIGHT) {
4919 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
4920 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
4921 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
4922
4923 generate_slow_guard(test_monitor, slow_region);
4924 } else {
4925 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4926 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val));
4927 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne));
4928
4929 generate_slow_guard(test_not_unlocked, slow_region);
4930 }
4931
4932 // Get the hash value and check to see that it has been properly assigned.
4933 // We depend on hash_mask being at most 32 bits and avoid the use of
4934 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4935 // vm: see markWord.hpp.
4936 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4937 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4938 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4939 // This hack lets the hash bits live anywhere in the mark object now, as long
4940 // as the shift drops the relevant bits into the low 32 bits. Note that
4941 // Java spec says that HashCode is an int so there's no point in capturing
4942 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4943 hshifted_header = ConvX2I(hshifted_header);
4944 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4945
4946 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4947 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4948 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4949
4950 generate_slow_guard(test_assigned, slow_region);
4951
4952 Node* init_mem = reset_memory();
4953 // fill in the rest of the null path:
4954 result_io ->init_req(_null_path, i_o());
4955 result_mem->init_req(_null_path, init_mem);
4956
4957 result_val->init_req(_fast_path, hash_val);
4958 result_reg->init_req(_fast_path, control());
4959 result_io ->init_req(_fast_path, i_o());
4960 result_mem->init_req(_fast_path, init_mem);
4961
4962 // Generate code for the slow case. We make a call to hashCode().
4963 set_control(_gvn.transform(slow_region));
4964 if (!stopped()) {
4965 // No need for PreserveJVMState, because we're using up the present state.
4966 set_all_memory(init_mem);
4967 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4968 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4969 Node* slow_result = set_results_for_java_call(slow_call);
4970 // this->control() comes from set_results_for_java_call
4971 result_reg->init_req(_slow_path, control());
4972 result_val->init_req(_slow_path, slow_result);
4973 result_io ->set_req(_slow_path, i_o());
4974 result_mem ->set_req(_slow_path, reset_memory());
4975 }
4976
4977 // Return the combined state.
4978 set_i_o( _gvn.transform(result_io) );
4979 set_all_memory( _gvn.transform(result_mem));
4980
4981 set_result(result_reg, result_val);
4982 return true;
4983 }
4984
4985 //---------------------------inline_native_getClass----------------------------
4986 // public final native Class<?> java.lang.Object.getClass();
4987 //
4988 // Build special case code for calls to getClass on an object.
4989 bool LibraryCallKit::inline_native_getClass() {
4990 Node* obj = argument(0);
4991 if (obj->is_InlineType()) {
4992 const Type* t = _gvn.type(obj);
4993 if (t->maybe_null()) {
4994 null_check(obj);
4995 }
4996 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror())));
4997 return true;
4998 }
4999 obj = null_check_receiver();
5000 if (stopped()) return true;
5001 set_result(load_mirror_from_klass(load_object_klass(obj)));
5002 return true;
5003 }
5004
5005 //-----------------inline_native_Reflection_getCallerClass---------------------
5006 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
5007 //
5008 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
5009 //
5010 // NOTE: This code must perform the same logic as JVM_GetCallerClass
5011 // in that it must skip particular security frames and checks for
5012 // caller sensitive methods.
5013 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
5014 #ifndef PRODUCT
5015 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5016 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
5017 }
5018 #endif
5019
5020 if (!jvms()->has_method()) {
5021 #ifndef PRODUCT
5022 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5023 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
5024 }
5025 #endif
5026 return false;
5027 }
5028
5029 // Walk back up the JVM state to find the caller at the required
5030 // depth.
5031 JVMState* caller_jvms = jvms();
5032
5033 // Cf. JVM_GetCallerClass
5034 // NOTE: Start the loop at depth 1 because the current JVM state does
5035 // not include the Reflection.getCallerClass() frame.
5036 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
5037 ciMethod* m = caller_jvms->method();
5038 switch (n) {
5039 case 0:
5040 fatal("current JVM state does not include the Reflection.getCallerClass frame");
5041 break;
5042 case 1:
5043 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
5044 if (!m->caller_sensitive()) {
5045 #ifndef PRODUCT
5046 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5047 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
5048 }
5049 #endif
5050 return false; // bail-out; let JVM_GetCallerClass do the work
5051 }
5052 break;
5053 default:
5054 if (!m->is_ignored_by_security_stack_walk()) {
5055 // We have reached the desired frame; return the holder class.
5056 // Acquire method holder as java.lang.Class and push as constant.
5057 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
5058 ciInstance* caller_mirror = caller_klass->java_mirror();
5059 set_result(makecon(TypeInstPtr::make(caller_mirror)));
5060
5061 #ifndef PRODUCT
5062 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5063 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());
5064 tty->print_cr(" JVM state at this point:");
5065 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5066 ciMethod* m = jvms()->of_depth(i)->method();
5067 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5068 }
5069 }
5070 #endif
5071 return true;
5072 }
5073 break;
5074 }
5075 }
5076
5077 #ifndef PRODUCT
5078 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5079 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
5080 tty->print_cr(" JVM state at this point:");
5081 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5082 ciMethod* m = jvms()->of_depth(i)->method();
5083 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5084 }
5085 }
5086 #endif
5087
5088 return false; // bail-out; let JVM_GetCallerClass do the work
5089 }
5090
5091 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
5092 Node* arg = argument(0);
5093 Node* result = nullptr;
5094
5095 switch (id) {
5096 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
5097 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
5098 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5099 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5100 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5101 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5102
5103 case vmIntrinsics::_doubleToLongBits: {
5104 // two paths (plus control) merge in a wood
5105 RegionNode *r = new RegionNode(3);
5106 Node *phi = new PhiNode(r, TypeLong::LONG);
5107
5108 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5109 // Build the boolean node
5110 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5111
5112 // Branch either way.
5113 // NaN case is less traveled, which makes all the difference.
5114 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5115 Node *opt_isnan = _gvn.transform(ifisnan);
5116 assert( opt_isnan->is_If(), "Expect an IfNode");
5117 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5118 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5119
5120 set_control(iftrue);
5121
5122 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5123 Node *slow_result = longcon(nan_bits); // return NaN
5124 phi->init_req(1, _gvn.transform( slow_result ));
5125 r->init_req(1, iftrue);
5126
5127 // Else fall through
5128 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5129 set_control(iffalse);
5130
5131 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5132 r->init_req(2, iffalse);
5133
5134 // Post merge
5135 set_control(_gvn.transform(r));
5136 record_for_igvn(r);
5137
5138 C->set_has_split_ifs(true); // Has chance for split-if optimization
5139 result = phi;
5140 assert(result->bottom_type()->isa_long(), "must be");
5141 break;
5142 }
5143
5144 case vmIntrinsics::_floatToIntBits: {
5145 // two paths (plus control) merge in a wood
5146 RegionNode *r = new RegionNode(3);
5147 Node *phi = new PhiNode(r, TypeInt::INT);
5148
5149 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5150 // Build the boolean node
5151 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5152
5153 // Branch either way.
5154 // NaN case is less traveled, which makes all the difference.
5155 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5156 Node *opt_isnan = _gvn.transform(ifisnan);
5157 assert( opt_isnan->is_If(), "Expect an IfNode");
5158 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5159 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5160
5161 set_control(iftrue);
5162
5163 static const jint nan_bits = 0x7fc00000;
5164 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5165 phi->init_req(1, _gvn.transform( slow_result ));
5166 r->init_req(1, iftrue);
5167
5168 // Else fall through
5169 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5170 set_control(iffalse);
5171
5172 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5173 r->init_req(2, iffalse);
5174
5175 // Post merge
5176 set_control(_gvn.transform(r));
5177 record_for_igvn(r);
5178
5179 C->set_has_split_ifs(true); // Has chance for split-if optimization
5180 result = phi;
5181 assert(result->bottom_type()->isa_int(), "must be");
5182 break;
5183 }
5184
5185 default:
5186 fatal_unexpected_iid(id);
5187 break;
5188 }
5189 set_result(_gvn.transform(result));
5190 return true;
5191 }
5192
5193 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5194 Node* arg = argument(0);
5195 Node* result = nullptr;
5196
5197 switch (id) {
5198 case vmIntrinsics::_floatIsInfinite:
5199 result = new IsInfiniteFNode(arg);
5200 break;
5201 case vmIntrinsics::_floatIsFinite:
5202 result = new IsFiniteFNode(arg);
5203 break;
5204 case vmIntrinsics::_doubleIsInfinite:
5205 result = new IsInfiniteDNode(arg);
5206 break;
5207 case vmIntrinsics::_doubleIsFinite:
5208 result = new IsFiniteDNode(arg);
5209 break;
5210 default:
5211 fatal_unexpected_iid(id);
5212 break;
5213 }
5214 set_result(_gvn.transform(result));
5215 return true;
5216 }
5217
5218 //----------------------inline_unsafe_copyMemory-------------------------
5219 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5220
5221 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5222 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5223 const Type* base_t = gvn.type(base);
5224
5225 bool in_native = (base_t == TypePtr::NULL_PTR);
5226 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5227 bool is_mixed = !in_heap && !in_native;
5228
5229 if (is_mixed) {
5230 return true; // mixed accesses can touch both on-heap and off-heap memory
5231 }
5232 if (in_heap) {
5233 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5234 if (!is_prim_array) {
5235 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5236 // there's not enough type information available to determine proper memory slice for it.
5237 return true;
5238 }
5239 }
5240 return false;
5241 }
5242
5243 bool LibraryCallKit::inline_unsafe_copyMemory() {
5244 if (callee()->is_static()) return false; // caller must have the capability!
5245 null_check_receiver(); // null-check receiver
5246 if (stopped()) return true;
5247
5248 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5249
5250 Node* src_base = argument(1); // type: oop
5251 Node* src_off = ConvL2X(argument(2)); // type: long
5252 Node* dst_base = argument(4); // type: oop
5253 Node* dst_off = ConvL2X(argument(5)); // type: long
5254 Node* size = ConvL2X(argument(7)); // type: long
5255
5256 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5257 "fieldOffset must be byte-scaled");
5258
5259 Node* src_addr = make_unsafe_address(src_base, src_off);
5260 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5261
5262 Node* thread = _gvn.transform(new ThreadLocalNode());
5263 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5264 BasicType doing_unsafe_access_bt = T_BYTE;
5265 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5266
5267 // update volatile field
5268 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5269
5270 int flags = RC_LEAF | RC_NO_FP;
5271
5272 const TypePtr* dst_type = TypePtr::BOTTOM;
5273
5274 // Adjust memory effects of the runtime call based on input values.
5275 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5276 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5277 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5278
5279 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5280 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5281 flags |= RC_NARROW_MEM; // narrow in memory
5282 }
5283 }
5284
5285 // Call it. Note that the length argument is not scaled.
5286 make_runtime_call(flags,
5287 OptoRuntime::fast_arraycopy_Type(),
5288 StubRoutines::unsafe_arraycopy(),
5289 "unsafe_arraycopy",
5290 dst_type,
5291 src_addr, dst_addr, size XTOP);
5292
5293 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5294
5295 return true;
5296 }
5297
5298 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5299 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5300 bool LibraryCallKit::inline_unsafe_setMemory() {
5301 if (callee()->is_static()) return false; // caller must have the capability!
5302 null_check_receiver(); // null-check receiver
5303 if (stopped()) return true;
5304
5305 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5306
5307 Node* dst_base = argument(1); // type: oop
5308 Node* dst_off = ConvL2X(argument(2)); // type: long
5309 Node* size = ConvL2X(argument(4)); // type: long
5310 Node* byte = argument(6); // type: byte
5311
5312 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5313 "fieldOffset must be byte-scaled");
5314
5315 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5316
5317 Node* thread = _gvn.transform(new ThreadLocalNode());
5318 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5319 BasicType doing_unsafe_access_bt = T_BYTE;
5320 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5321
5322 // update volatile field
5323 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5324
5325 int flags = RC_LEAF | RC_NO_FP;
5326
5327 const TypePtr* dst_type = TypePtr::BOTTOM;
5328
5329 // Adjust memory effects of the runtime call based on input values.
5330 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5331 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5332
5333 flags |= RC_NARROW_MEM; // narrow in memory
5334 }
5335
5336 // Call it. Note that the length argument is not scaled.
5337 make_runtime_call(flags,
5338 OptoRuntime::make_setmemory_Type(),
5339 StubRoutines::unsafe_setmemory(),
5340 "unsafe_setmemory",
5341 dst_type,
5342 dst_addr, size XTOP, byte);
5343
5344 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5345
5346 return true;
5347 }
5348
5349 #undef XTOP
5350
5351 //----------------------inline_unsafe_isFlatArray------------------------
5352 // public native boolean Unsafe.isFlatArray(Class<?> arrayClass);
5353 // This intrinsic exploits assumptions made by the native implementation
5354 // (arrayClass is neither null nor primitive) to avoid unnecessary null checks.
5355 bool LibraryCallKit::inline_unsafe_isFlatArray() {
5356 Node* cls = argument(1);
5357 Node* p = basic_plus_adr(cls, java_lang_Class::klass_offset());
5358 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p,
5359 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT));
5360 Node* result = flat_array_test(kls);
5361 set_result(result);
5362 return true;
5363 }
5364
5365 //------------------------clone_coping-----------------------------------
5366 // Helper function for inline_native_clone.
5367 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5368 assert(obj_size != nullptr, "");
5369 Node* raw_obj = alloc_obj->in(1);
5370 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5371
5372 AllocateNode* alloc = nullptr;
5373 if (ReduceBulkZeroing &&
5374 // If we are implementing an array clone without knowing its source type
5375 // (can happen when compiling the array-guarded branch of a reflective
5376 // Object.clone() invocation), initialize the array within the allocation.
5377 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5378 // to a runtime clone call that assumes fully initialized source arrays.
5379 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5380 // We will be completely responsible for initializing this object -
5381 // mark Initialize node as complete.
5382 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5383 // The object was just allocated - there should be no any stores!
5384 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5385 // Mark as complete_with_arraycopy so that on AllocateNode
5386 // expansion, we know this AllocateNode is initialized by an array
5387 // copy and a StoreStore barrier exists after the array copy.
5388 alloc->initialization()->set_complete_with_arraycopy();
5389 }
5390
5391 Node* size = _gvn.transform(obj_size);
5392 access_clone(obj, alloc_obj, size, is_array);
5393
5394 // Do not let reads from the cloned object float above the arraycopy.
5395 if (alloc != nullptr) {
5396 // Do not let stores that initialize this object be reordered with
5397 // a subsequent store that would make this object accessible by
5398 // other threads.
5399 // Record what AllocateNode this StoreStore protects so that
5400 // escape analysis can go from the MemBarStoreStoreNode to the
5401 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5402 // based on the escape status of the AllocateNode.
5403 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5404 } else {
5405 insert_mem_bar(Op_MemBarCPUOrder);
5406 }
5407 }
5408
5409 //------------------------inline_native_clone----------------------------
5410 // protected native Object java.lang.Object.clone();
5411 //
5412 // Here are the simple edge cases:
5413 // null receiver => normal trap
5414 // virtual and clone was overridden => slow path to out-of-line clone
5415 // not cloneable or finalizer => slow path to out-of-line Object.clone
5416 //
5417 // The general case has two steps, allocation and copying.
