1 /*
2 * Copyright (c) 1999, 2024, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "ci/ciFlatArrayKlass.hpp"
28 #include "ci/ciUtilities.inline.hpp"
29 #include "classfile/vmIntrinsics.hpp"
30 #include "compiler/compileBroker.hpp"
31 #include "compiler/compileLog.hpp"
32 #include "gc/shared/barrierSet.hpp"
33 #include "jfr/support/jfrIntrinsics.hpp"
34 #include "memory/resourceArea.hpp"
35 #include "oops/klass.inline.hpp"
36 #include "oops/objArrayKlass.hpp"
37 #include "opto/addnode.hpp"
38 #include "opto/arraycopynode.hpp"
39 #include "opto/c2compiler.hpp"
40 #include "opto/castnode.hpp"
41 #include "opto/cfgnode.hpp"
42 #include "opto/convertnode.hpp"
43 #include "opto/countbitsnode.hpp"
44 #include "opto/idealKit.hpp"
45 #include "opto/library_call.hpp"
46 #include "opto/mathexactnode.hpp"
47 #include "opto/mulnode.hpp"
48 #include "opto/narrowptrnode.hpp"
49 #include "opto/opaquenode.hpp"
50 #include "opto/parse.hpp"
51 #include "opto/runtime.hpp"
52 #include "opto/rootnode.hpp"
53 #include "opto/subnode.hpp"
54 #include "prims/jvmtiExport.hpp"
55 #include "prims/jvmtiThreadState.hpp"
56 #include "prims/unsafe.hpp"
57 #include "runtime/jniHandles.inline.hpp"
58 #include "runtime/objectMonitor.hpp"
59 #include "runtime/sharedRuntime.hpp"
60 #include "runtime/stubRoutines.hpp"
61 #include "utilities/macros.hpp"
62 #include "utilities/powerOfTwo.hpp"
63
64 //---------------------------make_vm_intrinsic----------------------------
65 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
66 vmIntrinsicID id = m->intrinsic_id();
67 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
68
69 if (!m->is_loaded()) {
70 // Do not attempt to inline unloaded methods.
71 return nullptr;
72 }
73
74 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
75 bool is_available = false;
76
77 {
78 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
79 // the compiler must transition to '_thread_in_vm' state because both
80 // methods access VM-internal data.
81 VM_ENTRY_MARK;
82 methodHandle mh(THREAD, m->get_Method());
83 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
84 if (is_available && is_virtual) {
85 is_available = vmIntrinsics::does_virtual_dispatch(id);
86 }
87 }
88
89 if (is_available) {
90 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
91 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
92 return new LibraryIntrinsic(m, is_virtual,
93 vmIntrinsics::predicates_needed(id),
94 vmIntrinsics::does_virtual_dispatch(id),
95 id);
96 } else {
97 return nullptr;
98 }
99 }
100
101 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
102 LibraryCallKit kit(jvms, this);
103 Compile* C = kit.C;
104 int nodes = C->unique();
105 #ifndef PRODUCT
106 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
107 char buf[1000];
108 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
109 tty->print_cr("Intrinsic %s", str);
110 }
111 #endif
112 ciMethod* callee = kit.callee();
113 const int bci = kit.bci();
114 #ifdef ASSERT
115 Node* ctrl = kit.control();
116 #endif
117 // Try to inline the intrinsic.
118 if (callee->check_intrinsic_candidate() &&
119 kit.try_to_inline(_last_predicate)) {
120 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
121 : "(intrinsic)";
122 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
123 if (C->print_intrinsics() || C->print_inlining()) {
124 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
125 }
126 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
127 if (C->log()) {
128 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
129 vmIntrinsics::name_at(intrinsic_id()),
130 (is_virtual() ? " virtual='1'" : ""),
131 C->unique() - nodes);
132 }
133 // Push the result from the inlined method onto the stack.
134 kit.push_result();
135 C->print_inlining_update(this);
136 return kit.transfer_exceptions_into_jvms();
137 }
138
139 // The intrinsic bailed out
140 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
141 if (jvms->has_method()) {
142 // Not a root compile.
143 const char* msg;
144 if (callee->intrinsic_candidate()) {
145 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
146 } else {
147 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
148 : "failed to inline (intrinsic), method not annotated";
149 }
150 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
151 if (C->print_intrinsics() || C->print_inlining()) {
152 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
153 }
154 } else {
155 // Root compile
156 ResourceMark rm;
157 stringStream msg_stream;
158 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
159 vmIntrinsics::name_at(intrinsic_id()),
160 is_virtual() ? " (virtual)" : "", bci);
161 const char *msg = msg_stream.freeze();
162 log_debug(jit, inlining)("%s", msg);
163 if (C->print_intrinsics() || C->print_inlining()) {
164 tty->print("%s", msg);
165 }
166 }
167 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
168 C->print_inlining_update(this);
169
170 return nullptr;
171 }
172
173 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
174 LibraryCallKit kit(jvms, this);
175 Compile* C = kit.C;
176 int nodes = C->unique();
177 _last_predicate = predicate;
178 #ifndef PRODUCT
179 assert(is_predicated() && predicate < predicates_count(), "sanity");
180 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
181 char buf[1000];
182 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
183 tty->print_cr("Predicate for intrinsic %s", str);
184 }
185 #endif
186 ciMethod* callee = kit.callee();
187 const int bci = kit.bci();
188
189 Node* slow_ctl = kit.try_to_predicate(predicate);
190 if (!kit.failing()) {
191 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
192 : "(intrinsic, predicate)";
193 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
194 if (C->print_intrinsics() || C->print_inlining()) {
195 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
196 }
197 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
198 if (C->log()) {
199 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
200 vmIntrinsics::name_at(intrinsic_id()),
201 (is_virtual() ? " virtual='1'" : ""),
202 C->unique() - nodes);
203 }
204 return slow_ctl; // Could be null if the check folds.
205 }
206
207 // The intrinsic bailed out
208 if (jvms->has_method()) {
209 // Not a root compile.
210 const char* msg = "failed to generate predicate for intrinsic";
211 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
212 if (C->print_intrinsics() || C->print_inlining()) {
213 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
214 }
215 } else {
216 // Root compile
217 ResourceMark rm;
218 stringStream msg_stream;
219 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
220 vmIntrinsics::name_at(intrinsic_id()),
221 is_virtual() ? " (virtual)" : "", bci);
222 const char *msg = msg_stream.freeze();
223 log_debug(jit, inlining)("%s", msg);
224 if (C->print_intrinsics() || C->print_inlining()) {
225 C->print_inlining_stream()->print("%s", msg);
226 }
227 }
228 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
229 return nullptr;
230 }
231
232 bool LibraryCallKit::try_to_inline(int predicate) {
233 // Handle symbolic names for otherwise undistinguished boolean switches:
234 const bool is_store = true;
235 const bool is_compress = true;
236 const bool is_static = true;
237 const bool is_volatile = true;
238
239 if (!jvms()->has_method()) {
240 // Root JVMState has a null method.
241 assert(map()->memory()->Opcode() == Op_Parm, "");
242 // Insert the memory aliasing node
243 set_all_memory(reset_memory());
244 }
245 assert(merged_memory(), "");
246
247 switch (intrinsic_id()) {
248 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
249 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
250 case vmIntrinsics::_getClass: return inline_native_getClass();
251
252 case vmIntrinsics::_ceil:
253 case vmIntrinsics::_floor:
254 case vmIntrinsics::_rint:
255 case vmIntrinsics::_dsin:
256 case vmIntrinsics::_dcos:
257 case vmIntrinsics::_dtan:
258 case vmIntrinsics::_dtanh:
259 case vmIntrinsics::_dabs:
260 case vmIntrinsics::_fabs:
261 case vmIntrinsics::_iabs:
262 case vmIntrinsics::_labs:
263 case vmIntrinsics::_datan2:
264 case vmIntrinsics::_dsqrt:
265 case vmIntrinsics::_dsqrt_strict:
266 case vmIntrinsics::_dexp:
267 case vmIntrinsics::_dlog:
268 case vmIntrinsics::_dlog10:
269 case vmIntrinsics::_dpow:
270 case vmIntrinsics::_dcopySign:
271 case vmIntrinsics::_fcopySign:
272 case vmIntrinsics::_dsignum:
273 case vmIntrinsics::_roundF:
274 case vmIntrinsics::_roundD:
275 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
276
277 case vmIntrinsics::_notify:
278 case vmIntrinsics::_notifyAll:
279 return inline_notify(intrinsic_id());
280
281 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
282 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
283 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
284 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
285 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
286 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
287 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
288 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
289 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
290 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
291 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
292 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
293 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
294 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
295
296 case vmIntrinsics::_arraycopy: return inline_arraycopy();
297
298 case vmIntrinsics::_arraySort: return inline_array_sort();
299 case vmIntrinsics::_arrayPartition: return inline_array_partition();
300
301 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
302 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
303 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
304 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
305
306 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
307 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
308 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
309 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
310 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
311 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
312 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
313 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
314
315 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
316
317 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
318
319 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
320 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
321 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
322 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
323
324 case vmIntrinsics::_compressStringC:
325 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
326 case vmIntrinsics::_inflateStringC:
327 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
328
329 case vmIntrinsics::_makePrivateBuffer: return inline_unsafe_make_private_buffer();
330 case vmIntrinsics::_finishPrivateBuffer: return inline_unsafe_finish_private_buffer();
331 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
332 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
333 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
334 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
335 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
336 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
337 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
338 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
339 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
340 case vmIntrinsics::_getValue: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false, true);
341
342 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
343 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
344 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
345 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
346 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
347 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
348 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
349 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
350 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
351 case vmIntrinsics::_putValue: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false, true);
352
353 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
354 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
355 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
356 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
357 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
358 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
359 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
360 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
361 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
362
363 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
364 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
365 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
366 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
367 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
368 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
369 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
370 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
371 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
372
373 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
374 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
375 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
376 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
377
378 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
379 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
380 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
381 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
382
383 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
384 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
385 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
386 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
387 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
388 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
389 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
390 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
391 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
392
393 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
394 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
395 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
396 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
397 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
398 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
399 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
400 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
401 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
402
403 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
404 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
405 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
406 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
407 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
408 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
409 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
410 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
411 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
412
413 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
414 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
415 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
416 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
417 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
418 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
419 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
420 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
421 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
422
423 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
424 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
425 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
426 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
427 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
428
429 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
430 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
431 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
432 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
433 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
434 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
435 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
436 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
437 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
438 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
439 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
440 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
441 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
442 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
443 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
444 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
445 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
446 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
447 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
448 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
449
450 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
451 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
452 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
453 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
454 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
455 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
456 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
457 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
458 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
459 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
460 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
461 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
462 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
463 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
464 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
465
466 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
467 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
468 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
469 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
470
471 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
472 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
473 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
474 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
475 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
476
477 case vmIntrinsics::_loadFence:
478 case vmIntrinsics::_storeFence:
479 case vmIntrinsics::_storeStoreFence:
480 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
481
482 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
483
484 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
485 case vmIntrinsics::_currentThread: return inline_native_currentThread();
486 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
487
488 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
489 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
490
491 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
492 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
493
494 #if INCLUDE_JVMTI
495 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
496 "notifyJvmtiStart", true, false);
497 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
498 "notifyJvmtiEnd", false, true);
499 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
500 "notifyJvmtiMount", false, false);
501 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
502 "notifyJvmtiUnmount", false, false);
503 case vmIntrinsics::_notifyJvmtiVThreadHideFrames: return inline_native_notify_jvmti_hide();
504 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
505 #endif
506
507 #ifdef JFR_HAVE_INTRINSICS
508 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
509 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
510 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
511 #endif
512 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
513 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
514 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
515 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
516 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
517 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
518 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
519 case vmIntrinsics::_isFlatArray: return inline_unsafe_isFlatArray();
520 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
521 case vmIntrinsics::_getLength: return inline_native_getLength();
522 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
523 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
524 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
525 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
526 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
527 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
528 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
529
530 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
531 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
532 case vmIntrinsics::_newNullRestrictedArray: return inline_newNullRestrictedArray();
533
534 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
535
536 case vmIntrinsics::_isInstance:
537 case vmIntrinsics::_getModifiers:
538 case vmIntrinsics::_isInterface:
539 case vmIntrinsics::_isArray:
540 case vmIntrinsics::_isPrimitive:
541 case vmIntrinsics::_isHidden:
542 case vmIntrinsics::_getSuperclass:
543 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
544
545 case vmIntrinsics::_floatToRawIntBits:
546 case vmIntrinsics::_floatToIntBits:
547 case vmIntrinsics::_intBitsToFloat:
548 case vmIntrinsics::_doubleToRawLongBits:
549 case vmIntrinsics::_doubleToLongBits:
550 case vmIntrinsics::_longBitsToDouble:
551 case vmIntrinsics::_floatToFloat16:
552 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
553
554 case vmIntrinsics::_floatIsFinite:
555 case vmIntrinsics::_floatIsInfinite:
556 case vmIntrinsics::_doubleIsFinite:
557 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
558
559 case vmIntrinsics::_numberOfLeadingZeros_i:
560 case vmIntrinsics::_numberOfLeadingZeros_l:
561 case vmIntrinsics::_numberOfTrailingZeros_i:
562 case vmIntrinsics::_numberOfTrailingZeros_l:
563 case vmIntrinsics::_bitCount_i:
564 case vmIntrinsics::_bitCount_l:
565 case vmIntrinsics::_reverse_i:
566 case vmIntrinsics::_reverse_l:
567 case vmIntrinsics::_reverseBytes_i:
568 case vmIntrinsics::_reverseBytes_l:
569 case vmIntrinsics::_reverseBytes_s:
570 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
571
572 case vmIntrinsics::_compress_i:
573 case vmIntrinsics::_compress_l:
574 case vmIntrinsics::_expand_i:
575 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
576
577 case vmIntrinsics::_compareUnsigned_i:
578 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
579
580 case vmIntrinsics::_divideUnsigned_i:
581 case vmIntrinsics::_divideUnsigned_l:
582 case vmIntrinsics::_remainderUnsigned_i:
583 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
584
585 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
586
587 case vmIntrinsics::_Reference_get: return inline_reference_get();
588 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
589 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
590 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
591 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
592
593 case vmIntrinsics::_Class_cast: return inline_Class_cast();
594
595 case vmIntrinsics::_aescrypt_encryptBlock:
596 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
597
598 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
599 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
600 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
601
602 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
603 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
604 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
605
606 case vmIntrinsics::_counterMode_AESCrypt:
607 return inline_counterMode_AESCrypt(intrinsic_id());
608
609 case vmIntrinsics::_galoisCounterMode_AESCrypt:
610 return inline_galoisCounterMode_AESCrypt();
611
612 case vmIntrinsics::_md5_implCompress:
613 case vmIntrinsics::_sha_implCompress:
614 case vmIntrinsics::_sha2_implCompress:
615 case vmIntrinsics::_sha5_implCompress:
616 case vmIntrinsics::_sha3_implCompress:
617 return inline_digestBase_implCompress(intrinsic_id());
618
619 case vmIntrinsics::_digestBase_implCompressMB:
620 return inline_digestBase_implCompressMB(predicate);
621
622 case vmIntrinsics::_multiplyToLen:
623 return inline_multiplyToLen();
624
625 case vmIntrinsics::_squareToLen:
626 return inline_squareToLen();
627
628 case vmIntrinsics::_mulAdd:
629 return inline_mulAdd();
630
631 case vmIntrinsics::_montgomeryMultiply:
632 return inline_montgomeryMultiply();
633 case vmIntrinsics::_montgomerySquare:
634 return inline_montgomerySquare();
635
636 case vmIntrinsics::_bigIntegerRightShiftWorker:
637 return inline_bigIntegerShift(true);
638 case vmIntrinsics::_bigIntegerLeftShiftWorker:
639 return inline_bigIntegerShift(false);
640
641 case vmIntrinsics::_vectorizedMismatch:
642 return inline_vectorizedMismatch();
643
644 case vmIntrinsics::_ghash_processBlocks:
645 return inline_ghash_processBlocks();
646 case vmIntrinsics::_chacha20Block:
647 return inline_chacha20Block();
648 case vmIntrinsics::_base64_encodeBlock:
649 return inline_base64_encodeBlock();
650 case vmIntrinsics::_base64_decodeBlock:
651 return inline_base64_decodeBlock();
652 case vmIntrinsics::_poly1305_processBlocks:
653 return inline_poly1305_processBlocks();
654 case vmIntrinsics::_intpoly_montgomeryMult_P256:
655 return inline_intpoly_montgomeryMult_P256();
656 case vmIntrinsics::_intpoly_assign:
657 return inline_intpoly_assign();
658 case vmIntrinsics::_encodeISOArray:
659 case vmIntrinsics::_encodeByteISOArray:
660 return inline_encodeISOArray(false);
661 case vmIntrinsics::_encodeAsciiArray:
662 return inline_encodeISOArray(true);
663
664 case vmIntrinsics::_updateCRC32:
665 return inline_updateCRC32();
666 case vmIntrinsics::_updateBytesCRC32:
667 return inline_updateBytesCRC32();
668 case vmIntrinsics::_updateByteBufferCRC32:
669 return inline_updateByteBufferCRC32();
670
671 case vmIntrinsics::_updateBytesCRC32C:
672 return inline_updateBytesCRC32C();
673 case vmIntrinsics::_updateDirectByteBufferCRC32C:
674 return inline_updateDirectByteBufferCRC32C();
675
676 case vmIntrinsics::_updateBytesAdler32:
677 return inline_updateBytesAdler32();
678 case vmIntrinsics::_updateByteBufferAdler32:
679 return inline_updateByteBufferAdler32();
680
681 case vmIntrinsics::_profileBoolean:
682 return inline_profileBoolean();
683 case vmIntrinsics::_isCompileConstant:
684 return inline_isCompileConstant();
685
686 case vmIntrinsics::_countPositives:
687 return inline_countPositives();
688
689 case vmIntrinsics::_fmaD:
690 case vmIntrinsics::_fmaF:
691 return inline_fma(intrinsic_id());
692
693 case vmIntrinsics::_isDigit:
694 case vmIntrinsics::_isLowerCase:
695 case vmIntrinsics::_isUpperCase:
696 case vmIntrinsics::_isWhitespace:
697 return inline_character_compare(intrinsic_id());
698
699 case vmIntrinsics::_min:
700 case vmIntrinsics::_max:
701 case vmIntrinsics::_min_strict:
702 case vmIntrinsics::_max_strict:
703 return inline_min_max(intrinsic_id());
704
705 case vmIntrinsics::_maxF:
706 case vmIntrinsics::_minF:
707 case vmIntrinsics::_maxD:
708 case vmIntrinsics::_minD:
709 case vmIntrinsics::_maxF_strict:
710 case vmIntrinsics::_minF_strict:
711 case vmIntrinsics::_maxD_strict:
712 case vmIntrinsics::_minD_strict:
713 return inline_fp_min_max(intrinsic_id());
714
715 case vmIntrinsics::_VectorUnaryOp:
716 return inline_vector_nary_operation(1);
717 case vmIntrinsics::_VectorBinaryOp:
718 return inline_vector_nary_operation(2);
719 case vmIntrinsics::_VectorTernaryOp:
720 return inline_vector_nary_operation(3);
721 case vmIntrinsics::_VectorFromBitsCoerced:
722 return inline_vector_frombits_coerced();
723 case vmIntrinsics::_VectorShuffleIota:
724 return inline_vector_shuffle_iota();
725 case vmIntrinsics::_VectorMaskOp:
726 return inline_vector_mask_operation();
727 case vmIntrinsics::_VectorShuffleToVector:
728 return inline_vector_shuffle_to_vector();
729 case vmIntrinsics::_VectorWrapShuffleIndexes:
730 return inline_vector_wrap_shuffle_indexes();
731 case vmIntrinsics::_VectorLoadOp:
732 return inline_vector_mem_operation(/*is_store=*/false);
733 case vmIntrinsics::_VectorLoadMaskedOp:
734 return inline_vector_mem_masked_operation(/*is_store*/false);
735 case vmIntrinsics::_VectorStoreOp:
736 return inline_vector_mem_operation(/*is_store=*/true);
737 case vmIntrinsics::_VectorStoreMaskedOp:
738 return inline_vector_mem_masked_operation(/*is_store=*/true);
739 case vmIntrinsics::_VectorGatherOp:
740 return inline_vector_gather_scatter(/*is_scatter*/ false);
741 case vmIntrinsics::_VectorScatterOp:
742 return inline_vector_gather_scatter(/*is_scatter*/ true);
743 case vmIntrinsics::_VectorReductionCoerced:
744 return inline_vector_reduction();
745 case vmIntrinsics::_VectorTest:
746 return inline_vector_test();
747 case vmIntrinsics::_VectorBlend:
748 return inline_vector_blend();
749 case vmIntrinsics::_VectorRearrange:
750 return inline_vector_rearrange();
751 case vmIntrinsics::_VectorSelectFrom:
752 return inline_vector_select_from();
753 case vmIntrinsics::_VectorCompare:
754 return inline_vector_compare();
755 case vmIntrinsics::_VectorBroadcastInt:
756 return inline_vector_broadcast_int();
757 case vmIntrinsics::_VectorConvert:
758 return inline_vector_convert();
759 case vmIntrinsics::_VectorInsert:
760 return inline_vector_insert();
761 case vmIntrinsics::_VectorExtract:
762 return inline_vector_extract();
763 case vmIntrinsics::_VectorCompressExpand:
764 return inline_vector_compress_expand();
765 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
766 return inline_vector_select_from_two_vectors();
767 case vmIntrinsics::_IndexVector:
768 return inline_index_vector();
769 case vmIntrinsics::_IndexPartiallyInUpperRange:
770 return inline_index_partially_in_upper_range();
771
772 case vmIntrinsics::_getObjectSize:
773 return inline_getObjectSize();
774
775 case vmIntrinsics::_blackhole:
776 return inline_blackhole();
777
778 default:
779 // If you get here, it may be that someone has added a new intrinsic
780 // to the list in vmIntrinsics.hpp without implementing it here.
781 #ifndef PRODUCT
782 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
783 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
784 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
785 }
786 #endif
787 return false;
788 }
789 }
790
791 Node* LibraryCallKit::try_to_predicate(int predicate) {
792 if (!jvms()->has_method()) {
793 // Root JVMState has a null method.
794 assert(map()->memory()->Opcode() == Op_Parm, "");
795 // Insert the memory aliasing node
796 set_all_memory(reset_memory());
797 }
798 assert(merged_memory(), "");
799
800 switch (intrinsic_id()) {
801 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
802 return inline_cipherBlockChaining_AESCrypt_predicate(false);
803 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
804 return inline_cipherBlockChaining_AESCrypt_predicate(true);
805 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
806 return inline_electronicCodeBook_AESCrypt_predicate(false);
807 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
808 return inline_electronicCodeBook_AESCrypt_predicate(true);
809 case vmIntrinsics::_counterMode_AESCrypt:
810 return inline_counterMode_AESCrypt_predicate();
811 case vmIntrinsics::_digestBase_implCompressMB:
812 return inline_digestBase_implCompressMB_predicate(predicate);
813 case vmIntrinsics::_galoisCounterMode_AESCrypt:
814 return inline_galoisCounterMode_AESCrypt_predicate();
815
816 default:
817 // If you get here, it may be that someone has added a new intrinsic
818 // to the list in vmIntrinsics.hpp without implementing it here.
819 #ifndef PRODUCT
820 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
821 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
822 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
823 }
824 #endif
825 Node* slow_ctl = control();
826 set_control(top()); // No fast path intrinsic
827 return slow_ctl;
828 }
829 }
830
831 //------------------------------set_result-------------------------------
832 // Helper function for finishing intrinsics.
833 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
834 record_for_igvn(region);
835 set_control(_gvn.transform(region));
836 set_result( _gvn.transform(value));
837 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
838 }
839
840 //------------------------------generate_guard---------------------------
841 // Helper function for generating guarded fast-slow graph structures.
842 // The given 'test', if true, guards a slow path. If the test fails
843 // then a fast path can be taken. (We generally hope it fails.)
844 // In all cases, GraphKit::control() is updated to the fast path.
845 // The returned value represents the control for the slow path.
846 // The return value is never 'top'; it is either a valid control
847 // or null if it is obvious that the slow path can never be taken.
848 // Also, if region and the slow control are not null, the slow edge
849 // is appended to the region.
850 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
851 if (stopped()) {
852 // Already short circuited.
853 return nullptr;
854 }
855
856 // Build an if node and its projections.
857 // If test is true we take the slow path, which we assume is uncommon.
858 if (_gvn.type(test) == TypeInt::ZERO) {
859 // The slow branch is never taken. No need to build this guard.
860 return nullptr;
861 }
862
863 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
864
865 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
866 if (if_slow == top()) {
867 // The slow branch is never taken. No need to build this guard.
868 return nullptr;
869 }
870
871 if (region != nullptr)
872 region->add_req(if_slow);
873
874 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
875 set_control(if_fast);
876
877 return if_slow;
878 }
879
880 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
881 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
882 }
883 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
884 return generate_guard(test, region, PROB_FAIR);
885 }
886
887 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
888 Node* *pos_index) {
889 if (stopped())
890 return nullptr; // already stopped
891 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
892 return nullptr; // index is already adequately typed
893 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
894 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
895 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
896 if (is_neg != nullptr && pos_index != nullptr) {
897 // Emulate effect of Parse::adjust_map_after_if.
898 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
899 (*pos_index) = _gvn.transform(ccast);
900 }
901 return is_neg;
902 }
903
904 // Make sure that 'position' is a valid limit index, in [0..length].
905 // There are two equivalent plans for checking this:
906 // A. (offset + copyLength) unsigned<= arrayLength
907 // B. offset <= (arrayLength - copyLength)
908 // We require that all of the values above, except for the sum and
909 // difference, are already known to be non-negative.
910 // Plan A is robust in the face of overflow, if offset and copyLength
911 // are both hugely positive.
912 //
913 // Plan B is less direct and intuitive, but it does not overflow at
914 // all, since the difference of two non-negatives is always
915 // representable. Whenever Java methods must perform the equivalent
916 // check they generally use Plan B instead of Plan A.
917 // For the moment we use Plan A.
918 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
919 Node* subseq_length,
920 Node* array_length,
921 RegionNode* region) {
922 if (stopped())
923 return nullptr; // already stopped
924 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
925 if (zero_offset && subseq_length->eqv_uncast(array_length))
926 return nullptr; // common case of whole-array copy
927 Node* last = subseq_length;
928 if (!zero_offset) // last += offset
929 last = _gvn.transform(new AddINode(last, offset));
930 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
931 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
932 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
933 return is_over;
934 }
935
936 // Emit range checks for the given String.value byte array
937 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
938 if (stopped()) {
939 return; // already stopped
940 }
941 RegionNode* bailout = new RegionNode(1);
942 record_for_igvn(bailout);
943 if (char_count) {
944 // Convert char count to byte count
945 count = _gvn.transform(new LShiftINode(count, intcon(1)));
946 }
947
948 // Offset and count must not be negative
949 generate_negative_guard(offset, bailout);
950 generate_negative_guard(count, bailout);
951 // Offset + count must not exceed length of array
952 generate_limit_guard(offset, count, load_array_length(array), bailout);
953
954 if (bailout->req() > 1) {
955 PreserveJVMState pjvms(this);
956 set_control(_gvn.transform(bailout));
957 uncommon_trap(Deoptimization::Reason_intrinsic,
958 Deoptimization::Action_maybe_recompile);
959 }
960 }
961
962 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
963 bool is_immutable) {
964 ciKlass* thread_klass = env()->Thread_klass();
965 const Type* thread_type
966 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
967
968 Node* thread = _gvn.transform(new ThreadLocalNode());
969 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
970 tls_output = thread;
971
972 Node* thread_obj_handle
973 = (is_immutable
974 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
975 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
976 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
977 thread_obj_handle = _gvn.transform(thread_obj_handle);
978
979 DecoratorSet decorators = IN_NATIVE;
980 if (is_immutable) {
981 decorators |= C2_IMMUTABLE_MEMORY;
982 }
983 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
984 }
985
986 //--------------------------generate_current_thread--------------------
987 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
988 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
989 /*is_immutable*/false);
990 }
991
992 //--------------------------generate_virtual_thread--------------------
993 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
994 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
995 !C->method()->changes_current_thread());
996 }
997
998 //------------------------------make_string_method_node------------------------
999 // Helper method for String intrinsic functions. This version is called with
1000 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1001 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1002 // containing the lengths of str1 and str2.
