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 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
491 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
492
493 #if INCLUDE_JVMTI
494 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
495 "notifyJvmtiStart", true, false);
496 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
497 "notifyJvmtiEnd", false, true);
498 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
499 "notifyJvmtiMount", false, false);
500 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
501 "notifyJvmtiUnmount", false, false);
502 case vmIntrinsics::_notifyJvmtiVThreadHideFrames: return inline_native_notify_jvmti_hide();
503 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
504 #endif
505
506 #ifdef JFR_HAVE_INTRINSICS
507 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
508 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
509 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
510 #endif
511 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
512 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
513 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
514 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
515 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
516 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
517 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
518 case vmIntrinsics::_isFlatArray: return inline_unsafe_isFlatArray();
519 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
520 case vmIntrinsics::_getLength: return inline_native_getLength();
521 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
522 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
523 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
524 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
525 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
526 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
527 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
528
529 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
530 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
531 case vmIntrinsics::_newNullRestrictedArray: return inline_newNullRestrictedArray();
532
533 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
534
535 case vmIntrinsics::_isInstance:
536 case vmIntrinsics::_getModifiers:
537 case vmIntrinsics::_isInterface:
538 case vmIntrinsics::_isArray:
539 case vmIntrinsics::_isPrimitive:
540 case vmIntrinsics::_isHidden:
541 case vmIntrinsics::_getSuperclass:
542 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
543
544 case vmIntrinsics::_floatToRawIntBits:
545 case vmIntrinsics::_floatToIntBits:
546 case vmIntrinsics::_intBitsToFloat:
547 case vmIntrinsics::_doubleToRawLongBits:
548 case vmIntrinsics::_doubleToLongBits:
549 case vmIntrinsics::_longBitsToDouble:
550 case vmIntrinsics::_floatToFloat16:
551 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
552
553 case vmIntrinsics::_floatIsFinite:
554 case vmIntrinsics::_floatIsInfinite:
555 case vmIntrinsics::_doubleIsFinite:
556 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
557
558 case vmIntrinsics::_numberOfLeadingZeros_i:
559 case vmIntrinsics::_numberOfLeadingZeros_l:
560 case vmIntrinsics::_numberOfTrailingZeros_i:
561 case vmIntrinsics::_numberOfTrailingZeros_l:
562 case vmIntrinsics::_bitCount_i:
563 case vmIntrinsics::_bitCount_l:
564 case vmIntrinsics::_reverse_i:
565 case vmIntrinsics::_reverse_l:
566 case vmIntrinsics::_reverseBytes_i:
567 case vmIntrinsics::_reverseBytes_l:
568 case vmIntrinsics::_reverseBytes_s:
569 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
570
571 case vmIntrinsics::_compress_i:
572 case vmIntrinsics::_compress_l:
573 case vmIntrinsics::_expand_i:
574 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
575
576 case vmIntrinsics::_compareUnsigned_i:
577 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
578
579 case vmIntrinsics::_divideUnsigned_i:
580 case vmIntrinsics::_divideUnsigned_l:
581 case vmIntrinsics::_remainderUnsigned_i:
582 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
583
584 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
585
586 case vmIntrinsics::_Reference_get: return inline_reference_get();
587 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
588 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
589
590 case vmIntrinsics::_Class_cast: return inline_Class_cast();
591
592 case vmIntrinsics::_aescrypt_encryptBlock:
593 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
594
595 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
596 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
597 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
598
599 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
600 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
601 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
602
603 case vmIntrinsics::_counterMode_AESCrypt:
604 return inline_counterMode_AESCrypt(intrinsic_id());
605
606 case vmIntrinsics::_galoisCounterMode_AESCrypt:
607 return inline_galoisCounterMode_AESCrypt();
608
609 case vmIntrinsics::_md5_implCompress:
610 case vmIntrinsics::_sha_implCompress:
611 case vmIntrinsics::_sha2_implCompress:
612 case vmIntrinsics::_sha5_implCompress:
613 case vmIntrinsics::_sha3_implCompress:
614 return inline_digestBase_implCompress(intrinsic_id());
615
616 case vmIntrinsics::_digestBase_implCompressMB:
617 return inline_digestBase_implCompressMB(predicate);
618
619 case vmIntrinsics::_multiplyToLen:
620 return inline_multiplyToLen();
621
622 case vmIntrinsics::_squareToLen:
623 return inline_squareToLen();
624
625 case vmIntrinsics::_mulAdd:
626 return inline_mulAdd();
627
628 case vmIntrinsics::_montgomeryMultiply:
629 return inline_montgomeryMultiply();
630 case vmIntrinsics::_montgomerySquare:
631 return inline_montgomerySquare();
632
633 case vmIntrinsics::_bigIntegerRightShiftWorker:
634 return inline_bigIntegerShift(true);
635 case vmIntrinsics::_bigIntegerLeftShiftWorker:
636 return inline_bigIntegerShift(false);
637
638 case vmIntrinsics::_vectorizedMismatch:
639 return inline_vectorizedMismatch();
640
641 case vmIntrinsics::_ghash_processBlocks:
642 return inline_ghash_processBlocks();
643 case vmIntrinsics::_chacha20Block:
644 return inline_chacha20Block();
645 case vmIntrinsics::_base64_encodeBlock:
646 return inline_base64_encodeBlock();
647 case vmIntrinsics::_base64_decodeBlock:
648 return inline_base64_decodeBlock();
649 case vmIntrinsics::_poly1305_processBlocks:
650 return inline_poly1305_processBlocks();
651 case vmIntrinsics::_intpoly_montgomeryMult_P256:
652 return inline_intpoly_montgomeryMult_P256();
653 case vmIntrinsics::_intpoly_assign:
654 return inline_intpoly_assign();
655 case vmIntrinsics::_encodeISOArray:
656 case vmIntrinsics::_encodeByteISOArray:
657 return inline_encodeISOArray(false);
658 case vmIntrinsics::_encodeAsciiArray:
659 return inline_encodeISOArray(true);
660
661 case vmIntrinsics::_updateCRC32:
662 return inline_updateCRC32();
663 case vmIntrinsics::_updateBytesCRC32:
664 return inline_updateBytesCRC32();
665 case vmIntrinsics::_updateByteBufferCRC32:
666 return inline_updateByteBufferCRC32();
667
668 case vmIntrinsics::_updateBytesCRC32C:
669 return inline_updateBytesCRC32C();
670 case vmIntrinsics::_updateDirectByteBufferCRC32C:
671 return inline_updateDirectByteBufferCRC32C();
672
673 case vmIntrinsics::_updateBytesAdler32:
674 return inline_updateBytesAdler32();
675 case vmIntrinsics::_updateByteBufferAdler32:
676 return inline_updateByteBufferAdler32();
677
678 case vmIntrinsics::_profileBoolean:
679 return inline_profileBoolean();
680 case vmIntrinsics::_isCompileConstant:
681 return inline_isCompileConstant();
682
683 case vmIntrinsics::_countPositives:
684 return inline_countPositives();
685
686 case vmIntrinsics::_fmaD:
687 case vmIntrinsics::_fmaF:
688 return inline_fma(intrinsic_id());
689
690 case vmIntrinsics::_isDigit:
691 case vmIntrinsics::_isLowerCase:
692 case vmIntrinsics::_isUpperCase:
693 case vmIntrinsics::_isWhitespace:
694 return inline_character_compare(intrinsic_id());
695
696 case vmIntrinsics::_min:
697 case vmIntrinsics::_max:
698 case vmIntrinsics::_min_strict:
699 case vmIntrinsics::_max_strict:
700 return inline_min_max(intrinsic_id());
701
702 case vmIntrinsics::_maxF:
703 case vmIntrinsics::_minF:
704 case vmIntrinsics::_maxD:
705 case vmIntrinsics::_minD:
706 case vmIntrinsics::_maxF_strict:
707 case vmIntrinsics::_minF_strict:
708 case vmIntrinsics::_maxD_strict:
709 case vmIntrinsics::_minD_strict:
710 return inline_fp_min_max(intrinsic_id());
711
712 case vmIntrinsics::_VectorUnaryOp:
713 return inline_vector_nary_operation(1);
714 case vmIntrinsics::_VectorBinaryOp:
715 return inline_vector_nary_operation(2);
716 case vmIntrinsics::_VectorTernaryOp:
717 return inline_vector_nary_operation(3);
718 case vmIntrinsics::_VectorFromBitsCoerced:
719 return inline_vector_frombits_coerced();
720 case vmIntrinsics::_VectorShuffleIota:
721 return inline_vector_shuffle_iota();
722 case vmIntrinsics::_VectorMaskOp:
723 return inline_vector_mask_operation();
724 case vmIntrinsics::_VectorShuffleToVector:
725 return inline_vector_shuffle_to_vector();
726 case vmIntrinsics::_VectorLoadOp:
727 return inline_vector_mem_operation(/*is_store=*/false);
728 case vmIntrinsics::_VectorLoadMaskedOp:
729 return inline_vector_mem_masked_operation(/*is_store*/false);
730 case vmIntrinsics::_VectorStoreOp:
731 return inline_vector_mem_operation(/*is_store=*/true);
732 case vmIntrinsics::_VectorStoreMaskedOp:
733 return inline_vector_mem_masked_operation(/*is_store=*/true);
734 case vmIntrinsics::_VectorGatherOp:
735 return inline_vector_gather_scatter(/*is_scatter*/ false);
736 case vmIntrinsics::_VectorScatterOp:
737 return inline_vector_gather_scatter(/*is_scatter*/ true);
738 case vmIntrinsics::_VectorReductionCoerced:
739 return inline_vector_reduction();
740 case vmIntrinsics::_VectorTest:
741 return inline_vector_test();
742 case vmIntrinsics::_VectorBlend:
743 return inline_vector_blend();
744 case vmIntrinsics::_VectorRearrange:
745 return inline_vector_rearrange();
746 case vmIntrinsics::_VectorCompare:
747 return inline_vector_compare();
748 case vmIntrinsics::_VectorBroadcastInt:
749 return inline_vector_broadcast_int();
750 case vmIntrinsics::_VectorConvert:
751 return inline_vector_convert();
752 case vmIntrinsics::_VectorInsert:
753 return inline_vector_insert();
754 case vmIntrinsics::_VectorExtract:
755 return inline_vector_extract();
756 case vmIntrinsics::_VectorCompressExpand:
757 return inline_vector_compress_expand();
758 case vmIntrinsics::_IndexVector:
759 return inline_index_vector();
760 case vmIntrinsics::_IndexPartiallyInUpperRange:
761 return inline_index_partially_in_upper_range();
762
763 case vmIntrinsics::_getObjectSize:
764 return inline_getObjectSize();
765
766 case vmIntrinsics::_blackhole:
767 return inline_blackhole();
768
769 default:
770 // If you get here, it may be that someone has added a new intrinsic
771 // to the list in vmIntrinsics.hpp without implementing it here.
772 #ifndef PRODUCT
773 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
774 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
775 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
776 }
777 #endif
778 return false;
779 }
780 }
781
782 Node* LibraryCallKit::try_to_predicate(int predicate) {
783 if (!jvms()->has_method()) {
784 // Root JVMState has a null method.
785 assert(map()->memory()->Opcode() == Op_Parm, "");
786 // Insert the memory aliasing node
787 set_all_memory(reset_memory());
788 }
789 assert(merged_memory(), "");
790
791 switch (intrinsic_id()) {
792 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
793 return inline_cipherBlockChaining_AESCrypt_predicate(false);
794 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
795 return inline_cipherBlockChaining_AESCrypt_predicate(true);
796 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
797 return inline_electronicCodeBook_AESCrypt_predicate(false);
798 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
799 return inline_electronicCodeBook_AESCrypt_predicate(true);
800 case vmIntrinsics::_counterMode_AESCrypt:
801 return inline_counterMode_AESCrypt_predicate();
802 case vmIntrinsics::_digestBase_implCompressMB:
803 return inline_digestBase_implCompressMB_predicate(predicate);
804 case vmIntrinsics::_galoisCounterMode_AESCrypt:
805 return inline_galoisCounterMode_AESCrypt_predicate();
806
807 default:
808 // If you get here, it may be that someone has added a new intrinsic
809 // to the list in vmIntrinsics.hpp without implementing it here.
810 #ifndef PRODUCT
811 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
812 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
813 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
814 }
815 #endif
816 Node* slow_ctl = control();
817 set_control(top()); // No fast path intrinsic
818 return slow_ctl;
819 }
820 }
821
822 //------------------------------set_result-------------------------------
823 // Helper function for finishing intrinsics.
824 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
825 record_for_igvn(region);
826 set_control(_gvn.transform(region));
827 set_result( _gvn.transform(value));
828 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
829 }
830
831 //------------------------------generate_guard---------------------------
832 // Helper function for generating guarded fast-slow graph structures.
833 // The given 'test', if true, guards a slow path. If the test fails
834 // then a fast path can be taken. (We generally hope it fails.)
835 // In all cases, GraphKit::control() is updated to the fast path.
836 // The returned value represents the control for the slow path.
837 // The return value is never 'top'; it is either a valid control
838 // or null if it is obvious that the slow path can never be taken.
839 // Also, if region and the slow control are not null, the slow edge
840 // is appended to the region.
841 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
842 if (stopped()) {
843 // Already short circuited.
844 return nullptr;
845 }
846
847 // Build an if node and its projections.
848 // If test is true we take the slow path, which we assume is uncommon.
849 if (_gvn.type(test) == TypeInt::ZERO) {
850 // The slow branch is never taken. No need to build this guard.
851 return nullptr;
852 }
853
854 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
855
856 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
857 if (if_slow == top()) {
858 // The slow branch is never taken. No need to build this guard.
859 return nullptr;
860 }
861
862 if (region != nullptr)
863 region->add_req(if_slow);
864
865 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
866 set_control(if_fast);
867
868 return if_slow;
869 }
870
871 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
872 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
873 }
874 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
875 return generate_guard(test, region, PROB_FAIR);
876 }
877
878 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
879 Node* *pos_index) {
880 if (stopped())
881 return nullptr; // already stopped
882 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
883 return nullptr; // index is already adequately typed
884 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
885 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
886 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
887 if (is_neg != nullptr && pos_index != nullptr) {
888 // Emulate effect of Parse::adjust_map_after_if.
889 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
890 (*pos_index) = _gvn.transform(ccast);
891 }
892 return is_neg;
893 }
894
895 // Make sure that 'position' is a valid limit index, in [0..length].
896 // There are two equivalent plans for checking this:
897 // A. (offset + copyLength) unsigned<= arrayLength
898 // B. offset <= (arrayLength - copyLength)
899 // We require that all of the values above, except for the sum and
900 // difference, are already known to be non-negative.
901 // Plan A is robust in the face of overflow, if offset and copyLength
902 // are both hugely positive.
903 //
904 // Plan B is less direct and intuitive, but it does not overflow at
905 // all, since the difference of two non-negatives is always
906 // representable. Whenever Java methods must perform the equivalent
907 // check they generally use Plan B instead of Plan A.
908 // For the moment we use Plan A.
909 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
910 Node* subseq_length,
911 Node* array_length,
912 RegionNode* region) {
913 if (stopped())
914 return nullptr; // already stopped
915 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
916 if (zero_offset && subseq_length->eqv_uncast(array_length))
917 return nullptr; // common case of whole-array copy
918 Node* last = subseq_length;
919 if (!zero_offset) // last += offset
920 last = _gvn.transform(new AddINode(last, offset));
921 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
922 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
923 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
924 return is_over;
925 }
926
927 // Emit range checks for the given String.value byte array
928 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
929 if (stopped()) {
930 return; // already stopped
931 }
932 RegionNode* bailout = new RegionNode(1);
933 record_for_igvn(bailout);
934 if (char_count) {
935 // Convert char count to byte count
936 count = _gvn.transform(new LShiftINode(count, intcon(1)));
937 }
938
939 // Offset and count must not be negative
940 generate_negative_guard(offset, bailout);
941 generate_negative_guard(count, bailout);
942 // Offset + count must not exceed length of array
943 generate_limit_guard(offset, count, load_array_length(array), bailout);
944
945 if (bailout->req() > 1) {
946 PreserveJVMState pjvms(this);
947 set_control(_gvn.transform(bailout));
948 uncommon_trap(Deoptimization::Reason_intrinsic,
949 Deoptimization::Action_maybe_recompile);
950 }
951 }
952
953 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
954 bool is_immutable) {
955 ciKlass* thread_klass = env()->Thread_klass();
956 const Type* thread_type
957 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
958
959 Node* thread = _gvn.transform(new ThreadLocalNode());
960 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
961 tls_output = thread;
962
963 Node* thread_obj_handle
964 = (is_immutable
965 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
966 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
967 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
968 thread_obj_handle = _gvn.transform(thread_obj_handle);
969
970 DecoratorSet decorators = IN_NATIVE;
971 if (is_immutable) {
972 decorators |= C2_IMMUTABLE_MEMORY;
973 }
974 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
975 }
976
977 //--------------------------generate_current_thread--------------------
978 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
979 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
980 /*is_immutable*/false);
981 }
982
983 //--------------------------generate_virtual_thread--------------------
984 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
985 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
986 !C->method()->changes_current_thread());
987 }
988
989 //------------------------------make_string_method_node------------------------
990 // Helper method for String intrinsic functions. This version is called with
991 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
992 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
993 // containing the lengths of str1 and str2.
994 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
995 Node* result = nullptr;
996 switch (opcode) {
997 case Op_StrIndexOf:
998 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
999 str1_start, cnt1, str2_start, cnt2, ae);
1000 break;
1001 case Op_StrComp:
1002 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1003 str1_start, cnt1, str2_start, cnt2, ae);
1004 break;
1005 case Op_StrEquals:
1006 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1007 // Use the constant length if there is one because optimized match rule may exist.
1008 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1009 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1010 break;
1011 default:
1012 ShouldNotReachHere();
1013 return nullptr;
1014 }
1015
1016 // All these intrinsics have checks.
1017 C->set_has_split_ifs(true); // Has chance for split-if optimization
1018 clear_upper_avx();
1019
1020 return _gvn.transform(result);
1021 }
1022
1023 //------------------------------inline_string_compareTo------------------------
1024 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1025 Node* arg1 = argument(0);
1026 Node* arg2 = argument(1);
1027
1028 arg1 = must_be_not_null(arg1, true);
1029 arg2 = must_be_not_null(arg2, true);
1030
1031 // Get start addr and length of first argument
1032 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1033 Node* arg1_cnt = load_array_length(arg1);
1034
1035 // Get start addr and length of second argument
1036 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1037 Node* arg2_cnt = load_array_length(arg2);
1038
1039 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1040 set_result(result);
1041 return true;
1042 }
1043
1044 //------------------------------inline_string_equals------------------------
1045 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1046 Node* arg1 = argument(0);
1047 Node* arg2 = argument(1);
1048
1049 // paths (plus control) merge
1050 RegionNode* region = new RegionNode(3);
1051 Node* phi = new PhiNode(region, TypeInt::BOOL);
1052
1053 if (!stopped()) {
1054
1055 arg1 = must_be_not_null(arg1, true);
1056 arg2 = must_be_not_null(arg2, true);
1057
1058 // Get start addr and length of first argument
1059 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1060 Node* arg1_cnt = load_array_length(arg1);
1061
1062 // Get start addr and length of second argument
1063 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1064 Node* arg2_cnt = load_array_length(arg2);
1065
1066 // Check for arg1_cnt != arg2_cnt
1067 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1068 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1069 Node* if_ne = generate_slow_guard(bol, nullptr);
1070 if (if_ne != nullptr) {
1071 phi->init_req(2, intcon(0));
1072 region->init_req(2, if_ne);
1073 }
1074
1075 // Check for count == 0 is done by assembler code for StrEquals.
1076
1077 if (!stopped()) {
1078 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1079 phi->init_req(1, equals);
1080 region->init_req(1, control());
1081 }
1082 }
1083
1084 // post merge
1085 set_control(_gvn.transform(region));
1086 record_for_igvn(region);
1087
1088 set_result(_gvn.transform(phi));
1089 return true;
1090 }
1091
1092 //------------------------------inline_array_equals----------------------------
1093 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1094 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1095 Node* arg1 = argument(0);
1096 Node* arg2 = argument(1);
1097
1098 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1099 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1100 clear_upper_avx();
1101
1102 return true;
1103 }
1104
1105
1106 //------------------------------inline_countPositives------------------------------
1107 bool LibraryCallKit::inline_countPositives() {
1108 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1109 return false;
1110 }
1111
1112 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1113 // no receiver since it is static method
1114 Node* ba = argument(0);
1115 Node* offset = argument(1);
1116 Node* len = argument(2);
1117
1118 ba = must_be_not_null(ba, true);
1119
1120 // Range checks
1121 generate_string_range_check(ba, offset, len, false);
1122 if (stopped()) {
1123 return true;
1124 }
1125 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1126 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1127 set_result(_gvn.transform(result));
1128 clear_upper_avx();
1129 return true;
1130 }
1131
1132 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1133 Node* index = argument(0);
1134 Node* length = bt == T_INT ? argument(1) : argument(2);
1135 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1136 return false;
1137 }
1138
1139 // check that length is positive
1140 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1141 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1142
1143 {
1144 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1145 uncommon_trap(Deoptimization::Reason_intrinsic,
1146 Deoptimization::Action_make_not_entrant);
1147 }
1148
1149 if (stopped()) {
1150 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1151 return true;
1152 }
1153
1154 // length is now known positive, add a cast node to make this explicit
1155 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1156 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1157 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1158 ConstraintCastNode::RegularDependency, bt);
1159 casted_length = _gvn.transform(casted_length);
1160 replace_in_map(length, casted_length);
1161 length = casted_length;
1162
1163 // Use an unsigned comparison for the range check itself
1164 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1165 BoolTest::mask btest = BoolTest::lt;
1166 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1167 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1168 _gvn.set_type(rc, rc->Value(&_gvn));
1169 if (!rc_bool->is_Con()) {
1170 record_for_igvn(rc);
1171 }
1172 set_control(_gvn.transform(new IfTrueNode(rc)));
1173 {
1174 PreserveJVMState pjvms(this);
1175 set_control(_gvn.transform(new IfFalseNode(rc)));
1176 uncommon_trap(Deoptimization::Reason_range_check,
1177 Deoptimization::Action_make_not_entrant);
1178 }
1179
1180 if (stopped()) {
1181 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1182 return true;
1183 }
1184
1185 // index is now known to be >= 0 and < length, cast it
1186 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1187 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1188 ConstraintCastNode::RegularDependency, bt);
1189 result = _gvn.transform(result);
1190 set_result(result);
1191 replace_in_map(index, result);
1192 return true;
1193 }
1194
1195 //------------------------------inline_string_indexOf------------------------
1196 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1197 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1198 return false;
1199 }
1200 Node* src = argument(0);
1201 Node* tgt = argument(1);
1202
1203 // Make the merge point
1204 RegionNode* result_rgn = new RegionNode(4);
1205 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1206
1207 src = must_be_not_null(src, true);
1208 tgt = must_be_not_null(tgt, true);
1209
1210 // Get start addr and length of source string
1211 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1212 Node* src_count = load_array_length(src);
1213
1214 // Get start addr and length of substring
1215 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1216 Node* tgt_count = load_array_length(tgt);
1217
1218 Node* result = nullptr;
1219 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1220
1221 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1222 // Divide src size by 2 if String is UTF16 encoded
1223 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1224 }
1225 if (ae == StrIntrinsicNode::UU) {
1226 // Divide substring size by 2 if String is UTF16 encoded
1227 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1228 }
1229
1230 if (call_opt_stub) {
1231 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1232 StubRoutines::_string_indexof_array[ae],
1233 "stringIndexOf", TypePtr::BOTTOM, src_start,
1234 src_count, tgt_start, tgt_count);
1235 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1236 } else {
1237 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1238 result_rgn, result_phi, ae);
1239 }
1240 if (result != nullptr) {
1241 result_phi->init_req(3, result);
1242 result_rgn->init_req(3, control());
1243 }
1244 set_control(_gvn.transform(result_rgn));
1245 record_for_igvn(result_rgn);
1246 set_result(_gvn.transform(result_phi));
1247
1248 return true;
1249 }
1250
1251 //-----------------------------inline_string_indexOfI-----------------------
1252 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1253 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1254 return false;
1255 }
1256 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1257 return false;
1258 }
1259
1260 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1261 Node* src = argument(0); // byte[]
1262 Node* src_count = argument(1); // char count
1263 Node* tgt = argument(2); // byte[]
1264 Node* tgt_count = argument(3); // char count
1265 Node* from_index = argument(4); // char index
1266
1267 src = must_be_not_null(src, true);
1268 tgt = must_be_not_null(tgt, true);
1269
1270 // Multiply byte array index by 2 if String is UTF16 encoded
1271 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1272 src_count = _gvn.transform(new SubINode(src_count, from_index));
1273 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1274 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1275
1276 // Range checks
1277 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1278 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1279 if (stopped()) {
1280 return true;
1281 }
1282
1283 RegionNode* region = new RegionNode(5);
1284 Node* phi = new PhiNode(region, TypeInt::INT);
1285 Node* result = nullptr;
1286
1287 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1288
1289 if (call_opt_stub) {
1290 assert(arrayOopDesc::base_offset_in_bytes(T_BYTE) >= 16, "Needed for indexOf");
1291 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1292 StubRoutines::_string_indexof_array[ae],
1293 "stringIndexOf", TypePtr::BOTTOM, src_start,
1294 src_count, tgt_start, tgt_count);
1295 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1296 } else {
1297 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1298 region, phi, ae);
1299 }
1300 if (result != nullptr) {
1301 // The result is index relative to from_index if substring was found, -1 otherwise.
1302 // Generate code which will fold into cmove.
1303 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1304 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1305
1306 Node* if_lt = generate_slow_guard(bol, nullptr);
1307 if (if_lt != nullptr) {
1308 // result == -1
1309 phi->init_req(3, result);
1310 region->init_req(3, if_lt);
1311 }
1312 if (!stopped()) {
1313 result = _gvn.transform(new AddINode(result, from_index));
1314 phi->init_req(4, result);
1315 region->init_req(4, control());
1316 }
1317 }
1318
1319 set_control(_gvn.transform(region));
1320 record_for_igvn(region);
1321 set_result(_gvn.transform(phi));
1322 clear_upper_avx();
1323
1324 return true;
1325 }
1326
1327 // Create StrIndexOfNode with fast path checks
1328 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1329 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1330 // Check for substr count > string count
1331 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1332 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1333 Node* if_gt = generate_slow_guard(bol, nullptr);
1334 if (if_gt != nullptr) {
1335 phi->init_req(1, intcon(-1));
1336 region->init_req(1, if_gt);
1337 }
1338 if (!stopped()) {
1339 // Check for substr count == 0
1340 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1341 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1342 Node* if_zero = generate_slow_guard(bol, nullptr);
1343 if (if_zero != nullptr) {
1344 phi->init_req(2, intcon(0));
1345 region->init_req(2, if_zero);
1346 }
1347 }
1348 if (!stopped()) {
1349 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1350 }
1351 return nullptr;
1352 }
1353
1354 //-----------------------------inline_string_indexOfChar-----------------------
1355 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1356 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1357 return false;
1358 }
1359 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1360 return false;
1361 }
1362 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1363 Node* src = argument(0); // byte[]
1364 Node* int_ch = argument(1);
1365 Node* from_index = argument(2);
1366 Node* max = argument(3);
1367
1368 src = must_be_not_null(src, true);
1369
1370 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1371 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1372 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1373
1374 // Range checks
1375 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1376
1377 // Check for int_ch >= 0
1378 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1379 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1380 {
1381 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1382 uncommon_trap(Deoptimization::Reason_intrinsic,
1383 Deoptimization::Action_maybe_recompile);
1384 }
1385 if (stopped()) {
1386 return true;
1387 }
1388
1389 RegionNode* region = new RegionNode(3);
1390 Node* phi = new PhiNode(region, TypeInt::INT);
1391
1392 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1393 C->set_has_split_ifs(true); // Has chance for split-if optimization
1394 _gvn.transform(result);
1395
1396 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1397 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1398
1399 Node* if_lt = generate_slow_guard(bol, nullptr);
1400 if (if_lt != nullptr) {
1401 // result == -1
1402 phi->init_req(2, result);
1403 region->init_req(2, if_lt);
1404 }
1405 if (!stopped()) {
1406 result = _gvn.transform(new AddINode(result, from_index));
1407 phi->init_req(1, result);
1408 region->init_req(1, control());
1409 }
1410 set_control(_gvn.transform(region));
1411 record_for_igvn(region);
1412 set_result(_gvn.transform(phi));
1413 clear_upper_avx();
1414
1415 return true;
1416 }
1417 //---------------------------inline_string_copy---------------------
1418 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1419 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1420 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1421 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1422 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1423 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1424 bool LibraryCallKit::inline_string_copy(bool compress) {
1425 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1426 return false;
1427 }
1428 int nargs = 5; // 2 oops, 3 ints
1429 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1430
1431 Node* src = argument(0);
1432 Node* src_offset = argument(1);
1433 Node* dst = argument(2);
1434 Node* dst_offset = argument(3);
1435 Node* length = argument(4);
1436
1437 // Check for allocation before we add nodes that would confuse
1438 // tightly_coupled_allocation()
1439 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1440
1441 // Figure out the size and type of the elements we will be copying.
1442 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1443 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1444 if (src_type == nullptr || dst_type == nullptr) {
1445 return false;
1446 }
1447 BasicType src_elem = src_type->elem()->array_element_basic_type();
1448 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1449 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1450 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1451 "Unsupported array types for inline_string_copy");
1452
1453 src = must_be_not_null(src, true);
1454 dst = must_be_not_null(dst, true);
1455
1456 // Convert char[] offsets to byte[] offsets
1457 bool convert_src = (compress && src_elem == T_BYTE);
1458 bool convert_dst = (!compress && dst_elem == T_BYTE);
1459 if (convert_src) {
1460 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1461 } else if (convert_dst) {
1462 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1463 }
1464
1465 // Range checks
1466 generate_string_range_check(src, src_offset, length, convert_src);
1467 generate_string_range_check(dst, dst_offset, length, convert_dst);
1468 if (stopped()) {
1469 return true;
1470 }
1471
1472 Node* src_start = array_element_address(src, src_offset, src_elem);
1473 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1474 // 'src_start' points to src array + scaled offset
1475 // 'dst_start' points to dst array + scaled offset
1476 Node* count = nullptr;
1477 if (compress) {
1478 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1479 } else {
1480 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1481 }
1482
1483 if (alloc != nullptr) {
1484 if (alloc->maybe_set_complete(&_gvn)) {
1485 // "You break it, you buy it."
1486 InitializeNode* init = alloc->initialization();
1487 assert(init->is_complete(), "we just did this");
1488 init->set_complete_with_arraycopy();
1489 assert(dst->is_CheckCastPP(), "sanity");
1490 assert(dst->in(0)->in(0) == init, "dest pinned");
1491 }
1492 // Do not let stores that initialize this object be reordered with
1493 // a subsequent store that would make this object accessible by
1494 // other threads.
1495 // Record what AllocateNode this StoreStore protects so that
1496 // escape analysis can go from the MemBarStoreStoreNode to the
1497 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1498 // based on the escape status of the AllocateNode.
1499 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1500 }
1501 if (compress) {
1502 set_result(_gvn.transform(count));
1503 }
1504 clear_upper_avx();
1505
1506 return true;
1507 }
1508
1509 #ifdef _LP64
1510 #define XTOP ,top() /*additional argument*/
1511 #else //_LP64
1512 #define XTOP /*no additional argument*/
1513 #endif //_LP64
1514
1515 //------------------------inline_string_toBytesU--------------------------
1516 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1517 bool LibraryCallKit::inline_string_toBytesU() {
1518 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1519 return false;
1520 }
1521 // Get the arguments.
1522 Node* value = argument(0);
1523 Node* offset = argument(1);
1524 Node* length = argument(2);
1525
1526 Node* newcopy = nullptr;
1527
1528 // Set the original stack and the reexecute bit for the interpreter to reexecute
1529 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1530 { PreserveReexecuteState preexecs(this);
1531 jvms()->set_should_reexecute(true);
1532
1533 // Check if a null path was taken unconditionally.