5418 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5419 //
5420 // Copying also has two cases, oop arrays and everything else.
5421 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5422 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5423 //
5424 // These steps fold up nicely if and when the cloned object's klass
5425 // can be sharply typed as an object array, a type array, or an instance.
5426 //
5427 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5428 PhiNode* result_val;
5429
5430 // Set the reexecute bit for the interpreter to reexecute
5431 // the bytecode that invokes Object.clone if deoptimization happens.
5432 { PreserveReexecuteState preexecs(this);
5433 jvms()->set_should_reexecute(true);
5434
5435 Node* obj = argument(0);
5436 obj = null_check_receiver();
5437 if (stopped()) return true;
5438
5439 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5440 if (obj_type->is_inlinetypeptr()) {
5441 // If the object to clone is an inline type, we can simply return it (i.e. a nop) since inline types have
5442 // no identity.
5443 set_result(obj);
5444 return true;
5445 }
5446
5447 // If we are going to clone an instance, we need its exact type to
5448 // know the number and types of fields to convert the clone to
5449 // loads/stores. Maybe a speculative type can help us.
5450 if (!obj_type->klass_is_exact() &&
5451 obj_type->speculative_type() != nullptr &&
5452 obj_type->speculative_type()->is_instance_klass() &&
5453 !obj_type->speculative_type()->is_inlinetype()) {
5454 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5455 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5456 !spec_ik->has_injected_fields()) {
5457 if (!obj_type->isa_instptr() ||
5458 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5459 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5460 }
5461 }
5462 }
5463
5464 // Conservatively insert a memory barrier on all memory slices.
5465 // Do not let writes into the original float below the clone.
5466 insert_mem_bar(Op_MemBarCPUOrder);
5467
5468 // paths into result_reg:
5469 enum {
5470 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5471 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5472 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5473 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5474 PATH_LIMIT
5475 };
5476 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5477 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5478 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5479 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5480 record_for_igvn(result_reg);
5481
5482 Node* obj_klass = load_object_klass(obj);
5483 // We only go to the fast case code if we pass a number of guards.
5484 // The paths which do not pass are accumulated in the slow_region.
5485 RegionNode* slow_region = new RegionNode(1);
5486 record_for_igvn(slow_region);
5487
5488 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr);
5489 if (array_ctl != nullptr) {
5490 // It's an array.
5491 PreserveJVMState pjvms(this);
5492 set_control(array_ctl);
5493
5494 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5495 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr();
5496 if (UseFlatArray && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) &&
5497 obj_type->can_be_inline_array() &&
5498 (ary_ptr == nullptr || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) {
5499 // Flat inline type array may have object field that would require a
5500 // write barrier. Conservatively, go to slow path.
5501 generate_fair_guard(flat_array_test(obj_klass), slow_region);
5502 }
5503
5504 if (!stopped()) {
5505 Node* obj_length = load_array_length(obj);
5506 Node* array_size = nullptr; // Size of the array without object alignment padding.
5507 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5508
5509 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5510 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5511 // If it is an oop array, it requires very special treatment,
5512 // because gc barriers are required when accessing the array.
5513 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5514 if (is_obja != nullptr) {
5515 PreserveJVMState pjvms2(this);
5516 set_control(is_obja);
5517 // Generate a direct call to the right arraycopy function(s).
5518 // Clones are always tightly coupled.
5519 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5520 ac->set_clone_oop_array();
5521 Node* n = _gvn.transform(ac);
5522 assert(n == ac, "cannot disappear");
5523 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5524
5525 result_reg->init_req(_objArray_path, control());
5526 result_val->init_req(_objArray_path, alloc_obj);
5527 result_i_o ->set_req(_objArray_path, i_o());
5528 result_mem ->set_req(_objArray_path, reset_memory());
5529 }
5530 }
5531 // Otherwise, there are no barriers to worry about.
5532 // (We can dispense with card marks if we know the allocation
5533 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5534 // causes the non-eden paths to take compensating steps to
5535 // simulate a fresh allocation, so that no further
5536 // card marks are required in compiled code to initialize
5537 // the object.)
5538
5539 if (!stopped()) {
5540 copy_to_clone(obj, alloc_obj, array_size, true);
5541
5542 // Present the results of the copy.
5543 result_reg->init_req(_array_path, control());
5544 result_val->init_req(_array_path, alloc_obj);
5545 result_i_o ->set_req(_array_path, i_o());
5546 result_mem ->set_req(_array_path, reset_memory());
5547 }
5548 }
5549 }
5550
5551 if (!stopped()) {
5552 // It's an instance (we did array above). Make the slow-path tests.
5553 // If this is a virtual call, we generate a funny guard. We grab
5554 // the vtable entry corresponding to clone() from the target object.
5555 // If the target method which we are calling happens to be the
5556 // Object clone() method, we pass the guard. We do not need this
5557 // guard for non-virtual calls; the caller is known to be the native
5558 // Object clone().
5559 if (is_virtual) {
5560 generate_virtual_guard(obj_klass, slow_region);
5561 }
5562
5563 // The object must be easily cloneable and must not have a finalizer.
5564 // Both of these conditions may be checked in a single test.
5565 // We could optimize the test further, but we don't care.
5566 generate_access_flags_guard(obj_klass,
5567 // Test both conditions:
5568 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
5569 // Must be cloneable but not finalizer:
5570 JVM_ACC_IS_CLONEABLE_FAST,
5571 slow_region);
5572 }
5573
5574 if (!stopped()) {
5575 // It's an instance, and it passed the slow-path tests.
5576 PreserveJVMState pjvms(this);
5577 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5578 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5579 // is reexecuted if deoptimization occurs and there could be problems when merging
5580 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5581 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5582
5583 copy_to_clone(obj, alloc_obj, obj_size, false);
5584
5585 // Present the results of the slow call.
5586 result_reg->init_req(_instance_path, control());
5587 result_val->init_req(_instance_path, alloc_obj);
5588 result_i_o ->set_req(_instance_path, i_o());
5589 result_mem ->set_req(_instance_path, reset_memory());
5590 }
5591
5592 // Generate code for the slow case. We make a call to clone().
5593 set_control(_gvn.transform(slow_region));
5594 if (!stopped()) {
5595 PreserveJVMState pjvms(this);
5596 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5597 // We need to deoptimize on exception (see comment above)
5598 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5599 // this->control() comes from set_results_for_java_call
5600 result_reg->init_req(_slow_path, control());
5601 result_val->init_req(_slow_path, slow_result);
5602 result_i_o ->set_req(_slow_path, i_o());
5603 result_mem ->set_req(_slow_path, reset_memory());
5604 }
5605
5606 // Return the combined state.
5607 set_control( _gvn.transform(result_reg));
5608 set_i_o( _gvn.transform(result_i_o));
5609 set_all_memory( _gvn.transform(result_mem));
5610 } // original reexecute is set back here
5611
5612 set_result(_gvn.transform(result_val));
5613 return true;
5614 }
5615
5616 // If we have a tightly coupled allocation, the arraycopy may take care
5617 // of the array initialization. If one of the guards we insert between
5618 // the allocation and the arraycopy causes a deoptimization, an
5619 // uninitialized array will escape the compiled method. To prevent that
5620 // we set the JVM state for uncommon traps between the allocation and
5621 // the arraycopy to the state before the allocation so, in case of
5622 // deoptimization, we'll reexecute the allocation and the
5623 // initialization.
5624 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5625 if (alloc != nullptr) {
5626 ciMethod* trap_method = alloc->jvms()->method();
5627 int trap_bci = alloc->jvms()->bci();
5628
5629 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5630 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5631 // Make sure there's no store between the allocation and the
5632 // arraycopy otherwise visible side effects could be rexecuted
5633 // in case of deoptimization and cause incorrect execution.
5634 bool no_interfering_store = true;
5635 Node* mem = alloc->in(TypeFunc::Memory);
5636 if (mem->is_MergeMem()) {
5637 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5638 Node* n = mms.memory();
5639 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5640 assert(n->is_Store(), "what else?");
5641 no_interfering_store = false;
5642 break;
5643 }
5644 }
5645 } else {
5646 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5647 Node* n = mms.memory();
5648 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5649 assert(n->is_Store(), "what else?");
5650 no_interfering_store = false;
5651 break;
5652 }
5653 }
5654 }
5655
5656 if (no_interfering_store) {
5657 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5658
5659 JVMState* saved_jvms = jvms();
5660 saved_reexecute_sp = _reexecute_sp;
5661
5662 set_jvms(sfpt->jvms());
5663 _reexecute_sp = jvms()->sp();
5664
5665 return saved_jvms;
5666 }
5667 }
5668 }
5669 return nullptr;
5670 }
5671
5672 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5673 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5674 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5675 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5676 uint size = alloc->req();
5677 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5678 old_jvms->set_map(sfpt);
5679 for (uint i = 0; i < size; i++) {
5680 sfpt->init_req(i, alloc->in(i));
5681 }
5682 int adjustment = 1;
5683 const TypeAryKlassPtr* ary_klass_ptr = alloc->in(AllocateNode::KlassNode)->bottom_type()->is_aryklassptr();
5684 if (ary_klass_ptr->is_null_free()) {
5685 // A null-free, tightly coupled array allocation can only come from LibraryCallKit::inline_newNullRestrictedArray
5686 // which requires both the component type and the array length on stack for re-execution. Re-create and push
5687 // the component type.
5688 ciArrayKlass* klass = ary_klass_ptr->exact_klass()->as_array_klass();
5689 ciInstance* instance = klass->component_mirror_instance();
5690 const TypeInstPtr* t_instance = TypeInstPtr::make(instance);
5691 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), makecon(t_instance));
5692 adjustment++;
5693 }
5694 // re-push array length for deoptimization
5695 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment - 1, alloc->in(AllocateNode::ALength));
5696 old_jvms->set_sp(old_jvms->sp() + adjustment);
5697 old_jvms->set_monoff(old_jvms->monoff() + adjustment);
5698 old_jvms->set_scloff(old_jvms->scloff() + adjustment);
5699 old_jvms->set_endoff(old_jvms->endoff() + adjustment);
5700 old_jvms->set_should_reexecute(true);
5701
5702 sfpt->set_i_o(map()->i_o());
5703 sfpt->set_memory(map()->memory());
5704 sfpt->set_control(map()->control());
5705 return sfpt;
5706 }
5707
5708 // In case of a deoptimization, we restart execution at the
5709 // allocation, allocating a new array. We would leave an uninitialized
5710 // array in the heap that GCs wouldn't expect. Move the allocation
5711 // after the traps so we don't allocate the array if we
5712 // deoptimize. This is possible because tightly_coupled_allocation()
5713 // guarantees there's no observer of the allocated array at this point
5714 // and the control flow is simple enough.
5715 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5716 int saved_reexecute_sp, uint new_idx) {
5717 if (saved_jvms_before_guards != nullptr && !stopped()) {
5718 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5719
5720 assert(alloc != nullptr, "only with a tightly coupled allocation");
5721 // restore JVM state to the state at the arraycopy
5722 saved_jvms_before_guards->map()->set_control(map()->control());
5723 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5724 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5725 // If we've improved the types of some nodes (null check) while
5726 // emitting the guards, propagate them to the current state
5727 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5728 set_jvms(saved_jvms_before_guards);
5729 _reexecute_sp = saved_reexecute_sp;
5730
5731 // Remove the allocation from above the guards
5732 CallProjections* callprojs = alloc->extract_projections(true);
5733 InitializeNode* init = alloc->initialization();
5734 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5735 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5736 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5737
5738 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5739 // the allocation (i.e. is only valid if the allocation succeeds):
5740 // 1) replace CastIINode with AllocateArrayNode's length here
5741 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5742 //
5743 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5744 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5745 Node* init_control = init->proj_out(TypeFunc::Control);
5746 Node* alloc_length = alloc->Ideal_length();
5747 #ifdef ASSERT
5748 Node* prev_cast = nullptr;
5749 #endif
5750 for (uint i = 0; i < init_control->outcnt(); i++) {
5751 Node* init_out = init_control->raw_out(i);
5752 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5753 #ifdef ASSERT
5754 if (prev_cast == nullptr) {
5755 prev_cast = init_out;
5756 } else {
5757 if (prev_cast->cmp(*init_out) == false) {
5758 prev_cast->dump();
5759 init_out->dump();
5760 assert(false, "not equal CastIINode");
5761 }
5762 }
5763 #endif
5764 C->gvn_replace_by(init_out, alloc_length);
5765 }
5766 }
5767 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5768
5769 // move the allocation here (after the guards)
5770 _gvn.hash_delete(alloc);
5771 alloc->set_req(TypeFunc::Control, control());
5772 alloc->set_req(TypeFunc::I_O, i_o());
5773 Node *mem = reset_memory();
5774 set_all_memory(mem);
5775 alloc->set_req(TypeFunc::Memory, mem);
5776 set_control(init->proj_out_or_null(TypeFunc::Control));
5777 set_i_o(callprojs->fallthrough_ioproj);
5778
5779 // Update memory as done in GraphKit::set_output_for_allocation()
5780 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5781 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5782 if (ary_type->isa_aryptr() && length_type != nullptr) {
5783 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5784 }
5785 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5786 int elemidx = C->get_alias_index(telemref);
5787 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5788 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5789
5790 Node* allocx = _gvn.transform(alloc);
5791 assert(allocx == alloc, "where has the allocation gone?");
5792 assert(dest->is_CheckCastPP(), "not an allocation result?");
5793
5794 _gvn.hash_delete(dest);
5795 dest->set_req(0, control());
5796 Node* destx = _gvn.transform(dest);
5797 assert(destx == dest, "where has the allocation result gone?");
5798
5799 array_ideal_length(alloc, ary_type, true);
5800 }
5801 }
5802
5803 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5804 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5805 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5806 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5807 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5808 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5809 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5810 JVMState* saved_jvms_before_guards) {
5811 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5812 // There is at least one unrelated uncommon trap which needs to be replaced.
5813 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5814
5815 JVMState* saved_jvms = jvms();
5816 const int saved_reexecute_sp = _reexecute_sp;
5817 set_jvms(sfpt->jvms());
5818 _reexecute_sp = jvms()->sp();
5819
5820 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5821
5822 // Restore state
5823 set_jvms(saved_jvms);
5824 _reexecute_sp = saved_reexecute_sp;
5825 }
5826 }
5827
5828 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5829 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5830 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5831 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5832 while (if_proj->is_IfProj()) {
5833 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5834 if (uncommon_trap != nullptr) {
5835 create_new_uncommon_trap(uncommon_trap);
5836 }
5837 assert(if_proj->in(0)->is_If(), "must be If");
5838 if_proj = if_proj->in(0)->in(0);
5839 }
5840 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5841 "must have reached control projection of init node");
5842 }
5843
5844 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5845 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5846 assert(trap_request != 0, "no valid UCT trap request");
5847 PreserveJVMState pjvms(this);
5848 set_control(uncommon_trap_call->in(0));
5849 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5850 Deoptimization::trap_request_action(trap_request));
5851 assert(stopped(), "Should be stopped");
5852 _gvn.hash_delete(uncommon_trap_call);
5853 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5854 }
5855
5856 // Common checks for array sorting intrinsics arguments.
