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/ciUtilities.inline.hpp"
28 #include "classfile/vmIntrinsics.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "gc/shared/barrierSet.hpp"
32 #include "jfr/support/jfrIntrinsics.hpp"
33 #include "memory/resourceArea.hpp"
34 #include "oops/klass.inline.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "opto/addnode.hpp"
37 #include "opto/arraycopynode.hpp"
38 #include "opto/c2compiler.hpp"
39 #include "opto/castnode.hpp"
40 #include "opto/cfgnode.hpp"
41 #include "opto/convertnode.hpp"
42 #include "opto/countbitsnode.hpp"
43 #include "opto/idealKit.hpp"
44 #include "opto/library_call.hpp"
45 #include "opto/mathexactnode.hpp"
46 #include "opto/mulnode.hpp"
47 #include "opto/narrowptrnode.hpp"
48 #include "opto/opaquenode.hpp"
49 #include "opto/parse.hpp"
50 #include "opto/runtime.hpp"
51 #include "opto/rootnode.hpp"
52 #include "opto/subnode.hpp"
53 #include "prims/jvmtiExport.hpp"
54 #include "prims/jvmtiThreadState.hpp"
55 #include "prims/unsafe.hpp"
56 #include "runtime/jniHandles.inline.hpp"
57 #include "runtime/objectMonitor.hpp"
58 #include "runtime/sharedRuntime.hpp"
59 #include "runtime/stubRoutines.hpp"
60 #include "utilities/macros.hpp"
61 #include "utilities/powerOfTwo.hpp"
62
63 //---------------------------make_vm_intrinsic----------------------------
64 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
65 vmIntrinsicID id = m->intrinsic_id();
66 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
67
68 if (!m->is_loaded()) {
69 // Do not attempt to inline unloaded methods.
70 return nullptr;
71 }
72
73 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
74 bool is_available = false;
75
76 {
77 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
78 // the compiler must transition to '_thread_in_vm' state because both
79 // methods access VM-internal data.
80 VM_ENTRY_MARK;
81 methodHandle mh(THREAD, m->get_Method());
82 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
83 if (is_available && is_virtual) {
84 is_available = vmIntrinsics::does_virtual_dispatch(id);
85 }
86 }
87
88 if (is_available) {
89 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
90 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
91 return new LibraryIntrinsic(m, is_virtual,
92 vmIntrinsics::predicates_needed(id),
93 vmIntrinsics::does_virtual_dispatch(id),
94 id);
95 } else {
96 return nullptr;
97 }
98 }
99
100 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
101 LibraryCallKit kit(jvms, this);
102 Compile* C = kit.C;
103 int nodes = C->unique();
104 #ifndef PRODUCT
105 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
106 char buf[1000];
107 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
108 tty->print_cr("Intrinsic %s", str);
109 }
110 #endif
111 ciMethod* callee = kit.callee();
112 const int bci = kit.bci();
113 #ifdef ASSERT
114 Node* ctrl = kit.control();
115 #endif
116 // Try to inline the intrinsic.
117 if (callee->check_intrinsic_candidate() &&
118 kit.try_to_inline(_last_predicate)) {
119 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
120 : "(intrinsic)";
121 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
122 if (C->print_intrinsics() || C->print_inlining()) {
123 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
124 }
125 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
126 if (C->log()) {
127 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
128 vmIntrinsics::name_at(intrinsic_id()),
129 (is_virtual() ? " virtual='1'" : ""),
130 C->unique() - nodes);
131 }
132 // Push the result from the inlined method onto the stack.
133 kit.push_result();
134 C->print_inlining_update(this);
135 return kit.transfer_exceptions_into_jvms();
136 }
137
138 // The intrinsic bailed out
139 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
140 if (jvms->has_method()) {
141 // Not a root compile.
142 const char* msg;
143 if (callee->intrinsic_candidate()) {
144 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
145 } else {
146 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
147 : "failed to inline (intrinsic), method not annotated";
148 }
149 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
150 if (C->print_intrinsics() || C->print_inlining()) {
151 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
152 }
153 } else {
154 // Root compile
155 ResourceMark rm;
156 stringStream msg_stream;
157 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
158 vmIntrinsics::name_at(intrinsic_id()),
159 is_virtual() ? " (virtual)" : "", bci);
160 const char *msg = msg_stream.freeze();
161 log_debug(jit, inlining)("%s", msg);
162 if (C->print_intrinsics() || C->print_inlining()) {
163 tty->print("%s", msg);
164 }
165 }
166 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
167 C->print_inlining_update(this);
168
169 return nullptr;
170 }
171
172 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
173 LibraryCallKit kit(jvms, this);
174 Compile* C = kit.C;
175 int nodes = C->unique();
176 _last_predicate = predicate;
177 #ifndef PRODUCT
178 assert(is_predicated() && predicate < predicates_count(), "sanity");
179 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
180 char buf[1000];
181 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
182 tty->print_cr("Predicate for intrinsic %s", str);
183 }
184 #endif
185 ciMethod* callee = kit.callee();
186 const int bci = kit.bci();
187
188 Node* slow_ctl = kit.try_to_predicate(predicate);
189 if (!kit.failing()) {
190 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
191 : "(intrinsic, predicate)";
192 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
193 if (C->print_intrinsics() || C->print_inlining()) {
194 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
195 }
196 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
197 if (C->log()) {
198 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
199 vmIntrinsics::name_at(intrinsic_id()),
200 (is_virtual() ? " virtual='1'" : ""),
201 C->unique() - nodes);
202 }
203 return slow_ctl; // Could be null if the check folds.
204 }
205
206 // The intrinsic bailed out
207 if (jvms->has_method()) {
208 // Not a root compile.
209 const char* msg = "failed to generate predicate for intrinsic";
210 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
211 if (C->print_intrinsics() || C->print_inlining()) {
212 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
213 }
214 } else {
215 // Root compile
216 ResourceMark rm;
217 stringStream msg_stream;
218 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
219 vmIntrinsics::name_at(intrinsic_id()),
220 is_virtual() ? " (virtual)" : "", bci);
221 const char *msg = msg_stream.freeze();
222 log_debug(jit, inlining)("%s", msg);
223 if (C->print_intrinsics() || C->print_inlining()) {
224 C->print_inlining_stream()->print("%s", msg);
225 }
226 }
227 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
228 return nullptr;
229 }
230
231 bool LibraryCallKit::try_to_inline(int predicate) {
232 // Handle symbolic names for otherwise undistinguished boolean switches:
233 const bool is_store = true;
234 const bool is_compress = true;
235 const bool is_static = true;
236 const bool is_volatile = true;
237
238 if (!jvms()->has_method()) {
239 // Root JVMState has a null method.
240 assert(map()->memory()->Opcode() == Op_Parm, "");
241 // Insert the memory aliasing node
242 set_all_memory(reset_memory());
243 }
244 assert(merged_memory(), "");
245
246 switch (intrinsic_id()) {
247 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
248 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
249 case vmIntrinsics::_getClass: return inline_native_getClass();
250
251 case vmIntrinsics::_ceil:
252 case vmIntrinsics::_floor:
253 case vmIntrinsics::_rint:
254 case vmIntrinsics::_dsin:
255 case vmIntrinsics::_dcos:
256 case vmIntrinsics::_dtan:
257 case vmIntrinsics::_dtanh:
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::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
329 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
330 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
331 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
332 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
333 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
334 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
335 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
336 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
337
338 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
339 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
340 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
341 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
342 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
343 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
344 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
345 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
346 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
347
348 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
349 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
350 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
351 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
352 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
353 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
354 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
355 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
356 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
357
358 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
359 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
360 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
361 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
362 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
363 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
364 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
365 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
366 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
367
368 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
369 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
370 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
371 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
372
373 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
374 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
375 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
376 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
377
378 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
379 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
380 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
381 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
382 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
383 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
384 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
385 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
386 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
387
388 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
389 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
390 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
391 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
392 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
393 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
394 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
395 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
396 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
397
398 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
399 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
400 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
401 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
402 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
403 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
404 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
405 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
406 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
407
408 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
409 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
410 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
411 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
412 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
413 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
414 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
415 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
416 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
417
418 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
419 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
420 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
421 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
422 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
423
424 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
425 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
426 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
427 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
428 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
429 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
430 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
431 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
432 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
433 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
434 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
435 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
436 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
437 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
438 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
439 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
440 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
441 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
442 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
443 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
444
445 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
446 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
447 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
448 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
449 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
450 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
451 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
452 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
453 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
454 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
455 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
456 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
457 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
458 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
459 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
460
461 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
462 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
463 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
464 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
465
466 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
467 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
468 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
469 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
470 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
471
472 case vmIntrinsics::_loadFence:
473 case vmIntrinsics::_storeFence:
474 case vmIntrinsics::_storeStoreFence:
475 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
476
477 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
478
479 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
480 case vmIntrinsics::_currentThread: return inline_native_currentThread();
481 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
482
483 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
484 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
485
486 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
487 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
488
489 #if INCLUDE_JVMTI
490 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
491 "notifyJvmtiStart", true, false);
492 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
493 "notifyJvmtiEnd", false, true);
494 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
495 "notifyJvmtiMount", false, false);
496 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
497 "notifyJvmtiUnmount", false, false);
498 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
499 #endif
500
501 #ifdef JFR_HAVE_INTRINSICS
502 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
503 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
504 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
505 #endif
506 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
507 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
508 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
509 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
510 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
511 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
512 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
513 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
514 case vmIntrinsics::_getLength: return inline_native_getLength();
515 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
516 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
517 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
518 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
519 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
520 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
521 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
522
523 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
524 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
525
526 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
527
528 case vmIntrinsics::_isInstance:
529 case vmIntrinsics::_getModifiers:
530 case vmIntrinsics::_isInterface:
531 case vmIntrinsics::_isArray:
532 case vmIntrinsics::_isPrimitive:
533 case vmIntrinsics::_isHidden:
534 case vmIntrinsics::_getSuperclass:
535 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
536
537 case vmIntrinsics::_floatToRawIntBits:
538 case vmIntrinsics::_floatToIntBits:
539 case vmIntrinsics::_intBitsToFloat:
540 case vmIntrinsics::_doubleToRawLongBits:
541 case vmIntrinsics::_doubleToLongBits:
542 case vmIntrinsics::_longBitsToDouble:
543 case vmIntrinsics::_floatToFloat16:
544 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
545
546 case vmIntrinsics::_floatIsFinite:
547 case vmIntrinsics::_floatIsInfinite:
548 case vmIntrinsics::_doubleIsFinite:
549 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
550
551 case vmIntrinsics::_numberOfLeadingZeros_i:
552 case vmIntrinsics::_numberOfLeadingZeros_l:
553 case vmIntrinsics::_numberOfTrailingZeros_i:
554 case vmIntrinsics::_numberOfTrailingZeros_l:
555 case vmIntrinsics::_bitCount_i:
556 case vmIntrinsics::_bitCount_l:
557 case vmIntrinsics::_reverse_i:
558 case vmIntrinsics::_reverse_l:
559 case vmIntrinsics::_reverseBytes_i:
560 case vmIntrinsics::_reverseBytes_l:
561 case vmIntrinsics::_reverseBytes_s:
562 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
563
564 case vmIntrinsics::_compress_i:
565 case vmIntrinsics::_compress_l:
566 case vmIntrinsics::_expand_i:
567 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
568
569 case vmIntrinsics::_compareUnsigned_i:
570 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
571
572 case vmIntrinsics::_divideUnsigned_i:
573 case vmIntrinsics::_divideUnsigned_l:
574 case vmIntrinsics::_remainderUnsigned_i:
575 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
576
577 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
578
579 case vmIntrinsics::_Reference_get: return inline_reference_get();
580 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
581 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
582 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
583 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
584
585 case vmIntrinsics::_Class_cast: return inline_Class_cast();
586
587 case vmIntrinsics::_aescrypt_encryptBlock:
588 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
589
590 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
591 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
592 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
593
594 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
595 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
596 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
597
598 case vmIntrinsics::_counterMode_AESCrypt:
599 return inline_counterMode_AESCrypt(intrinsic_id());
600
601 case vmIntrinsics::_galoisCounterMode_AESCrypt:
602 return inline_galoisCounterMode_AESCrypt();
603
604 case vmIntrinsics::_md5_implCompress:
605 case vmIntrinsics::_sha_implCompress:
606 case vmIntrinsics::_sha2_implCompress:
607 case vmIntrinsics::_sha5_implCompress:
608 case vmIntrinsics::_sha3_implCompress:
609 return inline_digestBase_implCompress(intrinsic_id());
610
611 case vmIntrinsics::_digestBase_implCompressMB:
612 return inline_digestBase_implCompressMB(predicate);
613
614 case vmIntrinsics::_multiplyToLen:
615 return inline_multiplyToLen();
616
617 case vmIntrinsics::_squareToLen:
618 return inline_squareToLen();
619
620 case vmIntrinsics::_mulAdd:
621 return inline_mulAdd();
622
623 case vmIntrinsics::_montgomeryMultiply:
624 return inline_montgomeryMultiply();
625 case vmIntrinsics::_montgomerySquare:
626 return inline_montgomerySquare();
627
628 case vmIntrinsics::_bigIntegerRightShiftWorker:
629 return inline_bigIntegerShift(true);
630 case vmIntrinsics::_bigIntegerLeftShiftWorker:
631 return inline_bigIntegerShift(false);
632
633 case vmIntrinsics::_vectorizedMismatch:
634 return inline_vectorizedMismatch();
635
636 case vmIntrinsics::_ghash_processBlocks:
637 return inline_ghash_processBlocks();
638 case vmIntrinsics::_chacha20Block:
639 return inline_chacha20Block();
640 case vmIntrinsics::_base64_encodeBlock:
641 return inline_base64_encodeBlock();
642 case vmIntrinsics::_base64_decodeBlock:
643 return inline_base64_decodeBlock();
644 case vmIntrinsics::_poly1305_processBlocks:
645 return inline_poly1305_processBlocks();
646 case vmIntrinsics::_intpoly_montgomeryMult_P256:
647 return inline_intpoly_montgomeryMult_P256();
648 case vmIntrinsics::_intpoly_assign:
649 return inline_intpoly_assign();
650 case vmIntrinsics::_encodeISOArray:
651 case vmIntrinsics::_encodeByteISOArray:
652 return inline_encodeISOArray(false);
653 case vmIntrinsics::_encodeAsciiArray:
654 return inline_encodeISOArray(true);
655
656 case vmIntrinsics::_updateCRC32:
657 return inline_updateCRC32();
658 case vmIntrinsics::_updateBytesCRC32:
659 return inline_updateBytesCRC32();
660 case vmIntrinsics::_updateByteBufferCRC32:
661 return inline_updateByteBufferCRC32();
662
663 case vmIntrinsics::_updateBytesCRC32C:
664 return inline_updateBytesCRC32C();
665 case vmIntrinsics::_updateDirectByteBufferCRC32C:
666 return inline_updateDirectByteBufferCRC32C();
667
668 case vmIntrinsics::_updateBytesAdler32:
669 return inline_updateBytesAdler32();
670 case vmIntrinsics::_updateByteBufferAdler32:
671 return inline_updateByteBufferAdler32();
672
673 case vmIntrinsics::_profileBoolean:
674 return inline_profileBoolean();
675 case vmIntrinsics::_isCompileConstant:
676 return inline_isCompileConstant();
677
678 case vmIntrinsics::_countPositives:
679 return inline_countPositives();
680
681 case vmIntrinsics::_fmaD:
682 case vmIntrinsics::_fmaF:
683 return inline_fma(intrinsic_id());
684
685 case vmIntrinsics::_isDigit:
686 case vmIntrinsics::_isLowerCase:
687 case vmIntrinsics::_isUpperCase:
688 case vmIntrinsics::_isWhitespace:
689 return inline_character_compare(intrinsic_id());
690
691 case vmIntrinsics::_min:
692 case vmIntrinsics::_max:
693 case vmIntrinsics::_min_strict:
694 case vmIntrinsics::_max_strict:
695 return inline_min_max(intrinsic_id());
696
697 case vmIntrinsics::_maxF:
698 case vmIntrinsics::_minF:
699 case vmIntrinsics::_maxD:
700 case vmIntrinsics::_minD:
701 case vmIntrinsics::_maxF_strict:
702 case vmIntrinsics::_minF_strict:
703 case vmIntrinsics::_maxD_strict:
704 case vmIntrinsics::_minD_strict:
705 return inline_fp_min_max(intrinsic_id());
706
707 case vmIntrinsics::_VectorUnaryOp:
708 return inline_vector_nary_operation(1);
709 case vmIntrinsics::_VectorBinaryOp:
710 return inline_vector_nary_operation(2);
711 case vmIntrinsics::_VectorTernaryOp:
712 return inline_vector_nary_operation(3);
713 case vmIntrinsics::_VectorFromBitsCoerced:
714 return inline_vector_frombits_coerced();
715 case vmIntrinsics::_VectorShuffleIota:
716 return inline_vector_shuffle_iota();
717 case vmIntrinsics::_VectorMaskOp:
718 return inline_vector_mask_operation();
719 case vmIntrinsics::_VectorShuffleToVector:
720 return inline_vector_shuffle_to_vector();
721 case vmIntrinsics::_VectorWrapShuffleIndexes:
722 return inline_vector_wrap_shuffle_indexes();
723 case vmIntrinsics::_VectorLoadOp:
724 return inline_vector_mem_operation(/*is_store=*/false);
725 case vmIntrinsics::_VectorLoadMaskedOp:
726 return inline_vector_mem_masked_operation(/*is_store*/false);
727 case vmIntrinsics::_VectorStoreOp:
728 return inline_vector_mem_operation(/*is_store=*/true);
729 case vmIntrinsics::_VectorStoreMaskedOp:
730 return inline_vector_mem_masked_operation(/*is_store=*/true);
731 case vmIntrinsics::_VectorGatherOp:
732 return inline_vector_gather_scatter(/*is_scatter*/ false);
733 case vmIntrinsics::_VectorScatterOp:
734 return inline_vector_gather_scatter(/*is_scatter*/ true);
735 case vmIntrinsics::_VectorReductionCoerced:
736 return inline_vector_reduction();
737 case vmIntrinsics::_VectorTest:
738 return inline_vector_test();
739 case vmIntrinsics::_VectorBlend:
740 return inline_vector_blend();
741 case vmIntrinsics::_VectorRearrange:
742 return inline_vector_rearrange();
743 case vmIntrinsics::_VectorSelectFrom:
744 return inline_vector_select_from();
745 case vmIntrinsics::_VectorCompare:
746 return inline_vector_compare();
747 case vmIntrinsics::_VectorBroadcastInt:
748 return inline_vector_broadcast_int();
749 case vmIntrinsics::_VectorConvert:
750 return inline_vector_convert();
751 case vmIntrinsics::_VectorInsert:
752 return inline_vector_insert();
753 case vmIntrinsics::_VectorExtract:
754 return inline_vector_extract();
755 case vmIntrinsics::_VectorCompressExpand:
756 return inline_vector_compress_expand();
757 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
758 return inline_vector_select_from_two_vectors();
759 case vmIntrinsics::_IndexVector:
760 return inline_index_vector();
761 case vmIntrinsics::_IndexPartiallyInUpperRange:
762 return inline_index_partially_in_upper_range();
763
764 case vmIntrinsics::_getObjectSize:
765 return inline_getObjectSize();
766
767 case vmIntrinsics::_blackhole:
768 return inline_blackhole();
769
770 default:
771 // If you get here, it may be that someone has added a new intrinsic
772 // to the list in vmIntrinsics.hpp without implementing it here.
773 #ifndef PRODUCT
774 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
775 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
776 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
777 }
778 #endif
779 return false;
780 }
781 }
782
783 Node* LibraryCallKit::try_to_predicate(int predicate) {
784 if (!jvms()->has_method()) {
785 // Root JVMState has a null method.
786 assert(map()->memory()->Opcode() == Op_Parm, "");
787 // Insert the memory aliasing node
788 set_all_memory(reset_memory());
789 }
790 assert(merged_memory(), "");
791
792 switch (intrinsic_id()) {
793 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
794 return inline_cipherBlockChaining_AESCrypt_predicate(false);
795 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
796 return inline_cipherBlockChaining_AESCrypt_predicate(true);
797 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
798 return inline_electronicCodeBook_AESCrypt_predicate(false);
799 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
800 return inline_electronicCodeBook_AESCrypt_predicate(true);
801 case vmIntrinsics::_counterMode_AESCrypt:
802 return inline_counterMode_AESCrypt_predicate();
803 case vmIntrinsics::_digestBase_implCompressMB:
804 return inline_digestBase_implCompressMB_predicate(predicate);
805 case vmIntrinsics::_galoisCounterMode_AESCrypt:
806 return inline_galoisCounterMode_AESCrypt_predicate();
807
808 default:
809 // If you get here, it may be that someone has added a new intrinsic
810 // to the list in vmIntrinsics.hpp without implementing it here.
811 #ifndef PRODUCT
812 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
813 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
814 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
815 }
816 #endif
817 Node* slow_ctl = control();
818 set_control(top()); // No fast path intrinsic
819 return slow_ctl;
820 }
821 }
822
823 //------------------------------set_result-------------------------------
824 // Helper function for finishing intrinsics.
825 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
826 record_for_igvn(region);
827 set_control(_gvn.transform(region));
828 set_result( _gvn.transform(value));
829 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
830 }
831
832 //------------------------------generate_guard---------------------------
833 // Helper function for generating guarded fast-slow graph structures.
834 // The given 'test', if true, guards a slow path. If the test fails
835 // then a fast path can be taken. (We generally hope it fails.)
836 // In all cases, GraphKit::control() is updated to the fast path.
837 // The returned value represents the control for the slow path.
838 // The return value is never 'top'; it is either a valid control
839 // or null if it is obvious that the slow path can never be taken.
840 // Also, if region and the slow control are not null, the slow edge
841 // is appended to the region.
842 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
843 if (stopped()) {
844 // Already short circuited.
845 return nullptr;
846 }
847
848 // Build an if node and its projections.
849 // If test is true we take the slow path, which we assume is uncommon.
850 if (_gvn.type(test) == TypeInt::ZERO) {
851 // The slow branch is never taken. No need to build this guard.
852 return nullptr;
853 }
854
855 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
856
857 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
858 if (if_slow == top()) {
859 // The slow branch is never taken. No need to build this guard.
860 return nullptr;
861 }
862
863 if (region != nullptr)
864 region->add_req(if_slow);
865
866 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
867 set_control(if_fast);
868
869 return if_slow;
870 }
871
872 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
873 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
874 }
875 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
876 return generate_guard(test, region, PROB_FAIR);
877 }
878
879 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
880 Node* *pos_index) {
881 if (stopped())
882 return nullptr; // already stopped
883 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
884 return nullptr; // index is already adequately typed
885 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
886 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
887 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
888 if (is_neg != nullptr && pos_index != nullptr) {
889 // Emulate effect of Parse::adjust_map_after_if.
890 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
891 (*pos_index) = _gvn.transform(ccast);
892 }
893 return is_neg;
894 }
895
896 // Make sure that 'position' is a valid limit index, in [0..length].
897 // There are two equivalent plans for checking this:
898 // A. (offset + copyLength) unsigned<= arrayLength
899 // B. offset <= (arrayLength - copyLength)
900 // We require that all of the values above, except for the sum and
901 // difference, are already known to be non-negative.
902 // Plan A is robust in the face of overflow, if offset and copyLength
903 // are both hugely positive.
904 //
905 // Plan B is less direct and intuitive, but it does not overflow at
906 // all, since the difference of two non-negatives is always
907 // representable. Whenever Java methods must perform the equivalent
908 // check they generally use Plan B instead of Plan A.
909 // For the moment we use Plan A.
910 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
911 Node* subseq_length,
912 Node* array_length,
913 RegionNode* region) {
914 if (stopped())
915 return nullptr; // already stopped
916 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
917 if (zero_offset && subseq_length->eqv_uncast(array_length))
918 return nullptr; // common case of whole-array copy
919 Node* last = subseq_length;
920 if (!zero_offset) // last += offset
921 last = _gvn.transform(new AddINode(last, offset));
922 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
923 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
924 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
925 return is_over;
926 }
927
928 // Emit range checks for the given String.value byte array
929 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
930 if (stopped()) {
931 return; // already stopped
932 }
933 RegionNode* bailout = new RegionNode(1);
934 record_for_igvn(bailout);
935 if (char_count) {
936 // Convert char count to byte count
937 count = _gvn.transform(new LShiftINode(count, intcon(1)));
938 }
939
940 // Offset and count must not be negative
941 generate_negative_guard(offset, bailout);
942 generate_negative_guard(count, bailout);
943 // Offset + count must not exceed length of array
944 generate_limit_guard(offset, count, load_array_length(array), bailout);
945
946 if (bailout->req() > 1) {
947 PreserveJVMState pjvms(this);
948 set_control(_gvn.transform(bailout));
949 uncommon_trap(Deoptimization::Reason_intrinsic,
950 Deoptimization::Action_maybe_recompile);
951 }
952 }
953
954 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
955 bool is_immutable) {
956 ciKlass* thread_klass = env()->Thread_klass();
957 const Type* thread_type
958 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
959
960 Node* thread = _gvn.transform(new ThreadLocalNode());
961 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
962 tls_output = thread;
963
964 Node* thread_obj_handle
965 = (is_immutable
966 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
967 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
968 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
969 thread_obj_handle = _gvn.transform(thread_obj_handle);
970
971 DecoratorSet decorators = IN_NATIVE;
972 if (is_immutable) {
973 decorators |= C2_IMMUTABLE_MEMORY;
974 }
975 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
976 }
977
978 //--------------------------generate_current_thread--------------------
979 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
980 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
981 /*is_immutable*/false);
982 }
983
984 //--------------------------generate_virtual_thread--------------------
985 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
986 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
987 !C->method()->changes_current_thread());
988 }
989
990 //------------------------------make_string_method_node------------------------
991 // Helper method for String intrinsic functions. This version is called with
992 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
993 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
994 // containing the lengths of str1 and str2.
995 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
996 Node* result = nullptr;
997 switch (opcode) {
998 case Op_StrIndexOf:
999 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1000 str1_start, cnt1, str2_start, cnt2, ae);
1001 break;
1002 case Op_StrComp:
1003 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1004 str1_start, cnt1, str2_start, cnt2, ae);
1005 break;
1006 case Op_StrEquals:
1007 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1008 // Use the constant length if there is one because optimized match rule may exist.
1009 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1010 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1011 break;
1012 default:
1013 ShouldNotReachHere();
1014 return nullptr;
1015 }
1016
1017 // All these intrinsics have checks.
1018 C->set_has_split_ifs(true); // Has chance for split-if optimization
1019 clear_upper_avx();
1020
1021 return _gvn.transform(result);
1022 }
1023
1024 //------------------------------inline_string_compareTo------------------------
1025 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1026 Node* arg1 = argument(0);
1027 Node* arg2 = argument(1);
1028
1029 arg1 = must_be_not_null(arg1, true);
1030 arg2 = must_be_not_null(arg2, true);
1031
1032 // Get start addr and length of first argument
1033 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1034 Node* arg1_cnt = load_array_length(arg1);
1035
1036 // Get start addr and length of second argument
1037 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1038 Node* arg2_cnt = load_array_length(arg2);
1039
1040 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1041 set_result(result);
1042 return true;
1043 }
1044
1045 //------------------------------inline_string_equals------------------------
1046 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1047 Node* arg1 = argument(0);
1048 Node* arg2 = argument(1);
1049
1050 // paths (plus control) merge
1051 RegionNode* region = new RegionNode(3);
1052 Node* phi = new PhiNode(region, TypeInt::BOOL);
1053
1054 if (!stopped()) {
1055
1056 arg1 = must_be_not_null(arg1, true);
1057 arg2 = must_be_not_null(arg2, true);
1058
1059 // Get start addr and length of first argument
1060 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1061 Node* arg1_cnt = load_array_length(arg1);
1062
1063 // Get start addr and length of second argument
1064 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1065 Node* arg2_cnt = load_array_length(arg2);
1066
1067 // Check for arg1_cnt != arg2_cnt
1068 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1069 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1070 Node* if_ne = generate_slow_guard(bol, nullptr);
1071 if (if_ne != nullptr) {
1072 phi->init_req(2, intcon(0));
1073 region->init_req(2, if_ne);
1074 }
1075
1076 // Check for count == 0 is done by assembler code for StrEquals.
1077
1078 if (!stopped()) {
1079 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1080 phi->init_req(1, equals);
1081 region->init_req(1, control());
1082 }
1083 }
1084
1085 // post merge
1086 set_control(_gvn.transform(region));
1087 record_for_igvn(region);
1088
1089 set_result(_gvn.transform(phi));
1090 return true;
1091 }
1092
1093 //------------------------------inline_array_equals----------------------------
1094 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1095 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1096 Node* arg1 = argument(0);
1097 Node* arg2 = argument(1);
1098
1099 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1100 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1101 clear_upper_avx();
1102
1103 return true;
1104 }
1105
1106
1107 //------------------------------inline_countPositives------------------------------
1108 bool LibraryCallKit::inline_countPositives() {
1109 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1110 return false;
1111 }
1112
1113 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1114 // no receiver since it is static method
1115 Node* ba = argument(0);
1116 Node* offset = argument(1);
1117 Node* len = argument(2);
1118
1119 ba = must_be_not_null(ba, true);
1120
1121 // Range checks
1122 generate_string_range_check(ba, offset, len, false);
1123 if (stopped()) {
1124 return true;
1125 }
1126 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1127 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1128 set_result(_gvn.transform(result));
1129 clear_upper_avx();
1130 return true;
1131 }
1132
1133 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1134 Node* index = argument(0);
1135 Node* length = bt == T_INT ? argument(1) : argument(2);
1136 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1137 return false;
1138 }
1139
1140 // check that length is positive
1141 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1142 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1143
1144 {
1145 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1146 uncommon_trap(Deoptimization::Reason_intrinsic,
1147 Deoptimization::Action_make_not_entrant);
1148 }
1149
1150 if (stopped()) {
1151 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1152 return true;
1153 }
1154
1155 // length is now known positive, add a cast node to make this explicit
1156 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1157 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1158 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1159 ConstraintCastNode::RegularDependency, bt);
1160 casted_length = _gvn.transform(casted_length);
1161 replace_in_map(length, casted_length);
1162 length = casted_length;
1163
1164 // Use an unsigned comparison for the range check itself
1165 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1166 BoolTest::mask btest = BoolTest::lt;
1167 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1168 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1169 _gvn.set_type(rc, rc->Value(&_gvn));
1170 if (!rc_bool->is_Con()) {
1171 record_for_igvn(rc);
1172 }
1173 set_control(_gvn.transform(new IfTrueNode(rc)));
1174 {
1175 PreserveJVMState pjvms(this);
1176 set_control(_gvn.transform(new IfFalseNode(rc)));
1177 uncommon_trap(Deoptimization::Reason_range_check,
1178 Deoptimization::Action_make_not_entrant);
1179 }
1180
1181 if (stopped()) {
1182 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1183 return true;
1184 }
1185
1186 // index is now known to be >= 0 and < length, cast it
1187 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1188 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1189 ConstraintCastNode::RegularDependency, bt);
1190 result = _gvn.transform(result);
1191 set_result(result);
1192 replace_in_map(index, result);
1193 return true;
1194 }
1195
1196 //------------------------------inline_string_indexOf------------------------
1197 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1198 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1199 return false;
1200 }
1201 Node* src = argument(0);
1202 Node* tgt = argument(1);
1203
1204 // Make the merge point
1205 RegionNode* result_rgn = new RegionNode(4);
1206 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1207
1208 src = must_be_not_null(src, true);
1209 tgt = must_be_not_null(tgt, true);
1210
1211 // Get start addr and length of source string
1212 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1213 Node* src_count = load_array_length(src);
1214
1215 // Get start addr and length of substring
1216 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1217 Node* tgt_count = load_array_length(tgt);
1218
1219 Node* result = nullptr;
1220 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1221
1222 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1223 // Divide src size by 2 if String is UTF16 encoded
1224 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1225 }
1226 if (ae == StrIntrinsicNode::UU) {
1227 // Divide substring size by 2 if String is UTF16 encoded
1228 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1229 }
1230
1231 if (call_opt_stub) {
1232 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1233 StubRoutines::_string_indexof_array[ae],
1234 "stringIndexOf", TypePtr::BOTTOM, src_start,
1235 src_count, tgt_start, tgt_count);
1236 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1237 } else {
1238 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1239 result_rgn, result_phi, ae);
1240 }
1241 if (result != nullptr) {
1242 result_phi->init_req(3, result);
1243 result_rgn->init_req(3, control());
1244 }
1245 set_control(_gvn.transform(result_rgn));
1246 record_for_igvn(result_rgn);
1247 set_result(_gvn.transform(result_phi));
1248
1249 return true;
1250 }
1251
1252 //-----------------------------inline_string_indexOfI-----------------------
1253 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1254 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1255 return false;
1256 }
1257 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1258 return false;
1259 }
1260
1261 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1262 Node* src = argument(0); // byte[]
1263 Node* src_count = argument(1); // char count
1264 Node* tgt = argument(2); // byte[]
1265 Node* tgt_count = argument(3); // char count
1266 Node* from_index = argument(4); // char index
1267
1268 src = must_be_not_null(src, true);
1269 tgt = must_be_not_null(tgt, true);
1270
1271 // Multiply byte array index by 2 if String is UTF16 encoded
1272 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1273 src_count = _gvn.transform(new SubINode(src_count, from_index));
1274 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1275 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1276
1277 // Range checks
1278 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1279 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1280 if (stopped()) {
1281 return true;
1282 }
1283
1284 RegionNode* region = new RegionNode(5);
1285 Node* phi = new PhiNode(region, TypeInt::INT);
1286 Node* result = nullptr;
1287
1288 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1289
1290 if (call_opt_stub) {
1291 assert(arrayOopDesc::base_offset_in_bytes(T_BYTE) >= 16, "Needed for indexOf");
1292 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1293 StubRoutines::_string_indexof_array[ae],
1294 "stringIndexOf", TypePtr::BOTTOM, src_start,
1295 src_count, tgt_start, tgt_count);
1296 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1297 } else {
1298 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1299 region, phi, ae);
1300 }
1301 if (result != nullptr) {
1302 // The result is index relative to from_index if substring was found, -1 otherwise.