1534 value = null_check(value);
1535
1536 RegionNode* bailout = new RegionNode(1);
1537 record_for_igvn(bailout);
1538
1539 // Range checks
1540 generate_negative_guard(offset, bailout);
1541 generate_negative_guard(length, bailout);
1542 generate_limit_guard(offset, length, load_array_length(value), bailout);
1543 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1544 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1545
1546 if (bailout->req() > 1) {
1547 PreserveJVMState pjvms(this);
1548 set_control(_gvn.transform(bailout));
1549 uncommon_trap(Deoptimization::Reason_intrinsic,
1550 Deoptimization::Action_maybe_recompile);
1551 }
1552 if (stopped()) {
1553 return true;
1554 }
1555
1556 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1557 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1558 newcopy = new_array(klass_node, size, 0); // no arguments to push
1559 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1560 guarantee(alloc != nullptr, "created above");
1561
1562 // Calculate starting addresses.
1563 Node* src_start = array_element_address(value, offset, T_CHAR);
1564 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1565
1566 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1567 const TypeInt* toffset = gvn().type(offset)->is_int();
1568 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1569
1570 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1571 const char* copyfunc_name = "arraycopy";
1572 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1573 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1574 OptoRuntime::fast_arraycopy_Type(),
1575 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1576 src_start, dst_start, ConvI2X(length) XTOP);
1577 // Do not let reads from the cloned object float above the arraycopy.
1578 if (alloc->maybe_set_complete(&_gvn)) {
1579 // "You break it, you buy it."
1580 InitializeNode* init = alloc->initialization();
1581 assert(init->is_complete(), "we just did this");
1582 init->set_complete_with_arraycopy();
1583 assert(newcopy->is_CheckCastPP(), "sanity");
1584 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1585 }
1586 // Do not let stores that initialize this object be reordered with
1587 // a subsequent store that would make this object accessible by
1588 // other threads.
1589 // Record what AllocateNode this StoreStore protects so that
1590 // escape analysis can go from the MemBarStoreStoreNode to the
1591 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1592 // based on the escape status of the AllocateNode.
1593 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1594 } // original reexecute is set back here
1595
1596 C->set_has_split_ifs(true); // Has chance for split-if optimization
1597 if (!stopped()) {
1598 set_result(newcopy);
1599 }
1600 clear_upper_avx();
1601
1602 return true;
1603 }
1604
1605 //------------------------inline_string_getCharsU--------------------------
1606 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1607 bool LibraryCallKit::inline_string_getCharsU() {
1608 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1609 return false;
1610 }
1611
1612 // Get the arguments.
1613 Node* src = argument(0);
1614 Node* src_begin = argument(1);
1615 Node* src_end = argument(2); // exclusive offset (i < src_end)
1616 Node* dst = argument(3);
1617 Node* dst_begin = argument(4);
1618
1619 // Check for allocation before we add nodes that would confuse
1620 // tightly_coupled_allocation()
1621 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1622
1623 // Check if a null path was taken unconditionally.
1624 src = null_check(src);
1625 dst = null_check(dst);
1626 if (stopped()) {
1627 return true;
1628 }
1629
1630 // Get length and convert char[] offset to byte[] offset
1631 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1632 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1633
1634 // Range checks
1635 generate_string_range_check(src, src_begin, length, true);
1636 generate_string_range_check(dst, dst_begin, length, false);
1637 if (stopped()) {
1638 return true;
1639 }
1640
1641 if (!stopped()) {
1642 // Calculate starting addresses.
1643 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1644 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1645
1646 // Check if array addresses are aligned to HeapWordSize
1647 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1648 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1649 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1650 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1651
1652 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1653 const char* copyfunc_name = "arraycopy";
1654 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1655 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1656 OptoRuntime::fast_arraycopy_Type(),
1657 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1658 src_start, dst_start, ConvI2X(length) XTOP);
1659 // Do not let reads from the cloned object float above the arraycopy.
1660 if (alloc != nullptr) {
1661 if (alloc->maybe_set_complete(&_gvn)) {
1662 // "You break it, you buy it."
1663 InitializeNode* init = alloc->initialization();
1664 assert(init->is_complete(), "we just did this");
1665 init->set_complete_with_arraycopy();
1666 assert(dst->is_CheckCastPP(), "sanity");
1667 assert(dst->in(0)->in(0) == init, "dest pinned");
1668 }
1669 // Do not let stores that initialize this object be reordered with
1670 // a subsequent store that would make this object accessible by
1671 // other threads.
1672 // Record what AllocateNode this StoreStore protects so that
1673 // escape analysis can go from the MemBarStoreStoreNode to the
1674 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1675 // based on the escape status of the AllocateNode.
1676 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1677 } else {
1678 insert_mem_bar(Op_MemBarCPUOrder);
1679 }
1680 }
1681
1682 C->set_has_split_ifs(true); // Has chance for split-if optimization
1683 return true;
1684 }
1685
1686 //----------------------inline_string_char_access----------------------------
1687 // Store/Load char to/from byte[] array.
1688 // static void StringUTF16.putChar(byte[] val, int index, int c)
1689 // static char StringUTF16.getChar(byte[] val, int index)
1690 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1691 Node* value = argument(0);
1692 Node* index = argument(1);
1693 Node* ch = is_store ? argument(2) : nullptr;
1694
1695 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1696 // correctly requires matched array shapes.
1697 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1698 "sanity: byte[] and char[] bases agree");
1699 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1700 "sanity: byte[] and char[] scales agree");
1701
1702 // Bail when getChar over constants is requested: constant folding would
1703 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1704 // Java method would constant fold nicely instead.
1705 if (!is_store && value->is_Con() && index->is_Con()) {
1706 return false;
1707 }
1708
1709 // Save state and restore on bailout
1710 uint old_sp = sp();
1711 SafePointNode* old_map = clone_map();
1712
1713 value = must_be_not_null(value, true);
1714
1715 Node* adr = array_element_address(value, index, T_CHAR);
1716 if (adr->is_top()) {
1717 set_map(old_map);
1718 set_sp(old_sp);
1719 return false;
1720 }
1721 destruct_map_clone(old_map);
1722 if (is_store) {
1723 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1724 } else {
1725 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);
1726 set_result(ch);
1727 }
1728 return true;
1729 }
1730
1731 //--------------------------round_double_node--------------------------------
1732 // Round a double node if necessary.
1733 Node* LibraryCallKit::round_double_node(Node* n) {
1734 if (Matcher::strict_fp_requires_explicit_rounding) {
1735 #ifdef IA32
1736 if (UseSSE < 2) {
1737 n = _gvn.transform(new RoundDoubleNode(nullptr, n));
1738 }
1739 #else
1740 Unimplemented();
1741 #endif // IA32
1742 }
1743 return n;
1744 }
1745
1746 //------------------------------inline_math-----------------------------------
1747 // public static double Math.abs(double)
1748 // public static double Math.sqrt(double)
1749 // public static double Math.log(double)
1750 // public static double Math.log10(double)
1751 // public static double Math.round(double)
1752 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1753 Node* arg = round_double_node(argument(0));
1754 Node* n = nullptr;
1755 switch (id) {
1756 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1757 case vmIntrinsics::_dsqrt:
1758 case vmIntrinsics::_dsqrt_strict:
1759 n = new SqrtDNode(C, control(), arg); break;
1760 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1761 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1762 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1763 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1764 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1765 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1766 default: fatal_unexpected_iid(id); break;
1767 }
1768 set_result(_gvn.transform(n));
1769 return true;
1770 }
1771
1772 //------------------------------inline_math-----------------------------------
1773 // public static float Math.abs(float)
1774 // public static int Math.abs(int)
1775 // public static long Math.abs(long)
1776 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1777 Node* arg = argument(0);
1778 Node* n = nullptr;
1779 switch (id) {
1780 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1781 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1782 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1783 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1784 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1785 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1786 default: fatal_unexpected_iid(id); break;
1787 }
1788 set_result(_gvn.transform(n));
1789 return true;
1790 }
1791
1792 //------------------------------runtime_math-----------------------------
1793 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1794 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1795 "must be (DD)D or (D)D type");
1796
1797 // Inputs
1798 Node* a = round_double_node(argument(0));
1799 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : nullptr;
1800
1801 const TypePtr* no_memory_effects = nullptr;
1802 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1803 no_memory_effects,
1804 a, top(), b, b ? top() : nullptr);
1805 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1806 #ifdef ASSERT
1807 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1808 assert(value_top == top(), "second value must be top");
1809 #endif
1810
1811 set_result(value);
1812 return true;
1813 }
1814
1815 //------------------------------inline_math_pow-----------------------------
1816 bool LibraryCallKit::inline_math_pow() {
1817 Node* exp = round_double_node(argument(2));
1818 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1819 if (d != nullptr) {
1820 if (d->getd() == 2.0) {
1821 // Special case: pow(x, 2.0) => x * x
1822 Node* base = round_double_node(argument(0));
1823 set_result(_gvn.transform(new MulDNode(base, base)));
1824 return true;
1825 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1826 // Special case: pow(x, 0.5) => sqrt(x)
1827 Node* base = round_double_node(argument(0));
1828 Node* zero = _gvn.zerocon(T_DOUBLE);
1829
1830 RegionNode* region = new RegionNode(3);
1831 Node* phi = new PhiNode(region, Type::DOUBLE);
1832
1833 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1834 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1835 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1836 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1837 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1838
1839 Node* if_pow = generate_slow_guard(test, nullptr);
1840 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1841 phi->init_req(1, value_sqrt);
1842 region->init_req(1, control());
1843
1844 if (if_pow != nullptr) {
1845 set_control(if_pow);
1846 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1847 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1848 const TypePtr* no_memory_effects = nullptr;
1849 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1850 no_memory_effects, base, top(), exp, top());
1851 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1852 #ifdef ASSERT
1853 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1854 assert(value_top == top(), "second value must be top");
1855 #endif
1856 phi->init_req(2, value_pow);
1857 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1858 }
1859
1860 C->set_has_split_ifs(true); // Has chance for split-if optimization
1861 set_control(_gvn.transform(region));
1862 record_for_igvn(region);
1863 set_result(_gvn.transform(phi));
1864
1865 return true;
1866 }
1867 }
1868
1869 return StubRoutines::dpow() != nullptr ?
1870 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1871 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1872 }
1873
1874 //------------------------------inline_math_native-----------------------------
1875 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1876 switch (id) {
1877 case vmIntrinsics::_dsin:
1878 return StubRoutines::dsin() != nullptr ?
1879 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1880 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1881 case vmIntrinsics::_dcos:
1882 return StubRoutines::dcos() != nullptr ?
1883 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1884 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1885 case vmIntrinsics::_dtan:
1886 return StubRoutines::dtan() != nullptr ?
1887 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1888 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1889 case vmIntrinsics::_dexp:
1890 return StubRoutines::dexp() != nullptr ?
1891 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1892 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1893 case vmIntrinsics::_dlog:
1894 return StubRoutines::dlog() != nullptr ?
1895 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1896 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1897 case vmIntrinsics::_dlog10:
1898 return StubRoutines::dlog10() != nullptr ?
1899 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1900 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1901
1902 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1903 case vmIntrinsics::_ceil:
1904 case vmIntrinsics::_floor:
1905 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1906
1907 case vmIntrinsics::_dsqrt:
1908 case vmIntrinsics::_dsqrt_strict:
1909 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1910 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1911 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1912 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1913 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1914
1915 case vmIntrinsics::_dpow: return inline_math_pow();
1916 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1917 case vmIntrinsics::_fcopySign: return inline_math(id);
1918 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1919 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1920 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1921
1922 // These intrinsics are not yet correctly implemented
1923 case vmIntrinsics::_datan2:
1924 return false;
1925
1926 default:
1927 fatal_unexpected_iid(id);
1928 return false;
1929 }
1930 }
1931
1932 //----------------------------inline_notify-----------------------------------*
1933 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1934 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1935 address func;
1936 if (id == vmIntrinsics::_notify) {
1937 func = OptoRuntime::monitor_notify_Java();
1938 } else {
1939 func = OptoRuntime::monitor_notifyAll_Java();
1940 }
1941 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1942 make_slow_call_ex(call, env()->Throwable_klass(), false);
1943 return true;
1944 }
1945
1946
1947 //----------------------------inline_min_max-----------------------------------
1948 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1949 set_result(generate_min_max(id, argument(0), argument(1)));
1950 return true;
1951 }
1952
1953 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1954 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1955 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1956 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1957 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1958
1959 {
1960 PreserveJVMState pjvms(this);
1961 PreserveReexecuteState preexecs(this);
1962 jvms()->set_should_reexecute(true);
1963
1964 set_control(slow_path);
1965 set_i_o(i_o());
1966
1967 uncommon_trap(Deoptimization::Reason_intrinsic,
1968 Deoptimization::Action_none);
1969 }
1970
1971 set_control(fast_path);
1972 set_result(math);
1973 }
1974
1975 template <typename OverflowOp>
1976 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1977 typedef typename OverflowOp::MathOp MathOp;
1978
1979 MathOp* mathOp = new MathOp(arg1, arg2);
1980 Node* operation = _gvn.transform( mathOp );
1981 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1982 inline_math_mathExact(operation, ofcheck);
1983 return true;
1984 }
1985
1986 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1987 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1988 }
1989
1990 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1991 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1992 }
1993
1994 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1995 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1996 }
1997
1998 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1999 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2000 }
2001
2002 bool LibraryCallKit::inline_math_negateExactI() {
2003 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2004 }
2005
2006 bool LibraryCallKit::inline_math_negateExactL() {
2007 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2008 }
2009
2010 bool LibraryCallKit::inline_math_multiplyExactI() {
2011 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2012 }
2013
2014 bool LibraryCallKit::inline_math_multiplyExactL() {
2015 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2016 }
2017
2018 bool LibraryCallKit::inline_math_multiplyHigh() {
2019 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2020 return true;
2021 }
2022
2023 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2024 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2025 return true;
2026 }
2027
2028 Node*
2029 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
2030 Node* result_val = nullptr;
2031 switch (id) {
2032 case vmIntrinsics::_min:
2033 case vmIntrinsics::_min_strict:
2034 result_val = _gvn.transform(new MinINode(x0, y0));
2035 break;
2036 case vmIntrinsics::_max:
2037 case vmIntrinsics::_max_strict:
2038 result_val = _gvn.transform(new MaxINode(x0, y0));
2039 break;
2040 default:
2041 fatal_unexpected_iid(id);
2042 break;
2043 }
2044 return result_val;
2045 }
2046
2047 inline int
2048 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2049 const TypePtr* base_type = TypePtr::NULL_PTR;
2050 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2051 if (base_type == nullptr) {
2052 // Unknown type.
2053 return Type::AnyPtr;
2054 } else if (base_type == TypePtr::NULL_PTR) {
2055 // Since this is a null+long form, we have to switch to a rawptr.
2056 base = _gvn.transform(new CastX2PNode(offset));
2057 offset = MakeConX(0);
2058 return Type::RawPtr;
2059 } else if (base_type->base() == Type::RawPtr) {
2060 return Type::RawPtr;
2061 } else if (base_type->isa_oopptr()) {
2062 // Base is never null => always a heap address.
2063 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2064 return Type::OopPtr;
2065 }
2066 // Offset is small => always a heap address.
2067 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2068 if (offset_type != nullptr &&
2069 base_type->offset() == 0 && // (should always be?)
2070 offset_type->_lo >= 0 &&
2071 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2072 return Type::OopPtr;
2073 } else if (type == T_OBJECT) {
2074 // off heap access to an oop doesn't make any sense. Has to be on
2075 // heap.
2076 return Type::OopPtr;
2077 }
2078 // Otherwise, it might either be oop+off or null+addr.
2079 return Type::AnyPtr;
2080 } else {
2081 // No information:
2082 return Type::AnyPtr;
2083 }
2084 }
2085
2086 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2087 Node* uncasted_base = base;
2088 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2089 if (kind == Type::RawPtr) {
2090 return basic_plus_adr(top(), uncasted_base, offset);
2091 } else if (kind == Type::AnyPtr) {
2092 assert(base == uncasted_base, "unexpected base change");
2093 if (can_cast) {
2094 if (!_gvn.type(base)->speculative_maybe_null() &&
2095 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2096 // According to profiling, this access is always on
2097 // heap. Casting the base to not null and thus avoiding membars
2098 // around the access should allow better optimizations
2099 Node* null_ctl = top();
2100 base = null_check_oop(base, &null_ctl, true, true, true);
2101 assert(null_ctl->is_top(), "no null control here");
2102 return basic_plus_adr(base, offset);
2103 } else if (_gvn.type(base)->speculative_always_null() &&
2104 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2105 // According to profiling, this access is always off
2106 // heap.
2107 base = null_assert(base);
2108 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2109 offset = MakeConX(0);
2110 return basic_plus_adr(top(), raw_base, offset);
2111 }
2112 }
2113 // We don't know if it's an on heap or off heap access. Fall back
2114 // to raw memory access.
2115 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2116 return basic_plus_adr(top(), raw, offset);
2117 } else {
2118 assert(base == uncasted_base, "unexpected base change");
2119 // We know it's an on heap access so base can't be null
2120 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2121 base = must_be_not_null(base, true);
2122 }
2123 return basic_plus_adr(base, offset);
2124 }
2125 }
2126
2127 //--------------------------inline_number_methods-----------------------------
2128 // inline int Integer.numberOfLeadingZeros(int)
2129 // inline int Long.numberOfLeadingZeros(long)
2130 //
2131 // inline int Integer.numberOfTrailingZeros(int)
2132 // inline int Long.numberOfTrailingZeros(long)
2133 //
2134 // inline int Integer.bitCount(int)
2135 // inline int Long.bitCount(long)
2136 //
2137 // inline char Character.reverseBytes(char)
2138 // inline short Short.reverseBytes(short)
2139 // inline int Integer.reverseBytes(int)
2140 // inline long Long.reverseBytes(long)
2141 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2142 Node* arg = argument(0);
2143 Node* n = nullptr;
2144 switch (id) {
2145 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2146 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2147 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2148 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2149 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2150 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2151 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(nullptr, arg); break;
2152 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( nullptr, arg); break;
2153 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( nullptr, arg); break;
2154 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( nullptr, arg); break;
2155 case vmIntrinsics::_reverse_i: n = new ReverseINode(nullptr, arg); break;
2156 case vmIntrinsics::_reverse_l: n = new ReverseLNode(nullptr, arg); break;
2157 default: fatal_unexpected_iid(id); break;
2158 }
2159 set_result(_gvn.transform(n));
2160 return true;
2161 }
2162
2163 //--------------------------inline_bitshuffle_methods-----------------------------
2164 // inline int Integer.compress(int, int)
2165 // inline int Integer.expand(int, int)
2166 // inline long Long.compress(long, long)
2167 // inline long Long.expand(long, long)
2168 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2169 Node* n = nullptr;
2170 switch (id) {
2171 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2172 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2173 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2174 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2175 default: fatal_unexpected_iid(id); break;
2176 }
2177 set_result(_gvn.transform(n));
2178 return true;
2179 }
2180
2181 //--------------------------inline_number_methods-----------------------------
2182 // inline int Integer.compareUnsigned(int, int)
2183 // inline int Long.compareUnsigned(long, long)
2184 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2185 Node* arg1 = argument(0);
2186 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2187 Node* n = nullptr;
2188 switch (id) {
2189 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2190 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2191 default: fatal_unexpected_iid(id); break;
2192 }
2193 set_result(_gvn.transform(n));
2194 return true;
2195 }
2196
2197 //--------------------------inline_unsigned_divmod_methods-----------------------------
2198 // inline int Integer.divideUnsigned(int, int)
2199 // inline int Integer.remainderUnsigned(int, int)
2200 // inline long Long.divideUnsigned(long, long)
2201 // inline long Long.remainderUnsigned(long, long)
2202 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2203 Node* n = nullptr;
2204 switch (id) {
2205 case vmIntrinsics::_divideUnsigned_i: {
2206 zero_check_int(argument(1));
2207 // Compile-time detect of null-exception
2208 if (stopped()) {
2209 return true; // keep the graph constructed so far
2210 }
2211 n = new UDivINode(control(), argument(0), argument(1));
2212 break;
2213 }
2214 case vmIntrinsics::_divideUnsigned_l: {
2215 zero_check_long(argument(2));
2216 // Compile-time detect of null-exception
2217 if (stopped()) {
2218 return true; // keep the graph constructed so far
2219 }
2220 n = new UDivLNode(control(), argument(0), argument(2));
2221 break;
2222 }
2223 case vmIntrinsics::_remainderUnsigned_i: {
2224 zero_check_int(argument(1));
2225 // Compile-time detect of null-exception
2226 if (stopped()) {
2227 return true; // keep the graph constructed so far
2228 }
2229 n = new UModINode(control(), argument(0), argument(1));
2230 break;
2231 }
2232 case vmIntrinsics::_remainderUnsigned_l: {
2233 zero_check_long(argument(2));
2234 // Compile-time detect of null-exception
2235 if (stopped()) {
2236 return true; // keep the graph constructed so far
2237 }
2238 n = new UModLNode(control(), argument(0), argument(2));
2239 break;
2240 }
2241 default: fatal_unexpected_iid(id); break;
2242 }
2243 set_result(_gvn.transform(n));
2244 return true;
2245 }
2246
2247 //----------------------------inline_unsafe_access----------------------------
2248
2249 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2250 // Attempt to infer a sharper value type from the offset and base type.
2251 ciKlass* sharpened_klass = nullptr;
2252 bool null_free = false;
2253
2254 // See if it is an instance field, with an object type.
2255 if (alias_type->field() != nullptr) {
2256 if (alias_type->field()->type()->is_klass()) {
2257 sharpened_klass = alias_type->field()->type()->as_klass();
2258 null_free = alias_type->field()->is_null_free();
2259 }
2260 }
2261
2262 const TypeOopPtr* result = nullptr;
2263 // See if it is a narrow oop array.
2264 if (adr_type->isa_aryptr()) {
2265 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2266 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2267 null_free = adr_type->is_aryptr()->is_null_free();
2268 if (elem_type != nullptr && elem_type->is_loaded()) {
2269 // Sharpen the value type.
2270 result = elem_type;
2271 }
2272 }
2273 }
2274
2275 // The sharpened class might be unloaded if there is no class loader
2276 // contraint in place.
2277 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2278 // Sharpen the value type.
2279 result = TypeOopPtr::make_from_klass(sharpened_klass);
2280 if (null_free) {
2281 result = result->join_speculative(TypePtr::NOTNULL)->is_oopptr();
2282 }
2283 }
2284 if (result != nullptr) {
2285 #ifndef PRODUCT
2286 if (C->print_intrinsics() || C->print_inlining()) {
2287 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2288 tty->print(" sharpened value: "); result->dump(); tty->cr();
2289 }
2290 #endif
2291 }
2292 return result;
2293 }
2294
2295 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2296 switch (kind) {
2297 case Relaxed:
2298 return MO_UNORDERED;
2299 case Opaque:
2300 return MO_RELAXED;
2301 case Acquire:
2302 return MO_ACQUIRE;
2303 case Release:
2304 return MO_RELEASE;
2305 case Volatile:
2306 return MO_SEQ_CST;
2307 default:
2308 ShouldNotReachHere();
2309 return 0;
2310 }
2311 }
2312
2313 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned, const bool is_flat) {
2314 if (callee()->is_static()) return false; // caller must have the capability!
2315 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2316 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2317 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2318 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2319
2320 if (is_reference_type(type)) {
2321 decorators |= ON_UNKNOWN_OOP_REF;
2322 }
2323
2324 if (unaligned) {
2325 decorators |= C2_UNALIGNED;
2326 }
2327
2328 #ifndef PRODUCT
2329 {
2330 ResourceMark rm;
2331 // Check the signatures.
2332 ciSignature* sig = callee()->signature();
2333 #ifdef ASSERT
2334 if (!is_store) {
2335 // Object getReference(Object base, int/long offset), etc.
2336 BasicType rtype = sig->return_type()->basic_type();
2337 assert(rtype == type, "getter must return the expected value");
2338 assert(sig->count() == 2 || (is_flat && sig->count() == 3), "oop getter has 2 or 3 arguments");
2339 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2340 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2341 } else {
2342 // void putReference(Object base, int/long offset, Object x), etc.
2343 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2344 assert(sig->count() == 3 || (is_flat && sig->count() == 4), "oop putter has 3 arguments");
2345 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2346 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2347 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2348 assert(vtype == type, "putter must accept the expected value");
2349 }
2350 #endif // ASSERT
2351 }
2352 #endif //PRODUCT
2353
2354 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2355
2356 Node* receiver = argument(0); // type: oop
2357
2358 // Build address expression.
2359 Node* heap_base_oop = top();
2360
2361 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2362 Node* base = argument(1); // type: oop
2363 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2364 Node* offset = argument(2); // type: long
2365 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2366 // to be plain byte offsets, which are also the same as those accepted
2367 // by oopDesc::field_addr.
2368 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2369 "fieldOffset must be byte-scaled");
2370
2371 ciInlineKlass* inline_klass = nullptr;
2372 if (is_flat) {
2373 const TypeInstPtr* cls = _gvn.type(argument(4))->isa_instptr();
2374 if (cls == nullptr || cls->const_oop() == nullptr) {
2375 return false;
2376 }
2377 ciType* mirror_type = cls->const_oop()->as_instance()->java_mirror_type();
2378 if (!mirror_type->is_inlinetype()) {
2379 return false;
2380 }
2381 inline_klass = mirror_type->as_inline_klass();
2382 }
2383
2384 if (base->is_InlineType()) {
2385 InlineTypeNode* vt = base->as_InlineType();
2386 if (is_store) {
2387 if (!vt->is_allocated(&_gvn)) {
2388 return false;
2389 }
2390 base = vt->get_oop();
2391 } else {
2392 if (offset->is_Con()) {
2393 long off = find_long_con(offset, 0);
2394 ciInlineKlass* vk = vt->type()->inline_klass();
2395 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2396 return false;
2397 }
2398
2399 ciField* field = vk->get_non_flat_field_by_offset(off);
2400 if (field != nullptr) {
2401 BasicType bt = type2field[field->type()->basic_type()];
2402 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2403 bt = T_OBJECT;
2404 }
2405 if (bt == type && (!field->is_flat() || field->type() == inline_klass)) {
2406 Node* value = vt->field_value_by_offset(off, false);
2407 if (value->is_InlineType()) {
2408 value = value->as_InlineType()->adjust_scalarization_depth(this);
2409 }
2410 set_result(value);
2411 return true;
2412 }
2413 }
2414 }
2415 {
2416 // Re-execute the unsafe access if allocation triggers deoptimization.
2417 PreserveReexecuteState preexecs(this);
2418 jvms()->set_should_reexecute(true);
2419 vt = vt->buffer(this);
2420 }
2421 base = vt->get_oop();
2422 }
2423 }
2424
2425 // 32-bit machines ignore the high half!
2426 offset = ConvL2X(offset);
2427
2428 // Save state and restore on bailout
2429 uint old_sp = sp();
2430 SafePointNode* old_map = clone_map();
2431
2432 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2433
2434 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) {
2435 if (type != T_OBJECT && (inline_klass == nullptr || !inline_klass->has_object_fields())) {
2436 decorators |= IN_NATIVE; // off-heap primitive access
2437 } else {
2438 set_map(old_map);
2439 set_sp(old_sp);
2440 return false; // off-heap oop accesses are not supported
2441 }
2442 } else {
2443 heap_base_oop = base; // on-heap or mixed access
2444 }
2445
2446 // Can base be null? Otherwise, always on-heap access.
2447 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2448
2449 if (!can_access_non_heap) {
2450 decorators |= IN_HEAP;
2451 }
2452
2453 Node* val = is_store ? argument(4 + (is_flat ? 1 : 0)) : nullptr;
2454
2455 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2456 if (adr_type == TypePtr::NULL_PTR) {
2457 set_map(old_map);
2458 set_sp(old_sp);
2459 return false; // off-heap access with zero address
2460 }
2461
2462 // Try to categorize the address.
2463 Compile::AliasType* alias_type = C->alias_type(adr_type);
2464 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2465
2466 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2467 alias_type->adr_type() == TypeAryPtr::RANGE) {
2468 set_map(old_map);
2469 set_sp(old_sp);
2470 return false; // not supported
2471 }
2472
2473 bool mismatched = false;
2474 BasicType bt = T_ILLEGAL;
2475 ciField* field = nullptr;
2476 if (adr_type->isa_instptr()) {
2477 const TypeInstPtr* instptr = adr_type->is_instptr();
2478 ciInstanceKlass* k = instptr->instance_klass();
2479 int off = instptr->offset();
2480 if (instptr->const_oop() != nullptr &&
2481 k == ciEnv::current()->Class_klass() &&
2482 instptr->offset() >= (k->size_helper() * wordSize)) {
2483 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2484 field = k->get_field_by_offset(off, true);
2485 } else {
2486 field = k->get_non_flat_field_by_offset(off);
2487 }
2488 if (field != nullptr) {
2489 bt = type2field[field->type()->basic_type()];
2490 }
2491 assert(bt == alias_type->basic_type() || is_flat, "should match");
2492 } else {
2493 bt = alias_type->basic_type();
2494 }
2495
2496 if (bt != T_ILLEGAL) {
2497 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2498 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2499 // Alias type doesn't differentiate between byte[] and boolean[]).
2500 // Use address type to get the element type.
2501 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2502 }
2503 if (is_reference_type(bt, true)) {
2504 // accessing an array field with getReference is not a mismatch
2505 bt = T_OBJECT;
2506 }
2507 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2508 // Don't intrinsify mismatched object accesses
2509 set_map(old_map);
2510 set_sp(old_sp);
2511 return false;
2512 }
2513 mismatched = (bt != type);
2514 } else if (alias_type->adr_type()->isa_oopptr()) {
2515 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2516 }
2517
2518 if (is_flat) {
2519 if (adr_type->isa_instptr()) {
2520 if (field == nullptr || field->type() != inline_klass) {
2521 mismatched = true;
2522 }
2523 } else if (adr_type->isa_aryptr()) {
2524 const Type* elem = adr_type->is_aryptr()->elem();
2525 if (!adr_type->is_flat() || elem->inline_klass() != inline_klass) {
2526 mismatched = true;
2527 }
2528 } else {
2529 mismatched = true;
2530 }
2531 if (is_store) {
2532 const Type* val_t = _gvn.type(val);
2533 if (!val_t->is_inlinetypeptr() || val_t->inline_klass() != inline_klass) {
2534 set_map(old_map);
2535 set_sp(old_sp);
2536 return false;
2537 }
2538 }
2539 }
2540
2541 destruct_map_clone(old_map);
2542 assert(!mismatched || is_flat || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2543
2544 if (mismatched) {
2545 decorators |= C2_MISMATCHED;
2546 }
2547
2548 // First guess at the value type.
2549 const Type *value_type = Type::get_const_basic_type(type);
2550
2551 // Figure out the memory ordering.
2552 decorators |= mo_decorator_for_access_kind(kind);
2553
2554 if (!is_store) {
2555 if (type == T_OBJECT && !is_flat) {
2556 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2557 if (tjp != nullptr) {
2558 value_type = tjp;
2559 }
2560 }
2561 }
2562
2563 receiver = null_check(receiver);
2564 if (stopped()) {
2565 return true;
2566 }
2567 // Heap pointers get a null-check from the interpreter,
2568 // as a courtesy. However, this is not guaranteed by Unsafe,
2569 // and it is not possible to fully distinguish unintended nulls
2570 // from intended ones in this API.
2571
2572 if (!is_store) {
2573 Node* p = nullptr;
2574 // Try to constant fold a load from a constant field
2575
2576 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !field->is_flat() && !mismatched) {
2577 // final or stable field
2578 p = make_constant_from_field(field, heap_base_oop);
2579 }
2580
2581 if (p == nullptr) { // Could not constant fold the load
2582 if (is_flat) {
2583 if (adr_type->isa_instptr() && !mismatched) {
2584 ciInstanceKlass* holder = adr_type->is_instptr()->instance_klass();
2585 int offset = adr_type->is_instptr()->offset();
2586 p = InlineTypeNode::make_from_flat(this, inline_klass, base, base, holder, offset, decorators);
2587 } else {
2588 p = InlineTypeNode::make_from_flat(this, inline_klass, base, adr, nullptr, 0, decorators);
2589 }
2590 } else {
2591 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2592 const TypeOopPtr* ptr = value_type->make_oopptr();
2593 if (ptr != nullptr && ptr->is_inlinetypeptr()) {
2594 // Load a non-flattened inline type from memory
2595 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass(), !ptr->maybe_null());
2596 }
2597 }
2598 // Normalize the value returned by getBoolean in the following cases
2599 if (type == T_BOOLEAN &&
2600 (mismatched ||
2601 heap_base_oop == top() || // - heap_base_oop is null or
2602 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2603 // and the unsafe access is made to large offset
2604 // (i.e., larger than the maximum offset necessary for any
2605 // field access)
2606 ) {
2607 IdealKit ideal = IdealKit(this);
2608 #define __ ideal.