5857 // Returns `true` if checks passed.
5858 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
5859 // check address of the class
5860 if (elementType == nullptr || elementType->is_top()) {
5861 return false; // dead path
5862 }
5863 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5864 if (elem_klass == nullptr) {
5865 return false; // dead path
5866 }
5867 // java_mirror_type() returns non-null for compile-time Class constants only
5868 ciType* elem_type = elem_klass->java_mirror_type();
5869 if (elem_type == nullptr) {
5870 return false;
5871 }
5872 bt = elem_type->basic_type();
5873 // Disable the intrinsic if the CPU does not support SIMD sort
5874 if (!Matcher::supports_simd_sort(bt)) {
5875 return false;
5876 }
5877 // check address of the array
5878 if (obj == nullptr || obj->is_top()) {
5879 return false; // dead path
5880 }
5881 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5882 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
5883 return false; // failed input validation
5884 }
5885 return true;
5886 }
5887
5888 //------------------------------inline_array_partition-----------------------
5889 bool LibraryCallKit::inline_array_partition() {
5890 address stubAddr = StubRoutines::select_array_partition_function();
5891 if (stubAddr == nullptr) {
5892 return false; // Intrinsic's stub is not implemented on this platform
5893 }
5894 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
5895
5896 // no receiver because it is a static method
5897 Node* elementType = argument(0);
5898 Node* obj = argument(1);
5899 Node* offset = argument(2); // long
5900 Node* fromIndex = argument(4);
5901 Node* toIndex = argument(5);
5902 Node* indexPivot1 = argument(6);
5903 Node* indexPivot2 = argument(7);
5904 // PartitionOperation: argument(8) is ignored
5905
5906 Node* pivotIndices = nullptr;
5907 BasicType bt = T_ILLEGAL;
5908
5909 if (!check_array_sort_arguments(elementType, obj, bt)) {
5910 return false;
5911 }
5912 null_check(obj);
5913 // If obj is dead, only null-path is taken.
5914 if (stopped()) {
5915 return true;
5916 }
5917 // Set the original stack and the reexecute bit for the interpreter to reexecute
5918 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
5919 { PreserveReexecuteState preexecs(this);
5920 jvms()->set_should_reexecute(true);
5921
5922 Node* obj_adr = make_unsafe_address(obj, offset);
5923
5924 // create the pivotIndices array of type int and size = 2
5925 Node* size = intcon(2);
5926 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
5927 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
5928 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
5929 guarantee(alloc != nullptr, "created above");
5930 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
5931
5932 // pass the basic type enum to the stub
5933 Node* elemType = intcon(bt);
5934
5935 // Call the stub
5936 const char *stubName = "array_partition_stub";
5937 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
5938 stubAddr, stubName, TypePtr::BOTTOM,
5939 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
5940 indexPivot1, indexPivot2);
5941
5942 } // original reexecute is set back here
5943
5944 if (!stopped()) {
5945 set_result(pivotIndices);
5946 }
5947
5948 return true;
5949 }
5950
5951
5952 //------------------------------inline_array_sort-----------------------
5953 bool LibraryCallKit::inline_array_sort() {
5954 address stubAddr = StubRoutines::select_arraysort_function();
5955 if (stubAddr == nullptr) {
5956 return false; // Intrinsic's stub is not implemented on this platform
5957 }
5958 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
5959
5960 // no receiver because it is a static method
5961 Node* elementType = argument(0);
5962 Node* obj = argument(1);
5963 Node* offset = argument(2); // long
5964 Node* fromIndex = argument(4);
5965 Node* toIndex = argument(5);
5966 // SortOperation: argument(6) is ignored
5967
5968 BasicType bt = T_ILLEGAL;
5969
5970 if (!check_array_sort_arguments(elementType, obj, bt)) {
5971 return false;
5972 }
5973 null_check(obj);
5974 // If obj is dead, only null-path is taken.
5975 if (stopped()) {
5976 return true;
5977 }
5978 Node* obj_adr = make_unsafe_address(obj, offset);
5979
5980 // pass the basic type enum to the stub
5981 Node* elemType = intcon(bt);
5982
5983 // Call the stub.
5984 const char *stubName = "arraysort_stub";
5985 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
5986 stubAddr, stubName, TypePtr::BOTTOM,
5987 obj_adr, elemType, fromIndex, toIndex);
5988
5989 return true;
5990 }
5991
5992
5993 //------------------------------inline_arraycopy-----------------------
5994 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5995 // Object dest, int destPos,
5996 // int length);
5997 bool LibraryCallKit::inline_arraycopy() {
5998 // Get the arguments.
5999 Node* src = argument(0); // type: oop
6000 Node* src_offset = argument(1); // type: int
6001 Node* dest = argument(2); // type: oop
6002 Node* dest_offset = argument(3); // type: int
6003 Node* length = argument(4); // type: int
6004
6005 uint new_idx = C->unique();
6006
6007 // Check for allocation before we add nodes that would confuse
6008 // tightly_coupled_allocation()
6009 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
6010
6011 int saved_reexecute_sp = -1;
6012 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
6013 // See arraycopy_restore_alloc_state() comment
6014 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
6015 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
6016 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
6017 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
6018
6019 // The following tests must be performed
6020 // (1) src and dest are arrays.
6021 // (2) src and dest arrays must have elements of the same BasicType
6022 // (3) src and dest must not be null.
6023 // (4) src_offset must not be negative.
6024 // (5) dest_offset must not be negative.
6025 // (6) length must not be negative.
6026 // (7) src_offset + length must not exceed length of src.
6027 // (8) dest_offset + length must not exceed length of dest.
6028 // (9) each element of an oop array must be assignable
6029
6030 // (3) src and dest must not be null.
6031 // always do this here because we need the JVM state for uncommon traps
6032 Node* null_ctl = top();
6033 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
6034 assert(null_ctl->is_top(), "no null control here");
6035 dest = null_check(dest, T_ARRAY);
6036
6037 if (!can_emit_guards) {
6038 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
6039 // guards but the arraycopy node could still take advantage of a
6040 // tightly allocated allocation. tightly_coupled_allocation() is
6041 // called again to make sure it takes the null check above into
6042 // account: the null check is mandatory and if it caused an
6043 // uncommon trap to be emitted then the allocation can't be
6044 // considered tightly coupled in this context.
6045 alloc = tightly_coupled_allocation(dest);
6046 }
6047
6048 bool validated = false;
6049
6050 const Type* src_type = _gvn.type(src);
6051 const Type* dest_type = _gvn.type(dest);
6052 const TypeAryPtr* top_src = src_type->isa_aryptr();
6053 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6054
6055 // Do we have the type of src?
6056 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6057 // Do we have the type of dest?
6058 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6059 // Is the type for src from speculation?
6060 bool src_spec = false;
6061 // Is the type for dest from speculation?
6062 bool dest_spec = false;
6063
6064 if ((!has_src || !has_dest) && can_emit_guards) {
6065 // We don't have sufficient type information, let's see if
6066 // speculative types can help. We need to have types for both src
6067 // and dest so that it pays off.
6068
6069 // Do we already have or could we have type information for src
6070 bool could_have_src = has_src;
6071 // Do we already have or could we have type information for dest
6072 bool could_have_dest = has_dest;
6073
6074 ciKlass* src_k = nullptr;
6075 if (!has_src) {
6076 src_k = src_type->speculative_type_not_null();
6077 if (src_k != nullptr && src_k->is_array_klass()) {
6078 could_have_src = true;
6079 }
6080 }
6081
6082 ciKlass* dest_k = nullptr;
6083 if (!has_dest) {
6084 dest_k = dest_type->speculative_type_not_null();
6085 if (dest_k != nullptr && dest_k->is_array_klass()) {
6086 could_have_dest = true;
6087 }
6088 }
6089
6090 if (could_have_src && could_have_dest) {
6091 // This is going to pay off so emit the required guards
6092 if (!has_src) {
6093 src = maybe_cast_profiled_obj(src, src_k, true);
6094 src_type = _gvn.type(src);
6095 top_src = src_type->isa_aryptr();
6096 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6097 src_spec = true;
6098 }
6099 if (!has_dest) {
6100 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6101 dest_type = _gvn.type(dest);
6102 top_dest = dest_type->isa_aryptr();
6103 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6104 dest_spec = true;
6105 }
6106 }
6107 }
6108
6109 if (has_src && has_dest && can_emit_guards) {
6110 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
6111 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
6112 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
6113 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
6114
6115 if (src_elem == dest_elem && top_src->is_flat() == top_dest->is_flat() && src_elem == T_OBJECT) {
6116 // If both arrays are object arrays then having the exact types
6117 // for both will remove the need for a subtype check at runtime
6118 // before the call and may make it possible to pick a faster copy
6119 // routine (without a subtype check on every element)
6120 // Do we have the exact type of src?
6121 bool could_have_src = src_spec;
6122 // Do we have the exact type of dest?
6123 bool could_have_dest = dest_spec;
6124 ciKlass* src_k = nullptr;
6125 ciKlass* dest_k = nullptr;
6126 if (!src_spec) {
6127 src_k = src_type->speculative_type_not_null();
6128 if (src_k != nullptr && src_k->is_array_klass()) {
6129 could_have_src = true;
6130 }
6131 }
6132 if (!dest_spec) {
6133 dest_k = dest_type->speculative_type_not_null();
6134 if (dest_k != nullptr && dest_k->is_array_klass()) {
6135 could_have_dest = true;
6136 }
6137 }
6138 if (could_have_src && could_have_dest) {
6139 // If we can have both exact types, emit the missing guards
6140 if (could_have_src && !src_spec) {
6141 src = maybe_cast_profiled_obj(src, src_k, true);
6142 src_type = _gvn.type(src);
6143 top_src = src_type->isa_aryptr();
6144 }
6145 if (could_have_dest && !dest_spec) {
6146 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6147 dest_type = _gvn.type(dest);
6148 top_dest = dest_type->isa_aryptr();
6149 }
6150 }
6151 }
6152 }
6153
6154 ciMethod* trap_method = method();
6155 int trap_bci = bci();
6156 if (saved_jvms_before_guards != nullptr) {
6157 trap_method = alloc->jvms()->method();
6158 trap_bci = alloc->jvms()->bci();
6159 }
6160
6161 bool negative_length_guard_generated = false;
6162
6163 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6164 can_emit_guards && !src->is_top() && !dest->is_top()) {
6165 // validate arguments: enables transformation the ArrayCopyNode
6166 validated = true;
6167
6168 RegionNode* slow_region = new RegionNode(1);
6169 record_for_igvn(slow_region);
6170
6171 // (1) src and dest are arrays.
6172 generate_non_array_guard(load_object_klass(src), slow_region);
6173 generate_non_array_guard(load_object_klass(dest), slow_region);
6174
6175 // (2) src and dest arrays must have elements of the same BasicType
6176 // done at macro expansion or at Ideal transformation time
6177
6178 // (4) src_offset must not be negative.
6179 generate_negative_guard(src_offset, slow_region);
6180
6181 // (5) dest_offset must not be negative.
6182 generate_negative_guard(dest_offset, slow_region);
6183
6184 // (7) src_offset + length must not exceed length of src.
6185 generate_limit_guard(src_offset, length,
6186 load_array_length(src),
6187 slow_region);
6188
6189 // (8) dest_offset + length must not exceed length of dest.
6190 generate_limit_guard(dest_offset, length,
6191 load_array_length(dest),
6192 slow_region);
6193
6194 // (6) length must not be negative.
6195 // This is also checked in generate_arraycopy() during macro expansion, but
6196 // we also have to check it here for the case where the ArrayCopyNode will
6197 // be eliminated by Escape Analysis.
6198 if (EliminateAllocations) {
6199 generate_negative_guard(length, slow_region);
6200 negative_length_guard_generated = true;
6201 }
6202
6203 // (9) each element of an oop array must be assignable
6204 Node* dest_klass = load_object_klass(dest);
6205 if (src != dest) {
6206 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6207 slow_region->add_req(not_subtype_ctrl);
6208 }
6209
6210 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
6211 const Type* toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6212 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6213 src_type = _gvn.type(src);
6214 top_src = src_type->isa_aryptr();
6215
6216 // Handle flat inline type arrays (null-free arrays are handled by the subtype check above)
6217 if (!stopped() && UseFlatArray) {
6218 // If dest is flat, src must be flat as well (guaranteed by src <: dest check). Handle flat src here.
6219 assert(top_dest == nullptr || !top_dest->is_flat() || top_src->is_flat(), "src array must be flat");
6220 if (top_src != nullptr && top_src->is_flat()) {
6221 // Src is flat, check that dest is flat as well
6222 if (top_dest != nullptr && !top_dest->is_flat()) {
6223 generate_fair_guard(flat_array_test(dest_klass, /* flat = */ false), slow_region);
6224 // Since dest is flat and src <: dest, dest must have the same type as src.
6225 top_dest = top_src->cast_to_exactness(false);
6226 assert(top_dest->is_flat(), "dest must be flat");
6227 dest = _gvn.transform(new CheckCastPPNode(control(), dest, top_dest));
6228 }
6229 } else if (top_src == nullptr || !top_src->is_not_flat()) {
6230 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
6231 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
6232 assert(top_dest == nullptr || !top_dest->is_flat(), "dest array must not be flat");
6233 generate_fair_guard(flat_array_test(src), slow_region);
6234 if (top_src != nullptr) {
6235 top_src = top_src->cast_to_not_flat();
6236 src = _gvn.transform(new CheckCastPPNode(control(), src, top_src));
6237 }
6238 }
6239 }
6240
6241 {
6242 PreserveJVMState pjvms(this);
6243 set_control(_gvn.transform(slow_region));
6244 uncommon_trap(Deoptimization::Reason_intrinsic,
6245 Deoptimization::Action_make_not_entrant);
6246 assert(stopped(), "Should be stopped");
6247 }
6248 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6249 }
6250
6251 if (stopped()) {
6252 return true;
6253 }
6254
6255 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6256 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6257 // so the compiler has a chance to eliminate them: during macro expansion,
6258 // we have to set their control (CastPP nodes are eliminated).
6259 load_object_klass(src), load_object_klass(dest),
6260 load_array_length(src), load_array_length(dest));
6261
6262 ac->set_arraycopy(validated);
6263
6264 Node* n = _gvn.transform(ac);
6265 if (n == ac) {
6266 ac->connect_outputs(this);
6267 } else {
6268 assert(validated, "shouldn't transform if all arguments not validated");
6269 set_all_memory(n);
6270 }
6271 clear_upper_avx();
6272
6273
6274 return true;
6275 }
6276
6277
6278 // Helper function which determines if an arraycopy immediately follows
6279 // an allocation, with no intervening tests or other escapes for the object.
6280 AllocateArrayNode*
6281 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6282 if (stopped()) return nullptr; // no fast path
6283 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6284
6285 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6286 if (alloc == nullptr) return nullptr;
6287
6288 Node* rawmem = memory(Compile::AliasIdxRaw);
6289 // Is the allocation's memory state untouched?