1003 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1004 Node* result = nullptr;
1005 switch (opcode) {
1006 case Op_StrIndexOf:
1007 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1008 str1_start, cnt1, str2_start, cnt2, ae);
1009 break;
1010 case Op_StrComp:
1011 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1012 str1_start, cnt1, str2_start, cnt2, ae);
1013 break;
1014 case Op_StrEquals:
1015 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1016 // Use the constant length if there is one because optimized match rule may exist.
1017 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1018 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1019 break;
1020 default:
1021 ShouldNotReachHere();
1022 return nullptr;
1023 }
1024
1025 // All these intrinsics have checks.
1026 C->set_has_split_ifs(true); // Has chance for split-if optimization
1027 clear_upper_avx();
1028
1029 return _gvn.transform(result);
1030 }
1031
1032 //------------------------------inline_string_compareTo------------------------
1033 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1034 Node* arg1 = argument(0);
1035 Node* arg2 = argument(1);
1036
1037 arg1 = must_be_not_null(arg1, true);
1038 arg2 = must_be_not_null(arg2, true);
1039
1040 // Get start addr and length of first argument
1041 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1042 Node* arg1_cnt = load_array_length(arg1);
1043
1044 // Get start addr and length of second argument
1045 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1046 Node* arg2_cnt = load_array_length(arg2);
1047
1048 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1049 set_result(result);
1050 return true;
1051 }
1052
1053 //------------------------------inline_string_equals------------------------
1054 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1055 Node* arg1 = argument(0);
1056 Node* arg2 = argument(1);
1057
1058 // paths (plus control) merge
1059 RegionNode* region = new RegionNode(3);
1060 Node* phi = new PhiNode(region, TypeInt::BOOL);
1061
1062 if (!stopped()) {
1063
1064 arg1 = must_be_not_null(arg1, true);
1065 arg2 = must_be_not_null(arg2, true);
1066
1067 // Get start addr and length of first argument
1068 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1069 Node* arg1_cnt = load_array_length(arg1);
1070
1071 // Get start addr and length of second argument
1072 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1073 Node* arg2_cnt = load_array_length(arg2);
1074
1075 // Check for arg1_cnt != arg2_cnt
1076 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1077 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1078 Node* if_ne = generate_slow_guard(bol, nullptr);
1079 if (if_ne != nullptr) {
1080 phi->init_req(2, intcon(0));
1081 region->init_req(2, if_ne);
1082 }
1083
1084 // Check for count == 0 is done by assembler code for StrEquals.
1085
1086 if (!stopped()) {
1087 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1088 phi->init_req(1, equals);
1089 region->init_req(1, control());
1090 }
1091 }
1092
1093 // post merge
1094 set_control(_gvn.transform(region));
1095 record_for_igvn(region);
1096
1097 set_result(_gvn.transform(phi));
1098 return true;
1099 }
1100
1101 //------------------------------inline_array_equals----------------------------
1102 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1103 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1104 Node* arg1 = argument(0);
1105 Node* arg2 = argument(1);
1106
1107 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1108 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1109 clear_upper_avx();
1110
1111 return true;
1112 }
1113
1114
1115 //------------------------------inline_countPositives------------------------------
1116 bool LibraryCallKit::inline_countPositives() {
1117 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1118 return false;
1119 }
1120
1121 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1122 // no receiver since it is static method
1123 Node* ba = argument(0);
1124 Node* offset = argument(1);
1125 Node* len = argument(2);
1126
1127 ba = must_be_not_null(ba, true);
1128
1129 // Range checks
1130 generate_string_range_check(ba, offset, len, false);
1131 if (stopped()) {
1132 return true;
1133 }
1134 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1135 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1136 set_result(_gvn.transform(result));
1137 clear_upper_avx();
1138 return true;
1139 }
1140
1141 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1142 Node* index = argument(0);
1143 Node* length = bt == T_INT ? argument(1) : argument(2);
1144 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1145 return false;
1146 }
1147
1148 // check that length is positive
1149 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1150 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1151
1152 {
1153 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1154 uncommon_trap(Deoptimization::Reason_intrinsic,
1155 Deoptimization::Action_make_not_entrant);
1156 }
1157
1158 if (stopped()) {
1159 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1160 return true;
1161 }
1162
1163 // length is now known positive, add a cast node to make this explicit
1164 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1165 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1166 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1167 ConstraintCastNode::RegularDependency, bt);
1168 casted_length = _gvn.transform(casted_length);
1169 replace_in_map(length, casted_length);
1170 length = casted_length;
1171
1172 // Use an unsigned comparison for the range check itself
1173 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1174 BoolTest::mask btest = BoolTest::lt;
1175 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1176 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1177 _gvn.set_type(rc, rc->Value(&_gvn));
1178 if (!rc_bool->is_Con()) {
1179 record_for_igvn(rc);
1180 }
1181 set_control(_gvn.transform(new IfTrueNode(rc)));
1182 {
1183 PreserveJVMState pjvms(this);
1184 set_control(_gvn.transform(new IfFalseNode(rc)));
1185 uncommon_trap(Deoptimization::Reason_range_check,
1186 Deoptimization::Action_make_not_entrant);
1187 }
1188
1189 if (stopped()) {
1190 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1191 return true;
1192 }
1193
1194 // index is now known to be >= 0 and < length, cast it
1195 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1196 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1197 ConstraintCastNode::RegularDependency, bt);
1198 result = _gvn.transform(result);
1199 set_result(result);
1200 replace_in_map(index, result);
1201 return true;
1202 }
1203
1204 //------------------------------inline_string_indexOf------------------------
1205 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1206 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1207 return false;
1208 }
1209 Node* src = argument(0);
1210 Node* tgt = argument(1);
1211
1212 // Make the merge point
1213 RegionNode* result_rgn = new RegionNode(4);
1214 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1215
1216 src = must_be_not_null(src, true);
1217 tgt = must_be_not_null(tgt, true);
1218
1219 // Get start addr and length of source string
1220 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1221 Node* src_count = load_array_length(src);
1222
1223 // Get start addr and length of substring
1224 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1225 Node* tgt_count = load_array_length(tgt);
1226
1227 Node* result = nullptr;
1228 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1229
1230 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1231 // Divide src size by 2 if String is UTF16 encoded
1232 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1233 }
1234 if (ae == StrIntrinsicNode::UU) {
1235 // Divide substring size by 2 if String is UTF16 encoded
1236 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1237 }
1238
1239 if (call_opt_stub) {
1240 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1241 StubRoutines::_string_indexof_array[ae],
1242 "stringIndexOf", TypePtr::BOTTOM, src_start,
1243 src_count, tgt_start, tgt_count);
1244 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1245 } else {
1246 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1247 result_rgn, result_phi, ae);
1248 }
1249 if (result != nullptr) {
1250 result_phi->init_req(3, result);
1251 result_rgn->init_req(3, control());
1252 }
1253 set_control(_gvn.transform(result_rgn));
1254 record_for_igvn(result_rgn);
1255 set_result(_gvn.transform(result_phi));
1256
1257 return true;
1258 }
1259
1260 //-----------------------------inline_string_indexOfI-----------------------
1261 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1262 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1263 return false;
1264 }
1265 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1266 return false;
1267 }
1268
1269 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1270 Node* src = argument(0); // byte[]
1271 Node* src_count = argument(1); // char count
1272 Node* tgt = argument(2); // byte[]
1273 Node* tgt_count = argument(3); // char count
1274 Node* from_index = argument(4); // char index
1275
1276 src = must_be_not_null(src, true);
1277 tgt = must_be_not_null(tgt, true);
1278
1279 // Multiply byte array index by 2 if String is UTF16 encoded
1280 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1281 src_count = _gvn.transform(new SubINode(src_count, from_index));
1282 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1283 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1284
1285 // Range checks
1286 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1287 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1288 if (stopped()) {
1289 return true;
1290 }
1291
1292 RegionNode* region = new RegionNode(5);
1293 Node* phi = new PhiNode(region, TypeInt::INT);
1294 Node* result = nullptr;
1295
1296 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1297
1298 if (call_opt_stub) {
1299 assert(arrayOopDesc::base_offset_in_bytes(T_BYTE) >= 16, "Needed for indexOf");
1300 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1301 StubRoutines::_string_indexof_array[ae],
1302 "stringIndexOf", TypePtr::BOTTOM, src_start,
1303 src_count, tgt_start, tgt_count);
1304 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1305 } else {
1306 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1307 region, phi, ae);
1308 }
1309 if (result != nullptr) {
1310 // The result is index relative to from_index if substring was found, -1 otherwise.
1311 // Generate code which will fold into cmove.
1312 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1313 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1314
1315 Node* if_lt = generate_slow_guard(bol, nullptr);
1316 if (if_lt != nullptr) {
1317 // result == -1
1318 phi->init_req(3, result);
1319 region->init_req(3, if_lt);
1320 }
1321 if (!stopped()) {
1322 result = _gvn.transform(new AddINode(result, from_index));
1323 phi->init_req(4, result);
1324 region->init_req(4, control());
1325 }
1326 }
1327
1328 set_control(_gvn.transform(region));
1329 record_for_igvn(region);
1330 set_result(_gvn.transform(phi));
1331 clear_upper_avx();
1332
1333 return true;
1334 }
1335
1336 // Create StrIndexOfNode with fast path checks
1337 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1338 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1339 // Check for substr count > string count
1340 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1341 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1342 Node* if_gt = generate_slow_guard(bol, nullptr);
1343 if (if_gt != nullptr) {
1344 phi->init_req(1, intcon(-1));
1345 region->init_req(1, if_gt);
1346 }
1347 if (!stopped()) {
1348 // Check for substr count == 0
1349 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1350 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1351 Node* if_zero = generate_slow_guard(bol, nullptr);
1352 if (if_zero != nullptr) {
1353 phi->init_req(2, intcon(0));
1354 region->init_req(2, if_zero);
1355 }
1356 }
1357 if (!stopped()) {
1358 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1359 }
1360 return nullptr;
1361 }
1362
1363 //-----------------------------inline_string_indexOfChar-----------------------
1364 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1365 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1366 return false;
1367 }
1368 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1369 return false;
1370 }
1371 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1372 Node* src = argument(0); // byte[]
1373 Node* int_ch = argument(1);
1374 Node* from_index = argument(2);
1375 Node* max = argument(3);
1376
1377 src = must_be_not_null(src, true);
1378
1379 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1380 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1381 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1382
1383 // Range checks
1384 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1385
1386 // Check for int_ch >= 0
1387 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1388 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1389 {
1390 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1391 uncommon_trap(Deoptimization::Reason_intrinsic,
1392 Deoptimization::Action_maybe_recompile);
1393 }
1394 if (stopped()) {
1395 return true;
1396 }
1397
1398 RegionNode* region = new RegionNode(3);
1399 Node* phi = new PhiNode(region, TypeInt::INT);
1400
1401 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1402 C->set_has_split_ifs(true); // Has chance for split-if optimization
1403 _gvn.transform(result);
1404
1405 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1406 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1407
1408 Node* if_lt = generate_slow_guard(bol, nullptr);
1409 if (if_lt != nullptr) {
1410 // result == -1
1411 phi->init_req(2, result);
1412 region->init_req(2, if_lt);
1413 }
1414 if (!stopped()) {
1415 result = _gvn.transform(new AddINode(result, from_index));
1416 phi->init_req(1, result);
1417 region->init_req(1, control());
1418 }
1419 set_control(_gvn.transform(region));
1420 record_for_igvn(region);
1421 set_result(_gvn.transform(phi));
1422 clear_upper_avx();
1423
1424 return true;
1425 }
1426 //---------------------------inline_string_copy---------------------
1427 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1428 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1429 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1430 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1431 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1432 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1433 bool LibraryCallKit::inline_string_copy(bool compress) {
1434 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1435 return false;
1436 }
1437 int nargs = 5; // 2 oops, 3 ints
1438 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1439
1440 Node* src = argument(0);
1441 Node* src_offset = argument(1);
1442 Node* dst = argument(2);
1443 Node* dst_offset = argument(3);
1444 Node* length = argument(4);
1445
1446 // Check for allocation before we add nodes that would confuse
1447 // tightly_coupled_allocation()
1448 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1449
1450 // Figure out the size and type of the elements we will be copying.
1451 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1452 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1453 if (src_type == nullptr || dst_type == nullptr) {
1454 return false;
1455 }
1456 BasicType src_elem = src_type->elem()->array_element_basic_type();
1457 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1458 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1459 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1460 "Unsupported array types for inline_string_copy");
1461
1462 src = must_be_not_null(src, true);
1463 dst = must_be_not_null(dst, true);
1464
1465 // Convert char[] offsets to byte[] offsets
1466 bool convert_src = (compress && src_elem == T_BYTE);
1467 bool convert_dst = (!compress && dst_elem == T_BYTE);
1468 if (convert_src) {
1469 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1470 } else if (convert_dst) {
1471 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1472 }
1473
1474 // Range checks
1475 generate_string_range_check(src, src_offset, length, convert_src);
1476 generate_string_range_check(dst, dst_offset, length, convert_dst);
1477 if (stopped()) {
1478 return true;
1479 }
1480
1481 Node* src_start = array_element_address(src, src_offset, src_elem);
1482 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1483 // 'src_start' points to src array + scaled offset
1484 // 'dst_start' points to dst array + scaled offset
1485 Node* count = nullptr;
1486 if (compress) {
1487 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1488 } else {
1489 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1490 }
1491
1492 if (alloc != nullptr) {
1493 if (alloc->maybe_set_complete(&_gvn)) {
1494 // "You break it, you buy it."
1495 InitializeNode* init = alloc->initialization();
1496 assert(init->is_complete(), "we just did this");
1497 init->set_complete_with_arraycopy();
1498 assert(dst->is_CheckCastPP(), "sanity");
1499 assert(dst->in(0)->in(0) == init, "dest pinned");
1500 }
1501 // Do not let stores that initialize this object be reordered with
1502 // a subsequent store that would make this object accessible by
1503 // other threads.
1504 // Record what AllocateNode this StoreStore protects so that
1505 // escape analysis can go from the MemBarStoreStoreNode to the
1506 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1507 // based on the escape status of the AllocateNode.
1508 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1509 }
1510 if (compress) {
1511 set_result(_gvn.transform(count));
1512 }
1513 clear_upper_avx();
1514
1515 return true;
1516 }
1517
1518 #ifdef _LP64
1519 #define XTOP ,top() /*additional argument*/
1520 #else //_LP64
1521 #define XTOP /*no additional argument*/
1522 #endif //_LP64
1523
1524 //------------------------inline_string_toBytesU--------------------------
1525 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1526 bool LibraryCallKit::inline_string_toBytesU() {
1527 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1528 return false;
1529 }
1530 // Get the arguments.
1531 Node* value = argument(0);
1532 Node* offset = argument(1);
1533 Node* length = argument(2);
1534
1535 Node* newcopy = nullptr;
1536
1537 // Set the original stack and the reexecute bit for the interpreter to reexecute
1538 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1539 { PreserveReexecuteState preexecs(this);
1540 jvms()->set_should_reexecute(true);
1541
1542 // Check if a null path was taken unconditionally.
1543 value = null_check(value);
1544
1545 RegionNode* bailout = new RegionNode(1);
1546 record_for_igvn(bailout);
1547
1548 // Range checks
1549 generate_negative_guard(offset, bailout);
1550 generate_negative_guard(length, bailout);
1551 generate_limit_guard(offset, length, load_array_length(value), bailout);
1552 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1553 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1554
1555 if (bailout->req() > 1) {
1556 PreserveJVMState pjvms(this);
1557 set_control(_gvn.transform(bailout));
1558 uncommon_trap(Deoptimization::Reason_intrinsic,
1559 Deoptimization::Action_maybe_recompile);
1560 }
1561 if (stopped()) {
1562 return true;
1563 }
1564
1565 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1566 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1567 newcopy = new_array(klass_node, size, 0); // no arguments to push
1568 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1569 guarantee(alloc != nullptr, "created above");
1570
1571 // Calculate starting addresses.
1572 Node* src_start = array_element_address(value, offset, T_CHAR);
1573 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1574
1575 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1576 const TypeInt* toffset = gvn().type(offset)->is_int();
1577 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1578
1579 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1580 const char* copyfunc_name = "arraycopy";
1581 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1582 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1583 OptoRuntime::fast_arraycopy_Type(),
1584 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1585 src_start, dst_start, ConvI2X(length) XTOP);
1586 // Do not let reads from the cloned object float above the arraycopy.
1587 if (alloc->maybe_set_complete(&_gvn)) {
1588 // "You break it, you buy it."
1589 InitializeNode* init = alloc->initialization();
1590 assert(init->is_complete(), "we just did this");
1591 init->set_complete_with_arraycopy();
1592 assert(newcopy->is_CheckCastPP(), "sanity");
1593 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1594 }
1595 // Do not let stores that initialize this object be reordered with
1596 // a subsequent store that would make this object accessible by
1597 // other threads.
1598 // Record what AllocateNode this StoreStore protects so that
1599 // escape analysis can go from the MemBarStoreStoreNode to the
1600 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1601 // based on the escape status of the AllocateNode.
1602 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1603 } // original reexecute is set back here
1604
1605 C->set_has_split_ifs(true); // Has chance for split-if optimization
1606 if (!stopped()) {
1607 set_result(newcopy);
1608 }
1609 clear_upper_avx();
1610
1611 return true;
1612 }
1613
1614 //------------------------inline_string_getCharsU--------------------------
1615 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1616 bool LibraryCallKit::inline_string_getCharsU() {
1617 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1618 return false;
1619 }
1620
1621 // Get the arguments.
1622 Node* src = argument(0);
1623 Node* src_begin = argument(1);
1624 Node* src_end = argument(2); // exclusive offset (i < src_end)
1625 Node* dst = argument(3);
1626 Node* dst_begin = argument(4);
1627
1628 // Check for allocation before we add nodes that would confuse
1629 // tightly_coupled_allocation()
1630 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1631
1632 // Check if a null path was taken unconditionally.
1633 src = null_check(src);
1634 dst = null_check(dst);
1635 if (stopped()) {
1636 return true;
1637 }
1638
1639 // Get length and convert char[] offset to byte[] offset
1640 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1641 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1642
1643 // Range checks
1644 generate_string_range_check(src, src_begin, length, true);
1645 generate_string_range_check(dst, dst_begin, length, false);
1646 if (stopped()) {
1647 return true;
1648 }
1649
1650 if (!stopped()) {
1651 // Calculate starting addresses.
1652 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1653 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1654
1655 // Check if array addresses are aligned to HeapWordSize
1656 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1657 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1658 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1659 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1660
1661 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1662 const char* copyfunc_name = "arraycopy";
1663 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1664 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1665 OptoRuntime::fast_arraycopy_Type(),
1666 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1667 src_start, dst_start, ConvI2X(length) XTOP);
1668 // Do not let reads from the cloned object float above the arraycopy.
1669 if (alloc != nullptr) {
1670 if (alloc->maybe_set_complete(&_gvn)) {
1671 // "You break it, you buy it."
1672 InitializeNode* init = alloc->initialization();
1673 assert(init->is_complete(), "we just did this");
1674 init->set_complete_with_arraycopy();
1675 assert(dst->is_CheckCastPP(), "sanity");
1676 assert(dst->in(0)->in(0) == init, "dest pinned");
1677 }
1678 // Do not let stores that initialize this object be reordered with
1679 // a subsequent store that would make this object accessible by
1680 // other threads.
1681 // Record what AllocateNode this StoreStore protects so that
1682 // escape analysis can go from the MemBarStoreStoreNode to the
1683 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1684 // based on the escape status of the AllocateNode.
1685 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1686 } else {
1687 insert_mem_bar(Op_MemBarCPUOrder);
1688 }
1689 }
1690
1691 C->set_has_split_ifs(true); // Has chance for split-if optimization
1692 return true;
1693 }
1694
1695 //----------------------inline_string_char_access----------------------------
1696 // Store/Load char to/from byte[] array.
1697 // static void StringUTF16.putChar(byte[] val, int index, int c)
1698 // static char StringUTF16.getChar(byte[] val, int index)
1699 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1700 Node* value = argument(0);
1701 Node* index = argument(1);
1702 Node* ch = is_store ? argument(2) : nullptr;
1703
1704 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1705 // correctly requires matched array shapes.
1706 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1707 "sanity: byte[] and char[] bases agree");
1708 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1709 "sanity: byte[] and char[] scales agree");
1710
1711 // Bail when getChar over constants is requested: constant folding would
1712 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1713 // Java method would constant fold nicely instead.
1714 if (!is_store && value->is_Con() && index->is_Con()) {
1715 return false;
1716 }
1717
1718 // Save state and restore on bailout
1719 uint old_sp = sp();
1720 SafePointNode* old_map = clone_map();
1721
1722 value = must_be_not_null(value, true);
1723
1724 Node* adr = array_element_address(value, index, T_CHAR);
1725 if (adr->is_top()) {
1726 set_map(old_map);
1727 set_sp(old_sp);
1728 return false;
1729 }
1730 destruct_map_clone(old_map);
1731 if (is_store) {
1732 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1733 } else {
1734 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);
1735 set_result(ch);
1736 }
1737 return true;
1738 }
1739
1740 //--------------------------round_double_node--------------------------------
1741 // Round a double node if necessary.
1742 Node* LibraryCallKit::round_double_node(Node* n) {
1743 if (Matcher::strict_fp_requires_explicit_rounding) {
1744 #ifdef IA32
1745 if (UseSSE < 2) {
1746 n = _gvn.transform(new RoundDoubleNode(nullptr, n));
1747 }
1748 #else
1749 Unimplemented();
1750 #endif // IA32
1751 }
1752 return n;
1753 }
1754
1755 //------------------------------inline_math-----------------------------------
1756 // public static double Math.abs(double)
1757 // public static double Math.sqrt(double)
1758 // public static double Math.log(double)
1759 // public static double Math.log10(double)
1760 // public static double Math.round(double)
1761 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1762 Node* arg = round_double_node(argument(0));
1763 Node* n = nullptr;
1764 switch (id) {
1765 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1766 case vmIntrinsics::_dsqrt:
1767 case vmIntrinsics::_dsqrt_strict:
1768 n = new SqrtDNode(C, control(), arg); break;
1769 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1770 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1771 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1772 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1773 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1774 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1775 default: fatal_unexpected_iid(id); break;
1776 }
1777 set_result(_gvn.transform(n));
1778 return true;
1779 }
1780
1781 //------------------------------inline_math-----------------------------------
1782 // public static float Math.abs(float)
1783 // public static int Math.abs(int)
1784 // public static long Math.abs(long)
1785 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1786 Node* arg = argument(0);
1787 Node* n = nullptr;
1788 switch (id) {
1789 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1790 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1791 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1792 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1793 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1794 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1795 default: fatal_unexpected_iid(id); break;
1796 }
1797 set_result(_gvn.transform(n));
1798 return true;
1799 }
1800
1801 //------------------------------runtime_math-----------------------------
1802 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1803 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1804 "must be (DD)D or (D)D type");
1805
1806 // Inputs
1807 Node* a = round_double_node(argument(0));
1808 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : nullptr;
1809
1810 const TypePtr* no_memory_effects = nullptr;
1811 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1812 no_memory_effects,
1813 a, top(), b, b ? top() : nullptr);
1814 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1815 #ifdef ASSERT
1816 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1817 assert(value_top == top(), "second value must be top");
1818 #endif
1819
1820 set_result(value);
1821 return true;
1822 }
1823
1824 //------------------------------inline_math_pow-----------------------------
1825 bool LibraryCallKit::inline_math_pow() {
1826 Node* exp = round_double_node(argument(2));
1827 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1828 if (d != nullptr) {
1829 if (d->getd() == 2.0) {
1830 // Special case: pow(x, 2.0) => x * x
1831 Node* base = round_double_node(argument(0));
1832 set_result(_gvn.transform(new MulDNode(base, base)));
1833 return true;
1834 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1835 // Special case: pow(x, 0.5) => sqrt(x)
1836 Node* base = round_double_node(argument(0));
1837 Node* zero = _gvn.zerocon(T_DOUBLE);
1838
1839 RegionNode* region = new RegionNode(3);
1840 Node* phi = new PhiNode(region, Type::DOUBLE);
1841
1842 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1843 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1844 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1845 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1846 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1847
1848 Node* if_pow = generate_slow_guard(test, nullptr);
1849 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1850 phi->init_req(1, value_sqrt);
1851 region->init_req(1, control());
1852
1853 if (if_pow != nullptr) {
1854 set_control(if_pow);
1855 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1856 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1857 const TypePtr* no_memory_effects = nullptr;
1858 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1859 no_memory_effects, base, top(), exp, top());
1860 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1861 #ifdef ASSERT
1862 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1863 assert(value_top == top(), "second value must be top");
1864 #endif
1865 phi->init_req(2, value_pow);
1866 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1867 }
1868
1869 C->set_has_split_ifs(true); // Has chance for split-if optimization
1870 set_control(_gvn.transform(region));
1871 record_for_igvn(region);
1872 set_result(_gvn.transform(phi));
1873
1874 return true;
1875 }
1876 }
1877
1878 return StubRoutines::dpow() != nullptr ?
1879 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1880 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1881 }
1882
1883 //------------------------------inline_math_native-----------------------------
1884 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1885 switch (id) {
1886 case vmIntrinsics::_dsin:
1887 return StubRoutines::dsin() != nullptr ?
1888 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1889 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1890 case vmIntrinsics::_dcos:
1891 return StubRoutines::dcos() != nullptr ?
1892 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1893 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1894 case vmIntrinsics::_dtan:
1895 return StubRoutines::dtan() != nullptr ?
1896 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1897 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1898 case vmIntrinsics::_dtanh:
1899 return StubRoutines::dtanh() != nullptr ?
1900 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1901 case vmIntrinsics::_dexp:
1902 return StubRoutines::dexp() != nullptr ?
1903 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1904 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1905 case vmIntrinsics::_dlog:
1906 return StubRoutines::dlog() != nullptr ?
1907 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1908 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1909 case vmIntrinsics::_dlog10:
1910 return StubRoutines::dlog10() != nullptr ?
1911 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1912 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1913
1914 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1915 case vmIntrinsics::_ceil:
1916 case vmIntrinsics::_floor:
1917 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1918
1919 case vmIntrinsics::_dsqrt:
1920 case vmIntrinsics::_dsqrt_strict:
1921 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1922 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1923 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1924 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1925 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1926
1927 case vmIntrinsics::_dpow: return inline_math_pow();
1928 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1929 case vmIntrinsics::_fcopySign: return inline_math(id);
1930 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1931 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1932 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1933
1934 // These intrinsics are not yet correctly implemented
1935 case vmIntrinsics::_datan2:
1936 return false;
1937
1938 default:
1939 fatal_unexpected_iid(id);
1940 return false;
1941 }
1942 }
1943
1944 //----------------------------inline_notify-----------------------------------*
1945 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1946 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1947 address func;
1948 if (id == vmIntrinsics::_notify) {
1949 func = OptoRuntime::monitor_notify_Java();
1950 } else {
1951 func = OptoRuntime::monitor_notifyAll_Java();
1952 }
1953 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1954 make_slow_call_ex(call, env()->Throwable_klass(), false);
1955 return true;
1956 }
1957
1958
1959 //----------------------------inline_min_max-----------------------------------
1960 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1961 set_result(generate_min_max(id, argument(0), argument(1)));
1962 return true;
1963 }
1964
1965 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1966 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1967 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1968 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1969 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1970
1971 {
1972 PreserveJVMState pjvms(this);
1973 PreserveReexecuteState preexecs(this);
1974 jvms()->set_should_reexecute(true);
1975
1976 set_control(slow_path);
1977 set_i_o(i_o());
1978
1979 uncommon_trap(Deoptimization::Reason_intrinsic,
1980 Deoptimization::Action_none);
1981 }
1982
1983 set_control(fast_path);
1984 set_result(math);
1985 }
1986
1987 template <typename OverflowOp>
1988 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1989 typedef typename OverflowOp::MathOp MathOp;
1990
1991 MathOp* mathOp = new MathOp(arg1, arg2);
1992 Node* operation = _gvn.transform( mathOp );
1993 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1994 inline_math_mathExact(operation, ofcheck);
1995 return true;
1996 }
1997
1998 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1999 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2000 }
2001
2002 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2003 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2004 }
2005
2006 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2007 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2008 }
2009
2010 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2011 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2012 }
2013
2014 bool LibraryCallKit::inline_math_negateExactI() {
2015 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2016 }
2017
2018 bool LibraryCallKit::inline_math_negateExactL() {
2019 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2020 }
2021
2022 bool LibraryCallKit::inline_math_multiplyExactI() {
2023 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2024 }
2025
2026 bool LibraryCallKit::inline_math_multiplyExactL() {
2027 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2028 }
2029
2030 bool LibraryCallKit::inline_math_multiplyHigh() {
2031 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2032 return true;
2033 }
2034
2035 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2036 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2037 return true;
2038 }
2039
2040 Node*
2041 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
2042 Node* result_val = nullptr;
2043 switch (id) {
2044 case vmIntrinsics::_min:
2045 case vmIntrinsics::_min_strict:
2046 result_val = _gvn.transform(new MinINode(x0, y0));
2047 break;
2048 case vmIntrinsics::_max:
2049 case vmIntrinsics::_max_strict:
2050 result_val = _gvn.transform(new MaxINode(x0, y0));
2051 break;
2052 default:
2053 fatal_unexpected_iid(id);
2054 break;
2055 }
2056 return result_val;
2057 }
2058
2059 inline int
2060 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2061 const TypePtr* base_type = TypePtr::NULL_PTR;
2062 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2063 if (base_type == nullptr) {
2064 // Unknown type.