2609 IdealVariable normalized_result(ideal);
2610 __ declarations_done();
2611 __ set(normalized_result, p);
2612 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2613 __ set(normalized_result, ideal.ConI(1));
2614 ideal.end_if();
2615 final_sync(ideal);
2616 p = __ value(normalized_result);
2617 #undef __
2618 }
2619 }
2620 if (type == T_ADDRESS) {
2621 p = gvn().transform(new CastP2XNode(nullptr, p));
2622 p = ConvX2UL(p);
2623 }
2624 // The load node has the control of the preceding MemBarCPUOrder. All
2625 // following nodes will have the control of the MemBarCPUOrder inserted at
2626 // the end of this method. So, pushing the load onto the stack at a later
2627 // point is fine.
2628 set_result(p);
2629 } else {
2630 if (bt == T_ADDRESS) {
2631 // Repackage the long as a pointer.
2632 val = ConvL2X(val);
2633 val = gvn().transform(new CastX2PNode(val));
2634 }
2635 if (is_flat) {
2636 if (adr_type->isa_instptr() && !mismatched) {
2637 ciInstanceKlass* holder = adr_type->is_instptr()->instance_klass();
2638 int offset = adr_type->is_instptr()->offset();
2639 val->as_InlineType()->store_flat(this, base, base, holder, offset, decorators);
2640 } else {
2641 val->as_InlineType()->store_flat(this, base, adr, nullptr, 0, decorators);
2642 }
2643 } else {
2644 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2645 }
2646 }
2647
2648 if (argument(1)->is_InlineType() && is_store) {
2649 InlineTypeNode* value = InlineTypeNode::make_from_oop(this, base, _gvn.type(argument(1))->inline_klass());
2650 value = value->make_larval(this, false);
2651 replace_in_map(argument(1), value);
2652 }
2653
2654 return true;
2655 }
2656
2657 bool LibraryCallKit::inline_unsafe_make_private_buffer() {
2658 Node* receiver = argument(0);
2659 Node* value = argument(1);
2660 if (!value->is_InlineType()) {
2661 return false;
2662 }
2663
2664 receiver = null_check(receiver);
2665 if (stopped()) {
2666 return true;
2667 }
2668
2669 set_result(value->as_InlineType()->make_larval(this, true));
2670 return true;
2671 }
2672
2673 bool LibraryCallKit::inline_unsafe_finish_private_buffer() {
2674 Node* receiver = argument(0);
2675 Node* buffer = argument(1);
2676 if (!buffer->is_InlineType()) {
2677 return false;
2678 }
2679 InlineTypeNode* vt = buffer->as_InlineType();
2680 if (!vt->is_allocated(&_gvn)) {
2681 return false;
2682 }
2683 // TODO 8239003 Why is this needed?
2684 if (AllocateNode::Ideal_allocation(vt->get_oop()) == nullptr) {
2685 return false;
2686 }
2687
2688 receiver = null_check(receiver);
2689 if (stopped()) {
2690 return true;
2691 }
2692
2693 set_result(vt->finish_larval(this));
2694 return true;
2695 }
2696
2697 //----------------------------inline_unsafe_load_store----------------------------
2698 // This method serves a couple of different customers (depending on LoadStoreKind):
2699 //
2700 // LS_cmp_swap:
2701 //
2702 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2703 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2704 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2705 //
2706 // LS_cmp_swap_weak:
2707 //
2708 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2709 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2710 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2711 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2712 //
2713 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2714 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2715 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2716 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2717 //
2718 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2719 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2720 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2721 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2722 //
2723 // LS_cmp_exchange:
2724 //
2725 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2726 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2727 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2728 //
2729 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2730 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2731 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2732 //
2733 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2734 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2735 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2736 //
2737 // LS_get_add:
2738 //
2739 // int getAndAddInt( Object o, long offset, int delta)
2740 // long getAndAddLong(Object o, long offset, long delta)
2741 //
2742 // LS_get_set:
2743 //
2744 // int getAndSet(Object o, long offset, int newValue)
2745 // long getAndSet(Object o, long offset, long newValue)
2746 // Object getAndSet(Object o, long offset, Object newValue)
2747 //
2748 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2749 // This basic scheme here is the same as inline_unsafe_access, but
2750 // differs in enough details that combining them would make the code
2751 // overly confusing. (This is a true fact! I originally combined
2752 // them, but even I was confused by it!) As much code/comments as
2753 // possible are retained from inline_unsafe_access though to make
2754 // the correspondences clearer. - dl
2755
2756 if (callee()->is_static()) return false; // caller must have the capability!
2757
2758 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2759 decorators |= mo_decorator_for_access_kind(access_kind);
2760
2761 #ifndef PRODUCT
2762 BasicType rtype;
2763 {
2764 ResourceMark rm;
2765 // Check the signatures.
2766 ciSignature* sig = callee()->signature();
2767 rtype = sig->return_type()->basic_type();
2768 switch(kind) {
2769 case LS_get_add:
2770 case LS_get_set: {
2771 // Check the signatures.
2772 #ifdef ASSERT
2773 assert(rtype == type, "get and set must return the expected type");
2774 assert(sig->count() == 3, "get and set has 3 arguments");
2775 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2776 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2777 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2778 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2779 #endif // ASSERT
2780 break;
2781 }
2782 case LS_cmp_swap:
2783 case LS_cmp_swap_weak: {
2784 // Check the signatures.
2785 #ifdef ASSERT
2786 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2787 assert(sig->count() == 4, "CAS has 4 arguments");
2788 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2789 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2790 #endif // ASSERT
2791 break;
2792 }
2793 case LS_cmp_exchange: {
2794 // Check the signatures.
2795 #ifdef ASSERT
2796 assert(rtype == type, "CAS must return the expected type");
2797 assert(sig->count() == 4, "CAS has 4 arguments");
2798 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2799 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2800 #endif // ASSERT
2801 break;
2802 }
2803 default:
2804 ShouldNotReachHere();
2805 }
2806 }
2807 #endif //PRODUCT
2808
2809 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2810
2811 // Get arguments:
2812 Node* receiver = nullptr;
2813 Node* base = nullptr;
2814 Node* offset = nullptr;
2815 Node* oldval = nullptr;
2816 Node* newval = nullptr;
2817 switch(kind) {
2818 case LS_cmp_swap:
2819 case LS_cmp_swap_weak:
2820 case LS_cmp_exchange: {
2821 const bool two_slot_type = type2size[type] == 2;
2822 receiver = argument(0); // type: oop
2823 base = argument(1); // type: oop
2824 offset = argument(2); // type: long
2825 oldval = argument(4); // type: oop, int, or long
2826 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2827 break;
2828 }
2829 case LS_get_add:
2830 case LS_get_set: {
2831 receiver = argument(0); // type: oop
2832 base = argument(1); // type: oop
2833 offset = argument(2); // type: long
2834 oldval = nullptr;
2835 newval = argument(4); // type: oop, int, or long
2836 break;
2837 }
2838 default:
2839 ShouldNotReachHere();
2840 }
2841
2842 // Build field offset expression.
2843 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2844 // to be plain byte offsets, which are also the same as those accepted
2845 // by oopDesc::field_addr.
2846 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2847 // 32-bit machines ignore the high half of long offsets
2848 offset = ConvL2X(offset);
2849 // Save state and restore on bailout
2850 uint old_sp = sp();
2851 SafePointNode* old_map = clone_map();
2852 Node* adr = make_unsafe_address(base, offset,type, false);
2853 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2854
2855 Compile::AliasType* alias_type = C->alias_type(adr_type);
2856 BasicType bt = alias_type->basic_type();
2857 if (bt != T_ILLEGAL &&
2858 (is_reference_type(bt) != (type == T_OBJECT))) {
2859 // Don't intrinsify mismatched object accesses.
2860 set_map(old_map);
2861 set_sp(old_sp);
2862 return false;
2863 }
2864
2865 destruct_map_clone(old_map);
2866
2867 // For CAS, unlike inline_unsafe_access, there seems no point in
2868 // trying to refine types. Just use the coarse types here.
2869 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2870 const Type *value_type = Type::get_const_basic_type(type);
2871
2872 switch (kind) {
2873 case LS_get_set:
2874 case LS_cmp_exchange: {
2875 if (type == T_OBJECT) {
2876 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2877 if (tjp != nullptr) {
2878 value_type = tjp;
2879 }
2880 }
2881 break;
2882 }
2883 case LS_cmp_swap:
2884 case LS_cmp_swap_weak:
2885 case LS_get_add:
2886 break;
2887 default:
2888 ShouldNotReachHere();
2889 }
2890
2891 // Null check receiver.
2892 receiver = null_check(receiver);
2893 if (stopped()) {
2894 return true;
2895 }
2896
2897 int alias_idx = C->get_alias_index(adr_type);
2898
2899 if (is_reference_type(type)) {
2900 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2901
2902 if (oldval != nullptr && oldval->is_InlineType()) {
2903 // Re-execute the unsafe access if allocation triggers deoptimization.
2904 PreserveReexecuteState preexecs(this);
2905 jvms()->set_should_reexecute(true);
2906 oldval = oldval->as_InlineType()->buffer(this)->get_oop();
2907 }
2908 if (newval != nullptr && newval->is_InlineType()) {
2909 // Re-execute the unsafe access if allocation triggers deoptimization.
2910 PreserveReexecuteState preexecs(this);
2911 jvms()->set_should_reexecute(true);
2912 newval = newval->as_InlineType()->buffer(this)->get_oop();
2913 }
2914
2915 // Transformation of a value which could be null pointer (CastPP #null)
2916 // could be delayed during Parse (for example, in adjust_map_after_if()).
2917 // Execute transformation here to avoid barrier generation in such case.
2918 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2919 newval = _gvn.makecon(TypePtr::NULL_PTR);
2920
2921 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2922 // Refine the value to a null constant, when it is known to be null
2923 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2924 }
2925 }
2926
2927 Node* result = nullptr;
2928 switch (kind) {
2929 case LS_cmp_exchange: {
2930 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2931 oldval, newval, value_type, type, decorators);
2932 break;
2933 }
2934 case LS_cmp_swap_weak:
2935 decorators |= C2_WEAK_CMPXCHG;
2936 case LS_cmp_swap: {
2937 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2938 oldval, newval, value_type, type, decorators);
2939 break;
2940 }
2941 case LS_get_set: {
2942 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2943 newval, value_type, type, decorators);
2944 break;
2945 }
2946 case LS_get_add: {
2947 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2948 newval, value_type, type, decorators);
2949 break;
2950 }
2951 default:
2952 ShouldNotReachHere();
2953 }
2954
2955 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2956 set_result(result);
2957 return true;
2958 }
2959
2960 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2961 // Regardless of form, don't allow previous ld/st to move down,
2962 // then issue acquire, release, or volatile mem_bar.
2963 insert_mem_bar(Op_MemBarCPUOrder);
2964 switch(id) {
2965 case vmIntrinsics::_loadFence:
2966 insert_mem_bar(Op_LoadFence);
2967 return true;
2968 case vmIntrinsics::_storeFence:
2969 insert_mem_bar(Op_StoreFence);
2970 return true;
2971 case vmIntrinsics::_storeStoreFence:
2972 insert_mem_bar(Op_StoreStoreFence);
2973 return true;
2974 case vmIntrinsics::_fullFence:
2975 insert_mem_bar(Op_MemBarVolatile);
2976 return true;
2977 default:
2978 fatal_unexpected_iid(id);
2979 return false;
2980 }
2981 }
2982
2983 bool LibraryCallKit::inline_onspinwait() {
2984 insert_mem_bar(Op_OnSpinWait);
2985 return true;
2986 }
2987
2988 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2989 if (!kls->is_Con()) {
2990 return true;
2991 }
2992 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2993 if (klsptr == nullptr) {
2994 return true;
2995 }
2996 ciInstanceKlass* ik = klsptr->instance_klass();
2997 // don't need a guard for a klass that is already initialized
2998 return !ik->is_initialized();
2999 }
3000
3001 //----------------------------inline_unsafe_writeback0-------------------------
3002 // public native void Unsafe.writeback0(long address)
3003 bool LibraryCallKit::inline_unsafe_writeback0() {
3004 if (!Matcher::has_match_rule(Op_CacheWB)) {
3005 return false;
3006 }
3007 #ifndef PRODUCT
3008 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
3009 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
3010 ciSignature* sig = callee()->signature();
3011 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
3012 #endif
3013 null_check_receiver(); // null-check, then ignore
3014 Node *addr = argument(1);
3015 addr = new CastX2PNode(addr);
3016 addr = _gvn.transform(addr);
3017 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
3018 flush = _gvn.transform(flush);
3019 set_memory(flush, TypeRawPtr::BOTTOM);
3020 return true;
3021 }
3022
3023 //----------------------------inline_unsafe_writeback0-------------------------
3024 // public native void Unsafe.writeback0(long address)
3025 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
3026 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
3027 return false;
3028 }
3029 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
3030 return false;
3031 }
3032 #ifndef PRODUCT
3033 assert(Matcher::has_match_rule(Op_CacheWB),
3034 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
3035 : "found match rule for CacheWBPostSync but not CacheWB"));
3036
3037 #endif
3038 null_check_receiver(); // null-check, then ignore
3039 Node *sync;
3040 if (is_pre) {
3041 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3042 } else {
3043 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3044 }
3045 sync = _gvn.transform(sync);
3046 set_memory(sync, TypeRawPtr::BOTTOM);
3047 return true;
3048 }
3049
3050 //----------------------------inline_unsafe_allocate---------------------------
3051 // public native Object Unsafe.allocateInstance(Class<?> cls);
3052 bool LibraryCallKit::inline_unsafe_allocate() {
3053
3054 #if INCLUDE_JVMTI
3055 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3056 return false;
3057 }
3058 #endif //INCLUDE_JVMTI
3059
3060 if (callee()->is_static()) return false; // caller must have the capability!
3061
3062 null_check_receiver(); // null-check, then ignore
3063 Node* cls = null_check(argument(1));
3064 if (stopped()) return true;
3065
3066 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
3067 kls = null_check(kls);
3068 if (stopped()) return true; // argument was like int.class
3069
3070 #if INCLUDE_JVMTI
3071 // Don't try to access new allocated obj in the intrinsic.
3072 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
3073 // Deoptimize and allocate in interpreter instead.
3074 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
3075 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
3076 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
3077 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
3078 {
3079 BuildCutout unless(this, tst, PROB_MAX);
3080 uncommon_trap(Deoptimization::Reason_intrinsic,
3081 Deoptimization::Action_make_not_entrant);
3082 }
3083 if (stopped()) {
3084 return true;
3085 }
3086 #endif //INCLUDE_JVMTI
3087
3088 Node* test = nullptr;
3089 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3090 // Note: The argument might still be an illegal value like
3091 // Serializable.class or Object[].class. The runtime will handle it.
3092 // But we must make an explicit check for initialization.
3093 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3094 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3095 // can generate code to load it as unsigned byte.
3096 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
3097 Node* bits = intcon(InstanceKlass::fully_initialized);
3098 test = _gvn.transform(new SubINode(inst, bits));
3099 // The 'test' is non-zero if we need to take a slow path.
3100 }
3101 Node* obj = nullptr;
3102 const TypeInstKlassPtr* tkls = _gvn.type(kls)->isa_instklassptr();
3103 if (tkls != nullptr && tkls->instance_klass()->is_inlinetype()) {
3104 obj = InlineTypeNode::make_default(_gvn, tkls->instance_klass()->as_inline_klass())->buffer(this);
3105 } else {
3106 obj = new_instance(kls, test);
3107 }
3108 set_result(obj);
3109 return true;
3110 }
3111
3112 //------------------------inline_native_time_funcs--------------
3113 // inline code for System.currentTimeMillis() and System.nanoTime()
3114 // these have the same type and signature
3115 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3116 const TypeFunc* tf = OptoRuntime::void_long_Type();
3117 const TypePtr* no_memory_effects = nullptr;
3118 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3119 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3120 #ifdef ASSERT
3121 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3122 assert(value_top == top(), "second value must be top");
3123 #endif
3124 set_result(value);
3125 return true;
3126 }
3127
3128
3129 #if INCLUDE_JVMTI
3130
3131 // When notifications are disabled then just update the VTMS transition bit and return.
3132 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
3133 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
3134 if (!DoJVMTIVirtualThreadTransitions) {
3135 return true;
3136 }
3137 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3138 IdealKit ideal(this);
3139
3140 Node* ONE = ideal.ConI(1);
3141 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
3142 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
3143 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3144
3145 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
3146 sync_kit(ideal);
3147 // if notifyJvmti enabled then make a call to the given SharedRuntime function
3148 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
3149 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
3150 ideal.sync_kit(this);
3151 } ideal.else_(); {
3152 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
3153 Node* thread = ideal.thread();
3154 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
3155 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
3156 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3157
3158 sync_kit(ideal);
3159 access_store_at(nullptr, jt_addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3160 access_store_at(nullptr, vt_addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3161
3162 ideal.sync_kit(this);
3163 } ideal.end_if();
3164 final_sync(ideal);
3165
3166 return true;
3167 }
3168
3169 // Always update the temporary VTMS transition bit.
3170 bool LibraryCallKit::inline_native_notify_jvmti_hide() {
3171 if (!DoJVMTIVirtualThreadTransitions) {
3172 return true;
3173 }
3174 IdealKit ideal(this);
3175
3176 {
3177 // unconditionally update the temporary VTMS transition bit in current JavaThread
3178 Node* thread = ideal.thread();
3179 Node* hide = _gvn.transform(argument(0)); // hide argument for temporary VTMS transition notification
3180 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_tmp_VTMS_transition_offset()));
3181 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3182
3183 sync_kit(ideal);
3184 access_store_at(nullptr, addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3185 ideal.sync_kit(this);
3186 }
3187 final_sync(ideal);
3188
3189 return true;
3190 }
3191
3192 // Always update the is_disable_suspend bit.
3193 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3194 if (!DoJVMTIVirtualThreadTransitions) {
3195 return true;
3196 }
3197 IdealKit ideal(this);
3198
3199 {
3200 // unconditionally update the is_disable_suspend bit in current JavaThread
3201 Node* thread = ideal.thread();
3202 Node* arg = _gvn.transform(argument(0)); // argument for notification
3203 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3204 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3205
3206 sync_kit(ideal);
3207 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3208 ideal.sync_kit(this);
3209 }
3210 final_sync(ideal);
3211
3212 return true;
3213 }
3214
3215 #endif // INCLUDE_JVMTI
3216
3217 #ifdef JFR_HAVE_INTRINSICS
3218
3219 /**
3220 * if oop->klass != null
3221 * // normal class
3222 * epoch = _epoch_state ? 2 : 1
3223 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3224 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3225 * }
3226 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3227 * else
3228 * // primitive class
3229 * if oop->array_klass != null
3230 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3231 * else
3232 * id = LAST_TYPE_ID + 1 // void class path
3233 * if (!signaled)
3234 * signaled = true
3235 */
3236 bool LibraryCallKit::inline_native_classID() {
3237 Node* cls = argument(0);
3238
3239 IdealKit ideal(this);
3240 #define __ ideal.
3241 IdealVariable result(ideal); __ declarations_done();
3242 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3243 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3244 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3245
3246
3247 __ if_then(kls, BoolTest::ne, null()); {
3248 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3249 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3250
3251 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3252 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3253 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3254 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3255 mask = _gvn.transform(new OrLNode(mask, epoch));
3256 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3257
3258 float unlikely = PROB_UNLIKELY(0.999);
3259 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3260 sync_kit(ideal);
3261 make_runtime_call(RC_LEAF,
3262 OptoRuntime::class_id_load_barrier_Type(),
3263 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3264 "class id load barrier",
3265 TypePtr::BOTTOM,
3266 kls);
3267 ideal.sync_kit(this);
3268 } __ end_if();
3269
3270 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3271 } __ else_(); {
3272 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3273 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3274 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3275 __ if_then(array_kls, BoolTest::ne, null()); {
3276 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3277 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3278 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3279 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3280 } __ else_(); {
3281 // void class case
3282 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3283 } __ end_if();
3284
3285 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3286 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3287 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3288 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3289 } __ end_if();
3290 } __ end_if();
3291
3292 final_sync(ideal);
3293 set_result(ideal.value(result));
3294 #undef __
3295 return true;
3296 }
3297
3298 //------------------------inline_native_jvm_commit------------------
3299 bool LibraryCallKit::inline_native_jvm_commit() {
3300 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3301
3302 // Save input memory and i_o state.
3303 Node* input_memory_state = reset_memory();
3304 set_all_memory(input_memory_state);
3305 Node* input_io_state = i_o();
3306
3307 // TLS.
3308 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3309 // Jfr java buffer.
3310 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3311 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3312 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3313
3314 // Load the current value of the notified field in the JfrThreadLocal.
3315 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3316 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3317
3318 // Test for notification.
3319 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3320 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3321 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3322
3323 // True branch, is notified.
3324 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3325 set_control(is_notified);
3326
3327 // Reset notified state.
3328 Node* notified_reset_memory = store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered);
3329
3330 // 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.
3331 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3332 // Convert the machine-word to a long.
3333 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3334
3335 // False branch, not notified.
3336 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3337 set_control(not_notified);
3338 set_all_memory(input_memory_state);
3339
3340 // Arg is the next position as a long.
3341 Node* arg = argument(0);
3342 // Convert long to machine-word.
3343 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3344
3345 // Store the next_position to the underlying jfr java buffer.
3346 Node* commit_memory;
3347 #ifdef _LP64
3348 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_LONG, Compile::AliasIdxRaw, MemNode::release);
3349 #else
3350 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_INT, Compile::AliasIdxRaw, MemNode::release);
3351 #endif
3352
3353 // 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.
3354 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3355 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3356 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3357
3358 // And flags with lease constant.
3359 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3360
3361 // Branch on lease to conditionalize returning the leased java buffer.
3362 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3363 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3364 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3365
3366 // False branch, not a lease.
3367 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3368
3369 // True branch, is lease.
3370 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3371 set_control(is_lease);
3372
3373 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3374 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3375 OptoRuntime::void_void_Type(),
3376 SharedRuntime::jfr_return_lease(),
3377 "return_lease", TypePtr::BOTTOM);
3378 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3379
3380 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3381 record_for_igvn(lease_compare_rgn);
3382 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3383 record_for_igvn(lease_compare_mem);
3384 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3385 record_for_igvn(lease_compare_io);
3386 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3387 record_for_igvn(lease_result_value);
3388
3389 // Update control and phi nodes.
3390 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3391 lease_compare_rgn->init_req(_false_path, not_lease);
3392
3393 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3394 lease_compare_mem->init_req(_false_path, commit_memory);
3395
3396 lease_compare_io->init_req(_true_path, i_o());
3397 lease_compare_io->init_req(_false_path, input_io_state);
3398
3399 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3400 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3401
3402 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3403 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3404 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3405 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3406
3407 // Update control and phi nodes.
3408 result_rgn->init_req(_true_path, is_notified);
3409 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3410
3411 result_mem->init_req(_true_path, notified_reset_memory);
3412 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3413
3414 result_io->init_req(_true_path, input_io_state);
3415 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3416
3417 result_value->init_req(_true_path, current_pos);
3418 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3419
3420 // Set output state.
3421 set_control(_gvn.transform(result_rgn));
3422 set_all_memory(_gvn.transform(result_mem));
3423 set_i_o(_gvn.transform(result_io));
3424 set_result(result_rgn, result_value);
3425 return true;
3426 }
3427
3428 /*
3429 * The intrinsic is a model of this pseudo-code:
3430 *
3431 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3432 * jobject h_event_writer = tl->java_event_writer();
3433 * if (h_event_writer == nullptr) {
3434 * return nullptr;
3435 * }
3436 * oop threadObj = Thread::threadObj();
3437 * oop vthread = java_lang_Thread::vthread(threadObj);
3438 * traceid tid;
3439 * bool pinVirtualThread;
3440 * bool excluded;
3441 * if (vthread != threadObj) { // i.e. current thread is virtual
3442 * tid = java_lang_Thread::tid(vthread);
3443 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3444 * pinVirtualThread = VMContinuations;
3445 * excluded = vthread_epoch_raw & excluded_mask;
3446 * if (!excluded) {
3447 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3448 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3449 * if (vthread_epoch != current_epoch) {
3450 * write_checkpoint();
3451 * }
3452 * }
3453 * } else {
3454 * tid = java_lang_Thread::tid(threadObj);
3455 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3456 * pinVirtualThread = false;
3457 * excluded = thread_epoch_raw & excluded_mask;
3458 * }
3459 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3460 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3461 * if (tid_in_event_writer != tid) {
3462 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3463 * setField(event_writer, "excluded", excluded);
3464 * setField(event_writer, "threadID", tid);
3465 * }
3466 * return event_writer
3467 */
3468 bool LibraryCallKit::inline_native_getEventWriter() {
3469 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3470
3471 // Save input memory and i_o state.
3472 Node* input_memory_state = reset_memory();
3473 set_all_memory(input_memory_state);
3474 Node* input_io_state = i_o();
3475
3476 Node* excluded_mask = _gvn.intcon(32768);
3477 Node* epoch_mask = _gvn.intcon(32767);
3478
3479 // TLS
3480 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3481
3482 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3483 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3484
3485 // Load the eventwriter jobject handle.
3486 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3487
3488 // Null check the jobject handle.
3489 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3490 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3491 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3492
3493 // False path, jobj is null.
3494 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3495
3496 // True path, jobj is not null.
3497 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3498
3499 set_control(jobj_is_not_null);
3500
3501 // Load the threadObj for the CarrierThread.
3502 Node* threadObj = generate_current_thread(tls_ptr);
3503
3504 // Load the vthread.
3505 Node* vthread = generate_virtual_thread(tls_ptr);
3506
3507 // If vthread != threadObj, this is a virtual thread.
3508 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3509 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3510 IfNode* iff_vthread_not_equal_threadObj =
3511 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3512
3513 // False branch, fallback to threadObj.
3514 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3515 set_control(vthread_equal_threadObj);
3516
3517 // Load the tid field from the vthread object.
3518 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3519
3520 // Load the raw epoch value from the threadObj.
3521 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3522 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3523 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3524
3525 // Mask off the excluded information from the epoch.
3526 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3527
3528 // True branch, this is a virtual thread.
3529 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3530 set_control(vthread_not_equal_threadObj);
3531
3532 // Load the tid field from the vthread object.
3533 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3534
3535 // Continuation support determines if a virtual thread should be pinned.
3536 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3537 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3538
3539 // Load the raw epoch value from the vthread.
3540 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3541 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3542 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3543
3544 // Mask off the excluded information from the epoch.
3545 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3546
3547 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3548 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3549 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3550 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3551
3552 // False branch, vthread is excluded, no need to write epoch info.
3553 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3554
3555 // True branch, vthread is included, update epoch info.
3556 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3557 set_control(included);
3558
3559 // Get epoch value.
3560 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3561
3562 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3563 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3564 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3565
3566 // Compare the epoch in the vthread to the current epoch generation.
3567 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3568 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3569 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3570
3571 // False path, epoch is equal, checkpoint information is valid.
3572 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3573
3574 // True path, epoch is not equal, write a checkpoint for the vthread.
3575 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3576
3577 set_control(epoch_is_not_equal);
3578
3579 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3580 // The call also updates the native thread local thread id and the vthread with the current epoch.
3581 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3582 OptoRuntime::jfr_write_checkpoint_Type(),
3583 SharedRuntime::jfr_write_checkpoint(),
3584 "write_checkpoint", TypePtr::BOTTOM);
3585 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3586
3587 // vthread epoch != current epoch
3588 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3589 record_for_igvn(epoch_compare_rgn);
3590 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3591 record_for_igvn(epoch_compare_mem);
3592 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3593 record_for_igvn(epoch_compare_io);
3594
3595 // Update control and phi nodes.
3596 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3597 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3598 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3599 epoch_compare_mem->init_req(_false_path, input_memory_state);
3600 epoch_compare_io->init_req(_true_path, i_o());
3601 epoch_compare_io->init_req(_false_path, input_io_state);
3602
3603 // excluded != true
3604 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3605 record_for_igvn(exclude_compare_rgn);
3606 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3607 record_for_igvn(exclude_compare_mem);
3608 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3609 record_for_igvn(exclude_compare_io);
3610
3611 // Update control and phi nodes.
3612 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3613 exclude_compare_rgn->init_req(_false_path, excluded);
3614 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3615 exclude_compare_mem->init_req(_false_path, input_memory_state);
3616 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3617 exclude_compare_io->init_req(_false_path, input_io_state);
3618
3619 // vthread != threadObj
3620 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3621 record_for_igvn(vthread_compare_rgn);
3622 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3623 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3624 record_for_igvn(vthread_compare_io);
3625 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3626 record_for_igvn(tid);
3627 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3628 record_for_igvn(exclusion);
3629 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3630 record_for_igvn(pinVirtualThread);
3631
3632 // Update control and phi nodes.
3633 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3634 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3635 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3636 vthread_compare_mem->init_req(_false_path, input_memory_state);
3637 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3638 vthread_compare_io->init_req(_false_path, input_io_state);
3639 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3640 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3641 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3642 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3643 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3644 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3645
3646 // Update branch state.
3647 set_control(_gvn.transform(vthread_compare_rgn));
3648 set_all_memory(_gvn.transform(vthread_compare_mem));
3649 set_i_o(_gvn.transform(vthread_compare_io));
3650
3651 // Load the event writer oop by dereferencing the jobject handle.
3652 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3653 assert(klass_EventWriter->is_loaded(), "invariant");
3654 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3655 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3656 const TypeOopPtr* const xtype = aklass->as_instance_type();
3657 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3658 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3659
3660 // Load the current thread id from the event writer object.
3661 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3662 // Get the field offset to, conditionally, store an updated tid value later.
3663 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3664 const TypePtr* event_writer_tid_field_type = _gvn.type(event_writer_tid_field)->isa_ptr();
3665 // Get the field offset to, conditionally, store an updated exclusion value later.
3666 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3667 const TypePtr* event_writer_excluded_field_type = _gvn.type(event_writer_excluded_field)->isa_ptr();
3668 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3669 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3670 const TypePtr* event_writer_pin_field_type = _gvn.type(event_writer_pin_field)->isa_ptr();
3671
3672 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3673 record_for_igvn(event_writer_tid_compare_rgn);
3674 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3675 record_for_igvn(event_writer_tid_compare_mem);
3676 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3677 record_for_igvn(event_writer_tid_compare_io);
3678
3679 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3680 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3681 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3682 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3683
3684 // False path, tids are the same.
3685 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3686
3687 // True path, tid is not equal, need to update the tid in the event writer.
3688 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3689 record_for_igvn(tid_is_not_equal);
3690
3691 // Store the pin state to the event writer.
3692 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, event_writer_pin_field_type, MemNode::unordered);
3693
3694 // Store the exclusion state to the event writer.
3695 store_to_memory(tid_is_not_equal, event_writer_excluded_field, _gvn.transform(exclusion), T_BOOLEAN, event_writer_excluded_field_type, MemNode::unordered);
3696
3697 // Store the tid to the event writer.
3698 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, event_writer_tid_field_type, MemNode::unordered);
3699
3700 // Update control and phi nodes.
3701 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3702 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3703 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3704 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3705 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3706 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3707
3708 // Result of top level CFG, Memory, IO and Value.
3709 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3710 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3711 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3712 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3713
3714 // Result control.
3715 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3716 result_rgn->init_req(_false_path, jobj_is_null);
3717
3718 // Result memory.
3719 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3720 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3721
3722 // Result IO.
3723 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3724 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3725
3726 // Result value.
3727 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3728 result_value->init_req(_false_path, null()); // return null
3729
3730 // Set output state.
3731 set_control(_gvn.transform(result_rgn));
3732 set_all_memory(_gvn.transform(result_mem));
3733 set_i_o(_gvn.transform(result_io));
3734 set_result(result_rgn, result_value);
3735 return true;
3736 }
3737
3738 /*
3739 * The intrinsic is a model of this pseudo-code:
3740 *
3741 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3742 * if (carrierThread != thread) { // is virtual thread
3743 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3744 * bool excluded = vthread_epoch_raw & excluded_mask;
3745 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3746 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3747 * if (!excluded) {
3748 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3749 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3750 * }
3751 * Atomic::release_store(&tl->_vthread, true);
3752 * return;
3753 * }
3754 * Atomic::release_store(&tl->_vthread, false);
3755 */
3756 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3757 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3758
3759 Node* input_memory_state = reset_memory();
3760 set_all_memory(input_memory_state);
3761
3762 Node* excluded_mask = _gvn.intcon(32768);
3763 Node* epoch_mask = _gvn.intcon(32767);
3764
3765 Node* const carrierThread = generate_current_thread(jt);
3766 // If thread != carrierThread, this is a virtual thread.