1303 // Generate code which will fold into cmove.
1304 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1305 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1306
1307 Node* if_lt = generate_slow_guard(bol, nullptr);
1308 if (if_lt != nullptr) {
1309 // result == -1
1310 phi->init_req(3, result);
1311 region->init_req(3, if_lt);
1312 }
1313 if (!stopped()) {
1314 result = _gvn.transform(new AddINode(result, from_index));
1315 phi->init_req(4, result);
1316 region->init_req(4, control());
1317 }
1318 }
1319
1320 set_control(_gvn.transform(region));
1321 record_for_igvn(region);
1322 set_result(_gvn.transform(phi));
1323 clear_upper_avx();
1324
1325 return true;
1326 }
1327
1328 // Create StrIndexOfNode with fast path checks
1329 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1330 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1331 // Check for substr count > string count
1332 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1333 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1334 Node* if_gt = generate_slow_guard(bol, nullptr);
1335 if (if_gt != nullptr) {
1336 phi->init_req(1, intcon(-1));
1337 region->init_req(1, if_gt);
1338 }
1339 if (!stopped()) {
1340 // Check for substr count == 0
1341 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1342 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1343 Node* if_zero = generate_slow_guard(bol, nullptr);
1344 if (if_zero != nullptr) {
1345 phi->init_req(2, intcon(0));
1346 region->init_req(2, if_zero);
1347 }
1348 }
1349 if (!stopped()) {
1350 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1351 }
1352 return nullptr;
1353 }
1354
1355 //-----------------------------inline_string_indexOfChar-----------------------
1356 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1357 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1358 return false;
1359 }
1360 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1361 return false;
1362 }
1363 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1364 Node* src = argument(0); // byte[]
1365 Node* int_ch = argument(1);
1366 Node* from_index = argument(2);
1367 Node* max = argument(3);
1368
1369 src = must_be_not_null(src, true);
1370
1371 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1372 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1373 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1374
1375 // Range checks
1376 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1377
1378 // Check for int_ch >= 0
1379 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1380 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1381 {
1382 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1383 uncommon_trap(Deoptimization::Reason_intrinsic,
1384 Deoptimization::Action_maybe_recompile);
1385 }
1386 if (stopped()) {
1387 return true;
1388 }
1389
1390 RegionNode* region = new RegionNode(3);
1391 Node* phi = new PhiNode(region, TypeInt::INT);
1392
1393 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1394 C->set_has_split_ifs(true); // Has chance for split-if optimization
1395 _gvn.transform(result);
1396
1397 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1398 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1399
1400 Node* if_lt = generate_slow_guard(bol, nullptr);
1401 if (if_lt != nullptr) {
1402 // result == -1
1403 phi->init_req(2, result);
1404 region->init_req(2, if_lt);
1405 }
1406 if (!stopped()) {
1407 result = _gvn.transform(new AddINode(result, from_index));
1408 phi->init_req(1, result);
1409 region->init_req(1, control());
1410 }
1411 set_control(_gvn.transform(region));
1412 record_for_igvn(region);
1413 set_result(_gvn.transform(phi));
1414 clear_upper_avx();
1415
1416 return true;
1417 }
1418 //---------------------------inline_string_copy---------------------
1419 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1420 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1421 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1422 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1423 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1424 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1425 bool LibraryCallKit::inline_string_copy(bool compress) {
1426 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1427 return false;
1428 }
1429 int nargs = 5; // 2 oops, 3 ints
1430 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1431
1432 Node* src = argument(0);
1433 Node* src_offset = argument(1);
1434 Node* dst = argument(2);
1435 Node* dst_offset = argument(3);
1436 Node* length = argument(4);
1437
1438 // Check for allocation before we add nodes that would confuse
1439 // tightly_coupled_allocation()
1440 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1441
1442 // Figure out the size and type of the elements we will be copying.
1443 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1444 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1445 if (src_type == nullptr || dst_type == nullptr) {
1446 return false;
1447 }
1448 BasicType src_elem = src_type->elem()->array_element_basic_type();
1449 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1450 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1451 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1452 "Unsupported array types for inline_string_copy");
1453
1454 src = must_be_not_null(src, true);
1455 dst = must_be_not_null(dst, true);
1456
1457 // Convert char[] offsets to byte[] offsets
1458 bool convert_src = (compress && src_elem == T_BYTE);
1459 bool convert_dst = (!compress && dst_elem == T_BYTE);
1460 if (convert_src) {
1461 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1462 } else if (convert_dst) {
1463 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1464 }
1465
1466 // Range checks
1467 generate_string_range_check(src, src_offset, length, convert_src);
1468 generate_string_range_check(dst, dst_offset, length, convert_dst);
1469 if (stopped()) {
1470 return true;
1471 }
1472
1473 Node* src_start = array_element_address(src, src_offset, src_elem);
1474 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1475 // 'src_start' points to src array + scaled offset
1476 // 'dst_start' points to dst array + scaled offset
1477 Node* count = nullptr;
1478 if (compress) {
1479 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1480 } else {
1481 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1482 }
1483
1484 if (alloc != nullptr) {
1485 if (alloc->maybe_set_complete(&_gvn)) {
1486 // "You break it, you buy it."
1487 InitializeNode* init = alloc->initialization();
1488 assert(init->is_complete(), "we just did this");
1489 init->set_complete_with_arraycopy();
1490 assert(dst->is_CheckCastPP(), "sanity");
1491 assert(dst->in(0)->in(0) == init, "dest pinned");
1492 }
1493 // Do not let stores that initialize this object be reordered with
1494 // a subsequent store that would make this object accessible by
1495 // other threads.
1496 // Record what AllocateNode this StoreStore protects so that
1497 // escape analysis can go from the MemBarStoreStoreNode to the
1498 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1499 // based on the escape status of the AllocateNode.
1500 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1501 }
1502 if (compress) {
1503 set_result(_gvn.transform(count));
1504 }
1505 clear_upper_avx();
1506
1507 return true;
1508 }
1509
1510 #ifdef _LP64
1511 #define XTOP ,top() /*additional argument*/
1512 #else //_LP64
1513 #define XTOP /*no additional argument*/
1514 #endif //_LP64
1515
1516 //------------------------inline_string_toBytesU--------------------------
1517 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1518 bool LibraryCallKit::inline_string_toBytesU() {
1519 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1520 return false;
1521 }
1522 // Get the arguments.
1523 Node* value = argument(0);
1524 Node* offset = argument(1);
1525 Node* length = argument(2);
1526
1527 Node* newcopy = nullptr;
1528
1529 // Set the original stack and the reexecute bit for the interpreter to reexecute
1530 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1531 { PreserveReexecuteState preexecs(this);
1532 jvms()->set_should_reexecute(true);
1533
1534 // Check if a null path was taken unconditionally.
1535 value = null_check(value);
1536
1537 RegionNode* bailout = new RegionNode(1);
1538 record_for_igvn(bailout);
1539
1540 // Range checks
1541 generate_negative_guard(offset, bailout);
1542 generate_negative_guard(length, bailout);
1543 generate_limit_guard(offset, length, load_array_length(value), bailout);
1544 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1545 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1546
1547 if (bailout->req() > 1) {
1548 PreserveJVMState pjvms(this);
1549 set_control(_gvn.transform(bailout));
1550 uncommon_trap(Deoptimization::Reason_intrinsic,
1551 Deoptimization::Action_maybe_recompile);
1552 }
1553 if (stopped()) {
1554 return true;
1555 }
1556
1557 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1558 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1559 newcopy = new_array(klass_node, size, 0); // no arguments to push
1560 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1561 guarantee(alloc != nullptr, "created above");
1562
1563 // Calculate starting addresses.
1564 Node* src_start = array_element_address(value, offset, T_CHAR);
1565 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1566
1567 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1568 const TypeInt* toffset = gvn().type(offset)->is_int();
1569 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1570
1571 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1572 const char* copyfunc_name = "arraycopy";
1573 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1574 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1575 OptoRuntime::fast_arraycopy_Type(),
1576 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1577 src_start, dst_start, ConvI2X(length) XTOP);
1578 // Do not let reads from the cloned object float above the arraycopy.
1579 if (alloc->maybe_set_complete(&_gvn)) {
1580 // "You break it, you buy it."
1581 InitializeNode* init = alloc->initialization();
1582 assert(init->is_complete(), "we just did this");
1583 init->set_complete_with_arraycopy();
1584 assert(newcopy->is_CheckCastPP(), "sanity");
1585 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1586 }
1587 // Do not let stores that initialize this object be reordered with
1588 // a subsequent store that would make this object accessible by
1589 // other threads.
1590 // Record what AllocateNode this StoreStore protects so that
1591 // escape analysis can go from the MemBarStoreStoreNode to the
1592 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1593 // based on the escape status of the AllocateNode.
1594 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1595 } // original reexecute is set back here
1596
1597 C->set_has_split_ifs(true); // Has chance for split-if optimization
1598 if (!stopped()) {
1599 set_result(newcopy);
1600 }
1601 clear_upper_avx();
1602
1603 return true;
1604 }
1605
1606 //------------------------inline_string_getCharsU--------------------------
1607 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1608 bool LibraryCallKit::inline_string_getCharsU() {
1609 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1610 return false;
1611 }
1612
1613 // Get the arguments.
1614 Node* src = argument(0);
1615 Node* src_begin = argument(1);
1616 Node* src_end = argument(2); // exclusive offset (i < src_end)
1617 Node* dst = argument(3);
1618 Node* dst_begin = argument(4);
1619
1620 // Check for allocation before we add nodes that would confuse
1621 // tightly_coupled_allocation()
1622 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1623
1624 // Check if a null path was taken unconditionally.
1625 src = null_check(src);
1626 dst = null_check(dst);
1627 if (stopped()) {
1628 return true;
1629 }
1630
1631 // Get length and convert char[] offset to byte[] offset
1632 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1633 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1634
1635 // Range checks
1636 generate_string_range_check(src, src_begin, length, true);
1637 generate_string_range_check(dst, dst_begin, length, false);
1638 if (stopped()) {
1639 return true;
1640 }
1641
1642 if (!stopped()) {
1643 // Calculate starting addresses.
1644 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1645 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1646
1647 // Check if array addresses are aligned to HeapWordSize
1648 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1649 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1650 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1651 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1652
1653 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1654 const char* copyfunc_name = "arraycopy";
1655 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1656 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1657 OptoRuntime::fast_arraycopy_Type(),
1658 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1659 src_start, dst_start, ConvI2X(length) XTOP);
1660 // Do not let reads from the cloned object float above the arraycopy.
1661 if (alloc != nullptr) {
1662 if (alloc->maybe_set_complete(&_gvn)) {
1663 // "You break it, you buy it."
1664 InitializeNode* init = alloc->initialization();
1665 assert(init->is_complete(), "we just did this");
1666 init->set_complete_with_arraycopy();
1667 assert(dst->is_CheckCastPP(), "sanity");
1668 assert(dst->in(0)->in(0) == init, "dest pinned");
1669 }
1670 // Do not let stores that initialize this object be reordered with
1671 // a subsequent store that would make this object accessible by
1672 // other threads.
1673 // Record what AllocateNode this StoreStore protects so that
1674 // escape analysis can go from the MemBarStoreStoreNode to the
1675 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1676 // based on the escape status of the AllocateNode.
1677 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1678 } else {
1679 insert_mem_bar(Op_MemBarCPUOrder);
1680 }
1681 }
1682
1683 C->set_has_split_ifs(true); // Has chance for split-if optimization
1684 return true;
1685 }
1686
1687 //----------------------inline_string_char_access----------------------------
1688 // Store/Load char to/from byte[] array.
1689 // static void StringUTF16.putChar(byte[] val, int index, int c)
1690 // static char StringUTF16.getChar(byte[] val, int index)
1691 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1692 Node* value = argument(0);
1693 Node* index = argument(1);
1694 Node* ch = is_store ? argument(2) : nullptr;
1695
1696 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1697 // correctly requires matched array shapes.
1698 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1699 "sanity: byte[] and char[] bases agree");
1700 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1701 "sanity: byte[] and char[] scales agree");
1702
1703 // Bail when getChar over constants is requested: constant folding would
1704 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1705 // Java method would constant fold nicely instead.
1706 if (!is_store && value->is_Con() && index->is_Con()) {
1707 return false;
1708 }
1709
1710 // Save state and restore on bailout
1711 uint old_sp = sp();
1712 SafePointNode* old_map = clone_map();
1713
1714 value = must_be_not_null(value, true);
1715
1716 Node* adr = array_element_address(value, index, T_CHAR);
1717 if (adr->is_top()) {
1718 set_map(old_map);
1719 set_sp(old_sp);
1720 return false;
1721 }
1722 destruct_map_clone(old_map);
1723 if (is_store) {
1724 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1725 } else {
1726 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);
1727 set_result(ch);
1728 }
1729 return true;
1730 }
1731
1732 //--------------------------round_double_node--------------------------------
1733 // Round a double node if necessary.
1734 Node* LibraryCallKit::round_double_node(Node* n) {
1735 if (Matcher::strict_fp_requires_explicit_rounding) {
1736 #ifdef IA32
1737 if (UseSSE < 2) {
1738 n = _gvn.transform(new RoundDoubleNode(nullptr, n));
1739 }
1740 #else
1741 Unimplemented();
1742 #endif // IA32
1743 }
1744 return n;
1745 }
1746
1747 //------------------------------inline_math-----------------------------------
1748 // public static double Math.abs(double)
1749 // public static double Math.sqrt(double)
1750 // public static double Math.log(double)
1751 // public static double Math.log10(double)
1752 // public static double Math.round(double)
1753 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1754 Node* arg = round_double_node(argument(0));
1755 Node* n = nullptr;
1756 switch (id) {
1757 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1758 case vmIntrinsics::_dsqrt:
1759 case vmIntrinsics::_dsqrt_strict:
1760 n = new SqrtDNode(C, control(), arg); break;
1761 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1762 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1763 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1764 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1765 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1766 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1767 default: fatal_unexpected_iid(id); break;
1768 }
1769 set_result(_gvn.transform(n));
1770 return true;
1771 }
1772
1773 //------------------------------inline_math-----------------------------------
1774 // public static float Math.abs(float)
1775 // public static int Math.abs(int)
1776 // public static long Math.abs(long)
1777 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1778 Node* arg = argument(0);
1779 Node* n = nullptr;
1780 switch (id) {
1781 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1782 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1783 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1784 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1785 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1786 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1787 default: fatal_unexpected_iid(id); break;
1788 }
1789 set_result(_gvn.transform(n));
1790 return true;
1791 }
1792
1793 //------------------------------runtime_math-----------------------------
1794 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1795 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1796 "must be (DD)D or (D)D type");
1797
1798 // Inputs
1799 Node* a = round_double_node(argument(0));
1800 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : nullptr;
1801
1802 const TypePtr* no_memory_effects = nullptr;
1803 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1804 no_memory_effects,
1805 a, top(), b, b ? top() : nullptr);
1806 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1807 #ifdef ASSERT
1808 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1809 assert(value_top == top(), "second value must be top");
1810 #endif
1811
1812 set_result(value);
1813 return true;
1814 }
1815
1816 //------------------------------inline_math_pow-----------------------------
1817 bool LibraryCallKit::inline_math_pow() {
1818 Node* exp = round_double_node(argument(2));
1819 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1820 if (d != nullptr) {
1821 if (d->getd() == 2.0) {
1822 // Special case: pow(x, 2.0) => x * x
1823 Node* base = round_double_node(argument(0));
1824 set_result(_gvn.transform(new MulDNode(base, base)));
1825 return true;
1826 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1827 // Special case: pow(x, 0.5) => sqrt(x)
1828 Node* base = round_double_node(argument(0));
1829 Node* zero = _gvn.zerocon(T_DOUBLE);
1830
1831 RegionNode* region = new RegionNode(3);
1832 Node* phi = new PhiNode(region, Type::DOUBLE);
1833
1834 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1835 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1836 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1837 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1838 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1839
1840 Node* if_pow = generate_slow_guard(test, nullptr);
1841 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1842 phi->init_req(1, value_sqrt);
1843 region->init_req(1, control());
1844
1845 if (if_pow != nullptr) {
1846 set_control(if_pow);
1847 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1848 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1849 const TypePtr* no_memory_effects = nullptr;
1850 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1851 no_memory_effects, base, top(), exp, top());
1852 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1853 #ifdef ASSERT
1854 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1855 assert(value_top == top(), "second value must be top");
1856 #endif
1857 phi->init_req(2, value_pow);
1858 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1859 }
1860
1861 C->set_has_split_ifs(true); // Has chance for split-if optimization
1862 set_control(_gvn.transform(region));
1863 record_for_igvn(region);
1864 set_result(_gvn.transform(phi));
1865
1866 return true;
1867 }
1868 }
1869
1870 return StubRoutines::dpow() != nullptr ?
1871 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1872 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1873 }
1874
1875 //------------------------------inline_math_native-----------------------------
1876 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1877 switch (id) {
1878 case vmIntrinsics::_dsin:
1879 return StubRoutines::dsin() != nullptr ?
1880 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1881 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1882 case vmIntrinsics::_dcos:
1883 return StubRoutines::dcos() != nullptr ?
1884 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1885 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1886 case vmIntrinsics::_dtan:
1887 return StubRoutines::dtan() != nullptr ?
1888 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1889 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1890 case vmIntrinsics::_dtanh:
1891 return StubRoutines::dtanh() != nullptr ?
1892 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1893 case vmIntrinsics::_dexp:
1894 return StubRoutines::dexp() != nullptr ?
1895 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1896 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1897 case vmIntrinsics::_dlog:
1898 return StubRoutines::dlog() != nullptr ?
1899 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1900 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1901 case vmIntrinsics::_dlog10:
1902 return StubRoutines::dlog10() != nullptr ?
1903 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1904 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1905
1906 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1907 case vmIntrinsics::_ceil:
1908 case vmIntrinsics::_floor:
1909 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1910
1911 case vmIntrinsics::_dsqrt:
1912 case vmIntrinsics::_dsqrt_strict:
1913 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1914 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1915 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1916 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1917 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1918
1919 case vmIntrinsics::_dpow: return inline_math_pow();
1920 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1921 case vmIntrinsics::_fcopySign: return inline_math(id);
1922 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1923 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1924 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1925
1926 // These intrinsics are not yet correctly implemented
1927 case vmIntrinsics::_datan2:
1928 return false;
1929
1930 default:
1931 fatal_unexpected_iid(id);
1932 return false;
1933 }
1934 }
1935
1936 //----------------------------inline_notify-----------------------------------*
1937 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1938 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1939 address func;
1940 if (id == vmIntrinsics::_notify) {
1941 func = OptoRuntime::monitor_notify_Java();
1942 } else {
1943 func = OptoRuntime::monitor_notifyAll_Java();
1944 }
1945 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1946 make_slow_call_ex(call, env()->Throwable_klass(), false);
1947 return true;
1948 }
1949
1950
1951 //----------------------------inline_min_max-----------------------------------
1952 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1953 set_result(generate_min_max(id, argument(0), argument(1)));
1954 return true;
1955 }
1956
1957 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1958 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1959 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1960 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1961 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1962
1963 {
1964 PreserveJVMState pjvms(this);
1965 PreserveReexecuteState preexecs(this);
1966 jvms()->set_should_reexecute(true);
1967
1968 set_control(slow_path);
1969 set_i_o(i_o());
1970
1971 uncommon_trap(Deoptimization::Reason_intrinsic,
1972 Deoptimization::Action_none);
1973 }
1974
1975 set_control(fast_path);
1976 set_result(math);
1977 }
1978
1979 template <typename OverflowOp>
1980 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1981 typedef typename OverflowOp::MathOp MathOp;
1982
1983 MathOp* mathOp = new MathOp(arg1, arg2);
1984 Node* operation = _gvn.transform( mathOp );
1985 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1986 inline_math_mathExact(operation, ofcheck);
1987 return true;
1988 }
1989
1990 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1991 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1992 }
1993
1994 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1995 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1996 }
1997
1998 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1999 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2000 }
2001
2002 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2003 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2004 }
2005
2006 bool LibraryCallKit::inline_math_negateExactI() {
2007 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2008 }
2009
2010 bool LibraryCallKit::inline_math_negateExactL() {
2011 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2012 }
2013
2014 bool LibraryCallKit::inline_math_multiplyExactI() {
2015 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2016 }
2017
2018 bool LibraryCallKit::inline_math_multiplyExactL() {
2019 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2020 }
2021
2022 bool LibraryCallKit::inline_math_multiplyHigh() {
2023 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2024 return true;
2025 }
2026
2027 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2028 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2029 return true;
2030 }
2031
2032 Node*
2033 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
2034 Node* result_val = nullptr;
2035 switch (id) {
2036 case vmIntrinsics::_min:
2037 case vmIntrinsics::_min_strict:
2038 result_val = _gvn.transform(new MinINode(x0, y0));
2039 break;
2040 case vmIntrinsics::_max:
2041 case vmIntrinsics::_max_strict:
2042 result_val = _gvn.transform(new MaxINode(x0, y0));
2043 break;
2044 default:
2045 fatal_unexpected_iid(id);
2046 break;
2047 }
2048 return result_val;
2049 }
2050
2051 inline int
2052 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2053 const TypePtr* base_type = TypePtr::NULL_PTR;
2054 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2055 if (base_type == nullptr) {
2056 // Unknown type.
2057 return Type::AnyPtr;
2058 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2059 // Since this is a null+long form, we have to switch to a rawptr.
2060 base = _gvn.transform(new CastX2PNode(offset));
2061 offset = MakeConX(0);
2062 return Type::RawPtr;
2063 } else if (base_type->base() == Type::RawPtr) {
2064 return Type::RawPtr;
2065 } else if (base_type->isa_oopptr()) {
2066 // Base is never null => always a heap address.
2067 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2068 return Type::OopPtr;
2069 }
2070 // Offset is small => always a heap address.
2071 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2072 if (offset_type != nullptr &&
2073 base_type->offset() == 0 && // (should always be?)
2074 offset_type->_lo >= 0 &&
2075 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2076 return Type::OopPtr;
2077 } else if (type == T_OBJECT) {
2078 // off heap access to an oop doesn't make any sense. Has to be on
2079 // heap.
2080 return Type::OopPtr;
2081 }
2082 // Otherwise, it might either be oop+off or null+addr.
2083 return Type::AnyPtr;
2084 } else {
2085 // No information:
2086 return Type::AnyPtr;
2087 }
2088 }
2089
2090 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2091 Node* uncasted_base = base;
2092 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2093 if (kind == Type::RawPtr) {
2094 return basic_plus_adr(top(), uncasted_base, offset);
2095 } else if (kind == Type::AnyPtr) {
2096 assert(base == uncasted_base, "unexpected base change");
2097 if (can_cast) {
2098 if (!_gvn.type(base)->speculative_maybe_null() &&
2099 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2100 // According to profiling, this access is always on
2101 // heap. Casting the base to not null and thus avoiding membars
2102 // around the access should allow better optimizations
2103 Node* null_ctl = top();
2104 base = null_check_oop(base, &null_ctl, true, true, true);
2105 assert(null_ctl->is_top(), "no null control here");
2106 return basic_plus_adr(base, offset);
2107 } else if (_gvn.type(base)->speculative_always_null() &&
2108 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2109 // According to profiling, this access is always off
2110 // heap.
2111 base = null_assert(base);
2112 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2113 offset = MakeConX(0);
2114 return basic_plus_adr(top(), raw_base, offset);
2115 }
2116 }
2117 // We don't know if it's an on heap or off heap access. Fall back
2118 // to raw memory access.
2119 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2120 return basic_plus_adr(top(), raw, offset);
2121 } else {
2122 assert(base == uncasted_base, "unexpected base change");
2123 // We know it's an on heap access so base can't be null
2124 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2125 base = must_be_not_null(base, true);
2126 }
2127 return basic_plus_adr(base, offset);
2128 }
2129 }
2130
2131 //--------------------------inline_number_methods-----------------------------
2132 // inline int Integer.numberOfLeadingZeros(int)
2133 // inline int Long.numberOfLeadingZeros(long)
2134 //
2135 // inline int Integer.numberOfTrailingZeros(int)
2136 // inline int Long.numberOfTrailingZeros(long)
2137 //
2138 // inline int Integer.bitCount(int)
2139 // inline int Long.bitCount(long)
2140 //
2141 // inline char Character.reverseBytes(char)
2142 // inline short Short.reverseBytes(short)
2143 // inline int Integer.reverseBytes(int)
2144 // inline long Long.reverseBytes(long)
2145 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2146 Node* arg = argument(0);
2147 Node* n = nullptr;
2148 switch (id) {
2149 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2150 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2151 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2152 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2153 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2154 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2155 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(nullptr, arg); break;
2156 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( nullptr, arg); break;
2157 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( nullptr, arg); break;
2158 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( nullptr, arg); break;
2159 case vmIntrinsics::_reverse_i: n = new ReverseINode(nullptr, arg); break;
2160 case vmIntrinsics::_reverse_l: n = new ReverseLNode(nullptr, arg); break;
2161 default: fatal_unexpected_iid(id); break;
2162 }
2163 set_result(_gvn.transform(n));
2164 return true;
2165 }
2166
2167 //--------------------------inline_bitshuffle_methods-----------------------------
2168 // inline int Integer.compress(int, int)
2169 // inline int Integer.expand(int, int)
2170 // inline long Long.compress(long, long)
2171 // inline long Long.expand(long, long)
2172 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2173 Node* n = nullptr;
2174 switch (id) {
2175 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2176 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2177 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2178 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2179 default: fatal_unexpected_iid(id); break;
2180 }
2181 set_result(_gvn.transform(n));
2182 return true;
2183 }
2184
2185 //--------------------------inline_number_methods-----------------------------
2186 // inline int Integer.compareUnsigned(int, int)
2187 // inline int Long.compareUnsigned(long, long)
2188 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2189 Node* arg1 = argument(0);
2190 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2191 Node* n = nullptr;
2192 switch (id) {
2193 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2194 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2195 default: fatal_unexpected_iid(id); break;
2196 }
2197 set_result(_gvn.transform(n));
2198 return true;
2199 }
2200
2201 //--------------------------inline_unsigned_divmod_methods-----------------------------
2202 // inline int Integer.divideUnsigned(int, int)
2203 // inline int Integer.remainderUnsigned(int, int)
2204 // inline long Long.divideUnsigned(long, long)
2205 // inline long Long.remainderUnsigned(long, long)
2206 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2207 Node* n = nullptr;
2208 switch (id) {
2209 case vmIntrinsics::_divideUnsigned_i: {
2210 zero_check_int(argument(1));
2211 // Compile-time detect of null-exception
2212 if (stopped()) {
2213 return true; // keep the graph constructed so far
2214 }
2215 n = new UDivINode(control(), argument(0), argument(1));
2216 break;
2217 }
2218 case vmIntrinsics::_divideUnsigned_l: {
2219 zero_check_long(argument(2));
2220 // Compile-time detect of null-exception
2221 if (stopped()) {
2222 return true; // keep the graph constructed so far
2223 }
2224 n = new UDivLNode(control(), argument(0), argument(2));
2225 break;
2226 }
2227 case vmIntrinsics::_remainderUnsigned_i: {
2228 zero_check_int(argument(1));
2229 // Compile-time detect of null-exception
2230 if (stopped()) {
2231 return true; // keep the graph constructed so far
2232 }
2233 n = new UModINode(control(), argument(0), argument(1));
2234 break;
2235 }
2236 case vmIntrinsics::_remainderUnsigned_l: {
2237 zero_check_long(argument(2));
2238 // Compile-time detect of null-exception
2239 if (stopped()) {
2240 return true; // keep the graph constructed so far
2241 }
2242 n = new UModLNode(control(), argument(0), argument(2));
2243 break;
2244 }
2245 default: fatal_unexpected_iid(id); break;
2246 }
2247 set_result(_gvn.transform(n));
2248 return true;
2249 }
2250
2251 //----------------------------inline_unsafe_access----------------------------
2252
2253 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2254 // Attempt to infer a sharper value type from the offset and base type.
2255 ciKlass* sharpened_klass = nullptr;
2256
2257 // See if it is an instance field, with an object type.
2258 if (alias_type->field() != nullptr) {
2259 if (alias_type->field()->type()->is_klass()) {
2260 sharpened_klass = alias_type->field()->type()->as_klass();
2261 }
2262 }
2263
2264 const TypeOopPtr* result = nullptr;
2265 // See if it is a narrow oop array.
2266 if (adr_type->isa_aryptr()) {
2267 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2268 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2269 if (elem_type != nullptr && elem_type->is_loaded()) {
2270 // Sharpen the value type.
2271 result = elem_type;
2272 }
2273 }
2274 }
2275
2276 // The sharpened class might be unloaded if there is no class loader
2277 // contraint in place.
2278 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2279 // Sharpen the value type.