6290 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6291 // Bail out if there have been raw-memory effects since the allocation.
6292 // (Example: There might have been a call or safepoint.)
6293 return nullptr;
6294 }
6295 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6296 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6297 return nullptr;
6298 }
6299
6300 // There must be no unexpected observers of this allocation.
6301 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6302 Node* obs = ptr->fast_out(i);
6303 if (obs != this->map()) {
6304 return nullptr;
6305 }
6306 }
6307
6308 // This arraycopy must unconditionally follow the allocation of the ptr.
6309 Node* alloc_ctl = ptr->in(0);
6310 Node* ctl = control();
6311 while (ctl != alloc_ctl) {
6312 // There may be guards which feed into the slow_region.
6313 // Any other control flow means that we might not get a chance
6314 // to finish initializing the allocated object.
6315 // Various low-level checks bottom out in uncommon traps. These
6316 // are considered safe since we've already checked above that
6317 // there is no unexpected observer of this allocation.
6318 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6319 assert(ctl->in(0)->is_If(), "must be If");
6320 ctl = ctl->in(0)->in(0);
6321 } else {
6322 return nullptr;
6323 }
6324 }
6325
6326 // If we get this far, we have an allocation which immediately
6327 // precedes the arraycopy, and we can take over zeroing the new object.
6328 // The arraycopy will finish the initialization, and provide
6329 // a new control state to which we will anchor the destination pointer.
6330
6331 return alloc;
6332 }
6333
6334 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6335 if (node->is_IfProj()) {
6336 Node* other_proj = node->as_IfProj()->other_if_proj();
6337 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6338 Node* obs = other_proj->fast_out(j);
6339 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6340 (obs->as_CallStaticJava()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
6341 return obs->as_CallStaticJava();
6342 }
6343 }
6344 }
6345 return nullptr;
6346 }
6347
6348 //-------------inline_encodeISOArray-----------------------------------
6349 // encode char[] to byte[] in ISO_8859_1 or ASCII
6350 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6351 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6352 // no receiver since it is static method
6353 Node *src = argument(0);
6354 Node *src_offset = argument(1);
6355 Node *dst = argument(2);
6356 Node *dst_offset = argument(3);
6357 Node *length = argument(4);
6358
6359 src = must_be_not_null(src, true);
6360 dst = must_be_not_null(dst, true);
6361
6362 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6363 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6364 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6365 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6366 // failed array check
6367 return false;
6368 }
6369
6370 // Figure out the size and type of the elements we will be copying.
6371 BasicType src_elem = src_type->elem()->array_element_basic_type();
6372 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6373 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6374 return false;
6375 }
6376
6377 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6378 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6379 // 'src_start' points to src array + scaled offset
6380 // 'dst_start' points to dst array + scaled offset
6381
6382 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6383 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6384 enc = _gvn.transform(enc);
6385 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6386 set_memory(res_mem, mtype);
6387 set_result(enc);
6388 clear_upper_avx();
6389
6390 return true;
6391 }
6392
6393 //-------------inline_multiplyToLen-----------------------------------
6394 bool LibraryCallKit::inline_multiplyToLen() {
6395 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6396
6397 address stubAddr = StubRoutines::multiplyToLen();
6398 if (stubAddr == nullptr) {
6399 return false; // Intrinsic's stub is not implemented on this platform
6400 }
6401 const char* stubName = "multiplyToLen";
6402
6403 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6404
6405 // no receiver because it is a static method
6406 Node* x = argument(0);
6407 Node* xlen = argument(1);
6408 Node* y = argument(2);
6409 Node* ylen = argument(3);
6410 Node* z = argument(4);
6411
6412 x = must_be_not_null(x, true);
6413 y = must_be_not_null(y, true);
6414
6415 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6416 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6417 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6418 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6419 // failed array check
6420 return false;
6421 }
6422
6423 BasicType x_elem = x_type->elem()->array_element_basic_type();
6424 BasicType y_elem = y_type->elem()->array_element_basic_type();
6425 if (x_elem != T_INT || y_elem != T_INT) {
6426 return false;
6427 }
6428
6429 Node* x_start = array_element_address(x, intcon(0), x_elem);
6430 Node* y_start = array_element_address(y, intcon(0), y_elem);
6431 // 'x_start' points to x array + scaled xlen
6432 // 'y_start' points to y array + scaled ylen
6433
6434 Node* z_start = array_element_address(z, intcon(0), T_INT);
6435
6436 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6437 OptoRuntime::multiplyToLen_Type(),
6438 stubAddr, stubName, TypePtr::BOTTOM,
6439 x_start, xlen, y_start, ylen, z_start);
6440
6441 C->set_has_split_ifs(true); // Has chance for split-if optimization
6442 set_result(z);
6443 return true;
6444 }
6445
6446 //-------------inline_squareToLen------------------------------------
6447 bool LibraryCallKit::inline_squareToLen() {
6448 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6449
6450 address stubAddr = StubRoutines::squareToLen();
6451 if (stubAddr == nullptr) {
6452 return false; // Intrinsic's stub is not implemented on this platform
6453 }
6454 const char* stubName = "squareToLen";
6455
6456 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6457
6458 Node* x = argument(0);
6459 Node* len = argument(1);
6460 Node* z = argument(2);
6461 Node* zlen = argument(3);
6462
6463 x = must_be_not_null(x, true);
6464 z = must_be_not_null(z, true);
6465
6466 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6467 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6468 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6469 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6470 // failed array check
6471 return false;
6472 }
6473
6474 BasicType x_elem = x_type->elem()->array_element_basic_type();
6475 BasicType z_elem = z_type->elem()->array_element_basic_type();
6476 if (x_elem != T_INT || z_elem != T_INT) {
6477 return false;
6478 }
6479
6480
6481 Node* x_start = array_element_address(x, intcon(0), x_elem);
6482 Node* z_start = array_element_address(z, intcon(0), z_elem);
6483
6484 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6485 OptoRuntime::squareToLen_Type(),
6486 stubAddr, stubName, TypePtr::BOTTOM,
6487 x_start, len, z_start, zlen);
6488
6489 set_result(z);
6490 return true;
6491 }
6492
6493 //-------------inline_mulAdd------------------------------------------
6494 bool LibraryCallKit::inline_mulAdd() {
6495 assert(UseMulAddIntrinsic, "not implemented on this platform");
6496
6497 address stubAddr = StubRoutines::mulAdd();
6498 if (stubAddr == nullptr) {
6499 return false; // Intrinsic's stub is not implemented on this platform
6500 }
6501 const char* stubName = "mulAdd";
6502
6503 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6504
6505 Node* out = argument(0);
6506 Node* in = argument(1);
6507 Node* offset = argument(2);
6508 Node* len = argument(3);
6509 Node* k = argument(4);
6510
6511 in = must_be_not_null(in, true);
6512 out = must_be_not_null(out, true);
6513
6514 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6515 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6516 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6517 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6518 // failed array check
6519 return false;
6520 }
6521
6522 BasicType out_elem = out_type->elem()->array_element_basic_type();
6523 BasicType in_elem = in_type->elem()->array_element_basic_type();
6524 if (out_elem != T_INT || in_elem != T_INT) {
6525 return false;
6526 }
6527
6528 Node* outlen = load_array_length(out);
6529 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6530 Node* out_start = array_element_address(out, intcon(0), out_elem);
6531 Node* in_start = array_element_address(in, intcon(0), in_elem);
6532
6533 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6534 OptoRuntime::mulAdd_Type(),
6535 stubAddr, stubName, TypePtr::BOTTOM,
6536 out_start,in_start, new_offset, len, k);
6537 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6538 set_result(result);
6539 return true;
6540 }
6541
6542 //-------------inline_montgomeryMultiply-----------------------------------
6543 bool LibraryCallKit::inline_montgomeryMultiply() {
6544 address stubAddr = StubRoutines::montgomeryMultiply();
6545 if (stubAddr == nullptr) {
6546 return false; // Intrinsic's stub is not implemented on this platform
6547 }
6548
6549 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6550 const char* stubName = "montgomery_multiply";
6551
6552 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6553
6554 Node* a = argument(0);
6555 Node* b = argument(1);
6556 Node* n = argument(2);
6557 Node* len = argument(3);
6558 Node* inv = argument(4);
6559 Node* m = argument(6);
6560
6561 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6562 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6563 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6564 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6565 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6566 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6567 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6568 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6569 // failed array check
6570 return false;
6571 }
6572
6573 BasicType a_elem = a_type->elem()->array_element_basic_type();
6574 BasicType b_elem = b_type->elem()->array_element_basic_type();
6575 BasicType n_elem = n_type->elem()->array_element_basic_type();
6576 BasicType m_elem = m_type->elem()->array_element_basic_type();
6577 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6578 return false;
6579 }
6580
6581 // Make the call
6582 {
6583 Node* a_start = array_element_address(a, intcon(0), a_elem);
6584 Node* b_start = array_element_address(b, intcon(0), b_elem);
6585 Node* n_start = array_element_address(n, intcon(0), n_elem);
6586 Node* m_start = array_element_address(m, intcon(0), m_elem);
6587
6588 Node* call = make_runtime_call(RC_LEAF,
6589 OptoRuntime::montgomeryMultiply_Type(),
6590 stubAddr, stubName, TypePtr::BOTTOM,
6591 a_start, b_start, n_start, len, inv, top(),
6592 m_start);
6593 set_result(m);
6594 }
6595
6596 return true;
6597 }
6598
6599 bool LibraryCallKit::inline_montgomerySquare() {
6600 address stubAddr = StubRoutines::montgomerySquare();
6601 if (stubAddr == nullptr) {
6602 return false; // Intrinsic's stub is not implemented on this platform
6603 }
6604
6605 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6606 const char* stubName = "montgomery_square";
6607
6608 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6609
6610 Node* a = argument(0);
6611 Node* n = argument(1);
6612 Node* len = argument(2);
6613 Node* inv = argument(3);
6614 Node* m = argument(5);
6615
6616 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6617 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6618 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6619 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6620 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6621 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6622 // failed array check
6623 return false;
6624 }
6625
6626 BasicType a_elem = a_type->elem()->array_element_basic_type();
6627 BasicType n_elem = n_type->elem()->array_element_basic_type();
6628 BasicType m_elem = m_type->elem()->array_element_basic_type();
6629 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6630 return false;
6631 }
6632
6633 // Make the call
6634 {
6635 Node* a_start = array_element_address(a, intcon(0), a_elem);
6636 Node* n_start = array_element_address(n, intcon(0), n_elem);
6637 Node* m_start = array_element_address(m, intcon(0), m_elem);
6638
6639 Node* call = make_runtime_call(RC_LEAF,
6640 OptoRuntime::montgomerySquare_Type(),
6641 stubAddr, stubName, TypePtr::BOTTOM,
6642 a_start, n_start, len, inv, top(),
6643 m_start);
6644 set_result(m);
6645 }
6646
6647 return true;
6648 }
6649
6650 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6651 address stubAddr = nullptr;
6652 const char* stubName = nullptr;
6653
6654 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6655 if (stubAddr == nullptr) {
6656 return false; // Intrinsic's stub is not implemented on this platform
6657 }
6658
6659 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6660
6661 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6662
6663 Node* newArr = argument(0);
6664 Node* oldArr = argument(1);
6665 Node* newIdx = argument(2);
6666 Node* shiftCount = argument(3);
6667 Node* numIter = argument(4);
6668
6669 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6670 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6671 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6672 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6673 return false;
6674 }
6675
6676 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6677 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6678 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6679 return false;
6680 }
6681
6682 // Make the call
6683 {
6684 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6685 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6686
6687 Node* call = make_runtime_call(RC_LEAF,
6688 OptoRuntime::bigIntegerShift_Type(),
6689 stubAddr,
6690 stubName,
6691 TypePtr::BOTTOM,
6692 newArr_start,
6693 oldArr_start,
6694 newIdx,
6695 shiftCount,
6696 numIter);
6697 }
6698
6699 return true;
6700 }
6701
6702 //-------------inline_vectorizedMismatch------------------------------
6703 bool LibraryCallKit::inline_vectorizedMismatch() {
6704 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6705
6706 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6707 Node* obja = argument(0); // Object
6708 Node* aoffset = argument(1); // long
6709 Node* objb = argument(3); // Object
6710 Node* boffset = argument(4); // long
6711 Node* length = argument(6); // int
6712 Node* scale = argument(7); // int
6713
6714 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6715 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6716 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6717 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6718 scale == top()) {
6719 return false; // failed input validation
6720 }
6721
6722 Node* obja_adr = make_unsafe_address(obja, aoffset);
6723 Node* objb_adr = make_unsafe_address(objb, boffset);
6724
6725 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6726 //
6727 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6728 // if (length <= inline_limit) {
6729 // inline_path:
6730 // vmask = VectorMaskGen length
6731 // vload1 = LoadVectorMasked obja, vmask
6732 // vload2 = LoadVectorMasked objb, vmask
6733 // result1 = VectorCmpMasked vload1, vload2, vmask
6734 // } else {
6735 // call_stub_path:
6736 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6737 // }
6738 // exit_block:
6739 // return Phi(result1, result2);
6740 //
6741 enum { inline_path = 1, // input is small enough to process it all at once
6742 stub_path = 2, // input is too large; call into the VM
6743 PATH_LIMIT = 3
6744 };
6745
6746 Node* exit_block = new RegionNode(PATH_LIMIT);
6747 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6748 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6749
6750 Node* call_stub_path = control();
6751
6752 BasicType elem_bt = T_ILLEGAL;
6753
6754 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6755 if (scale_t->is_con()) {
6756 switch (scale_t->get_con()) {
6757 case 0: elem_bt = T_BYTE; break;
6758 case 1: elem_bt = T_SHORT; break;
6759 case 2: elem_bt = T_INT; break;
6760 case 3: elem_bt = T_LONG; break;
6761
6762 default: elem_bt = T_ILLEGAL; break; // not supported
6763 }
6764 }
6765
6766 int inline_limit = 0;
6767 bool do_partial_inline = false;
6768
6769 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6770 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6771 do_partial_inline = inline_limit >= 16;
6772 }
6773
6774 if (do_partial_inline) {
6775 assert(elem_bt != T_ILLEGAL, "sanity");
6776
6777 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6778 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6779 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6780
6781 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6782 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6783 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6784
6785 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6786
6787 if (!stopped()) {
6788 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6789
6790 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6791 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6792 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6793 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6794
6795 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6796 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6797 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6798 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6799
6800 exit_block->init_req(inline_path, control());
6801 memory_phi->init_req(inline_path, map()->memory());
6802 result_phi->init_req(inline_path, result);
6803
6804 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6805 clear_upper_avx();
6806 }
6807 }
6808 }
6809
6810 if (call_stub_path != nullptr) {
6811 set_control(call_stub_path);
6812
6813 Node* call = make_runtime_call(RC_LEAF,
6814 OptoRuntime::vectorizedMismatch_Type(),
6815 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6816 obja_adr, objb_adr, length, scale);
6817
6818 exit_block->init_req(stub_path, control());
6819 memory_phi->init_req(stub_path, map()->memory());
6820 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6821 }
6822
6823 exit_block = _gvn.transform(exit_block);
6824 memory_phi = _gvn.transform(memory_phi);
6825 result_phi = _gvn.transform(result_phi);
6826
6827 set_control(exit_block);
6828 set_all_memory(memory_phi);
6829 set_result(result_phi);
6830
6831 return true;
6832 }
6833
6834 //------------------------------inline_vectorizedHashcode----------------------------
6835 bool LibraryCallKit::inline_vectorizedHashCode() {
6836 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6837
6838 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6839 Node* array = argument(0);
6840 Node* offset = argument(1);
6841 Node* length = argument(2);
6842 Node* initialValue = argument(3);
6843 Node* basic_type = argument(4);
6844
6845 if (basic_type == top()) {
6846 return false; // failed input validation
6847 }
6848
6849 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6850 if (!basic_type_t->is_con()) {
6851 return false; // Only intrinsify if mode argument is constant
6852 }
6853
6854 array = must_be_not_null(array, true);
6855
6856 BasicType bt = (BasicType)basic_type_t->get_con();
6857
6858 // Resolve address of first element
6859 Node* array_start = array_element_address(array, offset, bt);
6860
6861 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6862 array_start, length, initialValue, basic_type)));
6863 clear_upper_avx();
6864
6865 return true;
6866 }
6867
6868 /**
6869 * Calculate CRC32 for byte.