3767 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3768 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3769 IfNode* iff_thread_not_equal_carrierThread =
3770 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3771
3772 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3773
3774 // False branch, is carrierThread.
3775 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3776 // Store release
3777 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3778
3779 set_all_memory(input_memory_state);
3780
3781 // True branch, is virtual thread.
3782 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3783 set_control(thread_not_equal_carrierThread);
3784
3785 // Load the raw epoch value from the vthread.
3786 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3787 Node* epoch_raw = access_load_at(thread, epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3788 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3789
3790 // Mask off the excluded information from the epoch.
3791 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3792
3793 // Load the tid field from the thread.
3794 Node* tid = load_field_from_object(thread, "tid", "J");
3795
3796 // Store the vthread tid to the jfr thread local.
3797 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3798 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, Compile::AliasIdxRaw, MemNode::unordered, true);
3799
3800 // Branch is_excluded to conditionalize updating the epoch .
3801 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3802 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3803 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3804
3805 // True branch, vthread is excluded, no need to write epoch info.
3806 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3807 set_control(excluded);
3808 Node* vthread_is_excluded = _gvn.intcon(1);
3809
3810 // False branch, vthread is included, update epoch info.
3811 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3812 set_control(included);
3813 Node* vthread_is_included = _gvn.intcon(0);
3814
3815 // Get epoch value.
3816 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3817
3818 // Store the vthread epoch to the jfr thread local.
3819 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3820 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, Compile::AliasIdxRaw, MemNode::unordered, true);
3821
3822 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3823 record_for_igvn(excluded_rgn);
3824 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3825 record_for_igvn(excluded_mem);
3826 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3827 record_for_igvn(exclusion);
3828
3829 // Merge the excluded control and memory.
3830 excluded_rgn->init_req(_true_path, excluded);
3831 excluded_rgn->init_req(_false_path, included);
3832 excluded_mem->init_req(_true_path, tid_memory);
3833 excluded_mem->init_req(_false_path, included_memory);
3834 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3835 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3836
3837 // Set intermediate state.
3838 set_control(_gvn.transform(excluded_rgn));
3839 set_all_memory(excluded_mem);
3840
3841 // Store the vthread exclusion state to the jfr thread local.
3842 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3843 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered, true);
3844
3845 // Store release
3846 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3847
3848 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3849 record_for_igvn(thread_compare_rgn);
3850 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3851 record_for_igvn(thread_compare_mem);
3852 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3853 record_for_igvn(vthread);
3854
3855 // Merge the thread_compare control and memory.
3856 thread_compare_rgn->init_req(_true_path, control());
3857 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3858 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3859 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3860
3861 // Set output state.
3862 set_control(_gvn.transform(thread_compare_rgn));
3863 set_all_memory(_gvn.transform(thread_compare_mem));
3864 }
3865
3866 #endif // JFR_HAVE_INTRINSICS
3867
3868 //------------------------inline_native_currentCarrierThread------------------
3869 bool LibraryCallKit::inline_native_currentCarrierThread() {
3870 Node* junk = nullptr;
3871 set_result(generate_current_thread(junk));
3872 return true;
3873 }
3874
3875 //------------------------inline_native_currentThread------------------
3876 bool LibraryCallKit::inline_native_currentThread() {
3877 Node* junk = nullptr;
3878 set_result(generate_virtual_thread(junk));
3879 return true;
3880 }
3881
3882 //------------------------inline_native_setVthread------------------
3883 bool LibraryCallKit::inline_native_setCurrentThread() {
3884 assert(C->method()->changes_current_thread(),
3885 "method changes current Thread but is not annotated ChangesCurrentThread");
3886 Node* arr = argument(1);
3887 Node* thread = _gvn.transform(new ThreadLocalNode());
3888 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3889 Node* thread_obj_handle
3890 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3891 thread_obj_handle = _gvn.transform(thread_obj_handle);
3892 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3893 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3894 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3895 return true;
3896 }
3897
3898 const Type* LibraryCallKit::scopedValueCache_type() {
3899 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3900 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3901 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS, /* stable= */ false, /* flat= */ false, /* not_flat= */ true, /* not_null_free= */ true);
3902
3903 // Because we create the scopedValue cache lazily we have to make the
3904 // type of the result BotPTR.
3905 bool xk = etype->klass_is_exact();
3906 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, TypeAryPtr::Offset(0));
3907 return objects_type;
3908 }
3909
3910 Node* LibraryCallKit::scopedValueCache_helper() {
3911 Node* thread = _gvn.transform(new ThreadLocalNode());
3912 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3913 // We cannot use immutable_memory() because we might flip onto a
3914 // different carrier thread, at which point we'll need to use that
3915 // carrier thread's cache.
3916 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3917 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3918 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3919 }
3920
3921 //------------------------inline_native_scopedValueCache------------------
3922 bool LibraryCallKit::inline_native_scopedValueCache() {
3923 Node* cache_obj_handle = scopedValueCache_helper();
3924 const Type* objects_type = scopedValueCache_type();
3925 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3926
3927 return true;
3928 }
3929
3930 //------------------------inline_native_setScopedValueCache------------------
3931 bool LibraryCallKit::inline_native_setScopedValueCache() {
3932 Node* arr = argument(0);
3933 Node* cache_obj_handle = scopedValueCache_helper();
3934 const Type* objects_type = scopedValueCache_type();
3935
3936 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3937 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3938
3939 return true;
3940 }
3941
3942 //------------------------inline_native_Continuation_pin and unpin-----------
3943
3944 // Shared implementation routine for both pin and unpin.
3945 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
3946 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3947
3948 // Save input memory.
3949 Node* input_memory_state = reset_memory();
3950 set_all_memory(input_memory_state);
3951
3952 // TLS
3953 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3954 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
3955 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
3956
3957 // Null check the last continuation object.
3958 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
3959 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
3960 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3961
3962 // False path, last continuation is null.
3963 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
3964
3965 // True path, last continuation is not null.
3966 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
3967
3968 set_control(continuation_is_not_null);
3969
3970 // Load the pin count from the last continuation.
3971 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
3972 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
3973
3974 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
3975 Node* pin_count_rhs;
3976 if (unpin) {
3977 pin_count_rhs = _gvn.intcon(0);
3978 } else {
3979 pin_count_rhs = _gvn.intcon(UINT32_MAX);
3980 }
3981 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
3982 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
3983 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
3984
3985 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
3986 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
3987 set_control(valid_pin_count);
3988
3989 Node* next_pin_count;
3990 if (unpin) {
3991 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
3992 } else {
3993 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
3994 }
3995
3996 Node* updated_pin_count_memory = store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, Compile::AliasIdxRaw, MemNode::unordered);
3997
3998 // True branch, pin count over/underflow.
3999 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
4000 {
4001 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
4002 // which will throw IllegalStateException for pin count over/underflow.
4003 PreserveJVMState pjvms(this);
4004 set_control(pin_count_over_underflow);
4005 set_all_memory(input_memory_state);
4006 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
4007 Deoptimization::Action_none);
4008 assert(stopped(), "invariant");
4009 }
4010
4011 // Result of top level CFG and Memory.
4012 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
4013 record_for_igvn(result_rgn);
4014 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
4015 record_for_igvn(result_mem);
4016
4017 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
4018 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
4019 result_mem->init_req(_true_path, _gvn.transform(updated_pin_count_memory));
4020 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
4021
4022 // Set output state.
4023 set_control(_gvn.transform(result_rgn));
4024 set_all_memory(_gvn.transform(result_mem));
4025
4026 return true;
4027 }
4028
4029 //-----------------------load_klass_from_mirror_common-------------------------
4030 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
4031 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
4032 // and branch to the given path on the region.
4033 // If never_see_null, take an uncommon trap on null, so we can optimistically
4034 // compile for the non-null case.
4035 // If the region is null, force never_see_null = true.
4036 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
4037 bool never_see_null,
4038 RegionNode* region,
4039 int null_path,
4040 int offset) {
4041 if (region == nullptr) never_see_null = true;
4042 Node* p = basic_plus_adr(mirror, offset);
4043 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4044 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
4045 Node* null_ctl = top();
4046 kls = null_check_oop(kls, &null_ctl, never_see_null);
4047 if (region != nullptr) {
4048 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
4049 region->init_req(null_path, null_ctl);
4050 } else {
4051 assert(null_ctl == top(), "no loose ends");
4052 }
4053 return kls;
4054 }
4055
4056 //--------------------(inline_native_Class_query helpers)---------------------
4057 // Use this for JVM_ACC_INTERFACE.
4058 // Fall through if (mods & mask) == bits, take the guard otherwise.
4059 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
4060 ByteSize offset, const Type* type, BasicType bt) {
4061 // Branch around if the given klass has the given modifier bit set.
4062 // Like generate_guard, adds a new path onto the region.
4063 Node* modp = basic_plus_adr(kls, in_bytes(offset));
4064 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
4065 Node* mask = intcon(modifier_mask);
4066 Node* bits = intcon(modifier_bits);
4067 Node* mbit = _gvn.transform(new AndINode(mods, mask));
4068 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
4069 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
4070 return generate_fair_guard(bol, region);
4071 }
4072
4073 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
4074 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
4075 Klass::access_flags_offset(), TypeInt::INT, T_INT);
4076 }
4077
4078 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
4079 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
4080 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
4081 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
4082 }
4083
4084 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
4085 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
4086 }
4087
4088 //-------------------------inline_native_Class_query-------------------
4089 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
4090 const Type* return_type = TypeInt::BOOL;
4091 Node* prim_return_value = top(); // what happens if it's a primitive class?
4092 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4093 bool expect_prim = false; // most of these guys expect to work on refs
4094
4095 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
4096
4097 Node* mirror = argument(0);
4098 Node* obj = top();
4099
4100 switch (id) {
4101 case vmIntrinsics::_isInstance:
4102 // nothing is an instance of a primitive type
4103 prim_return_value = intcon(0);
4104 obj = argument(1);
4105 break;
4106 case vmIntrinsics::_getModifiers:
4107 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
4108 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
4109 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
4110 break;
4111 case vmIntrinsics::_isInterface:
4112 prim_return_value = intcon(0);
4113 break;
4114 case vmIntrinsics::_isArray:
4115 prim_return_value = intcon(0);
4116 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
4117 break;
4118 case vmIntrinsics::_isPrimitive:
4119 prim_return_value = intcon(1);
4120 expect_prim = true; // obviously
4121 break;
4122 case vmIntrinsics::_isHidden:
4123 prim_return_value = intcon(0);
4124 break;
4125 case vmIntrinsics::_getSuperclass:
4126 prim_return_value = null();
4127 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
4128 break;
4129 case vmIntrinsics::_getClassAccessFlags:
4130 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
4131 return_type = TypeInt::INT; // not bool! 6297094
4132 break;
4133 default:
4134 fatal_unexpected_iid(id);
4135 break;
4136 }
4137
4138 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4139 if (mirror_con == nullptr) return false; // cannot happen?
4140
4141 #ifndef PRODUCT
4142 if (C->print_intrinsics() || C->print_inlining()) {
4143 ciType* k = mirror_con->java_mirror_type();
4144 if (k) {
4145 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4146 k->print_name();
4147 tty->cr();
4148 }
4149 }
4150 #endif
4151
4152 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4153 RegionNode* region = new RegionNode(PATH_LIMIT);
4154 record_for_igvn(region);
4155 PhiNode* phi = new PhiNode(region, return_type);
4156
4157 // The mirror will never be null of Reflection.getClassAccessFlags, however
4158 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4159 // if it is. See bug 4774291.
4160
4161 // For Reflection.getClassAccessFlags(), the null check occurs in
4162 // the wrong place; see inline_unsafe_access(), above, for a similar
4163 // situation.
4164 mirror = null_check(mirror);
4165 // If mirror or obj is dead, only null-path is taken.
4166 if (stopped()) return true;
4167
4168 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4169
4170 // Now load the mirror's klass metaobject, and null-check it.
4171 // Side-effects region with the control path if the klass is null.
4172 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4173 // If kls is null, we have a primitive mirror.
4174 phi->init_req(_prim_path, prim_return_value);
4175 if (stopped()) { set_result(region, phi); return true; }
4176 bool safe_for_replace = (region->in(_prim_path) == top());
4177
4178 Node* p; // handy temp
4179 Node* null_ctl;
4180
4181 // Now that we have the non-null klass, we can perform the real query.
4182 // For constant classes, the query will constant-fold in LoadNode::Value.
4183 Node* query_value = top();
4184 switch (id) {
4185 case vmIntrinsics::_isInstance:
4186 // nothing is an instance of a primitive type
4187 query_value = gen_instanceof(obj, kls, safe_for_replace);
4188 break;
4189
4190 case vmIntrinsics::_getModifiers:
4191 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
4192 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
4193 break;
4194
4195 case vmIntrinsics::_isInterface:
4196 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4197 if (generate_interface_guard(kls, region) != nullptr)
4198 // A guard was added. If the guard is taken, it was an interface.
4199 phi->add_req(intcon(1));
4200 // If we fall through, it's a plain class.
4201 query_value = intcon(0);
4202 break;
4203
4204 case vmIntrinsics::_isArray:
4205 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
4206 if (generate_array_guard(kls, region) != nullptr)
4207 // A guard was added. If the guard is taken, it was an array.
4208 phi->add_req(intcon(1));
4209 // If we fall through, it's a plain class.
4210 query_value = intcon(0);
4211 break;
4212
4213 case vmIntrinsics::_isPrimitive:
4214 query_value = intcon(0); // "normal" path produces false
4215 break;
4216
4217 case vmIntrinsics::_isHidden:
4218 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4219 if (generate_hidden_class_guard(kls, region) != nullptr)
4220 // A guard was added. If the guard is taken, it was an hidden class.
4221 phi->add_req(intcon(1));
4222 // If we fall through, it's a plain class.
4223 query_value = intcon(0);
4224 break;
4225
4226
4227 case vmIntrinsics::_getSuperclass:
4228 // The rules here are somewhat unfortunate, but we can still do better
4229 // with random logic than with a JNI call.
4230 // Interfaces store null or Object as _super, but must report null.
4231 // Arrays store an intermediate super as _super, but must report Object.
4232 // Other types can report the actual _super.
4233 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4234 if (generate_interface_guard(kls, region) != nullptr)
4235 // A guard was added. If the guard is taken, it was an interface.
4236 phi->add_req(null());
4237 if (generate_array_guard(kls, region) != nullptr)
4238 // A guard was added. If the guard is taken, it was an array.
4239 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4240 // If we fall through, it's a plain class. Get its _super.
4241 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4242 kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4243 null_ctl = top();
4244 kls = null_check_oop(kls, &null_ctl);
4245 if (null_ctl != top()) {
4246 // If the guard is taken, Object.superClass is null (both klass and mirror).
4247 region->add_req(null_ctl);
4248 phi ->add_req(null());
4249 }
4250 if (!stopped()) {
4251 query_value = load_mirror_from_klass(kls);
4252 }
4253 break;
4254
4255 case vmIntrinsics::_getClassAccessFlags:
4256 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
4257 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
4258 break;
4259
4260 default:
4261 fatal_unexpected_iid(id);
4262 break;
4263 }
4264
4265 // Fall-through is the normal case of a query to a real class.
4266 phi->init_req(1, query_value);
4267 region->init_req(1, control());
4268
4269 C->set_has_split_ifs(true); // Has chance for split-if optimization
4270 set_result(region, phi);
4271 return true;
4272 }
4273
4274
4275 //-------------------------inline_Class_cast-------------------
4276 bool LibraryCallKit::inline_Class_cast() {
4277 Node* mirror = argument(0); // Class
4278 Node* obj = argument(1);
4279 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4280 if (mirror_con == nullptr) {
4281 return false; // dead path (mirror->is_top()).
4282 }
4283 if (obj == nullptr || obj->is_top()) {
4284 return false; // dead path
4285 }
4286 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4287
4288 // First, see if Class.cast() can be folded statically.
4289 // java_mirror_type() returns non-null for compile-time Class constants.
4290 bool is_null_free_array = false;
4291 ciType* tm = mirror_con->java_mirror_type(&is_null_free_array);
4292 if (tm != nullptr && tm->is_klass() &&
4293 tp != nullptr) {
4294 if (!tp->is_loaded()) {
4295 // Don't use intrinsic when class is not loaded.
4296 return false;
4297 } else {
4298 const TypeKlassPtr* tklass = TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces);
4299 if (is_null_free_array) {
4300 tklass = tklass->is_aryklassptr()->cast_to_null_free();
4301 }
4302 int static_res = C->static_subtype_check(tklass, tp->as_klass_type());
4303 if (static_res == Compile::SSC_always_true) {
4304 // isInstance() is true - fold the code.
4305 set_result(obj);
4306 return true;
4307 } else if (static_res == Compile::SSC_always_false) {
4308 // Don't use intrinsic, have to throw ClassCastException.
4309 // If the reference is null, the non-intrinsic bytecode will
4310 // be optimized appropriately.
4311 return false;
4312 }
4313 }
4314 }
4315
4316 // Bailout intrinsic and do normal inlining if exception path is frequent.
4317 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4318 return false;
4319 }
4320
4321 // Generate dynamic checks.
4322 // Class.cast() is java implementation of _checkcast bytecode.
4323 // Do checkcast (Parse::do_checkcast()) optimizations here.
4324
4325 mirror = null_check(mirror);
4326 // If mirror is dead, only null-path is taken.
4327 if (stopped()) {
4328 return true;
4329 }
4330
4331 // Not-subtype or the mirror's klass ptr is nullptr (in case it is a primitive).
4332 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
4333 RegionNode* region = new RegionNode(PATH_LIMIT);
4334 record_for_igvn(region);
4335
4336 // Now load the mirror's klass metaobject, and null-check it.
4337 // If kls is null, we have a primitive mirror and
4338 // nothing is an instance of a primitive type.
4339 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4340
4341 Node* res = top();
4342 Node* io = i_o();
4343 Node* mem = merged_memory();
4344 if (!stopped()) {
4345
4346 Node* bad_type_ctrl = top();
4347 // Do checkcast optimizations.
4348 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4349 region->init_req(_bad_type_path, bad_type_ctrl);
4350 }
4351 if (region->in(_prim_path) != top() ||
4352 region->in(_bad_type_path) != top() ||
4353 region->in(_npe_path) != top()) {
4354 // Let Interpreter throw ClassCastException.
4355 PreserveJVMState pjvms(this);
4356 set_control(_gvn.transform(region));
4357 // Set IO and memory because gen_checkcast may override them when buffering inline types
4358 set_i_o(io);
4359 set_all_memory(mem);
4360 uncommon_trap(Deoptimization::Reason_intrinsic,
4361 Deoptimization::Action_maybe_recompile);
4362 }
4363 if (!stopped()) {
4364 set_result(res);
4365 }
4366 return true;
4367 }
4368
4369
4370 //--------------------------inline_native_subtype_check------------------------
4371 // This intrinsic takes the JNI calls out of the heart of
4372 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4373 bool LibraryCallKit::inline_native_subtype_check() {
4374 // Pull both arguments off the stack.
4375 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4376 args[0] = argument(0);
4377 args[1] = argument(1);
4378 Node* klasses[2]; // corresponding Klasses: superk, subk
4379 klasses[0] = klasses[1] = top();
4380
4381 enum {
4382 // A full decision tree on {superc is prim, subc is prim}:
4383 _prim_0_path = 1, // {P,N} => false
4384 // {P,P} & superc!=subc => false
4385 _prim_same_path, // {P,P} & superc==subc => true
4386 _prim_1_path, // {N,P} => false
4387 _ref_subtype_path, // {N,N} & subtype check wins => true
4388 _both_ref_path, // {N,N} & subtype check loses => false
4389 PATH_LIMIT
4390 };
4391
4392 RegionNode* region = new RegionNode(PATH_LIMIT);
4393 RegionNode* prim_region = new RegionNode(2);
4394 Node* phi = new PhiNode(region, TypeInt::BOOL);
4395 record_for_igvn(region);
4396 record_for_igvn(prim_region);
4397
4398 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4399 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4400 int class_klass_offset = java_lang_Class::klass_offset();
4401
4402 // First null-check both mirrors and load each mirror's klass metaobject.
4403 int which_arg;
4404 for (which_arg = 0; which_arg <= 1; which_arg++) {
4405 Node* arg = args[which_arg];
4406 arg = null_check(arg);
4407 if (stopped()) break;
4408 args[which_arg] = arg;
4409
4410 Node* p = basic_plus_adr(arg, class_klass_offset);
4411 Node* kls = LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, adr_type, kls_type);
4412 klasses[which_arg] = _gvn.transform(kls);
4413 }
4414
4415 // Having loaded both klasses, test each for null.
4416 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4417 for (which_arg = 0; which_arg <= 1; which_arg++) {
4418 Node* kls = klasses[which_arg];
4419 Node* null_ctl = top();
4420 kls = null_check_oop(kls, &null_ctl, never_see_null);
4421 if (which_arg == 0) {
4422 prim_region->init_req(1, null_ctl);
4423 } else {
4424 region->init_req(_prim_1_path, null_ctl);
4425 }
4426 if (stopped()) break;
4427 klasses[which_arg] = kls;
4428 }
4429
4430 if (!stopped()) {
4431 // now we have two reference types, in klasses[0..1]
4432 Node* subk = klasses[1]; // the argument to isAssignableFrom
4433 Node* superk = klasses[0]; // the receiver
4434 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4435 region->set_req(_ref_subtype_path, control());
4436 }
4437
4438 // If both operands are primitive (both klasses null), then
4439 // we must return true when they are identical primitives.
4440 // It is convenient to test this after the first null klass check.
4441 // This path is also used if superc is a value mirror.
4442 set_control(_gvn.transform(prim_region));
4443 if (!stopped()) {
4444 // Since superc is primitive, make a guard for the superc==subc case.
4445 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4446 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4447 generate_fair_guard(bol_eq, region);
4448 if (region->req() == PATH_LIMIT+1) {
4449 // A guard was added. If the added guard is taken, superc==subc.
4450 region->swap_edges(PATH_LIMIT, _prim_same_path);
4451 region->del_req(PATH_LIMIT);
4452 }
4453 region->set_req(_prim_0_path, control()); // Not equal after all.
4454 }
4455
4456 // these are the only paths that produce 'true':
4457 phi->set_req(_prim_same_path, intcon(1));
4458 phi->set_req(_ref_subtype_path, intcon(1));
4459
4460 // pull together the cases:
4461 assert(region->req() == PATH_LIMIT, "sane region");
4462 for (uint i = 1; i < region->req(); i++) {
4463 Node* ctl = region->in(i);
4464 if (ctl == nullptr || ctl == top()) {
4465 region->set_req(i, top());
4466 phi ->set_req(i, top());
4467 } else if (phi->in(i) == nullptr) {
4468 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4469 }
4470 }
4471
4472 set_control(_gvn.transform(region));
4473 set_result(_gvn.transform(phi));
4474 return true;
4475 }
4476
4477 //---------------------generate_array_guard_common------------------------
4478 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind) {
4479
4480 if (stopped()) {
4481 return nullptr;
4482 }
4483
4484 // Like generate_guard, adds a new path onto the region.
4485 jint layout_con = 0;
4486 Node* layout_val = get_layout_helper(kls, layout_con);
4487 if (layout_val == nullptr) {
4488 bool query = 0;
4489 switch(kind) {
4490 case ObjectArray: query = Klass::layout_helper_is_objArray(layout_con); break;
4491 case NonObjectArray: query = !Klass::layout_helper_is_objArray(layout_con); break;
4492 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
4493 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
4494 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
4495 default:
4496 ShouldNotReachHere();
4497 }
4498 if (!query) {
4499 return nullptr; // never a branch
4500 } else { // always a branch
4501 Node* always_branch = control();
4502 if (region != nullptr)
4503 region->add_req(always_branch);
4504 set_control(top());
4505 return always_branch;
4506 }
4507 }
4508 unsigned int value = 0;
4509 BoolTest::mask btest = BoolTest::illegal;
4510 switch(kind) {
4511 case ObjectArray:
4512 case NonObjectArray: {
4513 value = Klass::_lh_array_tag_obj_value;
4514 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4515 btest = (kind == ObjectArray) ? BoolTest::eq : BoolTest::ne;
4516 break;
4517 }
4518 case TypeArray: {
4519 value = Klass::_lh_array_tag_type_value;
4520 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4521 btest = BoolTest::eq;
4522 break;
4523 }
4524 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
4525 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
4526 default:
4527 ShouldNotReachHere();
4528 }
4529 // Now test the correct condition.
4530 jint nval = (jint)value;
4531 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4532 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4533 return generate_fair_guard(bol, region);
4534 }
4535
4536 //-----------------------inline_newNullRestrictedArray--------------------------
4537 // public static native Object[] newNullRestrictedArray(Class<?> componentType, int length);
4538 bool LibraryCallKit::inline_newNullRestrictedArray() {
4539 Node* componentType = argument(0);
4540 Node* length = argument(1);
4541
4542 const TypeInstPtr* tp = _gvn.type(componentType)->isa_instptr();
4543 if (tp != nullptr) {
4544 ciInstanceKlass* ik = tp->instance_klass();
4545 if (ik == C->env()->Class_klass()) {
4546 ciType* t = tp->java_mirror_type();
4547 if (t != nullptr && t->is_inlinetype()) {
4548 ciArrayKlass* array_klass = ciArrayKlass::make(t, true);
4549 if (array_klass->is_loaded() && array_klass->element_klass()->as_inline_klass()->is_initialized()) {
4550 const TypeAryKlassPtr* array_klass_type = TypeKlassPtr::make(array_klass, Type::trust_interfaces)->is_aryklassptr();
4551 array_klass_type = array_klass_type->cast_to_null_free();
4552 Node* obj = new_array(makecon(array_klass_type), length, 0, nullptr, false); // no arguments to push
4553 set_result(obj);
4554 assert(gvn().type(obj)->is_aryptr()->is_null_free(), "must be null-free");
4555 return true;
4556 }
4557 }
4558 }
4559 }
4560 return false;
4561 }
4562
4563 //-----------------------inline_native_newArray--------------------------
4564 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
4565 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4566 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4567 Node* mirror;
4568 Node* count_val;
4569 if (uninitialized) {
4570 null_check_receiver();
4571 mirror = argument(1);
4572 count_val = argument(2);
4573 } else {
4574 mirror = argument(0);
4575 count_val = argument(1);
4576 }
4577
4578 mirror = null_check(mirror);
4579 // If mirror or obj is dead, only null-path is taken.
4580 if (stopped()) return true;
4581
4582 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4583 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4584 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4585 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4586 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4587
4588 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4589 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4590 result_reg, _slow_path);
4591 Node* normal_ctl = control();
4592 Node* no_array_ctl = result_reg->in(_slow_path);
4593
4594 // Generate code for the slow case. We make a call to newArray().
4595 set_control(no_array_ctl);
4596 if (!stopped()) {
4597 // Either the input type is void.class, or else the
4598 // array klass has not yet been cached. Either the
4599 // ensuing call will throw an exception, or else it
4600 // will cache the array klass for next time.
4601 PreserveJVMState pjvms(this);
4602 CallJavaNode* slow_call = nullptr;
4603 if (uninitialized) {
4604 // Generate optimized virtual call (holder class 'Unsafe' is final)
4605 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4606 } else {
4607 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4608 }
4609 Node* slow_result = set_results_for_java_call(slow_call);
4610 // this->control() comes from set_results_for_java_call
4611 result_reg->set_req(_slow_path, control());
4612 result_val->set_req(_slow_path, slow_result);
4613 result_io ->set_req(_slow_path, i_o());
4614 result_mem->set_req(_slow_path, reset_memory());
4615 }
4616
4617 set_control(normal_ctl);
4618 if (!stopped()) {
4619 // Normal case: The array type has been cached in the java.lang.Class.
4620 // The following call works fine even if the array type is polymorphic.
4621 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4622 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4623 result_reg->init_req(_normal_path, control());
4624 result_val->init_req(_normal_path, obj);
4625 result_io ->init_req(_normal_path, i_o());
4626 result_mem->init_req(_normal_path, reset_memory());
4627
4628 if (uninitialized) {
4629 // Mark the allocation so that zeroing is skipped
4630 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4631 alloc->maybe_set_complete(&_gvn);
4632 }
4633 }
4634
4635 // Return the combined state.
4636 set_i_o( _gvn.transform(result_io) );
4637 set_all_memory( _gvn.transform(result_mem));
4638
4639 C->set_has_split_ifs(true); // Has chance for split-if optimization
4640 set_result(result_reg, result_val);
4641 return true;
4642 }
4643
4644 //----------------------inline_native_getLength--------------------------
4645 // public static native int java.lang.reflect.Array.getLength(Object array);
4646 bool LibraryCallKit::inline_native_getLength() {
4647 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4648
4649 Node* array = null_check(argument(0));
4650 // If array is dead, only null-path is taken.
4651 if (stopped()) return true;
4652
4653 // Deoptimize if it is a non-array.
4654 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr);
4655
4656 if (non_array != nullptr) {
4657 PreserveJVMState pjvms(this);
4658 set_control(non_array);
4659 uncommon_trap(Deoptimization::Reason_intrinsic,
4660 Deoptimization::Action_maybe_recompile);
4661 }
4662
4663 // If control is dead, only non-array-path is taken.
4664 if (stopped()) return true;
4665
4666 // The works fine even if the array type is polymorphic.
4667 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4668 Node* result = load_array_length(array);
4669
4670 C->set_has_split_ifs(true); // Has chance for split-if optimization
4671 set_result(result);
4672 return true;
4673 }
4674
4675 //------------------------inline_array_copyOf----------------------------
4676 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4677 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4678 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4679 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4680
4681 // Get the arguments.
4682 Node* original = argument(0);
4683 Node* start = is_copyOfRange? argument(1): intcon(0);
4684 Node* end = is_copyOfRange? argument(2): argument(1);
4685 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4686
4687 Node* newcopy = nullptr;
4688
4689 // Set the original stack and the reexecute bit for the interpreter to reexecute
4690 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4691 { PreserveReexecuteState preexecs(this);
4692 jvms()->set_should_reexecute(true);
4693
4694 array_type_mirror = null_check(array_type_mirror);
4695 original = null_check(original);
4696
4697 // Check if a null path was taken unconditionally.
4698 if (stopped()) return true;
4699
4700 Node* orig_length = load_array_length(original);
4701
4702 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4703 klass_node = null_check(klass_node);
4704
4705 RegionNode* bailout = new RegionNode(1);
4706 record_for_igvn(bailout);
4707
4708 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4709 // Bail out if that is so.
4710 // Inline type array may have object field that would require a
4711 // write barrier. Conservatively, go to slow path.
4712 // TODO 8251971: Optimize for the case when flat src/dst are later found
4713 // to not contain oops (i.e., move this check to the macro expansion phase).
4714 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4715 const TypeAryPtr* orig_t = _gvn.type(original)->isa_aryptr();
4716 const TypeKlassPtr* tklass = _gvn.type(klass_node)->is_klassptr();
4717 bool exclude_flat = UseFlatArray && bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing) &&
4718 // Can src array be flat and contain oops?
4719 (orig_t == nullptr || (!orig_t->is_not_flat() && (!orig_t->is_flat() || orig_t->elem()->inline_klass()->contains_oops()))) &&
4720 // Can dest array be flat and contain oops?
4721 tklass->can_be_inline_array() && (!tklass->is_flat() || tklass->is_aryklassptr()->elem()->is_instklassptr()->instance_klass()->as_inline_klass()->contains_oops());
4722 Node* not_objArray = exclude_flat ? generate_non_objArray_guard(klass_node, bailout) : generate_typeArray_guard(klass_node, bailout);
4723 if (not_objArray != nullptr) {
4724 // Improve the klass node's type from the new optimistic assumption:
4725 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4726 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0));
4727 Node* cast = new CastPPNode(control(), klass_node, akls);
4728 klass_node = _gvn.transform(cast);
4729 }
4730
4731 // Bail out if either start or end is negative.
4732 generate_negative_guard(start, bailout, &start);
4733 generate_negative_guard(end, bailout, &end);
4734
4735 Node* length = end;
4736 if (_gvn.type(start) != TypeInt::ZERO) {
4737 length = _gvn.transform(new SubINode(end, start));
4738 }
4739
4740 // Bail out if length is negative (i.e., if start > end).