2065 return Type::AnyPtr;
2066 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2067 // Since this is a null+long form, we have to switch to a rawptr.
2068 base = _gvn.transform(new CastX2PNode(offset));
2069 offset = MakeConX(0);
2070 return Type::RawPtr;
2071 } else if (base_type->base() == Type::RawPtr) {
2072 return Type::RawPtr;
2073 } else if (base_type->isa_oopptr()) {
2074 // Base is never null => always a heap address.
2075 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2076 return Type::OopPtr;
2077 }
2078 // Offset is small => always a heap address.
2079 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2080 if (offset_type != nullptr &&
2081 base_type->offset() == 0 && // (should always be?)
2082 offset_type->_lo >= 0 &&
2083 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2084 return Type::OopPtr;
2085 } else if (type == T_OBJECT) {
2086 // off heap access to an oop doesn't make any sense. Has to be on
2087 // heap.
2088 return Type::OopPtr;
2089 }
2090 // Otherwise, it might either be oop+off or null+addr.
2091 return Type::AnyPtr;
2092 } else {
2093 // No information:
2094 return Type::AnyPtr;
2095 }
2096 }
2097
2098 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2099 Node* uncasted_base = base;
2100 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2101 if (kind == Type::RawPtr) {
2102 return basic_plus_adr(top(), uncasted_base, offset);
2103 } else if (kind == Type::AnyPtr) {
2104 assert(base == uncasted_base, "unexpected base change");
2105 if (can_cast) {
2106 if (!_gvn.type(base)->speculative_maybe_null() &&
2107 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2108 // According to profiling, this access is always on
2109 // heap. Casting the base to not null and thus avoiding membars
2110 // around the access should allow better optimizations
2111 Node* null_ctl = top();
2112 base = null_check_oop(base, &null_ctl, true, true, true);
2113 assert(null_ctl->is_top(), "no null control here");
2114 return basic_plus_adr(base, offset);
2115 } else if (_gvn.type(base)->speculative_always_null() &&
2116 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2117 // According to profiling, this access is always off
2118 // heap.
2119 base = null_assert(base);
2120 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2121 offset = MakeConX(0);
2122 return basic_plus_adr(top(), raw_base, offset);
2123 }
2124 }
2125 // We don't know if it's an on heap or off heap access. Fall back
2126 // to raw memory access.
2127 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2128 return basic_plus_adr(top(), raw, offset);
2129 } else {
2130 assert(base == uncasted_base, "unexpected base change");
2131 // We know it's an on heap access so base can't be null
2132 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2133 base = must_be_not_null(base, true);
2134 }
2135 return basic_plus_adr(base, offset);
2136 }
2137 }
2138
2139 //--------------------------inline_number_methods-----------------------------
2140 // inline int Integer.numberOfLeadingZeros(int)
2141 // inline int Long.numberOfLeadingZeros(long)
2142 //
2143 // inline int Integer.numberOfTrailingZeros(int)
2144 // inline int Long.numberOfTrailingZeros(long)
2145 //
2146 // inline int Integer.bitCount(int)
2147 // inline int Long.bitCount(long)
2148 //
2149 // inline char Character.reverseBytes(char)
2150 // inline short Short.reverseBytes(short)
2151 // inline int Integer.reverseBytes(int)
2152 // inline long Long.reverseBytes(long)
2153 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2154 Node* arg = argument(0);
2155 Node* n = nullptr;
2156 switch (id) {
2157 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2158 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2159 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2160 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2161 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2162 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2163 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(nullptr, arg); break;
2164 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( nullptr, arg); break;
2165 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( nullptr, arg); break;
2166 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( nullptr, arg); break;
2167 case vmIntrinsics::_reverse_i: n = new ReverseINode(nullptr, arg); break;
2168 case vmIntrinsics::_reverse_l: n = new ReverseLNode(nullptr, arg); break;
2169 default: fatal_unexpected_iid(id); break;
2170 }
2171 set_result(_gvn.transform(n));
2172 return true;
2173 }
2174
2175 //--------------------------inline_bitshuffle_methods-----------------------------
2176 // inline int Integer.compress(int, int)
2177 // inline int Integer.expand(int, int)
2178 // inline long Long.compress(long, long)
2179 // inline long Long.expand(long, long)
2180 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2181 Node* n = nullptr;
2182 switch (id) {
2183 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2184 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2185 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2186 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2187 default: fatal_unexpected_iid(id); break;
2188 }
2189 set_result(_gvn.transform(n));
2190 return true;
2191 }
2192
2193 //--------------------------inline_number_methods-----------------------------
2194 // inline int Integer.compareUnsigned(int, int)
2195 // inline int Long.compareUnsigned(long, long)
2196 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2197 Node* arg1 = argument(0);
2198 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2199 Node* n = nullptr;
2200 switch (id) {
2201 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2202 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2203 default: fatal_unexpected_iid(id); break;
2204 }
2205 set_result(_gvn.transform(n));
2206 return true;
2207 }
2208
2209 //--------------------------inline_unsigned_divmod_methods-----------------------------
2210 // inline int Integer.divideUnsigned(int, int)
2211 // inline int Integer.remainderUnsigned(int, int)
2212 // inline long Long.divideUnsigned(long, long)
2213 // inline long Long.remainderUnsigned(long, long)
2214 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2215 Node* n = nullptr;
2216 switch (id) {
2217 case vmIntrinsics::_divideUnsigned_i: {
2218 zero_check_int(argument(1));
2219 // Compile-time detect of null-exception
2220 if (stopped()) {
2221 return true; // keep the graph constructed so far
2222 }
2223 n = new UDivINode(control(), argument(0), argument(1));
2224 break;
2225 }
2226 case vmIntrinsics::_divideUnsigned_l: {
2227 zero_check_long(argument(2));
2228 // Compile-time detect of null-exception
2229 if (stopped()) {
2230 return true; // keep the graph constructed so far
2231 }
2232 n = new UDivLNode(control(), argument(0), argument(2));
2233 break;
2234 }
2235 case vmIntrinsics::_remainderUnsigned_i: {
2236 zero_check_int(argument(1));
2237 // Compile-time detect of null-exception
2238 if (stopped()) {
2239 return true; // keep the graph constructed so far
2240 }
2241 n = new UModINode(control(), argument(0), argument(1));
2242 break;
2243 }
2244 case vmIntrinsics::_remainderUnsigned_l: {
2245 zero_check_long(argument(2));
2246 // Compile-time detect of null-exception
2247 if (stopped()) {
2248 return true; // keep the graph constructed so far
2249 }
2250 n = new UModLNode(control(), argument(0), argument(2));
2251 break;
2252 }
2253 default: fatal_unexpected_iid(id); break;
2254 }
2255 set_result(_gvn.transform(n));
2256 return true;
2257 }
2258
2259 //----------------------------inline_unsafe_access----------------------------
2260
2261 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2262 // Attempt to infer a sharper value type from the offset and base type.
2263 ciKlass* sharpened_klass = nullptr;
2264 bool null_free = false;
2265
2266 // See if it is an instance field, with an object type.
2267 if (alias_type->field() != nullptr) {
2268 if (alias_type->field()->type()->is_klass()) {
2269 sharpened_klass = alias_type->field()->type()->as_klass();
2270 null_free = alias_type->field()->is_null_free();
2271 }
2272 }
2273
2274 const TypeOopPtr* result = nullptr;
2275 // See if it is a narrow oop array.
2276 if (adr_type->isa_aryptr()) {
2277 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2278 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2279 null_free = adr_type->is_aryptr()->is_null_free();
2280 if (elem_type != nullptr && elem_type->is_loaded()) {
2281 // Sharpen the value type.
2282 result = elem_type;
2283 }
2284 }
2285 }
2286
2287 // The sharpened class might be unloaded if there is no class loader
2288 // contraint in place.
2289 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2290 // Sharpen the value type.
2291 result = TypeOopPtr::make_from_klass(sharpened_klass);
2292 if (null_free) {
2293 result = result->join_speculative(TypePtr::NOTNULL)->is_oopptr();
2294 }
2295 }
2296 if (result != nullptr) {
2297 #ifndef PRODUCT
2298 if (C->print_intrinsics() || C->print_inlining()) {
2299 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2300 tty->print(" sharpened value: "); result->dump(); tty->cr();
2301 }
2302 #endif
2303 }
2304 return result;
2305 }
2306
2307 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2308 switch (kind) {
2309 case Relaxed:
2310 return MO_UNORDERED;
2311 case Opaque:
2312 return MO_RELAXED;
2313 case Acquire:
2314 return MO_ACQUIRE;
2315 case Release:
2316 return MO_RELEASE;
2317 case Volatile:
2318 return MO_SEQ_CST;
2319 default:
2320 ShouldNotReachHere();
2321 return 0;
2322 }
2323 }
2324
2325 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned, const bool is_flat) {
2326 if (callee()->is_static()) return false; // caller must have the capability!
2327 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2328 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2329 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2330 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2331
2332 if (is_reference_type(type)) {
2333 decorators |= ON_UNKNOWN_OOP_REF;
2334 }
2335
2336 if (unaligned) {
2337 decorators |= C2_UNALIGNED;
2338 }
2339
2340 #ifndef PRODUCT
2341 {
2342 ResourceMark rm;
2343 // Check the signatures.
2344 ciSignature* sig = callee()->signature();
2345 #ifdef ASSERT
2346 if (!is_store) {
2347 // Object getReference(Object base, int/long offset), etc.
2348 BasicType rtype = sig->return_type()->basic_type();
2349 assert(rtype == type, "getter must return the expected value");
2350 assert(sig->count() == 2 || (is_flat && sig->count() == 3), "oop getter has 2 or 3 arguments");
2351 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2352 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2353 } else {
2354 // void putReference(Object base, int/long offset, Object x), etc.
2355 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2356 assert(sig->count() == 3 || (is_flat && sig->count() == 4), "oop putter has 3 arguments");
2357 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2358 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2359 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2360 assert(vtype == type, "putter must accept the expected value");
2361 }
2362 #endif // ASSERT
2363 }
2364 #endif //PRODUCT
2365
2366 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2367
2368 Node* receiver = argument(0); // type: oop
2369
2370 // Build address expression.
2371 Node* heap_base_oop = top();
2372
2373 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2374 Node* base = argument(1); // type: oop
2375 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2376 Node* offset = argument(2); // type: long
2377 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2378 // to be plain byte offsets, which are also the same as those accepted
2379 // by oopDesc::field_addr.
2380 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2381 "fieldOffset must be byte-scaled");
2382
2383 ciInlineKlass* inline_klass = nullptr;
2384 if (is_flat) {
2385 const TypeInstPtr* cls = _gvn.type(argument(4))->isa_instptr();
2386 if (cls == nullptr || cls->const_oop() == nullptr) {
2387 return false;
2388 }
2389 ciType* mirror_type = cls->const_oop()->as_instance()->java_mirror_type();
2390 if (!mirror_type->is_inlinetype()) {
2391 return false;
2392 }
2393 inline_klass = mirror_type->as_inline_klass();
2394 }
2395
2396 if (base->is_InlineType()) {
2397 InlineTypeNode* vt = base->as_InlineType();
2398 if (is_store) {
2399 if (!vt->is_allocated(&_gvn)) {
2400 return false;
2401 }
2402 base = vt->get_oop();
2403 } else {
2404 if (offset->is_Con()) {
2405 long off = find_long_con(offset, 0);
2406 ciInlineKlass* vk = vt->type()->inline_klass();
2407 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2408 return false;
2409 }
2410
2411 ciField* field = vk->get_non_flat_field_by_offset(off);
2412 if (field != nullptr) {
2413 BasicType bt = type2field[field->type()->basic_type()];
2414 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2415 bt = T_OBJECT;
2416 }
2417 if (bt == type && (!field->is_flat() || field->type() == inline_klass)) {
2418 Node* value = vt->field_value_by_offset(off, false);
2419 if (value->is_InlineType()) {
2420 value = value->as_InlineType()->adjust_scalarization_depth(this);
2421 }
2422 set_result(value);
2423 return true;
2424 }
2425 }
2426 }
2427 {
2428 // Re-execute the unsafe access if allocation triggers deoptimization.
2429 PreserveReexecuteState preexecs(this);
2430 jvms()->set_should_reexecute(true);
2431 vt = vt->buffer(this);
2432 }
2433 base = vt->get_oop();
2434 }
2435 }
2436
2437 // 32-bit machines ignore the high half!
2438 offset = ConvL2X(offset);
2439
2440 // Save state and restore on bailout
2441 uint old_sp = sp();
2442 SafePointNode* old_map = clone_map();
2443
2444 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2445 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2446
2447 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2448 if (type != T_OBJECT && (inline_klass == nullptr || !inline_klass->has_object_fields())) {
2449 decorators |= IN_NATIVE; // off-heap primitive access
2450 } else {
2451 set_map(old_map);
2452 set_sp(old_sp);
2453 return false; // off-heap oop accesses are not supported
2454 }
2455 } else {
2456 heap_base_oop = base; // on-heap or mixed access
2457 }
2458
2459 // Can base be null? Otherwise, always on-heap access.
2460 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2461
2462 if (!can_access_non_heap) {
2463 decorators |= IN_HEAP;
2464 }
2465
2466 Node* val = is_store ? argument(4 + (is_flat ? 1 : 0)) : nullptr;
2467
2468 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2469 if (adr_type == TypePtr::NULL_PTR) {
2470 set_map(old_map);
2471 set_sp(old_sp);
2472 return false; // off-heap access with zero address
2473 }
2474
2475 // Try to categorize the address.
2476 Compile::AliasType* alias_type = C->alias_type(adr_type);
2477 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2478
2479 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2480 alias_type->adr_type() == TypeAryPtr::RANGE) {
2481 set_map(old_map);
2482 set_sp(old_sp);
2483 return false; // not supported
2484 }
2485
2486 bool mismatched = false;
2487 BasicType bt = T_ILLEGAL;
2488 ciField* field = nullptr;
2489 if (adr_type->isa_instptr()) {
2490 const TypeInstPtr* instptr = adr_type->is_instptr();
2491 ciInstanceKlass* k = instptr->instance_klass();
2492 int off = instptr->offset();
2493 if (instptr->const_oop() != nullptr &&
2494 k == ciEnv::current()->Class_klass() &&
2495 instptr->offset() >= (k->size_helper() * wordSize)) {
2496 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2497 field = k->get_field_by_offset(off, true);
2498 } else {
2499 field = k->get_non_flat_field_by_offset(off);
2500 }
2501 if (field != nullptr) {
2502 bt = type2field[field->type()->basic_type()];
2503 }
2504 assert(bt == alias_type->basic_type() || is_flat, "should match");
2505 } else {
2506 bt = alias_type->basic_type();
2507 }
2508
2509 if (bt != T_ILLEGAL) {
2510 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2511 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2512 // Alias type doesn't differentiate between byte[] and boolean[]).
2513 // Use address type to get the element type.
2514 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2515 }
2516 if (is_reference_type(bt, true)) {
2517 // accessing an array field with getReference is not a mismatch
2518 bt = T_OBJECT;
2519 }
2520 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2521 // Don't intrinsify mismatched object accesses
2522 set_map(old_map);
2523 set_sp(old_sp);
2524 return false;
2525 }
2526 mismatched = (bt != type);
2527 } else if (alias_type->adr_type()->isa_oopptr()) {
2528 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2529 }
2530
2531 if (is_flat) {
2532 if (adr_type->isa_instptr()) {
2533 if (field == nullptr || field->type() != inline_klass) {
2534 mismatched = true;
2535 }
2536 } else if (adr_type->isa_aryptr()) {
2537 const Type* elem = adr_type->is_aryptr()->elem();
2538 if (!adr_type->is_flat() || elem->inline_klass() != inline_klass) {
2539 mismatched = true;
2540 }
2541 } else {
2542 mismatched = true;
2543 }
2544 if (is_store) {
2545 const Type* val_t = _gvn.type(val);
2546 if (!val_t->is_inlinetypeptr() || val_t->inline_klass() != inline_klass) {
2547 set_map(old_map);
2548 set_sp(old_sp);
2549 return false;
2550 }
2551 }
2552 }
2553
2554 destruct_map_clone(old_map);
2555 assert(!mismatched || is_flat || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2556
2557 if (mismatched) {
2558 decorators |= C2_MISMATCHED;
2559 }
2560
2561 // First guess at the value type.
2562 const Type *value_type = Type::get_const_basic_type(type);
2563
2564 // Figure out the memory ordering.
2565 decorators |= mo_decorator_for_access_kind(kind);
2566
2567 if (!is_store) {
2568 if (type == T_OBJECT && !is_flat) {
2569 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2570 if (tjp != nullptr) {
2571 value_type = tjp;
2572 }
2573 }
2574 }
2575
2576 receiver = null_check(receiver);
2577 if (stopped()) {
2578 return true;
2579 }
2580 // Heap pointers get a null-check from the interpreter,
2581 // as a courtesy. However, this is not guaranteed by Unsafe,
2582 // and it is not possible to fully distinguish unintended nulls
2583 // from intended ones in this API.
2584
2585 if (!is_store) {
2586 Node* p = nullptr;
2587 // Try to constant fold a load from a constant field
2588
2589 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !field->is_flat() && !mismatched) {
2590 // final or stable field
2591 p = make_constant_from_field(field, heap_base_oop);
2592 }
2593
2594 if (p == nullptr) { // Could not constant fold the load
2595 if (is_flat) {
2596 if (adr_type->isa_instptr() && !mismatched) {
2597 ciInstanceKlass* holder = adr_type->is_instptr()->instance_klass();
2598 int offset = adr_type->is_instptr()->offset();
2599 p = InlineTypeNode::make_from_flat(this, inline_klass, base, base, holder, offset, decorators);
2600 } else {
2601 p = InlineTypeNode::make_from_flat(this, inline_klass, base, adr, nullptr, 0, decorators);
2602 }
2603 } else {
2604 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2605 const TypeOopPtr* ptr = value_type->make_oopptr();
2606 if (ptr != nullptr && ptr->is_inlinetypeptr()) {
2607 // Load a non-flattened inline type from memory
2608 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass(), !ptr->maybe_null());
2609 }
2610 }
2611 // Normalize the value returned by getBoolean in the following cases
2612 if (type == T_BOOLEAN &&
2613 (mismatched ||
2614 heap_base_oop == top() || // - heap_base_oop is null or
2615 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2616 // and the unsafe access is made to large offset
2617 // (i.e., larger than the maximum offset necessary for any
2618 // field access)
2619 ) {
2620 IdealKit ideal = IdealKit(this);
2621 #define __ ideal.
2622 IdealVariable normalized_result(ideal);
2623 __ declarations_done();
2624 __ set(normalized_result, p);
2625 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2626 __ set(normalized_result, ideal.ConI(1));
2627 ideal.end_if();
2628 final_sync(ideal);
2629 p = __ value(normalized_result);
2630 #undef __
2631 }
2632 }
2633 if (type == T_ADDRESS) {
2634 p = gvn().transform(new CastP2XNode(nullptr, p));
2635 p = ConvX2UL(p);
2636 }
2637 // The load node has the control of the preceding MemBarCPUOrder. All
2638 // following nodes will have the control of the MemBarCPUOrder inserted at
2639 // the end of this method. So, pushing the load onto the stack at a later
2640 // point is fine.
2641 set_result(p);
2642 } else {
2643 if (bt == T_ADDRESS) {
2644 // Repackage the long as a pointer.
2645 val = ConvL2X(val);
2646 val = gvn().transform(new CastX2PNode(val));
2647 }
2648 if (is_flat) {
2649 if (adr_type->isa_instptr() && !mismatched) {
2650 ciInstanceKlass* holder = adr_type->is_instptr()->instance_klass();
2651 int offset = adr_type->is_instptr()->offset();
2652 val->as_InlineType()->store_flat(this, base, base, holder, offset, decorators);
2653 } else {
2654 val->as_InlineType()->store_flat(this, base, adr, nullptr, 0, decorators);
2655 }
2656 } else {
2657 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2658 }
2659 }
2660
2661 if (argument(1)->is_InlineType() && is_store) {
2662 InlineTypeNode* value = InlineTypeNode::make_from_oop(this, base, _gvn.type(argument(1))->inline_klass());
2663 value = value->make_larval(this, false);
2664 replace_in_map(argument(1), value);
2665 }
2666
2667 return true;
2668 }
2669
2670 bool LibraryCallKit::inline_unsafe_make_private_buffer() {
2671 Node* receiver = argument(0);
2672 Node* value = argument(1);
2673 if (!value->is_InlineType()) {
2674 return false;
2675 }
2676
2677 receiver = null_check(receiver);
2678 if (stopped()) {
2679 return true;
2680 }
2681
2682 set_result(value->as_InlineType()->make_larval(this, true));
2683 return true;
2684 }
2685
2686 bool LibraryCallKit::inline_unsafe_finish_private_buffer() {
2687 Node* receiver = argument(0);
2688 Node* buffer = argument(1);
2689 if (!buffer->is_InlineType()) {
2690 return false;
2691 }
2692 InlineTypeNode* vt = buffer->as_InlineType();
2693 if (!vt->is_allocated(&_gvn)) {
2694 return false;
2695 }
2696 // TODO 8239003 Why is this needed?
2697 if (AllocateNode::Ideal_allocation(vt->get_oop()) == nullptr) {
2698 return false;
2699 }
2700
2701 receiver = null_check(receiver);
2702 if (stopped()) {
2703 return true;
2704 }
2705
2706 set_result(vt->finish_larval(this));
2707 return true;
2708 }
2709
2710 //----------------------------inline_unsafe_load_store----------------------------
2711 // This method serves a couple of different customers (depending on LoadStoreKind):
2712 //
2713 // LS_cmp_swap:
2714 //
2715 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2716 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2717 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2718 //
2719 // LS_cmp_swap_weak:
2720 //
2721 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2722 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2723 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2724 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2725 //
2726 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2727 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2728 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2729 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2730 //
2731 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2732 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2733 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2734 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2735 //
2736 // LS_cmp_exchange:
2737 //
2738 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2739 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2740 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2741 //
2742 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2743 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2744 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2745 //
2746 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2747 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2748 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2749 //
2750 // LS_get_add:
2751 //
2752 // int getAndAddInt( Object o, long offset, int delta)
2753 // long getAndAddLong(Object o, long offset, long delta)
2754 //
2755 // LS_get_set:
2756 //
2757 // int getAndSet(Object o, long offset, int newValue)
2758 // long getAndSet(Object o, long offset, long newValue)
2759 // Object getAndSet(Object o, long offset, Object newValue)
2760 //
2761 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2762 // This basic scheme here is the same as inline_unsafe_access, but
2763 // differs in enough details that combining them would make the code
2764 // overly confusing. (This is a true fact! I originally combined
2765 // them, but even I was confused by it!) As much code/comments as
2766 // possible are retained from inline_unsafe_access though to make
2767 // the correspondences clearer. - dl
2768
2769 if (callee()->is_static()) return false; // caller must have the capability!
2770
2771 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2772 decorators |= mo_decorator_for_access_kind(access_kind);
2773
2774 #ifndef PRODUCT
2775 BasicType rtype;
2776 {
2777 ResourceMark rm;
2778 // Check the signatures.
2779 ciSignature* sig = callee()->signature();
2780 rtype = sig->return_type()->basic_type();
2781 switch(kind) {
2782 case LS_get_add:
2783 case LS_get_set: {
2784 // Check the signatures.
2785 #ifdef ASSERT
2786 assert(rtype == type, "get and set must return the expected type");
2787 assert(sig->count() == 3, "get and set has 3 arguments");
2788 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2789 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2790 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2791 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2792 #endif // ASSERT
2793 break;
2794 }
2795 case LS_cmp_swap:
2796 case LS_cmp_swap_weak: {
2797 // Check the signatures.
2798 #ifdef ASSERT
2799 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2800 assert(sig->count() == 4, "CAS has 4 arguments");
2801 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2802 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2803 #endif // ASSERT
2804 break;
2805 }
2806 case LS_cmp_exchange: {
2807 // Check the signatures.
2808 #ifdef ASSERT
2809 assert(rtype == type, "CAS must return the expected type");
2810 assert(sig->count() == 4, "CAS has 4 arguments");
2811 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2812 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2813 #endif // ASSERT
2814 break;
2815 }
2816 default:
2817 ShouldNotReachHere();
2818 }
2819 }
2820 #endif //PRODUCT
2821
2822 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2823
2824 // Get arguments:
2825 Node* receiver = nullptr;
2826 Node* base = nullptr;
2827 Node* offset = nullptr;
2828 Node* oldval = nullptr;
2829 Node* newval = nullptr;
2830 switch(kind) {
2831 case LS_cmp_swap:
2832 case LS_cmp_swap_weak:
2833 case LS_cmp_exchange: {
2834 const bool two_slot_type = type2size[type] == 2;
2835 receiver = argument(0); // type: oop
2836 base = argument(1); // type: oop
2837 offset = argument(2); // type: long
2838 oldval = argument(4); // type: oop, int, or long
2839 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2840 break;
2841 }
2842 case LS_get_add:
2843 case LS_get_set: {
2844 receiver = argument(0); // type: oop
2845 base = argument(1); // type: oop
2846 offset = argument(2); // type: long
2847 oldval = nullptr;
2848 newval = argument(4); // type: oop, int, or long
2849 break;
2850 }
2851 default:
2852 ShouldNotReachHere();
2853 }
2854
2855 // Build field offset expression.
2856 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2857 // to be plain byte offsets, which are also the same as those accepted
2858 // by oopDesc::field_addr.
2859 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2860 // 32-bit machines ignore the high half of long offsets
2861 offset = ConvL2X(offset);
2862 // Save state and restore on bailout
2863 uint old_sp = sp();
2864 SafePointNode* old_map = clone_map();
2865 Node* adr = make_unsafe_address(base, offset,type, false);
2866 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2867
2868 Compile::AliasType* alias_type = C->alias_type(adr_type);
2869 BasicType bt = alias_type->basic_type();
2870 if (bt != T_ILLEGAL &&
2871 (is_reference_type(bt) != (type == T_OBJECT))) {
2872 // Don't intrinsify mismatched object accesses.
2873 set_map(old_map);
2874 set_sp(old_sp);
2875 return false;
2876 }
2877
2878 destruct_map_clone(old_map);
2879
2880 // For CAS, unlike inline_unsafe_access, there seems no point in
2881 // trying to refine types. Just use the coarse types here.
2882 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2883 const Type *value_type = Type::get_const_basic_type(type);
2884
2885 switch (kind) {
2886 case LS_get_set:
2887 case LS_cmp_exchange: {
2888 if (type == T_OBJECT) {
2889 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2890 if (tjp != nullptr) {
2891 value_type = tjp;
2892 }
2893 }
2894 break;
2895 }
2896 case LS_cmp_swap:
2897 case LS_cmp_swap_weak:
2898 case LS_get_add:
2899 break;
2900 default:
2901 ShouldNotReachHere();
2902 }
2903
2904 // Null check receiver.
2905 receiver = null_check(receiver);
2906 if (stopped()) {
2907 return true;
2908 }
2909
2910 int alias_idx = C->get_alias_index(adr_type);
2911
2912 if (is_reference_type(type)) {
2913 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2914
2915 if (oldval != nullptr && oldval->is_InlineType()) {
2916 // Re-execute the unsafe access if allocation triggers deoptimization.