2280 result = TypeOopPtr::make_from_klass(sharpened_klass);
2281 }
2282 if (result != nullptr) {
2283 #ifndef PRODUCT
2284 if (C->print_intrinsics() || C->print_inlining()) {
2285 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2286 tty->print(" sharpened value: "); result->dump(); tty->cr();
2287 }
2288 #endif
2289 }
2290 return result;
2291 }
2292
2293 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2294 switch (kind) {
2295 case Relaxed:
2296 return MO_UNORDERED;
2297 case Opaque:
2298 return MO_RELAXED;
2299 case Acquire:
2300 return MO_ACQUIRE;
2301 case Release:
2302 return MO_RELEASE;
2303 case Volatile:
2304 return MO_SEQ_CST;
2305 default:
2306 ShouldNotReachHere();
2307 return 0;
2308 }
2309 }
2310
2311 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2312 if (callee()->is_static()) return false; // caller must have the capability!
2313 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2314 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2315 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2316 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2317
2318 if (is_reference_type(type)) {
2319 decorators |= ON_UNKNOWN_OOP_REF;
2320 }
2321
2322 if (unaligned) {
2323 decorators |= C2_UNALIGNED;
2324 }
2325
2326 #ifndef PRODUCT
2327 {
2328 ResourceMark rm;
2329 // Check the signatures.
2330 ciSignature* sig = callee()->signature();
2331 #ifdef ASSERT
2332 if (!is_store) {
2333 // Object getReference(Object base, int/long offset), etc.
2334 BasicType rtype = sig->return_type()->basic_type();
2335 assert(rtype == type, "getter must return the expected value");
2336 assert(sig->count() == 2, "oop getter has 2 arguments");
2337 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2338 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2339 } else {
2340 // void putReference(Object base, int/long offset, Object x), etc.
2341 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2342 assert(sig->count() == 3, "oop putter has 3 arguments");
2343 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2344 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2345 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2346 assert(vtype == type, "putter must accept the expected value");
2347 }
2348 #endif // ASSERT
2349 }
2350 #endif //PRODUCT
2351
2352 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2353
2354 Node* receiver = argument(0); // type: oop
2355
2356 // Build address expression.
2357 Node* heap_base_oop = top();
2358
2359 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2360 Node* base = argument(1); // type: oop
2361 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2362 Node* offset = argument(2); // type: long
2363 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2364 // to be plain byte offsets, which are also the same as those accepted
2365 // by oopDesc::field_addr.
2366 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2367 "fieldOffset must be byte-scaled");
2368 // 32-bit machines ignore the high half!
2369 offset = ConvL2X(offset);
2370
2371 // Save state and restore on bailout
2372 uint old_sp = sp();
2373 SafePointNode* old_map = clone_map();
2374
2375 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2376 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2377
2378 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2379 if (type != T_OBJECT) {
2380 decorators |= IN_NATIVE; // off-heap primitive access
2381 } else {
2382 set_map(old_map);
2383 set_sp(old_sp);
2384 return false; // off-heap oop accesses are not supported
2385 }
2386 } else {
2387 heap_base_oop = base; // on-heap or mixed access
2388 }
2389
2390 // Can base be null? Otherwise, always on-heap access.
2391 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2392
2393 if (!can_access_non_heap) {
2394 decorators |= IN_HEAP;
2395 }
2396
2397 Node* val = is_store ? argument(4) : nullptr;
2398
2399 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2400 if (adr_type == TypePtr::NULL_PTR) {
2401 set_map(old_map);
2402 set_sp(old_sp);
2403 return false; // off-heap access with zero address
2404 }
2405
2406 // Try to categorize the address.
2407 Compile::AliasType* alias_type = C->alias_type(adr_type);
2408 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2409
2410 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2411 alias_type->adr_type() == TypeAryPtr::RANGE) {
2412 set_map(old_map);
2413 set_sp(old_sp);
2414 return false; // not supported
2415 }
2416
2417 bool mismatched = false;
2418 BasicType bt = alias_type->basic_type();
2419 if (bt != T_ILLEGAL) {
2420 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2421 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2422 // Alias type doesn't differentiate between byte[] and boolean[]).
2423 // Use address type to get the element type.
2424 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2425 }
2426 if (is_reference_type(bt, true)) {
2427 // accessing an array field with getReference is not a mismatch
2428 bt = T_OBJECT;
2429 }
2430 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2431 // Don't intrinsify mismatched object accesses
2432 set_map(old_map);
2433 set_sp(old_sp);
2434 return false;
2435 }
2436 mismatched = (bt != type);
2437 } else if (alias_type->adr_type()->isa_oopptr()) {
2438 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2439 }
2440
2441 destruct_map_clone(old_map);
2442 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2443
2444 if (mismatched) {
2445 decorators |= C2_MISMATCHED;
2446 }
2447
2448 // First guess at the value type.
2449 const Type *value_type = Type::get_const_basic_type(type);
2450
2451 // Figure out the memory ordering.
2452 decorators |= mo_decorator_for_access_kind(kind);
2453
2454 if (!is_store && type == T_OBJECT) {
2455 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2456 if (tjp != nullptr) {
2457 value_type = tjp;
2458 }
2459 }
2460
2461 receiver = null_check(receiver);
2462 if (stopped()) {
2463 return true;
2464 }
2465 // Heap pointers get a null-check from the interpreter,
2466 // as a courtesy. However, this is not guaranteed by Unsafe,
2467 // and it is not possible to fully distinguish unintended nulls
2468 // from intended ones in this API.
2469
2470 if (!is_store) {
2471 Node* p = nullptr;
2472 // Try to constant fold a load from a constant field
2473 ciField* field = alias_type->field();
2474 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) {
2475 // final or stable field
2476 p = make_constant_from_field(field, heap_base_oop);
2477 }
2478
2479 if (p == nullptr) { // Could not constant fold the load
2480 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2481 // Normalize the value returned by getBoolean in the following cases
2482 if (type == T_BOOLEAN &&
2483 (mismatched ||
2484 heap_base_oop == top() || // - heap_base_oop is null or
2485 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2486 // and the unsafe access is made to large offset
2487 // (i.e., larger than the maximum offset necessary for any
2488 // field access)
2489 ) {
2490 IdealKit ideal = IdealKit(this);
2491 #define __ ideal.
2492 IdealVariable normalized_result(ideal);
2493 __ declarations_done();
2494 __ set(normalized_result, p);
2495 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2496 __ set(normalized_result, ideal.ConI(1));
2497 ideal.end_if();
2498 final_sync(ideal);
2499 p = __ value(normalized_result);
2500 #undef __
2501 }
2502 }
2503 if (type == T_ADDRESS) {
2504 p = gvn().transform(new CastP2XNode(nullptr, p));
2505 p = ConvX2UL(p);
2506 }
2507 // The load node has the control of the preceding MemBarCPUOrder. All
2508 // following nodes will have the control of the MemBarCPUOrder inserted at
2509 // the end of this method. So, pushing the load onto the stack at a later
2510 // point is fine.
2511 set_result(p);
2512 } else {
2513 if (bt == T_ADDRESS) {
2514 // Repackage the long as a pointer.
2515 val = ConvL2X(val);
2516 val = gvn().transform(new CastX2PNode(val));
2517 }
2518 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2519 }
2520
2521 return true;
2522 }
2523
2524 //----------------------------inline_unsafe_load_store----------------------------
2525 // This method serves a couple of different customers (depending on LoadStoreKind):
2526 //
2527 // LS_cmp_swap:
2528 //
2529 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2530 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2531 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2532 //
2533 // LS_cmp_swap_weak:
2534 //
2535 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2536 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2537 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2538 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2539 //
2540 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2541 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2542 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2543 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2544 //
2545 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2546 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2547 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2548 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2549 //
2550 // LS_cmp_exchange:
2551 //
2552 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2553 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2554 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2555 //
2556 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2557 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2558 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2559 //
2560 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2561 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2562 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2563 //
2564 // LS_get_add:
2565 //
2566 // int getAndAddInt( Object o, long offset, int delta)
2567 // long getAndAddLong(Object o, long offset, long delta)
2568 //
2569 // LS_get_set:
2570 //
2571 // int getAndSet(Object o, long offset, int newValue)
2572 // long getAndSet(Object o, long offset, long newValue)
2573 // Object getAndSet(Object o, long offset, Object newValue)
2574 //
2575 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2576 // This basic scheme here is the same as inline_unsafe_access, but
2577 // differs in enough details that combining them would make the code
2578 // overly confusing. (This is a true fact! I originally combined
2579 // them, but even I was confused by it!) As much code/comments as
2580 // possible are retained from inline_unsafe_access though to make
2581 // the correspondences clearer. - dl
2582
2583 if (callee()->is_static()) return false; // caller must have the capability!
2584
2585 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2586 decorators |= mo_decorator_for_access_kind(access_kind);
2587
2588 #ifndef PRODUCT
2589 BasicType rtype;
2590 {
2591 ResourceMark rm;
2592 // Check the signatures.
2593 ciSignature* sig = callee()->signature();
2594 rtype = sig->return_type()->basic_type();
2595 switch(kind) {
2596 case LS_get_add:
2597 case LS_get_set: {
2598 // Check the signatures.
2599 #ifdef ASSERT
2600 assert(rtype == type, "get and set must return the expected type");
2601 assert(sig->count() == 3, "get and set has 3 arguments");
2602 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2603 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2604 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2605 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2606 #endif // ASSERT
2607 break;
2608 }
2609 case LS_cmp_swap:
2610 case LS_cmp_swap_weak: {
2611 // Check the signatures.
2612 #ifdef ASSERT
2613 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2614 assert(sig->count() == 4, "CAS has 4 arguments");
2615 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2616 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2617 #endif // ASSERT
2618 break;
2619 }
2620 case LS_cmp_exchange: {
2621 // Check the signatures.
2622 #ifdef ASSERT
2623 assert(rtype == type, "CAS must return the expected type");
2624 assert(sig->count() == 4, "CAS has 4 arguments");
2625 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2626 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2627 #endif // ASSERT
2628 break;
2629 }
2630 default:
2631 ShouldNotReachHere();
2632 }
2633 }
2634 #endif //PRODUCT
2635
2636 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2637
2638 // Get arguments:
2639 Node* receiver = nullptr;
2640 Node* base = nullptr;
2641 Node* offset = nullptr;
2642 Node* oldval = nullptr;
2643 Node* newval = nullptr;
2644 switch(kind) {
2645 case LS_cmp_swap:
2646 case LS_cmp_swap_weak:
2647 case LS_cmp_exchange: {
2648 const bool two_slot_type = type2size[type] == 2;
2649 receiver = argument(0); // type: oop
2650 base = argument(1); // type: oop
2651 offset = argument(2); // type: long
2652 oldval = argument(4); // type: oop, int, or long
2653 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2654 break;
2655 }
2656 case LS_get_add:
2657 case LS_get_set: {
2658 receiver = argument(0); // type: oop
2659 base = argument(1); // type: oop
2660 offset = argument(2); // type: long
2661 oldval = nullptr;
2662 newval = argument(4); // type: oop, int, or long
2663 break;
2664 }
2665 default:
2666 ShouldNotReachHere();
2667 }
2668
2669 // Build field offset expression.
2670 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2671 // to be plain byte offsets, which are also the same as those accepted
2672 // by oopDesc::field_addr.
2673 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2674 // 32-bit machines ignore the high half of long offsets
2675 offset = ConvL2X(offset);
2676 // Save state and restore on bailout
2677 uint old_sp = sp();
2678 SafePointNode* old_map = clone_map();
2679 Node* adr = make_unsafe_address(base, offset,type, false);
2680 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2681
2682 Compile::AliasType* alias_type = C->alias_type(adr_type);
2683 BasicType bt = alias_type->basic_type();
2684 if (bt != T_ILLEGAL &&
2685 (is_reference_type(bt) != (type == T_OBJECT))) {
2686 // Don't intrinsify mismatched object accesses.
2687 set_map(old_map);
2688 set_sp(old_sp);
2689 return false;
2690 }
2691
2692 destruct_map_clone(old_map);
2693
2694 // For CAS, unlike inline_unsafe_access, there seems no point in
2695 // trying to refine types. Just use the coarse types here.
2696 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2697 const Type *value_type = Type::get_const_basic_type(type);
2698
2699 switch (kind) {
2700 case LS_get_set:
2701 case LS_cmp_exchange: {
2702 if (type == T_OBJECT) {
2703 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2704 if (tjp != nullptr) {
2705 value_type = tjp;
2706 }
2707 }
2708 break;
2709 }
2710 case LS_cmp_swap:
2711 case LS_cmp_swap_weak:
2712 case LS_get_add:
2713 break;
2714 default:
2715 ShouldNotReachHere();
2716 }
2717
2718 // Null check receiver.
2719 receiver = null_check(receiver);
2720 if (stopped()) {
2721 return true;
2722 }
2723
2724 int alias_idx = C->get_alias_index(adr_type);
2725
2726 if (is_reference_type(type)) {
2727 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2728
2729 // Transformation of a value which could be null pointer (CastPP #null)
2730 // could be delayed during Parse (for example, in adjust_map_after_if()).
2731 // Execute transformation here to avoid barrier generation in such case.
2732 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2733 newval = _gvn.makecon(TypePtr::NULL_PTR);
2734
2735 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2736 // Refine the value to a null constant, when it is known to be null
2737 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2738 }
2739 }
2740
2741 Node* result = nullptr;
2742 switch (kind) {
2743 case LS_cmp_exchange: {
2744 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2745 oldval, newval, value_type, type, decorators);
2746 break;
2747 }
2748 case LS_cmp_swap_weak:
2749 decorators |= C2_WEAK_CMPXCHG;
2750 case LS_cmp_swap: {
2751 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2752 oldval, newval, value_type, type, decorators);
2753 break;
2754 }
2755 case LS_get_set: {
2756 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2757 newval, value_type, type, decorators);
2758 break;
2759 }
2760 case LS_get_add: {
2761 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2762 newval, value_type, type, decorators);
2763 break;
2764 }
2765 default:
2766 ShouldNotReachHere();
2767 }
2768
2769 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2770 set_result(result);
2771 return true;
2772 }
2773
2774 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2775 // Regardless of form, don't allow previous ld/st to move down,
2776 // then issue acquire, release, or volatile mem_bar.
2777 insert_mem_bar(Op_MemBarCPUOrder);
2778 switch(id) {
2779 case vmIntrinsics::_loadFence:
2780 insert_mem_bar(Op_LoadFence);
2781 return true;
2782 case vmIntrinsics::_storeFence:
2783 insert_mem_bar(Op_StoreFence);
2784 return true;
2785 case vmIntrinsics::_storeStoreFence:
2786 insert_mem_bar(Op_StoreStoreFence);
2787 return true;
2788 case vmIntrinsics::_fullFence:
2789 insert_mem_bar(Op_MemBarVolatile);
2790 return true;
2791 default:
2792 fatal_unexpected_iid(id);
2793 return false;
2794 }
2795 }
2796
2797 bool LibraryCallKit::inline_onspinwait() {
2798 insert_mem_bar(Op_OnSpinWait);
2799 return true;
2800 }
2801
2802 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2803 if (!kls->is_Con()) {
2804 return true;
2805 }
2806 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2807 if (klsptr == nullptr) {
2808 return true;
2809 }
2810 ciInstanceKlass* ik = klsptr->instance_klass();
2811 // don't need a guard for a klass that is already initialized
2812 return !ik->is_initialized();
2813 }
2814
2815 //----------------------------inline_unsafe_writeback0-------------------------
2816 // public native void Unsafe.writeback0(long address)
2817 bool LibraryCallKit::inline_unsafe_writeback0() {
2818 if (!Matcher::has_match_rule(Op_CacheWB)) {
2819 return false;
2820 }
2821 #ifndef PRODUCT
2822 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2823 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2824 ciSignature* sig = callee()->signature();
2825 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2826 #endif
2827 null_check_receiver(); // null-check, then ignore
2828 Node *addr = argument(1);
2829 addr = new CastX2PNode(addr);
2830 addr = _gvn.transform(addr);
2831 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2832 flush = _gvn.transform(flush);
2833 set_memory(flush, TypeRawPtr::BOTTOM);
2834 return true;
2835 }
2836
2837 //----------------------------inline_unsafe_writeback0-------------------------
2838 // public native void Unsafe.writeback0(long address)
2839 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2840 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2841 return false;
2842 }
2843 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2844 return false;
2845 }
2846 #ifndef PRODUCT
2847 assert(Matcher::has_match_rule(Op_CacheWB),
2848 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2849 : "found match rule for CacheWBPostSync but not CacheWB"));
2850
2851 #endif
2852 null_check_receiver(); // null-check, then ignore
2853 Node *sync;
2854 if (is_pre) {
2855 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2856 } else {
2857 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2858 }
2859 sync = _gvn.transform(sync);
2860 set_memory(sync, TypeRawPtr::BOTTOM);
2861 return true;
2862 }
2863
2864 //----------------------------inline_unsafe_allocate---------------------------
2865 // public native Object Unsafe.allocateInstance(Class<?> cls);
2866 bool LibraryCallKit::inline_unsafe_allocate() {
2867
2868 #if INCLUDE_JVMTI
2869 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
2870 return false;
2871 }
2872 #endif //INCLUDE_JVMTI
2873
2874 if (callee()->is_static()) return false; // caller must have the capability!
2875
2876 null_check_receiver(); // null-check, then ignore
2877 Node* cls = null_check(argument(1));
2878 if (stopped()) return true;
2879
2880 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
2881 kls = null_check(kls);
2882 if (stopped()) return true; // argument was like int.class
2883
2884 #if INCLUDE_JVMTI
2885 // Don't try to access new allocated obj in the intrinsic.
2886 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
2887 // Deoptimize and allocate in interpreter instead.
2888 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
2889 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
2890 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
2891 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
2892 {
2893 BuildCutout unless(this, tst, PROB_MAX);
2894 uncommon_trap(Deoptimization::Reason_intrinsic,
2895 Deoptimization::Action_make_not_entrant);
2896 }
2897 if (stopped()) {
2898 return true;
2899 }
2900 #endif //INCLUDE_JVMTI
2901
2902 Node* test = nullptr;
2903 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2904 // Note: The argument might still be an illegal value like
2905 // Serializable.class or Object[].class. The runtime will handle it.
2906 // But we must make an explicit check for initialization.
2907 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2908 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2909 // can generate code to load it as unsigned byte.
2910 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
2911 Node* bits = intcon(InstanceKlass::fully_initialized);
2912 test = _gvn.transform(new SubINode(inst, bits));
2913 // The 'test' is non-zero if we need to take a slow path.
2914 }
2915
2916 Node* obj = new_instance(kls, test);
2917 set_result(obj);
2918 return true;
2919 }
2920
2921 //------------------------inline_native_time_funcs--------------
2922 // inline code for System.currentTimeMillis() and System.nanoTime()
2923 // these have the same type and signature
2924 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2925 const TypeFunc* tf = OptoRuntime::void_long_Type();
2926 const TypePtr* no_memory_effects = nullptr;
2927 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2928 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2929 #ifdef ASSERT
2930 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2931 assert(value_top == top(), "second value must be top");
2932 #endif
2933 set_result(value);
2934 return true;
2935 }
2936
2937
2938 #if INCLUDE_JVMTI
2939
2940 // When notifications are disabled then just update the VTMS transition bit and return.
2941 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
2942 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
2943 if (!DoJVMTIVirtualThreadTransitions) {
2944 return true;
2945 }
2946 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
2947 IdealKit ideal(this);
2948
2949 Node* ONE = ideal.ConI(1);
2950 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
2951 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
2952 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
2953
2954 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
2955 sync_kit(ideal);
2956 // if notifyJvmti enabled then make a call to the given SharedRuntime function
2957 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
2958 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
2959 ideal.sync_kit(this);
2960 } ideal.else_(); {
2961 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
2962 Node* thread = ideal.thread();
2963 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
2964 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
2965
2966 sync_kit(ideal);
2967 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2968 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2969
2970 ideal.sync_kit(this);
2971 } ideal.end_if();
2972 final_sync(ideal);
2973
2974 return true;
2975 }
2976
2977 // Always update the is_disable_suspend bit.
2978 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
2979 if (!DoJVMTIVirtualThreadTransitions) {
2980 return true;
2981 }
2982 IdealKit ideal(this);
2983
2984 {
2985 // unconditionally update the is_disable_suspend bit in current JavaThread
2986 Node* thread = ideal.thread();
2987 Node* arg = _gvn.transform(argument(0)); // argument for notification
2988 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
2989 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
2990
2991 sync_kit(ideal);
2992 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2993 ideal.sync_kit(this);
2994 }
2995 final_sync(ideal);
2996
2997 return true;
2998 }
2999
3000 #endif // INCLUDE_JVMTI
3001
3002 #ifdef JFR_HAVE_INTRINSICS
3003
3004 /**
3005 * if oop->klass != null
3006 * // normal class
3007 * epoch = _epoch_state ? 2 : 1
3008 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3009 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3010 * }
3011 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3012 * else
3013 * // primitive class
3014 * if oop->array_klass != null
3015 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3016 * else
3017 * id = LAST_TYPE_ID + 1 // void class path
3018 * if (!signaled)
3019 * signaled = true
3020 */
3021 bool LibraryCallKit::inline_native_classID() {
3022 Node* cls = argument(0);
3023
3024 IdealKit ideal(this);
3025 #define __ ideal.
3026 IdealVariable result(ideal); __ declarations_done();
3027 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3028 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3029 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3030
3031
3032 __ if_then(kls, BoolTest::ne, null()); {
3033 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3034 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3035
3036 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3037 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3038 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3039 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3040 mask = _gvn.transform(new OrLNode(mask, epoch));
3041 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3042
3043 float unlikely = PROB_UNLIKELY(0.999);
3044 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3045 sync_kit(ideal);
3046 make_runtime_call(RC_LEAF,
3047 OptoRuntime::class_id_load_barrier_Type(),
3048 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3049 "class id load barrier",
3050 TypePtr::BOTTOM,
3051 kls);
3052 ideal.sync_kit(this);
3053 } __ end_if();
3054
3055 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3056 } __ else_(); {
3057 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3058 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3059 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3060 __ if_then(array_kls, BoolTest::ne, null()); {
3061 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3062 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3063 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3064 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3065 } __ else_(); {
3066 // void class case
3067 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3068 } __ end_if();
3069
3070 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3071 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3072 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3073 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3074 } __ end_if();
3075 } __ end_if();
3076
3077 final_sync(ideal);
3078 set_result(ideal.value(result));
3079 #undef __
3080 return true;
3081 }
3082
3083 //------------------------inline_native_jvm_commit------------------
3084 bool LibraryCallKit::inline_native_jvm_commit() {
3085 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3086
3087 // Save input memory and i_o state.
3088 Node* input_memory_state = reset_memory();
3089 set_all_memory(input_memory_state);
3090 Node* input_io_state = i_o();
3091
3092 // TLS.
3093 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3094 // Jfr java buffer.
3095 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3096 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3097 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3098
3099 // Load the current value of the notified field in the JfrThreadLocal.
3100 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3101 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3102
3103 // Test for notification.
3104 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3105 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3106 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3107
3108 // True branch, is notified.
3109 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3110 set_control(is_notified);
3111
3112 // Reset notified state.
3113 Node* notified_reset_memory = store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered);
3114
3115 // 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.
3116 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3117 // Convert the machine-word to a long.
3118 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3119
3120 // False branch, not notified.
3121 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3122 set_control(not_notified);
3123 set_all_memory(input_memory_state);
3124
3125 // Arg is the next position as a long.
3126 Node* arg = argument(0);
3127 // Convert long to machine-word.
3128 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3129
3130 // Store the next_position to the underlying jfr java buffer.
3131 Node* commit_memory;
3132 #ifdef _LP64
3133 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_LONG, Compile::AliasIdxRaw, MemNode::release);
3134 #else
3135 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_INT, Compile::AliasIdxRaw, MemNode::release);
3136 #endif
3137
3138 // 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.
3139 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3140 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3141 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3142
3143 // And flags with lease constant.
3144 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3145
3146 // Branch on lease to conditionalize returning the leased java buffer.
3147 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3148 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3149 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3150
3151 // False branch, not a lease.
3152 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3153
3154 // True branch, is lease.
3155 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3156 set_control(is_lease);
3157
3158 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3159 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3160 OptoRuntime::void_void_Type(),
3161 SharedRuntime::jfr_return_lease(),
3162 "return_lease", TypePtr::BOTTOM);
3163 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3164
3165 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3166 record_for_igvn(lease_compare_rgn);
3167 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3168 record_for_igvn(lease_compare_mem);
3169 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3170 record_for_igvn(lease_compare_io);
3171 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3172 record_for_igvn(lease_result_value);
3173
3174 // Update control and phi nodes.
3175 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3176 lease_compare_rgn->init_req(_false_path, not_lease);
3177
3178 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3179 lease_compare_mem->init_req(_false_path, commit_memory);
3180
3181 lease_compare_io->init_req(_true_path, i_o());
3182 lease_compare_io->init_req(_false_path, input_io_state);
3183
3184 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3185 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3186
3187 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3188 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3189 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3190 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3191
3192 // Update control and phi nodes.
3193 result_rgn->init_req(_true_path, is_notified);
3194 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3195
3196 result_mem->init_req(_true_path, notified_reset_memory);
3197 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3198
3199 result_io->init_req(_true_path, input_io_state);
3200 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3201
3202 result_value->init_req(_true_path, current_pos);
3203 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3204
3205 // Set output state.
3206 set_control(_gvn.transform(result_rgn));
3207 set_all_memory(_gvn.transform(result_mem));
3208 set_i_o(_gvn.transform(result_io));
3209 set_result(result_rgn, result_value);
3210 return true;
3211 }
3212
3213 /*
3214 * The intrinsic is a model of this pseudo-code:
3215 *
3216 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3217 * jobject h_event_writer = tl->java_event_writer();
3218 * if (h_event_writer == nullptr) {
3219 * return nullptr;
3220 * }
3221 * oop threadObj = Thread::threadObj();
3222 * oop vthread = java_lang_Thread::vthread(threadObj);
3223 * traceid tid;
3224 * bool pinVirtualThread;
3225 * bool excluded;
3226 * if (vthread != threadObj) { // i.e. current thread is virtual
3227 * tid = java_lang_Thread::tid(vthread);
3228 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3229 * pinVirtualThread = VMContinuations;
3230 * excluded = vthread_epoch_raw & excluded_mask;
3231 * if (!excluded) {
3232 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3233 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3234 * if (vthread_epoch != current_epoch) {
3235 * write_checkpoint();
3236 * }
3237 * }
3238 * } else {
3239 * tid = java_lang_Thread::tid(threadObj);
3240 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3241 * pinVirtualThread = false;
3242 * excluded = thread_epoch_raw & excluded_mask;
3243 * }
3244 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3245 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3246 * if (tid_in_event_writer != tid) {
3247 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3248 * setField(event_writer, "excluded", excluded);
3249 * setField(event_writer, "threadID", tid);
3250 * }
3251 * return event_writer
3252 */
3253 bool LibraryCallKit::inline_native_getEventWriter() {
3254 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3255
3256 // Save input memory and i_o state.
3257 Node* input_memory_state = reset_memory();
3258 set_all_memory(input_memory_state);
3259 Node* input_io_state = i_o();
3260
3261 Node* excluded_mask = _gvn.intcon(32768);
3262 Node* epoch_mask = _gvn.intcon(32767);
3263
3264 // TLS
3265 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3266
3267 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3268 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3269
3270 // Load the eventwriter jobject handle.
3271 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3272
3273 // Null check the jobject handle.
3274 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3275 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3276 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3277
3278 // False path, jobj is null.
3279 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3280
3281 // True path, jobj is not null.
3282 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3283
3284 set_control(jobj_is_not_null);
3285
3286 // Load the threadObj for the CarrierThread.
3287 Node* threadObj = generate_current_thread(tls_ptr);
3288
3289 // Load the vthread.
3290 Node* vthread = generate_virtual_thread(tls_ptr);
3291
3292 // If vthread != threadObj, this is a virtual thread.
3293 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3294 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3295 IfNode* iff_vthread_not_equal_threadObj =
3296 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3297
3298 // False branch, fallback to threadObj.
3299 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3300 set_control(vthread_equal_threadObj);
3301
3302 // Load the tid field from the vthread object.
3303 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3304
3305 // Load the raw epoch value from the threadObj.
3306 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3307 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3308 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3309 TypeInt::CHAR, T_CHAR,
3310 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3311
3312 // Mask off the excluded information from the epoch.
3313 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3314
3315 // True branch, this is a virtual thread.
3316 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3317 set_control(vthread_not_equal_threadObj);
3318
3319 // Load the tid field from the vthread object.
3320 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3321
3322 // Continuation support determines if a virtual thread should be pinned.
3323 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3324 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3325
3326 // Load the raw epoch value from the vthread.
3327 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3328 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3329 TypeInt::CHAR, T_CHAR,
3330 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3331
3332 // Mask off the excluded information from the epoch.
3333 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3334
3335 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3336 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3337 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3338 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3339
3340 // False branch, vthread is excluded, no need to write epoch info.
3341 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3342
3343 // True branch, vthread is included, update epoch info.
3344 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3345 set_control(included);
3346
3347 // Get epoch value.
3348 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3349
3350 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3351 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3352 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3353
3354 // Compare the epoch in the vthread to the current epoch generation.
3355 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3356 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3357 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3358
3359 // False path, epoch is equal, checkpoint information is valid.
3360 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3361
3362 // True path, epoch is not equal, write a checkpoint for the vthread.
3363 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3364
3365 set_control(epoch_is_not_equal);
3366
3367 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3368 // The call also updates the native thread local thread id and the vthread with the current epoch.
3369 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3370 OptoRuntime::jfr_write_checkpoint_Type(),
3371 SharedRuntime::jfr_write_checkpoint(),
3372 "write_checkpoint", TypePtr::BOTTOM);
3373 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3374
3375 // vthread epoch != current epoch
3376 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3377 record_for_igvn(epoch_compare_rgn);
3378 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3379 record_for_igvn(epoch_compare_mem);
3380 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3381 record_for_igvn(epoch_compare_io);
3382
3383 // Update control and phi nodes.
3384 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3385 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3386 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3387 epoch_compare_mem->init_req(_false_path, input_memory_state);
3388 epoch_compare_io->init_req(_true_path, i_o());
3389 epoch_compare_io->init_req(_false_path, input_io_state);
3390
3391 // excluded != true
3392 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3393 record_for_igvn(exclude_compare_rgn);
3394 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3395 record_for_igvn(exclude_compare_mem);
3396 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3397 record_for_igvn(exclude_compare_io);
3398
3399 // Update control and phi nodes.
3400 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3401 exclude_compare_rgn->init_req(_false_path, excluded);
3402 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3403 exclude_compare_mem->init_req(_false_path, input_memory_state);
3404 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3405 exclude_compare_io->init_req(_false_path, input_io_state);
3406
3407 // vthread != threadObj
3408 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3409 record_for_igvn(vthread_compare_rgn);
3410 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3411 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3412 record_for_igvn(vthread_compare_io);
3413 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3414 record_for_igvn(tid);
3415 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3416 record_for_igvn(exclusion);
3417 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3418 record_for_igvn(pinVirtualThread);
3419
3420 // Update control and phi nodes.
3421 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3422 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3423 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3424 vthread_compare_mem->init_req(_false_path, input_memory_state);
3425 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3426 vthread_compare_io->init_req(_false_path, input_io_state);
3427 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3428 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3429 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3430 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3431 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3432 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3433
3434 // Update branch state.