4741 // Without this the new_array would throw
4742 // NegativeArraySizeException but IllegalArgumentException is what
4743 // should be thrown
4744 generate_negative_guard(length, bailout, &length);
4745
4746 // Handle inline type arrays
4747 bool can_validate = !too_many_traps(Deoptimization::Reason_class_check);
4748 if (!stopped()) {
4749 // TODO JDK-8329224
4750 if (!orig_t->is_null_free()) {
4751 // Not statically known to be null free, add a check
4752 generate_fair_guard(null_free_array_test(original), bailout);
4753 }
4754 orig_t = _gvn.type(original)->isa_aryptr();
4755 if (orig_t != nullptr && orig_t->is_flat()) {
4756 // Src is flat, check that dest is flat as well
4757 if (exclude_flat) {
4758 // Dest can't be flat, bail out
4759 bailout->add_req(control());
4760 set_control(top());
4761 } else {
4762 generate_fair_guard(flat_array_test(klass_node, /* flat = */ false), bailout);
4763 }
4764 } else if (UseFlatArray && (orig_t == nullptr || !orig_t->is_not_flat()) &&
4765 // If dest is flat, src must be flat as well (guaranteed by src <: dest check if validated).
4766 ((!tklass->is_flat() && tklass->can_be_inline_array()) || !can_validate)) {
4767 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
4768 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
4769 generate_fair_guard(flat_array_test(load_object_klass(original)), bailout);
4770 if (orig_t != nullptr) {
4771 orig_t = orig_t->cast_to_not_flat();
4772 original = _gvn.transform(new CheckCastPPNode(control(), original, orig_t));
4773 }
4774 }
4775 if (!can_validate) {
4776 // No validation. The subtype check emitted at macro expansion time will not go to the slow
4777 // path but call checkcast_arraycopy which can not handle flat/null-free inline type arrays.
4778 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat/null-free.
4779 generate_fair_guard(flat_array_test(klass_node), bailout);
4780 generate_fair_guard(null_free_array_test(original), bailout);
4781 }
4782 }
4783
4784 // Bail out if start is larger than the original length
4785 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4786 generate_negative_guard(orig_tail, bailout, &orig_tail);
4787
4788 if (bailout->req() > 1) {
4789 PreserveJVMState pjvms(this);
4790 set_control(_gvn.transform(bailout));
4791 uncommon_trap(Deoptimization::Reason_intrinsic,
4792 Deoptimization::Action_maybe_recompile);
4793 }
4794
4795 if (!stopped()) {
4796 // How many elements will we copy from the original?
4797 // The answer is MinI(orig_tail, length).
4798 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4799
4800 // Generate a direct call to the right arraycopy function(s).
4801 // We know the copy is disjoint but we might not know if the
4802 // oop stores need checking.
4803 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4804 // This will fail a store-check if x contains any non-nulls.
4805
4806 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4807 // loads/stores but it is legal only if we're sure the
4808 // Arrays.copyOf would succeed. So we need all input arguments
4809 // to the copyOf to be validated, including that the copy to the
4810 // new array won't trigger an ArrayStoreException. That subtype
4811 // check can be optimized if we know something on the type of
4812 // the input array from type speculation.
4813 if (_gvn.type(klass_node)->singleton()) {
4814 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4815 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4816
4817 int test = C->static_subtype_check(superk, subk);
4818 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4819 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4820 if (t_original->speculative_type() != nullptr) {
4821 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4822 }
4823 }
4824 }
4825
4826 bool validated = false;
4827 // Reason_class_check rather than Reason_intrinsic because we
4828 // want to intrinsify even if this traps.
4829 if (can_validate) {
4830 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4831
4832 if (not_subtype_ctrl != top()) {
4833 PreserveJVMState pjvms(this);
4834 set_control(not_subtype_ctrl);
4835 uncommon_trap(Deoptimization::Reason_class_check,
4836 Deoptimization::Action_make_not_entrant);
4837 assert(stopped(), "Should be stopped");
4838 }
4839 validated = true;
4840 }
4841
4842 if (!stopped()) {
4843 newcopy = new_array(klass_node, length, 0); // no arguments to push
4844
4845 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4846 load_object_klass(original), klass_node);
4847 if (!is_copyOfRange) {
4848 ac->set_copyof(validated);
4849 } else {
4850 ac->set_copyofrange(validated);
4851 }
4852 Node* n = _gvn.transform(ac);
4853 if (n == ac) {
4854 ac->connect_outputs(this);
4855 } else {
4856 assert(validated, "shouldn't transform if all arguments not validated");
4857 set_all_memory(n);
4858 }
4859 }
4860 }
4861 } // original reexecute is set back here
4862
4863 C->set_has_split_ifs(true); // Has chance for split-if optimization
4864 if (!stopped()) {
4865 set_result(newcopy);
4866 }
4867 return true;
4868 }
4869
4870
4871 //----------------------generate_virtual_guard---------------------------
4872 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4873 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4874 RegionNode* slow_region) {
4875 ciMethod* method = callee();
4876 int vtable_index = method->vtable_index();
4877 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4878 "bad index %d", vtable_index);
4879 // Get the Method* out of the appropriate vtable entry.
4880 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4881 vtable_index*vtableEntry::size_in_bytes() +
4882 in_bytes(vtableEntry::method_offset());
4883 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4884 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4885
4886 // Compare the target method with the expected method (e.g., Object.hashCode).
4887 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4888
4889 Node* native_call = makecon(native_call_addr);
4890 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4891 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4892
4893 return generate_slow_guard(test_native, slow_region);
4894 }
4895
4896 //-----------------------generate_method_call----------------------------
4897 // Use generate_method_call to make a slow-call to the real
4898 // method if the fast path fails. An alternative would be to
4899 // use a stub like OptoRuntime::slow_arraycopy_Java.
4900 // This only works for expanding the current library call,
4901 // not another intrinsic. (E.g., don't use this for making an
4902 // arraycopy call inside of the copyOf intrinsic.)
4903 CallJavaNode*
4904 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4905 // When compiling the intrinsic method itself, do not use this technique.
4906 guarantee(callee() != C->method(), "cannot make slow-call to self");
4907
4908 ciMethod* method = callee();
4909 // ensure the JVMS we have will be correct for this call
4910 guarantee(method_id == method->intrinsic_id(), "must match");
4911
4912 const TypeFunc* tf = TypeFunc::make(method);
4913 if (res_not_null) {
4914 assert(tf->return_type() == T_OBJECT, "");
4915 const TypeTuple* range = tf->range_cc();
4916 const Type** fields = TypeTuple::fields(range->cnt());
4917 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4918 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4919 tf = TypeFunc::make(tf->domain_cc(), new_range);
4920 }
4921 CallJavaNode* slow_call;
4922 if (is_static) {
4923 assert(!is_virtual, "");
4924 slow_call = new CallStaticJavaNode(C, tf,
4925 SharedRuntime::get_resolve_static_call_stub(), method);
4926 } else if (is_virtual) {
4927 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4928 int vtable_index = Method::invalid_vtable_index;
4929 if (UseInlineCaches) {
4930 // Suppress the vtable call
4931 } else {
4932 // hashCode and clone are not a miranda methods,
4933 // so the vtable index is fixed.
4934 // No need to use the linkResolver to get it.
4935 vtable_index = method->vtable_index();
4936 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4937 "bad index %d", vtable_index);
4938 }
4939 slow_call = new CallDynamicJavaNode(tf,
4940 SharedRuntime::get_resolve_virtual_call_stub(),
4941 method, vtable_index);
4942 } else { // neither virtual nor static: opt_virtual
4943 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4944 slow_call = new CallStaticJavaNode(C, tf,
4945 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4946 slow_call->set_optimized_virtual(true);
4947 }
4948 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4949 // To be able to issue a direct call (optimized virtual or virtual)
4950 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4951 // about the method being invoked should be attached to the call site to
4952 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4953 slow_call->set_override_symbolic_info(true);
4954 }
4955 set_arguments_for_java_call(slow_call);
4956 set_edges_for_java_call(slow_call);
4957 return slow_call;
4958 }
4959
4960
4961 /**
4962 * Build special case code for calls to hashCode on an object. This call may
4963 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4964 * slightly different code.
4965 */
4966 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4967 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4968 assert(!(is_virtual && is_static), "either virtual, special, or static");
4969
4970 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4971
4972 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4973 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4974 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4975 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4976 Node* obj = argument(0);
4977
4978 // Don't intrinsify hashcode on inline types for now.
4979 // The "is locked" runtime check below also serves as inline type check and goes to the slow path.
4980 if (gvn().type(obj)->is_inlinetypeptr()) {
4981 return false;
4982 }
4983
4984 if (!is_static) {
4985 // Check for hashing null object
4986 obj = null_check_receiver();
4987 if (stopped()) return true; // unconditionally null
4988 result_reg->init_req(_null_path, top());
4989 result_val->init_req(_null_path, top());
4990 } else {
4991 // Do a null check, and return zero if null.
4992 // System.identityHashCode(null) == 0
4993 Node* null_ctl = top();
4994 obj = null_check_oop(obj, &null_ctl);
4995 result_reg->init_req(_null_path, null_ctl);
4996 result_val->init_req(_null_path, _gvn.intcon(0));
4997 }
4998
4999 // Unconditionally null? Then return right away.
5000 if (stopped()) {
5001 set_control( result_reg->in(_null_path));
5002 if (!stopped())
5003 set_result(result_val->in(_null_path));
5004 return true;
5005 }
5006
5007 // We only go to the fast case code if we pass a number of guards. The
5008 // paths which do not pass are accumulated in the slow_region.
5009 RegionNode* slow_region = new RegionNode(1);
5010 record_for_igvn(slow_region);
5011
5012 // If this is a virtual call, we generate a funny guard. We pull out
5013 // the vtable entry corresponding to hashCode() from the target object.
5014 // If the target method which we are calling happens to be the native
5015 // Object hashCode() method, we pass the guard. We do not need this
5016 // guard for non-virtual calls -- the caller is known to be the native
5017 // Object hashCode().
5018 if (is_virtual) {
5019 // After null check, get the object's klass.
5020 Node* obj_klass = load_object_klass(obj);
5021 generate_virtual_guard(obj_klass, slow_region);
5022 }
5023
5024 // Get the header out of the object, use LoadMarkNode when available
5025 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
5026 // The control of the load must be null. Otherwise, the load can move before
5027 // the null check after castPP removal.
5028 Node* no_ctrl = nullptr;
5029 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
5030
5031 if (!UseObjectMonitorTable) {
5032 // Test the header to see if it is safe to read w.r.t. locking.
5033 // This also serves as guard against inline types
5034 Node *lock_mask = _gvn.MakeConX(markWord::inline_type_mask_in_place);
5035 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
5036 if (LockingMode == LM_LIGHTWEIGHT) {
5037 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
5038 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
5039 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
5040
5041 generate_slow_guard(test_monitor, slow_region);
5042 } else {
5043 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
5044 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val));
5045 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne));
5046
5047 generate_slow_guard(test_not_unlocked, slow_region);
5048 }
5049 }
5050
5051 // Get the hash value and check to see that it has been properly assigned.
5052 // We depend on hash_mask being at most 32 bits and avoid the use of
5053 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
5054 // vm: see markWord.hpp.
5055 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
5056 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
5057 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
5058 // This hack lets the hash bits live anywhere in the mark object now, as long
5059 // as the shift drops the relevant bits into the low 32 bits. Note that
5060 // Java spec says that HashCode is an int so there's no point in capturing
5061 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
5062 hshifted_header = ConvX2I(hshifted_header);
5063 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
5064
5065 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
5066 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
5067 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
5068
5069 generate_slow_guard(test_assigned, slow_region);
5070
5071 Node* init_mem = reset_memory();
5072 // fill in the rest of the null path:
5073 result_io ->init_req(_null_path, i_o());
5074 result_mem->init_req(_null_path, init_mem);
5075
5076 result_val->init_req(_fast_path, hash_val);
5077 result_reg->init_req(_fast_path, control());
5078 result_io ->init_req(_fast_path, i_o());
5079 result_mem->init_req(_fast_path, init_mem);
5080
5081 // Generate code for the slow case. We make a call to hashCode().
5082 set_control(_gvn.transform(slow_region));
5083 if (!stopped()) {
5084 // No need for PreserveJVMState, because we're using up the present state.
5085 set_all_memory(init_mem);
5086 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
5087 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
5088 Node* slow_result = set_results_for_java_call(slow_call);
5089 // this->control() comes from set_results_for_java_call
5090 result_reg->init_req(_slow_path, control());
5091 result_val->init_req(_slow_path, slow_result);
5092 result_io ->set_req(_slow_path, i_o());
5093 result_mem ->set_req(_slow_path, reset_memory());
5094 }
5095
5096 // Return the combined state.
5097 set_i_o( _gvn.transform(result_io) );
5098 set_all_memory( _gvn.transform(result_mem));
5099
5100 set_result(result_reg, result_val);
5101 return true;
5102 }
5103
5104 //---------------------------inline_native_getClass----------------------------
5105 // public final native Class<?> java.lang.Object.getClass();
5106 //
5107 // Build special case code for calls to getClass on an object.
5108 bool LibraryCallKit::inline_native_getClass() {
5109 Node* obj = argument(0);
5110 if (obj->is_InlineType()) {
5111 const Type* t = _gvn.type(obj);
5112 if (t->maybe_null()) {
5113 null_check(obj);
5114 }
5115 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror())));
5116 return true;
5117 }
5118 obj = null_check_receiver();
5119 if (stopped()) return true;
5120 set_result(load_mirror_from_klass(load_object_klass(obj)));
5121 return true;
5122 }
5123
5124 //-----------------inline_native_Reflection_getCallerClass---------------------
5125 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
5126 //
5127 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
5128 //
5129 // NOTE: This code must perform the same logic as JVM_GetCallerClass
5130 // in that it must skip particular security frames and checks for
5131 // caller sensitive methods.
5132 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
5133 #ifndef PRODUCT
5134 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5135 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
5136 }
5137 #endif
5138
5139 if (!jvms()->has_method()) {
5140 #ifndef PRODUCT
5141 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5142 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
5143 }
5144 #endif
5145 return false;
5146 }
5147
5148 // Walk back up the JVM state to find the caller at the required
5149 // depth.
5150 JVMState* caller_jvms = jvms();
5151
5152 // Cf. JVM_GetCallerClass
5153 // NOTE: Start the loop at depth 1 because the current JVM state does
5154 // not include the Reflection.getCallerClass() frame.
5155 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
5156 ciMethod* m = caller_jvms->method();
5157 switch (n) {
5158 case 0:
5159 fatal("current JVM state does not include the Reflection.getCallerClass frame");
5160 break;
5161 case 1:
5162 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
5163 if (!m->caller_sensitive()) {
5164 #ifndef PRODUCT
5165 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5166 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
5167 }
5168 #endif
5169 return false; // bail-out; let JVM_GetCallerClass do the work
5170 }
5171 break;
5172 default:
5173 if (!m->is_ignored_by_security_stack_walk()) {
5174 // We have reached the desired frame; return the holder class.
5175 // Acquire method holder as java.lang.Class and push as constant.
5176 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
5177 ciInstance* caller_mirror = caller_klass->java_mirror();
5178 set_result(makecon(TypeInstPtr::make(caller_mirror)));
5179
5180 #ifndef PRODUCT
5181 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5182 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());
5183 tty->print_cr(" JVM state at this point:");
5184 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5185 ciMethod* m = jvms()->of_depth(i)->method();
5186 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5187 }
5188 }
5189 #endif
5190 return true;
5191 }
5192 break;
5193 }
5194 }
5195
5196 #ifndef PRODUCT
5197 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5198 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
5199 tty->print_cr(" JVM state at this point:");
5200 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5201 ciMethod* m = jvms()->of_depth(i)->method();
5202 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5203 }
5204 }
5205 #endif
5206
5207 return false; // bail-out; let JVM_GetCallerClass do the work
5208 }
5209
5210 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
5211 Node* arg = argument(0);
5212 Node* result = nullptr;
5213
5214 switch (id) {
5215 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
5216 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
5217 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5218 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5219 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5220 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5221
5222 case vmIntrinsics::_doubleToLongBits: {
5223 // two paths (plus control) merge in a wood
5224 RegionNode *r = new RegionNode(3);
5225 Node *phi = new PhiNode(r, TypeLong::LONG);
5226
5227 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5228 // Build the boolean node
5229 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5230
5231 // Branch either way.
5232 // NaN case is less traveled, which makes all the difference.
5233 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5234 Node *opt_isnan = _gvn.transform(ifisnan);
5235 assert( opt_isnan->is_If(), "Expect an IfNode");
5236 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5237 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5238
5239 set_control(iftrue);
5240
5241 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5242 Node *slow_result = longcon(nan_bits); // return NaN
5243 phi->init_req(1, _gvn.transform( slow_result ));
5244 r->init_req(1, iftrue);
5245
5246 // Else fall through
5247 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5248 set_control(iffalse);
5249
5250 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5251 r->init_req(2, iffalse);
5252
5253 // Post merge
5254 set_control(_gvn.transform(r));
5255 record_for_igvn(r);
5256
5257 C->set_has_split_ifs(true); // Has chance for split-if optimization
5258 result = phi;
5259 assert(result->bottom_type()->isa_long(), "must be");
5260 break;
5261 }
5262
5263 case vmIntrinsics::_floatToIntBits: {
5264 // two paths (plus control) merge in a wood
5265 RegionNode *r = new RegionNode(3);
5266 Node *phi = new PhiNode(r, TypeInt::INT);
5267
5268 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5269 // Build the boolean node
5270 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5271
5272 // Branch either way.
5273 // NaN case is less traveled, which makes all the difference.
5274 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5275 Node *opt_isnan = _gvn.transform(ifisnan);
5276 assert( opt_isnan->is_If(), "Expect an IfNode");
5277 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5278 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5279
5280 set_control(iftrue);
5281
5282 static const jint nan_bits = 0x7fc00000;
5283 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5284 phi->init_req(1, _gvn.transform( slow_result ));
5285 r->init_req(1, iftrue);
5286
5287 // Else fall through
5288 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5289 set_control(iffalse);
5290
5291 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5292 r->init_req(2, iffalse);
5293
5294 // Post merge
5295 set_control(_gvn.transform(r));
5296 record_for_igvn(r);
5297
5298 C->set_has_split_ifs(true); // Has chance for split-if optimization
5299 result = phi;
5300 assert(result->bottom_type()->isa_int(), "must be");
5301 break;
5302 }
5303
5304 default:
5305 fatal_unexpected_iid(id);
5306 break;
5307 }
5308 set_result(_gvn.transform(result));
5309 return true;
5310 }
5311
5312 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5313 Node* arg = argument(0);
5314 Node* result = nullptr;
5315
5316 switch (id) {
5317 case vmIntrinsics::_floatIsInfinite:
5318 result = new IsInfiniteFNode(arg);
5319 break;
5320 case vmIntrinsics::_floatIsFinite:
5321 result = new IsFiniteFNode(arg);
5322 break;
5323 case vmIntrinsics::_doubleIsInfinite:
5324 result = new IsInfiniteDNode(arg);
5325 break;
5326 case vmIntrinsics::_doubleIsFinite:
5327 result = new IsFiniteDNode(arg);
5328 break;
5329 default:
5330 fatal_unexpected_iid(id);
5331 break;
5332 }
5333 set_result(_gvn.transform(result));
5334 return true;
5335 }
5336
5337 //----------------------inline_unsafe_copyMemory-------------------------
5338 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5339
5340 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5341 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5342 const Type* base_t = gvn.type(base);
5343
5344 bool in_native = (base_t == TypePtr::NULL_PTR);
5345 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5346 bool is_mixed = !in_heap && !in_native;
5347
5348 if (is_mixed) {
5349 return true; // mixed accesses can touch both on-heap and off-heap memory
5350 }
5351 if (in_heap) {
5352 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5353 if (!is_prim_array) {
5354 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5355 // there's not enough type information available to determine proper memory slice for it.
5356 return true;
5357 }
5358 }
5359 return false;
5360 }
5361
5362 bool LibraryCallKit::inline_unsafe_copyMemory() {
5363 if (callee()->is_static()) return false; // caller must have the capability!
5364 null_check_receiver(); // null-check receiver
5365 if (stopped()) return true;
5366
5367 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5368
5369 Node* src_base = argument(1); // type: oop
5370 Node* src_off = ConvL2X(argument(2)); // type: long
5371 Node* dst_base = argument(4); // type: oop
5372 Node* dst_off = ConvL2X(argument(5)); // type: long
5373 Node* size = ConvL2X(argument(7)); // type: long
5374
5375 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5376 "fieldOffset must be byte-scaled");
5377
5378 Node* src_addr = make_unsafe_address(src_base, src_off);
5379 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5380
5381 Node* thread = _gvn.transform(new ThreadLocalNode());
5382 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5383 BasicType doing_unsafe_access_bt = T_BYTE;
5384 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5385
5386 // update volatile field
5387 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5388
5389 int flags = RC_LEAF | RC_NO_FP;
5390
5391 const TypePtr* dst_type = TypePtr::BOTTOM;
5392
5393 // Adjust memory effects of the runtime call based on input values.
5394 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5395 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5396 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5397
5398 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5399 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5400 flags |= RC_NARROW_MEM; // narrow in memory
5401 }
5402 }
5403
5404 // Call it. Note that the length argument is not scaled.
5405 make_runtime_call(flags,
5406 OptoRuntime::fast_arraycopy_Type(),
5407 StubRoutines::unsafe_arraycopy(),
5408 "unsafe_arraycopy",
5409 dst_type,
5410 src_addr, dst_addr, size XTOP);
5411
5412 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5413
5414 return true;
5415 }
5416
5417 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5418 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5419 bool LibraryCallKit::inline_unsafe_setMemory() {
5420 if (callee()->is_static()) return false; // caller must have the capability!
5421 null_check_receiver(); // null-check receiver
5422 if (stopped()) return true;
5423
5424 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5425
5426 Node* dst_base = argument(1); // type: oop
5427 Node* dst_off = ConvL2X(argument(2)); // type: long
5428 Node* size = ConvL2X(argument(4)); // type: long
5429 Node* byte = argument(6); // type: byte
5430
5431 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5432 "fieldOffset must be byte-scaled");
5433
5434 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5435
5436 Node* thread = _gvn.transform(new ThreadLocalNode());
5437 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5438 BasicType doing_unsafe_access_bt = T_BYTE;
5439 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5440
5441 // update volatile field
5442 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5443
5444 int flags = RC_LEAF | RC_NO_FP;
5445
5446 const TypePtr* dst_type = TypePtr::BOTTOM;
5447
5448 // Adjust memory effects of the runtime call based on input values.
5449 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5450 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5451
5452 flags |= RC_NARROW_MEM; // narrow in memory
5453 }
5454
5455 // Call it. Note that the length argument is not scaled.
5456 make_runtime_call(flags,
5457 OptoRuntime::make_setmemory_Type(),
5458 StubRoutines::unsafe_setmemory(),
5459 "unsafe_setmemory",
5460 dst_type,
5461 dst_addr, size XTOP, byte);
5462
5463 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5464
5465 return true;
5466 }
5467
5468 #undef XTOP
5469
5470 //----------------------inline_unsafe_isFlatArray------------------------
5471 // public native boolean Unsafe.isFlatArray(Class<?> arrayClass);
5472 // This intrinsic exploits assumptions made by the native implementation
5473 // (arrayClass is neither null nor primitive) to avoid unnecessary null checks.
5474 bool LibraryCallKit::inline_unsafe_isFlatArray() {
5475 Node* cls = argument(1);
5476 Node* p = basic_plus_adr(cls, java_lang_Class::klass_offset());
5477 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p,
5478 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT));
5479 Node* result = flat_array_test(kls);
5480 set_result(result);
5481 return true;
5482 }
5483
5484 //------------------------clone_coping-----------------------------------
5485 // Helper function for inline_native_clone.
5486 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5487 assert(obj_size != nullptr, "");
5488 Node* raw_obj = alloc_obj->in(1);
5489 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5490
5491 AllocateNode* alloc = nullptr;
5492 if (ReduceBulkZeroing &&
5493 // If we are implementing an array clone without knowing its source type
5494 // (can happen when compiling the array-guarded branch of a reflective
5495 // Object.clone() invocation), initialize the array within the allocation.
5496 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5497 // to a runtime clone call that assumes fully initialized source arrays.
5498 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5499 // We will be completely responsible for initializing this object -
5500 // mark Initialize node as complete.
5501 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5502 // The object was just allocated - there should be no any stores!
5503 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5504 // Mark as complete_with_arraycopy so that on AllocateNode
5505 // expansion, we know this AllocateNode is initialized by an array
5506 // copy and a StoreStore barrier exists after the array copy.
5507 alloc->initialization()->set_complete_with_arraycopy();
5508 }
5509
5510 Node* size = _gvn.transform(obj_size);
5511 access_clone(obj, alloc_obj, size, is_array);
5512
5513 // Do not let reads from the cloned object float above the arraycopy.
5514 if (alloc != nullptr) {
5515 // Do not let stores that initialize this object be reordered with
5516 // a subsequent store that would make this object accessible by
5517 // other threads.
5518 // Record what AllocateNode this StoreStore protects so that
5519 // escape analysis can go from the MemBarStoreStoreNode to the
5520 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5521 // based on the escape status of the AllocateNode.
5522 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5523 } else {
5524 insert_mem_bar(Op_MemBarCPUOrder);
5525 }
5526 }
5527
5528 //------------------------inline_native_clone----------------------------
5529 // protected native Object java.lang.Object.clone();
5530 //
5531 // Here are the simple edge cases:
5532 // null receiver => normal trap
5533 // virtual and clone was overridden => slow path to out-of-line clone
5534 // not cloneable or finalizer => slow path to out-of-line Object.clone
5535 //
5536 // The general case has two steps, allocation and copying.
5537 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5538 //
5539 // Copying also has two cases, oop arrays and everything else.
5540 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5541 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5542 //
5543 // These steps fold up nicely if and when the cloned object's klass
5544 // can be sharply typed as an object array, a type array, or an instance.
5545 //
5546 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5547 PhiNode* result_val;
5548
5549 // Set the reexecute bit for the interpreter to reexecute
5550 // the bytecode that invokes Object.clone if deoptimization happens.
5551 { PreserveReexecuteState preexecs(this);
5552 jvms()->set_should_reexecute(true);
5553
5554 Node* obj = argument(0);
5555 obj = null_check_receiver();
5556 if (stopped()) return true;
5557
5558 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5559 if (obj_type->is_inlinetypeptr()) {
5560 // If the object to clone is an inline type, we can simply return it (i.e. a nop) since inline types have
5561 // no identity.
5562 set_result(obj);
5563 return true;
5564 }
5565
5566 // If we are going to clone an instance, we need its exact type to
5567 // know the number and types of fields to convert the clone to
5568 // loads/stores. Maybe a speculative type can help us.
5569 if (!obj_type->klass_is_exact() &&
5570 obj_type->speculative_type() != nullptr &&
5571 obj_type->speculative_type()->is_instance_klass() &&
5572 !obj_type->speculative_type()->is_inlinetype()) {
5573 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5574 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5575 !spec_ik->has_injected_fields()) {
5576 if (!obj_type->isa_instptr() ||
5577 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5578 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5579 }
5580 }
5581 }
5582
5583 // Conservatively insert a memory barrier on all memory slices.
5584 // Do not let writes into the original float below the clone.
5585 insert_mem_bar(Op_MemBarCPUOrder);
5586
5587 // paths into result_reg:
5588 enum {
5589 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5590 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5591 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5592 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5593 PATH_LIMIT
5594 };
5595 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5596 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5597 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5598 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5599 record_for_igvn(result_reg);
5600
5601 Node* obj_klass = load_object_klass(obj);
5602 // We only go to the fast case code if we pass a number of guards.
5603 // The paths which do not pass are accumulated in the slow_region.
5604 RegionNode* slow_region = new RegionNode(1);
5605 record_for_igvn(slow_region);
5606
5607 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr);
5608 if (array_ctl != nullptr) {
5609 // It's an array.
5610 PreserveJVMState pjvms(this);
5611 set_control(array_ctl);
5612
5613 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5614 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr();
5615 if (UseFlatArray && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) &&
5616 obj_type->can_be_inline_array() &&
5617 (ary_ptr == nullptr || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) {
5618 // Flat inline type array may have object field that would require a
5619 // write barrier. Conservatively, go to slow path.
5620 generate_fair_guard(flat_array_test(obj_klass), slow_region);
5621 }
5622
5623 if (!stopped()) {
5624 Node* obj_length = load_array_length(obj);
5625 Node* array_size = nullptr; // Size of the array without object alignment padding.
5626 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5627
5628 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5629 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5630 // If it is an oop array, it requires very special treatment,
5631 // because gc barriers are required when accessing the array.
5632 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5633 if (is_obja != nullptr) {
5634 PreserveJVMState pjvms2(this);
5635 set_control(is_obja);
5636 // Generate a direct call to the right arraycopy function(s).
5637 // Clones are always tightly coupled.
5638 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5639 ac->set_clone_oop_array();
5640 Node* n = _gvn.transform(ac);
5641 assert(n == ac, "cannot disappear");
5642 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5643
5644 result_reg->init_req(_objArray_path, control());
5645 result_val->init_req(_objArray_path, alloc_obj);
5646 result_i_o ->set_req(_objArray_path, i_o());
5647 result_mem ->set_req(_objArray_path, reset_memory());
5648 }
5649 }
5650 // Otherwise, there are no barriers to worry about.
5651 // (We can dispense with card marks if we know the allocation
5652 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5653 // causes the non-eden paths to take compensating steps to
5654 // simulate a fresh allocation, so that no further
5655 // card marks are required in compiled code to initialize
5656 // the object.)
5657
5658 if (!stopped()) {
5659 copy_to_clone(obj, alloc_obj, array_size, true);
5660
5661 // Present the results of the copy.
5662 result_reg->init_req(_array_path, control());
5663 result_val->init_req(_array_path, alloc_obj);
5664 result_i_o ->set_req(_array_path, i_o());
5665 result_mem ->set_req(_array_path, reset_memory());
5666 }
5667 }
5668 }
5669
5670 if (!stopped()) {
5671 // It's an instance (we did array above). Make the slow-path tests.
5672 // If this is a virtual call, we generate a funny guard. We grab
5673 // the vtable entry corresponding to clone() from the target object.
5674 // If the target method which we are calling happens to be the
5675 // Object clone() method, we pass the guard. We do not need this
5676 // guard for non-virtual calls; the caller is known to be the native
5677 // Object clone().
5678 if (is_virtual) {
5679 generate_virtual_guard(obj_klass, slow_region);
5680 }
5681
5682 // The object must be easily cloneable and must not have a finalizer.
5683 // Both of these conditions may be checked in a single test.
5684 // We could optimize the test further, but we don't care.
5685 generate_misc_flags_guard(obj_klass,
5686 // Test both conditions:
5687 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5688 // Must be cloneable but not finalizer:
5689 KlassFlags::_misc_is_cloneable_fast,
5690 slow_region);
5691 }
5692
5693 if (!stopped()) {
5694 // It's an instance, and it passed the slow-path tests.
5695 PreserveJVMState pjvms(this);
5696 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5697 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5698 // is reexecuted if deoptimization occurs and there could be problems when merging
5699 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5700 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5701
5702 copy_to_clone(obj, alloc_obj, obj_size, false);
5703
5704 // Present the results of the slow call.
5705 result_reg->init_req(_instance_path, control());
5706 result_val->init_req(_instance_path, alloc_obj);
5707 result_i_o ->set_req(_instance_path, i_o());
5708 result_mem ->set_req(_instance_path, reset_memory());
5709 }
5710
5711 // Generate code for the slow case. We make a call to clone().
5712 set_control(_gvn.transform(slow_region));
5713 if (!stopped()) {
5714 PreserveJVMState pjvms(this);
5715 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5716 // We need to deoptimize on exception (see comment above)
5717 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5718 // this->control() comes from set_results_for_java_call
5719 result_reg->init_req(_slow_path, control());
5720 result_val->init_req(_slow_path, slow_result);
5721 result_i_o ->set_req(_slow_path, i_o());
5722 result_mem ->set_req(_slow_path, reset_memory());
5723 }
5724
5725 // Return the combined state.
5726 set_control( _gvn.transform(result_reg));
5727 set_i_o( _gvn.transform(result_i_o));
5728 set_all_memory( _gvn.transform(result_mem));
5729 } // original reexecute is set back here
5730
5731 set_result(_gvn.transform(result_val));
5732 return true;
5733 }
5734
5735 // If we have a tightly coupled allocation, the arraycopy may take care
5736 // of the array initialization. If one of the guards we insert between
5737 // the allocation and the arraycopy causes a deoptimization, an
5738 // uninitialized array will escape the compiled method. To prevent that
5739 // we set the JVM state for uncommon traps between the allocation and
5740 // the arraycopy to the state before the allocation so, in case of
5741 // deoptimization, we'll reexecute the allocation and the
5742 // initialization.