2917 PreserveReexecuteState preexecs(this);
2918 jvms()->set_should_reexecute(true);
2919 oldval = oldval->as_InlineType()->buffer(this)->get_oop();
2920 }
2921 if (newval != nullptr && newval->is_InlineType()) {
2922 // Re-execute the unsafe access if allocation triggers deoptimization.
2923 PreserveReexecuteState preexecs(this);
2924 jvms()->set_should_reexecute(true);
2925 newval = newval->as_InlineType()->buffer(this)->get_oop();
2926 }
2927
2928 // Transformation of a value which could be null pointer (CastPP #null)
2929 // could be delayed during Parse (for example, in adjust_map_after_if()).
2930 // Execute transformation here to avoid barrier generation in such case.
2931 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2932 newval = _gvn.makecon(TypePtr::NULL_PTR);
2933
2934 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2935 // Refine the value to a null constant, when it is known to be null
2936 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2937 }
2938 }
2939
2940 Node* result = nullptr;
2941 switch (kind) {
2942 case LS_cmp_exchange: {
2943 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2944 oldval, newval, value_type, type, decorators);
2945 break;
2946 }
2947 case LS_cmp_swap_weak:
2948 decorators |= C2_WEAK_CMPXCHG;
2949 case LS_cmp_swap: {
2950 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2951 oldval, newval, value_type, type, decorators);
2952 break;
2953 }
2954 case LS_get_set: {
2955 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2956 newval, value_type, type, decorators);
2957 break;
2958 }
2959 case LS_get_add: {
2960 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2961 newval, value_type, type, decorators);
2962 break;
2963 }
2964 default:
2965 ShouldNotReachHere();
2966 }
2967
2968 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2969 set_result(result);
2970 return true;
2971 }
2972
2973 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2974 // Regardless of form, don't allow previous ld/st to move down,
2975 // then issue acquire, release, or volatile mem_bar.
2976 insert_mem_bar(Op_MemBarCPUOrder);
2977 switch(id) {
2978 case vmIntrinsics::_loadFence:
2979 insert_mem_bar(Op_LoadFence);
2980 return true;
2981 case vmIntrinsics::_storeFence:
2982 insert_mem_bar(Op_StoreFence);
2983 return true;
2984 case vmIntrinsics::_storeStoreFence:
2985 insert_mem_bar(Op_StoreStoreFence);
2986 return true;
2987 case vmIntrinsics::_fullFence:
2988 insert_mem_bar(Op_MemBarVolatile);
2989 return true;
2990 default:
2991 fatal_unexpected_iid(id);
2992 return false;
2993 }
2994 }
2995
2996 bool LibraryCallKit::inline_onspinwait() {
2997 insert_mem_bar(Op_OnSpinWait);
2998 return true;
2999 }
3000
3001 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3002 if (!kls->is_Con()) {
3003 return true;
3004 }
3005 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
3006 if (klsptr == nullptr) {
3007 return true;
3008 }
3009 ciInstanceKlass* ik = klsptr->instance_klass();
3010 // don't need a guard for a klass that is already initialized
3011 return !ik->is_initialized();
3012 }
3013
3014 //----------------------------inline_unsafe_writeback0-------------------------
3015 // public native void Unsafe.writeback0(long address)
3016 bool LibraryCallKit::inline_unsafe_writeback0() {
3017 if (!Matcher::has_match_rule(Op_CacheWB)) {
3018 return false;
3019 }
3020 #ifndef PRODUCT
3021 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
3022 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
3023 ciSignature* sig = callee()->signature();
3024 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
3025 #endif
3026 null_check_receiver(); // null-check, then ignore
3027 Node *addr = argument(1);
3028 addr = new CastX2PNode(addr);
3029 addr = _gvn.transform(addr);
3030 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
3031 flush = _gvn.transform(flush);
3032 set_memory(flush, TypeRawPtr::BOTTOM);
3033 return true;
3034 }
3035
3036 //----------------------------inline_unsafe_writeback0-------------------------
3037 // public native void Unsafe.writeback0(long address)
3038 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
3039 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
3040 return false;
3041 }
3042 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
3043 return false;
3044 }
3045 #ifndef PRODUCT
3046 assert(Matcher::has_match_rule(Op_CacheWB),
3047 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
3048 : "found match rule for CacheWBPostSync but not CacheWB"));
3049
3050 #endif
3051 null_check_receiver(); // null-check, then ignore
3052 Node *sync;
3053 if (is_pre) {
3054 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3055 } else {
3056 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3057 }
3058 sync = _gvn.transform(sync);
3059 set_memory(sync, TypeRawPtr::BOTTOM);
3060 return true;
3061 }
3062
3063 //----------------------------inline_unsafe_allocate---------------------------
3064 // public native Object Unsafe.allocateInstance(Class<?> cls);
3065 bool LibraryCallKit::inline_unsafe_allocate() {
3066
3067 #if INCLUDE_JVMTI
3068 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3069 return false;
3070 }
3071 #endif //INCLUDE_JVMTI
3072
3073 if (callee()->is_static()) return false; // caller must have the capability!
3074
3075 null_check_receiver(); // null-check, then ignore
3076 Node* cls = null_check(argument(1));
3077 if (stopped()) return true;
3078
3079 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
3080 kls = null_check(kls);
3081 if (stopped()) return true; // argument was like int.class
3082
3083 #if INCLUDE_JVMTI
3084 // Don't try to access new allocated obj in the intrinsic.
3085 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
3086 // Deoptimize and allocate in interpreter instead.
3087 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
3088 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
3089 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
3090 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
3091 {
3092 BuildCutout unless(this, tst, PROB_MAX);
3093 uncommon_trap(Deoptimization::Reason_intrinsic,
3094 Deoptimization::Action_make_not_entrant);
3095 }
3096 if (stopped()) {
3097 return true;
3098 }
3099 #endif //INCLUDE_JVMTI
3100
3101 Node* test = nullptr;
3102 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3103 // Note: The argument might still be an illegal value like
3104 // Serializable.class or Object[].class. The runtime will handle it.
3105 // But we must make an explicit check for initialization.
3106 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3107 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3108 // can generate code to load it as unsigned byte.
3109 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
3110 Node* bits = intcon(InstanceKlass::fully_initialized);
3111 test = _gvn.transform(new SubINode(inst, bits));
3112 // The 'test' is non-zero if we need to take a slow path.
3113 }
3114 Node* obj = nullptr;
3115 const TypeInstKlassPtr* tkls = _gvn.type(kls)->isa_instklassptr();
3116 if (tkls != nullptr && tkls->instance_klass()->is_inlinetype()) {
3117 obj = InlineTypeNode::make_default(_gvn, tkls->instance_klass()->as_inline_klass())->buffer(this);
3118 } else {
3119 obj = new_instance(kls, test);
3120 }
3121 set_result(obj);
3122 return true;
3123 }
3124
3125 //------------------------inline_native_time_funcs--------------
3126 // inline code for System.currentTimeMillis() and System.nanoTime()
3127 // these have the same type and signature
3128 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3129 const TypeFunc* tf = OptoRuntime::void_long_Type();
3130 const TypePtr* no_memory_effects = nullptr;
3131 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3132 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3133 #ifdef ASSERT
3134 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3135 assert(value_top == top(), "second value must be top");
3136 #endif
3137 set_result(value);
3138 return true;
3139 }
3140
3141
3142 #if INCLUDE_JVMTI
3143
3144 // When notifications are disabled then just update the VTMS transition bit and return.
3145 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
3146 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
3147 if (!DoJVMTIVirtualThreadTransitions) {
3148 return true;
3149 }
3150 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3151 IdealKit ideal(this);
3152
3153 Node* ONE = ideal.ConI(1);
3154 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
3155 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
3156 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3157
3158 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
3159 sync_kit(ideal);
3160 // if notifyJvmti enabled then make a call to the given SharedRuntime function
3161 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
3162 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
3163 ideal.sync_kit(this);
3164 } ideal.else_(); {
3165 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
3166 Node* thread = ideal.thread();
3167 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
3168 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
3169
3170 sync_kit(ideal);
3171 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3172 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3173
3174 ideal.sync_kit(this);
3175 } ideal.end_if();
3176 final_sync(ideal);
3177
3178 return true;
3179 }
3180
3181 // Always update the temporary VTMS transition bit.
3182 bool LibraryCallKit::inline_native_notify_jvmti_hide() {
3183 if (!DoJVMTIVirtualThreadTransitions) {
3184 return true;
3185 }
3186 IdealKit ideal(this);
3187
3188 {
3189 // unconditionally update the temporary VTMS transition bit in current JavaThread
3190 Node* thread = ideal.thread();
3191 Node* hide = _gvn.transform(argument(0)); // hide argument for temporary VTMS transition notification
3192 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_tmp_VTMS_transition_offset()));
3193 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3194
3195 sync_kit(ideal);
3196 access_store_at(nullptr, addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3197 ideal.sync_kit(this);
3198 }
3199 final_sync(ideal);
3200
3201 return true;
3202 }
3203
3204 // Always update the is_disable_suspend bit.
3205 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3206 if (!DoJVMTIVirtualThreadTransitions) {
3207 return true;
3208 }
3209 IdealKit ideal(this);
3210
3211 {
3212 // unconditionally update the is_disable_suspend bit in current JavaThread
3213 Node* thread = ideal.thread();
3214 Node* arg = _gvn.transform(argument(0)); // argument for notification
3215 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3216 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3217
3218 sync_kit(ideal);
3219 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3220 ideal.sync_kit(this);
3221 }
3222 final_sync(ideal);
3223
3224 return true;
3225 }
3226
3227 #endif // INCLUDE_JVMTI
3228
3229 #ifdef JFR_HAVE_INTRINSICS
3230
3231 /**
3232 * if oop->klass != null
3233 * // normal class
3234 * epoch = _epoch_state ? 2 : 1
3235 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3236 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3237 * }
3238 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3239 * else
3240 * // primitive class
3241 * if oop->array_klass != null
3242 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3243 * else
3244 * id = LAST_TYPE_ID + 1 // void class path
3245 * if (!signaled)
3246 * signaled = true
3247 */
3248 bool LibraryCallKit::inline_native_classID() {
3249 Node* cls = argument(0);
3250
3251 IdealKit ideal(this);
3252 #define __ ideal.
3253 IdealVariable result(ideal); __ declarations_done();
3254 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3255 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3256 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3257
3258
3259 __ if_then(kls, BoolTest::ne, null()); {
3260 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3261 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3262
3263 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3264 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3265 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3266 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3267 mask = _gvn.transform(new OrLNode(mask, epoch));
3268 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3269
3270 float unlikely = PROB_UNLIKELY(0.999);
3271 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3272 sync_kit(ideal);
3273 make_runtime_call(RC_LEAF,
3274 OptoRuntime::class_id_load_barrier_Type(),
3275 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3276 "class id load barrier",
3277 TypePtr::BOTTOM,
3278 kls);
3279 ideal.sync_kit(this);
3280 } __ end_if();
3281
3282 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3283 } __ else_(); {
3284 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3285 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3286 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3287 __ if_then(array_kls, BoolTest::ne, null()); {
3288 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3289 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3290 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3291 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3292 } __ else_(); {
3293 // void class case
3294 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3295 } __ end_if();
3296
3297 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3298 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3299 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3300 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3301 } __ end_if();
3302 } __ end_if();
3303
3304 final_sync(ideal);
3305 set_result(ideal.value(result));
3306 #undef __
3307 return true;
3308 }
3309
3310 //------------------------inline_native_jvm_commit------------------
3311 bool LibraryCallKit::inline_native_jvm_commit() {
3312 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3313
3314 // Save input memory and i_o state.
3315 Node* input_memory_state = reset_memory();
3316 set_all_memory(input_memory_state);
3317 Node* input_io_state = i_o();
3318
3319 // TLS.
3320 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3321 // Jfr java buffer.
3322 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3323 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3324 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3325
3326 // Load the current value of the notified field in the JfrThreadLocal.
3327 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3328 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3329
3330 // Test for notification.
3331 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3332 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3333 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3334
3335 // True branch, is notified.
3336 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3337 set_control(is_notified);
3338
3339 // Reset notified state.
3340 Node* notified_reset_memory = store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered);
3341
3342 // 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.
3343 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3344 // Convert the machine-word to a long.
3345 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3346
3347 // False branch, not notified.
3348 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3349 set_control(not_notified);
3350 set_all_memory(input_memory_state);
3351
3352 // Arg is the next position as a long.
3353 Node* arg = argument(0);
3354 // Convert long to machine-word.
3355 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3356
3357 // Store the next_position to the underlying jfr java buffer.
3358 Node* commit_memory;
3359 #ifdef _LP64
3360 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_LONG, Compile::AliasIdxRaw, MemNode::release);
3361 #else
3362 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_INT, Compile::AliasIdxRaw, MemNode::release);
3363 #endif
3364
3365 // 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.
3366 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3367 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3368 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3369
3370 // And flags with lease constant.
3371 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3372
3373 // Branch on lease to conditionalize returning the leased java buffer.
3374 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3375 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3376 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3377
3378 // False branch, not a lease.
3379 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3380
3381 // True branch, is lease.
3382 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3383 set_control(is_lease);
3384
3385 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3386 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3387 OptoRuntime::void_void_Type(),
3388 SharedRuntime::jfr_return_lease(),
3389 "return_lease", TypePtr::BOTTOM);
3390 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3391
3392 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3393 record_for_igvn(lease_compare_rgn);
3394 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3395 record_for_igvn(lease_compare_mem);
3396 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3397 record_for_igvn(lease_compare_io);
3398 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3399 record_for_igvn(lease_result_value);
3400
3401 // Update control and phi nodes.
3402 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3403 lease_compare_rgn->init_req(_false_path, not_lease);
3404
3405 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3406 lease_compare_mem->init_req(_false_path, commit_memory);
3407
3408 lease_compare_io->init_req(_true_path, i_o());
3409 lease_compare_io->init_req(_false_path, input_io_state);
3410
3411 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3412 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3413
3414 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3415 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3416 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3417 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3418
3419 // Update control and phi nodes.
3420 result_rgn->init_req(_true_path, is_notified);
3421 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3422
3423 result_mem->init_req(_true_path, notified_reset_memory);
3424 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3425
3426 result_io->init_req(_true_path, input_io_state);
3427 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3428
3429 result_value->init_req(_true_path, current_pos);
3430 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3431
3432 // Set output state.
3433 set_control(_gvn.transform(result_rgn));
3434 set_all_memory(_gvn.transform(result_mem));
3435 set_i_o(_gvn.transform(result_io));
3436 set_result(result_rgn, result_value);
3437 return true;
3438 }
3439
3440 /*
3441 * The intrinsic is a model of this pseudo-code:
3442 *
3443 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3444 * jobject h_event_writer = tl->java_event_writer();
3445 * if (h_event_writer == nullptr) {
3446 * return nullptr;
3447 * }
3448 * oop threadObj = Thread::threadObj();
3449 * oop vthread = java_lang_Thread::vthread(threadObj);
3450 * traceid tid;
3451 * bool pinVirtualThread;
3452 * bool excluded;
3453 * if (vthread != threadObj) { // i.e. current thread is virtual
3454 * tid = java_lang_Thread::tid(vthread);
3455 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3456 * pinVirtualThread = VMContinuations;
3457 * excluded = vthread_epoch_raw & excluded_mask;
3458 * if (!excluded) {
3459 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3460 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3461 * if (vthread_epoch != current_epoch) {
3462 * write_checkpoint();
3463 * }
3464 * }
3465 * } else {
3466 * tid = java_lang_Thread::tid(threadObj);
3467 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3468 * pinVirtualThread = false;
3469 * excluded = thread_epoch_raw & excluded_mask;
3470 * }
3471 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3472 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3473 * if (tid_in_event_writer != tid) {
3474 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3475 * setField(event_writer, "excluded", excluded);
3476 * setField(event_writer, "threadID", tid);
3477 * }
3478 * return event_writer
3479 */
3480 bool LibraryCallKit::inline_native_getEventWriter() {
3481 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3482
3483 // Save input memory and i_o state.
3484 Node* input_memory_state = reset_memory();
3485 set_all_memory(input_memory_state);
3486 Node* input_io_state = i_o();
3487
3488 Node* excluded_mask = _gvn.intcon(32768);
3489 Node* epoch_mask = _gvn.intcon(32767);
3490
3491 // TLS
3492 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3493
3494 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3495 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3496
3497 // Load the eventwriter jobject handle.
3498 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3499
3500 // Null check the jobject handle.
3501 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3502 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3503 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3504
3505 // False path, jobj is null.
3506 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3507
3508 // True path, jobj is not null.
3509 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3510
3511 set_control(jobj_is_not_null);
3512
3513 // Load the threadObj for the CarrierThread.
3514 Node* threadObj = generate_current_thread(tls_ptr);
3515
3516 // Load the vthread.
3517 Node* vthread = generate_virtual_thread(tls_ptr);
3518
3519 // If vthread != threadObj, this is a virtual thread.
3520 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3521 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3522 IfNode* iff_vthread_not_equal_threadObj =
3523 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3524
3525 // False branch, fallback to threadObj.
3526 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3527 set_control(vthread_equal_threadObj);
3528
3529 // Load the tid field from the vthread object.
3530 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3531
3532 // Load the raw epoch value from the threadObj.
3533 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3534 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3535 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3536 TypeInt::CHAR, T_CHAR,
3537 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3538
3539 // Mask off the excluded information from the epoch.
3540 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3541
3542 // True branch, this is a virtual thread.
3543 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3544 set_control(vthread_not_equal_threadObj);
3545
3546 // Load the tid field from the vthread object.
3547 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3548
3549 // Continuation support determines if a virtual thread should be pinned.
3550 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3551 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3552
3553 // Load the raw epoch value from the vthread.
3554 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3555 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3556 TypeInt::CHAR, T_CHAR,
3557 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3558
3559 // Mask off the excluded information from the epoch.
3560 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3561
3562 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3563 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3564 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3565 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3566
3567 // False branch, vthread is excluded, no need to write epoch info.
3568 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3569
3570 // True branch, vthread is included, update epoch info.
3571 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3572 set_control(included);
3573
3574 // Get epoch value.
3575 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3576
3577 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3578 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3579 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3580
3581 // Compare the epoch in the vthread to the current epoch generation.
3582 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3583 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3584 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3585
3586 // False path, epoch is equal, checkpoint information is valid.
3587 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3588
3589 // True path, epoch is not equal, write a checkpoint for the vthread.
3590 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3591
3592 set_control(epoch_is_not_equal);
3593
3594 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3595 // The call also updates the native thread local thread id and the vthread with the current epoch.
3596 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3597 OptoRuntime::jfr_write_checkpoint_Type(),
3598 SharedRuntime::jfr_write_checkpoint(),
3599 "write_checkpoint", TypePtr::BOTTOM);
3600 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3601
3602 // vthread epoch != current epoch
3603 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3604 record_for_igvn(epoch_compare_rgn);
3605 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3606 record_for_igvn(epoch_compare_mem);
3607 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3608 record_for_igvn(epoch_compare_io);
3609
3610 // Update control and phi nodes.
3611 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3612 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3613 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3614 epoch_compare_mem->init_req(_false_path, input_memory_state);
3615 epoch_compare_io->init_req(_true_path, i_o());
3616 epoch_compare_io->init_req(_false_path, input_io_state);
3617
3618 // excluded != true
3619 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3620 record_for_igvn(exclude_compare_rgn);
3621 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3622 record_for_igvn(exclude_compare_mem);
3623 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3624 record_for_igvn(exclude_compare_io);
3625
3626 // Update control and phi nodes.
3627 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3628 exclude_compare_rgn->init_req(_false_path, excluded);
3629 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3630 exclude_compare_mem->init_req(_false_path, input_memory_state);
3631 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3632 exclude_compare_io->init_req(_false_path, input_io_state);
3633
3634 // vthread != threadObj
3635 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3636 record_for_igvn(vthread_compare_rgn);
3637 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3638 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3639 record_for_igvn(vthread_compare_io);
3640 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3641 record_for_igvn(tid);
3642 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3643 record_for_igvn(exclusion);
3644 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3645 record_for_igvn(pinVirtualThread);
3646
3647 // Update control and phi nodes.
3648 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3649 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3650 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3651 vthread_compare_mem->init_req(_false_path, input_memory_state);
3652 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3653 vthread_compare_io->init_req(_false_path, input_io_state);
3654 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3655 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3656 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3657 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3658 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3659 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3660
3661 // Update branch state.
3662 set_control(_gvn.transform(vthread_compare_rgn));
3663 set_all_memory(_gvn.transform(vthread_compare_mem));
3664 set_i_o(_gvn.transform(vthread_compare_io));
3665
3666 // Load the event writer oop by dereferencing the jobject handle.
3667 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3668 assert(klass_EventWriter->is_loaded(), "invariant");
3669 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3670 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3671 const TypeOopPtr* const xtype = aklass->as_instance_type();
3672 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3673 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3674
3675 // Load the current thread id from the event writer object.
3676 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3677 // Get the field offset to, conditionally, store an updated tid value later.
3678 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3679 const TypePtr* event_writer_tid_field_type = _gvn.type(event_writer_tid_field)->isa_ptr();
3680 // Get the field offset to, conditionally, store an updated exclusion value later.
3681 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3682 const TypePtr* event_writer_excluded_field_type = _gvn.type(event_writer_excluded_field)->isa_ptr();
3683 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3684 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3685 const TypePtr* event_writer_pin_field_type = _gvn.type(event_writer_pin_field)->isa_ptr();
3686
3687 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3688 record_for_igvn(event_writer_tid_compare_rgn);
3689 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3690 record_for_igvn(event_writer_tid_compare_mem);
3691 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3692 record_for_igvn(event_writer_tid_compare_io);
3693
3694 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3695 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3696 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3697 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3698
3699 // False path, tids are the same.
3700 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3701
3702 // True path, tid is not equal, need to update the tid in the event writer.
3703 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3704 record_for_igvn(tid_is_not_equal);
3705
3706 // Store the pin state to the event writer.
3707 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, event_writer_pin_field_type, MemNode::unordered);
3708
3709 // Store the exclusion state to the event writer.
3710 store_to_memory(tid_is_not_equal, event_writer_excluded_field, _gvn.transform(exclusion), T_BOOLEAN, event_writer_excluded_field_type, MemNode::unordered);
3711
3712 // Store the tid to the event writer.
3713 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, event_writer_tid_field_type, MemNode::unordered);
3714
3715 // Update control and phi nodes.
3716 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3717 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3718 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3719 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3720 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3721 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3722
3723 // Result of top level CFG, Memory, IO and Value.
3724 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3725 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3726 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3727 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3728
3729 // Result control.
3730 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3731 result_rgn->init_req(_false_path, jobj_is_null);
3732
3733 // Result memory.
3734 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3735 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3736
3737 // Result IO.
3738 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3739 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3740
3741 // Result value.
3742 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3743 result_value->init_req(_false_path, null()); // return null
3744
3745 // Set output state.
3746 set_control(_gvn.transform(result_rgn));
3747 set_all_memory(_gvn.transform(result_mem));
3748 set_i_o(_gvn.transform(result_io));
3749 set_result(result_rgn, result_value);
3750 return true;
3751 }
3752
3753 /*
3754 * The intrinsic is a model of this pseudo-code:
3755 *
3756 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3757 * if (carrierThread != thread) { // is virtual thread
3758 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3759 * bool excluded = vthread_epoch_raw & excluded_mask;
3760 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3761 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3762 * if (!excluded) {
3763 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3764 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3765 * }
3766 * Atomic::release_store(&tl->_vthread, true);
3767 * return;
3768 * }
3769 * Atomic::release_store(&tl->_vthread, false);
3770 */
3771 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3772 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3773
3774 Node* input_memory_state = reset_memory();
3775 set_all_memory(input_memory_state);
3776
3777 Node* excluded_mask = _gvn.intcon(32768);
3778 Node* epoch_mask = _gvn.intcon(32767);
3779
3780 Node* const carrierThread = generate_current_thread(jt);
3781 // If thread != carrierThread, this is a virtual thread.
3782 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3783 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3784 IfNode* iff_thread_not_equal_carrierThread =
3785 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3786
3787 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3788
3789 // False branch, is carrierThread.
3790 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3791 // Store release
3792 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3793
3794 set_all_memory(input_memory_state);
3795
3796 // True branch, is virtual thread.
3797 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3798 set_control(thread_not_equal_carrierThread);
3799
3800 // Load the raw epoch value from the vthread.
3801 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3802 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
3803 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3804
3805 // Mask off the excluded information from the epoch.
3806 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3807
3808 // Load the tid field from the thread.
3809 Node* tid = load_field_from_object(thread, "tid", "J");
3810
3811 // Store the vthread tid to the jfr thread local.
3812 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3813 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, Compile::AliasIdxRaw, MemNode::unordered, true);
3814
3815 // Branch is_excluded to conditionalize updating the epoch .
3816 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3817 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3818 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3819
3820 // True branch, vthread is excluded, no need to write epoch info.
3821 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3822 set_control(excluded);
3823 Node* vthread_is_excluded = _gvn.intcon(1);
3824
3825 // False branch, vthread is included, update epoch info.
3826 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3827 set_control(included);
3828 Node* vthread_is_included = _gvn.intcon(0);
3829
3830 // Get epoch value.
3831 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3832
3833 // Store the vthread epoch to the jfr thread local.
3834 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3835 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, Compile::AliasIdxRaw, MemNode::unordered, true);
3836
3837 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3838 record_for_igvn(excluded_rgn);
3839 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3840 record_for_igvn(excluded_mem);
3841 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3842 record_for_igvn(exclusion);
3843
3844 // Merge the excluded control and memory.
3845 excluded_rgn->init_req(_true_path, excluded);
3846 excluded_rgn->init_req(_false_path, included);
3847 excluded_mem->init_req(_true_path, tid_memory);
3848 excluded_mem->init_req(_false_path, included_memory);
3849 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3850 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3851
3852 // Set intermediate state.
3853 set_control(_gvn.transform(excluded_rgn));
3854 set_all_memory(excluded_mem);
3855
3856 // Store the vthread exclusion state to the jfr thread local.
3857 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3858 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered, true);
3859
3860 // Store release
3861 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3862
3863 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3864 record_for_igvn(thread_compare_rgn);
3865 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3866 record_for_igvn(thread_compare_mem);
3867 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3868 record_for_igvn(vthread);
3869
3870 // Merge the thread_compare control and memory.
3871 thread_compare_rgn->init_req(_true_path, control());
3872 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3873 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3874 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3875
3876 // Set output state.
3877 set_control(_gvn.transform(thread_compare_rgn));
3878 set_all_memory(_gvn.transform(thread_compare_mem));
3879 }
3880
3881 #endif // JFR_HAVE_INTRINSICS
3882
3883 //------------------------inline_native_currentCarrierThread------------------
3884 bool LibraryCallKit::inline_native_currentCarrierThread() {
3885 Node* junk = nullptr;
3886 set_result(generate_current_thread(junk));
3887 return true;
3888 }
3889
3890 //------------------------inline_native_currentThread------------------
3891 bool LibraryCallKit::inline_native_currentThread() {
3892 Node* junk = nullptr;
3893 set_result(generate_virtual_thread(junk));
3894 return true;
3895 }
3896
3897 //------------------------inline_native_setVthread------------------
3898 bool LibraryCallKit::inline_native_setCurrentThread() {
3899 assert(C->method()->changes_current_thread(),
3900 "method changes current Thread but is not annotated ChangesCurrentThread");
3901 Node* arr = argument(1);
3902 Node* thread = _gvn.transform(new ThreadLocalNode());
3903 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3904 Node* thread_obj_handle
3905 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3906 thread_obj_handle = _gvn.transform(thread_obj_handle);
3907 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3908 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3909 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3910 return true;
3911 }
3912
3913 const Type* LibraryCallKit::scopedValueCache_type() {
3914 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3915 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3916 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS, /* stable= */ false, /* flat= */ false, /* not_flat= */ true, /* not_null_free= */ true);
3917
3918 // Because we create the scopedValue cache lazily we have to make the
3919 // type of the result BotPTR.
3920 bool xk = etype->klass_is_exact();
3921 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, TypeAryPtr::Offset(0));
3922 return objects_type;
3923 }
3924
3925 Node* LibraryCallKit::scopedValueCache_helper() {
3926 Node* thread = _gvn.transform(new ThreadLocalNode());
3927 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3928 // We cannot use immutable_memory() because we might flip onto a
3929 // different carrier thread, at which point we'll need to use that
3930 // carrier thread's cache.