3435 set_control(_gvn.transform(vthread_compare_rgn));
3436 set_all_memory(_gvn.transform(vthread_compare_mem));
3437 set_i_o(_gvn.transform(vthread_compare_io));
3438
3439 // Load the event writer oop by dereferencing the jobject handle.
3440 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3441 assert(klass_EventWriter->is_loaded(), "invariant");
3442 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3443 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3444 const TypeOopPtr* const xtype = aklass->as_instance_type();
3445 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3446 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3447
3448 // Load the current thread id from the event writer object.
3449 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3450 // Get the field offset to, conditionally, store an updated tid value later.
3451 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3452 const TypePtr* event_writer_tid_field_type = _gvn.type(event_writer_tid_field)->isa_ptr();
3453 // Get the field offset to, conditionally, store an updated exclusion value later.
3454 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3455 const TypePtr* event_writer_excluded_field_type = _gvn.type(event_writer_excluded_field)->isa_ptr();
3456 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3457 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3458 const TypePtr* event_writer_pin_field_type = _gvn.type(event_writer_pin_field)->isa_ptr();
3459
3460 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3461 record_for_igvn(event_writer_tid_compare_rgn);
3462 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3463 record_for_igvn(event_writer_tid_compare_mem);
3464 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3465 record_for_igvn(event_writer_tid_compare_io);
3466
3467 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3468 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3469 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3470 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3471
3472 // False path, tids are the same.
3473 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3474
3475 // True path, tid is not equal, need to update the tid in the event writer.
3476 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3477 record_for_igvn(tid_is_not_equal);
3478
3479 // Store the pin state to the event writer.
3480 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, event_writer_pin_field_type, MemNode::unordered);
3481
3482 // Store the exclusion state to the event writer.
3483 store_to_memory(tid_is_not_equal, event_writer_excluded_field, _gvn.transform(exclusion), T_BOOLEAN, event_writer_excluded_field_type, MemNode::unordered);
3484
3485 // Store the tid to the event writer.
3486 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, event_writer_tid_field_type, MemNode::unordered);
3487
3488 // Update control and phi nodes.
3489 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3490 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3491 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3492 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3493 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3494 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3495
3496 // Result of top level CFG, Memory, IO and Value.
3497 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3498 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3499 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3500 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3501
3502 // Result control.
3503 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3504 result_rgn->init_req(_false_path, jobj_is_null);
3505
3506 // Result memory.
3507 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3508 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3509
3510 // Result IO.
3511 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3512 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3513
3514 // Result value.
3515 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3516 result_value->init_req(_false_path, null()); // return null
3517
3518 // Set output state.
3519 set_control(_gvn.transform(result_rgn));
3520 set_all_memory(_gvn.transform(result_mem));
3521 set_i_o(_gvn.transform(result_io));
3522 set_result(result_rgn, result_value);
3523 return true;
3524 }
3525
3526 /*
3527 * The intrinsic is a model of this pseudo-code:
3528 *
3529 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3530 * if (carrierThread != thread) { // is virtual thread
3531 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3532 * bool excluded = vthread_epoch_raw & excluded_mask;
3533 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3534 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3535 * if (!excluded) {
3536 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3537 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3538 * }
3539 * Atomic::release_store(&tl->_vthread, true);
3540 * return;
3541 * }
3542 * Atomic::release_store(&tl->_vthread, false);
3543 */
3544 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3545 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3546
3547 Node* input_memory_state = reset_memory();
3548 set_all_memory(input_memory_state);
3549
3550 Node* excluded_mask = _gvn.intcon(32768);
3551 Node* epoch_mask = _gvn.intcon(32767);
3552
3553 Node* const carrierThread = generate_current_thread(jt);
3554 // If thread != carrierThread, this is a virtual thread.
3555 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3556 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3557 IfNode* iff_thread_not_equal_carrierThread =
3558 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3559
3560 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3561
3562 // False branch, is carrierThread.
3563 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3564 // Store release
3565 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3566
3567 set_all_memory(input_memory_state);
3568
3569 // True branch, is virtual thread.
3570 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3571 set_control(thread_not_equal_carrierThread);
3572
3573 // Load the raw epoch value from the vthread.
3574 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3575 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
3576 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3577
3578 // Mask off the excluded information from the epoch.
3579 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3580
3581 // Load the tid field from the thread.
3582 Node* tid = load_field_from_object(thread, "tid", "J");
3583
3584 // Store the vthread tid to the jfr thread local.
3585 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3586 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, Compile::AliasIdxRaw, MemNode::unordered, true);
3587
3588 // Branch is_excluded to conditionalize updating the epoch .
3589 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3590 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3591 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3592
3593 // True branch, vthread is excluded, no need to write epoch info.
3594 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3595 set_control(excluded);
3596 Node* vthread_is_excluded = _gvn.intcon(1);
3597
3598 // False branch, vthread is included, update epoch info.
3599 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3600 set_control(included);
3601 Node* vthread_is_included = _gvn.intcon(0);
3602
3603 // Get epoch value.
3604 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3605
3606 // Store the vthread epoch to the jfr thread local.
3607 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3608 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, Compile::AliasIdxRaw, MemNode::unordered, true);
3609
3610 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3611 record_for_igvn(excluded_rgn);
3612 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3613 record_for_igvn(excluded_mem);
3614 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3615 record_for_igvn(exclusion);
3616
3617 // Merge the excluded control and memory.
3618 excluded_rgn->init_req(_true_path, excluded);
3619 excluded_rgn->init_req(_false_path, included);
3620 excluded_mem->init_req(_true_path, tid_memory);
3621 excluded_mem->init_req(_false_path, included_memory);
3622 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3623 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3624
3625 // Set intermediate state.
3626 set_control(_gvn.transform(excluded_rgn));
3627 set_all_memory(excluded_mem);
3628
3629 // Store the vthread exclusion state to the jfr thread local.
3630 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3631 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered, true);
3632
3633 // Store release
3634 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3635
3636 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3637 record_for_igvn(thread_compare_rgn);
3638 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3639 record_for_igvn(thread_compare_mem);
3640 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3641 record_for_igvn(vthread);
3642
3643 // Merge the thread_compare control and memory.
3644 thread_compare_rgn->init_req(_true_path, control());
3645 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3646 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3647 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3648
3649 // Set output state.
3650 set_control(_gvn.transform(thread_compare_rgn));
3651 set_all_memory(_gvn.transform(thread_compare_mem));
3652 }
3653
3654 #endif // JFR_HAVE_INTRINSICS
3655
3656 //------------------------inline_native_currentCarrierThread------------------
3657 bool LibraryCallKit::inline_native_currentCarrierThread() {
3658 Node* junk = nullptr;
3659 set_result(generate_current_thread(junk));
3660 return true;
3661 }
3662
3663 //------------------------inline_native_currentThread------------------
3664 bool LibraryCallKit::inline_native_currentThread() {
3665 Node* junk = nullptr;
3666 set_result(generate_virtual_thread(junk));
3667 return true;
3668 }
3669
3670 //------------------------inline_native_setVthread------------------
3671 bool LibraryCallKit::inline_native_setCurrentThread() {
3672 assert(C->method()->changes_current_thread(),
3673 "method changes current Thread but is not annotated ChangesCurrentThread");
3674 Node* arr = argument(1);
3675 Node* thread = _gvn.transform(new ThreadLocalNode());
3676 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3677 Node* thread_obj_handle
3678 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3679 thread_obj_handle = _gvn.transform(thread_obj_handle);
3680 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3681 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3682 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3683 return true;
3684 }
3685
3686 const Type* LibraryCallKit::scopedValueCache_type() {
3687 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3688 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3689 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3690
3691 // Because we create the scopedValue cache lazily we have to make the
3692 // type of the result BotPTR.
3693 bool xk = etype->klass_is_exact();
3694 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3695 return objects_type;
3696 }
3697
3698 Node* LibraryCallKit::scopedValueCache_helper() {
3699 Node* thread = _gvn.transform(new ThreadLocalNode());
3700 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3701 // We cannot use immutable_memory() because we might flip onto a
3702 // different carrier thread, at which point we'll need to use that
3703 // carrier thread's cache.
3704 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3705 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3706 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3707 }
3708
3709 //------------------------inline_native_scopedValueCache------------------
3710 bool LibraryCallKit::inline_native_scopedValueCache() {
3711 Node* cache_obj_handle = scopedValueCache_helper();
3712 const Type* objects_type = scopedValueCache_type();
3713 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3714
3715 return true;
3716 }
3717
3718 //------------------------inline_native_setScopedValueCache------------------
3719 bool LibraryCallKit::inline_native_setScopedValueCache() {
3720 Node* arr = argument(0);
3721 Node* cache_obj_handle = scopedValueCache_helper();
3722 const Type* objects_type = scopedValueCache_type();
3723
3724 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3725 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3726
3727 return true;
3728 }
3729
3730 //------------------------inline_native_Continuation_pin and unpin-----------
3731
3732 // Shared implementation routine for both pin and unpin.
3733 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
3734 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3735
3736 // Save input memory.
3737 Node* input_memory_state = reset_memory();
3738 set_all_memory(input_memory_state);
3739
3740 // TLS
3741 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3742 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
3743 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
3744
3745 // Null check the last continuation object.
3746 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
3747 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
3748 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3749
3750 // False path, last continuation is null.
3751 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
3752
3753 // True path, last continuation is not null.
3754 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
3755
3756 set_control(continuation_is_not_null);
3757
3758 // Load the pin count from the last continuation.
3759 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
3760 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
3761
3762 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
3763 Node* pin_count_rhs;
3764 if (unpin) {
3765 pin_count_rhs = _gvn.intcon(0);
3766 } else {
3767 pin_count_rhs = _gvn.intcon(UINT32_MAX);
3768 }
3769 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
3770 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
3771 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
3772
3773 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
3774 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
3775 set_control(valid_pin_count);
3776
3777 Node* next_pin_count;
3778 if (unpin) {
3779 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
3780 } else {
3781 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
3782 }
3783
3784 Node* updated_pin_count_memory = store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, Compile::AliasIdxRaw, MemNode::unordered);
3785
3786 // True branch, pin count over/underflow.
3787 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
3788 {
3789 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
3790 // which will throw IllegalStateException for pin count over/underflow.
3791 PreserveJVMState pjvms(this);
3792 set_control(pin_count_over_underflow);
3793 set_all_memory(input_memory_state);
3794 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
3795 Deoptimization::Action_none);
3796 assert(stopped(), "invariant");
3797 }
3798
3799 // Result of top level CFG and Memory.
3800 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3801 record_for_igvn(result_rgn);
3802 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3803 record_for_igvn(result_mem);
3804
3805 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
3806 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
3807 result_mem->init_req(_true_path, _gvn.transform(updated_pin_count_memory));
3808 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3809
3810 // Set output state.
3811 set_control(_gvn.transform(result_rgn));
3812 set_all_memory(_gvn.transform(result_mem));
3813
3814 return true;
3815 }
3816
3817 //---------------------------load_mirror_from_klass----------------------------
3818 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3819 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3820 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3821 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3822 // mirror = ((OopHandle)mirror)->resolve();
3823 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3824 }
3825
3826 //-----------------------load_klass_from_mirror_common-------------------------
3827 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3828 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3829 // and branch to the given path on the region.
3830 // If never_see_null, take an uncommon trap on null, so we can optimistically
3831 // compile for the non-null case.
3832 // If the region is null, force never_see_null = true.
3833 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3834 bool never_see_null,
3835 RegionNode* region,
3836 int null_path,
3837 int offset) {
3838 if (region == nullptr) never_see_null = true;
3839 Node* p = basic_plus_adr(mirror, offset);
3840 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3841 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3842 Node* null_ctl = top();
3843 kls = null_check_oop(kls, &null_ctl, never_see_null);
3844 if (region != nullptr) {
3845 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3846 region->init_req(null_path, null_ctl);
3847 } else {
3848 assert(null_ctl == top(), "no loose ends");
3849 }
3850 return kls;
3851 }
3852
3853 //--------------------(inline_native_Class_query helpers)---------------------
3854 // Use this for JVM_ACC_INTERFACE.
3855 // Fall through if (mods & mask) == bits, take the guard otherwise.
3856 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
3857 ByteSize offset, const Type* type, BasicType bt) {
3858 // Branch around if the given klass has the given modifier bit set.
3859 // Like generate_guard, adds a new path onto the region.
3860 Node* modp = basic_plus_adr(kls, in_bytes(offset));
3861 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
3862 Node* mask = intcon(modifier_mask);
3863 Node* bits = intcon(modifier_bits);
3864 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3865 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3866 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3867 return generate_fair_guard(bol, region);
3868 }
3869 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3870 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
3871 Klass::access_flags_offset(), TypeInt::INT, T_INT);
3872 }
3873
3874 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
3875 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3876 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
3877 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
3878 }
3879
3880 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3881 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
3882 }
3883
3884 //-------------------------inline_native_Class_query-------------------
3885 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3886 const Type* return_type = TypeInt::BOOL;
3887 Node* prim_return_value = top(); // what happens if it's a primitive class?
3888 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3889 bool expect_prim = false; // most of these guys expect to work on refs
3890
3891 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3892
3893 Node* mirror = argument(0);
3894 Node* obj = top();
3895
3896 switch (id) {
3897 case vmIntrinsics::_isInstance:
3898 // nothing is an instance of a primitive type
3899 prim_return_value = intcon(0);
3900 obj = argument(1);
3901 break;
3902 case vmIntrinsics::_getModifiers:
3903 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3904 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3905 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3906 break;
3907 case vmIntrinsics::_isInterface:
3908 prim_return_value = intcon(0);
3909 break;
3910 case vmIntrinsics::_isArray:
3911 prim_return_value = intcon(0);
3912 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3913 break;
3914 case vmIntrinsics::_isPrimitive:
3915 prim_return_value = intcon(1);
3916 expect_prim = true; // obviously
3917 break;
3918 case vmIntrinsics::_isHidden:
3919 prim_return_value = intcon(0);
3920 break;
3921 case vmIntrinsics::_getSuperclass:
3922 prim_return_value = null();
3923 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3924 break;
3925 case vmIntrinsics::_getClassAccessFlags:
3926 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3927 return_type = TypeInt::INT; // not bool! 6297094
3928 break;
3929 default:
3930 fatal_unexpected_iid(id);
3931 break;
3932 }
3933
3934 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3935 if (mirror_con == nullptr) return false; // cannot happen?
3936
3937 #ifndef PRODUCT
3938 if (C->print_intrinsics() || C->print_inlining()) {
3939 ciType* k = mirror_con->java_mirror_type();
3940 if (k) {
3941 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3942 k->print_name();
3943 tty->cr();
3944 }
3945 }
3946 #endif
3947
3948 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3949 RegionNode* region = new RegionNode(PATH_LIMIT);
3950 record_for_igvn(region);
3951 PhiNode* phi = new PhiNode(region, return_type);
3952
3953 // The mirror will never be null of Reflection.getClassAccessFlags, however
3954 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3955 // if it is. See bug 4774291.
3956
3957 // For Reflection.getClassAccessFlags(), the null check occurs in
3958 // the wrong place; see inline_unsafe_access(), above, for a similar
3959 // situation.
3960 mirror = null_check(mirror);
3961 // If mirror or obj is dead, only null-path is taken.
3962 if (stopped()) return true;
3963
3964 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3965
3966 // Now load the mirror's klass metaobject, and null-check it.
3967 // Side-effects region with the control path if the klass is null.
3968 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3969 // If kls is null, we have a primitive mirror.
3970 phi->init_req(_prim_path, prim_return_value);
3971 if (stopped()) { set_result(region, phi); return true; }
3972 bool safe_for_replace = (region->in(_prim_path) == top());
3973
3974 Node* p; // handy temp
3975 Node* null_ctl;
3976
3977 // Now that we have the non-null klass, we can perform the real query.
3978 // For constant classes, the query will constant-fold in LoadNode::Value.
3979 Node* query_value = top();
3980 switch (id) {
3981 case vmIntrinsics::_isInstance:
3982 // nothing is an instance of a primitive type
3983 query_value = gen_instanceof(obj, kls, safe_for_replace);
3984 break;
3985
3986 case vmIntrinsics::_getModifiers:
3987 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3988 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
3989 break;
3990
3991 case vmIntrinsics::_isInterface:
3992 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3993 if (generate_interface_guard(kls, region) != nullptr)
3994 // A guard was added. If the guard is taken, it was an interface.
3995 phi->add_req(intcon(1));
3996 // If we fall through, it's a plain class.
3997 query_value = intcon(0);
3998 break;
3999
4000 case vmIntrinsics::_isArray:
4001 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
4002 if (generate_array_guard(kls, region) != nullptr)
4003 // A guard was added. If the guard is taken, it was an array.
4004 phi->add_req(intcon(1));
4005 // If we fall through, it's a plain class.
4006 query_value = intcon(0);
4007 break;
4008
4009 case vmIntrinsics::_isPrimitive:
4010 query_value = intcon(0); // "normal" path produces false
4011 break;
4012
4013 case vmIntrinsics::_isHidden:
4014 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4015 if (generate_hidden_class_guard(kls, region) != nullptr)
4016 // A guard was added. If the guard is taken, it was an hidden class.
4017 phi->add_req(intcon(1));
4018 // If we fall through, it's a plain class.
4019 query_value = intcon(0);
4020 break;
4021
4022
4023 case vmIntrinsics::_getSuperclass:
4024 // The rules here are somewhat unfortunate, but we can still do better
4025 // with random logic than with a JNI call.
4026 // Interfaces store null or Object as _super, but must report null.
4027 // Arrays store an intermediate super as _super, but must report Object.
4028 // Other types can report the actual _super.
4029 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4030 if (generate_interface_guard(kls, region) != nullptr)
4031 // A guard was added. If the guard is taken, it was an interface.
4032 phi->add_req(null());
4033 if (generate_array_guard(kls, region) != nullptr)
4034 // A guard was added. If the guard is taken, it was an array.
4035 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4036 // If we fall through, it's a plain class. Get its _super.
4037 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4038 kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4039 null_ctl = top();
4040 kls = null_check_oop(kls, &null_ctl);
4041 if (null_ctl != top()) {
4042 // If the guard is taken, Object.superClass is null (both klass and mirror).
4043 region->add_req(null_ctl);
4044 phi ->add_req(null());
4045 }
4046 if (!stopped()) {
4047 query_value = load_mirror_from_klass(kls);
4048 }
4049 break;
4050
4051 case vmIntrinsics::_getClassAccessFlags:
4052 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
4053 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
4054 break;
4055
4056 default:
4057 fatal_unexpected_iid(id);
4058 break;
4059 }
4060
4061 // Fall-through is the normal case of a query to a real class.
4062 phi->init_req(1, query_value);
4063 region->init_req(1, control());
4064
4065 C->set_has_split_ifs(true); // Has chance for split-if optimization
4066 set_result(region, phi);
4067 return true;
4068 }
4069
4070 //-------------------------inline_Class_cast-------------------
4071 bool LibraryCallKit::inline_Class_cast() {
4072 Node* mirror = argument(0); // Class
4073 Node* obj = argument(1);
4074 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4075 if (mirror_con == nullptr) {
4076 return false; // dead path (mirror->is_top()).
4077 }
4078 if (obj == nullptr || obj->is_top()) {
4079 return false; // dead path
4080 }
4081 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4082
4083 // First, see if Class.cast() can be folded statically.
4084 // java_mirror_type() returns non-null for compile-time Class constants.
4085 ciType* tm = mirror_con->java_mirror_type();
4086 if (tm != nullptr && tm->is_klass() &&
4087 tp != nullptr) {
4088 if (!tp->is_loaded()) {
4089 // Don't use intrinsic when class is not loaded.
4090 return false;
4091 } else {
4092 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
4093 if (static_res == Compile::SSC_always_true) {
4094 // isInstance() is true - fold the code.
4095 set_result(obj);
4096 return true;
4097 } else if (static_res == Compile::SSC_always_false) {
4098 // Don't use intrinsic, have to throw ClassCastException.
4099 // If the reference is null, the non-intrinsic bytecode will
4100 // be optimized appropriately.
4101 return false;
4102 }
4103 }
4104 }
4105
4106 // Bailout intrinsic and do normal inlining if exception path is frequent.
4107 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4108 return false;
4109 }
4110
4111 // Generate dynamic checks.
4112 // Class.cast() is java implementation of _checkcast bytecode.
4113 // Do checkcast (Parse::do_checkcast()) optimizations here.
4114
4115 mirror = null_check(mirror);
4116 // If mirror is dead, only null-path is taken.
4117 if (stopped()) {
4118 return true;
4119 }
4120
4121 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive).
4122 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
4123 RegionNode* region = new RegionNode(PATH_LIMIT);
4124 record_for_igvn(region);
4125
4126 // Now load the mirror's klass metaobject, and null-check it.
4127 // If kls is null, we have a primitive mirror and
4128 // nothing is an instance of a primitive type.
4129 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4130
4131 Node* res = top();
4132 if (!stopped()) {
4133 Node* bad_type_ctrl = top();
4134 // Do checkcast optimizations.
4135 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4136 region->init_req(_bad_type_path, bad_type_ctrl);
4137 }
4138 if (region->in(_prim_path) != top() ||
4139 region->in(_bad_type_path) != top()) {
4140 // Let Interpreter throw ClassCastException.
4141 PreserveJVMState pjvms(this);
4142 set_control(_gvn.transform(region));
4143 uncommon_trap(Deoptimization::Reason_intrinsic,
4144 Deoptimization::Action_maybe_recompile);
4145 }
4146 if (!stopped()) {
4147 set_result(res);
4148 }
4149 return true;
4150 }
4151
4152
4153 //--------------------------inline_native_subtype_check------------------------
4154 // This intrinsic takes the JNI calls out of the heart of
4155 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4156 bool LibraryCallKit::inline_native_subtype_check() {
4157 // Pull both arguments off the stack.
4158 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4159 args[0] = argument(0);
4160 args[1] = argument(1);
4161 Node* klasses[2]; // corresponding Klasses: superk, subk
4162 klasses[0] = klasses[1] = top();
4163
4164 enum {
4165 // A full decision tree on {superc is prim, subc is prim}:
4166 _prim_0_path = 1, // {P,N} => false
4167 // {P,P} & superc!=subc => false
4168 _prim_same_path, // {P,P} & superc==subc => true
4169 _prim_1_path, // {N,P} => false
4170 _ref_subtype_path, // {N,N} & subtype check wins => true
4171 _both_ref_path, // {N,N} & subtype check loses => false
4172 PATH_LIMIT
4173 };
4174
4175 RegionNode* region = new RegionNode(PATH_LIMIT);
4176 Node* phi = new PhiNode(region, TypeInt::BOOL);
4177 record_for_igvn(region);
4178
4179 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4180 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4181 int class_klass_offset = java_lang_Class::klass_offset();
4182
4183 // First null-check both mirrors and load each mirror's klass metaobject.
4184 int which_arg;
4185 for (which_arg = 0; which_arg <= 1; which_arg++) {
4186 Node* arg = args[which_arg];
4187 arg = null_check(arg);
4188 if (stopped()) break;
4189 args[which_arg] = arg;
4190
4191 Node* p = basic_plus_adr(arg, class_klass_offset);
4192 Node* kls = LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, adr_type, kls_type);
4193 klasses[which_arg] = _gvn.transform(kls);
4194 }
4195
4196 // Having loaded both klasses, test each for null.
4197 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4198 for (which_arg = 0; which_arg <= 1; which_arg++) {
4199 Node* kls = klasses[which_arg];
4200 Node* null_ctl = top();
4201 kls = null_check_oop(kls, &null_ctl, never_see_null);
4202 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
4203 region->init_req(prim_path, null_ctl);
4204 if (stopped()) break;
4205 klasses[which_arg] = kls;
4206 }
4207
4208 if (!stopped()) {
4209 // now we have two reference types, in klasses[0..1]
4210 Node* subk = klasses[1]; // the argument to isAssignableFrom
4211 Node* superk = klasses[0]; // the receiver
4212 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4213 // now we have a successful reference subtype check
4214 region->set_req(_ref_subtype_path, control());
4215 }
4216
4217 // If both operands are primitive (both klasses null), then
4218 // we must return true when they are identical primitives.
4219 // It is convenient to test this after the first null klass check.
4220 set_control(region->in(_prim_0_path)); // go back to first null check
4221 if (!stopped()) {
4222 // Since superc is primitive, make a guard for the superc==subc case.
4223 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4224 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4225 generate_guard(bol_eq, region, PROB_FAIR);
4226 if (region->req() == PATH_LIMIT+1) {
4227 // A guard was added. If the added guard is taken, superc==subc.
4228 region->swap_edges(PATH_LIMIT, _prim_same_path);
4229 region->del_req(PATH_LIMIT);
4230 }
4231 region->set_req(_prim_0_path, control()); // Not equal after all.
4232 }
4233
4234 // these are the only paths that produce 'true':
4235 phi->set_req(_prim_same_path, intcon(1));
4236 phi->set_req(_ref_subtype_path, intcon(1));
4237
4238 // pull together the cases:
4239 assert(region->req() == PATH_LIMIT, "sane region");
4240 for (uint i = 1; i < region->req(); i++) {
4241 Node* ctl = region->in(i);
4242 if (ctl == nullptr || ctl == top()) {
4243 region->set_req(i, top());
4244 phi ->set_req(i, top());
4245 } else if (phi->in(i) == nullptr) {
4246 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4247 }
4248 }
4249
4250 set_control(_gvn.transform(region));
4251 set_result(_gvn.transform(phi));
4252 return true;
4253 }
4254
4255 //---------------------generate_array_guard_common------------------------
4256 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
4257 bool obj_array, bool not_array) {
4258
4259 if (stopped()) {
4260 return nullptr;
4261 }
4262
4263 // If obj_array/non_array==false/false:
4264 // Branch around if the given klass is in fact an array (either obj or prim).
4265 // If obj_array/non_array==false/true:
4266 // Branch around if the given klass is not an array klass of any kind.
4267 // If obj_array/non_array==true/true:
4268 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
4269 // If obj_array/non_array==true/false:
4270 // Branch around if the kls is an oop array (Object[] or subtype)
4271 //
4272 // Like generate_guard, adds a new path onto the region.
4273 jint layout_con = 0;
4274 Node* layout_val = get_layout_helper(kls, layout_con);
4275 if (layout_val == nullptr) {
4276 bool query = (obj_array
4277 ? Klass::layout_helper_is_objArray(layout_con)
4278 : Klass::layout_helper_is_array(layout_con));
4279 if (query == not_array) {
4280 return nullptr; // never a branch
4281 } else { // always a branch
4282 Node* always_branch = control();
4283 if (region != nullptr)
4284 region->add_req(always_branch);
4285 set_control(top());
4286 return always_branch;
4287 }
4288 }
4289 // Now test the correct condition.
4290 jint nval = (obj_array
4291 ? (jint)(Klass::_lh_array_tag_type_value
4292 << Klass::_lh_array_tag_shift)
4293 : Klass::_lh_neutral_value);
4294 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4295 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
4296 // invert the test if we are looking for a non-array
4297 if (not_array) btest = BoolTest(btest).negate();
4298 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4299 return generate_fair_guard(bol, region);
4300 }
4301
4302
4303 //-----------------------inline_native_newArray--------------------------
4304 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4305 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4306 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4307 Node* mirror;
4308 Node* count_val;
4309 if (uninitialized) {
4310 null_check_receiver();
4311 mirror = argument(1);
4312 count_val = argument(2);
4313 } else {
4314 mirror = argument(0);
4315 count_val = argument(1);
4316 }
4317
4318 mirror = null_check(mirror);
4319 // If mirror or obj is dead, only null-path is taken.
4320 if (stopped()) return true;
4321
4322 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4323 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4324 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4325 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4326 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4327
4328 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4329 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4330 result_reg, _slow_path);
4331 Node* normal_ctl = control();
4332 Node* no_array_ctl = result_reg->in(_slow_path);
4333
4334 // Generate code for the slow case. We make a call to newArray().
4335 set_control(no_array_ctl);
4336 if (!stopped()) {
4337 // Either the input type is void.class, or else the
4338 // array klass has not yet been cached. Either the
4339 // ensuing call will throw an exception, or else it
4340 // will cache the array klass for next time.
4341 PreserveJVMState pjvms(this);
4342 CallJavaNode* slow_call = nullptr;
4343 if (uninitialized) {
4344 // Generate optimized virtual call (holder class 'Unsafe' is final)
4345 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4346 } else {
4347 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4348 }
4349 Node* slow_result = set_results_for_java_call(slow_call);
4350 // this->control() comes from set_results_for_java_call
4351 result_reg->set_req(_slow_path, control());
4352 result_val->set_req(_slow_path, slow_result);
4353 result_io ->set_req(_slow_path, i_o());
4354 result_mem->set_req(_slow_path, reset_memory());
4355 }
4356
4357 set_control(normal_ctl);
4358 if (!stopped()) {
4359 // Normal case: The array type has been cached in the java.lang.Class.
4360 // The following call works fine even if the array type is polymorphic.
4361 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4362 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4363 result_reg->init_req(_normal_path, control());
4364 result_val->init_req(_normal_path, obj);
4365 result_io ->init_req(_normal_path, i_o());
4366 result_mem->init_req(_normal_path, reset_memory());
4367
4368 if (uninitialized) {
4369 // Mark the allocation so that zeroing is skipped
4370 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4371 alloc->maybe_set_complete(&_gvn);
4372 }
4373 }
4374
4375 // Return the combined state.
4376 set_i_o( _gvn.transform(result_io) );
4377 set_all_memory( _gvn.transform(result_mem));
4378
4379 C->set_has_split_ifs(true); // Has chance for split-if optimization
4380 set_result(result_reg, result_val);
4381 return true;
4382 }
4383
4384 //----------------------inline_native_getLength--------------------------
4385 // public static native int java.lang.reflect.Array.getLength(Object array);
4386 bool LibraryCallKit::inline_native_getLength() {
4387 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4388
4389 Node* array = null_check(argument(0));
4390 // If array is dead, only null-path is taken.
4391 if (stopped()) return true;
4392
4393 // Deoptimize if it is a non-array.
4394 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr);
4395
4396 if (non_array != nullptr) {
4397 PreserveJVMState pjvms(this);
4398 set_control(non_array);
4399 uncommon_trap(Deoptimization::Reason_intrinsic,
4400 Deoptimization::Action_maybe_recompile);
4401 }
4402
4403 // If control is dead, only non-array-path is taken.
4404 if (stopped()) return true;
4405
4406 // The works fine even if the array type is polymorphic.
4407 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4408 Node* result = load_array_length(array);
4409
4410 C->set_has_split_ifs(true); // Has chance for split-if optimization
4411 set_result(result);
4412 return true;
4413 }
4414
4415 //------------------------inline_array_copyOf----------------------------
4416 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4417 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4418 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4419 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4420
4421 // Get the arguments.
4422 Node* original = argument(0);
4423 Node* start = is_copyOfRange? argument(1): intcon(0);
4424 Node* end = is_copyOfRange? argument(2): argument(1);
4425 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4426
4427 Node* newcopy = nullptr;
4428
4429 // Set the original stack and the reexecute bit for the interpreter to reexecute
4430 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4431 { PreserveReexecuteState preexecs(this);
4432 jvms()->set_should_reexecute(true);
4433
4434 array_type_mirror = null_check(array_type_mirror);
4435 original = null_check(original);
4436
4437 // Check if a null path was taken unconditionally.
4438 if (stopped()) return true;
4439
4440 Node* orig_length = load_array_length(original);
4441
4442 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4443 klass_node = null_check(klass_node);
4444
4445 RegionNode* bailout = new RegionNode(1);
4446 record_for_igvn(bailout);
4447
4448 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4449 // Bail out if that is so.