6870 * int java.util.zip.CRC32.update(int crc, int b)
6871 */
6872 bool LibraryCallKit::inline_updateCRC32() {
6873 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6874 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6875 // no receiver since it is static method
6876 Node* crc = argument(0); // type: int
6877 Node* b = argument(1); // type: int
6878
6879 /*
6880 * int c = ~ crc;
6881 * b = timesXtoThe32[(b ^ c) & 0xFF];
6882 * b = b ^ (c >>> 8);
6883 * crc = ~b;
6884 */
6885
6886 Node* M1 = intcon(-1);
6887 crc = _gvn.transform(new XorINode(crc, M1));
6888 Node* result = _gvn.transform(new XorINode(crc, b));
6889 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6890
6891 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6892 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6893 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6894 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6895
6896 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6897 result = _gvn.transform(new XorINode(crc, result));
6898 result = _gvn.transform(new XorINode(result, M1));
6899 set_result(result);
6900 return true;
6901 }
6902
6903 /**
6904 * Calculate CRC32 for byte[] array.
6905 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6906 */
6907 bool LibraryCallKit::inline_updateBytesCRC32() {
6908 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6909 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6910 // no receiver since it is static method
6911 Node* crc = argument(0); // type: int
6912 Node* src = argument(1); // type: oop
6913 Node* offset = argument(2); // type: int
6914 Node* length = argument(3); // type: int
6915
6916 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6917 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6918 // failed array check
6919 return false;
6920 }
6921
6922 // Figure out the size and type of the elements we will be copying.
6923 BasicType src_elem = src_type->elem()->array_element_basic_type();
6924 if (src_elem != T_BYTE) {
6925 return false;
6926 }
6927
6928 // 'src_start' points to src array + scaled offset
6929 src = must_be_not_null(src, true);
6930 Node* src_start = array_element_address(src, offset, src_elem);
6931
6932 // We assume that range check is done by caller.
6933 // TODO: generate range check (offset+length < src.length) in debug VM.
6934
6935 // Call the stub.
6936 address stubAddr = StubRoutines::updateBytesCRC32();
6937 const char *stubName = "updateBytesCRC32";
6938
6939 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6940 stubAddr, stubName, TypePtr::BOTTOM,
6941 crc, src_start, length);
6942 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6943 set_result(result);
6944 return true;
6945 }
6946
6947 /**
6948 * Calculate CRC32 for ByteBuffer.
6949 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6950 */
6951 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6952 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6953 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6954 // no receiver since it is static method
6955 Node* crc = argument(0); // type: int
6956 Node* src = argument(1); // type: long
6957 Node* offset = argument(3); // type: int
6958 Node* length = argument(4); // type: int
6959
6960 src = ConvL2X(src); // adjust Java long to machine word
6961 Node* base = _gvn.transform(new CastX2PNode(src));
6962 offset = ConvI2X(offset);
6963
6964 // 'src_start' points to src array + scaled offset
6965 Node* src_start = basic_plus_adr(top(), base, offset);
6966
6967 // Call the stub.
6968 address stubAddr = StubRoutines::updateBytesCRC32();
6969 const char *stubName = "updateBytesCRC32";
6970
6971 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6972 stubAddr, stubName, TypePtr::BOTTOM,
6973 crc, src_start, length);
6974 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6975 set_result(result);
6976 return true;
6977 }
6978
6979 //------------------------------get_table_from_crc32c_class-----------------------
6980 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6981 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6982 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6983
6984 return table;
6985 }
6986
6987 //------------------------------inline_updateBytesCRC32C-----------------------
6988 //
6989 // Calculate CRC32C for byte[] array.
6990 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6991 //
6992 bool LibraryCallKit::inline_updateBytesCRC32C() {
6993 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6994 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6995 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6996 // no receiver since it is a static method
6997 Node* crc = argument(0); // type: int
6998 Node* src = argument(1); // type: oop
6999 Node* offset = argument(2); // type: int
7000 Node* end = argument(3); // type: int
7001
7002 Node* length = _gvn.transform(new SubINode(end, offset));
7003
7004 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7005 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7006 // failed array check
7007 return false;
7008 }
7009
7010 // Figure out the size and type of the elements we will be copying.
7011 BasicType src_elem = src_type->elem()->array_element_basic_type();
7012 if (src_elem != T_BYTE) {
7013 return false;
7014 }
7015
7016 // 'src_start' points to src array + scaled offset
7017 src = must_be_not_null(src, true);
7018 Node* src_start = array_element_address(src, offset, src_elem);
7019
7020 // static final int[] byteTable in class CRC32C
7021 Node* table = get_table_from_crc32c_class(callee()->holder());
7022 table = must_be_not_null(table, true);
7023 Node* table_start = array_element_address(table, intcon(0), T_INT);
7024
7025 // We assume that range check is done by caller.
7026 // TODO: generate range check (offset+length < src.length) in debug VM.
7027
7028 // Call the stub.
7029 address stubAddr = StubRoutines::updateBytesCRC32C();
7030 const char *stubName = "updateBytesCRC32C";
7031
7032 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7033 stubAddr, stubName, TypePtr::BOTTOM,
7034 crc, src_start, length, table_start);
7035 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7036 set_result(result);
7037 return true;
7038 }
7039
7040 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
7041 //
7042 // Calculate CRC32C for DirectByteBuffer.
7043 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
7044 //
7045 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
7046 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7047 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
7048 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7049 // no receiver since it is a static method
7050 Node* crc = argument(0); // type: int
7051 Node* src = argument(1); // type: long
7052 Node* offset = argument(3); // type: int
7053 Node* end = argument(4); // type: int
7054
7055 Node* length = _gvn.transform(new SubINode(end, offset));
7056
7057 src = ConvL2X(src); // adjust Java long to machine word
7058 Node* base = _gvn.transform(new CastX2PNode(src));
7059 offset = ConvI2X(offset);
7060
7061 // 'src_start' points to src array + scaled offset
7062 Node* src_start = basic_plus_adr(top(), base, offset);
7063
7064 // static final int[] byteTable in class CRC32C
7065 Node* table = get_table_from_crc32c_class(callee()->holder());
7066 table = must_be_not_null(table, true);
7067 Node* table_start = array_element_address(table, intcon(0), T_INT);
7068
7069 // Call the stub.
7070 address stubAddr = StubRoutines::updateBytesCRC32C();
7071 const char *stubName = "updateBytesCRC32C";
7072
7073 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7074 stubAddr, stubName, TypePtr::BOTTOM,
7075 crc, src_start, length, table_start);
7076 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7077 set_result(result);
7078 return true;
7079 }
7080
7081 //------------------------------inline_updateBytesAdler32----------------------
7082 //
7083 // Calculate Adler32 checksum for byte[] array.
7084 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
7085 //
7086 bool LibraryCallKit::inline_updateBytesAdler32() {
7087 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7088 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7089 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7090 // no receiver since it is static method
7091 Node* crc = argument(0); // type: int
7092 Node* src = argument(1); // type: oop
7093 Node* offset = argument(2); // type: int
7094 Node* length = argument(3); // type: int
7095
7096 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7097 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7098 // failed array check
7099 return false;
7100 }
7101
7102 // Figure out the size and type of the elements we will be copying.
7103 BasicType src_elem = src_type->elem()->array_element_basic_type();
7104 if (src_elem != T_BYTE) {
7105 return false;
7106 }
7107
7108 // 'src_start' points to src array + scaled offset
7109 Node* src_start = array_element_address(src, offset, src_elem);
7110
7111 // We assume that range check is done by caller.
7112 // TODO: generate range check (offset+length < src.length) in debug VM.
7113
7114 // Call the stub.
7115 address stubAddr = StubRoutines::updateBytesAdler32();
7116 const char *stubName = "updateBytesAdler32";
7117
7118 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7119 stubAddr, stubName, TypePtr::BOTTOM,
7120 crc, src_start, length);
7121 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7122 set_result(result);
7123 return true;
7124 }
7125
7126 //------------------------------inline_updateByteBufferAdler32---------------
7127 //
7128 // Calculate Adler32 checksum for DirectByteBuffer.
7129 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
7130 //
7131 bool LibraryCallKit::inline_updateByteBufferAdler32() {
7132 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7133 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7134 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7135 // no receiver since it is static method
7136 Node* crc = argument(0); // type: int
7137 Node* src = argument(1); // type: long
7138 Node* offset = argument(3); // type: int
7139 Node* length = argument(4); // type: int
7140
7141 src = ConvL2X(src); // adjust Java long to machine word
7142 Node* base = _gvn.transform(new CastX2PNode(src));
7143 offset = ConvI2X(offset);
7144
7145 // 'src_start' points to src array + scaled offset
7146 Node* src_start = basic_plus_adr(top(), base, offset);
7147
7148 // Call the stub.
7149 address stubAddr = StubRoutines::updateBytesAdler32();
7150 const char *stubName = "updateBytesAdler32";
7151
7152 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7153 stubAddr, stubName, TypePtr::BOTTOM,
7154 crc, src_start, length);
7155
7156 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7157 set_result(result);
7158 return true;
7159 }
7160
7161 //----------------------------inline_reference_get----------------------------
7162 // public T java.lang.ref.Reference.get();
7163 bool LibraryCallKit::inline_reference_get() {
7164 const int referent_offset = java_lang_ref_Reference::referent_offset();
7165
7166 // Get the argument:
7167 Node* reference_obj = null_check_receiver();
7168 if (stopped()) return true;
7169
7170 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7171 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7172 decorators, /*is_static*/ false, nullptr);
7173 if (result == nullptr) return false;
7174
7175 // Add memory barrier to prevent commoning reads from this field
7176 // across safepoint since GC can change its value.
7177 insert_mem_bar(Op_MemBarCPUOrder);
7178
7179 set_result(result);
7180 return true;
7181 }
7182
7183 //----------------------------inline_reference_refersTo0----------------------------
7184 // bool java.lang.ref.Reference.refersTo0();
7185 // bool java.lang.ref.PhantomReference.refersTo0();
7186 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7187 // Get arguments:
7188 Node* reference_obj = null_check_receiver();
7189 Node* other_obj = argument(1);
7190 if (stopped()) return true;
7191
7192 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7193 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7194 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7195 decorators, /*is_static*/ false, nullptr);
7196 if (referent == nullptr) return false;
7197
7198 // Add memory barrier to prevent commoning reads from this field
7199 // across safepoint since GC can change its value.
7200 insert_mem_bar(Op_MemBarCPUOrder);
7201
7202 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7203 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7204 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7205
7206 RegionNode* region = new RegionNode(3);
7207 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7208
7209 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7210 region->init_req(1, if_true);
7211 phi->init_req(1, intcon(1));
7212
7213 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7214 region->init_req(2, if_false);
7215 phi->init_req(2, intcon(0));
7216
7217 set_control(_gvn.transform(region));
7218 record_for_igvn(region);
7219 set_result(_gvn.transform(phi));
7220 return true;
7221 }
7222
7223
7224 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7225 DecoratorSet decorators, bool is_static,
7226 ciInstanceKlass* fromKls) {
7227 if (fromKls == nullptr) {
7228 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7229 assert(tinst != nullptr, "obj is null");
7230 assert(tinst->is_loaded(), "obj is not loaded");
7231 fromKls = tinst->instance_klass();
7232 } else {
7233 assert(is_static, "only for static field access");
7234 }
7235 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7236 ciSymbol::make(fieldTypeString),
7237 is_static);
7238
7239 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7240 if (field == nullptr) return (Node *) nullptr;
7241
7242 if (is_static) {
7243 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7244 fromObj = makecon(tip);
7245 }
7246
7247 // Next code copied from Parse::do_get_xxx():
7248
7249 // Compute address and memory type.
7250 int offset = field->offset_in_bytes();
7251 bool is_vol = field->is_volatile();
7252 ciType* field_klass = field->type();
7253 assert(field_klass->is_loaded(), "should be loaded");
7254 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7255 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7256 BasicType bt = field->layout_type();
7257
7258 // Build the resultant type of the load
7259 const Type *type;
7260 if (bt == T_OBJECT) {
7261 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7262 } else {
7263 type = Type::get_const_basic_type(bt);
7264 }
7265
7266 if (is_vol) {
7267 decorators |= MO_SEQ_CST;
7268 }
7269
7270 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7271 }
7272
7273 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7274 bool is_exact /* true */, bool is_static /* false */,
7275 ciInstanceKlass * fromKls /* nullptr */) {
7276 if (fromKls == nullptr) {
7277 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7278 assert(tinst != nullptr, "obj is null");
7279 assert(tinst->is_loaded(), "obj is not loaded");
7280 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7281 fromKls = tinst->instance_klass();
7282 }
7283 else {
7284 assert(is_static, "only for static field access");
7285 }
7286 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7287 ciSymbol::make(fieldTypeString),
7288 is_static);
7289
7290 assert(field != nullptr, "undefined field");
7291 assert(!field->is_volatile(), "not defined for volatile fields");
7292
7293 if (is_static) {
7294 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7295 fromObj = makecon(tip);
7296 }
7297
7298 // Next code copied from Parse::do_get_xxx():
7299
7300 // Compute address and memory type.
7301 int offset = field->offset_in_bytes();
7302 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7303
7304 return adr;
7305 }
7306
7307 //------------------------------inline_aescrypt_Block-----------------------
7308 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7309 address stubAddr = nullptr;
7310 const char *stubName;
7311 assert(UseAES, "need AES instruction support");
7312
7313 switch(id) {
7314 case vmIntrinsics::_aescrypt_encryptBlock:
7315 stubAddr = StubRoutines::aescrypt_encryptBlock();
7316 stubName = "aescrypt_encryptBlock";
7317 break;
7318 case vmIntrinsics::_aescrypt_decryptBlock:
7319 stubAddr = StubRoutines::aescrypt_decryptBlock();
7320 stubName = "aescrypt_decryptBlock";
7321 break;
7322 default:
7323 break;
7324 }
7325 if (stubAddr == nullptr) return false;
7326
7327 Node* aescrypt_object = argument(0);
7328 Node* src = argument(1);
7329 Node* src_offset = argument(2);
7330 Node* dest = argument(3);
7331 Node* dest_offset = argument(4);
7332
7333 src = must_be_not_null(src, true);
7334 dest = must_be_not_null(dest, true);
7335
7336 // (1) src and dest are arrays.