3931 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3932 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3933 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3934 }
3935
3936 //------------------------inline_native_scopedValueCache------------------
3937 bool LibraryCallKit::inline_native_scopedValueCache() {
3938 Node* cache_obj_handle = scopedValueCache_helper();
3939 const Type* objects_type = scopedValueCache_type();
3940 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3941
3942 return true;
3943 }
3944
3945 //------------------------inline_native_setScopedValueCache------------------
3946 bool LibraryCallKit::inline_native_setScopedValueCache() {
3947 Node* arr = argument(0);
3948 Node* cache_obj_handle = scopedValueCache_helper();
3949 const Type* objects_type = scopedValueCache_type();
3950
3951 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3952 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3953
3954 return true;
3955 }
3956
3957 //------------------------inline_native_Continuation_pin and unpin-----------
3958
3959 // Shared implementation routine for both pin and unpin.
3960 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
3961 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3962
3963 // Save input memory.
3964 Node* input_memory_state = reset_memory();
3965 set_all_memory(input_memory_state);
3966
3967 // TLS
3968 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3969 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
3970 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
3971
3972 // Null check the last continuation object.
3973 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
3974 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
3975 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3976
3977 // False path, last continuation is null.
3978 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
3979
3980 // True path, last continuation is not null.
3981 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
3982
3983 set_control(continuation_is_not_null);
3984
3985 // Load the pin count from the last continuation.
3986 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
3987 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
3988
3989 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
3990 Node* pin_count_rhs;
3991 if (unpin) {
3992 pin_count_rhs = _gvn.intcon(0);
3993 } else {
3994 pin_count_rhs = _gvn.intcon(UINT32_MAX);
3995 }
3996 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
3997 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
3998 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
3999
4000 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
4001 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
4002 set_control(valid_pin_count);
4003
4004 Node* next_pin_count;
4005 if (unpin) {
4006 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
4007 } else {
4008 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
4009 }
4010
4011 Node* updated_pin_count_memory = store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, Compile::AliasIdxRaw, MemNode::unordered);
4012
4013 // True branch, pin count over/underflow.
4014 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
4015 {
4016 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
4017 // which will throw IllegalStateException for pin count over/underflow.
4018 PreserveJVMState pjvms(this);
4019 set_control(pin_count_over_underflow);
4020 set_all_memory(input_memory_state);
4021 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
4022 Deoptimization::Action_none);
4023 assert(stopped(), "invariant");
4024 }
4025
4026 // Result of top level CFG and Memory.
4027 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
4028 record_for_igvn(result_rgn);
4029 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
4030 record_for_igvn(result_mem);
4031
4032 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
4033 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
4034 result_mem->init_req(_true_path, _gvn.transform(updated_pin_count_memory));
4035 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
4036
4037 // Set output state.
4038 set_control(_gvn.transform(result_rgn));
4039 set_all_memory(_gvn.transform(result_mem));
4040
4041 return true;
4042 }
4043
4044 //-----------------------load_klass_from_mirror_common-------------------------
4045 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
4046 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
4047 // and branch to the given path on the region.
4048 // If never_see_null, take an uncommon trap on null, so we can optimistically
4049 // compile for the non-null case.
4050 // If the region is null, force never_see_null = true.
4051 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
4052 bool never_see_null,
4053 RegionNode* region,
4054 int null_path,
4055 int offset) {
4056 if (region == nullptr) never_see_null = true;
4057 Node* p = basic_plus_adr(mirror, offset);
4058 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4059 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
4060 Node* null_ctl = top();
4061 kls = null_check_oop(kls, &null_ctl, never_see_null);
4062 if (region != nullptr) {
4063 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
4064 region->init_req(null_path, null_ctl);
4065 } else {
4066 assert(null_ctl == top(), "no loose ends");
4067 }
4068 return kls;
4069 }
4070
4071 //--------------------(inline_native_Class_query helpers)---------------------
4072 // Use this for JVM_ACC_INTERFACE.
4073 // Fall through if (mods & mask) == bits, take the guard otherwise.
4074 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
4075 ByteSize offset, const Type* type, BasicType bt) {
4076 // Branch around if the given klass has the given modifier bit set.
4077 // Like generate_guard, adds a new path onto the region.
4078 Node* modp = basic_plus_adr(kls, in_bytes(offset));
4079 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
4080 Node* mask = intcon(modifier_mask);
4081 Node* bits = intcon(modifier_bits);
4082 Node* mbit = _gvn.transform(new AndINode(mods, mask));
4083 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
4084 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
4085 return generate_fair_guard(bol, region);
4086 }
4087
4088 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
4089 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
4090 Klass::access_flags_offset(), TypeInt::INT, T_INT);
4091 }
4092
4093 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
4094 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
4095 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
4096 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
4097 }
4098
4099 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
4100 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
4101 }
4102
4103 //-------------------------inline_native_Class_query-------------------
4104 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
4105 const Type* return_type = TypeInt::BOOL;
4106 Node* prim_return_value = top(); // what happens if it's a primitive class?
4107 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4108 bool expect_prim = false; // most of these guys expect to work on refs
4109
4110 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
4111
4112 Node* mirror = argument(0);
4113 Node* obj = top();
4114
4115 switch (id) {
4116 case vmIntrinsics::_isInstance:
4117 // nothing is an instance of a primitive type
4118 prim_return_value = intcon(0);
4119 obj = argument(1);
4120 break;
4121 case vmIntrinsics::_getModifiers:
4122 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
4123 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
4124 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
4125 break;
4126 case vmIntrinsics::_isInterface:
4127 prim_return_value = intcon(0);
4128 break;
4129 case vmIntrinsics::_isArray:
4130 prim_return_value = intcon(0);
4131 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
4132 break;
4133 case vmIntrinsics::_isPrimitive:
4134 prim_return_value = intcon(1);
4135 expect_prim = true; // obviously
4136 break;
4137 case vmIntrinsics::_isHidden:
4138 prim_return_value = intcon(0);
4139 break;
4140 case vmIntrinsics::_getSuperclass:
4141 prim_return_value = null();
4142 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
4143 break;
4144 case vmIntrinsics::_getClassAccessFlags:
4145 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
4146 return_type = TypeInt::INT; // not bool! 6297094
4147 break;
4148 default:
4149 fatal_unexpected_iid(id);
4150 break;
4151 }
4152
4153 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4154 if (mirror_con == nullptr) return false; // cannot happen?
4155
4156 #ifndef PRODUCT
4157 if (C->print_intrinsics() || C->print_inlining()) {
4158 ciType* k = mirror_con->java_mirror_type();
4159 if (k) {
4160 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4161 k->print_name();
4162 tty->cr();
4163 }
4164 }
4165 #endif
4166
4167 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4168 RegionNode* region = new RegionNode(PATH_LIMIT);
4169 record_for_igvn(region);
4170 PhiNode* phi = new PhiNode(region, return_type);
4171
4172 // The mirror will never be null of Reflection.getClassAccessFlags, however
4173 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4174 // if it is. See bug 4774291.
4175
4176 // For Reflection.getClassAccessFlags(), the null check occurs in
4177 // the wrong place; see inline_unsafe_access(), above, for a similar
4178 // situation.
4179 mirror = null_check(mirror);
4180 // If mirror or obj is dead, only null-path is taken.
4181 if (stopped()) return true;
4182
4183 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4184
4185 // Now load the mirror's klass metaobject, and null-check it.
4186 // Side-effects region with the control path if the klass is null.
4187 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4188 // If kls is null, we have a primitive mirror.
4189 phi->init_req(_prim_path, prim_return_value);
4190 if (stopped()) { set_result(region, phi); return true; }
4191 bool safe_for_replace = (region->in(_prim_path) == top());
4192
4193 Node* p; // handy temp
4194 Node* null_ctl;
4195
4196 // Now that we have the non-null klass, we can perform the real query.
4197 // For constant classes, the query will constant-fold in LoadNode::Value.
4198 Node* query_value = top();
4199 switch (id) {
4200 case vmIntrinsics::_isInstance:
4201 // nothing is an instance of a primitive type
4202 query_value = gen_instanceof(obj, kls, safe_for_replace);
4203 break;
4204
4205 case vmIntrinsics::_getModifiers:
4206 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
4207 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
4208 break;
4209
4210 case vmIntrinsics::_isInterface:
4211 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4212 if (generate_interface_guard(kls, region) != nullptr)
4213 // A guard was added. If the guard is taken, it was an interface.
4214 phi->add_req(intcon(1));
4215 // If we fall through, it's a plain class.
4216 query_value = intcon(0);
4217 break;
4218
4219 case vmIntrinsics::_isArray:
4220 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
4221 if (generate_array_guard(kls, region) != nullptr)
4222 // A guard was added. If the guard is taken, it was an array.
4223 phi->add_req(intcon(1));
4224 // If we fall through, it's a plain class.
4225 query_value = intcon(0);
4226 break;
4227
4228 case vmIntrinsics::_isPrimitive:
4229 query_value = intcon(0); // "normal" path produces false
4230 break;
4231
4232 case vmIntrinsics::_isHidden:
4233 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4234 if (generate_hidden_class_guard(kls, region) != nullptr)
4235 // A guard was added. If the guard is taken, it was an hidden class.
4236 phi->add_req(intcon(1));
4237 // If we fall through, it's a plain class.
4238 query_value = intcon(0);
4239 break;
4240
4241
4242 case vmIntrinsics::_getSuperclass:
4243 // The rules here are somewhat unfortunate, but we can still do better
4244 // with random logic than with a JNI call.
4245 // Interfaces store null or Object as _super, but must report null.
4246 // Arrays store an intermediate super as _super, but must report Object.
4247 // Other types can report the actual _super.
4248 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4249 if (generate_interface_guard(kls, region) != nullptr)
4250 // A guard was added. If the guard is taken, it was an interface.
4251 phi->add_req(null());
4252 if (generate_array_guard(kls, region) != nullptr)
4253 // A guard was added. If the guard is taken, it was an array.
4254 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4255 // If we fall through, it's a plain class. Get its _super.
4256 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4257 kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4258 null_ctl = top();
4259 kls = null_check_oop(kls, &null_ctl);
4260 if (null_ctl != top()) {
4261 // If the guard is taken, Object.superClass is null (both klass and mirror).
4262 region->add_req(null_ctl);
4263 phi ->add_req(null());
4264 }
4265 if (!stopped()) {
4266 query_value = load_mirror_from_klass(kls);
4267 }
4268 break;
4269
4270 case vmIntrinsics::_getClassAccessFlags:
4271 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
4272 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
4273 break;
4274
4275 default:
4276 fatal_unexpected_iid(id);
4277 break;
4278 }
4279
4280 // Fall-through is the normal case of a query to a real class.
4281 phi->init_req(1, query_value);
4282 region->init_req(1, control());
4283
4284 C->set_has_split_ifs(true); // Has chance for split-if optimization
4285 set_result(region, phi);
4286 return true;
4287 }
4288
4289
4290 //-------------------------inline_Class_cast-------------------
4291 bool LibraryCallKit::inline_Class_cast() {
4292 Node* mirror = argument(0); // Class
4293 Node* obj = argument(1);
4294 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4295 if (mirror_con == nullptr) {
4296 return false; // dead path (mirror->is_top()).
4297 }
4298 if (obj == nullptr || obj->is_top()) {
4299 return false; // dead path
4300 }
4301 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4302
4303 // First, see if Class.cast() can be folded statically.
4304 // java_mirror_type() returns non-null for compile-time Class constants.
4305 bool is_null_free_array = false;
4306 ciType* tm = mirror_con->java_mirror_type(&is_null_free_array);
4307 if (tm != nullptr && tm->is_klass() &&
4308 tp != nullptr) {
4309 if (!tp->is_loaded()) {
4310 // Don't use intrinsic when class is not loaded.
4311 return false;
4312 } else {
4313 const TypeKlassPtr* tklass = TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces);
4314 if (is_null_free_array) {
4315 tklass = tklass->is_aryklassptr()->cast_to_null_free();
4316 }
4317 int static_res = C->static_subtype_check(tklass, tp->as_klass_type());
4318 if (static_res == Compile::SSC_always_true) {
4319 // isInstance() is true - fold the code.
4320 set_result(obj);
4321 return true;
4322 } else if (static_res == Compile::SSC_always_false) {
4323 // Don't use intrinsic, have to throw ClassCastException.
4324 // If the reference is null, the non-intrinsic bytecode will
4325 // be optimized appropriately.
4326 return false;
4327 }
4328 }
4329 }
4330
4331 // Bailout intrinsic and do normal inlining if exception path is frequent.
4332 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4333 return false;
4334 }
4335
4336 // Generate dynamic checks.
4337 // Class.cast() is java implementation of _checkcast bytecode.
4338 // Do checkcast (Parse::do_checkcast()) optimizations here.
4339
4340 mirror = null_check(mirror);
4341 // If mirror is dead, only null-path is taken.
4342 if (stopped()) {
4343 return true;
4344 }
4345
4346 // Not-subtype or the mirror's klass ptr is nullptr (in case it is a primitive).
4347 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
4348 RegionNode* region = new RegionNode(PATH_LIMIT);
4349 record_for_igvn(region);
4350
4351 // Now load the mirror's klass metaobject, and null-check it.
4352 // If kls is null, we have a primitive mirror and
4353 // nothing is an instance of a primitive type.
4354 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4355
4356 Node* res = top();
4357 Node* io = i_o();
4358 Node* mem = merged_memory();
4359 if (!stopped()) {
4360
4361 Node* bad_type_ctrl = top();
4362 // Do checkcast optimizations.
4363 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4364 region->init_req(_bad_type_path, bad_type_ctrl);
4365 }
4366 if (region->in(_prim_path) != top() ||
4367 region->in(_bad_type_path) != top() ||
4368 region->in(_npe_path) != top()) {
4369 // Let Interpreter throw ClassCastException.
4370 PreserveJVMState pjvms(this);
4371 set_control(_gvn.transform(region));
4372 // Set IO and memory because gen_checkcast may override them when buffering inline types
4373 set_i_o(io);
4374 set_all_memory(mem);
4375 uncommon_trap(Deoptimization::Reason_intrinsic,
4376 Deoptimization::Action_maybe_recompile);
4377 }
4378 if (!stopped()) {
4379 set_result(res);
4380 }
4381 return true;
4382 }
4383
4384
4385 //--------------------------inline_native_subtype_check------------------------
4386 // This intrinsic takes the JNI calls out of the heart of
4387 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4388 bool LibraryCallKit::inline_native_subtype_check() {
4389 // Pull both arguments off the stack.
4390 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4391 args[0] = argument(0);
4392 args[1] = argument(1);
4393 Node* klasses[2]; // corresponding Klasses: superk, subk
4394 klasses[0] = klasses[1] = top();
4395
4396 enum {
4397 // A full decision tree on {superc is prim, subc is prim}:
4398 _prim_0_path = 1, // {P,N} => false
4399 // {P,P} & superc!=subc => false
4400 _prim_same_path, // {P,P} & superc==subc => true
4401 _prim_1_path, // {N,P} => false
4402 _ref_subtype_path, // {N,N} & subtype check wins => true
4403 _both_ref_path, // {N,N} & subtype check loses => false
4404 PATH_LIMIT
4405 };
4406
4407 RegionNode* region = new RegionNode(PATH_LIMIT);
4408 RegionNode* prim_region = new RegionNode(2);
4409 Node* phi = new PhiNode(region, TypeInt::BOOL);
4410 record_for_igvn(region);
4411 record_for_igvn(prim_region);
4412
4413 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4414 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4415 int class_klass_offset = java_lang_Class::klass_offset();
4416
4417 // First null-check both mirrors and load each mirror's klass metaobject.
4418 int which_arg;
4419 for (which_arg = 0; which_arg <= 1; which_arg++) {
4420 Node* arg = args[which_arg];
4421 arg = null_check(arg);
4422 if (stopped()) break;
4423 args[which_arg] = arg;
4424
4425 Node* p = basic_plus_adr(arg, class_klass_offset);
4426 Node* kls = LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, adr_type, kls_type);
4427 klasses[which_arg] = _gvn.transform(kls);
4428 }
4429
4430 // Having loaded both klasses, test each for null.
4431 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4432 for (which_arg = 0; which_arg <= 1; which_arg++) {
4433 Node* kls = klasses[which_arg];
4434 Node* null_ctl = top();
4435 kls = null_check_oop(kls, &null_ctl, never_see_null);
4436 if (which_arg == 0) {
4437 prim_region->init_req(1, null_ctl);
4438 } else {
4439 region->init_req(_prim_1_path, null_ctl);
4440 }
4441 if (stopped()) break;
4442 klasses[which_arg] = kls;
4443 }
4444
4445 if (!stopped()) {
4446 // now we have two reference types, in klasses[0..1]
4447 Node* subk = klasses[1]; // the argument to isAssignableFrom
4448 Node* superk = klasses[0]; // the receiver
4449 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4450 region->set_req(_ref_subtype_path, control());
4451 }
4452
4453 // If both operands are primitive (both klasses null), then
4454 // we must return true when they are identical primitives.
4455 // It is convenient to test this after the first null klass check.
4456 // This path is also used if superc is a value mirror.
4457 set_control(_gvn.transform(prim_region));
4458 if (!stopped()) {
4459 // Since superc is primitive, make a guard for the superc==subc case.
4460 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4461 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4462 generate_fair_guard(bol_eq, region);
4463 if (region->req() == PATH_LIMIT+1) {
4464 // A guard was added. If the added guard is taken, superc==subc.
4465 region->swap_edges(PATH_LIMIT, _prim_same_path);
4466 region->del_req(PATH_LIMIT);
4467 }
4468 region->set_req(_prim_0_path, control()); // Not equal after all.
4469 }
4470
4471 // these are the only paths that produce 'true':
4472 phi->set_req(_prim_same_path, intcon(1));
4473 phi->set_req(_ref_subtype_path, intcon(1));
4474
4475 // pull together the cases:
4476 assert(region->req() == PATH_LIMIT, "sane region");
4477 for (uint i = 1; i < region->req(); i++) {
4478 Node* ctl = region->in(i);
4479 if (ctl == nullptr || ctl == top()) {
4480 region->set_req(i, top());
4481 phi ->set_req(i, top());
4482 } else if (phi->in(i) == nullptr) {
4483 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4484 }
4485 }
4486
4487 set_control(_gvn.transform(region));
4488 set_result(_gvn.transform(phi));
4489 return true;
4490 }
4491
4492 //---------------------generate_array_guard_common------------------------
4493 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind) {
4494
4495 if (stopped()) {
4496 return nullptr;
4497 }
4498
4499 // Like generate_guard, adds a new path onto the region.
4500 jint layout_con = 0;
4501 Node* layout_val = get_layout_helper(kls, layout_con);
4502 if (layout_val == nullptr) {
4503 bool query = 0;
4504 switch(kind) {
4505 case ObjectArray: query = Klass::layout_helper_is_objArray(layout_con); break;
4506 case NonObjectArray: query = !Klass::layout_helper_is_objArray(layout_con); break;
4507 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
4508 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
4509 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
4510 default:
4511 ShouldNotReachHere();
4512 }
4513 if (!query) {
4514 return nullptr; // never a branch
4515 } else { // always a branch
4516 Node* always_branch = control();
4517 if (region != nullptr)
4518 region->add_req(always_branch);
4519 set_control(top());
4520 return always_branch;
4521 }
4522 }
4523 unsigned int value = 0;
4524 BoolTest::mask btest = BoolTest::illegal;
4525 switch(kind) {
4526 case ObjectArray:
4527 case NonObjectArray: {
4528 value = Klass::_lh_array_tag_obj_value;
4529 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4530 btest = (kind == ObjectArray) ? BoolTest::eq : BoolTest::ne;
4531 break;
4532 }
4533 case TypeArray: {
4534 value = Klass::_lh_array_tag_type_value;
4535 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4536 btest = BoolTest::eq;
4537 break;
4538 }
4539 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
4540 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
4541 default:
4542 ShouldNotReachHere();
4543 }
4544 // Now test the correct condition.
4545 jint nval = (jint)value;
4546 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4547 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4548 return generate_fair_guard(bol, region);
4549 }
4550
4551 //-----------------------inline_newNullRestrictedArray--------------------------
4552 // public static native Object[] newNullRestrictedArray(Class<?> componentType, int length);
4553 bool LibraryCallKit::inline_newNullRestrictedArray() {
4554 Node* componentType = argument(0);
4555 Node* length = argument(1);
4556
4557 const TypeInstPtr* tp = _gvn.type(componentType)->isa_instptr();
4558 if (tp != nullptr) {
4559 ciInstanceKlass* ik = tp->instance_klass();
4560 if (ik == C->env()->Class_klass()) {
4561 ciType* t = tp->java_mirror_type();
4562 if (t != nullptr && t->is_inlinetype()) {
4563 ciArrayKlass* array_klass = ciArrayKlass::make(t, true);
4564 if (array_klass->is_loaded() && array_klass->element_klass()->as_inline_klass()->is_initialized()) {
4565 const TypeAryKlassPtr* array_klass_type = TypeKlassPtr::make(array_klass, Type::trust_interfaces)->is_aryklassptr();
4566 array_klass_type = array_klass_type->cast_to_null_free();
4567 Node* obj = new_array(makecon(array_klass_type), length, 0, nullptr, false); // no arguments to push
4568 set_result(obj);
4569 assert(gvn().type(obj)->is_aryptr()->is_null_free(), "must be null-free");
4570 return true;
4571 }
4572 }
4573 }
4574 }
4575 return false;
4576 }
4577
4578 //-----------------------inline_native_newArray--------------------------
4579 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
4580 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4581 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4582 Node* mirror;
4583 Node* count_val;
4584 if (uninitialized) {
4585 null_check_receiver();
4586 mirror = argument(1);
4587 count_val = argument(2);
4588 } else {
4589 mirror = argument(0);
4590 count_val = argument(1);
4591 }
4592
4593 mirror = null_check(mirror);
4594 // If mirror or obj is dead, only null-path is taken.
4595 if (stopped()) return true;
4596
4597 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4598 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4599 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4600 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4601 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4602
4603 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4604 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4605 result_reg, _slow_path);
4606 Node* normal_ctl = control();
4607 Node* no_array_ctl = result_reg->in(_slow_path);
4608
4609 // Generate code for the slow case. We make a call to newArray().
4610 set_control(no_array_ctl);
4611 if (!stopped()) {
4612 // Either the input type is void.class, or else the
4613 // array klass has not yet been cached. Either the
4614 // ensuing call will throw an exception, or else it
4615 // will cache the array klass for next time.
4616 PreserveJVMState pjvms(this);
4617 CallJavaNode* slow_call = nullptr;
4618 if (uninitialized) {
4619 // Generate optimized virtual call (holder class 'Unsafe' is final)
4620 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4621 } else {
4622 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4623 }
4624 Node* slow_result = set_results_for_java_call(slow_call);
4625 // this->control() comes from set_results_for_java_call
4626 result_reg->set_req(_slow_path, control());
4627 result_val->set_req(_slow_path, slow_result);
4628 result_io ->set_req(_slow_path, i_o());
4629 result_mem->set_req(_slow_path, reset_memory());
4630 }
4631
4632 set_control(normal_ctl);
4633 if (!stopped()) {
4634 // Normal case: The array type has been cached in the java.lang.Class.
4635 // The following call works fine even if the array type is polymorphic.
4636 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4637 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4638 result_reg->init_req(_normal_path, control());
4639 result_val->init_req(_normal_path, obj);
4640 result_io ->init_req(_normal_path, i_o());
4641 result_mem->init_req(_normal_path, reset_memory());
4642
4643 if (uninitialized) {
4644 // Mark the allocation so that zeroing is skipped
4645 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4646 alloc->maybe_set_complete(&_gvn);
4647 }
4648 }
4649
4650 // Return the combined state.
4651 set_i_o( _gvn.transform(result_io) );
4652 set_all_memory( _gvn.transform(result_mem));
4653
4654 C->set_has_split_ifs(true); // Has chance for split-if optimization
4655 set_result(result_reg, result_val);
4656 return true;
4657 }
4658
4659 //----------------------inline_native_getLength--------------------------
4660 // public static native int java.lang.reflect.Array.getLength(Object array);
4661 bool LibraryCallKit::inline_native_getLength() {
4662 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4663
4664 Node* array = null_check(argument(0));
4665 // If array is dead, only null-path is taken.
4666 if (stopped()) return true;
4667
4668 // Deoptimize if it is a non-array.
4669 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr);
4670
4671 if (non_array != nullptr) {
4672 PreserveJVMState pjvms(this);
4673 set_control(non_array);
4674 uncommon_trap(Deoptimization::Reason_intrinsic,
4675 Deoptimization::Action_maybe_recompile);
4676 }
4677
4678 // If control is dead, only non-array-path is taken.
4679 if (stopped()) return true;
4680
4681 // The works fine even if the array type is polymorphic.
4682 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4683 Node* result = load_array_length(array);
4684
4685 C->set_has_split_ifs(true); // Has chance for split-if optimization
4686 set_result(result);
4687 return true;
4688 }
4689
4690 //------------------------inline_array_copyOf----------------------------
4691 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4692 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4693 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4694 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4695
4696 // Get the arguments.
4697 Node* original = argument(0);
4698 Node* start = is_copyOfRange? argument(1): intcon(0);
4699 Node* end = is_copyOfRange? argument(2): argument(1);
4700 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4701
4702 Node* newcopy = nullptr;
4703
4704 // Set the original stack and the reexecute bit for the interpreter to reexecute
4705 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4706 { PreserveReexecuteState preexecs(this);
4707 jvms()->set_should_reexecute(true);
4708
4709 array_type_mirror = null_check(array_type_mirror);
4710 original = null_check(original);
4711
4712 // Check if a null path was taken unconditionally.
4713 if (stopped()) return true;
4714
4715 Node* orig_length = load_array_length(original);
4716
4717 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4718 klass_node = null_check(klass_node);
4719
4720 RegionNode* bailout = new RegionNode(1);
4721 record_for_igvn(bailout);
4722
4723 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4724 // Bail out if that is so.
4725 // Inline type array may have object field that would require a
4726 // write barrier. Conservatively, go to slow path.
4727 // TODO 8251971: Optimize for the case when flat src/dst are later found
4728 // to not contain oops (i.e., move this check to the macro expansion phase).
4729 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4730 const TypeAryPtr* orig_t = _gvn.type(original)->isa_aryptr();
4731 const TypeKlassPtr* tklass = _gvn.type(klass_node)->is_klassptr();
4732 bool exclude_flat = UseFlatArray && bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing) &&
4733 // Can src array be flat and contain oops?
4734 (orig_t == nullptr || (!orig_t->is_not_flat() && (!orig_t->is_flat() || orig_t->elem()->inline_klass()->contains_oops()))) &&
4735 // Can dest array be flat and contain oops?
4736 tklass->can_be_inline_array() && (!tklass->is_flat() || tklass->is_aryklassptr()->elem()->is_instklassptr()->instance_klass()->as_inline_klass()->contains_oops());
4737 Node* not_objArray = exclude_flat ? generate_non_objArray_guard(klass_node, bailout) : generate_typeArray_guard(klass_node, bailout);
4738 if (not_objArray != nullptr) {
4739 // Improve the klass node's type from the new optimistic assumption:
4740 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4741 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0));
4742 Node* cast = new CastPPNode(control(), klass_node, akls);
4743 klass_node = _gvn.transform(cast);
4744 }
4745
4746 // Bail out if either start or end is negative.
4747 generate_negative_guard(start, bailout, &start);
4748 generate_negative_guard(end, bailout, &end);
4749
4750 Node* length = end;
4751 if (_gvn.type(start) != TypeInt::ZERO) {
4752 length = _gvn.transform(new SubINode(end, start));
4753 }
4754
4755 // Bail out if length is negative (i.e., if start > end).
4756 // Without this the new_array would throw
4757 // NegativeArraySizeException but IllegalArgumentException is what
4758 // should be thrown
4759 generate_negative_guard(length, bailout, &length);
4760
4761 // Handle inline type arrays
4762 bool can_validate = !too_many_traps(Deoptimization::Reason_class_check);
4763 if (!stopped()) {
4764 // TODO JDK-8329224
4765 if (!orig_t->is_null_free()) {
4766 // Not statically known to be null free, add a check
4767 generate_fair_guard(null_free_array_test(original), bailout);
4768 }
4769 orig_t = _gvn.type(original)->isa_aryptr();
4770 if (orig_t != nullptr && orig_t->is_flat()) {
4771 // Src is flat, check that dest is flat as well
4772 if (exclude_flat) {
4773 // Dest can't be flat, bail out
4774 bailout->add_req(control());
4775 set_control(top());
4776 } else {
4777 generate_fair_guard(flat_array_test(klass_node, /* flat = */ false), bailout);
4778 }
4779 } else if (UseFlatArray && (orig_t == nullptr || !orig_t->is_not_flat()) &&
4780 // If dest is flat, src must be flat as well (guaranteed by src <: dest check if validated).