4450 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4451 if (not_objArray != nullptr) {
4452 // Improve the klass node's type from the new optimistic assumption:
4453 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4454 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4455 Node* cast = new CastPPNode(control(), klass_node, akls);
4456 klass_node = _gvn.transform(cast);
4457 }
4458
4459 // Bail out if either start or end is negative.
4460 generate_negative_guard(start, bailout, &start);
4461 generate_negative_guard(end, bailout, &end);
4462
4463 Node* length = end;
4464 if (_gvn.type(start) != TypeInt::ZERO) {
4465 length = _gvn.transform(new SubINode(end, start));
4466 }
4467
4468 // Bail out if length is negative (i.e., if start > end).
4469 // Without this the new_array would throw
4470 // NegativeArraySizeException but IllegalArgumentException is what
4471 // should be thrown
4472 generate_negative_guard(length, bailout, &length);
4473
4474 // Bail out if start is larger than the original length
4475 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4476 generate_negative_guard(orig_tail, bailout, &orig_tail);
4477
4478 if (bailout->req() > 1) {
4479 PreserveJVMState pjvms(this);
4480 set_control(_gvn.transform(bailout));
4481 uncommon_trap(Deoptimization::Reason_intrinsic,
4482 Deoptimization::Action_maybe_recompile);
4483 }
4484
4485 if (!stopped()) {
4486 // How many elements will we copy from the original?
4487 // The answer is MinI(orig_tail, length).
4488 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4489
4490 // Generate a direct call to the right arraycopy function(s).
4491 // We know the copy is disjoint but we might not know if the
4492 // oop stores need checking.
4493 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4494 // This will fail a store-check if x contains any non-nulls.
4495
4496 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4497 // loads/stores but it is legal only if we're sure the
4498 // Arrays.copyOf would succeed. So we need all input arguments
4499 // to the copyOf to be validated, including that the copy to the
4500 // new array won't trigger an ArrayStoreException. That subtype
4501 // check can be optimized if we know something on the type of
4502 // the input array from type speculation.
4503 if (_gvn.type(klass_node)->singleton()) {
4504 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4505 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4506
4507 int test = C->static_subtype_check(superk, subk);
4508 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4509 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4510 if (t_original->speculative_type() != nullptr) {
4511 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4512 }
4513 }
4514 }
4515
4516 bool validated = false;
4517 // Reason_class_check rather than Reason_intrinsic because we
4518 // want to intrinsify even if this traps.
4519 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4520 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4521
4522 if (not_subtype_ctrl != top()) {
4523 PreserveJVMState pjvms(this);
4524 set_control(not_subtype_ctrl);
4525 uncommon_trap(Deoptimization::Reason_class_check,
4526 Deoptimization::Action_make_not_entrant);
4527 assert(stopped(), "Should be stopped");
4528 }
4529 validated = true;
4530 }
4531
4532 if (!stopped()) {
4533 newcopy = new_array(klass_node, length, 0); // no arguments to push
4534
4535 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4536 load_object_klass(original), klass_node);
4537 if (!is_copyOfRange) {
4538 ac->set_copyof(validated);
4539 } else {
4540 ac->set_copyofrange(validated);
4541 }
4542 Node* n = _gvn.transform(ac);
4543 if (n == ac) {
4544 ac->connect_outputs(this);
4545 } else {
4546 assert(validated, "shouldn't transform if all arguments not validated");
4547 set_all_memory(n);
4548 }
4549 }
4550 }
4551 } // original reexecute is set back here
4552
4553 C->set_has_split_ifs(true); // Has chance for split-if optimization
4554 if (!stopped()) {
4555 set_result(newcopy);
4556 }
4557 return true;
4558 }
4559
4560
4561 //----------------------generate_virtual_guard---------------------------
4562 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4563 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4564 RegionNode* slow_region) {
4565 ciMethod* method = callee();
4566 int vtable_index = method->vtable_index();
4567 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4568 "bad index %d", vtable_index);
4569 // Get the Method* out of the appropriate vtable entry.
4570 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4571 vtable_index*vtableEntry::size_in_bytes() +
4572 in_bytes(vtableEntry::method_offset());
4573 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4574 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4575
4576 // Compare the target method with the expected method (e.g., Object.hashCode).
4577 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4578
4579 Node* native_call = makecon(native_call_addr);
4580 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4581 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4582
4583 return generate_slow_guard(test_native, slow_region);
4584 }
4585
4586 //-----------------------generate_method_call----------------------------
4587 // Use generate_method_call to make a slow-call to the real
4588 // method if the fast path fails. An alternative would be to
4589 // use a stub like OptoRuntime::slow_arraycopy_Java.
4590 // This only works for expanding the current library call,
4591 // not another intrinsic. (E.g., don't use this for making an
4592 // arraycopy call inside of the copyOf intrinsic.)
4593 CallJavaNode*
4594 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4595 // When compiling the intrinsic method itself, do not use this technique.
4596 guarantee(callee() != C->method(), "cannot make slow-call to self");
4597
4598 ciMethod* method = callee();
4599 // ensure the JVMS we have will be correct for this call
4600 guarantee(method_id == method->intrinsic_id(), "must match");
4601
4602 const TypeFunc* tf = TypeFunc::make(method);
4603 if (res_not_null) {
4604 assert(tf->return_type() == T_OBJECT, "");
4605 const TypeTuple* range = tf->range();
4606 const Type** fields = TypeTuple::fields(range->cnt());
4607 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4608 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4609 tf = TypeFunc::make(tf->domain(), new_range);
4610 }
4611 CallJavaNode* slow_call;
4612 if (is_static) {
4613 assert(!is_virtual, "");
4614 slow_call = new CallStaticJavaNode(C, tf,
4615 SharedRuntime::get_resolve_static_call_stub(), method);
4616 } else if (is_virtual) {
4617 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4618 int vtable_index = Method::invalid_vtable_index;
4619 if (UseInlineCaches) {
4620 // Suppress the vtable call
4621 } else {
4622 // hashCode and clone are not a miranda methods,
4623 // so the vtable index is fixed.
4624 // No need to use the linkResolver to get it.
4625 vtable_index = method->vtable_index();
4626 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4627 "bad index %d", vtable_index);
4628 }
4629 slow_call = new CallDynamicJavaNode(tf,
4630 SharedRuntime::get_resolve_virtual_call_stub(),
4631 method, vtable_index);
4632 } else { // neither virtual nor static: opt_virtual
4633 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4634 slow_call = new CallStaticJavaNode(C, tf,
4635 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4636 slow_call->set_optimized_virtual(true);
4637 }
4638 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4639 // To be able to issue a direct call (optimized virtual or virtual)
4640 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4641 // about the method being invoked should be attached to the call site to
4642 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4643 slow_call->set_override_symbolic_info(true);
4644 }
4645 set_arguments_for_java_call(slow_call);
4646 set_edges_for_java_call(slow_call);
4647 return slow_call;
4648 }
4649
4650
4651 /**
4652 * Build special case code for calls to hashCode on an object. This call may
4653 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4654 * slightly different code.
4655 */
4656 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4657 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4658 assert(!(is_virtual && is_static), "either virtual, special, or static");
4659
4660 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4661
4662 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4663 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4664 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4665 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4666 Node* obj = nullptr;
4667 if (!is_static) {
4668 // Check for hashing null object
4669 obj = null_check_receiver();
4670 if (stopped()) return true; // unconditionally null
4671 result_reg->init_req(_null_path, top());
4672 result_val->init_req(_null_path, top());
4673 } else {
4674 // Do a null check, and return zero if null.
4675 // System.identityHashCode(null) == 0
4676 obj = argument(0);
4677 Node* null_ctl = top();
4678 obj = null_check_oop(obj, &null_ctl);
4679 result_reg->init_req(_null_path, null_ctl);
4680 result_val->init_req(_null_path, _gvn.intcon(0));
4681 }
4682
4683 // Unconditionally null? Then return right away.
4684 if (stopped()) {
4685 set_control( result_reg->in(_null_path));
4686 if (!stopped())
4687 set_result(result_val->in(_null_path));
4688 return true;
4689 }
4690
4691 // We only go to the fast case code if we pass a number of guards. The
4692 // paths which do not pass are accumulated in the slow_region.
4693 RegionNode* slow_region = new RegionNode(1);
4694 record_for_igvn(slow_region);
4695
4696 // If this is a virtual call, we generate a funny guard. We pull out
4697 // the vtable entry corresponding to hashCode() from the target object.
4698 // If the target method which we are calling happens to be the native
4699 // Object hashCode() method, we pass the guard. We do not need this
4700 // guard for non-virtual calls -- the caller is known to be the native
4701 // Object hashCode().
4702 if (is_virtual) {
4703 // After null check, get the object's klass.
4704 Node* obj_klass = load_object_klass(obj);
4705 generate_virtual_guard(obj_klass, slow_region);
4706 }
4707
4708 // Get the header out of the object, use LoadMarkNode when available
4709 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4710 // The control of the load must be null. Otherwise, the load can move before
4711 // the null check after castPP removal.
4712 Node* no_ctrl = nullptr;
4713 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4714
4715 if (!UseObjectMonitorTable) {
4716 // Test the header to see if it is safe to read w.r.t. locking.
4717 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4718 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4719 if (LockingMode == LM_LIGHTWEIGHT) {
4720 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
4721 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
4722 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
4723
4724 generate_slow_guard(test_monitor, slow_region);
4725 } else {
4726 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4727 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val));
4728 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne));
4729
4730 generate_slow_guard(test_not_unlocked, slow_region);
4731 }
4732 }
4733
4734 // Get the hash value and check to see that it has been properly assigned.
4735 // We depend on hash_mask being at most 32 bits and avoid the use of
4736 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4737 // vm: see markWord.hpp.
4738 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4739 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4740 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4741 // This hack lets the hash bits live anywhere in the mark object now, as long
4742 // as the shift drops the relevant bits into the low 32 bits. Note that
4743 // Java spec says that HashCode is an int so there's no point in capturing
4744 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4745 hshifted_header = ConvX2I(hshifted_header);
4746 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4747
4748 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4749 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4750 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4751
4752 generate_slow_guard(test_assigned, slow_region);
4753
4754 Node* init_mem = reset_memory();
4755 // fill in the rest of the null path:
4756 result_io ->init_req(_null_path, i_o());
4757 result_mem->init_req(_null_path, init_mem);
4758
4759 result_val->init_req(_fast_path, hash_val);
4760 result_reg->init_req(_fast_path, control());
4761 result_io ->init_req(_fast_path, i_o());
4762 result_mem->init_req(_fast_path, init_mem);
4763
4764 // Generate code for the slow case. We make a call to hashCode().
4765 set_control(_gvn.transform(slow_region));
4766 if (!stopped()) {
4767 // No need for PreserveJVMState, because we're using up the present state.
4768 set_all_memory(init_mem);
4769 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4770 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4771 Node* slow_result = set_results_for_java_call(slow_call);
4772 // this->control() comes from set_results_for_java_call
4773 result_reg->init_req(_slow_path, control());
4774 result_val->init_req(_slow_path, slow_result);
4775 result_io ->set_req(_slow_path, i_o());
4776 result_mem ->set_req(_slow_path, reset_memory());
4777 }
4778
4779 // Return the combined state.
4780 set_i_o( _gvn.transform(result_io) );
4781 set_all_memory( _gvn.transform(result_mem));
4782
4783 set_result(result_reg, result_val);
4784 return true;
4785 }
4786
4787 //---------------------------inline_native_getClass----------------------------
4788 // public final native Class<?> java.lang.Object.getClass();
4789 //
4790 // Build special case code for calls to getClass on an object.
4791 bool LibraryCallKit::inline_native_getClass() {
4792 Node* obj = null_check_receiver();
4793 if (stopped()) return true;
4794 set_result(load_mirror_from_klass(load_object_klass(obj)));
4795 return true;
4796 }
4797
4798 //-----------------inline_native_Reflection_getCallerClass---------------------
4799 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4800 //
4801 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4802 //
4803 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4804 // in that it must skip particular security frames and checks for
4805 // caller sensitive methods.
4806 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4807 #ifndef PRODUCT
4808 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4809 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4810 }
4811 #endif
4812
4813 if (!jvms()->has_method()) {
4814 #ifndef PRODUCT
4815 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4816 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4817 }
4818 #endif
4819 return false;
4820 }
4821
4822 // Walk back up the JVM state to find the caller at the required
4823 // depth.
4824 JVMState* caller_jvms = jvms();
4825
4826 // Cf. JVM_GetCallerClass
4827 // NOTE: Start the loop at depth 1 because the current JVM state does
4828 // not include the Reflection.getCallerClass() frame.
4829 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4830 ciMethod* m = caller_jvms->method();
4831 switch (n) {
4832 case 0:
4833 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4834 break;
4835 case 1:
4836 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4837 if (!m->caller_sensitive()) {
4838 #ifndef PRODUCT
4839 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4840 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4841 }
4842 #endif
4843 return false; // bail-out; let JVM_GetCallerClass do the work
4844 }
4845 break;
4846 default:
4847 if (!m->is_ignored_by_security_stack_walk()) {
4848 // We have reached the desired frame; return the holder class.
4849 // Acquire method holder as java.lang.Class and push as constant.
4850 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4851 ciInstance* caller_mirror = caller_klass->java_mirror();
4852 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4853
4854 #ifndef PRODUCT
4855 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4856 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());
4857 tty->print_cr(" JVM state at this point:");
4858 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4859 ciMethod* m = jvms()->of_depth(i)->method();
4860 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4861 }
4862 }
4863 #endif
4864 return true;
4865 }
4866 break;
4867 }
4868 }
4869
4870 #ifndef PRODUCT
4871 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4872 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4873 tty->print_cr(" JVM state at this point:");
4874 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4875 ciMethod* m = jvms()->of_depth(i)->method();
4876 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4877 }
4878 }
4879 #endif
4880
4881 return false; // bail-out; let JVM_GetCallerClass do the work
4882 }
4883
4884 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4885 Node* arg = argument(0);
4886 Node* result = nullptr;
4887
4888 switch (id) {
4889 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4890 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4891 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4892 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4893 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
4894 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
4895
4896 case vmIntrinsics::_doubleToLongBits: {
4897 // two paths (plus control) merge in a wood
4898 RegionNode *r = new RegionNode(3);
4899 Node *phi = new PhiNode(r, TypeLong::LONG);
4900
4901 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4902 // Build the boolean node
4903 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4904
4905 // Branch either way.
4906 // NaN case is less traveled, which makes all the difference.
4907 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4908 Node *opt_isnan = _gvn.transform(ifisnan);
4909 assert( opt_isnan->is_If(), "Expect an IfNode");
4910 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4911 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4912
4913 set_control(iftrue);
4914
4915 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4916 Node *slow_result = longcon(nan_bits); // return NaN
4917 phi->init_req(1, _gvn.transform( slow_result ));
4918 r->init_req(1, iftrue);
4919
4920 // Else fall through
4921 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4922 set_control(iffalse);
4923
4924 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4925 r->init_req(2, iffalse);
4926
4927 // Post merge
4928 set_control(_gvn.transform(r));
4929 record_for_igvn(r);
4930
4931 C->set_has_split_ifs(true); // Has chance for split-if optimization
4932 result = phi;
4933 assert(result->bottom_type()->isa_long(), "must be");
4934 break;
4935 }
4936
4937 case vmIntrinsics::_floatToIntBits: {
4938 // two paths (plus control) merge in a wood
4939 RegionNode *r = new RegionNode(3);
4940 Node *phi = new PhiNode(r, TypeInt::INT);
4941
4942 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4943 // Build the boolean node
4944 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4945
4946 // Branch either way.
4947 // NaN case is less traveled, which makes all the difference.
4948 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4949 Node *opt_isnan = _gvn.transform(ifisnan);
4950 assert( opt_isnan->is_If(), "Expect an IfNode");
4951 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4952 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4953
4954 set_control(iftrue);
4955
4956 static const jint nan_bits = 0x7fc00000;
4957 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4958 phi->init_req(1, _gvn.transform( slow_result ));
4959 r->init_req(1, iftrue);
4960
4961 // Else fall through
4962 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4963 set_control(iffalse);
4964
4965 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4966 r->init_req(2, iffalse);
4967
4968 // Post merge
4969 set_control(_gvn.transform(r));
4970 record_for_igvn(r);
4971
4972 C->set_has_split_ifs(true); // Has chance for split-if optimization
4973 result = phi;
4974 assert(result->bottom_type()->isa_int(), "must be");
4975 break;
4976 }
4977
4978 default:
4979 fatal_unexpected_iid(id);
4980 break;
4981 }
4982 set_result(_gvn.transform(result));
4983 return true;
4984 }
4985
4986 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
4987 Node* arg = argument(0);
4988 Node* result = nullptr;
4989
4990 switch (id) {
4991 case vmIntrinsics::_floatIsInfinite:
4992 result = new IsInfiniteFNode(arg);
4993 break;
4994 case vmIntrinsics::_floatIsFinite:
4995 result = new IsFiniteFNode(arg);
4996 break;
4997 case vmIntrinsics::_doubleIsInfinite:
4998 result = new IsInfiniteDNode(arg);
4999 break;
5000 case vmIntrinsics::_doubleIsFinite:
5001 result = new IsFiniteDNode(arg);
5002 break;
5003 default:
5004 fatal_unexpected_iid(id);
5005 break;
5006 }
5007 set_result(_gvn.transform(result));
5008 return true;
5009 }
5010
5011 //----------------------inline_unsafe_copyMemory-------------------------
5012 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5013
5014 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5015 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5016 const Type* base_t = gvn.type(base);
5017
5018 bool in_native = (base_t == TypePtr::NULL_PTR);
5019 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5020 bool is_mixed = !in_heap && !in_native;
5021
5022 if (is_mixed) {
5023 return true; // mixed accesses can touch both on-heap and off-heap memory
5024 }
5025 if (in_heap) {
5026 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5027 if (!is_prim_array) {
5028 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5029 // there's not enough type information available to determine proper memory slice for it.
5030 return true;
5031 }
5032 }
5033 return false;
5034 }
5035
5036 bool LibraryCallKit::inline_unsafe_copyMemory() {
5037 if (callee()->is_static()) return false; // caller must have the capability!
5038 null_check_receiver(); // null-check receiver
5039 if (stopped()) return true;
5040
5041 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5042
5043 Node* src_base = argument(1); // type: oop
5044 Node* src_off = ConvL2X(argument(2)); // type: long
5045 Node* dst_base = argument(4); // type: oop
5046 Node* dst_off = ConvL2X(argument(5)); // type: long
5047 Node* size = ConvL2X(argument(7)); // type: long
5048
5049 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5050 "fieldOffset must be byte-scaled");
5051
5052 Node* src_addr = make_unsafe_address(src_base, src_off);
5053 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5054
5055 Node* thread = _gvn.transform(new ThreadLocalNode());
5056 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5057 BasicType doing_unsafe_access_bt = T_BYTE;
5058 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5059
5060 // update volatile field
5061 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5062
5063 int flags = RC_LEAF | RC_NO_FP;
5064
5065 const TypePtr* dst_type = TypePtr::BOTTOM;
5066
5067 // Adjust memory effects of the runtime call based on input values.
5068 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5069 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5070 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5071
5072 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5073 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5074 flags |= RC_NARROW_MEM; // narrow in memory
5075 }
5076 }
5077
5078 // Call it. Note that the length argument is not scaled.
5079 make_runtime_call(flags,
5080 OptoRuntime::fast_arraycopy_Type(),
5081 StubRoutines::unsafe_arraycopy(),
5082 "unsafe_arraycopy",
5083 dst_type,
5084 src_addr, dst_addr, size XTOP);
5085
5086 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5087
5088 return true;
5089 }
5090
5091 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5092 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5093 bool LibraryCallKit::inline_unsafe_setMemory() {
5094 if (callee()->is_static()) return false; // caller must have the capability!
5095 null_check_receiver(); // null-check receiver
5096 if (stopped()) return true;
5097
5098 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5099
5100 Node* dst_base = argument(1); // type: oop
5101 Node* dst_off = ConvL2X(argument(2)); // type: long
5102 Node* size = ConvL2X(argument(4)); // type: long
5103 Node* byte = argument(6); // type: byte
5104
5105 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5106 "fieldOffset must be byte-scaled");
5107
5108 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5109
5110 Node* thread = _gvn.transform(new ThreadLocalNode());
5111 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5112 BasicType doing_unsafe_access_bt = T_BYTE;
5113 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5114
5115 // update volatile field
5116 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5117
5118 int flags = RC_LEAF | RC_NO_FP;
5119
5120 const TypePtr* dst_type = TypePtr::BOTTOM;
5121
5122 // Adjust memory effects of the runtime call based on input values.
5123 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5124 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5125
5126 flags |= RC_NARROW_MEM; // narrow in memory
5127 }
5128
5129 // Call it. Note that the length argument is not scaled.
5130 make_runtime_call(flags,
5131 OptoRuntime::make_setmemory_Type(),
5132 StubRoutines::unsafe_setmemory(),
5133 "unsafe_setmemory",
5134 dst_type,
5135 dst_addr, size XTOP, byte);
5136
5137 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5138
5139 return true;
5140 }
5141
5142 #undef XTOP
5143
5144 //------------------------clone_coping-----------------------------------
5145 // Helper function for inline_native_clone.
5146 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5147 assert(obj_size != nullptr, "");
5148 Node* raw_obj = alloc_obj->in(1);
5149 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5150
5151 AllocateNode* alloc = nullptr;
5152 if (ReduceBulkZeroing &&
5153 // If we are implementing an array clone without knowing its source type
5154 // (can happen when compiling the array-guarded branch of a reflective
5155 // Object.clone() invocation), initialize the array within the allocation.
5156 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5157 // to a runtime clone call that assumes fully initialized source arrays.
5158 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5159 // We will be completely responsible for initializing this object -
5160 // mark Initialize node as complete.
5161 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5162 // The object was just allocated - there should be no any stores!
5163 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5164 // Mark as complete_with_arraycopy so that on AllocateNode
5165 // expansion, we know this AllocateNode is initialized by an array
5166 // copy and a StoreStore barrier exists after the array copy.
5167 alloc->initialization()->set_complete_with_arraycopy();
5168 }
5169
5170 Node* size = _gvn.transform(obj_size);
5171 access_clone(obj, alloc_obj, size, is_array);
5172
5173 // Do not let reads from the cloned object float above the arraycopy.
5174 if (alloc != nullptr) {
5175 // Do not let stores that initialize this object be reordered with
5176 // a subsequent store that would make this object accessible by
5177 // other threads.
5178 // Record what AllocateNode this StoreStore protects so that
5179 // escape analysis can go from the MemBarStoreStoreNode to the
5180 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5181 // based on the escape status of the AllocateNode.
5182 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5183 } else {
5184 insert_mem_bar(Op_MemBarCPUOrder);
5185 }
5186 }
5187
5188 //------------------------inline_native_clone----------------------------
5189 // protected native Object java.lang.Object.clone();
5190 //
5191 // Here are the simple edge cases:
5192 // null receiver => normal trap
5193 // virtual and clone was overridden => slow path to out-of-line clone
5194 // not cloneable or finalizer => slow path to out-of-line Object.clone
5195 //
5196 // The general case has two steps, allocation and copying.
5197 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5198 //
5199 // Copying also has two cases, oop arrays and everything else.
5200 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5201 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5202 //
5203 // These steps fold up nicely if and when the cloned object's klass
5204 // can be sharply typed as an object array, a type array, or an instance.
5205 //
5206 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5207 PhiNode* result_val;
5208
5209 // Set the reexecute bit for the interpreter to reexecute
5210 // the bytecode that invokes Object.clone if deoptimization happens.
5211 { PreserveReexecuteState preexecs(this);
5212 jvms()->set_should_reexecute(true);
5213
5214 Node* obj = null_check_receiver();
5215 if (stopped()) return true;
5216
5217 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5218
5219 // If we are going to clone an instance, we need its exact type to
5220 // know the number and types of fields to convert the clone to
5221 // loads/stores. Maybe a speculative type can help us.
5222 if (!obj_type->klass_is_exact() &&
5223 obj_type->speculative_type() != nullptr &&
5224 obj_type->speculative_type()->is_instance_klass()) {
5225 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5226 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5227 !spec_ik->has_injected_fields()) {
5228 if (!obj_type->isa_instptr() ||
5229 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5230 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5231 }
5232 }
5233 }
5234
5235 // Conservatively insert a memory barrier on all memory slices.
5236 // Do not let writes into the original float below the clone.
5237 insert_mem_bar(Op_MemBarCPUOrder);
5238
5239 // paths into result_reg:
5240 enum {
5241 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5242 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5243 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5244 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5245 PATH_LIMIT
5246 };
5247 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5248 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5249 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5250 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5251 record_for_igvn(result_reg);
5252
5253 Node* obj_klass = load_object_klass(obj);
5254 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr);
5255 if (array_ctl != nullptr) {
5256 // It's an array.
5257 PreserveJVMState pjvms(this);
5258 set_control(array_ctl);
5259 Node* obj_length = load_array_length(obj);
5260 Node* array_size = nullptr; // Size of the array without object alignment padding.
5261 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5262
5263 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5264 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5265 // If it is an oop array, it requires very special treatment,
5266 // because gc barriers are required when accessing the array.
5267 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5268 if (is_obja != nullptr) {
5269 PreserveJVMState pjvms2(this);
5270 set_control(is_obja);
5271 // Generate a direct call to the right arraycopy function(s).
5272 // Clones are always tightly coupled.
5273 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5274 ac->set_clone_oop_array();
5275 Node* n = _gvn.transform(ac);
5276 assert(n == ac, "cannot disappear");
5277 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5278
5279 result_reg->init_req(_objArray_path, control());
5280 result_val->init_req(_objArray_path, alloc_obj);
5281 result_i_o ->set_req(_objArray_path, i_o());
5282 result_mem ->set_req(_objArray_path, reset_memory());
5283 }
5284 }
5285 // Otherwise, there are no barriers to worry about.
5286 // (We can dispense with card marks if we know the allocation
5287 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5288 // causes the non-eden paths to take compensating steps to
5289 // simulate a fresh allocation, so that no further
5290 // card marks are required in compiled code to initialize
5291 // the object.)
5292
5293 if (!stopped()) {
5294 copy_to_clone(obj, alloc_obj, array_size, true);
5295
5296 // Present the results of the copy.
5297 result_reg->init_req(_array_path, control());
5298 result_val->init_req(_array_path, alloc_obj);
5299 result_i_o ->set_req(_array_path, i_o());
5300 result_mem ->set_req(_array_path, reset_memory());
5301 }
5302 }
5303
5304 // We only go to the instance fast case code if we pass a number of guards.
5305 // The paths which do not pass are accumulated in the slow_region.
5306 RegionNode* slow_region = new RegionNode(1);
5307 record_for_igvn(slow_region);
5308 if (!stopped()) {
5309 // It's an instance (we did array above). Make the slow-path tests.
5310 // If this is a virtual call, we generate a funny guard. We grab
5311 // the vtable entry corresponding to clone() from the target object.
5312 // If the target method which we are calling happens to be the
5313 // Object clone() method, we pass the guard. We do not need this
5314 // guard for non-virtual calls; the caller is known to be the native
5315 // Object clone().
5316 if (is_virtual) {
5317 generate_virtual_guard(obj_klass, slow_region);
5318 }
5319
5320 // The object must be easily cloneable and must not have a finalizer.
5321 // Both of these conditions may be checked in a single test.
5322 // We could optimize the test further, but we don't care.
5323 generate_misc_flags_guard(obj_klass,
5324 // Test both conditions:
5325 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5326 // Must be cloneable but not finalizer:
5327 KlassFlags::_misc_is_cloneable_fast,
5328 slow_region);
5329 }
5330
5331 if (!stopped()) {
5332 // It's an instance, and it passed the slow-path tests.
5333 PreserveJVMState pjvms(this);
5334 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5335 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5336 // is reexecuted if deoptimization occurs and there could be problems when merging
5337 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5338 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5339
5340 copy_to_clone(obj, alloc_obj, obj_size, false);
5341
5342 // Present the results of the slow call.
5343 result_reg->init_req(_instance_path, control());
5344 result_val->init_req(_instance_path, alloc_obj);
5345 result_i_o ->set_req(_instance_path, i_o());
5346 result_mem ->set_req(_instance_path, reset_memory());
5347 }
5348
5349 // Generate code for the slow case. We make a call to clone().
5350 set_control(_gvn.transform(slow_region));
5351 if (!stopped()) {
5352 PreserveJVMState pjvms(this);
5353 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5354 // We need to deoptimize on exception (see comment above)
5355 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5356 // this->control() comes from set_results_for_java_call
5357 result_reg->init_req(_slow_path, control());
5358 result_val->init_req(_slow_path, slow_result);
5359 result_i_o ->set_req(_slow_path, i_o());
5360 result_mem ->set_req(_slow_path, reset_memory());
5361 }
5362
5363 // Return the combined state.
5364 set_control( _gvn.transform(result_reg));
5365 set_i_o( _gvn.transform(result_i_o));
5366 set_all_memory( _gvn.transform(result_mem));
5367 } // original reexecute is set back here
5368
5369 set_result(_gvn.transform(result_val));
5370 return true;
5371 }
5372
5373 // If we have a tightly coupled allocation, the arraycopy may take care
5374 // of the array initialization. If one of the guards we insert between
5375 // the allocation and the arraycopy causes a deoptimization, an
5376 // uninitialized array will escape the compiled method. To prevent that
5377 // we set the JVM state for uncommon traps between the allocation and
5378 // the arraycopy to the state before the allocation so, in case of
5379 // deoptimization, we'll reexecute the allocation and the
5380 // initialization.
5381 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5382 if (alloc != nullptr) {
5383 ciMethod* trap_method = alloc->jvms()->method();
5384 int trap_bci = alloc->jvms()->bci();
5385
5386 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5387 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5388 // Make sure there's no store between the allocation and the
5389 // arraycopy otherwise visible side effects could be rexecuted
5390 // in case of deoptimization and cause incorrect execution.
5391 bool no_interfering_store = true;
5392 Node* mem = alloc->in(TypeFunc::Memory);
5393 if (mem->is_MergeMem()) {
5394 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5395 Node* n = mms.memory();
5396 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5397 assert(n->is_Store(), "what else?");
5398 no_interfering_store = false;
5399 break;
5400 }
5401 }
5402 } else {
5403 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5404 Node* n = mms.memory();
5405 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5406 assert(n->is_Store(), "what else?");
5407 no_interfering_store = false;
5408 break;
5409 }
5410 }
5411 }
5412
5413 if (no_interfering_store) {
5414 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5415
5416 JVMState* saved_jvms = jvms();
5417 saved_reexecute_sp = _reexecute_sp;
5418
5419 set_jvms(sfpt->jvms());
5420 _reexecute_sp = jvms()->sp();
5421
5422 return saved_jvms;
5423 }
5424 }
5425 }
5426 return nullptr;
5427 }
5428
5429 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5430 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5431 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5432 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5433 uint size = alloc->req();
5434 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5435 old_jvms->set_map(sfpt);
5436 for (uint i = 0; i < size; i++) {
5437 sfpt->init_req(i, alloc->in(i));
5438 }
5439 // re-push array length for deoptimization
5440 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5441 old_jvms->set_sp(old_jvms->sp()+1);
5442 old_jvms->set_monoff(old_jvms->monoff()+1);
5443 old_jvms->set_scloff(old_jvms->scloff()+1);
5444 old_jvms->set_endoff(old_jvms->endoff()+1);
5445 old_jvms->set_should_reexecute(true);
5446
5447 sfpt->set_i_o(map()->i_o());
5448 sfpt->set_memory(map()->memory());
5449 sfpt->set_control(map()->control());
5450 return sfpt;
5451 }
5452
5453 // In case of a deoptimization, we restart execution at the
5454 // allocation, allocating a new array. We would leave an uninitialized
5455 // array in the heap that GCs wouldn't expect. Move the allocation
5456 // after the traps so we don't allocate the array if we
5457 // deoptimize. This is possible because tightly_coupled_allocation()
5458 // guarantees there's no observer of the allocated array at this point
5459 // and the control flow is simple enough.