7337 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7338 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7339 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7340 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7341
7342 // for the quick and dirty code we will skip all the checks.
7343 // we are just trying to get the call to be generated.
7344 Node* src_start = src;
7345 Node* dest_start = dest;
7346 if (src_offset != nullptr || dest_offset != nullptr) {
7347 assert(src_offset != nullptr && dest_offset != nullptr, "");
7348 src_start = array_element_address(src, src_offset, T_BYTE);
7349 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7350 }
7351
7352 // now need to get the start of its expanded key array
7353 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7354 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7355 if (k_start == nullptr) return false;
7356
7357 // Call the stub.
7358 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7359 stubAddr, stubName, TypePtr::BOTTOM,
7360 src_start, dest_start, k_start);
7361
7362 return true;
7363 }
7364
7365 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7366 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7367 address stubAddr = nullptr;
7368 const char *stubName = nullptr;
7369
7370 assert(UseAES, "need AES instruction support");
7371
7372 switch(id) {
7373 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7374 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7375 stubName = "cipherBlockChaining_encryptAESCrypt";
7376 break;
7377 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7378 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7379 stubName = "cipherBlockChaining_decryptAESCrypt";
7380 break;
7381 default:
7382 break;
7383 }
7384 if (stubAddr == nullptr) return false;
7385
7386 Node* cipherBlockChaining_object = argument(0);
7387 Node* src = argument(1);
7388 Node* src_offset = argument(2);
7389 Node* len = argument(3);
7390 Node* dest = argument(4);
7391 Node* dest_offset = argument(5);
7392
7393 src = must_be_not_null(src, false);
7394 dest = must_be_not_null(dest, false);
7395
7396 // (1) src and dest are arrays.
7397 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7398 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7399 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7400 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7401
7402 // checks are the responsibility of the caller
7403 Node* src_start = src;
7404 Node* dest_start = dest;
7405 if (src_offset != nullptr || dest_offset != nullptr) {
7406 assert(src_offset != nullptr && dest_offset != nullptr, "");
7407 src_start = array_element_address(src, src_offset, T_BYTE);
7408 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7409 }
7410
7411 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7412 // (because of the predicated logic executed earlier).
7413 // so we cast it here safely.
7414 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7415
7416 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7417 if (embeddedCipherObj == nullptr) return false;
7418
7419 // cast it to what we know it will be at runtime
7420 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7421 assert(tinst != nullptr, "CBC obj is null");
7422 assert(tinst->is_loaded(), "CBC obj is not loaded");
7423 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7424 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7425
7426 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7427 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7428 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7429 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7430 aescrypt_object = _gvn.transform(aescrypt_object);
7431
7432 // we need to get the start of the aescrypt_object's expanded key array
7433 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7434 if (k_start == nullptr) return false;
7435
7436 // similarly, get the start address of the r vector
7437 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7438 if (objRvec == nullptr) return false;
7439 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7440
7441 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7442 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7443 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7444 stubAddr, stubName, TypePtr::BOTTOM,
7445 src_start, dest_start, k_start, r_start, len);
7446
7447 // return cipher length (int)
7448 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7449 set_result(retvalue);
7450 return true;
7451 }
7452
7453 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7454 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7455 address stubAddr = nullptr;
7456 const char *stubName = nullptr;
7457
7458 assert(UseAES, "need AES instruction support");
7459
7460 switch (id) {
7461 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7462 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7463 stubName = "electronicCodeBook_encryptAESCrypt";
7464 break;
7465 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7466 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7467 stubName = "electronicCodeBook_decryptAESCrypt";
7468 break;
7469 default:
7470 break;
7471 }
7472
7473 if (stubAddr == nullptr) return false;
7474
7475 Node* electronicCodeBook_object = argument(0);
7476 Node* src = argument(1);
7477 Node* src_offset = argument(2);
7478 Node* len = argument(3);
7479 Node* dest = argument(4);
7480 Node* dest_offset = argument(5);
7481
7482 // (1) src and dest are arrays.
7483 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7484 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7485 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7486 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7487
7488 // checks are the responsibility of the caller
7489 Node* src_start = src;
7490 Node* dest_start = dest;
7491 if (src_offset != nullptr || dest_offset != nullptr) {
7492 assert(src_offset != nullptr && dest_offset != nullptr, "");
7493 src_start = array_element_address(src, src_offset, T_BYTE);
7494 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7495 }
7496
7497 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7498 // (because of the predicated logic executed earlier).
7499 // so we cast it here safely.
7500 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7501
7502 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7503 if (embeddedCipherObj == nullptr) return false;
7504
7505 // cast it to what we know it will be at runtime
7506 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7507 assert(tinst != nullptr, "ECB obj is null");
7508 assert(tinst->is_loaded(), "ECB obj is not loaded");
7509 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7510 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7511
7512 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7513 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7514 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7515 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7516 aescrypt_object = _gvn.transform(aescrypt_object);
7517
7518 // we need to get the start of the aescrypt_object's expanded key array
7519 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7520 if (k_start == nullptr) return false;
7521
7522 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7523 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7524 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7525 stubAddr, stubName, TypePtr::BOTTOM,
7526 src_start, dest_start, k_start, len);
7527
7528 // return cipher length (int)
7529 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7530 set_result(retvalue);
7531 return true;
7532 }
7533
7534 //------------------------------inline_counterMode_AESCrypt-----------------------
7535 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7536 assert(UseAES, "need AES instruction support");
7537 if (!UseAESCTRIntrinsics) return false;
7538
7539 address stubAddr = nullptr;
7540 const char *stubName = nullptr;
7541 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7542 stubAddr = StubRoutines::counterMode_AESCrypt();
7543 stubName = "counterMode_AESCrypt";
7544 }
7545 if (stubAddr == nullptr) return false;
7546
7547 Node* counterMode_object = argument(0);
7548 Node* src = argument(1);
7549 Node* src_offset = argument(2);
7550 Node* len = argument(3);
7551 Node* dest = argument(4);
7552 Node* dest_offset = argument(5);
7553
7554 // (1) src and dest are arrays.
7555 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7556 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7557 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7558 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7559
7560 // checks are the responsibility of the caller
7561 Node* src_start = src;
7562 Node* dest_start = dest;
7563 if (src_offset != nullptr || dest_offset != nullptr) {
7564 assert(src_offset != nullptr && dest_offset != nullptr, "");
7565 src_start = array_element_address(src, src_offset, T_BYTE);
7566 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7567 }
7568
7569 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7570 // (because of the predicated logic executed earlier).
7571 // so we cast it here safely.
7572 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7573 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7574 if (embeddedCipherObj == nullptr) return false;
7575 // cast it to what we know it will be at runtime
7576 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7577 assert(tinst != nullptr, "CTR obj is null");
7578 assert(tinst->is_loaded(), "CTR obj is not loaded");
7579 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7580 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7581 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7582 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7583 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7584 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7585 aescrypt_object = _gvn.transform(aescrypt_object);
7586 // we need to get the start of the aescrypt_object's expanded key array
7587 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7588 if (k_start == nullptr) return false;
7589 // similarly, get the start address of the r vector
7590 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7591 if (obj_counter == nullptr) return false;
7592 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7593
7594 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7595 if (saved_encCounter == nullptr) return false;
7596 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7597 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7598
7599 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7600 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7601 OptoRuntime::counterMode_aescrypt_Type(),
7602 stubAddr, stubName, TypePtr::BOTTOM,
7603 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7604
7605 // return cipher length (int)
7606 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7607 set_result(retvalue);
7608 return true;
7609 }
7610
7611 //------------------------------get_key_start_from_aescrypt_object-----------------------
7612 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7613 #if defined(PPC64) || defined(S390)
7614 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7615 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7616 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7617 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7618 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7619 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7620 if (objSessionK == nullptr) {
7621 return (Node *) nullptr;
7622 }
7623 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7624 #else
7625 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7626 #endif // PPC64
7627 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7628 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7629
7630 // now have the array, need to get the start address of the K array
7631 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7632 return k_start;
7633 }
7634
7635 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7636 // Return node representing slow path of predicate check.
7637 // the pseudo code we want to emulate with this predicate is:
7638 // for encryption:
7639 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7640 // for decryption:
7641 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7642 // note cipher==plain is more conservative than the original java code but that's OK
7643 //
7644 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7645 // The receiver was checked for null already.
7646 Node* objCBC = argument(0);
7647
7648 Node* src = argument(1);
7649 Node* dest = argument(4);
7650
7651 // Load embeddedCipher field of CipherBlockChaining object.
7652 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7653
7654 // get AESCrypt klass for instanceOf check
7655 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7656 // will have same classloader as CipherBlockChaining object
7657 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7658 assert(tinst != nullptr, "CBCobj is null");
7659 assert(tinst->is_loaded(), "CBCobj is not loaded");
7660
7661 // we want to do an instanceof comparison against the AESCrypt class
7662 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7663 if (!klass_AESCrypt->is_loaded()) {
7664 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7665 Node* ctrl = control();
7666 set_control(top()); // no regular fast path
7667 return ctrl;
7668 }
7669
7670 src = must_be_not_null(src, true);
7671 dest = must_be_not_null(dest, true);
7672
7673 // Resolve oops to stable for CmpP below.
7674 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7675
7676 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7677 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7678 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7679
7680 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7681
7682 // for encryption, we are done
7683 if (!decrypting)
7684 return instof_false; // even if it is null
7685
7686 // for decryption, we need to add a further check to avoid
7687 // taking the intrinsic path when cipher and plain are the same
7688 // see the original java code for why.
7689 RegionNode* region = new RegionNode(3);
7690 region->init_req(1, instof_false);
7691
7692 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7693 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7694 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7695 region->init_req(2, src_dest_conjoint);
7696
7697 record_for_igvn(region);
7698 return _gvn.transform(region);
7699 }
7700
7701 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7702 // Return node representing slow path of predicate check.
7703 // the pseudo code we want to emulate with this predicate is:
7704 // for encryption:
7705 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7706 // for decryption:
7707 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7708 // note cipher==plain is more conservative than the original java code but that's OK
7709 //
7710 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7711 // The receiver was checked for null already.
7712 Node* objECB = argument(0);
7713
7714 // Load embeddedCipher field of ElectronicCodeBook object.
7715 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7716
7717 // get AESCrypt klass for instanceOf check
7718 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7719 // will have same classloader as ElectronicCodeBook object
7720 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7721 assert(tinst != nullptr, "ECBobj is null");
7722 assert(tinst->is_loaded(), "ECBobj is not loaded");
7723
7724 // we want to do an instanceof comparison against the AESCrypt class
7725 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7726 if (!klass_AESCrypt->is_loaded()) {
7727 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7728 Node* ctrl = control();
7729 set_control(top()); // no regular fast path
7730 return ctrl;
7731 }
7732 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7733
7734 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7735 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7736 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7737
7738 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7739
7740 // for encryption, we are done
7741 if (!decrypting)
7742 return instof_false; // even if it is null
7743
7744 // for decryption, we need to add a further check to avoid
7745 // taking the intrinsic path when cipher and plain are the same
7746 // see the original java code for why.
7747 RegionNode* region = new RegionNode(3);
7748 region->init_req(1, instof_false);
7749 Node* src = argument(1);
7750 Node* dest = argument(4);
7751 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7752 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7753 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7754 region->init_req(2, src_dest_conjoint);
7755
7756 record_for_igvn(region);
7757 return _gvn.transform(region);
7758 }
7759
7760 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7761 // Return node representing slow path of predicate check.
7762 // the pseudo code we want to emulate with this predicate is:
7763 // for encryption:
7764 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7765 // for decryption:
7766 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7767 // note cipher==plain is more conservative than the original java code but that's OK
7768 //
7769
7770 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7771 // The receiver was checked for null already.
7772 Node* objCTR = argument(0);
7773
7774 // Load embeddedCipher field of CipherBlockChaining object.