4781 ((!tklass->is_flat() && tklass->can_be_inline_array()) || !can_validate)) {
4782 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
4783 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
4784 generate_fair_guard(flat_array_test(load_object_klass(original)), bailout);
4785 if (orig_t != nullptr) {
4786 orig_t = orig_t->cast_to_not_flat();
4787 original = _gvn.transform(new CheckCastPPNode(control(), original, orig_t));
4788 }
4789 }
4790 if (!can_validate) {
4791 // No validation. The subtype check emitted at macro expansion time will not go to the slow
4792 // path but call checkcast_arraycopy which can not handle flat/null-free inline type arrays.
4793 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat/null-free.
4794 generate_fair_guard(flat_array_test(klass_node), bailout);
4795 generate_fair_guard(null_free_array_test(original), bailout);
4796 }
4797 }
4798
4799 // Bail out if start is larger than the original length
4800 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4801 generate_negative_guard(orig_tail, bailout, &orig_tail);
4802
4803 if (bailout->req() > 1) {
4804 PreserveJVMState pjvms(this);
4805 set_control(_gvn.transform(bailout));
4806 uncommon_trap(Deoptimization::Reason_intrinsic,
4807 Deoptimization::Action_maybe_recompile);
4808 }
4809
4810 if (!stopped()) {
4811 // How many elements will we copy from the original?
4812 // The answer is MinI(orig_tail, length).
4813 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4814
4815 // Generate a direct call to the right arraycopy function(s).
4816 // We know the copy is disjoint but we might not know if the
4817 // oop stores need checking.
4818 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4819 // This will fail a store-check if x contains any non-nulls.
4820
4821 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4822 // loads/stores but it is legal only if we're sure the
4823 // Arrays.copyOf would succeed. So we need all input arguments
4824 // to the copyOf to be validated, including that the copy to the
4825 // new array won't trigger an ArrayStoreException. That subtype
4826 // check can be optimized if we know something on the type of
4827 // the input array from type speculation.
4828 if (_gvn.type(klass_node)->singleton()) {
4829 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4830 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4831
4832 int test = C->static_subtype_check(superk, subk);
4833 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4834 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4835 if (t_original->speculative_type() != nullptr) {
4836 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4837 }
4838 }
4839 }
4840
4841 bool validated = false;
4842 // Reason_class_check rather than Reason_intrinsic because we
4843 // want to intrinsify even if this traps.
4844 if (can_validate) {
4845 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4846
4847 if (not_subtype_ctrl != top()) {
4848 PreserveJVMState pjvms(this);
4849 set_control(not_subtype_ctrl);
4850 uncommon_trap(Deoptimization::Reason_class_check,
4851 Deoptimization::Action_make_not_entrant);
4852 assert(stopped(), "Should be stopped");
4853 }
4854 validated = true;
4855 }
4856
4857 if (!stopped()) {
4858 newcopy = new_array(klass_node, length, 0); // no arguments to push
4859
4860 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4861 load_object_klass(original), klass_node);
4862 if (!is_copyOfRange) {
4863 ac->set_copyof(validated);
4864 } else {
4865 ac->set_copyofrange(validated);
4866 }
4867 Node* n = _gvn.transform(ac);
4868 if (n == ac) {
4869 ac->connect_outputs(this);
4870 } else {
4871 assert(validated, "shouldn't transform if all arguments not validated");
4872 set_all_memory(n);
4873 }
4874 }
4875 }
4876 } // original reexecute is set back here
4877
4878 C->set_has_split_ifs(true); // Has chance for split-if optimization
4879 if (!stopped()) {
4880 set_result(newcopy);
4881 }
4882 return true;
4883 }
4884
4885
4886 //----------------------generate_virtual_guard---------------------------
4887 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4888 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4889 RegionNode* slow_region) {
4890 ciMethod* method = callee();
4891 int vtable_index = method->vtable_index();
4892 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4893 "bad index %d", vtable_index);
4894 // Get the Method* out of the appropriate vtable entry.
4895 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4896 vtable_index*vtableEntry::size_in_bytes() +
4897 in_bytes(vtableEntry::method_offset());
4898 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4899 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4900
4901 // Compare the target method with the expected method (e.g., Object.hashCode).
4902 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4903
4904 Node* native_call = makecon(native_call_addr);
4905 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4906 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4907
4908 return generate_slow_guard(test_native, slow_region);
4909 }
4910
4911 //-----------------------generate_method_call----------------------------
4912 // Use generate_method_call to make a slow-call to the real
4913 // method if the fast path fails. An alternative would be to
4914 // use a stub like OptoRuntime::slow_arraycopy_Java.
4915 // This only works for expanding the current library call,
4916 // not another intrinsic. (E.g., don't use this for making an
4917 // arraycopy call inside of the copyOf intrinsic.)
4918 CallJavaNode*
4919 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4920 // When compiling the intrinsic method itself, do not use this technique.
4921 guarantee(callee() != C->method(), "cannot make slow-call to self");
4922
4923 ciMethod* method = callee();
4924 // ensure the JVMS we have will be correct for this call
4925 guarantee(method_id == method->intrinsic_id(), "must match");
4926
4927 const TypeFunc* tf = TypeFunc::make(method);
4928 if (res_not_null) {
4929 assert(tf->return_type() == T_OBJECT, "");
4930 const TypeTuple* range = tf->range_cc();
4931 const Type** fields = TypeTuple::fields(range->cnt());
4932 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4933 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4934 tf = TypeFunc::make(tf->domain_cc(), new_range);
4935 }
4936 CallJavaNode* slow_call;
4937 if (is_static) {
4938 assert(!is_virtual, "");
4939 slow_call = new CallStaticJavaNode(C, tf,
4940 SharedRuntime::get_resolve_static_call_stub(), method);
4941 } else if (is_virtual) {
4942 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4943 int vtable_index = Method::invalid_vtable_index;
4944 if (UseInlineCaches) {
4945 // Suppress the vtable call
4946 } else {
4947 // hashCode and clone are not a miranda methods,
4948 // so the vtable index is fixed.
4949 // No need to use the linkResolver to get it.
4950 vtable_index = method->vtable_index();
4951 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4952 "bad index %d", vtable_index);
4953 }
4954 slow_call = new CallDynamicJavaNode(tf,
4955 SharedRuntime::get_resolve_virtual_call_stub(),
4956 method, vtable_index);
4957 } else { // neither virtual nor static: opt_virtual
4958 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4959 slow_call = new CallStaticJavaNode(C, tf,
4960 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4961 slow_call->set_optimized_virtual(true);
4962 }
4963 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4964 // To be able to issue a direct call (optimized virtual or virtual)
4965 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4966 // about the method being invoked should be attached to the call site to
4967 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4968 slow_call->set_override_symbolic_info(true);
4969 }
4970 set_arguments_for_java_call(slow_call);
4971 set_edges_for_java_call(slow_call);
4972 return slow_call;
4973 }
4974
4975
4976 /**
4977 * Build special case code for calls to hashCode on an object. This call may
4978 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4979 * slightly different code.
4980 */
4981 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4982 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4983 assert(!(is_virtual && is_static), "either virtual, special, or static");
4984
4985 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4986
4987 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4988 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4989 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4990 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4991 Node* obj = argument(0);
4992
4993 // Don't intrinsify hashcode on inline types for now.
4994 // The "is locked" runtime check below also serves as inline type check and goes to the slow path.
4995 if (gvn().type(obj)->is_inlinetypeptr()) {
4996 return false;
4997 }
4998
4999 if (!is_static) {
5000 // Check for hashing null object
5001 obj = null_check_receiver();
5002 if (stopped()) return true; // unconditionally null
5003 result_reg->init_req(_null_path, top());
5004 result_val->init_req(_null_path, top());
5005 } else {
5006 // Do a null check, and return zero if null.
5007 // System.identityHashCode(null) == 0
5008 Node* null_ctl = top();
5009 obj = null_check_oop(obj, &null_ctl);
5010 result_reg->init_req(_null_path, null_ctl);
5011 result_val->init_req(_null_path, _gvn.intcon(0));
5012 }
5013
5014 // Unconditionally null? Then return right away.
5015 if (stopped()) {
5016 set_control( result_reg->in(_null_path));
5017 if (!stopped())
5018 set_result(result_val->in(_null_path));
5019 return true;
5020 }
5021
5022 // We only go to the fast case code if we pass a number of guards. The
5023 // paths which do not pass are accumulated in the slow_region.
5024 RegionNode* slow_region = new RegionNode(1);
5025 record_for_igvn(slow_region);
5026
5027 // If this is a virtual call, we generate a funny guard. We pull out
5028 // the vtable entry corresponding to hashCode() from the target object.
5029 // If the target method which we are calling happens to be the native
5030 // Object hashCode() method, we pass the guard. We do not need this
5031 // guard for non-virtual calls -- the caller is known to be the native
5032 // Object hashCode().
5033 if (is_virtual) {
5034 // After null check, get the object's klass.
5035 Node* obj_klass = load_object_klass(obj);
5036 generate_virtual_guard(obj_klass, slow_region);
5037 }
5038
5039 // Get the header out of the object, use LoadMarkNode when available
5040 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
5041 // The control of the load must be null. Otherwise, the load can move before
5042 // the null check after castPP removal.
5043 Node* no_ctrl = nullptr;
5044 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
5045
5046 if (!UseObjectMonitorTable) {
5047 // Test the header to see if it is safe to read w.r.t. locking.
5048 // This also serves as guard against inline types
5049 Node *lock_mask = _gvn.MakeConX(markWord::inline_type_mask_in_place);
5050 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
5051 if (LockingMode == LM_LIGHTWEIGHT) {
5052 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
5053 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
5054 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
5055
5056 generate_slow_guard(test_monitor, slow_region);
5057 } else {
5058 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
5059 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val));
5060 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne));
5061
5062 generate_slow_guard(test_not_unlocked, slow_region);
5063 }
5064 }
5065
5066 // Get the hash value and check to see that it has been properly assigned.
5067 // We depend on hash_mask being at most 32 bits and avoid the use of
5068 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
5069 // vm: see markWord.hpp.
5070 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
5071 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
5072 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
5073 // This hack lets the hash bits live anywhere in the mark object now, as long
5074 // as the shift drops the relevant bits into the low 32 bits. Note that
5075 // Java spec says that HashCode is an int so there's no point in capturing
5076 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
5077 hshifted_header = ConvX2I(hshifted_header);
5078 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
5079
5080 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
5081 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
5082 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
5083
5084 generate_slow_guard(test_assigned, slow_region);
5085
5086 Node* init_mem = reset_memory();
5087 // fill in the rest of the null path:
5088 result_io ->init_req(_null_path, i_o());
5089 result_mem->init_req(_null_path, init_mem);
5090
5091 result_val->init_req(_fast_path, hash_val);
5092 result_reg->init_req(_fast_path, control());
5093 result_io ->init_req(_fast_path, i_o());
5094 result_mem->init_req(_fast_path, init_mem);
5095
5096 // Generate code for the slow case. We make a call to hashCode().
5097 set_control(_gvn.transform(slow_region));
5098 if (!stopped()) {
5099 // No need for PreserveJVMState, because we're using up the present state.
5100 set_all_memory(init_mem);
5101 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
5102 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
5103 Node* slow_result = set_results_for_java_call(slow_call);
5104 // this->control() comes from set_results_for_java_call
5105 result_reg->init_req(_slow_path, control());
5106 result_val->init_req(_slow_path, slow_result);
5107 result_io ->set_req(_slow_path, i_o());
5108 result_mem ->set_req(_slow_path, reset_memory());
5109 }
5110
5111 // Return the combined state.
5112 set_i_o( _gvn.transform(result_io) );
5113 set_all_memory( _gvn.transform(result_mem));
5114
5115 set_result(result_reg, result_val);
5116 return true;
5117 }
5118
5119 //---------------------------inline_native_getClass----------------------------
5120 // public final native Class<?> java.lang.Object.getClass();
5121 //
5122 // Build special case code for calls to getClass on an object.
5123 bool LibraryCallKit::inline_native_getClass() {
5124 Node* obj = argument(0);
5125 if (obj->is_InlineType()) {
5126 const Type* t = _gvn.type(obj);
5127 if (t->maybe_null()) {
5128 null_check(obj);
5129 }
5130 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror())));
5131 return true;
5132 }
5133 obj = null_check_receiver();
5134 if (stopped()) return true;
5135 set_result(load_mirror_from_klass(load_object_klass(obj)));
5136 return true;
5137 }
5138
5139 //-----------------inline_native_Reflection_getCallerClass---------------------
5140 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
5141 //
5142 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
5143 //
5144 // NOTE: This code must perform the same logic as JVM_GetCallerClass
5145 // in that it must skip particular security frames and checks for
5146 // caller sensitive methods.
5147 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
5148 #ifndef PRODUCT
5149 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5150 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
5151 }
5152 #endif
5153
5154 if (!jvms()->has_method()) {
5155 #ifndef PRODUCT
5156 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5157 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
5158 }
5159 #endif
5160 return false;
5161 }
5162
5163 // Walk back up the JVM state to find the caller at the required
5164 // depth.
5165 JVMState* caller_jvms = jvms();
5166
5167 // Cf. JVM_GetCallerClass
5168 // NOTE: Start the loop at depth 1 because the current JVM state does
5169 // not include the Reflection.getCallerClass() frame.
5170 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
5171 ciMethod* m = caller_jvms->method();
5172 switch (n) {
5173 case 0:
5174 fatal("current JVM state does not include the Reflection.getCallerClass frame");
5175 break;
5176 case 1:
5177 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
5178 if (!m->caller_sensitive()) {
5179 #ifndef PRODUCT
5180 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5181 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
5182 }
5183 #endif
5184 return false; // bail-out; let JVM_GetCallerClass do the work
5185 }
5186 break;
5187 default:
5188 if (!m->is_ignored_by_security_stack_walk()) {
5189 // We have reached the desired frame; return the holder class.
5190 // Acquire method holder as java.lang.Class and push as constant.
5191 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
5192 ciInstance* caller_mirror = caller_klass->java_mirror();
5193 set_result(makecon(TypeInstPtr::make(caller_mirror)));
5194
5195 #ifndef PRODUCT
5196 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5197 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());
5198 tty->print_cr(" JVM state at this point:");
5199 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5200 ciMethod* m = jvms()->of_depth(i)->method();
5201 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5202 }
5203 }
5204 #endif
5205 return true;
5206 }
5207 break;
5208 }
5209 }
5210
5211 #ifndef PRODUCT
5212 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5213 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
5214 tty->print_cr(" JVM state at this point:");
5215 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5216 ciMethod* m = jvms()->of_depth(i)->method();
5217 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5218 }
5219 }
5220 #endif
5221
5222 return false; // bail-out; let JVM_GetCallerClass do the work
5223 }
5224
5225 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
5226 Node* arg = argument(0);
5227 Node* result = nullptr;
5228
5229 switch (id) {
5230 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
5231 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
5232 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5233 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5234 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5235 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5236
5237 case vmIntrinsics::_doubleToLongBits: {
5238 // two paths (plus control) merge in a wood
5239 RegionNode *r = new RegionNode(3);
5240 Node *phi = new PhiNode(r, TypeLong::LONG);
5241
5242 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5243 // Build the boolean node
5244 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5245
5246 // Branch either way.
5247 // NaN case is less traveled, which makes all the difference.
5248 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5249 Node *opt_isnan = _gvn.transform(ifisnan);
5250 assert( opt_isnan->is_If(), "Expect an IfNode");
5251 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5252 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5253
5254 set_control(iftrue);
5255
5256 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5257 Node *slow_result = longcon(nan_bits); // return NaN
5258 phi->init_req(1, _gvn.transform( slow_result ));
5259 r->init_req(1, iftrue);
5260
5261 // Else fall through
5262 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5263 set_control(iffalse);
5264
5265 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5266 r->init_req(2, iffalse);
5267
5268 // Post merge
5269 set_control(_gvn.transform(r));
5270 record_for_igvn(r);
5271
5272 C->set_has_split_ifs(true); // Has chance for split-if optimization
5273 result = phi;
5274 assert(result->bottom_type()->isa_long(), "must be");
5275 break;
5276 }
5277
5278 case vmIntrinsics::_floatToIntBits: {
5279 // two paths (plus control) merge in a wood
5280 RegionNode *r = new RegionNode(3);
5281 Node *phi = new PhiNode(r, TypeInt::INT);
5282
5283 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5284 // Build the boolean node
5285 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5286
5287 // Branch either way.
5288 // NaN case is less traveled, which makes all the difference.
5289 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5290 Node *opt_isnan = _gvn.transform(ifisnan);
5291 assert( opt_isnan->is_If(), "Expect an IfNode");
5292 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5293 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5294
5295 set_control(iftrue);
5296
5297 static const jint nan_bits = 0x7fc00000;
5298 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5299 phi->init_req(1, _gvn.transform( slow_result ));
5300 r->init_req(1, iftrue);
5301
5302 // Else fall through
5303 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5304 set_control(iffalse);
5305
5306 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5307 r->init_req(2, iffalse);
5308
5309 // Post merge
5310 set_control(_gvn.transform(r));
5311 record_for_igvn(r);
5312
5313 C->set_has_split_ifs(true); // Has chance for split-if optimization
5314 result = phi;
5315 assert(result->bottom_type()->isa_int(), "must be");
5316 break;
5317 }
5318
5319 default:
5320 fatal_unexpected_iid(id);
5321 break;
5322 }
5323 set_result(_gvn.transform(result));
5324 return true;
5325 }
5326
5327 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5328 Node* arg = argument(0);
5329 Node* result = nullptr;
5330
5331 switch (id) {
5332 case vmIntrinsics::_floatIsInfinite:
5333 result = new IsInfiniteFNode(arg);
5334 break;
5335 case vmIntrinsics::_floatIsFinite:
5336 result = new IsFiniteFNode(arg);
5337 break;
5338 case vmIntrinsics::_doubleIsInfinite:
5339 result = new IsInfiniteDNode(arg);
5340 break;
5341 case vmIntrinsics::_doubleIsFinite:
5342 result = new IsFiniteDNode(arg);
5343 break;
5344 default:
5345 fatal_unexpected_iid(id);
5346 break;
5347 }
5348 set_result(_gvn.transform(result));
5349 return true;
5350 }
5351
5352 //----------------------inline_unsafe_copyMemory-------------------------
5353 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5354
5355 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5356 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5357 const Type* base_t = gvn.type(base);
5358
5359 bool in_native = (base_t == TypePtr::NULL_PTR);
5360 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5361 bool is_mixed = !in_heap && !in_native;
5362
5363 if (is_mixed) {
5364 return true; // mixed accesses can touch both on-heap and off-heap memory
5365 }
5366 if (in_heap) {
5367 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5368 if (!is_prim_array) {
5369 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5370 // there's not enough type information available to determine proper memory slice for it.
5371 return true;
5372 }
5373 }
5374 return false;
5375 }
5376
5377 bool LibraryCallKit::inline_unsafe_copyMemory() {
5378 if (callee()->is_static()) return false; // caller must have the capability!
5379 null_check_receiver(); // null-check receiver
5380 if (stopped()) return true;
5381
5382 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5383
5384 Node* src_base = argument(1); // type: oop
5385 Node* src_off = ConvL2X(argument(2)); // type: long
5386 Node* dst_base = argument(4); // type: oop
5387 Node* dst_off = ConvL2X(argument(5)); // type: long
5388 Node* size = ConvL2X(argument(7)); // type: long
5389
5390 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5391 "fieldOffset must be byte-scaled");
5392
5393 Node* src_addr = make_unsafe_address(src_base, src_off);
5394 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5395
5396 Node* thread = _gvn.transform(new ThreadLocalNode());
5397 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5398 BasicType doing_unsafe_access_bt = T_BYTE;
5399 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5400
5401 // update volatile field
5402 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5403
5404 int flags = RC_LEAF | RC_NO_FP;
5405
5406 const TypePtr* dst_type = TypePtr::BOTTOM;
5407
5408 // Adjust memory effects of the runtime call based on input values.
5409 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5410 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5411 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5412
5413 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5414 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5415 flags |= RC_NARROW_MEM; // narrow in memory
5416 }
5417 }
5418
5419 // Call it. Note that the length argument is not scaled.
5420 make_runtime_call(flags,
5421 OptoRuntime::fast_arraycopy_Type(),
5422 StubRoutines::unsafe_arraycopy(),
5423 "unsafe_arraycopy",
5424 dst_type,
5425 src_addr, dst_addr, size XTOP);
5426
5427 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5428
5429 return true;
5430 }
5431
5432 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5433 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5434 bool LibraryCallKit::inline_unsafe_setMemory() {
5435 if (callee()->is_static()) return false; // caller must have the capability!
5436 null_check_receiver(); // null-check receiver
5437 if (stopped()) return true;
5438
5439 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5440
5441 Node* dst_base = argument(1); // type: oop
5442 Node* dst_off = ConvL2X(argument(2)); // type: long
5443 Node* size = ConvL2X(argument(4)); // type: long
5444 Node* byte = argument(6); // type: byte
5445
5446 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5447 "fieldOffset must be byte-scaled");
5448
5449 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5450
5451 Node* thread = _gvn.transform(new ThreadLocalNode());
5452 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5453 BasicType doing_unsafe_access_bt = T_BYTE;
5454 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5455
5456 // update volatile field
5457 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5458
5459 int flags = RC_LEAF | RC_NO_FP;
5460
5461 const TypePtr* dst_type = TypePtr::BOTTOM;
5462
5463 // Adjust memory effects of the runtime call based on input values.
5464 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5465 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5466
5467 flags |= RC_NARROW_MEM; // narrow in memory
5468 }
5469
5470 // Call it. Note that the length argument is not scaled.
5471 make_runtime_call(flags,
5472 OptoRuntime::make_setmemory_Type(),
5473 StubRoutines::unsafe_setmemory(),
5474 "unsafe_setmemory",
5475 dst_type,
5476 dst_addr, size XTOP, byte);
5477
5478 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5479
5480 return true;
5481 }
5482
5483 #undef XTOP
5484
5485 //----------------------inline_unsafe_isFlatArray------------------------
5486 // public native boolean Unsafe.isFlatArray(Class<?> arrayClass);
5487 // This intrinsic exploits assumptions made by the native implementation
5488 // (arrayClass is neither null nor primitive) to avoid unnecessary null checks.
5489 bool LibraryCallKit::inline_unsafe_isFlatArray() {
5490 Node* cls = argument(1);
5491 Node* p = basic_plus_adr(cls, java_lang_Class::klass_offset());
5492 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p,
5493 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT));
5494 Node* result = flat_array_test(kls);
5495 set_result(result);
5496 return true;
5497 }
5498
5499 //------------------------clone_coping-----------------------------------
5500 // Helper function for inline_native_clone.
5501 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5502 assert(obj_size != nullptr, "");
5503 Node* raw_obj = alloc_obj->in(1);
5504 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5505
5506 AllocateNode* alloc = nullptr;
5507 if (ReduceBulkZeroing &&
5508 // If we are implementing an array clone without knowing its source type
5509 // (can happen when compiling the array-guarded branch of a reflective
5510 // Object.clone() invocation), initialize the array within the allocation.
5511 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5512 // to a runtime clone call that assumes fully initialized source arrays.
5513 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5514 // We will be completely responsible for initializing this object -
5515 // mark Initialize node as complete.
5516 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5517 // The object was just allocated - there should be no any stores!
5518 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5519 // Mark as complete_with_arraycopy so that on AllocateNode
5520 // expansion, we know this AllocateNode is initialized by an array
5521 // copy and a StoreStore barrier exists after the array copy.
5522 alloc->initialization()->set_complete_with_arraycopy();
5523 }
5524
5525 Node* size = _gvn.transform(obj_size);
5526 access_clone(obj, alloc_obj, size, is_array);
5527
5528 // Do not let reads from the cloned object float above the arraycopy.
5529 if (alloc != nullptr) {
5530 // Do not let stores that initialize this object be reordered with
5531 // a subsequent store that would make this object accessible by
5532 // other threads.
5533 // Record what AllocateNode this StoreStore protects so that
5534 // escape analysis can go from the MemBarStoreStoreNode to the
5535 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5536 // based on the escape status of the AllocateNode.
5537 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5538 } else {
5539 insert_mem_bar(Op_MemBarCPUOrder);
5540 }
5541 }
5542
5543 //------------------------inline_native_clone----------------------------
5544 // protected native Object java.lang.Object.clone();
5545 //
5546 // Here are the simple edge cases:
5547 // null receiver => normal trap
5548 // virtual and clone was overridden => slow path to out-of-line clone
5549 // not cloneable or finalizer => slow path to out-of-line Object.clone
5550 //
5551 // The general case has two steps, allocation and copying.
5552 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5553 //
5554 // Copying also has two cases, oop arrays and everything else.
5555 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5556 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5557 //
5558 // These steps fold up nicely if and when the cloned object's klass
5559 // can be sharply typed as an object array, a type array, or an instance.
5560 //
5561 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5562 PhiNode* result_val;
5563
5564 // Set the reexecute bit for the interpreter to reexecute
5565 // the bytecode that invokes Object.clone if deoptimization happens.
5566 { PreserveReexecuteState preexecs(this);
5567 jvms()->set_should_reexecute(true);
5568
5569 Node* obj = argument(0);
5570 obj = null_check_receiver();
5571 if (stopped()) return true;
5572
5573 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5574 if (obj_type->is_inlinetypeptr()) {
5575 // If the object to clone is an inline type, we can simply return it (i.e. a nop) since inline types have
5576 // no identity.
5577 set_result(obj);
5578 return true;
5579 }
5580
5581 // If we are going to clone an instance, we need its exact type to
5582 // know the number and types of fields to convert the clone to
5583 // loads/stores. Maybe a speculative type can help us.
5584 if (!obj_type->klass_is_exact() &&
5585 obj_type->speculative_type() != nullptr &&
5586 obj_type->speculative_type()->is_instance_klass() &&
5587 !obj_type->speculative_type()->is_inlinetype()) {
5588 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5589 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5590 !spec_ik->has_injected_fields()) {
5591 if (!obj_type->isa_instptr() ||
5592 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5593 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5594 }
5595 }
5596 }
5597
5598 // Conservatively insert a memory barrier on all memory slices.
5599 // Do not let writes into the original float below the clone.
5600 insert_mem_bar(Op_MemBarCPUOrder);
5601
5602 // paths into result_reg:
5603 enum {
5604 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5605 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5606 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5607 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5608 PATH_LIMIT
5609 };
5610 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5611 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5612 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5613 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5614 record_for_igvn(result_reg);
5615
5616 Node* obj_klass = load_object_klass(obj);
5617 // We only go to the fast case code if we pass a number of guards.
5618 // The paths which do not pass are accumulated in the slow_region.
5619 RegionNode* slow_region = new RegionNode(1);
5620 record_for_igvn(slow_region);
5621
5622 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr);
5623 if (array_ctl != nullptr) {
5624 // It's an array.
5625 PreserveJVMState pjvms(this);
5626 set_control(array_ctl);
5627
5628 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5629 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr();
5630 if (UseFlatArray && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) &&
5631 obj_type->can_be_inline_array() &&
5632 (ary_ptr == nullptr || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) {
5633 // Flat inline type array may have object field that would require a
5634 // write barrier. Conservatively, go to slow path.
5635 generate_fair_guard(flat_array_test(obj_klass), slow_region);
5636 }
5637
5638 if (!stopped()) {
5639 Node* obj_length = load_array_length(obj);
5640 Node* array_size = nullptr; // Size of the array without object alignment padding.
5641 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5642
5643 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5644 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5645 // If it is an oop array, it requires very special treatment,
5646 // because gc barriers are required when accessing the array.
5647 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5648 if (is_obja != nullptr) {
5649 PreserveJVMState pjvms2(this);
5650 set_control(is_obja);
5651 // Generate a direct call to the right arraycopy function(s).
5652 // Clones are always tightly coupled.