5460 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5461 int saved_reexecute_sp, uint new_idx) {
5462 if (saved_jvms_before_guards != nullptr && !stopped()) {
5463 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5464
5465 assert(alloc != nullptr, "only with a tightly coupled allocation");
5466 // restore JVM state to the state at the arraycopy
5467 saved_jvms_before_guards->map()->set_control(map()->control());
5468 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5469 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5470 // If we've improved the types of some nodes (null check) while
5471 // emitting the guards, propagate them to the current state
5472 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5473 set_jvms(saved_jvms_before_guards);
5474 _reexecute_sp = saved_reexecute_sp;
5475
5476 // Remove the allocation from above the guards
5477 CallProjections callprojs;
5478 alloc->extract_projections(&callprojs, true);
5479 InitializeNode* init = alloc->initialization();
5480 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5481 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5482 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5483
5484 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5485 // the allocation (i.e. is only valid if the allocation succeeds):
5486 // 1) replace CastIINode with AllocateArrayNode's length here
5487 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5488 //
5489 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5490 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5491 Node* init_control = init->proj_out(TypeFunc::Control);
5492 Node* alloc_length = alloc->Ideal_length();
5493 #ifdef ASSERT
5494 Node* prev_cast = nullptr;
5495 #endif
5496 for (uint i = 0; i < init_control->outcnt(); i++) {
5497 Node* init_out = init_control->raw_out(i);
5498 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5499 #ifdef ASSERT
5500 if (prev_cast == nullptr) {
5501 prev_cast = init_out;
5502 } else {
5503 if (prev_cast->cmp(*init_out) == false) {
5504 prev_cast->dump();
5505 init_out->dump();
5506 assert(false, "not equal CastIINode");
5507 }
5508 }
5509 #endif
5510 C->gvn_replace_by(init_out, alloc_length);
5511 }
5512 }
5513 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5514
5515 // move the allocation here (after the guards)
5516 _gvn.hash_delete(alloc);
5517 alloc->set_req(TypeFunc::Control, control());
5518 alloc->set_req(TypeFunc::I_O, i_o());
5519 Node *mem = reset_memory();
5520 set_all_memory(mem);
5521 alloc->set_req(TypeFunc::Memory, mem);
5522 set_control(init->proj_out_or_null(TypeFunc::Control));
5523 set_i_o(callprojs.fallthrough_ioproj);
5524
5525 // Update memory as done in GraphKit::set_output_for_allocation()
5526 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5527 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5528 if (ary_type->isa_aryptr() && length_type != nullptr) {
5529 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5530 }
5531 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5532 int elemidx = C->get_alias_index(telemref);
5533 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5534 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5535
5536 Node* allocx = _gvn.transform(alloc);
5537 assert(allocx == alloc, "where has the allocation gone?");
5538 assert(dest->is_CheckCastPP(), "not an allocation result?");
5539
5540 _gvn.hash_delete(dest);
5541 dest->set_req(0, control());
5542 Node* destx = _gvn.transform(dest);
5543 assert(destx == dest, "where has the allocation result gone?");
5544
5545 array_ideal_length(alloc, ary_type, true);
5546 }
5547 }
5548
5549 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5550 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5551 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5552 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5553 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5554 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5555 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5556 JVMState* saved_jvms_before_guards) {
5557 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5558 // There is at least one unrelated uncommon trap which needs to be replaced.
5559 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5560
5561 JVMState* saved_jvms = jvms();
5562 const int saved_reexecute_sp = _reexecute_sp;
5563 set_jvms(sfpt->jvms());
5564 _reexecute_sp = jvms()->sp();
5565
5566 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5567
5568 // Restore state
5569 set_jvms(saved_jvms);
5570 _reexecute_sp = saved_reexecute_sp;
5571 }
5572 }
5573
5574 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5575 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5576 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5577 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5578 while (if_proj->is_IfProj()) {
5579 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5580 if (uncommon_trap != nullptr) {
5581 create_new_uncommon_trap(uncommon_trap);
5582 }
5583 assert(if_proj->in(0)->is_If(), "must be If");
5584 if_proj = if_proj->in(0)->in(0);
5585 }
5586 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5587 "must have reached control projection of init node");
5588 }
5589
5590 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5591 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5592 assert(trap_request != 0, "no valid UCT trap request");
5593 PreserveJVMState pjvms(this);
5594 set_control(uncommon_trap_call->in(0));
5595 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5596 Deoptimization::trap_request_action(trap_request));
5597 assert(stopped(), "Should be stopped");
5598 _gvn.hash_delete(uncommon_trap_call);
5599 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5600 }
5601
5602 // Common checks for array sorting intrinsics arguments.
5603 // Returns `true` if checks passed.
5604 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
5605 // check address of the class
5606 if (elementType == nullptr || elementType->is_top()) {
5607 return false; // dead path
5608 }
5609 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5610 if (elem_klass == nullptr) {
5611 return false; // dead path
5612 }
5613 // java_mirror_type() returns non-null for compile-time Class constants only
5614 ciType* elem_type = elem_klass->java_mirror_type();
5615 if (elem_type == nullptr) {
5616 return false;
5617 }
5618 bt = elem_type->basic_type();
5619 // Disable the intrinsic if the CPU does not support SIMD sort
5620 if (!Matcher::supports_simd_sort(bt)) {
5621 return false;
5622 }
5623 // check address of the array
5624 if (obj == nullptr || obj->is_top()) {
5625 return false; // dead path
5626 }
5627 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5628 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
5629 return false; // failed input validation
5630 }
5631 return true;
5632 }
5633
5634 //------------------------------inline_array_partition-----------------------
5635 bool LibraryCallKit::inline_array_partition() {
5636 address stubAddr = StubRoutines::select_array_partition_function();
5637 if (stubAddr == nullptr) {
5638 return false; // Intrinsic's stub is not implemented on this platform
5639 }
5640 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
5641
5642 // no receiver because it is a static method
5643 Node* elementType = argument(0);
5644 Node* obj = argument(1);
5645 Node* offset = argument(2); // long
5646 Node* fromIndex = argument(4);
5647 Node* toIndex = argument(5);
5648 Node* indexPivot1 = argument(6);
5649 Node* indexPivot2 = argument(7);
5650 // PartitionOperation: argument(8) is ignored
5651
5652 Node* pivotIndices = nullptr;
5653 BasicType bt = T_ILLEGAL;
5654
5655 if (!check_array_sort_arguments(elementType, obj, bt)) {
5656 return false;
5657 }
5658 null_check(obj);
5659 // If obj is dead, only null-path is taken.
5660 if (stopped()) {
5661 return true;
5662 }
5663 // Set the original stack and the reexecute bit for the interpreter to reexecute
5664 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
5665 { PreserveReexecuteState preexecs(this);
5666 jvms()->set_should_reexecute(true);
5667
5668 Node* obj_adr = make_unsafe_address(obj, offset);
5669
5670 // create the pivotIndices array of type int and size = 2
5671 Node* size = intcon(2);
5672 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
5673 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
5674 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
5675 guarantee(alloc != nullptr, "created above");
5676 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
5677
5678 // pass the basic type enum to the stub
5679 Node* elemType = intcon(bt);
5680
5681 // Call the stub
5682 const char *stubName = "array_partition_stub";
5683 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
5684 stubAddr, stubName, TypePtr::BOTTOM,
5685 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
5686 indexPivot1, indexPivot2);
5687
5688 } // original reexecute is set back here
5689
5690 if (!stopped()) {
5691 set_result(pivotIndices);
5692 }
5693
5694 return true;
5695 }
5696
5697
5698 //------------------------------inline_array_sort-----------------------
5699 bool LibraryCallKit::inline_array_sort() {
5700 address stubAddr = StubRoutines::select_arraysort_function();
5701 if (stubAddr == nullptr) {
5702 return false; // Intrinsic's stub is not implemented on this platform
5703 }
5704 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
5705
5706 // no receiver because it is a static method
5707 Node* elementType = argument(0);
5708 Node* obj = argument(1);
5709 Node* offset = argument(2); // long
5710 Node* fromIndex = argument(4);
5711 Node* toIndex = argument(5);
5712 // SortOperation: argument(6) is ignored
5713
5714 BasicType bt = T_ILLEGAL;
5715
5716 if (!check_array_sort_arguments(elementType, obj, bt)) {
5717 return false;
5718 }
5719 null_check(obj);
5720 // If obj is dead, only null-path is taken.
5721 if (stopped()) {
5722 return true;
5723 }
5724 Node* obj_adr = make_unsafe_address(obj, offset);
5725
5726 // pass the basic type enum to the stub
5727 Node* elemType = intcon(bt);
5728
5729 // Call the stub.
5730 const char *stubName = "arraysort_stub";
5731 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
5732 stubAddr, stubName, TypePtr::BOTTOM,
5733 obj_adr, elemType, fromIndex, toIndex);
5734
5735 return true;
5736 }
5737
5738
5739 //------------------------------inline_arraycopy-----------------------
5740 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5741 // Object dest, int destPos,
5742 // int length);
5743 bool LibraryCallKit::inline_arraycopy() {
5744 // Get the arguments.
5745 Node* src = argument(0); // type: oop
5746 Node* src_offset = argument(1); // type: int
5747 Node* dest = argument(2); // type: oop
5748 Node* dest_offset = argument(3); // type: int
5749 Node* length = argument(4); // type: int
5750
5751 uint new_idx = C->unique();
5752
5753 // Check for allocation before we add nodes that would confuse
5754 // tightly_coupled_allocation()
5755 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5756
5757 int saved_reexecute_sp = -1;
5758 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5759 // See arraycopy_restore_alloc_state() comment
5760 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5761 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5762 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5763 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5764
5765 // The following tests must be performed
5766 // (1) src and dest are arrays.
5767 // (2) src and dest arrays must have elements of the same BasicType
5768 // (3) src and dest must not be null.
5769 // (4) src_offset must not be negative.
5770 // (5) dest_offset must not be negative.
5771 // (6) length must not be negative.
5772 // (7) src_offset + length must not exceed length of src.
5773 // (8) dest_offset + length must not exceed length of dest.
5774 // (9) each element of an oop array must be assignable
5775
5776 // (3) src and dest must not be null.
5777 // always do this here because we need the JVM state for uncommon traps
5778 Node* null_ctl = top();
5779 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5780 assert(null_ctl->is_top(), "no null control here");
5781 dest = null_check(dest, T_ARRAY);
5782
5783 if (!can_emit_guards) {
5784 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5785 // guards but the arraycopy node could still take advantage of a
5786 // tightly allocated allocation. tightly_coupled_allocation() is
5787 // called again to make sure it takes the null check above into
5788 // account: the null check is mandatory and if it caused an
5789 // uncommon trap to be emitted then the allocation can't be
5790 // considered tightly coupled in this context.
5791 alloc = tightly_coupled_allocation(dest);
5792 }
5793
5794 bool validated = false;
5795
5796 const Type* src_type = _gvn.type(src);
5797 const Type* dest_type = _gvn.type(dest);
5798 const TypeAryPtr* top_src = src_type->isa_aryptr();
5799 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5800
5801 // Do we have the type of src?
5802 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5803 // Do we have the type of dest?
5804 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5805 // Is the type for src from speculation?
5806 bool src_spec = false;
5807 // Is the type for dest from speculation?
5808 bool dest_spec = false;
5809
5810 if ((!has_src || !has_dest) && can_emit_guards) {
5811 // We don't have sufficient type information, let's see if
5812 // speculative types can help. We need to have types for both src
5813 // and dest so that it pays off.
5814
5815 // Do we already have or could we have type information for src
5816 bool could_have_src = has_src;
5817 // Do we already have or could we have type information for dest
5818 bool could_have_dest = has_dest;
5819
5820 ciKlass* src_k = nullptr;
5821 if (!has_src) {
5822 src_k = src_type->speculative_type_not_null();
5823 if (src_k != nullptr && src_k->is_array_klass()) {
5824 could_have_src = true;
5825 }
5826 }
5827
5828 ciKlass* dest_k = nullptr;
5829 if (!has_dest) {
5830 dest_k = dest_type->speculative_type_not_null();
5831 if (dest_k != nullptr && dest_k->is_array_klass()) {
5832 could_have_dest = true;
5833 }
5834 }
5835
5836 if (could_have_src && could_have_dest) {
5837 // This is going to pay off so emit the required guards
5838 if (!has_src) {
5839 src = maybe_cast_profiled_obj(src, src_k, true);
5840 src_type = _gvn.type(src);
5841 top_src = src_type->isa_aryptr();
5842 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5843 src_spec = true;
5844 }
5845 if (!has_dest) {
5846 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5847 dest_type = _gvn.type(dest);
5848 top_dest = dest_type->isa_aryptr();
5849 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5850 dest_spec = true;
5851 }
5852 }
5853 }
5854
5855 if (has_src && has_dest && can_emit_guards) {
5856 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5857 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5858 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5859 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5860
5861 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5862 // If both arrays are object arrays then having the exact types
5863 // for both will remove the need for a subtype check at runtime
5864 // before the call and may make it possible to pick a faster copy
5865 // routine (without a subtype check on every element)
5866 // Do we have the exact type of src?
5867 bool could_have_src = src_spec;
5868 // Do we have the exact type of dest?
5869 bool could_have_dest = dest_spec;
5870 ciKlass* src_k = nullptr;
5871 ciKlass* dest_k = nullptr;
5872 if (!src_spec) {
5873 src_k = src_type->speculative_type_not_null();
5874 if (src_k != nullptr && src_k->is_array_klass()) {
5875 could_have_src = true;
5876 }
5877 }
5878 if (!dest_spec) {
5879 dest_k = dest_type->speculative_type_not_null();
5880 if (dest_k != nullptr && dest_k->is_array_klass()) {
5881 could_have_dest = true;
5882 }
5883 }
5884 if (could_have_src && could_have_dest) {
5885 // If we can have both exact types, emit the missing guards
5886 if (could_have_src && !src_spec) {
5887 src = maybe_cast_profiled_obj(src, src_k, true);
5888 }
5889 if (could_have_dest && !dest_spec) {
5890 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5891 }
5892 }
5893 }
5894 }
5895
5896 ciMethod* trap_method = method();
5897 int trap_bci = bci();
5898 if (saved_jvms_before_guards != nullptr) {
5899 trap_method = alloc->jvms()->method();
5900 trap_bci = alloc->jvms()->bci();
5901 }
5902
5903 bool negative_length_guard_generated = false;
5904
5905 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5906 can_emit_guards &&
5907 !src->is_top() && !dest->is_top()) {
5908 // validate arguments: enables transformation the ArrayCopyNode
5909 validated = true;
5910
5911 RegionNode* slow_region = new RegionNode(1);
5912 record_for_igvn(slow_region);
5913
5914 // (1) src and dest are arrays.
5915 generate_non_array_guard(load_object_klass(src), slow_region);
5916 generate_non_array_guard(load_object_klass(dest), slow_region);
5917
5918 // (2) src and dest arrays must have elements of the same BasicType
5919 // done at macro expansion or at Ideal transformation time
5920
5921 // (4) src_offset must not be negative.
5922 generate_negative_guard(src_offset, slow_region);
5923
5924 // (5) dest_offset must not be negative.
5925 generate_negative_guard(dest_offset, slow_region);
5926
5927 // (7) src_offset + length must not exceed length of src.
5928 generate_limit_guard(src_offset, length,
5929 load_array_length(src),
5930 slow_region);
5931
5932 // (8) dest_offset + length must not exceed length of dest.
5933 generate_limit_guard(dest_offset, length,
5934 load_array_length(dest),
5935 slow_region);
5936
5937 // (6) length must not be negative.
5938 // This is also checked in generate_arraycopy() during macro expansion, but
5939 // we also have to check it here for the case where the ArrayCopyNode will
5940 // be eliminated by Escape Analysis.
5941 if (EliminateAllocations) {
5942 generate_negative_guard(length, slow_region);
5943 negative_length_guard_generated = true;
5944 }
5945
5946 // (9) each element of an oop array must be assignable
5947 Node* dest_klass = load_object_klass(dest);
5948 if (src != dest) {
5949 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
5950
5951 if (not_subtype_ctrl != top()) {
5952 PreserveJVMState pjvms(this);
5953 set_control(not_subtype_ctrl);
5954 uncommon_trap(Deoptimization::Reason_intrinsic,
5955 Deoptimization::Action_make_not_entrant);
5956 assert(stopped(), "Should be stopped");
5957 }
5958 }
5959 {
5960 PreserveJVMState pjvms(this);
5961 set_control(_gvn.transform(slow_region));
5962 uncommon_trap(Deoptimization::Reason_intrinsic,
5963 Deoptimization::Action_make_not_entrant);
5964 assert(stopped(), "Should be stopped");
5965 }
5966
5967 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5968 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
5969 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5970 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
5971 }
5972
5973 if (stopped()) {
5974 return true;
5975 }
5976
5977 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
5978 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5979 // so the compiler has a chance to eliminate them: during macro expansion,
5980 // we have to set their control (CastPP nodes are eliminated).
5981 load_object_klass(src), load_object_klass(dest),
5982 load_array_length(src), load_array_length(dest));
5983
5984 ac->set_arraycopy(validated);
5985
5986 Node* n = _gvn.transform(ac);
5987 if (n == ac) {
5988 ac->connect_outputs(this);
5989 } else {
5990 assert(validated, "shouldn't transform if all arguments not validated");
5991 set_all_memory(n);
5992 }
5993 clear_upper_avx();
5994
5995
5996 return true;
5997 }
5998
5999
6000 // Helper function which determines if an arraycopy immediately follows
6001 // an allocation, with no intervening tests or other escapes for the object.
6002 AllocateArrayNode*
6003 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6004 if (stopped()) return nullptr; // no fast path
6005 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6006
6007 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6008 if (alloc == nullptr) return nullptr;
6009
6010 Node* rawmem = memory(Compile::AliasIdxRaw);
6011 // Is the allocation's memory state untouched?
6012 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6013 // Bail out if there have been raw-memory effects since the allocation.
6014 // (Example: There might have been a call or safepoint.)
6015 return nullptr;
6016 }
6017 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6018 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6019 return nullptr;
6020 }
6021
6022 // There must be no unexpected observers of this allocation.
6023 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6024 Node* obs = ptr->fast_out(i);
6025 if (obs != this->map()) {
6026 return nullptr;
6027 }
6028 }
6029
6030 // This arraycopy must unconditionally follow the allocation of the ptr.
6031 Node* alloc_ctl = ptr->in(0);
6032 Node* ctl = control();
6033 while (ctl != alloc_ctl) {
6034 // There may be guards which feed into the slow_region.
6035 // Any other control flow means that we might not get a chance
6036 // to finish initializing the allocated object.
6037 // Various low-level checks bottom out in uncommon traps. These
6038 // are considered safe since we've already checked above that
6039 // there is no unexpected observer of this allocation.
6040 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6041 assert(ctl->in(0)->is_If(), "must be If");
6042 ctl = ctl->in(0)->in(0);
6043 } else {
6044 return nullptr;
6045 }
6046 }
6047
6048 // If we get this far, we have an allocation which immediately
6049 // precedes the arraycopy, and we can take over zeroing the new object.
6050 // The arraycopy will finish the initialization, and provide
6051 // a new control state to which we will anchor the destination pointer.
6052
6053 return alloc;
6054 }
6055
6056 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6057 if (node->is_IfProj()) {
6058 Node* other_proj = node->as_IfProj()->other_if_proj();
6059 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6060 Node* obs = other_proj->fast_out(j);
6061 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6062 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6063 return obs->as_CallStaticJava();
6064 }
6065 }
6066 }
6067 return nullptr;
6068 }
6069
6070 //-------------inline_encodeISOArray-----------------------------------
6071 // encode char[] to byte[] in ISO_8859_1 or ASCII
6072 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6073 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6074 // no receiver since it is static method
6075 Node *src = argument(0);
6076 Node *src_offset = argument(1);
6077 Node *dst = argument(2);
6078 Node *dst_offset = argument(3);
6079 Node *length = argument(4);
6080
6081 src = must_be_not_null(src, true);
6082 dst = must_be_not_null(dst, true);
6083
6084 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6085 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6086 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6087 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6088 // failed array check
6089 return false;
6090 }
6091
6092 // Figure out the size and type of the elements we will be copying.
6093 BasicType src_elem = src_type->elem()->array_element_basic_type();
6094 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6095 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6096 return false;
6097 }
6098
6099 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6100 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6101 // 'src_start' points to src array + scaled offset
6102 // 'dst_start' points to dst array + scaled offset
6103
6104 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6105 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6106 enc = _gvn.transform(enc);
6107 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6108 set_memory(res_mem, mtype);
6109 set_result(enc);
6110 clear_upper_avx();
6111
6112 return true;
6113 }
6114
6115 //-------------inline_multiplyToLen-----------------------------------
6116 bool LibraryCallKit::inline_multiplyToLen() {
6117 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6118
6119 address stubAddr = StubRoutines::multiplyToLen();
6120 if (stubAddr == nullptr) {
6121 return false; // Intrinsic's stub is not implemented on this platform
6122 }
6123 const char* stubName = "multiplyToLen";
6124
6125 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6126
6127 // no receiver because it is a static method
6128 Node* x = argument(0);
6129 Node* xlen = argument(1);
6130 Node* y = argument(2);
6131 Node* ylen = argument(3);
6132 Node* z = argument(4);
6133
6134 x = must_be_not_null(x, true);
6135 y = must_be_not_null(y, true);
6136
6137 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6138 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6139 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6140 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6141 // failed array check
6142 return false;
6143 }
6144
6145 BasicType x_elem = x_type->elem()->array_element_basic_type();
6146 BasicType y_elem = y_type->elem()->array_element_basic_type();
6147 if (x_elem != T_INT || y_elem != T_INT) {
6148 return false;
6149 }
6150
6151 Node* x_start = array_element_address(x, intcon(0), x_elem);
6152 Node* y_start = array_element_address(y, intcon(0), y_elem);
6153 // 'x_start' points to x array + scaled xlen
6154 // 'y_start' points to y array + scaled ylen
6155
6156 Node* z_start = array_element_address(z, intcon(0), T_INT);
6157
6158 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6159 OptoRuntime::multiplyToLen_Type(),
6160 stubAddr, stubName, TypePtr::BOTTOM,
6161 x_start, xlen, y_start, ylen, z_start);
6162
6163 C->set_has_split_ifs(true); // Has chance for split-if optimization
6164 set_result(z);
6165 return true;
6166 }
6167
6168 //-------------inline_squareToLen------------------------------------
6169 bool LibraryCallKit::inline_squareToLen() {
6170 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6171
6172 address stubAddr = StubRoutines::squareToLen();
6173 if (stubAddr == nullptr) {
6174 return false; // Intrinsic's stub is not implemented on this platform
6175 }
6176 const char* stubName = "squareToLen";
6177
6178 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6179
6180 Node* x = argument(0);
6181 Node* len = argument(1);
6182 Node* z = argument(2);
6183 Node* zlen = argument(3);
6184
6185 x = must_be_not_null(x, true);
6186 z = must_be_not_null(z, true);
6187
6188 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6189 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6190 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6191 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6192 // failed array check
6193 return false;
6194 }
6195
6196 BasicType x_elem = x_type->elem()->array_element_basic_type();
6197 BasicType z_elem = z_type->elem()->array_element_basic_type();
6198 if (x_elem != T_INT || z_elem != T_INT) {
6199 return false;
6200 }
6201
6202
6203 Node* x_start = array_element_address(x, intcon(0), x_elem);
6204 Node* z_start = array_element_address(z, intcon(0), z_elem);
6205
6206 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6207 OptoRuntime::squareToLen_Type(),
6208 stubAddr, stubName, TypePtr::BOTTOM,
6209 x_start, len, z_start, zlen);
6210
6211 set_result(z);
6212 return true;
6213 }
6214
6215 //-------------inline_mulAdd------------------------------------------
6216 bool LibraryCallKit::inline_mulAdd() {
6217 assert(UseMulAddIntrinsic, "not implemented on this platform");
6218
6219 address stubAddr = StubRoutines::mulAdd();
6220 if (stubAddr == nullptr) {
6221 return false; // Intrinsic's stub is not implemented on this platform
6222 }
6223 const char* stubName = "mulAdd";
6224
6225 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6226
6227 Node* out = argument(0);
6228 Node* in = argument(1);
6229 Node* offset = argument(2);
6230 Node* len = argument(3);
6231 Node* k = argument(4);
6232
6233 in = must_be_not_null(in, true);
6234 out = must_be_not_null(out, true);
6235
6236 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6237 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6238 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6239 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6240 // failed array check
6241 return false;
6242 }
6243
6244 BasicType out_elem = out_type->elem()->array_element_basic_type();
6245 BasicType in_elem = in_type->elem()->array_element_basic_type();
6246 if (out_elem != T_INT || in_elem != T_INT) {
6247 return false;
6248 }
6249
6250 Node* outlen = load_array_length(out);
6251 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6252 Node* out_start = array_element_address(out, intcon(0), out_elem);
6253 Node* in_start = array_element_address(in, intcon(0), in_elem);
6254
6255 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6256 OptoRuntime::mulAdd_Type(),
6257 stubAddr, stubName, TypePtr::BOTTOM,
6258 out_start,in_start, new_offset, len, k);
6259 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6260 set_result(result);
6261 return true;
6262 }
6263
6264 //-------------inline_montgomeryMultiply-----------------------------------
6265 bool LibraryCallKit::inline_montgomeryMultiply() {
6266 address stubAddr = StubRoutines::montgomeryMultiply();
6267 if (stubAddr == nullptr) {
6268 return false; // Intrinsic's stub is not implemented on this platform
6269 }
6270
6271 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6272 const char* stubName = "montgomery_multiply";
6273
6274 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6275
6276 Node* a = argument(0);
6277 Node* b = argument(1);
6278 Node* n = argument(2);
6279 Node* len = argument(3);
6280 Node* inv = argument(4);
6281 Node* m = argument(6);
6282
6283 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6284 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6285 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6286 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6287 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6288 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6289 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6290 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6291 // failed array check
6292 return false;
6293 }
6294
6295 BasicType a_elem = a_type->elem()->array_element_basic_type();
6296 BasicType b_elem = b_type->elem()->array_element_basic_type();
6297 BasicType n_elem = n_type->elem()->array_element_basic_type();
6298 BasicType m_elem = m_type->elem()->array_element_basic_type();
6299 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6300 return false;
6301 }
6302
6303 // Make the call
6304 {
6305 Node* a_start = array_element_address(a, intcon(0), a_elem);
6306 Node* b_start = array_element_address(b, intcon(0), b_elem);
6307 Node* n_start = array_element_address(n, intcon(0), n_elem);
6308 Node* m_start = array_element_address(m, intcon(0), m_elem);
6309
6310 Node* call = make_runtime_call(RC_LEAF,
6311 OptoRuntime::montgomeryMultiply_Type(),
6312 stubAddr, stubName, TypePtr::BOTTOM,
6313 a_start, b_start, n_start, len, inv, top(),
6314 m_start);
6315 set_result(m);
6316 }
6317
6318 return true;
6319 }
6320
6321 bool LibraryCallKit::inline_montgomerySquare() {
6322 address stubAddr = StubRoutines::montgomerySquare();
6323 if (stubAddr == nullptr) {
6324 return false; // Intrinsic's stub is not implemented on this platform
6325 }
6326
6327 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6328 const char* stubName = "montgomery_square";
6329
6330 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6331
6332 Node* a = argument(0);
6333 Node* n = argument(1);
6334 Node* len = argument(2);
6335 Node* inv = argument(3);
6336 Node* m = argument(5);
6337
6338 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6339 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6340 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6341 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6342 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6343 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6344 // failed array check
6345 return false;
6346 }
6347
6348 BasicType a_elem = a_type->elem()->array_element_basic_type();
6349 BasicType n_elem = n_type->elem()->array_element_basic_type();
6350 BasicType m_elem = m_type->elem()->array_element_basic_type();
6351 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6352 return false;
6353 }
6354
6355 // Make the call
6356 {
6357 Node* a_start = array_element_address(a, intcon(0), a_elem);
6358 Node* n_start = array_element_address(n, intcon(0), n_elem);
6359 Node* m_start = array_element_address(m, intcon(0), m_elem);
6360
6361 Node* call = make_runtime_call(RC_LEAF,
6362 OptoRuntime::montgomerySquare_Type(),
6363 stubAddr, stubName, TypePtr::BOTTOM,
6364 a_start, n_start, len, inv, top(),
6365 m_start);
6366 set_result(m);
6367 }
6368
6369 return true;
6370 }
6371
6372 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6373 address stubAddr = nullptr;
6374 const char* stubName = nullptr;
6375
6376 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6377 if (stubAddr == nullptr) {
6378 return false; // Intrinsic's stub is not implemented on this platform
6379 }
6380
6381 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6382
6383 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6384
6385 Node* newArr = argument(0);
6386 Node* oldArr = argument(1);
6387 Node* newIdx = argument(2);
6388 Node* shiftCount = argument(3);
6389 Node* numIter = argument(4);
6390
6391 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6392 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6393 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6394 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6395 return false;
6396 }
6397
6398 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6399 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6400 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6401 return false;
6402 }
6403
6404 // Make the call
6405 {
6406 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6407 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6408
6409 Node* call = make_runtime_call(RC_LEAF,
6410 OptoRuntime::bigIntegerShift_Type(),
6411 stubAddr,
6412 stubName,
6413 TypePtr::BOTTOM,
6414 newArr_start,
6415 oldArr_start,
6416 newIdx,
6417 shiftCount,
6418 numIter);
6419 }
6420
6421 return true;
6422 }
6423
6424 //-------------inline_vectorizedMismatch------------------------------
6425 bool LibraryCallKit::inline_vectorizedMismatch() {
6426 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6427
6428 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6429 Node* obja = argument(0); // Object
6430 Node* aoffset = argument(1); // long
6431 Node* objb = argument(3); // Object
6432 Node* boffset = argument(4); // long
6433 Node* length = argument(6); // int
6434 Node* scale = argument(7); // int
6435
6436 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6437 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6438 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6439 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6440 scale == top()) {
6441 return false; // failed input validation
6442 }
6443
6444 Node* obja_adr = make_unsafe_address(obja, aoffset);
6445 Node* objb_adr = make_unsafe_address(objb, boffset);
6446
6447 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6448 //
6449 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6450 // if (length <= inline_limit) {
6451 // inline_path:
6452 // vmask = VectorMaskGen length
6453 // vload1 = LoadVectorMasked obja, vmask
6454 // vload2 = LoadVectorMasked objb, vmask
6455 // result1 = VectorCmpMasked vload1, vload2, vmask
6456 // } else {
6457 // call_stub_path:
6458 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6459 // }
6460 // exit_block:
6461 // return Phi(result1, result2);
6462 //
6463 enum { inline_path = 1, // input is small enough to process it all at once
6464 stub_path = 2, // input is too large; call into the VM
6465 PATH_LIMIT = 3
6466 };
6467
6468 Node* exit_block = new RegionNode(PATH_LIMIT);
6469 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6470 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6471
6472 Node* call_stub_path = control();
6473
6474 BasicType elem_bt = T_ILLEGAL;
6475
6476 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6477 if (scale_t->is_con()) {
6478 switch (scale_t->get_con()) {
6479 case 0: elem_bt = T_BYTE; break;
6480 case 1: elem_bt = T_SHORT; break;
6481 case 2: elem_bt = T_INT; break;
6482 case 3: elem_bt = T_LONG; break;
6483
6484 default: elem_bt = T_ILLEGAL; break; // not supported
6485 }
6486 }
6487
6488 int inline_limit = 0;
6489 bool do_partial_inline = false;
6490
6491 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6492 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6493 do_partial_inline = inline_limit >= 16;
6494 }
6495
6496 if (do_partial_inline) {
6497 assert(elem_bt != T_ILLEGAL, "sanity");
6498
6499 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6500 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6501 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6502
6503 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6504 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6505 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6506
6507 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6508
6509 if (!stopped()) {
6510 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6511
6512 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6513 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6514 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6515 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6516
6517 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6518 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6519 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6520 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6521
6522 exit_block->init_req(inline_path, control());
6523 memory_phi->init_req(inline_path, map()->memory());
6524 result_phi->init_req(inline_path, result);
6525
6526 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6527 clear_upper_avx();
6528 }
6529 }
6530 }
6531
6532 if (call_stub_path != nullptr) {
6533 set_control(call_stub_path);
6534
6535 Node* call = make_runtime_call(RC_LEAF,
6536 OptoRuntime::vectorizedMismatch_Type(),
6537 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6538 obja_adr, objb_adr, length, scale);
6539
6540 exit_block->init_req(stub_path, control());
6541 memory_phi->init_req(stub_path, map()->memory());
6542 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6543 }
6544
6545 exit_block = _gvn.transform(exit_block);
6546 memory_phi = _gvn.transform(memory_phi);
6547 result_phi = _gvn.transform(result_phi);
6548
6549 set_control(exit_block);
6550 set_all_memory(memory_phi);
6551 set_result(result_phi);
6552
6553 return true;
6554 }
6555
6556 //------------------------------inline_vectorizedHashcode----------------------------
6557 bool LibraryCallKit::inline_vectorizedHashCode() {
6558 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6559
6560 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6561 Node* array = argument(0);
6562 Node* offset = argument(1);
6563 Node* length = argument(2);
6564 Node* initialValue = argument(3);
6565 Node* basic_type = argument(4);
6566
6567 if (basic_type == top()) {
6568 return false; // failed input validation
6569 }
6570
6571 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6572 if (!basic_type_t->is_con()) {
6573 return false; // Only intrinsify if mode argument is constant
6574 }
6575
6576 array = must_be_not_null(array, true);
6577
6578 BasicType bt = (BasicType)basic_type_t->get_con();
6579
6580 // Resolve address of first element
6581 Node* array_start = array_element_address(array, offset, bt);
6582
6583 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6584 array_start, length, initialValue, basic_type)));
6585 clear_upper_avx();
6586
6587 return true;
6588 }
6589
6590 /**
6591 * Calculate CRC32 for byte.