7775 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7776
7777 // get AESCrypt klass for instanceOf check
7778 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7779 // will have same classloader as CipherBlockChaining object
7780 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7781 assert(tinst != nullptr, "CTRobj is null");
7782 assert(tinst->is_loaded(), "CTRobj is not loaded");
7783
7784 // we want to do an instanceof comparison against the AESCrypt class
7785 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7786 if (!klass_AESCrypt->is_loaded()) {
7787 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7788 Node* ctrl = control();
7789 set_control(top()); // no regular fast path
7790 return ctrl;
7791 }
7792
7793 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7794 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7795 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7796 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7797 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7798
7799 return instof_false; // even if it is null
7800 }
7801
7802 //------------------------------inline_ghash_processBlocks
7803 bool LibraryCallKit::inline_ghash_processBlocks() {
7804 address stubAddr;
7805 const char *stubName;
7806 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7807
7808 stubAddr = StubRoutines::ghash_processBlocks();
7809 stubName = "ghash_processBlocks";
7810
7811 Node* data = argument(0);
7812 Node* offset = argument(1);
7813 Node* len = argument(2);
7814 Node* state = argument(3);
7815 Node* subkeyH = argument(4);
7816
7817 state = must_be_not_null(state, true);
7818 subkeyH = must_be_not_null(subkeyH, true);
7819 data = must_be_not_null(data, true);
7820
7821 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7822 assert(state_start, "state is null");
7823 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7824 assert(subkeyH_start, "subkeyH is null");
7825 Node* data_start = array_element_address(data, offset, T_BYTE);
7826 assert(data_start, "data is null");
7827
7828 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7829 OptoRuntime::ghash_processBlocks_Type(),
7830 stubAddr, stubName, TypePtr::BOTTOM,
7831 state_start, subkeyH_start, data_start, len);
7832 return true;
7833 }
7834
7835 //------------------------------inline_chacha20Block
7836 bool LibraryCallKit::inline_chacha20Block() {
7837 address stubAddr;
7838 const char *stubName;
7839 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7840
7841 stubAddr = StubRoutines::chacha20Block();
7842 stubName = "chacha20Block";
7843
7844 Node* state = argument(0);
7845 Node* result = argument(1);
7846
7847 state = must_be_not_null(state, true);
7848 result = must_be_not_null(result, true);
7849
7850 Node* state_start = array_element_address(state, intcon(0), T_INT);
7851 assert(state_start, "state is null");
7852 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7853 assert(result_start, "result is null");
7854
7855 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7856 OptoRuntime::chacha20Block_Type(),
7857 stubAddr, stubName, TypePtr::BOTTOM,
7858 state_start, result_start);
7859 // return key stream length (int)
7860 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7861 set_result(retvalue);
7862 return true;
7863 }
7864
7865 bool LibraryCallKit::inline_base64_encodeBlock() {
7866 address stubAddr;
7867 const char *stubName;
7868 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7869 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
7870 stubAddr = StubRoutines::base64_encodeBlock();
7871 stubName = "encodeBlock";
7872
7873 if (!stubAddr) return false;
7874 Node* base64obj = argument(0);
7875 Node* src = argument(1);
7876 Node* offset = argument(2);
7877 Node* len = argument(3);
7878 Node* dest = argument(4);
7879 Node* dp = argument(5);
7880 Node* isURL = argument(6);
7881
7882 src = must_be_not_null(src, true);
7883 dest = must_be_not_null(dest, true);
7884
7885 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7886 assert(src_start, "source array is null");
7887 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7888 assert(dest_start, "destination array is null");
7889
7890 Node* base64 = make_runtime_call(RC_LEAF,
7891 OptoRuntime::base64_encodeBlock_Type(),
7892 stubAddr, stubName, TypePtr::BOTTOM,
7893 src_start, offset, len, dest_start, dp, isURL);
7894 return true;
7895 }
7896
7897 bool LibraryCallKit::inline_base64_decodeBlock() {
7898 address stubAddr;
7899 const char *stubName;
7900 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7901 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
7902 stubAddr = StubRoutines::base64_decodeBlock();
7903 stubName = "decodeBlock";
7904
7905 if (!stubAddr) return false;
7906 Node* base64obj = argument(0);
7907 Node* src = argument(1);
7908 Node* src_offset = argument(2);
7909 Node* len = argument(3);
7910 Node* dest = argument(4);
7911 Node* dest_offset = argument(5);
7912 Node* isURL = argument(6);
7913 Node* isMIME = argument(7);
7914
7915 src = must_be_not_null(src, true);
7916 dest = must_be_not_null(dest, true);
7917
7918 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7919 assert(src_start, "source array is null");
7920 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7921 assert(dest_start, "destination array is null");
7922
7923 Node* call = make_runtime_call(RC_LEAF,
7924 OptoRuntime::base64_decodeBlock_Type(),
7925 stubAddr, stubName, TypePtr::BOTTOM,
7926 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
7927 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7928 set_result(result);
7929 return true;
7930 }
7931
7932 bool LibraryCallKit::inline_poly1305_processBlocks() {
7933 address stubAddr;
7934 const char *stubName;
7935 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
7936 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
7937 stubAddr = StubRoutines::poly1305_processBlocks();
7938 stubName = "poly1305_processBlocks";
7939
7940 if (!stubAddr) return false;
7941 null_check_receiver(); // null-check receiver
7942 if (stopped()) return true;
7943
7944 Node* input = argument(1);
7945 Node* input_offset = argument(2);
7946 Node* len = argument(3);
7947 Node* alimbs = argument(4);
7948 Node* rlimbs = argument(5);
7949
7950 input = must_be_not_null(input, true);
7951 alimbs = must_be_not_null(alimbs, true);
7952 rlimbs = must_be_not_null(rlimbs, true);
7953
7954 Node* input_start = array_element_address(input, input_offset, T_BYTE);
7955 assert(input_start, "input array is null");
7956 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
7957 assert(acc_start, "acc array is null");
7958 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
7959 assert(r_start, "r array is null");
7960
7961 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7962 OptoRuntime::poly1305_processBlocks_Type(),
7963 stubAddr, stubName, TypePtr::BOTTOM,
7964 input_start, len, acc_start, r_start);
7965 return true;
7966 }
7967
7968 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
7969 address stubAddr;
7970 const char *stubName;
7971 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
7972 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
7973 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
7974 stubName = "intpoly_montgomeryMult_P256";
7975
7976 if (!stubAddr) return false;
7977 null_check_receiver(); // null-check receiver
7978 if (stopped()) return true;
7979
7980 Node* a = argument(1);
7981 Node* b = argument(2);
7982 Node* r = argument(3);
7983
7984 a = must_be_not_null(a, true);
7985 b = must_be_not_null(b, true);
7986 r = must_be_not_null(r, true);
7987
7988 Node* a_start = array_element_address(a, intcon(0), T_LONG);
7989 assert(a_start, "a array is NULL");
7990 Node* b_start = array_element_address(b, intcon(0), T_LONG);
7991 assert(b_start, "b array is NULL");
7992 Node* r_start = array_element_address(r, intcon(0), T_LONG);
7993 assert(r_start, "r array is NULL");
7994
7995 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7996 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
7997 stubAddr, stubName, TypePtr::BOTTOM,
7998 a_start, b_start, r_start);
7999 return true;
8000 }
8001
8002 bool LibraryCallKit::inline_intpoly_assign() {
8003 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8004 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
8005 const char *stubName = "intpoly_assign";
8006 address stubAddr = StubRoutines::intpoly_assign();
8007 if (!stubAddr) return false;
8008
8009 Node* set = argument(0);
8010 Node* a = argument(1);
8011 Node* b = argument(2);
8012 Node* arr_length = load_array_length(a);
8013
8014 a = must_be_not_null(a, true);
8015 b = must_be_not_null(b, true);
8016
8017 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8018 assert(a_start, "a array is NULL");
8019 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8020 assert(b_start, "b array is NULL");
8021
8022 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8023 OptoRuntime::intpoly_assign_Type(),
8024 stubAddr, stubName, TypePtr::BOTTOM,
8025 set, a_start, b_start, arr_length);
8026 return true;
8027 }
8028
8029 //------------------------------inline_digestBase_implCompress-----------------------
8030 //
8031 // Calculate MD5 for single-block byte[] array.
8032 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
8033 //
8034 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
8035 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
8036 //
8037 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
8038 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
8039 //
8040 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
8041 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
8042 //
8043 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
8044 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
8045 //
8046 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
8047 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
8048
8049 Node* digestBase_obj = argument(0);
8050 Node* src = argument(1); // type oop
8051 Node* ofs = argument(2); // type int
8052
8053 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8054 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8055 // failed array check
8056 return false;
8057 }
8058 // Figure out the size and type of the elements we will be copying.
8059 BasicType src_elem = src_type->elem()->array_element_basic_type();
8060 if (src_elem != T_BYTE) {
8061 return false;
8062 }
8063 // 'src_start' points to src array + offset
8064 src = must_be_not_null(src, true);
8065 Node* src_start = array_element_address(src, ofs, src_elem);
8066 Node* state = nullptr;
8067 Node* block_size = nullptr;
8068 address stubAddr;
8069 const char *stubName;
8070
8071 switch(id) {
8072 case vmIntrinsics::_md5_implCompress:
8073 assert(UseMD5Intrinsics, "need MD5 instruction support");
8074 state = get_state_from_digest_object(digestBase_obj, T_INT);
8075 stubAddr = StubRoutines::md5_implCompress();
8076 stubName = "md5_implCompress";
8077 break;
8078 case vmIntrinsics::_sha_implCompress:
8079 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
8080 state = get_state_from_digest_object(digestBase_obj, T_INT);
8081 stubAddr = StubRoutines::sha1_implCompress();
8082 stubName = "sha1_implCompress";
8083 break;
8084 case vmIntrinsics::_sha2_implCompress:
8085 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
8086 state = get_state_from_digest_object(digestBase_obj, T_INT);
8087 stubAddr = StubRoutines::sha256_implCompress();
8088 stubName = "sha256_implCompress";
8089 break;
8090 case vmIntrinsics::_sha5_implCompress:
8091 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
8092 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8093 stubAddr = StubRoutines::sha512_implCompress();
8094 stubName = "sha512_implCompress";
8095 break;
8096 case vmIntrinsics::_sha3_implCompress:
8097 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
8098 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8099 stubAddr = StubRoutines::sha3_implCompress();
8100 stubName = "sha3_implCompress";
8101 block_size = get_block_size_from_digest_object(digestBase_obj);
8102 if (block_size == nullptr) return false;
8103 break;
8104 default:
8105 fatal_unexpected_iid(id);
8106 return false;
8107 }
8108 if (state == nullptr) return false;
8109
8110 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
8111 if (stubAddr == nullptr) return false;
8112
8113 // Call the stub.
8114 Node* call;
8115 if (block_size == nullptr) {
8116 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
8117 stubAddr, stubName, TypePtr::BOTTOM,
8118 src_start, state);
8119 } else {
8120 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
8121 stubAddr, stubName, TypePtr::BOTTOM,
8122 src_start, state, block_size);
8123 }
8124
8125 return true;
8126 }
8127
8128 //------------------------------inline_digestBase_implCompressMB-----------------------
8129 //
8130 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
8131 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
8132 //
8133 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
8134 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8135 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8136 assert((uint)predicate < 5, "sanity");
8137 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
8138
8139 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
8140 Node* src = argument(1); // byte[] array
8141 Node* ofs = argument(2); // type int
8142 Node* limit = argument(3); // type int
8143
8144 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8145 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8146 // failed array check
8147 return false;
8148 }
8149 // Figure out the size and type of the elements we will be copying.
8150 BasicType src_elem = src_type->elem()->array_element_basic_type();
8151 if (src_elem != T_BYTE) {
8152 return false;
8153 }
8154 // 'src_start' points to src array + offset
8155 src = must_be_not_null(src, false);
8156 Node* src_start = array_element_address(src, ofs, src_elem);
8157
8158 const char* klass_digestBase_name = nullptr;
8159 const char* stub_name = nullptr;
8160 address stub_addr = nullptr;
8161 BasicType elem_type = T_INT;
8162
8163 switch (predicate) {
8164 case 0:
8165 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
8166 klass_digestBase_name = "sun/security/provider/MD5";
8167 stub_name = "md5_implCompressMB";
8168 stub_addr = StubRoutines::md5_implCompressMB();
8169 }
8170 break;
8171 case 1:
8172 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
8173 klass_digestBase_name = "sun/security/provider/SHA";
8174 stub_name = "sha1_implCompressMB";
8175 stub_addr = StubRoutines::sha1_implCompressMB();
8176 }
8177 break;
8178 case 2:
8179 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
8180 klass_digestBase_name = "sun/security/provider/SHA2";
8181 stub_name = "sha256_implCompressMB";
8182 stub_addr = StubRoutines::sha256_implCompressMB();
8183 }
8184 break;
8185 case 3:
8186 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
8187 klass_digestBase_name = "sun/security/provider/SHA5";
8188 stub_name = "sha512_implCompressMB";
8189 stub_addr = StubRoutines::sha512_implCompressMB();
8190 elem_type = T_LONG;
8191 }
8192 break;
8193 case 4:
8194 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
8195 klass_digestBase_name = "sun/security/provider/SHA3";
8196 stub_name = "sha3_implCompressMB";
8197 stub_addr = StubRoutines::sha3_implCompressMB();
8198 elem_type = T_LONG;
8199 }
8200 break;
8201 default:
8202 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
8203 }
8204 if (klass_digestBase_name != nullptr) {
8205 assert(stub_addr != nullptr, "Stub is generated");
8206 if (stub_addr == nullptr) return false;
8207
8208 // get DigestBase klass to lookup for SHA klass
8209 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
8210 assert(tinst != nullptr, "digestBase_obj is not instance???");
8211 assert(tinst->is_loaded(), "DigestBase is not loaded");
8212
8213 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
8214 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
8215 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
8216 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
8217 }
8218 return false;
8219 }
8220
8221 //------------------------------inline_digestBase_implCompressMB-----------------------
8222 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
8223 BasicType elem_type, address stubAddr, const char *stubName,
8224 Node* src_start, Node* ofs, Node* limit) {
8225 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
8226 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8227 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
8228 digest_obj = _gvn.transform(digest_obj);
8229
8230 Node* state = get_state_from_digest_object(digest_obj, elem_type);
8231 if (state == nullptr) return false;
8232
8233 Node* block_size = nullptr;
8234 if (strcmp("sha3_implCompressMB", stubName) == 0) {
8235 block_size = get_block_size_from_digest_object(digest_obj);
8236 if (block_size == nullptr) return false;
8237 }
8238
8239 // Call the stub.
8240 Node* call;
8241 if (block_size == nullptr) {
8242 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8243 OptoRuntime::digestBase_implCompressMB_Type(false),
8244 stubAddr, stubName, TypePtr::BOTTOM,
8245 src_start, state, ofs, limit);
8246 } else {
8247 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8248 OptoRuntime::digestBase_implCompressMB_Type(true),
8249 stubAddr, stubName, TypePtr::BOTTOM,
8250 src_start, state, block_size, ofs, limit);
8251 }
8252
8253 // return ofs (int)
8254 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8255 set_result(result);
8256
8257 return true;
8258 }
8259
8260 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8261 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8262 assert(UseAES, "need AES instruction support");
8263 address stubAddr = nullptr;
8264 const char *stubName = nullptr;
8265 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8266 stubName = "galoisCounterMode_AESCrypt";
8267
8268 if (stubAddr == nullptr) return false;
8269
8270 Node* in = argument(0);
8271 Node* inOfs = argument(1);
8272 Node* len = argument(2);
8273 Node* ct = argument(3);
8274 Node* ctOfs = argument(4);
8275 Node* out = argument(5);
8276 Node* outOfs = argument(6);
8277 Node* gctr_object = argument(7);
8278 Node* ghash_object = argument(8);
8279
8280 // (1) in, ct and out are arrays.
8281 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8282 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8283 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8284 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8285 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8286 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8287
8288 // checks are the responsibility of the caller
8289 Node* in_start = in;
8290 Node* ct_start = ct;
8291 Node* out_start = out;
8292 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8293 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8294 in_start = array_element_address(in, inOfs, T_BYTE);
8295 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8296 out_start = array_element_address(out, outOfs, T_BYTE);
8297 }
8298
8299 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8300 // (because of the predicated logic executed earlier).
8301 // so we cast it here safely.
8302 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8303 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8304 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8305 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8306 Node* state = load_field_from_object(ghash_object, "state", "[J");
8307
8308 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8309 return false;
8310 }
8311 // cast it to what we know it will be at runtime
8312 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8313 assert(tinst != nullptr, "GCTR obj is null");
8314 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8315 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8316 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8317 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8318 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8319 const TypeOopPtr* xtype = aklass->as_instance_type();
8320 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8321 aescrypt_object = _gvn.transform(aescrypt_object);
8322 // we need to get the start of the aescrypt_object's expanded key array
8323 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8324 if (k_start == nullptr) return false;
8325 // similarly, get the start address of the r vector
8326 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8327 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8328 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8329
8330
8331 // Call the stub, passing params
8332 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8333 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8334 stubAddr, stubName, TypePtr::BOTTOM,
8335 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8336
8337 // return cipher length (int)
8338 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8339 set_result(retvalue);
8340
8341 return true;
8342 }
8343
8344 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8345 // Return node representing slow path of predicate check.
8346 // the pseudo code we want to emulate with this predicate is:
8347 // for encryption:
8348 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8349 // for decryption:
8350 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8351 // note cipher==plain is more conservative than the original java code but that's OK
8352 //
8353
8354 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8355 // The receiver was checked for null already.
8356 Node* objGCTR = argument(7);
8357 // Load embeddedCipher field of GCTR object.