5653 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5654 ac->set_clone_oop_array();
5655 Node* n = _gvn.transform(ac);
5656 assert(n == ac, "cannot disappear");
5657 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5658
5659 result_reg->init_req(_objArray_path, control());
5660 result_val->init_req(_objArray_path, alloc_obj);
5661 result_i_o ->set_req(_objArray_path, i_o());
5662 result_mem ->set_req(_objArray_path, reset_memory());
5663 }
5664 }
5665 // Otherwise, there are no barriers to worry about.
5666 // (We can dispense with card marks if we know the allocation
5667 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5668 // causes the non-eden paths to take compensating steps to
5669 // simulate a fresh allocation, so that no further
5670 // card marks are required in compiled code to initialize
5671 // the object.)
5672
5673 if (!stopped()) {
5674 copy_to_clone(obj, alloc_obj, array_size, true);
5675
5676 // Present the results of the copy.
5677 result_reg->init_req(_array_path, control());
5678 result_val->init_req(_array_path, alloc_obj);
5679 result_i_o ->set_req(_array_path, i_o());
5680 result_mem ->set_req(_array_path, reset_memory());
5681 }
5682 }
5683 }
5684
5685 if (!stopped()) {
5686 // It's an instance (we did array above). Make the slow-path tests.
5687 // If this is a virtual call, we generate a funny guard. We grab
5688 // the vtable entry corresponding to clone() from the target object.
5689 // If the target method which we are calling happens to be the
5690 // Object clone() method, we pass the guard. We do not need this
5691 // guard for non-virtual calls; the caller is known to be the native
5692 // Object clone().
5693 if (is_virtual) {
5694 generate_virtual_guard(obj_klass, slow_region);
5695 }
5696
5697 // The object must be easily cloneable and must not have a finalizer.
5698 // Both of these conditions may be checked in a single test.
5699 // We could optimize the test further, but we don't care.
5700 generate_misc_flags_guard(obj_klass,
5701 // Test both conditions:
5702 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5703 // Must be cloneable but not finalizer:
5704 KlassFlags::_misc_is_cloneable_fast,
5705 slow_region);
5706 }
5707
5708 if (!stopped()) {
5709 // It's an instance, and it passed the slow-path tests.
5710 PreserveJVMState pjvms(this);
5711 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5712 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5713 // is reexecuted if deoptimization occurs and there could be problems when merging
5714 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5715 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5716
5717 copy_to_clone(obj, alloc_obj, obj_size, false);
5718
5719 // Present the results of the slow call.
5720 result_reg->init_req(_instance_path, control());
5721 result_val->init_req(_instance_path, alloc_obj);
5722 result_i_o ->set_req(_instance_path, i_o());
5723 result_mem ->set_req(_instance_path, reset_memory());
5724 }
5725
5726 // Generate code for the slow case. We make a call to clone().
5727 set_control(_gvn.transform(slow_region));
5728 if (!stopped()) {
5729 PreserveJVMState pjvms(this);
5730 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5731 // We need to deoptimize on exception (see comment above)
5732 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5733 // this->control() comes from set_results_for_java_call
5734 result_reg->init_req(_slow_path, control());
5735 result_val->init_req(_slow_path, slow_result);
5736 result_i_o ->set_req(_slow_path, i_o());
5737 result_mem ->set_req(_slow_path, reset_memory());
5738 }
5739
5740 // Return the combined state.
5741 set_control( _gvn.transform(result_reg));
5742 set_i_o( _gvn.transform(result_i_o));
5743 set_all_memory( _gvn.transform(result_mem));
5744 } // original reexecute is set back here
5745
5746 set_result(_gvn.transform(result_val));
5747 return true;
5748 }
5749
5750 // If we have a tightly coupled allocation, the arraycopy may take care
5751 // of the array initialization. If one of the guards we insert between
5752 // the allocation and the arraycopy causes a deoptimization, an
5753 // uninitialized array will escape the compiled method. To prevent that
5754 // we set the JVM state for uncommon traps between the allocation and
5755 // the arraycopy to the state before the allocation so, in case of
5756 // deoptimization, we'll reexecute the allocation and the
5757 // initialization.
5758 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5759 if (alloc != nullptr) {
5760 ciMethod* trap_method = alloc->jvms()->method();
5761 int trap_bci = alloc->jvms()->bci();
5762
5763 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5764 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5765 // Make sure there's no store between the allocation and the
5766 // arraycopy otherwise visible side effects could be rexecuted
5767 // in case of deoptimization and cause incorrect execution.
5768 bool no_interfering_store = true;
5769 Node* mem = alloc->in(TypeFunc::Memory);
5770 if (mem->is_MergeMem()) {
5771 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5772 Node* n = mms.memory();
5773 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5774 assert(n->is_Store(), "what else?");
5775 no_interfering_store = false;
5776 break;
5777 }
5778 }
5779 } else {
5780 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5781 Node* n = mms.memory();
5782 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5783 assert(n->is_Store(), "what else?");
5784 no_interfering_store = false;
5785 break;
5786 }
5787 }
5788 }
5789
5790 if (no_interfering_store) {
5791 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5792
5793 JVMState* saved_jvms = jvms();
5794 saved_reexecute_sp = _reexecute_sp;
5795
5796 set_jvms(sfpt->jvms());
5797 _reexecute_sp = jvms()->sp();
5798
5799 return saved_jvms;
5800 }
5801 }
5802 }
5803 return nullptr;
5804 }
5805
5806 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5807 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5808 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5809 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5810 uint size = alloc->req();
5811 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5812 old_jvms->set_map(sfpt);
5813 for (uint i = 0; i < size; i++) {
5814 sfpt->init_req(i, alloc->in(i));
5815 }
5816 int adjustment = 1;
5817 const TypeAryKlassPtr* ary_klass_ptr = alloc->in(AllocateNode::KlassNode)->bottom_type()->is_aryklassptr();
5818 if (ary_klass_ptr->is_null_free()) {
5819 // A null-free, tightly coupled array allocation can only come from LibraryCallKit::inline_newNullRestrictedArray
5820 // which requires both the component type and the array length on stack for re-execution. Re-create and push
5821 // the component type.
5822 ciArrayKlass* klass = ary_klass_ptr->exact_klass()->as_array_klass();
5823 ciInstance* instance = klass->component_mirror_instance();
5824 const TypeInstPtr* t_instance = TypeInstPtr::make(instance);
5825 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), makecon(t_instance));
5826 adjustment++;
5827 }
5828 // re-push array length for deoptimization
5829 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment - 1, alloc->in(AllocateNode::ALength));
5830 old_jvms->set_sp(old_jvms->sp() + adjustment);
5831 old_jvms->set_monoff(old_jvms->monoff() + adjustment);
5832 old_jvms->set_scloff(old_jvms->scloff() + adjustment);
5833 old_jvms->set_endoff(old_jvms->endoff() + adjustment);
5834 old_jvms->set_should_reexecute(true);
5835
5836 sfpt->set_i_o(map()->i_o());
5837 sfpt->set_memory(map()->memory());
5838 sfpt->set_control(map()->control());
5839 return sfpt;
5840 }
5841
5842 // In case of a deoptimization, we restart execution at the
5843 // allocation, allocating a new array. We would leave an uninitialized
5844 // array in the heap that GCs wouldn't expect. Move the allocation
5845 // after the traps so we don't allocate the array if we
5846 // deoptimize. This is possible because tightly_coupled_allocation()
5847 // guarantees there's no observer of the allocated array at this point
5848 // and the control flow is simple enough.
5849 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5850 int saved_reexecute_sp, uint new_idx) {
5851 if (saved_jvms_before_guards != nullptr && !stopped()) {
5852 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5853
5854 assert(alloc != nullptr, "only with a tightly coupled allocation");
5855 // restore JVM state to the state at the arraycopy
5856 saved_jvms_before_guards->map()->set_control(map()->control());
5857 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5858 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5859 // If we've improved the types of some nodes (null check) while
5860 // emitting the guards, propagate them to the current state
5861 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5862 set_jvms(saved_jvms_before_guards);
5863 _reexecute_sp = saved_reexecute_sp;
5864
5865 // Remove the allocation from above the guards
5866 CallProjections* callprojs = alloc->extract_projections(true);
5867 InitializeNode* init = alloc->initialization();
5868 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5869 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5870 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5871
5872 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5873 // the allocation (i.e. is only valid if the allocation succeeds):
5874 // 1) replace CastIINode with AllocateArrayNode's length here
5875 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5876 //
5877 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5878 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5879 Node* init_control = init->proj_out(TypeFunc::Control);
5880 Node* alloc_length = alloc->Ideal_length();
5881 #ifdef ASSERT
5882 Node* prev_cast = nullptr;
5883 #endif
5884 for (uint i = 0; i < init_control->outcnt(); i++) {
5885 Node* init_out = init_control->raw_out(i);
5886 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5887 #ifdef ASSERT
5888 if (prev_cast == nullptr) {
5889 prev_cast = init_out;
5890 } else {
5891 if (prev_cast->cmp(*init_out) == false) {
5892 prev_cast->dump();
5893 init_out->dump();
5894 assert(false, "not equal CastIINode");
5895 }
5896 }
5897 #endif
5898 C->gvn_replace_by(init_out, alloc_length);
5899 }
5900 }
5901 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5902
5903 // move the allocation here (after the guards)
5904 _gvn.hash_delete(alloc);
5905 alloc->set_req(TypeFunc::Control, control());
5906 alloc->set_req(TypeFunc::I_O, i_o());
5907 Node *mem = reset_memory();
5908 set_all_memory(mem);
5909 alloc->set_req(TypeFunc::Memory, mem);
5910 set_control(init->proj_out_or_null(TypeFunc::Control));
5911 set_i_o(callprojs->fallthrough_ioproj);
5912
5913 // Update memory as done in GraphKit::set_output_for_allocation()
5914 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5915 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5916 if (ary_type->isa_aryptr() && length_type != nullptr) {
5917 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5918 }
5919 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5920 int elemidx = C->get_alias_index(telemref);
5921 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5922 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5923
5924 Node* allocx = _gvn.transform(alloc);
5925 assert(allocx == alloc, "where has the allocation gone?");
5926 assert(dest->is_CheckCastPP(), "not an allocation result?");
5927
5928 _gvn.hash_delete(dest);
5929 dest->set_req(0, control());
5930 Node* destx = _gvn.transform(dest);
5931 assert(destx == dest, "where has the allocation result gone?");
5932
5933 array_ideal_length(alloc, ary_type, true);
5934 }
5935 }
5936
5937 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5938 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5939 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5940 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5941 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5942 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5943 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5944 JVMState* saved_jvms_before_guards) {
5945 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5946 // There is at least one unrelated uncommon trap which needs to be replaced.
5947 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5948
5949 JVMState* saved_jvms = jvms();
5950 const int saved_reexecute_sp = _reexecute_sp;
5951 set_jvms(sfpt->jvms());
5952 _reexecute_sp = jvms()->sp();
5953
5954 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5955
5956 // Restore state
5957 set_jvms(saved_jvms);
5958 _reexecute_sp = saved_reexecute_sp;
5959 }
5960 }
5961
5962 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5963 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5964 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5965 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5966 while (if_proj->is_IfProj()) {
5967 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5968 if (uncommon_trap != nullptr) {
5969 create_new_uncommon_trap(uncommon_trap);
5970 }
5971 assert(if_proj->in(0)->is_If(), "must be If");
5972 if_proj = if_proj->in(0)->in(0);
5973 }
5974 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5975 "must have reached control projection of init node");
5976 }
5977
5978 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5979 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5980 assert(trap_request != 0, "no valid UCT trap request");
5981 PreserveJVMState pjvms(this);
5982 set_control(uncommon_trap_call->in(0));
5983 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5984 Deoptimization::trap_request_action(trap_request));
5985 assert(stopped(), "Should be stopped");
5986 _gvn.hash_delete(uncommon_trap_call);
5987 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5988 }
5989
5990 // Common checks for array sorting intrinsics arguments.
5991 // Returns `true` if checks passed.
5992 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
5993 // check address of the class
5994 if (elementType == nullptr || elementType->is_top()) {
5995 return false; // dead path
5996 }
5997 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5998 if (elem_klass == nullptr) {
5999 return false; // dead path
6000 }
6001 // java_mirror_type() returns non-null for compile-time Class constants only
6002 ciType* elem_type = elem_klass->java_mirror_type();
6003 if (elem_type == nullptr) {
6004 return false;
6005 }
6006 bt = elem_type->basic_type();
6007 // Disable the intrinsic if the CPU does not support SIMD sort
6008 if (!Matcher::supports_simd_sort(bt)) {
6009 return false;
6010 }
6011 // check address of the array
6012 if (obj == nullptr || obj->is_top()) {
6013 return false; // dead path
6014 }
6015 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
6016 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
6017 return false; // failed input validation
6018 }
6019 return true;
6020 }
6021
6022 //------------------------------inline_array_partition-----------------------
6023 bool LibraryCallKit::inline_array_partition() {
6024 address stubAddr = StubRoutines::select_array_partition_function();
6025 if (stubAddr == nullptr) {
6026 return false; // Intrinsic's stub is not implemented on this platform
6027 }
6028 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
6029
6030 // no receiver because it is a static method
6031 Node* elementType = argument(0);
6032 Node* obj = argument(1);
6033 Node* offset = argument(2); // long
6034 Node* fromIndex = argument(4);
6035 Node* toIndex = argument(5);
6036 Node* indexPivot1 = argument(6);
6037 Node* indexPivot2 = argument(7);
6038 // PartitionOperation: argument(8) is ignored
6039
6040 Node* pivotIndices = nullptr;
6041 BasicType bt = T_ILLEGAL;
6042
6043 if (!check_array_sort_arguments(elementType, obj, bt)) {
6044 return false;
6045 }
6046 null_check(obj);
6047 // If obj is dead, only null-path is taken.
6048 if (stopped()) {
6049 return true;
6050 }
6051 // Set the original stack and the reexecute bit for the interpreter to reexecute
6052 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
6053 { PreserveReexecuteState preexecs(this);
6054 jvms()->set_should_reexecute(true);
6055
6056 Node* obj_adr = make_unsafe_address(obj, offset);
6057
6058 // create the pivotIndices array of type int and size = 2
6059 Node* size = intcon(2);
6060 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
6061 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
6062 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
6063 guarantee(alloc != nullptr, "created above");
6064 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
6065
6066 // pass the basic type enum to the stub
6067 Node* elemType = intcon(bt);
6068
6069 // Call the stub
6070 const char *stubName = "array_partition_stub";
6071 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
6072 stubAddr, stubName, TypePtr::BOTTOM,
6073 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
6074 indexPivot1, indexPivot2);
6075
6076 } // original reexecute is set back here
6077
6078 if (!stopped()) {
6079 set_result(pivotIndices);
6080 }
6081
6082 return true;
6083 }
6084
6085
6086 //------------------------------inline_array_sort-----------------------
6087 bool LibraryCallKit::inline_array_sort() {
6088 address stubAddr = StubRoutines::select_arraysort_function();
6089 if (stubAddr == nullptr) {
6090 return false; // Intrinsic's stub is not implemented on this platform
6091 }
6092 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
6093
6094 // no receiver because it is a static method
6095 Node* elementType = argument(0);
6096 Node* obj = argument(1);
6097 Node* offset = argument(2); // long
6098 Node* fromIndex = argument(4);
6099 Node* toIndex = argument(5);
6100 // SortOperation: argument(6) is ignored
6101
6102 BasicType bt = T_ILLEGAL;
6103
6104 if (!check_array_sort_arguments(elementType, obj, bt)) {
6105 return false;
6106 }
6107 null_check(obj);
6108 // If obj is dead, only null-path is taken.
6109 if (stopped()) {
6110 return true;
6111 }
6112 Node* obj_adr = make_unsafe_address(obj, offset);
6113
6114 // pass the basic type enum to the stub
6115 Node* elemType = intcon(bt);
6116
6117 // Call the stub.
6118 const char *stubName = "arraysort_stub";
6119 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
6120 stubAddr, stubName, TypePtr::BOTTOM,
6121 obj_adr, elemType, fromIndex, toIndex);
6122
6123 return true;
6124 }
6125
6126
6127 //------------------------------inline_arraycopy-----------------------
6128 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
6129 // Object dest, int destPos,
6130 // int length);
6131 bool LibraryCallKit::inline_arraycopy() {
6132 // Get the arguments.
6133 Node* src = argument(0); // type: oop
6134 Node* src_offset = argument(1); // type: int
6135 Node* dest = argument(2); // type: oop
6136 Node* dest_offset = argument(3); // type: int
6137 Node* length = argument(4); // type: int
6138
6139 uint new_idx = C->unique();
6140
6141 // Check for allocation before we add nodes that would confuse
6142 // tightly_coupled_allocation()
6143 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
6144
6145 int saved_reexecute_sp = -1;
6146 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
6147 // See arraycopy_restore_alloc_state() comment
6148 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
6149 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
6150 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
6151 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
6152
6153 // The following tests must be performed
6154 // (1) src and dest are arrays.
6155 // (2) src and dest arrays must have elements of the same BasicType
6156 // (3) src and dest must not be null.
6157 // (4) src_offset must not be negative.
6158 // (5) dest_offset must not be negative.
6159 // (6) length must not be negative.
6160 // (7) src_offset + length must not exceed length of src.
6161 // (8) dest_offset + length must not exceed length of dest.
6162 // (9) each element of an oop array must be assignable
6163
6164 // (3) src and dest must not be null.
6165 // always do this here because we need the JVM state for uncommon traps
6166 Node* null_ctl = top();
6167 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
6168 assert(null_ctl->is_top(), "no null control here");
6169 dest = null_check(dest, T_ARRAY);
6170
6171 if (!can_emit_guards) {
6172 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
6173 // guards but the arraycopy node could still take advantage of a
6174 // tightly allocated allocation. tightly_coupled_allocation() is
6175 // called again to make sure it takes the null check above into
6176 // account: the null check is mandatory and if it caused an
6177 // uncommon trap to be emitted then the allocation can't be
6178 // considered tightly coupled in this context.
6179 alloc = tightly_coupled_allocation(dest);
6180 }
6181
6182 bool validated = false;
6183
6184 const Type* src_type = _gvn.type(src);
6185 const Type* dest_type = _gvn.type(dest);
6186 const TypeAryPtr* top_src = src_type->isa_aryptr();
6187 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6188
6189 // Do we have the type of src?
6190 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6191 // Do we have the type of dest?
6192 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6193 // Is the type for src from speculation?
6194 bool src_spec = false;
6195 // Is the type for dest from speculation?
6196 bool dest_spec = false;
6197
6198 if ((!has_src || !has_dest) && can_emit_guards) {
6199 // We don't have sufficient type information, let's see if
6200 // speculative types can help. We need to have types for both src
6201 // and dest so that it pays off.
6202
6203 // Do we already have or could we have type information for src
6204 bool could_have_src = has_src;
6205 // Do we already have or could we have type information for dest
6206 bool could_have_dest = has_dest;
6207
6208 ciKlass* src_k = nullptr;
6209 if (!has_src) {
6210 src_k = src_type->speculative_type_not_null();
6211 if (src_k != nullptr && src_k->is_array_klass()) {
6212 could_have_src = true;
6213 }
6214 }
6215
6216 ciKlass* dest_k = nullptr;
6217 if (!has_dest) {
6218 dest_k = dest_type->speculative_type_not_null();
6219 if (dest_k != nullptr && dest_k->is_array_klass()) {
6220 could_have_dest = true;
6221 }
6222 }
6223
6224 if (could_have_src && could_have_dest) {
6225 // This is going to pay off so emit the required guards
6226 if (!has_src) {
6227 src = maybe_cast_profiled_obj(src, src_k, true);
6228 src_type = _gvn.type(src);
6229 top_src = src_type->isa_aryptr();
6230 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6231 src_spec = true;
6232 }
6233 if (!has_dest) {
6234 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6235 dest_type = _gvn.type(dest);
6236 top_dest = dest_type->isa_aryptr();
6237 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6238 dest_spec = true;
6239 }
6240 }
6241 }
6242
6243 if (has_src && has_dest && can_emit_guards) {
6244 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
6245 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
6246 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
6247 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
6248
6249 if (src_elem == dest_elem && top_src->is_flat() == top_dest->is_flat() && src_elem == T_OBJECT) {
6250 // If both arrays are object arrays then having the exact types
6251 // for both will remove the need for a subtype check at runtime
6252 // before the call and may make it possible to pick a faster copy
6253 // routine (without a subtype check on every element)
6254 // Do we have the exact type of src?
6255 bool could_have_src = src_spec;
6256 // Do we have the exact type of dest?
6257 bool could_have_dest = dest_spec;
6258 ciKlass* src_k = nullptr;
6259 ciKlass* dest_k = nullptr;
6260 if (!src_spec) {
6261 src_k = src_type->speculative_type_not_null();
6262 if (src_k != nullptr && src_k->is_array_klass()) {
6263 could_have_src = true;
6264 }
6265 }
6266 if (!dest_spec) {
6267 dest_k = dest_type->speculative_type_not_null();
6268 if (dest_k != nullptr && dest_k->is_array_klass()) {
6269 could_have_dest = true;
6270 }
6271 }
6272 if (could_have_src && could_have_dest) {
6273 // If we can have both exact types, emit the missing guards
6274 if (could_have_src && !src_spec) {
6275 src = maybe_cast_profiled_obj(src, src_k, true);
6276 src_type = _gvn.type(src);
6277 top_src = src_type->isa_aryptr();
6278 }
6279 if (could_have_dest && !dest_spec) {
6280 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6281 dest_type = _gvn.type(dest);
6282 top_dest = dest_type->isa_aryptr();
6283 }
6284 }
6285 }
6286 }
6287
6288 ciMethod* trap_method = method();
6289 int trap_bci = bci();
6290 if (saved_jvms_before_guards != nullptr) {
6291 trap_method = alloc->jvms()->method();
6292 trap_bci = alloc->jvms()->bci();
6293 }
6294
6295 bool negative_length_guard_generated = false;
6296
6297 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6298 can_emit_guards && !src->is_top() && !dest->is_top()) {
6299 // validate arguments: enables transformation the ArrayCopyNode
6300 validated = true;
6301
6302 RegionNode* slow_region = new RegionNode(1);
6303 record_for_igvn(slow_region);
6304
6305 // (1) src and dest are arrays.
6306 generate_non_array_guard(load_object_klass(src), slow_region);
6307 generate_non_array_guard(load_object_klass(dest), slow_region);
6308
6309 // (2) src and dest arrays must have elements of the same BasicType
6310 // done at macro expansion or at Ideal transformation time
6311
6312 // (4) src_offset must not be negative.
6313 generate_negative_guard(src_offset, slow_region);
6314
6315 // (5) dest_offset must not be negative.
6316 generate_negative_guard(dest_offset, slow_region);
6317
6318 // (7) src_offset + length must not exceed length of src.
6319 generate_limit_guard(src_offset, length,
6320 load_array_length(src),
6321 slow_region);
6322
6323 // (8) dest_offset + length must not exceed length of dest.
6324 generate_limit_guard(dest_offset, length,
6325 load_array_length(dest),
6326 slow_region);
6327
6328 // (6) length must not be negative.
6329 // This is also checked in generate_arraycopy() during macro expansion, but
6330 // we also have to check it here for the case where the ArrayCopyNode will
6331 // be eliminated by Escape Analysis.
6332 if (EliminateAllocations) {
6333 generate_negative_guard(length, slow_region);
6334 negative_length_guard_generated = true;
6335 }
6336
6337 // (9) each element of an oop array must be assignable
6338 Node* dest_klass = load_object_klass(dest);
6339 if (src != dest) {
6340 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6341 slow_region->add_req(not_subtype_ctrl);
6342 }
6343
6344 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
6345 const Type* toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6346 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6347 src_type = _gvn.type(src);
6348 top_src = src_type->isa_aryptr();
6349
6350 // Handle flat inline type arrays (null-free arrays are handled by the subtype check above)
6351 if (!stopped() && UseFlatArray) {
6352 // If dest is flat, src must be flat as well (guaranteed by src <: dest check). Handle flat src here.
6353 assert(top_dest == nullptr || !top_dest->is_flat() || top_src->is_flat(), "src array must be flat");
6354 if (top_src != nullptr && top_src->is_flat()) {
6355 // Src is flat, check that dest is flat as well
6356 if (top_dest != nullptr && !top_dest->is_flat()) {
6357 generate_fair_guard(flat_array_test(dest_klass, /* flat = */ false), slow_region);
6358 // Since dest is flat and src <: dest, dest must have the same type as src.
6359 top_dest = top_src->cast_to_exactness(false);
6360 assert(top_dest->is_flat(), "dest must be flat");
6361 dest = _gvn.transform(new CheckCastPPNode(control(), dest, top_dest));
6362 }
6363 } else if (top_src == nullptr || !top_src->is_not_flat()) {
6364 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
6365 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
6366 assert(top_dest == nullptr || !top_dest->is_flat(), "dest array must not be flat");
6367 generate_fair_guard(flat_array_test(src), slow_region);
6368 if (top_src != nullptr) {
6369 top_src = top_src->cast_to_not_flat();
6370 src = _gvn.transform(new CheckCastPPNode(control(), src, top_src));
6371 }
6372 }
6373 }
6374
6375 {
6376 PreserveJVMState pjvms(this);
6377 set_control(_gvn.transform(slow_region));
6378 uncommon_trap(Deoptimization::Reason_intrinsic,
6379 Deoptimization::Action_make_not_entrant);
6380 assert(stopped(), "Should be stopped");
6381 }
6382 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6383 }
6384
6385 if (stopped()) {
6386 return true;
6387 }
6388
6389 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6390 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6391 // so the compiler has a chance to eliminate them: during macro expansion,
6392 // we have to set their control (CastPP nodes are eliminated).
6393 load_object_klass(src), load_object_klass(dest),
6394 load_array_length(src), load_array_length(dest));
6395
6396 ac->set_arraycopy(validated);
6397
6398 Node* n = _gvn.transform(ac);
6399 if (n == ac) {
6400 ac->connect_outputs(this);
6401 } else {
6402 assert(validated, "shouldn't transform if all arguments not validated");
6403 set_all_memory(n);
6404 }
6405 clear_upper_avx();
6406
6407
6408 return true;
6409 }
6410
6411
6412 // Helper function which determines if an arraycopy immediately follows
6413 // an allocation, with no intervening tests or other escapes for the object.
6414 AllocateArrayNode*
6415 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6416 if (stopped()) return nullptr; // no fast path
6417 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6418
6419 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6420 if (alloc == nullptr) return nullptr;
6421
6422 Node* rawmem = memory(Compile::AliasIdxRaw);
6423 // Is the allocation's memory state untouched?
6424 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6425 // Bail out if there have been raw-memory effects since the allocation.
6426 // (Example: There might have been a call or safepoint.)
6427 return nullptr;
6428 }
6429 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6430 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6431 return nullptr;
6432 }
6433
6434 // There must be no unexpected observers of this allocation.
6435 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6436 Node* obs = ptr->fast_out(i);
6437 if (obs != this->map()) {
6438 return nullptr;
6439 }
6440 }
6441
6442 // This arraycopy must unconditionally follow the allocation of the ptr.
6443 Node* alloc_ctl = ptr->in(0);
6444 Node* ctl = control();
6445 while (ctl != alloc_ctl) {
6446 // There may be guards which feed into the slow_region.
6447 // Any other control flow means that we might not get a chance
6448 // to finish initializing the allocated object.
6449 // Various low-level checks bottom out in uncommon traps. These
6450 // are considered safe since we've already checked above that
6451 // there is no unexpected observer of this allocation.
6452 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6453 assert(ctl->in(0)->is_If(), "must be If");
6454 ctl = ctl->in(0)->in(0);
6455 } else {
6456 return nullptr;
6457 }
6458 }
6459
6460 // If we get this far, we have an allocation which immediately
6461 // precedes the arraycopy, and we can take over zeroing the new object.
6462 // The arraycopy will finish the initialization, and provide
6463 // a new control state to which we will anchor the destination pointer.
6464
6465 return alloc;
6466 }
6467
6468 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6469 if (node->is_IfProj()) {
6470 Node* other_proj = node->as_IfProj()->other_if_proj();
6471 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6472 Node* obs = other_proj->fast_out(j);
6473 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6474 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6475 return obs->as_CallStaticJava();
6476 }
6477 }
6478 }
6479 return nullptr;
6480 }
6481
6482 //-------------inline_encodeISOArray-----------------------------------
6483 // encode char[] to byte[] in ISO_8859_1 or ASCII
6484 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6485 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6486 // no receiver since it is static method
6487 Node *src = argument(0);
6488 Node *src_offset = argument(1);
6489 Node *dst = argument(2);
6490 Node *dst_offset = argument(3);
6491 Node *length = argument(4);
6492
6493 src = must_be_not_null(src, true);
6494 dst = must_be_not_null(dst, true);
6495
6496 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6497 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6498 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6499 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6500 // failed array check
6501 return false;
6502 }
6503
6504 // Figure out the size and type of the elements we will be copying.