5743 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5744 if (alloc != nullptr) {
5745 ciMethod* trap_method = alloc->jvms()->method();
5746 int trap_bci = alloc->jvms()->bci();
5747
5748 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5749 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5750 // Make sure there's no store between the allocation and the
5751 // arraycopy otherwise visible side effects could be rexecuted
5752 // in case of deoptimization and cause incorrect execution.
5753 bool no_interfering_store = true;
5754 Node* mem = alloc->in(TypeFunc::Memory);
5755 if (mem->is_MergeMem()) {
5756 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5757 Node* n = mms.memory();
5758 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5759 assert(n->is_Store(), "what else?");
5760 no_interfering_store = false;
5761 break;
5762 }
5763 }
5764 } else {
5765 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5766 Node* n = mms.memory();
5767 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5768 assert(n->is_Store(), "what else?");
5769 no_interfering_store = false;
5770 break;
5771 }
5772 }
5773 }
5774
5775 if (no_interfering_store) {
5776 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5777
5778 JVMState* saved_jvms = jvms();
5779 saved_reexecute_sp = _reexecute_sp;
5780
5781 set_jvms(sfpt->jvms());
5782 _reexecute_sp = jvms()->sp();
5783
5784 return saved_jvms;
5785 }
5786 }
5787 }
5788 return nullptr;
5789 }
5790
5791 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5792 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5793 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5794 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5795 uint size = alloc->req();
5796 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5797 old_jvms->set_map(sfpt);
5798 for (uint i = 0; i < size; i++) {
5799 sfpt->init_req(i, alloc->in(i));
5800 }
5801 int adjustment = 1;
5802 const TypeAryKlassPtr* ary_klass_ptr = alloc->in(AllocateNode::KlassNode)->bottom_type()->is_aryklassptr();
5803 if (ary_klass_ptr->is_null_free()) {
5804 // A null-free, tightly coupled array allocation can only come from LibraryCallKit::inline_newNullRestrictedArray
5805 // which requires both the component type and the array length on stack for re-execution. Re-create and push
5806 // the component type.
5807 ciArrayKlass* klass = ary_klass_ptr->exact_klass()->as_array_klass();
5808 ciInstance* instance = klass->component_mirror_instance();
5809 const TypeInstPtr* t_instance = TypeInstPtr::make(instance);
5810 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), makecon(t_instance));
5811 adjustment++;
5812 }
5813 // re-push array length for deoptimization
5814 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment - 1, alloc->in(AllocateNode::ALength));
5815 old_jvms->set_sp(old_jvms->sp() + adjustment);
5816 old_jvms->set_monoff(old_jvms->monoff() + adjustment);
5817 old_jvms->set_scloff(old_jvms->scloff() + adjustment);
5818 old_jvms->set_endoff(old_jvms->endoff() + adjustment);
5819 old_jvms->set_should_reexecute(true);
5820
5821 sfpt->set_i_o(map()->i_o());
5822 sfpt->set_memory(map()->memory());
5823 sfpt->set_control(map()->control());
5824 return sfpt;
5825 }
5826
5827 // In case of a deoptimization, we restart execution at the
5828 // allocation, allocating a new array. We would leave an uninitialized
5829 // array in the heap that GCs wouldn't expect. Move the allocation
5830 // after the traps so we don't allocate the array if we
5831 // deoptimize. This is possible because tightly_coupled_allocation()
5832 // guarantees there's no observer of the allocated array at this point
5833 // and the control flow is simple enough.
5834 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5835 int saved_reexecute_sp, uint new_idx) {
5836 if (saved_jvms_before_guards != nullptr && !stopped()) {
5837 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5838
5839 assert(alloc != nullptr, "only with a tightly coupled allocation");
5840 // restore JVM state to the state at the arraycopy
5841 saved_jvms_before_guards->map()->set_control(map()->control());
5842 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5843 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5844 // If we've improved the types of some nodes (null check) while
5845 // emitting the guards, propagate them to the current state
5846 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5847 set_jvms(saved_jvms_before_guards);
5848 _reexecute_sp = saved_reexecute_sp;
5849
5850 // Remove the allocation from above the guards
5851 CallProjections* callprojs = alloc->extract_projections(true);
5852 InitializeNode* init = alloc->initialization();
5853 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5854 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5855 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5856
5857 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5858 // the allocation (i.e. is only valid if the allocation succeeds):
5859 // 1) replace CastIINode with AllocateArrayNode's length here
5860 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5861 //
5862 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5863 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5864 Node* init_control = init->proj_out(TypeFunc::Control);
5865 Node* alloc_length = alloc->Ideal_length();
5866 #ifdef ASSERT
5867 Node* prev_cast = nullptr;
5868 #endif
5869 for (uint i = 0; i < init_control->outcnt(); i++) {
5870 Node* init_out = init_control->raw_out(i);
5871 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5872 #ifdef ASSERT
5873 if (prev_cast == nullptr) {
5874 prev_cast = init_out;
5875 } else {
5876 if (prev_cast->cmp(*init_out) == false) {
5877 prev_cast->dump();
5878 init_out->dump();
5879 assert(false, "not equal CastIINode");
5880 }
5881 }
5882 #endif
5883 C->gvn_replace_by(init_out, alloc_length);
5884 }
5885 }
5886 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5887
5888 // move the allocation here (after the guards)
5889 _gvn.hash_delete(alloc);
5890 alloc->set_req(TypeFunc::Control, control());
5891 alloc->set_req(TypeFunc::I_O, i_o());
5892 Node *mem = reset_memory();
5893 set_all_memory(mem);
5894 alloc->set_req(TypeFunc::Memory, mem);
5895 set_control(init->proj_out_or_null(TypeFunc::Control));
5896 set_i_o(callprojs->fallthrough_ioproj);
5897
5898 // Update memory as done in GraphKit::set_output_for_allocation()
5899 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5900 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5901 if (ary_type->isa_aryptr() && length_type != nullptr) {
5902 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5903 }
5904 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5905 int elemidx = C->get_alias_index(telemref);
5906 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5907 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5908
5909 Node* allocx = _gvn.transform(alloc);
5910 assert(allocx == alloc, "where has the allocation gone?");
5911 assert(dest->is_CheckCastPP(), "not an allocation result?");
5912
5913 _gvn.hash_delete(dest);
5914 dest->set_req(0, control());
5915 Node* destx = _gvn.transform(dest);
5916 assert(destx == dest, "where has the allocation result gone?");
5917
5918 array_ideal_length(alloc, ary_type, true);
5919 }
5920 }
5921
5922 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5923 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5924 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5925 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5926 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5927 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5928 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5929 JVMState* saved_jvms_before_guards) {
5930 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5931 // There is at least one unrelated uncommon trap which needs to be replaced.
5932 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5933
5934 JVMState* saved_jvms = jvms();
5935 const int saved_reexecute_sp = _reexecute_sp;
5936 set_jvms(sfpt->jvms());
5937 _reexecute_sp = jvms()->sp();
5938
5939 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5940
5941 // Restore state
5942 set_jvms(saved_jvms);
5943 _reexecute_sp = saved_reexecute_sp;
5944 }
5945 }
5946
5947 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5948 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5949 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5950 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5951 while (if_proj->is_IfProj()) {
5952 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5953 if (uncommon_trap != nullptr) {
5954 create_new_uncommon_trap(uncommon_trap);
5955 }
5956 assert(if_proj->in(0)->is_If(), "must be If");
5957 if_proj = if_proj->in(0)->in(0);
5958 }
5959 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5960 "must have reached control projection of init node");
5961 }
5962
5963 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5964 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5965 assert(trap_request != 0, "no valid UCT trap request");
5966 PreserveJVMState pjvms(this);
5967 set_control(uncommon_trap_call->in(0));
5968 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5969 Deoptimization::trap_request_action(trap_request));
5970 assert(stopped(), "Should be stopped");
5971 _gvn.hash_delete(uncommon_trap_call);
5972 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5973 }
5974
5975 // Common checks for array sorting intrinsics arguments.
5976 // Returns `true` if checks passed.
5977 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
5978 // check address of the class
5979 if (elementType == nullptr || elementType->is_top()) {
5980 return false; // dead path
5981 }
5982 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5983 if (elem_klass == nullptr) {
5984 return false; // dead path
5985 }
5986 // java_mirror_type() returns non-null for compile-time Class constants only
5987 ciType* elem_type = elem_klass->java_mirror_type();
5988 if (elem_type == nullptr) {
5989 return false;
5990 }
5991 bt = elem_type->basic_type();
5992 // Disable the intrinsic if the CPU does not support SIMD sort
5993 if (!Matcher::supports_simd_sort(bt)) {
5994 return false;
5995 }
5996 // check address of the array
5997 if (obj == nullptr || obj->is_top()) {
5998 return false; // dead path
5999 }
6000 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
6001 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
6002 return false; // failed input validation
6003 }
6004 return true;
6005 }
6006
6007 //------------------------------inline_array_partition-----------------------
6008 bool LibraryCallKit::inline_array_partition() {
6009 address stubAddr = StubRoutines::select_array_partition_function();
6010 if (stubAddr == nullptr) {
6011 return false; // Intrinsic's stub is not implemented on this platform
6012 }
6013 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
6014
6015 // no receiver because it is a static method
6016 Node* elementType = argument(0);
6017 Node* obj = argument(1);
6018 Node* offset = argument(2); // long
6019 Node* fromIndex = argument(4);
6020 Node* toIndex = argument(5);
6021 Node* indexPivot1 = argument(6);
6022 Node* indexPivot2 = argument(7);
6023 // PartitionOperation: argument(8) is ignored
6024
6025 Node* pivotIndices = nullptr;
6026 BasicType bt = T_ILLEGAL;
6027
6028 if (!check_array_sort_arguments(elementType, obj, bt)) {
6029 return false;
6030 }
6031 null_check(obj);
6032 // If obj is dead, only null-path is taken.
6033 if (stopped()) {
6034 return true;
6035 }
6036 // Set the original stack and the reexecute bit for the interpreter to reexecute
6037 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
6038 { PreserveReexecuteState preexecs(this);
6039 jvms()->set_should_reexecute(true);
6040
6041 Node* obj_adr = make_unsafe_address(obj, offset);
6042
6043 // create the pivotIndices array of type int and size = 2
6044 Node* size = intcon(2);
6045 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
6046 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
6047 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
6048 guarantee(alloc != nullptr, "created above");
6049 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
6050
6051 // pass the basic type enum to the stub
6052 Node* elemType = intcon(bt);
6053
6054 // Call the stub
6055 const char *stubName = "array_partition_stub";
6056 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
6057 stubAddr, stubName, TypePtr::BOTTOM,
6058 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
6059 indexPivot1, indexPivot2);
6060
6061 } // original reexecute is set back here
6062
6063 if (!stopped()) {
6064 set_result(pivotIndices);
6065 }
6066
6067 return true;
6068 }
6069
6070
6071 //------------------------------inline_array_sort-----------------------
6072 bool LibraryCallKit::inline_array_sort() {
6073 address stubAddr = StubRoutines::select_arraysort_function();
6074 if (stubAddr == nullptr) {
6075 return false; // Intrinsic's stub is not implemented on this platform
6076 }
6077 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
6078
6079 // no receiver because it is a static method
6080 Node* elementType = argument(0);
6081 Node* obj = argument(1);
6082 Node* offset = argument(2); // long
6083 Node* fromIndex = argument(4);
6084 Node* toIndex = argument(5);
6085 // SortOperation: argument(6) is ignored
6086
6087 BasicType bt = T_ILLEGAL;
6088
6089 if (!check_array_sort_arguments(elementType, obj, bt)) {
6090 return false;
6091 }
6092 null_check(obj);
6093 // If obj is dead, only null-path is taken.
6094 if (stopped()) {
6095 return true;
6096 }
6097 Node* obj_adr = make_unsafe_address(obj, offset);
6098
6099 // pass the basic type enum to the stub
6100 Node* elemType = intcon(bt);
6101
6102 // Call the stub.
6103 const char *stubName = "arraysort_stub";
6104 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
6105 stubAddr, stubName, TypePtr::BOTTOM,
6106 obj_adr, elemType, fromIndex, toIndex);
6107
6108 return true;
6109 }
6110
6111
6112 //------------------------------inline_arraycopy-----------------------
6113 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
6114 // Object dest, int destPos,
6115 // int length);
6116 bool LibraryCallKit::inline_arraycopy() {
6117 // Get the arguments.
6118 Node* src = argument(0); // type: oop
6119 Node* src_offset = argument(1); // type: int
6120 Node* dest = argument(2); // type: oop
6121 Node* dest_offset = argument(3); // type: int
6122 Node* length = argument(4); // type: int
6123
6124 uint new_idx = C->unique();
6125
6126 // Check for allocation before we add nodes that would confuse
6127 // tightly_coupled_allocation()
6128 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
6129
6130 int saved_reexecute_sp = -1;
6131 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
6132 // See arraycopy_restore_alloc_state() comment
6133 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
6134 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
6135 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
6136 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
6137
6138 // The following tests must be performed
6139 // (1) src and dest are arrays.
6140 // (2) src and dest arrays must have elements of the same BasicType
6141 // (3) src and dest must not be null.
6142 // (4) src_offset must not be negative.
6143 // (5) dest_offset must not be negative.
6144 // (6) length must not be negative.
6145 // (7) src_offset + length must not exceed length of src.
6146 // (8) dest_offset + length must not exceed length of dest.
6147 // (9) each element of an oop array must be assignable
6148
6149 // (3) src and dest must not be null.
6150 // always do this here because we need the JVM state for uncommon traps
6151 Node* null_ctl = top();
6152 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
6153 assert(null_ctl->is_top(), "no null control here");
6154 dest = null_check(dest, T_ARRAY);
6155
6156 if (!can_emit_guards) {
6157 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
6158 // guards but the arraycopy node could still take advantage of a
6159 // tightly allocated allocation. tightly_coupled_allocation() is
6160 // called again to make sure it takes the null check above into
6161 // account: the null check is mandatory and if it caused an
6162 // uncommon trap to be emitted then the allocation can't be
6163 // considered tightly coupled in this context.
6164 alloc = tightly_coupled_allocation(dest);
6165 }
6166
6167 bool validated = false;
6168
6169 const Type* src_type = _gvn.type(src);
6170 const Type* dest_type = _gvn.type(dest);
6171 const TypeAryPtr* top_src = src_type->isa_aryptr();
6172 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6173
6174 // Do we have the type of src?
6175 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6176 // Do we have the type of dest?
6177 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6178 // Is the type for src from speculation?
6179 bool src_spec = false;
6180 // Is the type for dest from speculation?
6181 bool dest_spec = false;
6182
6183 if ((!has_src || !has_dest) && can_emit_guards) {
6184 // We don't have sufficient type information, let's see if
6185 // speculative types can help. We need to have types for both src
6186 // and dest so that it pays off.
6187
6188 // Do we already have or could we have type information for src
6189 bool could_have_src = has_src;
6190 // Do we already have or could we have type information for dest
6191 bool could_have_dest = has_dest;
6192
6193 ciKlass* src_k = nullptr;
6194 if (!has_src) {
6195 src_k = src_type->speculative_type_not_null();
6196 if (src_k != nullptr && src_k->is_array_klass()) {
6197 could_have_src = true;
6198 }
6199 }
6200
6201 ciKlass* dest_k = nullptr;
6202 if (!has_dest) {
6203 dest_k = dest_type->speculative_type_not_null();
6204 if (dest_k != nullptr && dest_k->is_array_klass()) {
6205 could_have_dest = true;
6206 }
6207 }
6208
6209 if (could_have_src && could_have_dest) {
6210 // This is going to pay off so emit the required guards
6211 if (!has_src) {
6212 src = maybe_cast_profiled_obj(src, src_k, true);
6213 src_type = _gvn.type(src);
6214 top_src = src_type->isa_aryptr();
6215 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6216 src_spec = true;
6217 }
6218 if (!has_dest) {
6219 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6220 dest_type = _gvn.type(dest);
6221 top_dest = dest_type->isa_aryptr();
6222 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6223 dest_spec = true;
6224 }
6225 }
6226 }
6227
6228 if (has_src && has_dest && can_emit_guards) {
6229 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
6230 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
6231 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
6232 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
6233
6234 if (src_elem == dest_elem && top_src->is_flat() == top_dest->is_flat() && src_elem == T_OBJECT) {
6235 // If both arrays are object arrays then having the exact types
6236 // for both will remove the need for a subtype check at runtime
6237 // before the call and may make it possible to pick a faster copy
6238 // routine (without a subtype check on every element)
6239 // Do we have the exact type of src?
6240 bool could_have_src = src_spec;
6241 // Do we have the exact type of dest?
6242 bool could_have_dest = dest_spec;
6243 ciKlass* src_k = nullptr;
6244 ciKlass* dest_k = nullptr;
6245 if (!src_spec) {
6246 src_k = src_type->speculative_type_not_null();
6247 if (src_k != nullptr && src_k->is_array_klass()) {
6248 could_have_src = true;
6249 }
6250 }
6251 if (!dest_spec) {
6252 dest_k = dest_type->speculative_type_not_null();
6253 if (dest_k != nullptr && dest_k->is_array_klass()) {
6254 could_have_dest = true;
6255 }
6256 }
6257 if (could_have_src && could_have_dest) {
6258 // If we can have both exact types, emit the missing guards
6259 if (could_have_src && !src_spec) {
6260 src = maybe_cast_profiled_obj(src, src_k, true);
6261 src_type = _gvn.type(src);
6262 top_src = src_type->isa_aryptr();
6263 }
6264 if (could_have_dest && !dest_spec) {
6265 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6266 dest_type = _gvn.type(dest);
6267 top_dest = dest_type->isa_aryptr();
6268 }
6269 }
6270 }
6271 }
6272
6273 ciMethod* trap_method = method();
6274 int trap_bci = bci();
6275 if (saved_jvms_before_guards != nullptr) {
6276 trap_method = alloc->jvms()->method();
6277 trap_bci = alloc->jvms()->bci();
6278 }
6279
6280 bool negative_length_guard_generated = false;
6281
6282 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6283 can_emit_guards && !src->is_top() && !dest->is_top()) {
6284 // validate arguments: enables transformation the ArrayCopyNode
6285 validated = true;
6286
6287 RegionNode* slow_region = new RegionNode(1);
6288 record_for_igvn(slow_region);
6289
6290 // (1) src and dest are arrays.
6291 generate_non_array_guard(load_object_klass(src), slow_region);
6292 generate_non_array_guard(load_object_klass(dest), slow_region);
6293
6294 // (2) src and dest arrays must have elements of the same BasicType
6295 // done at macro expansion or at Ideal transformation time
6296
6297 // (4) src_offset must not be negative.
6298 generate_negative_guard(src_offset, slow_region);
6299
6300 // (5) dest_offset must not be negative.
6301 generate_negative_guard(dest_offset, slow_region);
6302
6303 // (7) src_offset + length must not exceed length of src.
6304 generate_limit_guard(src_offset, length,
6305 load_array_length(src),
6306 slow_region);
6307
6308 // (8) dest_offset + length must not exceed length of dest.
6309 generate_limit_guard(dest_offset, length,
6310 load_array_length(dest),
6311 slow_region);
6312
6313 // (6) length must not be negative.
6314 // This is also checked in generate_arraycopy() during macro expansion, but
6315 // we also have to check it here for the case where the ArrayCopyNode will
6316 // be eliminated by Escape Analysis.
6317 if (EliminateAllocations) {
6318 generate_negative_guard(length, slow_region);
6319 negative_length_guard_generated = true;
6320 }
6321
6322 // (9) each element of an oop array must be assignable
6323 Node* dest_klass = load_object_klass(dest);
6324 if (src != dest) {
6325 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6326 slow_region->add_req(not_subtype_ctrl);
6327 }
6328
6329 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
6330 const Type* toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6331 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6332 src_type = _gvn.type(src);
6333 top_src = src_type->isa_aryptr();
6334
6335 // Handle flat inline type arrays (null-free arrays are handled by the subtype check above)
6336 if (!stopped() && UseFlatArray) {
6337 // If dest is flat, src must be flat as well (guaranteed by src <: dest check). Handle flat src here.
6338 assert(top_dest == nullptr || !top_dest->is_flat() || top_src->is_flat(), "src array must be flat");
6339 if (top_src != nullptr && top_src->is_flat()) {
6340 // Src is flat, check that dest is flat as well
6341 if (top_dest != nullptr && !top_dest->is_flat()) {
6342 generate_fair_guard(flat_array_test(dest_klass, /* flat = */ false), slow_region);
6343 // Since dest is flat and src <: dest, dest must have the same type as src.
6344 top_dest = top_src->cast_to_exactness(false);
6345 assert(top_dest->is_flat(), "dest must be flat");
6346 dest = _gvn.transform(new CheckCastPPNode(control(), dest, top_dest));
6347 }
6348 } else if (top_src == nullptr || !top_src->is_not_flat()) {
6349 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
6350 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
6351 assert(top_dest == nullptr || !top_dest->is_flat(), "dest array must not be flat");
6352 generate_fair_guard(flat_array_test(src), slow_region);
6353 if (top_src != nullptr) {
6354 top_src = top_src->cast_to_not_flat();
6355 src = _gvn.transform(new CheckCastPPNode(control(), src, top_src));
6356 }
6357 }
6358 }
6359
6360 {
6361 PreserveJVMState pjvms(this);
6362 set_control(_gvn.transform(slow_region));
6363 uncommon_trap(Deoptimization::Reason_intrinsic,
6364 Deoptimization::Action_make_not_entrant);
6365 assert(stopped(), "Should be stopped");
6366 }
6367 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6368 }
6369
6370 if (stopped()) {
6371 return true;
6372 }
6373
6374 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6375 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6376 // so the compiler has a chance to eliminate them: during macro expansion,
6377 // we have to set their control (CastPP nodes are eliminated).
6378 load_object_klass(src), load_object_klass(dest),
6379 load_array_length(src), load_array_length(dest));
6380
6381 ac->set_arraycopy(validated);
6382
6383 Node* n = _gvn.transform(ac);
6384 if (n == ac) {
6385 ac->connect_outputs(this);
6386 } else {
6387 assert(validated, "shouldn't transform if all arguments not validated");
6388 set_all_memory(n);
6389 }
6390 clear_upper_avx();
6391
6392
6393 return true;
6394 }
6395
6396
6397 // Helper function which determines if an arraycopy immediately follows
6398 // an allocation, with no intervening tests or other escapes for the object.
6399 AllocateArrayNode*
6400 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6401 if (stopped()) return nullptr; // no fast path
6402 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6403
6404 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6405 if (alloc == nullptr) return nullptr;
6406
6407 Node* rawmem = memory(Compile::AliasIdxRaw);
6408 // Is the allocation's memory state untouched?
6409 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6410 // Bail out if there have been raw-memory effects since the allocation.
6411 // (Example: There might have been a call or safepoint.)
6412 return nullptr;
6413 }
6414 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6415 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6416 return nullptr;
6417 }
6418
6419 // There must be no unexpected observers of this allocation.
6420 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6421 Node* obs = ptr->fast_out(i);
6422 if (obs != this->map()) {
6423 return nullptr;
6424 }
6425 }
6426
6427 // This arraycopy must unconditionally follow the allocation of the ptr.
6428 Node* alloc_ctl = ptr->in(0);
6429 Node* ctl = control();
6430 while (ctl != alloc_ctl) {
6431 // There may be guards which feed into the slow_region.
6432 // Any other control flow means that we might not get a chance
6433 // to finish initializing the allocated object.
6434 // Various low-level checks bottom out in uncommon traps. These
6435 // are considered safe since we've already checked above that
6436 // there is no unexpected observer of this allocation.
6437 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6438 assert(ctl->in(0)->is_If(), "must be If");
6439 ctl = ctl->in(0)->in(0);
6440 } else {
6441 return nullptr;
6442 }
6443 }
6444
6445 // If we get this far, we have an allocation which immediately
6446 // precedes the arraycopy, and we can take over zeroing the new object.
6447 // The arraycopy will finish the initialization, and provide
6448 // a new control state to which we will anchor the destination pointer.
6449
6450 return alloc;
6451 }
6452
6453 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6454 if (node->is_IfProj()) {
6455 Node* other_proj = node->as_IfProj()->other_if_proj();
6456 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6457 Node* obs = other_proj->fast_out(j);
6458 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6459 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6460 return obs->as_CallStaticJava();
6461 }
6462 }
6463 }
6464 return nullptr;
6465 }
6466
6467 //-------------inline_encodeISOArray-----------------------------------
6468 // encode char[] to byte[] in ISO_8859_1 or ASCII
6469 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6470 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6471 // no receiver since it is static method
6472 Node *src = argument(0);
6473 Node *src_offset = argument(1);
6474 Node *dst = argument(2);
6475 Node *dst_offset = argument(3);
6476 Node *length = argument(4);
6477
6478 src = must_be_not_null(src, true);
6479 dst = must_be_not_null(dst, true);
6480
6481 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6482 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6483 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6484 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6485 // failed array check
6486 return false;
6487 }
6488
6489 // Figure out the size and type of the elements we will be copying.
6490 BasicType src_elem = src_type->elem()->array_element_basic_type();
6491 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6492 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6493 return false;
6494 }
6495
6496 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6497 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6498 // 'src_start' points to src array + scaled offset
6499 // 'dst_start' points to dst array + scaled offset
6500
6501 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6502 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6503 enc = _gvn.transform(enc);
6504 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6505 set_memory(res_mem, mtype);
6506 set_result(enc);
6507 clear_upper_avx();
6508
6509 return true;
6510 }
6511
6512 //-------------inline_multiplyToLen-----------------------------------
6513 bool LibraryCallKit::inline_multiplyToLen() {
6514 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6515
6516 address stubAddr = StubRoutines::multiplyToLen();
6517 if (stubAddr == nullptr) {
6518 return false; // Intrinsic's stub is not implemented on this platform
6519 }
6520 const char* stubName = "multiplyToLen";
6521
6522 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6523
6524 // no receiver because it is a static method
6525 Node* x = argument(0);
6526 Node* xlen = argument(1);
6527 Node* y = argument(2);
6528 Node* ylen = argument(3);
6529 Node* z = argument(4);
6530
6531 x = must_be_not_null(x, true);
6532 y = must_be_not_null(y, true);
6533
6534 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6535 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6536 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6537 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6538 // failed array check
6539 return false;
6540 }
6541
6542 BasicType x_elem = x_type->elem()->array_element_basic_type();
6543 BasicType y_elem = y_type->elem()->array_element_basic_type();
6544 if (x_elem != T_INT || y_elem != T_INT) {
6545 return false;
6546 }
6547
6548 Node* x_start = array_element_address(x, intcon(0), x_elem);
6549 Node* y_start = array_element_address(y, intcon(0), y_elem);
6550 // 'x_start' points to x array + scaled xlen
6551 // 'y_start' points to y array + scaled ylen
6552
6553 Node* z_start = array_element_address(z, intcon(0), T_INT);
6554
6555 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6556 OptoRuntime::multiplyToLen_Type(),
6557 stubAddr, stubName, TypePtr::BOTTOM,
6558 x_start, xlen, y_start, ylen, z_start);
6559
6560 C->set_has_split_ifs(true); // Has chance for split-if optimization
6561 set_result(z);
6562 return true;
6563 }
6564
6565 //-------------inline_squareToLen------------------------------------
6566 bool LibraryCallKit::inline_squareToLen() {
6567 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6568
6569 address stubAddr = StubRoutines::squareToLen();
6570 if (stubAddr == nullptr) {
6571 return false; // Intrinsic's stub is not implemented on this platform
6572 }
6573 const char* stubName = "squareToLen";
6574
6575 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6576
6577 Node* x = argument(0);
6578 Node* len = argument(1);
6579 Node* z = argument(2);
6580 Node* zlen = argument(3);
6581
6582 x = must_be_not_null(x, true);
6583 z = must_be_not_null(z, true);
6584
6585 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6586 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6587 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6588 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6589 // failed array check
6590 return false;
6591 }
6592
6593 BasicType x_elem = x_type->elem()->array_element_basic_type();
6594 BasicType z_elem = z_type->elem()->array_element_basic_type();
6595 if (x_elem != T_INT || z_elem != T_INT) {
6596 return false;
6597 }
6598
6599
6600 Node* x_start = array_element_address(x, intcon(0), x_elem);
6601 Node* z_start = array_element_address(z, intcon(0), z_elem);
6602
6603 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6604 OptoRuntime::squareToLen_Type(),
6605 stubAddr, stubName, TypePtr::BOTTOM,
6606 x_start, len, z_start, zlen);
6607
6608 set_result(z);
6609 return true;
6610 }
6611
6612 //-------------inline_mulAdd------------------------------------------
6613 bool LibraryCallKit::inline_mulAdd() {
6614 assert(UseMulAddIntrinsic, "not implemented on this platform");
6615
6616 address stubAddr = StubRoutines::mulAdd();
6617 if (stubAddr == nullptr) {
6618 return false; // Intrinsic's stub is not implemented on this platform
6619 }
6620 const char* stubName = "mulAdd";
6621
6622 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6623
6624 Node* out = argument(0);
6625 Node* in = argument(1);
6626 Node* offset = argument(2);
6627 Node* len = argument(3);
6628 Node* k = argument(4);
6629
6630 in = must_be_not_null(in, true);
6631 out = must_be_not_null(out, true);
6632
6633 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6634 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6635 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6636 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6637 // failed array check
6638 return false;
6639 }
6640
6641 BasicType out_elem = out_type->elem()->array_element_basic_type();
6642 BasicType in_elem = in_type->elem()->array_element_basic_type();
6643 if (out_elem != T_INT || in_elem != T_INT) {
6644 return false;
6645 }
6646
6647 Node* outlen = load_array_length(out);
6648 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6649 Node* out_start = array_element_address(out, intcon(0), out_elem);
6650 Node* in_start = array_element_address(in, intcon(0), in_elem);
6651
6652 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6653 OptoRuntime::mulAdd_Type(),
6654 stubAddr, stubName, TypePtr::BOTTOM,
6655 out_start,in_start, new_offset, len, k);
6656 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6657 set_result(result);
6658 return true;
6659 }
6660
6661 //-------------inline_montgomeryMultiply-----------------------------------
6662 bool LibraryCallKit::inline_montgomeryMultiply() {
6663 address stubAddr = StubRoutines::montgomeryMultiply();
6664 if (stubAddr == nullptr) {
6665 return false; // Intrinsic's stub is not implemented on this platform
6666 }
6667
6668 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6669 const char* stubName = "montgomery_multiply";
6670
6671 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6672
6673 Node* a = argument(0);
6674 Node* b = argument(1);
6675 Node* n = argument(2);
6676 Node* len = argument(3);
6677 Node* inv = argument(4);
6678 Node* m = argument(6);
6679
6680 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6681 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6682 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6683 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6684 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6685 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6686 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6687 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6688 // failed array check
6689 return false;
6690 }
6691
6692 BasicType a_elem = a_type->elem()->array_element_basic_type();
6693 BasicType b_elem = b_type->elem()->array_element_basic_type();
6694 BasicType n_elem = n_type->elem()->array_element_basic_type();
6695 BasicType m_elem = m_type->elem()->array_element_basic_type();
6696 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6697 return false;
6698 }
6699
6700 // Make the call
6701 {
6702 Node* a_start = array_element_address(a, intcon(0), a_elem);
6703 Node* b_start = array_element_address(b, intcon(0), b_elem);
6704 Node* n_start = array_element_address(n, intcon(0), n_elem);
6705 Node* m_start = array_element_address(m, intcon(0), m_elem);
6706
6707 Node* call = make_runtime_call(RC_LEAF,
6708 OptoRuntime::montgomeryMultiply_Type(),
6709 stubAddr, stubName, TypePtr::BOTTOM,
6710 a_start, b_start, n_start, len, inv, top(),
6711 m_start);
6712 set_result(m);
6713 }
6714
6715 return true;
6716 }
6717
6718 bool LibraryCallKit::inline_montgomerySquare() {
6719 address stubAddr = StubRoutines::montgomerySquare();
6720 if (stubAddr == nullptr) {
6721 return false; // Intrinsic's stub is not implemented on this platform
6722 }
6723
6724 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6725 const char* stubName = "montgomery_square";
6726
6727 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6728
6729 Node* a = argument(0);
6730 Node* n = argument(1);
6731 Node* len = argument(2);
6732 Node* inv = argument(3);
6733 Node* m = argument(5);
6734
6735 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6736 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6737 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6738 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6739 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6740 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6741 // failed array check
6742 return false;
6743 }
6744
6745 BasicType a_elem = a_type->elem()->array_element_basic_type();
6746 BasicType n_elem = n_type->elem()->array_element_basic_type();
6747 BasicType m_elem = m_type->elem()->array_element_basic_type();
6748 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6749 return false;
6750 }
6751
6752 // Make the call
6753 {
6754 Node* a_start = array_element_address(a, intcon(0), a_elem);
6755 Node* n_start = array_element_address(n, intcon(0), n_elem);
6756 Node* m_start = array_element_address(m, intcon(0), m_elem);
6757
6758 Node* call = make_runtime_call(RC_LEAF,
6759 OptoRuntime::montgomerySquare_Type(),
6760 stubAddr, stubName, TypePtr::BOTTOM,
6761 a_start, n_start, len, inv, top(),
6762 m_start);
6763 set_result(m);
6764 }
6765
6766 return true;
6767 }
6768
6769 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6770 address stubAddr = nullptr;
6771 const char* stubName = nullptr;
6772
6773 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6774 if (stubAddr == nullptr) {
6775 return false; // Intrinsic's stub is not implemented on this platform
6776 }
6777
6778 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6779
6780 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6781
6782 Node* newArr = argument(0);
6783 Node* oldArr = argument(1);
6784 Node* newIdx = argument(2);
6785 Node* shiftCount = argument(3);
6786 Node* numIter = argument(4);
6787
6788 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6789 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6790 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6791 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6792 return false;
6793 }
6794
6795 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6796 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6797 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6798 return false;
6799 }
6800
6801 // Make the call
6802 {
6803 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6804 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6805
6806 Node* call = make_runtime_call(RC_LEAF,
6807 OptoRuntime::bigIntegerShift_Type(),
6808 stubAddr,
6809 stubName,
6810 TypePtr::BOTTOM,
6811 newArr_start,
6812 oldArr_start,
6813 newIdx,
6814 shiftCount,
6815 numIter);
6816 }
6817
6818 return true;
6819 }
6820
6821 //-------------inline_vectorizedMismatch------------------------------
6822 bool LibraryCallKit::inline_vectorizedMismatch() {
6823 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6824
6825 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6826 Node* obja = argument(0); // Object
6827 Node* aoffset = argument(1); // long
6828 Node* objb = argument(3); // Object
6829 Node* boffset = argument(4); // long
6830 Node* length = argument(6); // int
6831 Node* scale = argument(7); // int
6832
6833 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6834 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6835 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6836 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6837 scale == top()) {
6838 return false; // failed input validation
6839 }
6840
6841 Node* obja_adr = make_unsafe_address(obja, aoffset);
6842 Node* objb_adr = make_unsafe_address(objb, boffset);
6843
6844 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6845 //
6846 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6847 // if (length <= inline_limit) {
6848 // inline_path:
6849 // vmask = VectorMaskGen length
6850 // vload1 = LoadVectorMasked obja, vmask
6851 // vload2 = LoadVectorMasked objb, vmask
6852 // result1 = VectorCmpMasked vload1, vload2, vmask
6853 // } else {
6854 // call_stub_path:
6855 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6856 // }
6857 // exit_block:
6858 // return Phi(result1, result2);
6859 //
6860 enum { inline_path = 1, // input is small enough to process it all at once
6861 stub_path = 2, // input is too large; call into the VM
6862 PATH_LIMIT = 3
6863 };
6864
6865 Node* exit_block = new RegionNode(PATH_LIMIT);
6866 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6867 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6868
6869 Node* call_stub_path = control();
6870
6871 BasicType elem_bt = T_ILLEGAL;
6872
6873 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6874 if (scale_t->is_con()) {
6875 switch (scale_t->get_con()) {
6876 case 0: elem_bt = T_BYTE; break;
6877 case 1: elem_bt = T_SHORT; break;
6878 case 2: elem_bt = T_INT; break;
6879 case 3: elem_bt = T_LONG; break;
6880
6881 default: elem_bt = T_ILLEGAL; break; // not supported
6882 }
6883 }
6884
6885 int inline_limit = 0;
6886 bool do_partial_inline = false;
6887
6888 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6889 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6890 do_partial_inline = inline_limit >= 16;
6891 }
6892
6893 if (do_partial_inline) {
6894 assert(elem_bt != T_ILLEGAL, "sanity");
6895
6896 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6897 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6898 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6899
6900 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6901 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6902 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6903
6904 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6905
6906 if (!stopped()) {
6907 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6908
6909 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6910 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6911 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6912 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6913
6914 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6915 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6916 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6917 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6918
6919 exit_block->init_req(inline_path, control());
6920 memory_phi->init_req(inline_path, map()->memory());
6921 result_phi->init_req(inline_path, result);
6922
6923 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6924 clear_upper_avx();
6925 }
6926 }
6927 }
6928
6929 if (call_stub_path != nullptr) {
6930 set_control(call_stub_path);
6931
6932 Node* call = make_runtime_call(RC_LEAF,
6933 OptoRuntime::vectorizedMismatch_Type(),
6934 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6935 obja_adr, objb_adr, length, scale);
6936
6937 exit_block->init_req(stub_path, control());
6938 memory_phi->init_req(stub_path, map()->memory());
6939 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6940 }
6941
6942 exit_block = _gvn.transform(exit_block);
6943 memory_phi = _gvn.transform(memory_phi);
6944 result_phi = _gvn.transform(result_phi);
6945
6946 set_control(exit_block);
6947 set_all_memory(memory_phi);
6948 set_result(result_phi);
6949
6950 return true;
6951 }
6952
6953 //------------------------------inline_vectorizedHashcode----------------------------
6954 bool LibraryCallKit::inline_vectorizedHashCode() {
6955 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6956
6957 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6958 Node* array = argument(0);
6959 Node* offset = argument(1);
6960 Node* length = argument(2);
6961 Node* initialValue = argument(3);
6962 Node* basic_type = argument(4);
6963
6964 if (basic_type == top()) {
6965 return false; // failed input validation
6966 }
6967
6968 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6969 if (!basic_type_t->is_con()) {
6970 return false; // Only intrinsify if mode argument is constant
6971 }
6972
6973 array = must_be_not_null(array, true);
6974
6975 BasicType bt = (BasicType)basic_type_t->get_con();
6976
6977 // Resolve address of first element
6978 Node* array_start = array_element_address(array, offset, bt);
6979
6980 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6981 array_start, length, initialValue, basic_type)));
6982 clear_upper_avx();
6983
6984 return true;
6985 }
6986
6987 /**
6988 * Calculate CRC32 for byte.