8358 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8359 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8360
8361 // get AESCrypt klass for instanceOf check
8362 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8363 // will have same classloader as CipherBlockChaining object
8364 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8365 assert(tinst != nullptr, "GCTR obj is null");
8366 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8367
8368 // we want to do an instanceof comparison against the AESCrypt class
8369 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8370 if (!klass_AESCrypt->is_loaded()) {
8371 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8372 Node* ctrl = control();
8373 set_control(top()); // no regular fast path
8374 return ctrl;
8375 }
8376
8377 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8378 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8379 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8380 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8381 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8382
8383 return instof_false; // even if it is null
8384 }
8385
8386 //------------------------------get_state_from_digest_object-----------------------
8387 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8388 const char* state_type;
8389 switch (elem_type) {
8390 case T_BYTE: state_type = "[B"; break;
8391 case T_INT: state_type = "[I"; break;
8392 case T_LONG: state_type = "[J"; break;
8393 default: ShouldNotReachHere();
8394 }
8395 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8396 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8397 if (digest_state == nullptr) return (Node *) nullptr;
8398
8399 // now have the array, need to get the start address of the state array
8400 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8401 return state;
8402 }
8403
8404 //------------------------------get_block_size_from_sha3_object----------------------------------
8405 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8406 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8407 assert (block_size != nullptr, "sanity");
8408 return block_size;
8409 }
8410
8411 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8412 // Return node representing slow path of predicate check.
8413 // the pseudo code we want to emulate with this predicate is:
8414 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8415 //
8416 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8417 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8418 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8419 assert((uint)predicate < 5, "sanity");
8420
8421 // The receiver was checked for null already.
8422 Node* digestBaseObj = argument(0);
8423
8424 // get DigestBase klass for instanceOf check
8425 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8426 assert(tinst != nullptr, "digestBaseObj is null");
8427 assert(tinst->is_loaded(), "DigestBase is not loaded");
8428
8429 const char* klass_name = nullptr;
8430 switch (predicate) {
8431 case 0:
8432 if (UseMD5Intrinsics) {
8433 // we want to do an instanceof comparison against the MD5 class
8434 klass_name = "sun/security/provider/MD5";
8435 }
8436 break;
8437 case 1:
8438 if (UseSHA1Intrinsics) {
8439 // we want to do an instanceof comparison against the SHA class
8440 klass_name = "sun/security/provider/SHA";
8441 }
8442 break;
8443 case 2:
8444 if (UseSHA256Intrinsics) {
8445 // we want to do an instanceof comparison against the SHA2 class
8446 klass_name = "sun/security/provider/SHA2";
8447 }
8448 break;
8449 case 3:
8450 if (UseSHA512Intrinsics) {
8451 // we want to do an instanceof comparison against the SHA5 class
8452 klass_name = "sun/security/provider/SHA5";
8453 }
8454 break;
8455 case 4:
8456 if (UseSHA3Intrinsics) {
8457 // we want to do an instanceof comparison against the SHA3 class
8458 klass_name = "sun/security/provider/SHA3";
8459 }
8460 break;
8461 default:
8462 fatal("unknown SHA intrinsic predicate: %d", predicate);
8463 }
8464
8465 ciKlass* klass = nullptr;
8466 if (klass_name != nullptr) {
8467 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8468 }
8469 if ((klass == nullptr) || !klass->is_loaded()) {
8470 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8471 Node* ctrl = control();
8472 set_control(top()); // no intrinsic path
8473 return ctrl;
8474 }
8475 ciInstanceKlass* instklass = klass->as_instance_klass();
8476
8477 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8478 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8479 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8480 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8481
8482 return instof_false; // even if it is null
8483 }
8484
8485 //-------------inline_fma-----------------------------------
8486 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8487 Node *a = nullptr;
8488 Node *b = nullptr;
8489 Node *c = nullptr;
8490 Node* result = nullptr;
8491 switch (id) {
8492 case vmIntrinsics::_fmaD:
8493 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8494 // no receiver since it is static method
8495 a = round_double_node(argument(0));
8496 b = round_double_node(argument(2));
8497 c = round_double_node(argument(4));
8498 result = _gvn.transform(new FmaDNode(control(), a, b, c));
8499 break;
8500 case vmIntrinsics::_fmaF:
8501 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8502 a = argument(0);
8503 b = argument(1);
8504 c = argument(2);
8505 result = _gvn.transform(new FmaFNode(control(), a, b, c));
8506 break;
8507 default:
8508 fatal_unexpected_iid(id); break;
8509 }
8510 set_result(result);
8511 return true;
8512 }
8513
8514 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8515 // argument(0) is receiver
8516 Node* codePoint = argument(1);
8517 Node* n = nullptr;
8518
8519 switch (id) {
8520 case vmIntrinsics::_isDigit :
8521 n = new DigitNode(control(), codePoint);
8522 break;
8523 case vmIntrinsics::_isLowerCase :
8524 n = new LowerCaseNode(control(), codePoint);
8525 break;
8526 case vmIntrinsics::_isUpperCase :
8527 n = new UpperCaseNode(control(), codePoint);
8528 break;
8529 case vmIntrinsics::_isWhitespace :
8530 n = new WhitespaceNode(control(), codePoint);
8531 break;
8532 default:
8533 fatal_unexpected_iid(id);
8534 }
8535
8536 set_result(_gvn.transform(n));
8537 return true;
8538 }
8539
8540 //------------------------------inline_fp_min_max------------------------------
8541 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
8542 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
8543
8544 // The intrinsic should be used only when the API branches aren't predictable,
8545 // the last one performing the most important comparison. The following heuristic
8546 // uses the branch statistics to eventually bail out if necessary.
8547
8548 ciMethodData *md = callee()->method_data();
8549
8550 if ( md != nullptr && md->is_mature() && md->invocation_count() > 0 ) {
8551 ciCallProfile cp = caller()->call_profile_at_bci(bci());
8552
8553 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
8554 // Bail out if the call-site didn't contribute enough to the statistics.
8555 return false;
8556 }
8557
8558 uint taken = 0, not_taken = 0;
8559
8560 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
8561 if (p->is_BranchData()) {
8562 taken = ((ciBranchData*)p)->taken();
8563 not_taken = ((ciBranchData*)p)->not_taken();
8564 }
8565 }
8566
8567 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
8568 balance = balance < 0 ? -balance : balance;
8569 if ( balance > 0.2 ) {
8570 // Bail out if the most important branch is predictable enough.
8571 return false;
8572 }
8573 }
8574 */
8575
8576 Node *a = nullptr;
8577 Node *b = nullptr;
8578 Node *n = nullptr;
8579 switch (id) {
8580 case vmIntrinsics::_maxF:
8581 case vmIntrinsics::_minF:
8582 case vmIntrinsics::_maxF_strict:
8583 case vmIntrinsics::_minF_strict:
8584 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
8585 a = argument(0);
8586 b = argument(1);
8587 break;
8588 case vmIntrinsics::_maxD:
8589 case vmIntrinsics::_minD:
8590 case vmIntrinsics::_maxD_strict:
8591 case vmIntrinsics::_minD_strict:
8592 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
8593 a = round_double_node(argument(0));
8594 b = round_double_node(argument(2));
8595 break;
8596 default:
8597 fatal_unexpected_iid(id);
8598 break;
8599 }
8600 switch (id) {
8601 case vmIntrinsics::_maxF:
8602 case vmIntrinsics::_maxF_strict:
8603 n = new MaxFNode(a, b);
8604 break;
8605 case vmIntrinsics::_minF:
8606 case vmIntrinsics::_minF_strict:
8607 n = new MinFNode(a, b);
8608 break;
8609 case vmIntrinsics::_maxD:
8610 case vmIntrinsics::_maxD_strict:
8611 n = new MaxDNode(a, b);
8612 break;
8613 case vmIntrinsics::_minD:
8614 case vmIntrinsics::_minD_strict:
8615 n = new MinDNode(a, b);
8616 break;
8617 default:
8618 fatal_unexpected_iid(id);
8619 break;
8620 }
8621 set_result(_gvn.transform(n));
8622 return true;
8623 }
8624
8625 bool LibraryCallKit::inline_profileBoolean() {
8626 Node* counts = argument(1);
8627 const TypeAryPtr* ary = nullptr;
8628 ciArray* aobj = nullptr;
8629 if (counts->is_Con()
8630 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8631 && (aobj = ary->const_oop()->as_array()) != nullptr
8632 && (aobj->length() == 2)) {
8633 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8634 jint false_cnt = aobj->element_value(0).as_int();
8635 jint true_cnt = aobj->element_value(1).as_int();
8636
8637 if (C->log() != nullptr) {
8638 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8639 false_cnt, true_cnt);
8640 }
8641
8642 if (false_cnt + true_cnt == 0) {
8643 // According to profile, never executed.
8644 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8645 Deoptimization::Action_reinterpret);
8646 return true;
8647 }
8648
8649 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8650 // is a number of each value occurrences.
8651 Node* result = argument(0);
8652 if (false_cnt == 0 || true_cnt == 0) {
8653 // According to profile, one value has been never seen.
8654 int expected_val = (false_cnt == 0) ? 1 : 0;
8655
8656 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8657 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8658
8659 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8660 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8661 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8662
8663 { // Slow path: uncommon trap for never seen value and then reexecute
8664 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8665 // the value has been seen at least once.
8666 PreserveJVMState pjvms(this);
8667 PreserveReexecuteState preexecs(this);
8668 jvms()->set_should_reexecute(true);
8669
8670 set_control(slow_path);
8671 set_i_o(i_o());
8672
8673 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8674 Deoptimization::Action_reinterpret);
8675 }
8676 // The guard for never seen value enables sharpening of the result and
8677 // returning a constant. It allows to eliminate branches on the same value
8678 // later on.
8679 set_control(fast_path);
8680 result = intcon(expected_val);
8681 }
8682 // Stop profiling.
8683 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8684 // By replacing method body with profile data (represented as ProfileBooleanNode
8685 // on IR level) we effectively disable profiling.
8686 // It enables full speed execution once optimized code is generated.
8687 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8688 C->record_for_igvn(profile);
8689 set_result(profile);
8690 return true;
8691 } else {
8692 // Continue profiling.
8693 // Profile data isn't available at the moment. So, execute method's bytecode version.
8694 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8695 // is compiled and counters aren't available since corresponding MethodHandle
8696 // isn't a compile-time constant.
8697 return false;
8698 }
8699 }
8700
8701 bool LibraryCallKit::inline_isCompileConstant() {
8702 Node* n = argument(0);
8703 set_result(n->is_Con() ? intcon(1) : intcon(0));
8704 return true;
8705 }
8706
8707 //------------------------------- inline_getObjectSize --------------------------------------
8708 //
8709 // Calculate the runtime size of the object/array.
8710 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8711 //
8712 bool LibraryCallKit::inline_getObjectSize() {
8713 Node* obj = argument(3);
8714 Node* klass_node = load_object_klass(obj);
8715
8716 jint layout_con = Klass::_lh_neutral_value;
8717 Node* layout_val = get_layout_helper(klass_node, layout_con);
8718 int layout_is_con = (layout_val == nullptr);
8719
8720 if (layout_is_con) {
8721 // Layout helper is constant, can figure out things at compile time.
8722
8723 if (Klass::layout_helper_is_instance(layout_con)) {
8724 // Instance case: layout_con contains the size itself.
8725 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8726 set_result(size);
8727 } else {
8728 // Array case: size is round(header + element_size*arraylength).
8729 // Since arraylength is different for every array instance, we have to
8730 // compute the whole thing at runtime.
8731
8732 Node* arr_length = load_array_length(obj);
8733
8734 int round_mask = MinObjAlignmentInBytes - 1;
8735 int hsize = Klass::layout_helper_header_size(layout_con);
8736 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8737
8738 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8739 round_mask = 0; // strength-reduce it if it goes away completely
8740 }
8741 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8742 Node* header_size = intcon(hsize + round_mask);
8743
8744 Node* lengthx = ConvI2X(arr_length);
8745 Node* headerx = ConvI2X(header_size);
8746
8747 Node* abody = lengthx;
8748 if (eshift != 0) {
8749 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8750 }
8751 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8752 if (round_mask != 0) {
8753 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8754 }
8755 size = ConvX2L(size);
8756 set_result(size);
8757 }
8758 } else {
8759 // Layout helper is not constant, need to test for array-ness at runtime.
8760
8761 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8762 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8763 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8764 record_for_igvn(result_reg);
8765
8766 Node* array_ctl = generate_array_guard(klass_node, nullptr);
8767 if (array_ctl != nullptr) {
8768 // Array case: size is round(header + element_size*arraylength).
8769 // Since arraylength is different for every array instance, we have to
8770 // compute the whole thing at runtime.
8771
8772 PreserveJVMState pjvms(this);
8773 set_control(array_ctl);
8774 Node* arr_length = load_array_length(obj);
8775
8776 int round_mask = MinObjAlignmentInBytes - 1;
8777 Node* mask = intcon(round_mask);
8778
8779 Node* hss = intcon(Klass::_lh_header_size_shift);
8780 Node* hsm = intcon(Klass::_lh_header_size_mask);
8781 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8782 header_size = _gvn.transform(new AndINode(header_size, hsm));
8783 header_size = _gvn.transform(new AddINode(header_size, mask));
8784
8785 // There is no need to mask or shift this value.
8786 // The semantics of LShiftINode include an implicit mask to 0x1F.
8787 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8788 Node* elem_shift = layout_val;
8789
8790 Node* lengthx = ConvI2X(arr_length);
8791 Node* headerx = ConvI2X(header_size);
8792
8793 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
8794 Node* size = _gvn.transform(new AddXNode(headerx, abody));
8795 if (round_mask != 0) {
8796 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
8797 }
8798 size = ConvX2L(size);
8799
8800 result_reg->init_req(_array_path, control());
8801 result_val->init_req(_array_path, size);
8802 }
8803
8804 if (!stopped()) {
8805 // Instance case: the layout helper gives us instance size almost directly,
8806 // but we need to mask out the _lh_instance_slow_path_bit.
8807 Node* size = ConvI2X(layout_val);
8808 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
8809 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
8810 size = _gvn.transform(new AndXNode(size, mask));
8811 size = ConvX2L(size);
8812
8813 result_reg->init_req(_instance_path, control());
8814 result_val->init_req(_instance_path, size);
8815 }
8816
8817 set_result(result_reg, result_val);
8818 }
8819
8820 return true;
8821 }
8822
8823 //------------------------------- inline_blackhole --------------------------------------
8824 //
8825 // Make sure all arguments to this node are alive.
8826 // This matches methods that were requested to be blackholed through compile commands.
8827 //
8828 bool LibraryCallKit::inline_blackhole() {
8829 assert(callee()->is_static(), "Should have been checked before: only static methods here");
8830 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
8831 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
8832
8833 // Blackhole node pinches only the control, not memory. This allows
8834 // the blackhole to be pinned in the loop that computes blackholed
8835 // values, but have no other side effects, like breaking the optimizations
8836 // across the blackhole.
8837
8838 Node* bh = _gvn.transform(new BlackholeNode(control()));
8839 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
8840
8841 // Bind call arguments as blackhole arguments to keep them alive
8842 uint nargs = callee()->arg_size();
8843 for (uint i = 0; i < nargs; i++) {
8844 bh->add_req(argument(i));
8845 }
8846
8847 return true;
8848 }