6592 * int java.util.zip.CRC32.update(int crc, int b)
6593 */
6594 bool LibraryCallKit::inline_updateCRC32() {
6595 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6596 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6597 // no receiver since it is static method
6598 Node* crc = argument(0); // type: int
6599 Node* b = argument(1); // type: int
6600
6601 /*
6602 * int c = ~ crc;
6603 * b = timesXtoThe32[(b ^ c) & 0xFF];
6604 * b = b ^ (c >>> 8);
6605 * crc = ~b;
6606 */
6607
6608 Node* M1 = intcon(-1);
6609 crc = _gvn.transform(new XorINode(crc, M1));
6610 Node* result = _gvn.transform(new XorINode(crc, b));
6611 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6612
6613 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6614 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6615 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6616 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6617
6618 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6619 result = _gvn.transform(new XorINode(crc, result));
6620 result = _gvn.transform(new XorINode(result, M1));
6621 set_result(result);
6622 return true;
6623 }
6624
6625 /**
6626 * Calculate CRC32 for byte[] array.
6627 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6628 */
6629 bool LibraryCallKit::inline_updateBytesCRC32() {
6630 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6631 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6632 // no receiver since it is static method
6633 Node* crc = argument(0); // type: int
6634 Node* src = argument(1); // type: oop
6635 Node* offset = argument(2); // type: int
6636 Node* length = argument(3); // type: int
6637
6638 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6639 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6640 // failed array check
6641 return false;
6642 }
6643
6644 // Figure out the size and type of the elements we will be copying.
6645 BasicType src_elem = src_type->elem()->array_element_basic_type();
6646 if (src_elem != T_BYTE) {
6647 return false;
6648 }
6649
6650 // 'src_start' points to src array + scaled offset
6651 src = must_be_not_null(src, true);
6652 Node* src_start = array_element_address(src, offset, src_elem);
6653
6654 // We assume that range check is done by caller.
6655 // TODO: generate range check (offset+length < src.length) in debug VM.
6656
6657 // Call the stub.
6658 address stubAddr = StubRoutines::updateBytesCRC32();
6659 const char *stubName = "updateBytesCRC32";
6660
6661 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6662 stubAddr, stubName, TypePtr::BOTTOM,
6663 crc, src_start, length);
6664 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6665 set_result(result);
6666 return true;
6667 }
6668
6669 /**
6670 * Calculate CRC32 for ByteBuffer.
6671 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6672 */
6673 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6674 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6675 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6676 // no receiver since it is static method
6677 Node* crc = argument(0); // type: int
6678 Node* src = argument(1); // type: long
6679 Node* offset = argument(3); // type: int
6680 Node* length = argument(4); // type: int
6681
6682 src = ConvL2X(src); // adjust Java long to machine word
6683 Node* base = _gvn.transform(new CastX2PNode(src));
6684 offset = ConvI2X(offset);
6685
6686 // 'src_start' points to src array + scaled offset
6687 Node* src_start = basic_plus_adr(top(), base, offset);
6688
6689 // Call the stub.
6690 address stubAddr = StubRoutines::updateBytesCRC32();
6691 const char *stubName = "updateBytesCRC32";
6692
6693 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6694 stubAddr, stubName, TypePtr::BOTTOM,
6695 crc, src_start, length);
6696 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6697 set_result(result);
6698 return true;
6699 }
6700
6701 //------------------------------get_table_from_crc32c_class-----------------------
6702 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6703 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6704 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6705
6706 return table;
6707 }
6708
6709 //------------------------------inline_updateBytesCRC32C-----------------------
6710 //
6711 // Calculate CRC32C for byte[] array.
6712 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6713 //
6714 bool LibraryCallKit::inline_updateBytesCRC32C() {
6715 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6716 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6717 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6718 // no receiver since it is a static method
6719 Node* crc = argument(0); // type: int
6720 Node* src = argument(1); // type: oop
6721 Node* offset = argument(2); // type: int
6722 Node* end = argument(3); // type: int
6723
6724 Node* length = _gvn.transform(new SubINode(end, offset));
6725
6726 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6727 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6728 // failed array check
6729 return false;
6730 }
6731
6732 // Figure out the size and type of the elements we will be copying.
6733 BasicType src_elem = src_type->elem()->array_element_basic_type();
6734 if (src_elem != T_BYTE) {
6735 return false;
6736 }
6737
6738 // 'src_start' points to src array + scaled offset
6739 src = must_be_not_null(src, true);
6740 Node* src_start = array_element_address(src, offset, src_elem);
6741
6742 // static final int[] byteTable in class CRC32C
6743 Node* table = get_table_from_crc32c_class(callee()->holder());
6744 table = must_be_not_null(table, true);
6745 Node* table_start = array_element_address(table, intcon(0), T_INT);
6746
6747 // We assume that range check is done by caller.
6748 // TODO: generate range check (offset+length < src.length) in debug VM.
6749
6750 // Call the stub.
6751 address stubAddr = StubRoutines::updateBytesCRC32C();
6752 const char *stubName = "updateBytesCRC32C";
6753
6754 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6755 stubAddr, stubName, TypePtr::BOTTOM,
6756 crc, src_start, length, table_start);
6757 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6758 set_result(result);
6759 return true;
6760 }
6761
6762 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6763 //
6764 // Calculate CRC32C for DirectByteBuffer.
6765 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6766 //
6767 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6768 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6769 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6770 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6771 // no receiver since it is a static method
6772 Node* crc = argument(0); // type: int
6773 Node* src = argument(1); // type: long
6774 Node* offset = argument(3); // type: int
6775 Node* end = argument(4); // type: int
6776
6777 Node* length = _gvn.transform(new SubINode(end, offset));
6778
6779 src = ConvL2X(src); // adjust Java long to machine word
6780 Node* base = _gvn.transform(new CastX2PNode(src));
6781 offset = ConvI2X(offset);
6782
6783 // 'src_start' points to src array + scaled offset
6784 Node* src_start = basic_plus_adr(top(), base, offset);
6785
6786 // static final int[] byteTable in class CRC32C
6787 Node* table = get_table_from_crc32c_class(callee()->holder());
6788 table = must_be_not_null(table, true);
6789 Node* table_start = array_element_address(table, intcon(0), T_INT);
6790
6791 // Call the stub.
6792 address stubAddr = StubRoutines::updateBytesCRC32C();
6793 const char *stubName = "updateBytesCRC32C";
6794
6795 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6796 stubAddr, stubName, TypePtr::BOTTOM,
6797 crc, src_start, length, table_start);
6798 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6799 set_result(result);
6800 return true;
6801 }
6802
6803 //------------------------------inline_updateBytesAdler32----------------------
6804 //
6805 // Calculate Adler32 checksum for byte[] array.
6806 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6807 //
6808 bool LibraryCallKit::inline_updateBytesAdler32() {
6809 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6810 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6811 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6812 // no receiver since it is static method
6813 Node* crc = argument(0); // type: int
6814 Node* src = argument(1); // type: oop
6815 Node* offset = argument(2); // type: int
6816 Node* length = argument(3); // type: int
6817
6818 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6819 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6820 // failed array check
6821 return false;
6822 }
6823
6824 // Figure out the size and type of the elements we will be copying.
6825 BasicType src_elem = src_type->elem()->array_element_basic_type();
6826 if (src_elem != T_BYTE) {
6827 return false;
6828 }
6829
6830 // 'src_start' points to src array + scaled offset
6831 Node* src_start = array_element_address(src, offset, src_elem);
6832
6833 // We assume that range check is done by caller.
6834 // TODO: generate range check (offset+length < src.length) in debug VM.
6835
6836 // Call the stub.
6837 address stubAddr = StubRoutines::updateBytesAdler32();
6838 const char *stubName = "updateBytesAdler32";
6839
6840 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6841 stubAddr, stubName, TypePtr::BOTTOM,
6842 crc, src_start, length);
6843 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6844 set_result(result);
6845 return true;
6846 }
6847
6848 //------------------------------inline_updateByteBufferAdler32---------------
6849 //
6850 // Calculate Adler32 checksum for DirectByteBuffer.
6851 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6852 //
6853 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6854 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6855 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6856 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6857 // no receiver since it is static method
6858 Node* crc = argument(0); // type: int
6859 Node* src = argument(1); // type: long
6860 Node* offset = argument(3); // type: int
6861 Node* length = argument(4); // type: int
6862
6863 src = ConvL2X(src); // adjust Java long to machine word
6864 Node* base = _gvn.transform(new CastX2PNode(src));
6865 offset = ConvI2X(offset);
6866
6867 // 'src_start' points to src array + scaled offset
6868 Node* src_start = basic_plus_adr(top(), base, offset);
6869
6870 // Call the stub.
6871 address stubAddr = StubRoutines::updateBytesAdler32();
6872 const char *stubName = "updateBytesAdler32";
6873
6874 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6875 stubAddr, stubName, TypePtr::BOTTOM,
6876 crc, src_start, length);
6877
6878 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6879 set_result(result);
6880 return true;
6881 }
6882
6883 //----------------------------inline_reference_get----------------------------
6884 // public T java.lang.ref.Reference.get();
6885 bool LibraryCallKit::inline_reference_get() {
6886 const int referent_offset = java_lang_ref_Reference::referent_offset();
6887
6888 // Get the argument:
6889 Node* reference_obj = null_check_receiver();
6890 if (stopped()) return true;
6891
6892 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6893 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6894 decorators, /*is_static*/ false, nullptr);
6895 if (result == nullptr) return false;
6896
6897 // Add memory barrier to prevent commoning reads from this field
6898 // across safepoint since GC can change its value.
6899 insert_mem_bar(Op_MemBarCPUOrder);
6900
6901 set_result(result);
6902 return true;
6903 }
6904
6905 //----------------------------inline_reference_refersTo0----------------------------
6906 // bool java.lang.ref.Reference.refersTo0();
6907 // bool java.lang.ref.PhantomReference.refersTo0();
6908 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6909 // Get arguments:
6910 Node* reference_obj = null_check_receiver();
6911 Node* other_obj = argument(1);
6912 if (stopped()) return true;
6913
6914 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6915 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6916 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6917 decorators, /*is_static*/ false, nullptr);
6918 if (referent == nullptr) return false;
6919
6920 // Add memory barrier to prevent commoning reads from this field
6921 // across safepoint since GC can change its value.
6922 insert_mem_bar(Op_MemBarCPUOrder);
6923
6924 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
6925 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6926 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
6927
6928 RegionNode* region = new RegionNode(3);
6929 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
6930
6931 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
6932 region->init_req(1, if_true);
6933 phi->init_req(1, intcon(1));
6934
6935 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
6936 region->init_req(2, if_false);
6937 phi->init_req(2, intcon(0));
6938
6939 set_control(_gvn.transform(region));
6940 record_for_igvn(region);
6941 set_result(_gvn.transform(phi));
6942 return true;
6943 }
6944
6945 //----------------------------inline_reference_clear0----------------------------
6946 // void java.lang.ref.Reference.clear0();
6947 // void java.lang.ref.PhantomReference.clear0();
6948 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
6949 // This matches the implementation in JVM_ReferenceClear, see the comments there.
6950
6951 // Get arguments
6952 Node* reference_obj = null_check_receiver();
6953 if (stopped()) return true;
6954
6955 // Common access parameters
6956 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6957 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6958 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
6959 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
6960 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
6961
6962 Node* referent = access_load_at(reference_obj,
6963 referent_field_addr,
6964 referent_field_addr_type,
6965 val_type,
6966 T_OBJECT,
6967 decorators);
6968
6969 IdealKit ideal(this);
6970 #define __ ideal.
6971 __ if_then(referent, BoolTest::ne, null());
6972 sync_kit(ideal);
6973 access_store_at(reference_obj,
6974 referent_field_addr,
6975 referent_field_addr_type,
6976 null(),
6977 val_type,
6978 T_OBJECT,
6979 decorators);
6980 __ sync_kit(this);
6981 __ end_if();
6982 final_sync(ideal);
6983 #undef __
6984
6985 return true;
6986 }
6987
6988 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
6989 DecoratorSet decorators, bool is_static,
6990 ciInstanceKlass* fromKls) {
6991 if (fromKls == nullptr) {
6992 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6993 assert(tinst != nullptr, "obj is null");
6994 assert(tinst->is_loaded(), "obj is not loaded");
6995 fromKls = tinst->instance_klass();
6996 } else {
6997 assert(is_static, "only for static field access");
6998 }
6999 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7000 ciSymbol::make(fieldTypeString),
7001 is_static);
7002
7003 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7004 if (field == nullptr) return (Node *) nullptr;
7005
7006 if (is_static) {
7007 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7008 fromObj = makecon(tip);
7009 }
7010
7011 // Next code copied from Parse::do_get_xxx():
7012
7013 // Compute address and memory type.
7014 int offset = field->offset_in_bytes();
7015 bool is_vol = field->is_volatile();
7016 ciType* field_klass = field->type();
7017 assert(field_klass->is_loaded(), "should be loaded");
7018 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7019 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7020 BasicType bt = field->layout_type();
7021
7022 // Build the resultant type of the load
7023 const Type *type;
7024 if (bt == T_OBJECT) {
7025 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7026 } else {
7027 type = Type::get_const_basic_type(bt);
7028 }
7029
7030 if (is_vol) {
7031 decorators |= MO_SEQ_CST;
7032 }
7033
7034 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7035 }
7036
7037 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7038 bool is_exact /* true */, bool is_static /* false */,
7039 ciInstanceKlass * fromKls /* nullptr */) {
7040 if (fromKls == nullptr) {
7041 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7042 assert(tinst != nullptr, "obj is null");
7043 assert(tinst->is_loaded(), "obj is not loaded");
7044 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7045 fromKls = tinst->instance_klass();
7046 }
7047 else {
7048 assert(is_static, "only for static field access");
7049 }
7050 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7051 ciSymbol::make(fieldTypeString),
7052 is_static);
7053
7054 assert(field != nullptr, "undefined field");
7055 assert(!field->is_volatile(), "not defined for volatile fields");
7056
7057 if (is_static) {
7058 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7059 fromObj = makecon(tip);
7060 }
7061
7062 // Next code copied from Parse::do_get_xxx():
7063
7064 // Compute address and memory type.
7065 int offset = field->offset_in_bytes();
7066 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7067
7068 return adr;
7069 }
7070
7071 //------------------------------inline_aescrypt_Block-----------------------
7072 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7073 address stubAddr = nullptr;
7074 const char *stubName;
7075 assert(UseAES, "need AES instruction support");
7076
7077 switch(id) {
7078 case vmIntrinsics::_aescrypt_encryptBlock:
7079 stubAddr = StubRoutines::aescrypt_encryptBlock();
7080 stubName = "aescrypt_encryptBlock";
7081 break;
7082 case vmIntrinsics::_aescrypt_decryptBlock:
7083 stubAddr = StubRoutines::aescrypt_decryptBlock();
7084 stubName = "aescrypt_decryptBlock";
7085 break;
7086 default:
7087 break;
7088 }
7089 if (stubAddr == nullptr) return false;
7090
7091 Node* aescrypt_object = argument(0);
7092 Node* src = argument(1);
7093 Node* src_offset = argument(2);
7094 Node* dest = argument(3);
7095 Node* dest_offset = argument(4);
7096
7097 src = must_be_not_null(src, true);
7098 dest = must_be_not_null(dest, true);
7099
7100 // (1) src and dest are arrays.
7101 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7102 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7103 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7104 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7105
7106 // for the quick and dirty code we will skip all the checks.
7107 // we are just trying to get the call to be generated.
7108 Node* src_start = src;
7109 Node* dest_start = dest;
7110 if (src_offset != nullptr || dest_offset != nullptr) {
7111 assert(src_offset != nullptr && dest_offset != nullptr, "");
7112 src_start = array_element_address(src, src_offset, T_BYTE);
7113 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7114 }
7115
7116 // now need to get the start of its expanded key array
7117 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7118 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7119 if (k_start == nullptr) return false;
7120
7121 // Call the stub.
7122 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7123 stubAddr, stubName, TypePtr::BOTTOM,
7124 src_start, dest_start, k_start);
7125
7126 return true;
7127 }
7128
7129 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7130 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7131 address stubAddr = nullptr;
7132 const char *stubName = nullptr;
7133
7134 assert(UseAES, "need AES instruction support");
7135
7136 switch(id) {
7137 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7138 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7139 stubName = "cipherBlockChaining_encryptAESCrypt";
7140 break;
7141 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7142 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7143 stubName = "cipherBlockChaining_decryptAESCrypt";
7144 break;
7145 default:
7146 break;
7147 }
7148 if (stubAddr == nullptr) return false;
7149
7150 Node* cipherBlockChaining_object = argument(0);
7151 Node* src = argument(1);
7152 Node* src_offset = argument(2);
7153 Node* len = argument(3);
7154 Node* dest = argument(4);
7155 Node* dest_offset = argument(5);
7156
7157 src = must_be_not_null(src, false);
7158 dest = must_be_not_null(dest, false);
7159
7160 // (1) src and dest are arrays.
7161 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7162 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7163 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7164 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7165
7166 // checks are the responsibility of the caller
7167 Node* src_start = src;
7168 Node* dest_start = dest;
7169 if (src_offset != nullptr || dest_offset != nullptr) {
7170 assert(src_offset != nullptr && dest_offset != nullptr, "");
7171 src_start = array_element_address(src, src_offset, T_BYTE);
7172 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7173 }
7174
7175 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7176 // (because of the predicated logic executed earlier).
7177 // so we cast it here safely.
7178 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7179
7180 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7181 if (embeddedCipherObj == nullptr) return false;
7182
7183 // cast it to what we know it will be at runtime
7184 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7185 assert(tinst != nullptr, "CBC obj is null");
7186 assert(tinst->is_loaded(), "CBC obj is not loaded");
7187 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7188 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7189
7190 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7191 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7192 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7193 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7194 aescrypt_object = _gvn.transform(aescrypt_object);
7195
7196 // we need to get the start of the aescrypt_object's expanded key array
7197 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7198 if (k_start == nullptr) return false;
7199
7200 // similarly, get the start address of the r vector
7201 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7202 if (objRvec == nullptr) return false;
7203 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7204
7205 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7206 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7207 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7208 stubAddr, stubName, TypePtr::BOTTOM,
7209 src_start, dest_start, k_start, r_start, len);
7210
7211 // return cipher length (int)
7212 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7213 set_result(retvalue);
7214 return true;
7215 }
7216
7217 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7218 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7219 address stubAddr = nullptr;
7220 const char *stubName = nullptr;
7221
7222 assert(UseAES, "need AES instruction support");
7223
7224 switch (id) {
7225 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7226 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7227 stubName = "electronicCodeBook_encryptAESCrypt";
7228 break;
7229 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7230 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7231 stubName = "electronicCodeBook_decryptAESCrypt";
7232 break;
7233 default:
7234 break;
7235 }
7236
7237 if (stubAddr == nullptr) return false;
7238
7239 Node* electronicCodeBook_object = argument(0);
7240 Node* src = argument(1);
7241 Node* src_offset = argument(2);
7242 Node* len = argument(3);
7243 Node* dest = argument(4);
7244 Node* dest_offset = argument(5);
7245
7246 // (1) src and dest are arrays.
7247 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7248 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7249 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7250 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7251
7252 // checks are the responsibility of the caller
7253 Node* src_start = src;
7254 Node* dest_start = dest;
7255 if (src_offset != nullptr || dest_offset != nullptr) {
7256 assert(src_offset != nullptr && dest_offset != nullptr, "");
7257 src_start = array_element_address(src, src_offset, T_BYTE);
7258 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7259 }
7260
7261 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7262 // (because of the predicated logic executed earlier).
7263 // so we cast it here safely.
7264 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7265
7266 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7267 if (embeddedCipherObj == nullptr) return false;
7268
7269 // cast it to what we know it will be at runtime
7270 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7271 assert(tinst != nullptr, "ECB obj is null");
7272 assert(tinst->is_loaded(), "ECB obj is not loaded");
7273 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7274 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7275
7276 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7277 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7278 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7279 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7280 aescrypt_object = _gvn.transform(aescrypt_object);
7281
7282 // we need to get the start of the aescrypt_object's expanded key array
7283 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7284 if (k_start == nullptr) return false;
7285
7286 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7287 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7288 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7289 stubAddr, stubName, TypePtr::BOTTOM,
7290 src_start, dest_start, k_start, len);
7291
7292 // return cipher length (int)
7293 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7294 set_result(retvalue);
7295 return true;
7296 }
7297
7298 //------------------------------inline_counterMode_AESCrypt-----------------------
7299 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7300 assert(UseAES, "need AES instruction support");
7301 if (!UseAESCTRIntrinsics) return false;
7302
7303 address stubAddr = nullptr;
7304 const char *stubName = nullptr;
7305 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7306 stubAddr = StubRoutines::counterMode_AESCrypt();
7307 stubName = "counterMode_AESCrypt";
7308 }
7309 if (stubAddr == nullptr) return false;
7310
7311 Node* counterMode_object = argument(0);
7312 Node* src = argument(1);
7313 Node* src_offset = argument(2);
7314 Node* len = argument(3);
7315 Node* dest = argument(4);
7316 Node* dest_offset = argument(5);
7317
7318 // (1) src and dest are arrays.
7319 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7320 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7321 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7322 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7323
7324 // checks are the responsibility of the caller
7325 Node* src_start = src;
7326 Node* dest_start = dest;
7327 if (src_offset != nullptr || dest_offset != nullptr) {
7328 assert(src_offset != nullptr && dest_offset != nullptr, "");
7329 src_start = array_element_address(src, src_offset, T_BYTE);
7330 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7331 }
7332
7333 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7334 // (because of the predicated logic executed earlier).
7335 // so we cast it here safely.
7336 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7337 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7338 if (embeddedCipherObj == nullptr) return false;
7339 // cast it to what we know it will be at runtime
7340 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7341 assert(tinst != nullptr, "CTR obj is null");
7342 assert(tinst->is_loaded(), "CTR obj is not loaded");
7343 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7344 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7345 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7346 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7347 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7348 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7349 aescrypt_object = _gvn.transform(aescrypt_object);
7350 // we need to get the start of the aescrypt_object's expanded key array
7351 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7352 if (k_start == nullptr) return false;
7353 // similarly, get the start address of the r vector
7354 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7355 if (obj_counter == nullptr) return false;
7356 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7357
7358 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7359 if (saved_encCounter == nullptr) return false;
7360 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7361 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7362
7363 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7364 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7365 OptoRuntime::counterMode_aescrypt_Type(),
7366 stubAddr, stubName, TypePtr::BOTTOM,
7367 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7368
7369 // return cipher length (int)
7370 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7371 set_result(retvalue);
7372 return true;
7373 }
7374
7375 //------------------------------get_key_start_from_aescrypt_object-----------------------
7376 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7377 #if defined(PPC64) || defined(S390) || defined(RISCV64)
7378 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7379 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7380 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7381 // The ppc64 and riscv64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7382 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7383 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7384 if (objSessionK == nullptr) {
7385 return (Node *) nullptr;
7386 }
7387 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7388 #else
7389 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7390 #endif // PPC64
7391 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7392 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7393
7394 // now have the array, need to get the start address of the K array
7395 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7396 return k_start;
7397 }
7398
7399 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7400 // Return node representing slow path of predicate check.
7401 // the pseudo code we want to emulate with this predicate is:
7402 // for encryption:
7403 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7404 // for decryption:
7405 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7406 // note cipher==plain is more conservative than the original java code but that's OK
7407 //
7408 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7409 // The receiver was checked for null already.
7410 Node* objCBC = argument(0);
7411
7412 Node* src = argument(1);
7413 Node* dest = argument(4);
7414
7415 // Load embeddedCipher field of CipherBlockChaining object.
7416 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7417
7418 // get AESCrypt klass for instanceOf check
7419 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7420 // will have same classloader as CipherBlockChaining object
7421 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7422 assert(tinst != nullptr, "CBCobj is null");
7423 assert(tinst->is_loaded(), "CBCobj is not loaded");
7424
7425 // we want to do an instanceof comparison against the AESCrypt class
7426 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7427 if (!klass_AESCrypt->is_loaded()) {
7428 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7429 Node* ctrl = control();
7430 set_control(top()); // no regular fast path
7431 return ctrl;
7432 }
7433
7434 src = must_be_not_null(src, true);
7435 dest = must_be_not_null(dest, true);
7436
7437 // Resolve oops to stable for CmpP below.
7438 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7439
7440 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7441 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7442 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7443
7444 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7445
7446 // for encryption, we are done
7447 if (!decrypting)
7448 return instof_false; // even if it is null
7449
7450 // for decryption, we need to add a further check to avoid
7451 // taking the intrinsic path when cipher and plain are the same
7452 // see the original java code for why.
7453 RegionNode* region = new RegionNode(3);
7454 region->init_req(1, instof_false);
7455
7456 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7457 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7458 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7459 region->init_req(2, src_dest_conjoint);
7460
7461 record_for_igvn(region);
7462 return _gvn.transform(region);
7463 }
7464
7465 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7466 // Return node representing slow path of predicate check.
7467 // the pseudo code we want to emulate with this predicate is:
7468 // for encryption:
7469 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7470 // for decryption:
7471 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7472 // note cipher==plain is more conservative than the original java code but that's OK
7473 //
7474 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7475 // The receiver was checked for null already.
7476 Node* objECB = argument(0);
7477
7478 // Load embeddedCipher field of ElectronicCodeBook object.
7479 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7480
7481 // get AESCrypt klass for instanceOf check
7482 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7483 // will have same classloader as ElectronicCodeBook object
7484 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7485 assert(tinst != nullptr, "ECBobj is null");
7486 assert(tinst->is_loaded(), "ECBobj is not loaded");
7487
7488 // we want to do an instanceof comparison against the AESCrypt class
7489 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7490 if (!klass_AESCrypt->is_loaded()) {
7491 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7492 Node* ctrl = control();
7493 set_control(top()); // no regular fast path
7494 return ctrl;
7495 }
7496 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7497
7498 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7499 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7500 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7501
7502 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7503
7504 // for encryption, we are done
7505 if (!decrypting)
7506 return instof_false; // even if it is null
7507
7508 // for decryption, we need to add a further check to avoid
7509 // taking the intrinsic path when cipher and plain are the same
7510 // see the original java code for why.
7511 RegionNode* region = new RegionNode(3);
7512 region->init_req(1, instof_false);
7513 Node* src = argument(1);
7514 Node* dest = argument(4);
7515 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7516 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7517 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7518 region->init_req(2, src_dest_conjoint);
7519
7520 record_for_igvn(region);
7521 return _gvn.transform(region);
7522 }
7523
7524 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7525 // Return node representing slow path of predicate check.
7526 // the pseudo code we want to emulate with this predicate is:
7527 // for encryption:
7528 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7529 // for decryption:
7530 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7531 // note cipher==plain is more conservative than the original java code but that's OK
7532 //
7533
7534 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7535 // The receiver was checked for null already.
7536 Node* objCTR = argument(0);
7537
7538 // Load embeddedCipher field of CipherBlockChaining object.