6989 * int java.util.zip.CRC32.update(int crc, int b)
6990 */
6991 bool LibraryCallKit::inline_updateCRC32() {
6992 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6993 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6994 // no receiver since it is static method
6995 Node* crc = argument(0); // type: int
6996 Node* b = argument(1); // type: int
6997
6998 /*
6999 * int c = ~ crc;
7000 * b = timesXtoThe32[(b ^ c) & 0xFF];
7001 * b = b ^ (c >>> 8);
7002 * crc = ~b;
7003 */
7004
7005 Node* M1 = intcon(-1);
7006 crc = _gvn.transform(new XorINode(crc, M1));
7007 Node* result = _gvn.transform(new XorINode(crc, b));
7008 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
7009
7010 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
7011 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
7012 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
7013 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
7014
7015 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
7016 result = _gvn.transform(new XorINode(crc, result));
7017 result = _gvn.transform(new XorINode(result, M1));
7018 set_result(result);
7019 return true;
7020 }
7021
7022 /**
7023 * Calculate CRC32 for byte[] array.
7024 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
7025 */
7026 bool LibraryCallKit::inline_updateBytesCRC32() {
7027 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
7028 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7029 // no receiver since it is static method
7030 Node* crc = argument(0); // type: int
7031 Node* src = argument(1); // type: oop
7032 Node* offset = argument(2); // type: int
7033 Node* length = argument(3); // type: int
7034
7035 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7036 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7037 // failed array check
7038 return false;
7039 }
7040
7041 // Figure out the size and type of the elements we will be copying.
7042 BasicType src_elem = src_type->elem()->array_element_basic_type();
7043 if (src_elem != T_BYTE) {
7044 return false;
7045 }
7046
7047 // 'src_start' points to src array + scaled offset
7048 src = must_be_not_null(src, true);
7049 Node* src_start = array_element_address(src, offset, src_elem);
7050
7051 // We assume that range check is done by caller.
7052 // TODO: generate range check (offset+length < src.length) in debug VM.
7053
7054 // Call the stub.
7055 address stubAddr = StubRoutines::updateBytesCRC32();
7056 const char *stubName = "updateBytesCRC32";
7057
7058 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7059 stubAddr, stubName, TypePtr::BOTTOM,
7060 crc, src_start, length);
7061 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7062 set_result(result);
7063 return true;
7064 }
7065
7066 /**
7067 * Calculate CRC32 for ByteBuffer.
7068 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
7069 */
7070 bool LibraryCallKit::inline_updateByteBufferCRC32() {
7071 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
7072 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7073 // no receiver since it is static method
7074 Node* crc = argument(0); // type: int
7075 Node* src = argument(1); // type: long
7076 Node* offset = argument(3); // type: int
7077 Node* length = argument(4); // type: int
7078
7079 src = ConvL2X(src); // adjust Java long to machine word
7080 Node* base = _gvn.transform(new CastX2PNode(src));
7081 offset = ConvI2X(offset);
7082
7083 // 'src_start' points to src array + scaled offset
7084 Node* src_start = basic_plus_adr(top(), base, offset);
7085
7086 // Call the stub.
7087 address stubAddr = StubRoutines::updateBytesCRC32();
7088 const char *stubName = "updateBytesCRC32";
7089
7090 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7091 stubAddr, stubName, TypePtr::BOTTOM,
7092 crc, src_start, length);
7093 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7094 set_result(result);
7095 return true;
7096 }
7097
7098 //------------------------------get_table_from_crc32c_class-----------------------
7099 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
7100 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
7101 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
7102
7103 return table;
7104 }
7105
7106 //------------------------------inline_updateBytesCRC32C-----------------------
7107 //
7108 // Calculate CRC32C for byte[] array.
7109 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
7110 //
7111 bool LibraryCallKit::inline_updateBytesCRC32C() {
7112 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7113 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7114 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7115 // no receiver since it is a static method
7116 Node* crc = argument(0); // type: int
7117 Node* src = argument(1); // type: oop
7118 Node* offset = argument(2); // type: int
7119 Node* end = argument(3); // type: int
7120
7121 Node* length = _gvn.transform(new SubINode(end, offset));
7122
7123 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7124 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7125 // failed array check
7126 return false;
7127 }
7128
7129 // Figure out the size and type of the elements we will be copying.
7130 BasicType src_elem = src_type->elem()->array_element_basic_type();
7131 if (src_elem != T_BYTE) {
7132 return false;
7133 }
7134
7135 // 'src_start' points to src array + scaled offset
7136 src = must_be_not_null(src, true);
7137 Node* src_start = array_element_address(src, offset, src_elem);
7138
7139 // static final int[] byteTable in class CRC32C
7140 Node* table = get_table_from_crc32c_class(callee()->holder());
7141 table = must_be_not_null(table, true);
7142 Node* table_start = array_element_address(table, intcon(0), T_INT);
7143
7144 // We assume that range check is done by caller.
7145 // TODO: generate range check (offset+length < src.length) in debug VM.
7146
7147 // Call the stub.
7148 address stubAddr = StubRoutines::updateBytesCRC32C();
7149 const char *stubName = "updateBytesCRC32C";
7150
7151 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7152 stubAddr, stubName, TypePtr::BOTTOM,
7153 crc, src_start, length, table_start);
7154 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7155 set_result(result);
7156 return true;
7157 }
7158
7159 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
7160 //
7161 // Calculate CRC32C for DirectByteBuffer.
7162 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
7163 //
7164 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
7165 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7166 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
7167 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7168 // no receiver since it is a static method
7169 Node* crc = argument(0); // type: int
7170 Node* src = argument(1); // type: long
7171 Node* offset = argument(3); // type: int
7172 Node* end = argument(4); // type: int
7173
7174 Node* length = _gvn.transform(new SubINode(end, offset));
7175
7176 src = ConvL2X(src); // adjust Java long to machine word
7177 Node* base = _gvn.transform(new CastX2PNode(src));
7178 offset = ConvI2X(offset);
7179
7180 // 'src_start' points to src array + scaled offset
7181 Node* src_start = basic_plus_adr(top(), base, offset);
7182
7183 // static final int[] byteTable in class CRC32C
7184 Node* table = get_table_from_crc32c_class(callee()->holder());
7185 table = must_be_not_null(table, true);
7186 Node* table_start = array_element_address(table, intcon(0), T_INT);
7187
7188 // Call the stub.
7189 address stubAddr = StubRoutines::updateBytesCRC32C();
7190 const char *stubName = "updateBytesCRC32C";
7191
7192 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7193 stubAddr, stubName, TypePtr::BOTTOM,
7194 crc, src_start, length, table_start);
7195 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7196 set_result(result);
7197 return true;
7198 }
7199
7200 //------------------------------inline_updateBytesAdler32----------------------
7201 //
7202 // Calculate Adler32 checksum for byte[] array.
7203 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
7204 //
7205 bool LibraryCallKit::inline_updateBytesAdler32() {
7206 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7207 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7208 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7209 // no receiver since it is static method
7210 Node* crc = argument(0); // type: int
7211 Node* src = argument(1); // type: oop
7212 Node* offset = argument(2); // type: int
7213 Node* length = argument(3); // type: int
7214
7215 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7216 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7217 // failed array check
7218 return false;
7219 }
7220
7221 // Figure out the size and type of the elements we will be copying.
7222 BasicType src_elem = src_type->elem()->array_element_basic_type();
7223 if (src_elem != T_BYTE) {
7224 return false;
7225 }
7226
7227 // 'src_start' points to src array + scaled offset
7228 Node* src_start = array_element_address(src, offset, src_elem);
7229
7230 // We assume that range check is done by caller.
7231 // TODO: generate range check (offset+length < src.length) in debug VM.
7232
7233 // Call the stub.
7234 address stubAddr = StubRoutines::updateBytesAdler32();
7235 const char *stubName = "updateBytesAdler32";
7236
7237 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7238 stubAddr, stubName, TypePtr::BOTTOM,
7239 crc, src_start, length);
7240 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7241 set_result(result);
7242 return true;
7243 }
7244
7245 //------------------------------inline_updateByteBufferAdler32---------------
7246 //
7247 // Calculate Adler32 checksum for DirectByteBuffer.
7248 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
7249 //
7250 bool LibraryCallKit::inline_updateByteBufferAdler32() {
7251 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7252 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7253 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7254 // no receiver since it is static method
7255 Node* crc = argument(0); // type: int
7256 Node* src = argument(1); // type: long
7257 Node* offset = argument(3); // type: int
7258 Node* length = argument(4); // type: int
7259
7260 src = ConvL2X(src); // adjust Java long to machine word
7261 Node* base = _gvn.transform(new CastX2PNode(src));
7262 offset = ConvI2X(offset);
7263
7264 // 'src_start' points to src array + scaled offset
7265 Node* src_start = basic_plus_adr(top(), base, offset);
7266
7267 // Call the stub.
7268 address stubAddr = StubRoutines::updateBytesAdler32();
7269 const char *stubName = "updateBytesAdler32";
7270
7271 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7272 stubAddr, stubName, TypePtr::BOTTOM,
7273 crc, src_start, length);
7274
7275 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7276 set_result(result);
7277 return true;
7278 }
7279
7280 //----------------------------inline_reference_get----------------------------
7281 // public T java.lang.ref.Reference.get();
7282 bool LibraryCallKit::inline_reference_get() {
7283 const int referent_offset = java_lang_ref_Reference::referent_offset();
7284
7285 // Get the argument:
7286 Node* reference_obj = null_check_receiver();
7287 if (stopped()) return true;
7288
7289 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7290 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7291 decorators, /*is_static*/ false, nullptr);
7292 if (result == nullptr) return false;
7293
7294 // Add memory barrier to prevent commoning reads from this field
7295 // across safepoint since GC can change its value.
7296 insert_mem_bar(Op_MemBarCPUOrder);
7297
7298 set_result(result);
7299 return true;
7300 }
7301
7302 //----------------------------inline_reference_refersTo0----------------------------
7303 // bool java.lang.ref.Reference.refersTo0();
7304 // bool java.lang.ref.PhantomReference.refersTo0();
7305 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7306 // Get arguments:
7307 Node* reference_obj = null_check_receiver();
7308 Node* other_obj = argument(1);
7309 if (stopped()) return true;
7310
7311 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7312 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7313 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7314 decorators, /*is_static*/ false, nullptr);
7315 if (referent == nullptr) return false;
7316
7317 // Add memory barrier to prevent commoning reads from this field
7318 // across safepoint since GC can change its value.
7319 insert_mem_bar(Op_MemBarCPUOrder);
7320
7321 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7322 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7323 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7324
7325 RegionNode* region = new RegionNode(3);
7326 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7327
7328 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7329 region->init_req(1, if_true);
7330 phi->init_req(1, intcon(1));
7331
7332 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7333 region->init_req(2, if_false);
7334 phi->init_req(2, intcon(0));
7335
7336 set_control(_gvn.transform(region));
7337 record_for_igvn(region);
7338 set_result(_gvn.transform(phi));
7339 return true;
7340 }
7341
7342
7343 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7344 DecoratorSet decorators, bool is_static,
7345 ciInstanceKlass* fromKls) {
7346 if (fromKls == nullptr) {
7347 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7348 assert(tinst != nullptr, "obj is null");
7349 assert(tinst->is_loaded(), "obj is not loaded");
7350 fromKls = tinst->instance_klass();
7351 } else {
7352 assert(is_static, "only for static field access");
7353 }
7354 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7355 ciSymbol::make(fieldTypeString),
7356 is_static);
7357
7358 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7359 if (field == nullptr) return (Node *) nullptr;
7360
7361 if (is_static) {
7362 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7363 fromObj = makecon(tip);
7364 }
7365
7366 // Next code copied from Parse::do_get_xxx():
7367
7368 // Compute address and memory type.
7369 int offset = field->offset_in_bytes();
7370 bool is_vol = field->is_volatile();
7371 ciType* field_klass = field->type();
7372 assert(field_klass->is_loaded(), "should be loaded");
7373 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7374 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7375 BasicType bt = field->layout_type();
7376
7377 // Build the resultant type of the load
7378 const Type *type;
7379 if (bt == T_OBJECT) {
7380 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7381 } else {
7382 type = Type::get_const_basic_type(bt);
7383 }
7384
7385 if (is_vol) {
7386 decorators |= MO_SEQ_CST;
7387 }
7388
7389 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7390 }
7391
7392 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7393 bool is_exact /* true */, bool is_static /* false */,
7394 ciInstanceKlass * fromKls /* nullptr */) {
7395 if (fromKls == nullptr) {
7396 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7397 assert(tinst != nullptr, "obj is null");
7398 assert(tinst->is_loaded(), "obj is not loaded");
7399 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7400 fromKls = tinst->instance_klass();
7401 }
7402 else {
7403 assert(is_static, "only for static field access");
7404 }
7405 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7406 ciSymbol::make(fieldTypeString),
7407 is_static);
7408
7409 assert(field != nullptr, "undefined field");
7410 assert(!field->is_volatile(), "not defined for volatile fields");
7411
7412 if (is_static) {
7413 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7414 fromObj = makecon(tip);
7415 }
7416
7417 // Next code copied from Parse::do_get_xxx():
7418
7419 // Compute address and memory type.
7420 int offset = field->offset_in_bytes();
7421 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7422
7423 return adr;
7424 }
7425
7426 //------------------------------inline_aescrypt_Block-----------------------
7427 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7428 address stubAddr = nullptr;
7429 const char *stubName;
7430 assert(UseAES, "need AES instruction support");
7431
7432 switch(id) {
7433 case vmIntrinsics::_aescrypt_encryptBlock:
7434 stubAddr = StubRoutines::aescrypt_encryptBlock();
7435 stubName = "aescrypt_encryptBlock";
7436 break;
7437 case vmIntrinsics::_aescrypt_decryptBlock:
7438 stubAddr = StubRoutines::aescrypt_decryptBlock();
7439 stubName = "aescrypt_decryptBlock";
7440 break;
7441 default:
7442 break;
7443 }
7444 if (stubAddr == nullptr) return false;
7445
7446 Node* aescrypt_object = argument(0);
7447 Node* src = argument(1);
7448 Node* src_offset = argument(2);
7449 Node* dest = argument(3);
7450 Node* dest_offset = argument(4);
7451
7452 src = must_be_not_null(src, true);
7453 dest = must_be_not_null(dest, true);
7454
7455 // (1) src and dest are arrays.
7456 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7457 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7458 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7459 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7460
7461 // for the quick and dirty code we will skip all the checks.
7462 // we are just trying to get the call to be generated.
7463 Node* src_start = src;
7464 Node* dest_start = dest;
7465 if (src_offset != nullptr || dest_offset != nullptr) {
7466 assert(src_offset != nullptr && dest_offset != nullptr, "");
7467 src_start = array_element_address(src, src_offset, T_BYTE);
7468 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7469 }
7470
7471 // now need to get the start of its expanded key array
7472 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7473 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7474 if (k_start == nullptr) return false;
7475
7476 // Call the stub.
7477 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7478 stubAddr, stubName, TypePtr::BOTTOM,
7479 src_start, dest_start, k_start);
7480
7481 return true;
7482 }
7483
7484 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7485 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7486 address stubAddr = nullptr;
7487 const char *stubName = nullptr;
7488
7489 assert(UseAES, "need AES instruction support");
7490
7491 switch(id) {
7492 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7493 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7494 stubName = "cipherBlockChaining_encryptAESCrypt";
7495 break;
7496 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7497 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7498 stubName = "cipherBlockChaining_decryptAESCrypt";
7499 break;
7500 default:
7501 break;
7502 }
7503 if (stubAddr == nullptr) return false;
7504
7505 Node* cipherBlockChaining_object = argument(0);
7506 Node* src = argument(1);
7507 Node* src_offset = argument(2);
7508 Node* len = argument(3);
7509 Node* dest = argument(4);
7510 Node* dest_offset = argument(5);
7511
7512 src = must_be_not_null(src, false);
7513 dest = must_be_not_null(dest, false);
7514
7515 // (1) src and dest are arrays.
7516 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7517 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7518 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7519 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7520
7521 // checks are the responsibility of the caller
7522 Node* src_start = src;
7523 Node* dest_start = dest;
7524 if (src_offset != nullptr || dest_offset != nullptr) {
7525 assert(src_offset != nullptr && dest_offset != nullptr, "");
7526 src_start = array_element_address(src, src_offset, T_BYTE);
7527 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7528 }
7529
7530 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7531 // (because of the predicated logic executed earlier).
7532 // so we cast it here safely.
7533 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7534
7535 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7536 if (embeddedCipherObj == nullptr) return false;
7537
7538 // cast it to what we know it will be at runtime
7539 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7540 assert(tinst != nullptr, "CBC obj is null");
7541 assert(tinst->is_loaded(), "CBC obj is not loaded");
7542 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7543 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7544
7545 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7546 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7547 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7548 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7549 aescrypt_object = _gvn.transform(aescrypt_object);
7550
7551 // we need to get the start of the aescrypt_object's expanded key array
7552 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7553 if (k_start == nullptr) return false;
7554
7555 // similarly, get the start address of the r vector
7556 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7557 if (objRvec == nullptr) return false;
7558 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7559
7560 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7561 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7562 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7563 stubAddr, stubName, TypePtr::BOTTOM,
7564 src_start, dest_start, k_start, r_start, len);
7565
7566 // return cipher length (int)
7567 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7568 set_result(retvalue);
7569 return true;
7570 }
7571
7572 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7573 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7574 address stubAddr = nullptr;
7575 const char *stubName = nullptr;
7576
7577 assert(UseAES, "need AES instruction support");
7578
7579 switch (id) {
7580 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7581 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7582 stubName = "electronicCodeBook_encryptAESCrypt";
7583 break;
7584 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7585 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7586 stubName = "electronicCodeBook_decryptAESCrypt";
7587 break;
7588 default:
7589 break;
7590 }
7591
7592 if (stubAddr == nullptr) return false;
7593
7594 Node* electronicCodeBook_object = argument(0);
7595 Node* src = argument(1);
7596 Node* src_offset = argument(2);
7597 Node* len = argument(3);
7598 Node* dest = argument(4);
7599 Node* dest_offset = argument(5);
7600
7601 // (1) src and dest are arrays.
7602 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7603 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7604 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7605 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7606
7607 // checks are the responsibility of the caller
7608 Node* src_start = src;
7609 Node* dest_start = dest;
7610 if (src_offset != nullptr || dest_offset != nullptr) {
7611 assert(src_offset != nullptr && dest_offset != nullptr, "");
7612 src_start = array_element_address(src, src_offset, T_BYTE);
7613 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7614 }
7615
7616 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7617 // (because of the predicated logic executed earlier).
7618 // so we cast it here safely.
7619 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7620
7621 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7622 if (embeddedCipherObj == nullptr) return false;
7623
7624 // cast it to what we know it will be at runtime
7625 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7626 assert(tinst != nullptr, "ECB obj is null");
7627 assert(tinst->is_loaded(), "ECB obj is not loaded");
7628 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7629 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7630
7631 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7632 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7633 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7634 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7635 aescrypt_object = _gvn.transform(aescrypt_object);
7636
7637 // we need to get the start of the aescrypt_object's expanded key array
7638 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7639 if (k_start == nullptr) return false;
7640
7641 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7642 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7643 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7644 stubAddr, stubName, TypePtr::BOTTOM,
7645 src_start, dest_start, k_start, len);
7646
7647 // return cipher length (int)
7648 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7649 set_result(retvalue);
7650 return true;
7651 }
7652
7653 //------------------------------inline_counterMode_AESCrypt-----------------------
7654 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7655 assert(UseAES, "need AES instruction support");
7656 if (!UseAESCTRIntrinsics) return false;
7657
7658 address stubAddr = nullptr;
7659 const char *stubName = nullptr;
7660 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7661 stubAddr = StubRoutines::counterMode_AESCrypt();
7662 stubName = "counterMode_AESCrypt";
7663 }
7664 if (stubAddr == nullptr) return false;
7665
7666 Node* counterMode_object = argument(0);
7667 Node* src = argument(1);
7668 Node* src_offset = argument(2);
7669 Node* len = argument(3);
7670 Node* dest = argument(4);
7671 Node* dest_offset = argument(5);
7672
7673 // (1) src and dest are arrays.
7674 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7675 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7676 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7677 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7678
7679 // checks are the responsibility of the caller
7680 Node* src_start = src;
7681 Node* dest_start = dest;
7682 if (src_offset != nullptr || dest_offset != nullptr) {
7683 assert(src_offset != nullptr && dest_offset != nullptr, "");
7684 src_start = array_element_address(src, src_offset, T_BYTE);
7685 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7686 }
7687
7688 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7689 // (because of the predicated logic executed earlier).
7690 // so we cast it here safely.
7691 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7692 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7693 if (embeddedCipherObj == nullptr) return false;
7694 // cast it to what we know it will be at runtime
7695 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7696 assert(tinst != nullptr, "CTR obj is null");
7697 assert(tinst->is_loaded(), "CTR obj is not loaded");
7698 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7699 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7700 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7701 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7702 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7703 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7704 aescrypt_object = _gvn.transform(aescrypt_object);
7705 // we need to get the start of the aescrypt_object's expanded key array
7706 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7707 if (k_start == nullptr) return false;
7708 // similarly, get the start address of the r vector
7709 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7710 if (obj_counter == nullptr) return false;
7711 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7712
7713 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7714 if (saved_encCounter == nullptr) return false;
7715 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7716 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7717
7718 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7719 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7720 OptoRuntime::counterMode_aescrypt_Type(),
7721 stubAddr, stubName, TypePtr::BOTTOM,
7722 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7723
7724 // return cipher length (int)
7725 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7726 set_result(retvalue);
7727 return true;
7728 }
7729
7730 //------------------------------get_key_start_from_aescrypt_object-----------------------
7731 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7732 #if defined(PPC64) || defined(S390)
7733 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7734 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7735 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7736 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7737 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7738 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7739 if (objSessionK == nullptr) {
7740 return (Node *) nullptr;
7741 }
7742 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7743 #else
7744 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7745 #endif // PPC64
7746 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7747 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7748
7749 // now have the array, need to get the start address of the K array
7750 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7751 return k_start;
7752 }
7753
7754 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7755 // Return node representing slow path of predicate check.
7756 // the pseudo code we want to emulate with this predicate is:
7757 // for encryption:
7758 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7759 // for decryption:
7760 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7761 // note cipher==plain is more conservative than the original java code but that's OK
7762 //
7763 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7764 // The receiver was checked for null already.
7765 Node* objCBC = argument(0);
7766
7767 Node* src = argument(1);
7768 Node* dest = argument(4);
7769
7770 // Load embeddedCipher field of CipherBlockChaining object.
7771 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7772
7773 // get AESCrypt klass for instanceOf check
7774 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7775 // will have same classloader as CipherBlockChaining object
7776 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7777 assert(tinst != nullptr, "CBCobj is null");
7778 assert(tinst->is_loaded(), "CBCobj is not loaded");
7779
7780 // we want to do an instanceof comparison against the AESCrypt class
7781 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7782 if (!klass_AESCrypt->is_loaded()) {
7783 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7784 Node* ctrl = control();
7785 set_control(top()); // no regular fast path
7786 return ctrl;
7787 }
7788
7789 src = must_be_not_null(src, true);
7790 dest = must_be_not_null(dest, true);
7791
7792 // Resolve oops to stable for CmpP below.
7793 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7794
7795 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7796 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7797 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7798
7799 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7800
7801 // for encryption, we are done
7802 if (!decrypting)
7803 return instof_false; // even if it is null
7804
7805 // for decryption, we need to add a further check to avoid
7806 // taking the intrinsic path when cipher and plain are the same
7807 // see the original java code for why.
7808 RegionNode* region = new RegionNode(3);
7809 region->init_req(1, instof_false);
7810
7811 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7812 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7813 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7814 region->init_req(2, src_dest_conjoint);
7815
7816 record_for_igvn(region);
7817 return _gvn.transform(region);
7818 }
7819
7820 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7821 // Return node representing slow path of predicate check.
7822 // the pseudo code we want to emulate with this predicate is:
7823 // for encryption:
7824 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7825 // for decryption:
7826 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7827 // note cipher==plain is more conservative than the original java code but that's OK
7828 //
7829 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7830 // The receiver was checked for null already.
7831 Node* objECB = argument(0);
7832
7833 // Load embeddedCipher field of ElectronicCodeBook object.
7834 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7835
7836 // get AESCrypt klass for instanceOf check
7837 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7838 // will have same classloader as ElectronicCodeBook object
7839 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7840 assert(tinst != nullptr, "ECBobj is null");
7841 assert(tinst->is_loaded(), "ECBobj is not loaded");
7842
7843 // we want to do an instanceof comparison against the AESCrypt class
7844 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7845 if (!klass_AESCrypt->is_loaded()) {
7846 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7847 Node* ctrl = control();
7848 set_control(top()); // no regular fast path
7849 return ctrl;
7850 }
7851 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7852
7853 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7854 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7855 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7856
7857 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7858
7859 // for encryption, we are done
7860 if (!decrypting)
7861 return instof_false; // even if it is null
7862
7863 // for decryption, we need to add a further check to avoid
7864 // taking the intrinsic path when cipher and plain are the same
7865 // see the original java code for why.
7866 RegionNode* region = new RegionNode(3);
7867 region->init_req(1, instof_false);
7868 Node* src = argument(1);
7869 Node* dest = argument(4);
7870 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7871 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7872 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7873 region->init_req(2, src_dest_conjoint);
7874
7875 record_for_igvn(region);
7876 return _gvn.transform(region);
7877 }
7878
7879 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7880 // Return node representing slow path of predicate check.
7881 // the pseudo code we want to emulate with this predicate is:
7882 // for encryption:
7883 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7884 // for decryption:
7885 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7886 // note cipher==plain is more conservative than the original java code but that's OK
7887 //
7888
7889 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7890 // The receiver was checked for null already.
7891 Node* objCTR = argument(0);
7892
7893 // Load embeddedCipher field of CipherBlockChaining object.
7894 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7895
7896 // get AESCrypt klass for instanceOf check
7897 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7898 // will have same classloader as CipherBlockChaining object
7899 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7900 assert(tinst != nullptr, "CTRobj is null");
7901 assert(tinst->is_loaded(), "CTRobj is not loaded");
7902
7903 // we want to do an instanceof comparison against the AESCrypt class
7904 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7905 if (!klass_AESCrypt->is_loaded()) {
7906 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7907 Node* ctrl = control();
7908 set_control(top()); // no regular fast path
7909 return ctrl;
7910 }
7911
7912 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7913 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7914 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7915 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7916 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7917
7918 return instof_false; // even if it is null
7919 }
7920
7921 //------------------------------inline_ghash_processBlocks
7922 bool LibraryCallKit::inline_ghash_processBlocks() {
7923 address stubAddr;
7924 const char *stubName;
7925 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7926
7927 stubAddr = StubRoutines::ghash_processBlocks();
7928 stubName = "ghash_processBlocks";
7929
7930 Node* data = argument(0);
7931 Node* offset = argument(1);
7932 Node* len = argument(2);
7933 Node* state = argument(3);
7934 Node* subkeyH = argument(4);
7935
7936 state = must_be_not_null(state, true);
7937 subkeyH = must_be_not_null(subkeyH, true);
7938 data = must_be_not_null(data, true);
7939
7940 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7941 assert(state_start, "state is null");
7942 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7943 assert(subkeyH_start, "subkeyH is null");
7944 Node* data_start = array_element_address(data, offset, T_BYTE);
7945 assert(data_start, "data is null");
7946
7947 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7948 OptoRuntime::ghash_processBlocks_Type(),
7949 stubAddr, stubName, TypePtr::BOTTOM,
7950 state_start, subkeyH_start, data_start, len);
7951 return true;
7952 }
7953
7954 //------------------------------inline_chacha20Block
7955 bool LibraryCallKit::inline_chacha20Block() {
7956 address stubAddr;
7957 const char *stubName;
7958 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7959
7960 stubAddr = StubRoutines::chacha20Block();
7961 stubName = "chacha20Block";
7962
7963 Node* state = argument(0);
7964 Node* result = argument(1);
7965
7966 state = must_be_not_null(state, true);
7967 result = must_be_not_null(result, true);
7968
7969 Node* state_start = array_element_address(state, intcon(0), T_INT);
7970 assert(state_start, "state is null");
7971 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7972 assert(result_start, "result is null");
7973
7974 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7975 OptoRuntime::chacha20Block_Type(),
7976 stubAddr, stubName, TypePtr::BOTTOM,
7977 state_start, result_start);
7978 // return key stream length (int)
7979 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7980 set_result(retvalue);
7981 return true;
7982 }
7983
7984 bool LibraryCallKit::inline_base64_encodeBlock() {
7985 address stubAddr;
7986 const char *stubName;
7987 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7988 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
7989 stubAddr = StubRoutines::base64_encodeBlock();
7990 stubName = "encodeBlock";
7991
7992 if (!stubAddr) return false;
7993 Node* base64obj = argument(0);
7994 Node* src = argument(1);
7995 Node* offset = argument(2);
7996 Node* len = argument(3);
7997 Node* dest = argument(4);
7998 Node* dp = argument(5);
7999 Node* isURL = argument(6);
8000
8001 src = must_be_not_null(src, true);
8002 dest = must_be_not_null(dest, true);
8003
8004 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8005 assert(src_start, "source array is null");
8006 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8007 assert(dest_start, "destination array is null");
8008
8009 Node* base64 = make_runtime_call(RC_LEAF,
8010 OptoRuntime::base64_encodeBlock_Type(),
8011 stubAddr, stubName, TypePtr::BOTTOM,
8012 src_start, offset, len, dest_start, dp, isURL);
8013 return true;
8014 }
8015
8016 bool LibraryCallKit::inline_base64_decodeBlock() {
8017 address stubAddr;
8018 const char *stubName;
8019 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8020 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8021 stubAddr = StubRoutines::base64_decodeBlock();
8022 stubName = "decodeBlock";
8023
8024 if (!stubAddr) return false;
8025 Node* base64obj = argument(0);
8026 Node* src = argument(1);
8027 Node* src_offset = argument(2);
8028 Node* len = argument(3);
8029 Node* dest = argument(4);
8030 Node* dest_offset = argument(5);
8031 Node* isURL = argument(6);
8032 Node* isMIME = argument(7);
8033
8034 src = must_be_not_null(src, true);
8035 dest = must_be_not_null(dest, true);
8036
8037 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8038 assert(src_start, "source array is null");
8039 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8040 assert(dest_start, "destination array is null");
8041
8042 Node* call = make_runtime_call(RC_LEAF,
8043 OptoRuntime::base64_decodeBlock_Type(),
8044 stubAddr, stubName, TypePtr::BOTTOM,
8045 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8046 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8047 set_result(result);
8048 return true;
8049 }
8050
8051 bool LibraryCallKit::inline_poly1305_processBlocks() {
8052 address stubAddr;
8053 const char *stubName;
8054 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8055 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8056 stubAddr = StubRoutines::poly1305_processBlocks();
8057 stubName = "poly1305_processBlocks";
8058
8059 if (!stubAddr) return false;
8060 null_check_receiver(); // null-check receiver
8061 if (stopped()) return true;
8062
8063 Node* input = argument(1);
8064 Node* input_offset = argument(2);
8065 Node* len = argument(3);
8066 Node* alimbs = argument(4);
8067 Node* rlimbs = argument(5);
8068
8069 input = must_be_not_null(input, true);
8070 alimbs = must_be_not_null(alimbs, true);
8071 rlimbs = must_be_not_null(rlimbs, true);
8072
8073 Node* input_start = array_element_address(input, input_offset, T_BYTE);
8074 assert(input_start, "input array is null");
8075 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
8076 assert(acc_start, "acc array is null");
8077 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
8078 assert(r_start, "r array is null");
8079
8080 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8081 OptoRuntime::poly1305_processBlocks_Type(),
8082 stubAddr, stubName, TypePtr::BOTTOM,
8083 input_start, len, acc_start, r_start);
8084 return true;
8085 }
8086
8087 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
8088 address stubAddr;
8089 const char *stubName;
8090 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8091 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
8092 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
8093 stubName = "intpoly_montgomeryMult_P256";
8094
8095 if (!stubAddr) return false;
8096 null_check_receiver(); // null-check receiver
8097 if (stopped()) return true;
8098
8099 Node* a = argument(1);
8100 Node* b = argument(2);
8101 Node* r = argument(3);
8102
8103 a = must_be_not_null(a, true);
8104 b = must_be_not_null(b, true);
8105 r = must_be_not_null(r, true);
8106
8107 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8108 assert(a_start, "a array is NULL");
8109 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8110 assert(b_start, "b array is NULL");
8111 Node* r_start = array_element_address(r, intcon(0), T_LONG);
8112 assert(r_start, "r array is NULL");
8113
8114 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8115 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
8116 stubAddr, stubName, TypePtr::BOTTOM,
8117 a_start, b_start, r_start);
8118 return true;
8119 }
8120
8121 bool LibraryCallKit::inline_intpoly_assign() {
8122 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8123 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
8124 const char *stubName = "intpoly_assign";
8125 address stubAddr = StubRoutines::intpoly_assign();
8126 if (!stubAddr) return false;
8127
8128 Node* set = argument(0);
8129 Node* a = argument(1);
8130 Node* b = argument(2);
8131 Node* arr_length = load_array_length(a);
8132
8133 a = must_be_not_null(a, true);
8134 b = must_be_not_null(b, true);
8135
8136 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8137 assert(a_start, "a array is NULL");
8138 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8139 assert(b_start, "b array is NULL");
8140
8141 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8142 OptoRuntime::intpoly_assign_Type(),
8143 stubAddr, stubName, TypePtr::BOTTOM,
8144 set, a_start, b_start, arr_length);
8145 return true;
8146 }
8147
8148 //------------------------------inline_digestBase_implCompress-----------------------
8149 //
8150 // Calculate MD5 for single-block byte[] array.
8151 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
8152 //
8153 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
8154 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
8155 //
8156 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
8157 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
8158 //
8159 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
8160 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
8161 //
8162 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
8163 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
8164 //
8165 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
8166 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
8167
8168 Node* digestBase_obj = argument(0);
8169 Node* src = argument(1); // type oop
8170 Node* ofs = argument(2); // type int
8171
8172 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8173 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8174 // failed array check
8175 return false;
8176 }
8177 // Figure out the size and type of the elements we will be copying.
8178 BasicType src_elem = src_type->elem()->array_element_basic_type();
8179 if (src_elem != T_BYTE) {
8180 return false;
8181 }
8182 // 'src_start' points to src array + offset
8183 src = must_be_not_null(src, true);
8184 Node* src_start = array_element_address(src, ofs, src_elem);
8185 Node* state = nullptr;
8186 Node* block_size = nullptr;
8187 address stubAddr;
8188 const char *stubName;
8189
8190 switch(id) {
8191 case vmIntrinsics::_md5_implCompress:
8192 assert(UseMD5Intrinsics, "need MD5 instruction support");
8193 state = get_state_from_digest_object(digestBase_obj, T_INT);
8194 stubAddr = StubRoutines::md5_implCompress();
8195 stubName = "md5_implCompress";
8196 break;
8197 case vmIntrinsics::_sha_implCompress:
8198 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
8199 state = get_state_from_digest_object(digestBase_obj, T_INT);
8200 stubAddr = StubRoutines::sha1_implCompress();
8201 stubName = "sha1_implCompress";
8202 break;
8203 case vmIntrinsics::_sha2_implCompress:
8204 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
8205 state = get_state_from_digest_object(digestBase_obj, T_INT);
8206 stubAddr = StubRoutines::sha256_implCompress();
8207 stubName = "sha256_implCompress";
8208 break;
8209 case vmIntrinsics::_sha5_implCompress:
8210 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
8211 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8212 stubAddr = StubRoutines::sha512_implCompress();
8213 stubName = "sha512_implCompress";
8214 break;
8215 case vmIntrinsics::_sha3_implCompress:
8216 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
8217 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8218 stubAddr = StubRoutines::sha3_implCompress();
8219 stubName = "sha3_implCompress";
8220 block_size = get_block_size_from_digest_object(digestBase_obj);
8221 if (block_size == nullptr) return false;
8222 break;
8223 default:
8224 fatal_unexpected_iid(id);
8225 return false;
8226 }
8227 if (state == nullptr) return false;
8228
8229 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
8230 if (stubAddr == nullptr) return false;
8231
8232 // Call the stub.
8233 Node* call;
8234 if (block_size == nullptr) {
8235 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
8236 stubAddr, stubName, TypePtr::BOTTOM,
8237 src_start, state);
8238 } else {
8239 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
8240 stubAddr, stubName, TypePtr::BOTTOM,
8241 src_start, state, block_size);
8242 }
8243
8244 return true;
8245 }
8246
8247 //------------------------------inline_digestBase_implCompressMB-----------------------
8248 //
8249 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
8250 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
8251 //
8252 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
8253 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8254 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8255 assert((uint)predicate < 5, "sanity");
8256 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
8257
8258 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
8259 Node* src = argument(1); // byte[] array
8260 Node* ofs = argument(2); // type int
8261 Node* limit = argument(3); // type int
8262
8263 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8264 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8265 // failed array check
8266 return false;
8267 }
8268 // Figure out the size and type of the elements we will be copying.
8269 BasicType src_elem = src_type->elem()->array_element_basic_type();
8270 if (src_elem != T_BYTE) {
8271 return false;
8272 }
8273 // 'src_start' points to src array + offset
8274 src = must_be_not_null(src, false);
8275 Node* src_start = array_element_address(src, ofs, src_elem);
8276
8277 const char* klass_digestBase_name = nullptr;
8278 const char* stub_name = nullptr;
8279 address stub_addr = nullptr;
8280 BasicType elem_type = T_INT;
8281
8282 switch (predicate) {
8283 case 0:
8284 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
8285 klass_digestBase_name = "sun/security/provider/MD5";
8286 stub_name = "md5_implCompressMB";
8287 stub_addr = StubRoutines::md5_implCompressMB();
8288 }
8289 break;
8290 case 1:
8291 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
8292 klass_digestBase_name = "sun/security/provider/SHA";
8293 stub_name = "sha1_implCompressMB";
8294 stub_addr = StubRoutines::sha1_implCompressMB();
8295 }
8296 break;
8297 case 2:
8298 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
8299 klass_digestBase_name = "sun/security/provider/SHA2";
8300 stub_name = "sha256_implCompressMB";
8301 stub_addr = StubRoutines::sha256_implCompressMB();
8302 }
8303 break;
8304 case 3:
8305 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
8306 klass_digestBase_name = "sun/security/provider/SHA5";
8307 stub_name = "sha512_implCompressMB";
8308 stub_addr = StubRoutines::sha512_implCompressMB();
8309 elem_type = T_LONG;
8310 }
8311 break;
8312 case 4:
8313 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
8314 klass_digestBase_name = "sun/security/provider/SHA3";
8315 stub_name = "sha3_implCompressMB";
8316 stub_addr = StubRoutines::sha3_implCompressMB();
8317 elem_type = T_LONG;
8318 }
8319 break;
8320 default:
8321 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
8322 }
8323 if (klass_digestBase_name != nullptr) {
8324 assert(stub_addr != nullptr, "Stub is generated");
8325 if (stub_addr == nullptr) return false;
8326
8327 // get DigestBase klass to lookup for SHA klass
8328 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
8329 assert(tinst != nullptr, "digestBase_obj is not instance???");
8330 assert(tinst->is_loaded(), "DigestBase is not loaded");
8331
8332 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
8333 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
8334 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
8335 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
8336 }
8337 return false;
8338 }
8339
8340 //------------------------------inline_digestBase_implCompressMB-----------------------
8341 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
8342 BasicType elem_type, address stubAddr, const char *stubName,
8343 Node* src_start, Node* ofs, Node* limit) {
8344 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
8345 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8346 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
8347 digest_obj = _gvn.transform(digest_obj);
8348
8349 Node* state = get_state_from_digest_object(digest_obj, elem_type);
8350 if (state == nullptr) return false;
8351
8352 Node* block_size = nullptr;
8353 if (strcmp("sha3_implCompressMB", stubName) == 0) {
8354 block_size = get_block_size_from_digest_object(digest_obj);
8355 if (block_size == nullptr) return false;
8356 }
8357
8358 // Call the stub.
8359 Node* call;
8360 if (block_size == nullptr) {
8361 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8362 OptoRuntime::digestBase_implCompressMB_Type(false),
8363 stubAddr, stubName, TypePtr::BOTTOM,
8364 src_start, state, ofs, limit);
8365 } else {
8366 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8367 OptoRuntime::digestBase_implCompressMB_Type(true),
8368 stubAddr, stubName, TypePtr::BOTTOM,
8369 src_start, state, block_size, ofs, limit);
8370 }
8371
8372 // return ofs (int)
8373 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8374 set_result(result);
8375
8376 return true;
8377 }
8378
8379 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8380 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8381 assert(UseAES, "need AES instruction support");
8382 address stubAddr = nullptr;
8383 const char *stubName = nullptr;
8384 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8385 stubName = "galoisCounterMode_AESCrypt";
8386
8387 if (stubAddr == nullptr) return false;
8388
8389 Node* in = argument(0);
8390 Node* inOfs = argument(1);
8391 Node* len = argument(2);
8392 Node* ct = argument(3);
8393 Node* ctOfs = argument(4);
8394 Node* out = argument(5);
8395 Node* outOfs = argument(6);
8396 Node* gctr_object = argument(7);
8397 Node* ghash_object = argument(8);
8398
8399 // (1) in, ct and out are arrays.
8400 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8401 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8402 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8403 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8404 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8405 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8406
8407 // checks are the responsibility of the caller
8408 Node* in_start = in;
8409 Node* ct_start = ct;
8410 Node* out_start = out;
8411 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8412 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8413 in_start = array_element_address(in, inOfs, T_BYTE);
8414 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8415 out_start = array_element_address(out, outOfs, T_BYTE);
8416 }
8417
8418 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8419 // (because of the predicated logic executed earlier).
8420 // so we cast it here safely.
8421 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8422 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8423 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8424 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8425 Node* state = load_field_from_object(ghash_object, "state", "[J");
8426
8427 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8428 return false;
8429 }
8430 // cast it to what we know it will be at runtime
8431 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8432 assert(tinst != nullptr, "GCTR obj is null");
8433 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8434 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8435 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8436 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8437 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8438 const TypeOopPtr* xtype = aklass->as_instance_type();
8439 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8440 aescrypt_object = _gvn.transform(aescrypt_object);
8441 // we need to get the start of the aescrypt_object's expanded key array
8442 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8443 if (k_start == nullptr) return false;
8444 // similarly, get the start address of the r vector
8445 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8446 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8447 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8448
8449
8450 // Call the stub, passing params
8451 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8452 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8453 stubAddr, stubName, TypePtr::BOTTOM,
8454 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8455
8456 // return cipher length (int)
8457 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8458 set_result(retvalue);
8459
8460 return true;
8461 }
8462
8463 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8464 // Return node representing slow path of predicate check.
8465 // the pseudo code we want to emulate with this predicate is:
8466 // for encryption:
8467 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8468 // for decryption:
8469 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8470 // note cipher==plain is more conservative than the original java code but that's OK
8471 //
8472
8473 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8474 // The receiver was checked for null already.
8475 Node* objGCTR = argument(7);
8476 // Load embeddedCipher field of GCTR object.
8477 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8478 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8479
8480 // get AESCrypt klass for instanceOf check
8481 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8482 // will have same classloader as CipherBlockChaining object
8483 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8484 assert(tinst != nullptr, "GCTR obj is null");
8485 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8486
8487 // we want to do an instanceof comparison against the AESCrypt class
8488 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8489 if (!klass_AESCrypt->is_loaded()) {
8490 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8491 Node* ctrl = control();
8492 set_control(top()); // no regular fast path
8493 return ctrl;
8494 }
8495
8496 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8497 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8498 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8499 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8500 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8501
8502 return instof_false; // even if it is null
8503 }
8504
8505 //------------------------------get_state_from_digest_object-----------------------
8506 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8507 const char* state_type;
8508 switch (elem_type) {
8509 case T_BYTE: state_type = "[B"; break;
8510 case T_INT: state_type = "[I"; break;
8511 case T_LONG: state_type = "[J"; break;
8512 default: ShouldNotReachHere();
8513 }
8514 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8515 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8516 if (digest_state == nullptr) return (Node *) nullptr;
8517
8518 // now have the array, need to get the start address of the state array
8519 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8520 return state;
8521 }
8522
8523 //------------------------------get_block_size_from_sha3_object----------------------------------
8524 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8525 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8526 assert (block_size != nullptr, "sanity");
8527 return block_size;
8528 }
8529
8530 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8531 // Return node representing slow path of predicate check.
8532 // the pseudo code we want to emulate with this predicate is:
8533 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8534 //
8535 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8536 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8537 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8538 assert((uint)predicate < 5, "sanity");
8539
8540 // The receiver was checked for null already.
8541 Node* digestBaseObj = argument(0);
8542
8543 // get DigestBase klass for instanceOf check
8544 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8545 assert(tinst != nullptr, "digestBaseObj is null");
8546 assert(tinst->is_loaded(), "DigestBase is not loaded");
8547
8548 const char* klass_name = nullptr;
8549 switch (predicate) {
8550 case 0:
8551 if (UseMD5Intrinsics) {
8552 // we want to do an instanceof comparison against the MD5 class
8553 klass_name = "sun/security/provider/MD5";
8554 }
8555 break;
8556 case 1:
8557 if (UseSHA1Intrinsics) {
8558 // we want to do an instanceof comparison against the SHA class
8559 klass_name = "sun/security/provider/SHA";
8560 }
8561 break;
8562 case 2:
8563 if (UseSHA256Intrinsics) {
8564 // we want to do an instanceof comparison against the SHA2 class
8565 klass_name = "sun/security/provider/SHA2";
8566 }
8567 break;
8568 case 3:
8569 if (UseSHA512Intrinsics) {
8570 // we want to do an instanceof comparison against the SHA5 class
8571 klass_name = "sun/security/provider/SHA5";
8572 }
8573 break;
8574 case 4:
8575 if (UseSHA3Intrinsics) {
8576 // we want to do an instanceof comparison against the SHA3 class
8577 klass_name = "sun/security/provider/SHA3";
8578 }
8579 break;
8580 default:
8581 fatal("unknown SHA intrinsic predicate: %d", predicate);
8582 }
8583
8584 ciKlass* klass = nullptr;
8585 if (klass_name != nullptr) {
8586 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8587 }
8588 if ((klass == nullptr) || !klass->is_loaded()) {
8589 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8590 Node* ctrl = control();
8591 set_control(top()); // no intrinsic path
8592 return ctrl;
8593 }
8594 ciInstanceKlass* instklass = klass->as_instance_klass();
8595
8596 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8597 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8598 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8599 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8600
8601 return instof_false; // even if it is null
8602 }
8603
8604 //-------------inline_fma-----------------------------------
8605 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8606 Node *a = nullptr;
8607 Node *b = nullptr;
8608 Node *c = nullptr;
8609 Node* result = nullptr;
8610 switch (id) {
8611 case vmIntrinsics::_fmaD:
8612 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8613 // no receiver since it is static method
8614 a = round_double_node(argument(0));
8615 b = round_double_node(argument(2));
8616 c = round_double_node(argument(4));
8617 result = _gvn.transform(new FmaDNode(control(), a, b, c));
8618 break;
8619 case vmIntrinsics::_fmaF:
8620 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8621 a = argument(0);
8622 b = argument(1);
8623 c = argument(2);
8624 result = _gvn.transform(new FmaFNode(control(), a, b, c));
8625 break;
8626 default:
8627 fatal_unexpected_iid(id); break;
8628 }
8629 set_result(result);
8630 return true;
8631 }
8632
8633 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8634 // argument(0) is receiver
8635 Node* codePoint = argument(1);
8636 Node* n = nullptr;
8637
8638 switch (id) {
8639 case vmIntrinsics::_isDigit :
8640 n = new DigitNode(control(), codePoint);
8641 break;
8642 case vmIntrinsics::_isLowerCase :
8643 n = new LowerCaseNode(control(), codePoint);
8644 break;
8645 case vmIntrinsics::_isUpperCase :
8646 n = new UpperCaseNode(control(), codePoint);
8647 break;
8648 case vmIntrinsics::_isWhitespace :
8649 n = new WhitespaceNode(control(), codePoint);
8650 break;
8651 default:
8652 fatal_unexpected_iid(id);
8653 }
8654
8655 set_result(_gvn.transform(n));
8656 return true;
8657 }
8658
8659 //------------------------------inline_fp_min_max------------------------------
8660 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
8661 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
8662
8663 // The intrinsic should be used only when the API branches aren't predictable,
8664 // the last one performing the most important comparison. The following heuristic
8665 // uses the branch statistics to eventually bail out if necessary.
8666
8667 ciMethodData *md = callee()->method_data();
8668
8669 if ( md != nullptr && md->is_mature() && md->invocation_count() > 0 ) {
8670 ciCallProfile cp = caller()->call_profile_at_bci(bci());
8671
8672 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
8673 // Bail out if the call-site didn't contribute enough to the statistics.
8674 return false;
8675 }
8676
8677 uint taken = 0, not_taken = 0;
8678
8679 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
8680 if (p->is_BranchData()) {
8681 taken = ((ciBranchData*)p)->taken();
8682 not_taken = ((ciBranchData*)p)->not_taken();
8683 }
8684 }
8685
8686 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
8687 balance = balance < 0 ? -balance : balance;
8688 if ( balance > 0.2 ) {
8689 // Bail out if the most important branch is predictable enough.
8690 return false;
8691 }
8692 }
8693 */
8694
8695 Node *a = nullptr;
8696 Node *b = nullptr;
8697 Node *n = nullptr;
8698 switch (id) {
8699 case vmIntrinsics::_maxF:
8700 case vmIntrinsics::_minF:
8701 case vmIntrinsics::_maxF_strict:
8702 case vmIntrinsics::_minF_strict:
8703 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
8704 a = argument(0);
8705 b = argument(1);
8706 break;
8707 case vmIntrinsics::_maxD:
8708 case vmIntrinsics::_minD:
8709 case vmIntrinsics::_maxD_strict:
8710 case vmIntrinsics::_minD_strict:
8711 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
8712 a = round_double_node(argument(0));
8713 b = round_double_node(argument(2));
8714 break;
8715 default:
8716 fatal_unexpected_iid(id);
8717 break;
8718 }
8719 switch (id) {
8720 case vmIntrinsics::_maxF:
8721 case vmIntrinsics::_maxF_strict:
8722 n = new MaxFNode(a, b);
8723 break;
8724 case vmIntrinsics::_minF:
8725 case vmIntrinsics::_minF_strict:
8726 n = new MinFNode(a, b);
8727 break;
8728 case vmIntrinsics::_maxD:
8729 case vmIntrinsics::_maxD_strict:
8730 n = new MaxDNode(a, b);
8731 break;
8732 case vmIntrinsics::_minD:
8733 case vmIntrinsics::_minD_strict:
8734 n = new MinDNode(a, b);
8735 break;
8736 default:
8737 fatal_unexpected_iid(id);
8738 break;
8739 }
8740 set_result(_gvn.transform(n));
8741 return true;
8742 }
8743
8744 bool LibraryCallKit::inline_profileBoolean() {
8745 Node* counts = argument(1);
8746 const TypeAryPtr* ary = nullptr;
8747 ciArray* aobj = nullptr;
8748 if (counts->is_Con()
8749 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8750 && (aobj = ary->const_oop()->as_array()) != nullptr
8751 && (aobj->length() == 2)) {
8752 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8753 jint false_cnt = aobj->element_value(0).as_int();
8754 jint true_cnt = aobj->element_value(1).as_int();
8755
8756 if (C->log() != nullptr) {
8757 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8758 false_cnt, true_cnt);
8759 }
8760
8761 if (false_cnt + true_cnt == 0) {
8762 // According to profile, never executed.
8763 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8764 Deoptimization::Action_reinterpret);
8765 return true;
8766 }
8767
8768 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8769 // is a number of each value occurrences.
8770 Node* result = argument(0);
8771 if (false_cnt == 0 || true_cnt == 0) {
8772 // According to profile, one value has been never seen.
8773 int expected_val = (false_cnt == 0) ? 1 : 0;
8774
8775 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8776 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8777
8778 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8779 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8780 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8781
8782 { // Slow path: uncommon trap for never seen value and then reexecute
8783 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8784 // the value has been seen at least once.
8785 PreserveJVMState pjvms(this);
8786 PreserveReexecuteState preexecs(this);
8787 jvms()->set_should_reexecute(true);
8788
8789 set_control(slow_path);
8790 set_i_o(i_o());
8791
8792 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8793 Deoptimization::Action_reinterpret);
8794 }
8795 // The guard for never seen value enables sharpening of the result and
8796 // returning a constant. It allows to eliminate branches on the same value
8797 // later on.
8798 set_control(fast_path);
8799 result = intcon(expected_val);
8800 }
8801 // Stop profiling.
8802 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8803 // By replacing method body with profile data (represented as ProfileBooleanNode
8804 // on IR level) we effectively disable profiling.
8805 // It enables full speed execution once optimized code is generated.
8806 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8807 C->record_for_igvn(profile);
8808 set_result(profile);
8809 return true;
8810 } else {
8811 // Continue profiling.
8812 // Profile data isn't available at the moment. So, execute method's bytecode version.
8813 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8814 // is compiled and counters aren't available since corresponding MethodHandle
8815 // isn't a compile-time constant.
8816 return false;
8817 }
8818 }
8819
8820 bool LibraryCallKit::inline_isCompileConstant() {
8821 Node* n = argument(0);
8822 set_result(n->is_Con() ? intcon(1) : intcon(0));
8823 return true;
8824 }
8825
8826 //------------------------------- inline_getObjectSize --------------------------------------
8827 //
8828 // Calculate the runtime size of the object/array.
8829 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8830 //
8831 bool LibraryCallKit::inline_getObjectSize() {
8832 Node* obj = argument(3);
8833 Node* klass_node = load_object_klass(obj);
8834
8835 jint layout_con = Klass::_lh_neutral_value;
8836 Node* layout_val = get_layout_helper(klass_node, layout_con);
8837 int layout_is_con = (layout_val == nullptr);
8838
8839 if (layout_is_con) {
8840 // Layout helper is constant, can figure out things at compile time.
8841
8842 if (Klass::layout_helper_is_instance(layout_con)) {
8843 // Instance case: layout_con contains the size itself.
8844 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8845 set_result(size);
8846 } else {
8847 // Array case: size is round(header + element_size*arraylength).
8848 // Since arraylength is different for every array instance, we have to
8849 // compute the whole thing at runtime.
8850
8851 Node* arr_length = load_array_length(obj);
8852
8853 int round_mask = MinObjAlignmentInBytes - 1;
8854 int hsize = Klass::layout_helper_header_size(layout_con);
8855 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8856
8857 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8858 round_mask = 0; // strength-reduce it if it goes away completely
8859 }
8860 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8861 Node* header_size = intcon(hsize + round_mask);
8862
8863 Node* lengthx = ConvI2X(arr_length);
8864 Node* headerx = ConvI2X(header_size);
8865
8866 Node* abody = lengthx;
8867 if (eshift != 0) {
8868 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8869 }
8870 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8871 if (round_mask != 0) {
8872 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8873 }
8874 size = ConvX2L(size);
8875 set_result(size);
8876 }
8877 } else {
8878 // Layout helper is not constant, need to test for array-ness at runtime.
8879
8880 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8881 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8882 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8883 record_for_igvn(result_reg);
8884
8885 Node* array_ctl = generate_array_guard(klass_node, nullptr);
8886 if (array_ctl != nullptr) {
8887 // Array case: size is round(header + element_size*arraylength).
8888 // Since arraylength is different for every array instance, we have to
8889 // compute the whole thing at runtime.
8890
8891 PreserveJVMState pjvms(this);
8892 set_control(array_ctl);
8893 Node* arr_length = load_array_length(obj);
8894
8895 int round_mask = MinObjAlignmentInBytes - 1;
8896 Node* mask = intcon(round_mask);
8897
8898 Node* hss = intcon(Klass::_lh_header_size_shift);
8899 Node* hsm = intcon(Klass::_lh_header_size_mask);
8900 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8901 header_size = _gvn.transform(new AndINode(header_size, hsm));
8902 header_size = _gvn.transform(new AddINode(header_size, mask));
8903
8904 // There is no need to mask or shift this value.
8905 // The semantics of LShiftINode include an implicit mask to 0x1F.
8906 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8907 Node* elem_shift = layout_val;
8908
8909 Node* lengthx = ConvI2X(arr_length);
8910 Node* headerx = ConvI2X(header_size);
8911
8912 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
8913 Node* size = _gvn.transform(new AddXNode(headerx, abody));
8914 if (round_mask != 0) {
8915 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
8916 }
8917 size = ConvX2L(size);
8918
8919 result_reg->init_req(_array_path, control());
8920 result_val->init_req(_array_path, size);
8921 }
8922
8923 if (!stopped()) {
8924 // Instance case: the layout helper gives us instance size almost directly,
8925 // but we need to mask out the _lh_instance_slow_path_bit.
8926 Node* size = ConvI2X(layout_val);
8927 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
8928 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
8929 size = _gvn.transform(new AndXNode(size, mask));
8930 size = ConvX2L(size);
8931
8932 result_reg->init_req(_instance_path, control());
8933 result_val->init_req(_instance_path, size);
8934 }
8935
8936 set_result(result_reg, result_val);
8937 }
8938
8939 return true;
8940 }
8941
8942 //------------------------------- inline_blackhole --------------------------------------
8943 //
8944 // Make sure all arguments to this node are alive.
8945 // This matches methods that were requested to be blackholed through compile commands.
8946 //
8947 bool LibraryCallKit::inline_blackhole() {
8948 assert(callee()->is_static(), "Should have been checked before: only static methods here");
8949 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
8950 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
8951
8952 // Blackhole node pinches only the control, not memory. This allows
8953 // the blackhole to be pinned in the loop that computes blackholed
8954 // values, but have no other side effects, like breaking the optimizations
8955 // across the blackhole.
8956
8957 Node* bh = _gvn.transform(new BlackholeNode(control()));
8958 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
8959
8960 // Bind call arguments as blackhole arguments to keep them alive
8961 uint nargs = callee()->arg_size();
8962 for (uint i = 0; i < nargs; i++) {
8963 bh->add_req(argument(i));
8964 }
8965
8966 return true;
8967 }