7539 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7540
7541 // get AESCrypt klass for instanceOf check
7542 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7543 // will have same classloader as CipherBlockChaining object
7544 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7545 assert(tinst != nullptr, "CTRobj is null");
7546 assert(tinst->is_loaded(), "CTRobj is not loaded");
7547
7548 // we want to do an instanceof comparison against the AESCrypt class
7549 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7550 if (!klass_AESCrypt->is_loaded()) {
7551 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7552 Node* ctrl = control();
7553 set_control(top()); // no regular fast path
7554 return ctrl;
7555 }
7556
7557 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7558 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7559 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7560 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7561 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7562
7563 return instof_false; // even if it is null
7564 }
7565
7566 //------------------------------inline_ghash_processBlocks
7567 bool LibraryCallKit::inline_ghash_processBlocks() {
7568 address stubAddr;
7569 const char *stubName;
7570 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7571
7572 stubAddr = StubRoutines::ghash_processBlocks();
7573 stubName = "ghash_processBlocks";
7574
7575 Node* data = argument(0);
7576 Node* offset = argument(1);
7577 Node* len = argument(2);
7578 Node* state = argument(3);
7579 Node* subkeyH = argument(4);
7580
7581 state = must_be_not_null(state, true);
7582 subkeyH = must_be_not_null(subkeyH, true);
7583 data = must_be_not_null(data, true);
7584
7585 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7586 assert(state_start, "state is null");
7587 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7588 assert(subkeyH_start, "subkeyH is null");
7589 Node* data_start = array_element_address(data, offset, T_BYTE);
7590 assert(data_start, "data is null");
7591
7592 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7593 OptoRuntime::ghash_processBlocks_Type(),
7594 stubAddr, stubName, TypePtr::BOTTOM,
7595 state_start, subkeyH_start, data_start, len);
7596 return true;
7597 }
7598
7599 //------------------------------inline_chacha20Block
7600 bool LibraryCallKit::inline_chacha20Block() {
7601 address stubAddr;
7602 const char *stubName;
7603 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7604
7605 stubAddr = StubRoutines::chacha20Block();
7606 stubName = "chacha20Block";
7607
7608 Node* state = argument(0);
7609 Node* result = argument(1);
7610
7611 state = must_be_not_null(state, true);
7612 result = must_be_not_null(result, true);
7613
7614 Node* state_start = array_element_address(state, intcon(0), T_INT);
7615 assert(state_start, "state is null");
7616 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7617 assert(result_start, "result is null");
7618
7619 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7620 OptoRuntime::chacha20Block_Type(),
7621 stubAddr, stubName, TypePtr::BOTTOM,
7622 state_start, result_start);
7623 // return key stream length (int)
7624 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7625 set_result(retvalue);
7626 return true;
7627 }
7628
7629 bool LibraryCallKit::inline_base64_encodeBlock() {
7630 address stubAddr;
7631 const char *stubName;
7632 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7633 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
7634 stubAddr = StubRoutines::base64_encodeBlock();
7635 stubName = "encodeBlock";
7636
7637 if (!stubAddr) return false;
7638 Node* base64obj = argument(0);
7639 Node* src = argument(1);
7640 Node* offset = argument(2);
7641 Node* len = argument(3);
7642 Node* dest = argument(4);
7643 Node* dp = argument(5);
7644 Node* isURL = argument(6);
7645
7646 src = must_be_not_null(src, true);
7647 dest = must_be_not_null(dest, true);
7648
7649 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7650 assert(src_start, "source array is null");
7651 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7652 assert(dest_start, "destination array is null");
7653
7654 Node* base64 = make_runtime_call(RC_LEAF,
7655 OptoRuntime::base64_encodeBlock_Type(),
7656 stubAddr, stubName, TypePtr::BOTTOM,
7657 src_start, offset, len, dest_start, dp, isURL);
7658 return true;
7659 }
7660
7661 bool LibraryCallKit::inline_base64_decodeBlock() {
7662 address stubAddr;
7663 const char *stubName;
7664 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7665 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
7666 stubAddr = StubRoutines::base64_decodeBlock();
7667 stubName = "decodeBlock";
7668
7669 if (!stubAddr) return false;
7670 Node* base64obj = argument(0);
7671 Node* src = argument(1);
7672 Node* src_offset = argument(2);
7673 Node* len = argument(3);
7674 Node* dest = argument(4);
7675 Node* dest_offset = argument(5);
7676 Node* isURL = argument(6);
7677 Node* isMIME = argument(7);
7678
7679 src = must_be_not_null(src, true);
7680 dest = must_be_not_null(dest, true);
7681
7682 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7683 assert(src_start, "source array is null");
7684 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7685 assert(dest_start, "destination array is null");
7686
7687 Node* call = make_runtime_call(RC_LEAF,
7688 OptoRuntime::base64_decodeBlock_Type(),
7689 stubAddr, stubName, TypePtr::BOTTOM,
7690 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
7691 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7692 set_result(result);
7693 return true;
7694 }
7695
7696 bool LibraryCallKit::inline_poly1305_processBlocks() {
7697 address stubAddr;
7698 const char *stubName;
7699 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
7700 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
7701 stubAddr = StubRoutines::poly1305_processBlocks();
7702 stubName = "poly1305_processBlocks";
7703
7704 if (!stubAddr) return false;
7705 null_check_receiver(); // null-check receiver
7706 if (stopped()) return true;
7707
7708 Node* input = argument(1);
7709 Node* input_offset = argument(2);
7710 Node* len = argument(3);
7711 Node* alimbs = argument(4);
7712 Node* rlimbs = argument(5);
7713
7714 input = must_be_not_null(input, true);
7715 alimbs = must_be_not_null(alimbs, true);
7716 rlimbs = must_be_not_null(rlimbs, true);
7717
7718 Node* input_start = array_element_address(input, input_offset, T_BYTE);
7719 assert(input_start, "input array is null");
7720 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
7721 assert(acc_start, "acc array is null");
7722 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
7723 assert(r_start, "r array is null");
7724
7725 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7726 OptoRuntime::poly1305_processBlocks_Type(),
7727 stubAddr, stubName, TypePtr::BOTTOM,
7728 input_start, len, acc_start, r_start);
7729 return true;
7730 }
7731
7732 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
7733 address stubAddr;
7734 const char *stubName;
7735 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
7736 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
7737 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
7738 stubName = "intpoly_montgomeryMult_P256";
7739
7740 if (!stubAddr) return false;
7741 null_check_receiver(); // null-check receiver
7742 if (stopped()) return true;
7743
7744 Node* a = argument(1);
7745 Node* b = argument(2);
7746 Node* r = argument(3);
7747
7748 a = must_be_not_null(a, true);
7749 b = must_be_not_null(b, true);
7750 r = must_be_not_null(r, true);
7751
7752 Node* a_start = array_element_address(a, intcon(0), T_LONG);
7753 assert(a_start, "a array is NULL");
7754 Node* b_start = array_element_address(b, intcon(0), T_LONG);
7755 assert(b_start, "b array is NULL");
7756 Node* r_start = array_element_address(r, intcon(0), T_LONG);
7757 assert(r_start, "r array is NULL");
7758
7759 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7760 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
7761 stubAddr, stubName, TypePtr::BOTTOM,
7762 a_start, b_start, r_start);
7763 return true;
7764 }
7765
7766 bool LibraryCallKit::inline_intpoly_assign() {
7767 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
7768 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
7769 const char *stubName = "intpoly_assign";
7770 address stubAddr = StubRoutines::intpoly_assign();
7771 if (!stubAddr) return false;
7772
7773 Node* set = argument(0);
7774 Node* a = argument(1);
7775 Node* b = argument(2);
7776 Node* arr_length = load_array_length(a);
7777
7778 a = must_be_not_null(a, true);
7779 b = must_be_not_null(b, true);
7780
7781 Node* a_start = array_element_address(a, intcon(0), T_LONG);
7782 assert(a_start, "a array is NULL");
7783 Node* b_start = array_element_address(b, intcon(0), T_LONG);
7784 assert(b_start, "b array is NULL");
7785
7786 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7787 OptoRuntime::intpoly_assign_Type(),
7788 stubAddr, stubName, TypePtr::BOTTOM,
7789 set, a_start, b_start, arr_length);
7790 return true;
7791 }
7792
7793 //------------------------------inline_digestBase_implCompress-----------------------
7794 //
7795 // Calculate MD5 for single-block byte[] array.
7796 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
7797 //
7798 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
7799 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
7800 //
7801 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
7802 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
7803 //
7804 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
7805 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
7806 //
7807 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
7808 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
7809 //
7810 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
7811 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
7812
7813 Node* digestBase_obj = argument(0);
7814 Node* src = argument(1); // type oop
7815 Node* ofs = argument(2); // type int
7816
7817 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7818 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7819 // failed array check
7820 return false;
7821 }
7822 // Figure out the size and type of the elements we will be copying.
7823 BasicType src_elem = src_type->elem()->array_element_basic_type();
7824 if (src_elem != T_BYTE) {
7825 return false;
7826 }
7827 // 'src_start' points to src array + offset
7828 src = must_be_not_null(src, true);
7829 Node* src_start = array_element_address(src, ofs, src_elem);
7830 Node* state = nullptr;
7831 Node* block_size = nullptr;
7832 address stubAddr;
7833 const char *stubName;
7834
7835 switch(id) {
7836 case vmIntrinsics::_md5_implCompress:
7837 assert(UseMD5Intrinsics, "need MD5 instruction support");
7838 state = get_state_from_digest_object(digestBase_obj, T_INT);
7839 stubAddr = StubRoutines::md5_implCompress();
7840 stubName = "md5_implCompress";
7841 break;
7842 case vmIntrinsics::_sha_implCompress:
7843 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
7844 state = get_state_from_digest_object(digestBase_obj, T_INT);
7845 stubAddr = StubRoutines::sha1_implCompress();
7846 stubName = "sha1_implCompress";
7847 break;
7848 case vmIntrinsics::_sha2_implCompress:
7849 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
7850 state = get_state_from_digest_object(digestBase_obj, T_INT);
7851 stubAddr = StubRoutines::sha256_implCompress();
7852 stubName = "sha256_implCompress";
7853 break;
7854 case vmIntrinsics::_sha5_implCompress:
7855 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
7856 state = get_state_from_digest_object(digestBase_obj, T_LONG);
7857 stubAddr = StubRoutines::sha512_implCompress();
7858 stubName = "sha512_implCompress";
7859 break;
7860 case vmIntrinsics::_sha3_implCompress:
7861 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
7862 state = get_state_from_digest_object(digestBase_obj, T_LONG);
7863 stubAddr = StubRoutines::sha3_implCompress();
7864 stubName = "sha3_implCompress";
7865 block_size = get_block_size_from_digest_object(digestBase_obj);
7866 if (block_size == nullptr) return false;
7867 break;
7868 default:
7869 fatal_unexpected_iid(id);
7870 return false;
7871 }
7872 if (state == nullptr) return false;
7873
7874 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
7875 if (stubAddr == nullptr) return false;
7876
7877 // Call the stub.
7878 Node* call;
7879 if (block_size == nullptr) {
7880 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
7881 stubAddr, stubName, TypePtr::BOTTOM,
7882 src_start, state);
7883 } else {
7884 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
7885 stubAddr, stubName, TypePtr::BOTTOM,
7886 src_start, state, block_size);
7887 }
7888
7889 return true;
7890 }
7891
7892 //------------------------------inline_digestBase_implCompressMB-----------------------
7893 //
7894 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
7895 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
7896 //
7897 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
7898 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7899 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7900 assert((uint)predicate < 5, "sanity");
7901 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
7902
7903 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
7904 Node* src = argument(1); // byte[] array
7905 Node* ofs = argument(2); // type int
7906 Node* limit = argument(3); // type int
7907
7908 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7909 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7910 // failed array check
7911 return false;
7912 }
7913 // Figure out the size and type of the elements we will be copying.
7914 BasicType src_elem = src_type->elem()->array_element_basic_type();
7915 if (src_elem != T_BYTE) {
7916 return false;
7917 }
7918 // 'src_start' points to src array + offset
7919 src = must_be_not_null(src, false);
7920 Node* src_start = array_element_address(src, ofs, src_elem);
7921
7922 const char* klass_digestBase_name = nullptr;
7923 const char* stub_name = nullptr;
7924 address stub_addr = nullptr;
7925 BasicType elem_type = T_INT;
7926
7927 switch (predicate) {
7928 case 0:
7929 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
7930 klass_digestBase_name = "sun/security/provider/MD5";
7931 stub_name = "md5_implCompressMB";
7932 stub_addr = StubRoutines::md5_implCompressMB();
7933 }
7934 break;
7935 case 1:
7936 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
7937 klass_digestBase_name = "sun/security/provider/SHA";
7938 stub_name = "sha1_implCompressMB";
7939 stub_addr = StubRoutines::sha1_implCompressMB();
7940 }
7941 break;
7942 case 2:
7943 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
7944 klass_digestBase_name = "sun/security/provider/SHA2";
7945 stub_name = "sha256_implCompressMB";
7946 stub_addr = StubRoutines::sha256_implCompressMB();
7947 }
7948 break;
7949 case 3:
7950 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
7951 klass_digestBase_name = "sun/security/provider/SHA5";
7952 stub_name = "sha512_implCompressMB";
7953 stub_addr = StubRoutines::sha512_implCompressMB();
7954 elem_type = T_LONG;
7955 }
7956 break;
7957 case 4:
7958 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
7959 klass_digestBase_name = "sun/security/provider/SHA3";
7960 stub_name = "sha3_implCompressMB";
7961 stub_addr = StubRoutines::sha3_implCompressMB();
7962 elem_type = T_LONG;
7963 }
7964 break;
7965 default:
7966 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
7967 }
7968 if (klass_digestBase_name != nullptr) {
7969 assert(stub_addr != nullptr, "Stub is generated");
7970 if (stub_addr == nullptr) return false;
7971
7972 // get DigestBase klass to lookup for SHA klass
7973 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
7974 assert(tinst != nullptr, "digestBase_obj is not instance???");
7975 assert(tinst->is_loaded(), "DigestBase is not loaded");
7976
7977 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
7978 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
7979 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
7980 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
7981 }
7982 return false;
7983 }
7984
7985 //------------------------------inline_digestBase_implCompressMB-----------------------
7986 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
7987 BasicType elem_type, address stubAddr, const char *stubName,
7988 Node* src_start, Node* ofs, Node* limit) {
7989 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
7990 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7991 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
7992 digest_obj = _gvn.transform(digest_obj);
7993
7994 Node* state = get_state_from_digest_object(digest_obj, elem_type);
7995 if (state == nullptr) return false;
7996
7997 Node* block_size = nullptr;
7998 if (strcmp("sha3_implCompressMB", stubName) == 0) {
7999 block_size = get_block_size_from_digest_object(digest_obj);
8000 if (block_size == nullptr) return false;
8001 }
8002
8003 // Call the stub.
8004 Node* call;
8005 if (block_size == nullptr) {
8006 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8007 OptoRuntime::digestBase_implCompressMB_Type(false),
8008 stubAddr, stubName, TypePtr::BOTTOM,
8009 src_start, state, ofs, limit);
8010 } else {
8011 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8012 OptoRuntime::digestBase_implCompressMB_Type(true),
8013 stubAddr, stubName, TypePtr::BOTTOM,
8014 src_start, state, block_size, ofs, limit);
8015 }
8016
8017 // return ofs (int)
8018 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8019 set_result(result);
8020
8021 return true;
8022 }
8023
8024 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8025 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8026 assert(UseAES, "need AES instruction support");
8027 address stubAddr = nullptr;
8028 const char *stubName = nullptr;
8029 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8030 stubName = "galoisCounterMode_AESCrypt";
8031
8032 if (stubAddr == nullptr) return false;
8033
8034 Node* in = argument(0);
8035 Node* inOfs = argument(1);
8036 Node* len = argument(2);
8037 Node* ct = argument(3);
8038 Node* ctOfs = argument(4);
8039 Node* out = argument(5);
8040 Node* outOfs = argument(6);
8041 Node* gctr_object = argument(7);
8042 Node* ghash_object = argument(8);
8043
8044 // (1) in, ct and out are arrays.
8045 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8046 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8047 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8048 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8049 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8050 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8051
8052 // checks are the responsibility of the caller
8053 Node* in_start = in;
8054 Node* ct_start = ct;
8055 Node* out_start = out;
8056 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8057 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8058 in_start = array_element_address(in, inOfs, T_BYTE);
8059 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8060 out_start = array_element_address(out, outOfs, T_BYTE);
8061 }
8062
8063 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8064 // (because of the predicated logic executed earlier).
8065 // so we cast it here safely.
8066 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8067 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8068 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8069 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8070 Node* state = load_field_from_object(ghash_object, "state", "[J");
8071
8072 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8073 return false;
8074 }
8075 // cast it to what we know it will be at runtime
8076 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8077 assert(tinst != nullptr, "GCTR obj is null");
8078 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8079 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8080 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8081 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8082 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8083 const TypeOopPtr* xtype = aklass->as_instance_type();
8084 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8085 aescrypt_object = _gvn.transform(aescrypt_object);
8086 // we need to get the start of the aescrypt_object's expanded key array
8087 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8088 if (k_start == nullptr) return false;
8089 // similarly, get the start address of the r vector
8090 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8091 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8092 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8093
8094
8095 // Call the stub, passing params
8096 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8097 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8098 stubAddr, stubName, TypePtr::BOTTOM,
8099 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8100
8101 // return cipher length (int)
8102 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8103 set_result(retvalue);
8104
8105 return true;
8106 }
8107
8108 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8109 // Return node representing slow path of predicate check.
8110 // the pseudo code we want to emulate with this predicate is:
8111 // for encryption:
8112 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8113 // for decryption:
8114 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8115 // note cipher==plain is more conservative than the original java code but that's OK
8116 //
8117
8118 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8119 // The receiver was checked for null already.
8120 Node* objGCTR = argument(7);
8121 // Load embeddedCipher field of GCTR object.
8122 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8123 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8124
8125 // get AESCrypt klass for instanceOf check
8126 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8127 // will have same classloader as CipherBlockChaining object
8128 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8129 assert(tinst != nullptr, "GCTR obj is null");
8130 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8131
8132 // we want to do an instanceof comparison against the AESCrypt class
8133 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8134 if (!klass_AESCrypt->is_loaded()) {
8135 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8136 Node* ctrl = control();
8137 set_control(top()); // no regular fast path
8138 return ctrl;
8139 }
8140
8141 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8142 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8143 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8144 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8145 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8146
8147 return instof_false; // even if it is null
8148 }
8149
8150 //------------------------------get_state_from_digest_object-----------------------
8151 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8152 const char* state_type;
8153 switch (elem_type) {
8154 case T_BYTE: state_type = "[B"; break;
8155 case T_INT: state_type = "[I"; break;
8156 case T_LONG: state_type = "[J"; break;
8157 default: ShouldNotReachHere();
8158 }
8159 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8160 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8161 if (digest_state == nullptr) return (Node *) nullptr;
8162
8163 // now have the array, need to get the start address of the state array
8164 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8165 return state;
8166 }
8167
8168 //------------------------------get_block_size_from_sha3_object----------------------------------
8169 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8170 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8171 assert (block_size != nullptr, "sanity");
8172 return block_size;
8173 }
8174
8175 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8176 // Return node representing slow path of predicate check.
8177 // the pseudo code we want to emulate with this predicate is:
8178 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8179 //
8180 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8181 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8182 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8183 assert((uint)predicate < 5, "sanity");
8184
8185 // The receiver was checked for null already.
8186 Node* digestBaseObj = argument(0);
8187
8188 // get DigestBase klass for instanceOf check
8189 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8190 assert(tinst != nullptr, "digestBaseObj is null");
8191 assert(tinst->is_loaded(), "DigestBase is not loaded");
8192
8193 const char* klass_name = nullptr;
8194 switch (predicate) {
8195 case 0:
8196 if (UseMD5Intrinsics) {
8197 // we want to do an instanceof comparison against the MD5 class
8198 klass_name = "sun/security/provider/MD5";
8199 }
8200 break;
8201 case 1:
8202 if (UseSHA1Intrinsics) {
8203 // we want to do an instanceof comparison against the SHA class
8204 klass_name = "sun/security/provider/SHA";
8205 }
8206 break;
8207 case 2:
8208 if (UseSHA256Intrinsics) {
8209 // we want to do an instanceof comparison against the SHA2 class
8210 klass_name = "sun/security/provider/SHA2";
8211 }
8212 break;
8213 case 3:
8214 if (UseSHA512Intrinsics) {
8215 // we want to do an instanceof comparison against the SHA5 class
8216 klass_name = "sun/security/provider/SHA5";
8217 }
8218 break;
8219 case 4:
8220 if (UseSHA3Intrinsics) {
8221 // we want to do an instanceof comparison against the SHA3 class
8222 klass_name = "sun/security/provider/SHA3";
8223 }
8224 break;
8225 default:
8226 fatal("unknown SHA intrinsic predicate: %d", predicate);
8227 }
8228
8229 ciKlass* klass = nullptr;
8230 if (klass_name != nullptr) {
8231 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8232 }
8233 if ((klass == nullptr) || !klass->is_loaded()) {
8234 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8235 Node* ctrl = control();
8236 set_control(top()); // no intrinsic path
8237 return ctrl;
8238 }
8239 ciInstanceKlass* instklass = klass->as_instance_klass();
8240
8241 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8242 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8243 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8244 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8245
8246 return instof_false; // even if it is null
8247 }
8248
8249 //-------------inline_fma-----------------------------------
8250 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8251 Node *a = nullptr;
8252 Node *b = nullptr;
8253 Node *c = nullptr;
8254 Node* result = nullptr;
8255 switch (id) {
8256 case vmIntrinsics::_fmaD:
8257 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8258 // no receiver since it is static method
8259 a = round_double_node(argument(0));
8260 b = round_double_node(argument(2));
8261 c = round_double_node(argument(4));
8262 result = _gvn.transform(new FmaDNode(control(), a, b, c));
8263 break;
8264 case vmIntrinsics::_fmaF:
8265 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8266 a = argument(0);
8267 b = argument(1);
8268 c = argument(2);
8269 result = _gvn.transform(new FmaFNode(control(), a, b, c));
8270 break;
8271 default:
8272 fatal_unexpected_iid(id); break;
8273 }
8274 set_result(result);
8275 return true;
8276 }
8277
8278 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8279 // argument(0) is receiver
8280 Node* codePoint = argument(1);
8281 Node* n = nullptr;
8282
8283 switch (id) {
8284 case vmIntrinsics::_isDigit :
8285 n = new DigitNode(control(), codePoint);
8286 break;
8287 case vmIntrinsics::_isLowerCase :
8288 n = new LowerCaseNode(control(), codePoint);
8289 break;
8290 case vmIntrinsics::_isUpperCase :
8291 n = new UpperCaseNode(control(), codePoint);
8292 break;
8293 case vmIntrinsics::_isWhitespace :
8294 n = new WhitespaceNode(control(), codePoint);
8295 break;
8296 default:
8297 fatal_unexpected_iid(id);
8298 }
8299
8300 set_result(_gvn.transform(n));
8301 return true;
8302 }
8303
8304 //------------------------------inline_fp_min_max------------------------------
8305 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
8306 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
8307
8308 // The intrinsic should be used only when the API branches aren't predictable,
8309 // the last one performing the most important comparison. The following heuristic
8310 // uses the branch statistics to eventually bail out if necessary.
8311
8312 ciMethodData *md = callee()->method_data();
8313
8314 if ( md != nullptr && md->is_mature() && md->invocation_count() > 0 ) {
8315 ciCallProfile cp = caller()->call_profile_at_bci(bci());
8316
8317 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
8318 // Bail out if the call-site didn't contribute enough to the statistics.
8319 return false;
8320 }
8321
8322 uint taken = 0, not_taken = 0;
8323
8324 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
8325 if (p->is_BranchData()) {
8326 taken = ((ciBranchData*)p)->taken();
8327 not_taken = ((ciBranchData*)p)->not_taken();
8328 }
8329 }
8330
8331 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
8332 balance = balance < 0 ? -balance : balance;
8333 if ( balance > 0.2 ) {
8334 // Bail out if the most important branch is predictable enough.
8335 return false;
8336 }
8337 }
8338 */
8339
8340 Node *a = nullptr;
8341 Node *b = nullptr;
8342 Node *n = nullptr;
8343 switch (id) {
8344 case vmIntrinsics::_maxF:
8345 case vmIntrinsics::_minF:
8346 case vmIntrinsics::_maxF_strict:
8347 case vmIntrinsics::_minF_strict:
8348 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
8349 a = argument(0);
8350 b = argument(1);
8351 break;
8352 case vmIntrinsics::_maxD:
8353 case vmIntrinsics::_minD:
8354 case vmIntrinsics::_maxD_strict:
8355 case vmIntrinsics::_minD_strict:
8356 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
8357 a = round_double_node(argument(0));
8358 b = round_double_node(argument(2));
8359 break;
8360 default:
8361 fatal_unexpected_iid(id);
8362 break;
8363 }
8364 switch (id) {
8365 case vmIntrinsics::_maxF:
8366 case vmIntrinsics::_maxF_strict:
8367 n = new MaxFNode(a, b);
8368 break;
8369 case vmIntrinsics::_minF:
8370 case vmIntrinsics::_minF_strict:
8371 n = new MinFNode(a, b);
8372 break;
8373 case vmIntrinsics::_maxD:
8374 case vmIntrinsics::_maxD_strict:
8375 n = new MaxDNode(a, b);
8376 break;
8377 case vmIntrinsics::_minD:
8378 case vmIntrinsics::_minD_strict:
8379 n = new MinDNode(a, b);
8380 break;
8381 default:
8382 fatal_unexpected_iid(id);
8383 break;
8384 }
8385 set_result(_gvn.transform(n));
8386 return true;
8387 }
8388
8389 bool LibraryCallKit::inline_profileBoolean() {
8390 Node* counts = argument(1);
8391 const TypeAryPtr* ary = nullptr;
8392 ciArray* aobj = nullptr;
8393 if (counts->is_Con()
8394 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8395 && (aobj = ary->const_oop()->as_array()) != nullptr
8396 && (aobj->length() == 2)) {
8397 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8398 jint false_cnt = aobj->element_value(0).as_int();
8399 jint true_cnt = aobj->element_value(1).as_int();
8400
8401 if (C->log() != nullptr) {
8402 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8403 false_cnt, true_cnt);
8404 }
8405
8406 if (false_cnt + true_cnt == 0) {
8407 // According to profile, never executed.
8408 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8409 Deoptimization::Action_reinterpret);
8410 return true;
8411 }
8412
8413 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8414 // is a number of each value occurrences.
8415 Node* result = argument(0);
8416 if (false_cnt == 0 || true_cnt == 0) {
8417 // According to profile, one value has been never seen.
8418 int expected_val = (false_cnt == 0) ? 1 : 0;
8419
8420 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8421 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8422
8423 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8424 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8425 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8426
8427 { // Slow path: uncommon trap for never seen value and then reexecute
8428 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8429 // the value has been seen at least once.
8430 PreserveJVMState pjvms(this);
8431 PreserveReexecuteState preexecs(this);
8432 jvms()->set_should_reexecute(true);
8433
8434 set_control(slow_path);
8435 set_i_o(i_o());
8436
8437 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8438 Deoptimization::Action_reinterpret);
8439 }
8440 // The guard for never seen value enables sharpening of the result and
8441 // returning a constant. It allows to eliminate branches on the same value
8442 // later on.
8443 set_control(fast_path);
8444 result = intcon(expected_val);
8445 }
8446 // Stop profiling.
8447 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8448 // By replacing method body with profile data (represented as ProfileBooleanNode
8449 // on IR level) we effectively disable profiling.
8450 // It enables full speed execution once optimized code is generated.
8451 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8452 C->record_for_igvn(profile);
8453 set_result(profile);
8454 return true;
8455 } else {
8456 // Continue profiling.
8457 // Profile data isn't available at the moment. So, execute method's bytecode version.
8458 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8459 // is compiled and counters aren't available since corresponding MethodHandle
8460 // isn't a compile-time constant.
8461 return false;
8462 }
8463 }
8464
8465 bool LibraryCallKit::inline_isCompileConstant() {
8466 Node* n = argument(0);
8467 set_result(n->is_Con() ? intcon(1) : intcon(0));
8468 return true;
8469 }
8470
8471 //------------------------------- inline_getObjectSize --------------------------------------
8472 //
8473 // Calculate the runtime size of the object/array.
8474 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8475 //
8476 bool LibraryCallKit::inline_getObjectSize() {
8477 Node* obj = argument(3);
8478 Node* klass_node = load_object_klass(obj);
8479
8480 jint layout_con = Klass::_lh_neutral_value;
8481 Node* layout_val = get_layout_helper(klass_node, layout_con);
8482 int layout_is_con = (layout_val == nullptr);
8483
8484 if (layout_is_con) {
8485 // Layout helper is constant, can figure out things at compile time.
8486
8487 if (Klass::layout_helper_is_instance(layout_con)) {
8488 // Instance case: layout_con contains the size itself.
8489 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8490 set_result(size);
8491 } else {
8492 // Array case: size is round(header + element_size*arraylength).
8493 // Since arraylength is different for every array instance, we have to
8494 // compute the whole thing at runtime.
8495
8496 Node* arr_length = load_array_length(obj);
8497
8498 int round_mask = MinObjAlignmentInBytes - 1;
8499 int hsize = Klass::layout_helper_header_size(layout_con);
8500 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8501
8502 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8503 round_mask = 0; // strength-reduce it if it goes away completely
8504 }
8505 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8506 Node* header_size = intcon(hsize + round_mask);
8507
8508 Node* lengthx = ConvI2X(arr_length);
8509 Node* headerx = ConvI2X(header_size);
8510
8511 Node* abody = lengthx;
8512 if (eshift != 0) {
8513 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8514 }
8515 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8516 if (round_mask != 0) {
8517 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8518 }
8519 size = ConvX2L(size);
8520 set_result(size);
8521 }
8522 } else {
8523 // Layout helper is not constant, need to test for array-ness at runtime.
8524
8525 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8526 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8527 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8528 record_for_igvn(result_reg);
8529
8530 Node* array_ctl = generate_array_guard(klass_node, nullptr);
8531 if (array_ctl != nullptr) {
8532 // Array case: size is round(header + element_size*arraylength).
8533 // Since arraylength is different for every array instance, we have to
8534 // compute the whole thing at runtime.
8535
8536 PreserveJVMState pjvms(this);
8537 set_control(array_ctl);
8538 Node* arr_length = load_array_length(obj);
8539
8540 int round_mask = MinObjAlignmentInBytes - 1;
8541 Node* mask = intcon(round_mask);
8542
8543 Node* hss = intcon(Klass::_lh_header_size_shift);
8544 Node* hsm = intcon(Klass::_lh_header_size_mask);
8545 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8546 header_size = _gvn.transform(new AndINode(header_size, hsm));
8547 header_size = _gvn.transform(new AddINode(header_size, mask));
8548
8549 // There is no need to mask or shift this value.
8550 // The semantics of LShiftINode include an implicit mask to 0x1F.
8551 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8552 Node* elem_shift = layout_val;
8553
8554 Node* lengthx = ConvI2X(arr_length);
8555 Node* headerx = ConvI2X(header_size);
8556
8557 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
8558 Node* size = _gvn.transform(new AddXNode(headerx, abody));
8559 if (round_mask != 0) {
8560 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
8561 }
8562 size = ConvX2L(size);
8563
8564 result_reg->init_req(_array_path, control());
8565 result_val->init_req(_array_path, size);
8566 }
8567
8568 if (!stopped()) {
8569 // Instance case: the layout helper gives us instance size almost directly,
8570 // but we need to mask out the _lh_instance_slow_path_bit.
8571 Node* size = ConvI2X(layout_val);
8572 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
8573 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
8574 size = _gvn.transform(new AndXNode(size, mask));
8575 size = ConvX2L(size);
8576
8577 result_reg->init_req(_instance_path, control());
8578 result_val->init_req(_instance_path, size);
8579 }
8580
8581 set_result(result_reg, result_val);
8582 }
8583
8584 return true;
8585 }
8586
8587 //------------------------------- inline_blackhole --------------------------------------
8588 //
8589 // Make sure all arguments to this node are alive.
8590 // This matches methods that were requested to be blackholed through compile commands.
8591 //
8592 bool LibraryCallKit::inline_blackhole() {
8593 assert(callee()->is_static(), "Should have been checked before: only static methods here");
8594 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
8595 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
8596
8597 // Blackhole node pinches only the control, not memory. This allows
8598 // the blackhole to be pinned in the loop that computes blackholed
8599 // values, but have no other side effects, like breaking the optimizations
8600 // across the blackhole.
8601
8602 Node* bh = _gvn.transform(new BlackholeNode(control()));
8603 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
8604
8605 // Bind call arguments as blackhole arguments to keep them alive
8606 uint nargs = callee()->arg_size();
8607 for (uint i = 0; i < nargs; i++) {
8608 bh->add_req(argument(i));
8609 }
8610
8611 return true;
8612 }