6505 BasicType src_elem = src_type->elem()->array_element_basic_type();
6506 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6507 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6508 return false;
6509 }
6510
6511 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6512 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6513 // 'src_start' points to src array + scaled offset
6514 // 'dst_start' points to dst array + scaled offset
6515
6516 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6517 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6518 enc = _gvn.transform(enc);
6519 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6520 set_memory(res_mem, mtype);
6521 set_result(enc);
6522 clear_upper_avx();
6523
6524 return true;
6525 }
6526
6527 //-------------inline_multiplyToLen-----------------------------------
6528 bool LibraryCallKit::inline_multiplyToLen() {
6529 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6530
6531 address stubAddr = StubRoutines::multiplyToLen();
6532 if (stubAddr == nullptr) {
6533 return false; // Intrinsic's stub is not implemented on this platform
6534 }
6535 const char* stubName = "multiplyToLen";
6536
6537 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6538
6539 // no receiver because it is a static method
6540 Node* x = argument(0);
6541 Node* xlen = argument(1);
6542 Node* y = argument(2);
6543 Node* ylen = argument(3);
6544 Node* z = argument(4);
6545
6546 x = must_be_not_null(x, true);
6547 y = must_be_not_null(y, true);
6548
6549 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6550 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6551 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6552 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6553 // failed array check
6554 return false;
6555 }
6556
6557 BasicType x_elem = x_type->elem()->array_element_basic_type();
6558 BasicType y_elem = y_type->elem()->array_element_basic_type();
6559 if (x_elem != T_INT || y_elem != T_INT) {
6560 return false;
6561 }
6562
6563 Node* x_start = array_element_address(x, intcon(0), x_elem);
6564 Node* y_start = array_element_address(y, intcon(0), y_elem);
6565 // 'x_start' points to x array + scaled xlen
6566 // 'y_start' points to y array + scaled ylen
6567
6568 Node* z_start = array_element_address(z, intcon(0), T_INT);
6569
6570 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6571 OptoRuntime::multiplyToLen_Type(),
6572 stubAddr, stubName, TypePtr::BOTTOM,
6573 x_start, xlen, y_start, ylen, z_start);
6574
6575 C->set_has_split_ifs(true); // Has chance for split-if optimization
6576 set_result(z);
6577 return true;
6578 }
6579
6580 //-------------inline_squareToLen------------------------------------
6581 bool LibraryCallKit::inline_squareToLen() {
6582 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6583
6584 address stubAddr = StubRoutines::squareToLen();
6585 if (stubAddr == nullptr) {
6586 return false; // Intrinsic's stub is not implemented on this platform
6587 }
6588 const char* stubName = "squareToLen";
6589
6590 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6591
6592 Node* x = argument(0);
6593 Node* len = argument(1);
6594 Node* z = argument(2);
6595 Node* zlen = argument(3);
6596
6597 x = must_be_not_null(x, true);
6598 z = must_be_not_null(z, true);
6599
6600 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6601 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6602 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6603 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6604 // failed array check
6605 return false;
6606 }
6607
6608 BasicType x_elem = x_type->elem()->array_element_basic_type();
6609 BasicType z_elem = z_type->elem()->array_element_basic_type();
6610 if (x_elem != T_INT || z_elem != T_INT) {
6611 return false;
6612 }
6613
6614
6615 Node* x_start = array_element_address(x, intcon(0), x_elem);
6616 Node* z_start = array_element_address(z, intcon(0), z_elem);
6617
6618 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6619 OptoRuntime::squareToLen_Type(),
6620 stubAddr, stubName, TypePtr::BOTTOM,
6621 x_start, len, z_start, zlen);
6622
6623 set_result(z);
6624 return true;
6625 }
6626
6627 //-------------inline_mulAdd------------------------------------------
6628 bool LibraryCallKit::inline_mulAdd() {
6629 assert(UseMulAddIntrinsic, "not implemented on this platform");
6630
6631 address stubAddr = StubRoutines::mulAdd();
6632 if (stubAddr == nullptr) {
6633 return false; // Intrinsic's stub is not implemented on this platform
6634 }
6635 const char* stubName = "mulAdd";
6636
6637 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6638
6639 Node* out = argument(0);
6640 Node* in = argument(1);
6641 Node* offset = argument(2);
6642 Node* len = argument(3);
6643 Node* k = argument(4);
6644
6645 in = must_be_not_null(in, true);
6646 out = must_be_not_null(out, true);
6647
6648 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6649 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6650 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6651 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6652 // failed array check
6653 return false;
6654 }
6655
6656 BasicType out_elem = out_type->elem()->array_element_basic_type();
6657 BasicType in_elem = in_type->elem()->array_element_basic_type();
6658 if (out_elem != T_INT || in_elem != T_INT) {
6659 return false;
6660 }
6661
6662 Node* outlen = load_array_length(out);
6663 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6664 Node* out_start = array_element_address(out, intcon(0), out_elem);
6665 Node* in_start = array_element_address(in, intcon(0), in_elem);
6666
6667 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6668 OptoRuntime::mulAdd_Type(),
6669 stubAddr, stubName, TypePtr::BOTTOM,
6670 out_start,in_start, new_offset, len, k);
6671 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6672 set_result(result);
6673 return true;
6674 }
6675
6676 //-------------inline_montgomeryMultiply-----------------------------------
6677 bool LibraryCallKit::inline_montgomeryMultiply() {
6678 address stubAddr = StubRoutines::montgomeryMultiply();
6679 if (stubAddr == nullptr) {
6680 return false; // Intrinsic's stub is not implemented on this platform
6681 }
6682
6683 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6684 const char* stubName = "montgomery_multiply";
6685
6686 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6687
6688 Node* a = argument(0);
6689 Node* b = argument(1);
6690 Node* n = argument(2);
6691 Node* len = argument(3);
6692 Node* inv = argument(4);
6693 Node* m = argument(6);
6694
6695 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6696 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6697 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6698 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6699 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6700 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6701 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6702 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6703 // failed array check
6704 return false;
6705 }
6706
6707 BasicType a_elem = a_type->elem()->array_element_basic_type();
6708 BasicType b_elem = b_type->elem()->array_element_basic_type();
6709 BasicType n_elem = n_type->elem()->array_element_basic_type();
6710 BasicType m_elem = m_type->elem()->array_element_basic_type();
6711 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6712 return false;
6713 }
6714
6715 // Make the call
6716 {
6717 Node* a_start = array_element_address(a, intcon(0), a_elem);
6718 Node* b_start = array_element_address(b, intcon(0), b_elem);
6719 Node* n_start = array_element_address(n, intcon(0), n_elem);
6720 Node* m_start = array_element_address(m, intcon(0), m_elem);
6721
6722 Node* call = make_runtime_call(RC_LEAF,
6723 OptoRuntime::montgomeryMultiply_Type(),
6724 stubAddr, stubName, TypePtr::BOTTOM,
6725 a_start, b_start, n_start, len, inv, top(),
6726 m_start);
6727 set_result(m);
6728 }
6729
6730 return true;
6731 }
6732
6733 bool LibraryCallKit::inline_montgomerySquare() {
6734 address stubAddr = StubRoutines::montgomerySquare();
6735 if (stubAddr == nullptr) {
6736 return false; // Intrinsic's stub is not implemented on this platform
6737 }
6738
6739 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6740 const char* stubName = "montgomery_square";
6741
6742 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6743
6744 Node* a = argument(0);
6745 Node* n = argument(1);
6746 Node* len = argument(2);
6747 Node* inv = argument(3);
6748 Node* m = argument(5);
6749
6750 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6751 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6752 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6753 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6754 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6755 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6756 // failed array check
6757 return false;
6758 }
6759
6760 BasicType a_elem = a_type->elem()->array_element_basic_type();
6761 BasicType n_elem = n_type->elem()->array_element_basic_type();
6762 BasicType m_elem = m_type->elem()->array_element_basic_type();
6763 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6764 return false;
6765 }
6766
6767 // Make the call
6768 {
6769 Node* a_start = array_element_address(a, intcon(0), a_elem);
6770 Node* n_start = array_element_address(n, intcon(0), n_elem);
6771 Node* m_start = array_element_address(m, intcon(0), m_elem);
6772
6773 Node* call = make_runtime_call(RC_LEAF,
6774 OptoRuntime::montgomerySquare_Type(),
6775 stubAddr, stubName, TypePtr::BOTTOM,
6776 a_start, n_start, len, inv, top(),
6777 m_start);
6778 set_result(m);
6779 }
6780
6781 return true;
6782 }
6783
6784 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6785 address stubAddr = nullptr;
6786 const char* stubName = nullptr;
6787
6788 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6789 if (stubAddr == nullptr) {
6790 return false; // Intrinsic's stub is not implemented on this platform
6791 }
6792
6793 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6794
6795 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6796
6797 Node* newArr = argument(0);
6798 Node* oldArr = argument(1);
6799 Node* newIdx = argument(2);
6800 Node* shiftCount = argument(3);
6801 Node* numIter = argument(4);
6802
6803 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6804 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6805 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6806 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6807 return false;
6808 }
6809
6810 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6811 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6812 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6813 return false;
6814 }
6815
6816 // Make the call
6817 {
6818 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6819 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6820
6821 Node* call = make_runtime_call(RC_LEAF,
6822 OptoRuntime::bigIntegerShift_Type(),
6823 stubAddr,
6824 stubName,
6825 TypePtr::BOTTOM,
6826 newArr_start,
6827 oldArr_start,
6828 newIdx,
6829 shiftCount,
6830 numIter);
6831 }
6832
6833 return true;
6834 }
6835
6836 //-------------inline_vectorizedMismatch------------------------------
6837 bool LibraryCallKit::inline_vectorizedMismatch() {
6838 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6839
6840 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6841 Node* obja = argument(0); // Object
6842 Node* aoffset = argument(1); // long
6843 Node* objb = argument(3); // Object
6844 Node* boffset = argument(4); // long
6845 Node* length = argument(6); // int
6846 Node* scale = argument(7); // int
6847
6848 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6849 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6850 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6851 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6852 scale == top()) {
6853 return false; // failed input validation
6854 }
6855
6856 Node* obja_adr = make_unsafe_address(obja, aoffset);
6857 Node* objb_adr = make_unsafe_address(objb, boffset);
6858
6859 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6860 //
6861 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6862 // if (length <= inline_limit) {
6863 // inline_path:
6864 // vmask = VectorMaskGen length
6865 // vload1 = LoadVectorMasked obja, vmask
6866 // vload2 = LoadVectorMasked objb, vmask
6867 // result1 = VectorCmpMasked vload1, vload2, vmask
6868 // } else {
6869 // call_stub_path:
6870 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6871 // }
6872 // exit_block:
6873 // return Phi(result1, result2);
6874 //
6875 enum { inline_path = 1, // input is small enough to process it all at once
6876 stub_path = 2, // input is too large; call into the VM
6877 PATH_LIMIT = 3
6878 };
6879
6880 Node* exit_block = new RegionNode(PATH_LIMIT);
6881 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6882 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6883
6884 Node* call_stub_path = control();
6885
6886 BasicType elem_bt = T_ILLEGAL;
6887
6888 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6889 if (scale_t->is_con()) {
6890 switch (scale_t->get_con()) {
6891 case 0: elem_bt = T_BYTE; break;
6892 case 1: elem_bt = T_SHORT; break;
6893 case 2: elem_bt = T_INT; break;
6894 case 3: elem_bt = T_LONG; break;
6895
6896 default: elem_bt = T_ILLEGAL; break; // not supported
6897 }
6898 }
6899
6900 int inline_limit = 0;
6901 bool do_partial_inline = false;
6902
6903 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6904 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6905 do_partial_inline = inline_limit >= 16;
6906 }
6907
6908 if (do_partial_inline) {
6909 assert(elem_bt != T_ILLEGAL, "sanity");
6910
6911 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6912 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6913 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6914
6915 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6916 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6917 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6918
6919 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6920
6921 if (!stopped()) {
6922 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6923
6924 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6925 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6926 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6927 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6928
6929 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6930 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6931 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6932 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6933
6934 exit_block->init_req(inline_path, control());
6935 memory_phi->init_req(inline_path, map()->memory());
6936 result_phi->init_req(inline_path, result);
6937
6938 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6939 clear_upper_avx();
6940 }
6941 }
6942 }
6943
6944 if (call_stub_path != nullptr) {
6945 set_control(call_stub_path);
6946
6947 Node* call = make_runtime_call(RC_LEAF,
6948 OptoRuntime::vectorizedMismatch_Type(),
6949 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6950 obja_adr, objb_adr, length, scale);
6951
6952 exit_block->init_req(stub_path, control());
6953 memory_phi->init_req(stub_path, map()->memory());
6954 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6955 }
6956
6957 exit_block = _gvn.transform(exit_block);
6958 memory_phi = _gvn.transform(memory_phi);
6959 result_phi = _gvn.transform(result_phi);
6960
6961 set_control(exit_block);
6962 set_all_memory(memory_phi);
6963 set_result(result_phi);
6964
6965 return true;
6966 }
6967
6968 //------------------------------inline_vectorizedHashcode----------------------------
6969 bool LibraryCallKit::inline_vectorizedHashCode() {
6970 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6971
6972 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6973 Node* array = argument(0);
6974 Node* offset = argument(1);
6975 Node* length = argument(2);
6976 Node* initialValue = argument(3);
6977 Node* basic_type = argument(4);
6978
6979 if (basic_type == top()) {
6980 return false; // failed input validation
6981 }
6982
6983 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6984 if (!basic_type_t->is_con()) {
6985 return false; // Only intrinsify if mode argument is constant
6986 }
6987
6988 array = must_be_not_null(array, true);
6989
6990 BasicType bt = (BasicType)basic_type_t->get_con();
6991
6992 // Resolve address of first element
6993 Node* array_start = array_element_address(array, offset, bt);
6994
6995 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6996 array_start, length, initialValue, basic_type)));
6997 clear_upper_avx();
6998
6999 return true;
7000 }
7001
7002 /**
7003 * Calculate CRC32 for byte.
7004 * int java.util.zip.CRC32.update(int crc, int b)
7005 */
7006 bool LibraryCallKit::inline_updateCRC32() {
7007 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
7008 assert(callee()->signature()->size() == 2, "update has 2 parameters");
7009 // no receiver since it is static method
7010 Node* crc = argument(0); // type: int
7011 Node* b = argument(1); // type: int
7012
7013 /*
7014 * int c = ~ crc;
7015 * b = timesXtoThe32[(b ^ c) & 0xFF];
7016 * b = b ^ (c >>> 8);
7017 * crc = ~b;
7018 */
7019
7020 Node* M1 = intcon(-1);
7021 crc = _gvn.transform(new XorINode(crc, M1));
7022 Node* result = _gvn.transform(new XorINode(crc, b));
7023 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
7024
7025 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
7026 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
7027 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
7028 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
7029
7030 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
7031 result = _gvn.transform(new XorINode(crc, result));
7032 result = _gvn.transform(new XorINode(result, M1));
7033 set_result(result);
7034 return true;
7035 }
7036
7037 /**
7038 * Calculate CRC32 for byte[] array.
7039 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
7040 */
7041 bool LibraryCallKit::inline_updateBytesCRC32() {
7042 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
7043 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7044 // no receiver since it is static method
7045 Node* crc = argument(0); // type: int
7046 Node* src = argument(1); // type: oop
7047 Node* offset = argument(2); // type: int
7048 Node* length = argument(3); // type: int
7049
7050 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7051 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7052 // failed array check
7053 return false;
7054 }
7055
7056 // Figure out the size and type of the elements we will be copying.
7057 BasicType src_elem = src_type->elem()->array_element_basic_type();
7058 if (src_elem != T_BYTE) {
7059 return false;
7060 }
7061
7062 // 'src_start' points to src array + scaled offset
7063 src = must_be_not_null(src, true);
7064 Node* src_start = array_element_address(src, offset, src_elem);
7065
7066 // We assume that range check is done by caller.
7067 // TODO: generate range check (offset+length < src.length) in debug VM.
7068
7069 // Call the stub.
7070 address stubAddr = StubRoutines::updateBytesCRC32();
7071 const char *stubName = "updateBytesCRC32";
7072
7073 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7074 stubAddr, stubName, TypePtr::BOTTOM,
7075 crc, src_start, length);
7076 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7077 set_result(result);
7078 return true;
7079 }
7080
7081 /**
7082 * Calculate CRC32 for ByteBuffer.
7083 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
7084 */
7085 bool LibraryCallKit::inline_updateByteBufferCRC32() {
7086 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
7087 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7088 // no receiver since it is static method
7089 Node* crc = argument(0); // type: int
7090 Node* src = argument(1); // type: long
7091 Node* offset = argument(3); // type: int
7092 Node* length = argument(4); // type: int
7093
7094 src = ConvL2X(src); // adjust Java long to machine word
7095 Node* base = _gvn.transform(new CastX2PNode(src));
7096 offset = ConvI2X(offset);
7097
7098 // 'src_start' points to src array + scaled offset
7099 Node* src_start = basic_plus_adr(top(), base, offset);
7100
7101 // Call the stub.
7102 address stubAddr = StubRoutines::updateBytesCRC32();
7103 const char *stubName = "updateBytesCRC32";
7104
7105 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7106 stubAddr, stubName, TypePtr::BOTTOM,
7107 crc, src_start, length);
7108 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7109 set_result(result);
7110 return true;
7111 }
7112
7113 //------------------------------get_table_from_crc32c_class-----------------------
7114 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
7115 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
7116 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
7117
7118 return table;
7119 }
7120
7121 //------------------------------inline_updateBytesCRC32C-----------------------
7122 //
7123 // Calculate CRC32C for byte[] array.
7124 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
7125 //
7126 bool LibraryCallKit::inline_updateBytesCRC32C() {
7127 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7128 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7129 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7130 // no receiver since it is a static method
7131 Node* crc = argument(0); // type: int
7132 Node* src = argument(1); // type: oop
7133 Node* offset = argument(2); // type: int
7134 Node* end = argument(3); // type: int
7135
7136 Node* length = _gvn.transform(new SubINode(end, offset));
7137
7138 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7139 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7140 // failed array check
7141 return false;
7142 }
7143
7144 // Figure out the size and type of the elements we will be copying.
7145 BasicType src_elem = src_type->elem()->array_element_basic_type();
7146 if (src_elem != T_BYTE) {
7147 return false;
7148 }
7149
7150 // 'src_start' points to src array + scaled offset
7151 src = must_be_not_null(src, true);
7152 Node* src_start = array_element_address(src, offset, src_elem);
7153
7154 // static final int[] byteTable in class CRC32C
7155 Node* table = get_table_from_crc32c_class(callee()->holder());
7156 table = must_be_not_null(table, true);
7157 Node* table_start = array_element_address(table, intcon(0), T_INT);
7158
7159 // We assume that range check is done by caller.
7160 // TODO: generate range check (offset+length < src.length) in debug VM.
7161
7162 // Call the stub.
7163 address stubAddr = StubRoutines::updateBytesCRC32C();
7164 const char *stubName = "updateBytesCRC32C";
7165
7166 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7167 stubAddr, stubName, TypePtr::BOTTOM,
7168 crc, src_start, length, table_start);
7169 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7170 set_result(result);
7171 return true;
7172 }
7173
7174 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
7175 //
7176 // Calculate CRC32C for DirectByteBuffer.
7177 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
7178 //
7179 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
7180 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7181 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
7182 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7183 // no receiver since it is a static method
7184 Node* crc = argument(0); // type: int
7185 Node* src = argument(1); // type: long
7186 Node* offset = argument(3); // type: int
7187 Node* end = argument(4); // type: int
7188
7189 Node* length = _gvn.transform(new SubINode(end, offset));
7190
7191 src = ConvL2X(src); // adjust Java long to machine word
7192 Node* base = _gvn.transform(new CastX2PNode(src));
7193 offset = ConvI2X(offset);
7194
7195 // 'src_start' points to src array + scaled offset
7196 Node* src_start = basic_plus_adr(top(), base, offset);
7197
7198 // static final int[] byteTable in class CRC32C
7199 Node* table = get_table_from_crc32c_class(callee()->holder());
7200 table = must_be_not_null(table, true);
7201 Node* table_start = array_element_address(table, intcon(0), T_INT);
7202
7203 // Call the stub.
7204 address stubAddr = StubRoutines::updateBytesCRC32C();
7205 const char *stubName = "updateBytesCRC32C";
7206
7207 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7208 stubAddr, stubName, TypePtr::BOTTOM,
7209 crc, src_start, length, table_start);
7210 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7211 set_result(result);
7212 return true;
7213 }
7214
7215 //------------------------------inline_updateBytesAdler32----------------------
7216 //
7217 // Calculate Adler32 checksum for byte[] array.
7218 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
7219 //
7220 bool LibraryCallKit::inline_updateBytesAdler32() {
7221 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7222 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7223 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7224 // no receiver since it is static method
7225 Node* crc = argument(0); // type: int
7226 Node* src = argument(1); // type: oop
7227 Node* offset = argument(2); // type: int
7228 Node* length = argument(3); // type: int
7229
7230 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7231 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7232 // failed array check
7233 return false;
7234 }
7235
7236 // Figure out the size and type of the elements we will be copying.
7237 BasicType src_elem = src_type->elem()->array_element_basic_type();
7238 if (src_elem != T_BYTE) {
7239 return false;
7240 }
7241
7242 // 'src_start' points to src array + scaled offset
7243 Node* src_start = array_element_address(src, offset, src_elem);
7244
7245 // We assume that range check is done by caller.
7246 // TODO: generate range check (offset+length < src.length) in debug VM.
7247
7248 // Call the stub.
7249 address stubAddr = StubRoutines::updateBytesAdler32();
7250 const char *stubName = "updateBytesAdler32";
7251
7252 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7253 stubAddr, stubName, TypePtr::BOTTOM,
7254 crc, src_start, length);
7255 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7256 set_result(result);
7257 return true;
7258 }
7259
7260 //------------------------------inline_updateByteBufferAdler32---------------
7261 //
7262 // Calculate Adler32 checksum for DirectByteBuffer.
7263 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
7264 //
7265 bool LibraryCallKit::inline_updateByteBufferAdler32() {
7266 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7267 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7268 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7269 // no receiver since it is static method
7270 Node* crc = argument(0); // type: int
7271 Node* src = argument(1); // type: long
7272 Node* offset = argument(3); // type: int
7273 Node* length = argument(4); // type: int
7274
7275 src = ConvL2X(src); // adjust Java long to machine word
7276 Node* base = _gvn.transform(new CastX2PNode(src));
7277 offset = ConvI2X(offset);
7278
7279 // 'src_start' points to src array + scaled offset
7280 Node* src_start = basic_plus_adr(top(), base, offset);
7281
7282 // Call the stub.
7283 address stubAddr = StubRoutines::updateBytesAdler32();
7284 const char *stubName = "updateBytesAdler32";
7285
7286 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7287 stubAddr, stubName, TypePtr::BOTTOM,
7288 crc, src_start, length);
7289
7290 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7291 set_result(result);
7292 return true;
7293 }
7294
7295 //----------------------------inline_reference_get----------------------------
7296 // public T java.lang.ref.Reference.get();
7297 bool LibraryCallKit::inline_reference_get() {
7298 const int referent_offset = java_lang_ref_Reference::referent_offset();
7299
7300 // Get the argument:
7301 Node* reference_obj = null_check_receiver();
7302 if (stopped()) return true;
7303
7304 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7305 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7306 decorators, /*is_static*/ false, nullptr);
7307 if (result == nullptr) return false;
7308
7309 // Add memory barrier to prevent commoning reads from this field
7310 // across safepoint since GC can change its value.
7311 insert_mem_bar(Op_MemBarCPUOrder);
7312
7313 set_result(result);
7314 return true;
7315 }
7316
7317 //----------------------------inline_reference_refersTo0----------------------------
7318 // bool java.lang.ref.Reference.refersTo0();
7319 // bool java.lang.ref.PhantomReference.refersTo0();
7320 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7321 // Get arguments:
7322 Node* reference_obj = null_check_receiver();
7323 Node* other_obj = argument(1);
7324 if (stopped()) return true;
7325
7326 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7327 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7328 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7329 decorators, /*is_static*/ false, nullptr);
7330 if (referent == nullptr) return false;
7331
7332 // Add memory barrier to prevent commoning reads from this field
7333 // across safepoint since GC can change its value.
7334 insert_mem_bar(Op_MemBarCPUOrder);
7335
7336 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7337 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7338 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7339
7340 RegionNode* region = new RegionNode(3);
7341 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7342
7343 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7344 region->init_req(1, if_true);
7345 phi->init_req(1, intcon(1));
7346
7347 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7348 region->init_req(2, if_false);
7349 phi->init_req(2, intcon(0));
7350
7351 set_control(_gvn.transform(region));
7352 record_for_igvn(region);
7353 set_result(_gvn.transform(phi));
7354 return true;
7355 }
7356
7357 //----------------------------inline_reference_clear0----------------------------
7358 // void java.lang.ref.Reference.clear0();
7359 // void java.lang.ref.PhantomReference.clear0();
7360 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7361 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7362
7363 // Get arguments
7364 Node* reference_obj = null_check_receiver();
7365 if (stopped()) return true;
7366
7367 // Common access parameters
7368 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7369 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7370 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7371 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7372 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7373
7374 Node* referent = access_load_at(reference_obj,
7375 referent_field_addr,
7376 referent_field_addr_type,
7377 val_type,
7378 T_OBJECT,
7379 decorators);
7380
7381 IdealKit ideal(this);
7382 #define __ ideal.
7383 __ if_then(referent, BoolTest::ne, null());
7384 sync_kit(ideal);
7385 access_store_at(reference_obj,
7386 referent_field_addr,
7387 referent_field_addr_type,
7388 null(),
7389 val_type,
7390 T_OBJECT,
7391 decorators);
7392 __ sync_kit(this);
7393 __ end_if();
7394 final_sync(ideal);
7395 #undef __
7396
7397 return true;
7398 }
7399
7400 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7401 DecoratorSet decorators, bool is_static,
7402 ciInstanceKlass* fromKls) {
7403 if (fromKls == nullptr) {
7404 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7405 assert(tinst != nullptr, "obj is null");
7406 assert(tinst->is_loaded(), "obj is not loaded");
7407 fromKls = tinst->instance_klass();
7408 } else {
7409 assert(is_static, "only for static field access");
7410 }
7411 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7412 ciSymbol::make(fieldTypeString),
7413 is_static);
7414
7415 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7416 if (field == nullptr) return (Node *) nullptr;
7417
7418 if (is_static) {
7419 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7420 fromObj = makecon(tip);
7421 }
7422
7423 // Next code copied from Parse::do_get_xxx():
7424
7425 // Compute address and memory type.
7426 int offset = field->offset_in_bytes();
7427 bool is_vol = field->is_volatile();
7428 ciType* field_klass = field->type();
7429 assert(field_klass->is_loaded(), "should be loaded");
7430 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7431 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7432 BasicType bt = field->layout_type();
7433
7434 // Build the resultant type of the load
7435 const Type *type;
7436 if (bt == T_OBJECT) {
7437 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7438 } else {
7439 type = Type::get_const_basic_type(bt);
7440 }
7441
7442 if (is_vol) {
7443 decorators |= MO_SEQ_CST;
7444 }
7445
7446 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7447 }
7448
7449 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7450 bool is_exact /* true */, bool is_static /* false */,
7451 ciInstanceKlass * fromKls /* nullptr */) {
7452 if (fromKls == nullptr) {
7453 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7454 assert(tinst != nullptr, "obj is null");
7455 assert(tinst->is_loaded(), "obj is not loaded");
7456 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7457 fromKls = tinst->instance_klass();
7458 }
7459 else {
7460 assert(is_static, "only for static field access");
7461 }
7462 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7463 ciSymbol::make(fieldTypeString),
7464 is_static);
7465
7466 assert(field != nullptr, "undefined field");
7467 assert(!field->is_volatile(), "not defined for volatile fields");
7468
7469 if (is_static) {
7470 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7471 fromObj = makecon(tip);
7472 }
7473
7474 // Next code copied from Parse::do_get_xxx():
7475
7476 // Compute address and memory type.
7477 int offset = field->offset_in_bytes();
7478 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7479
7480 return adr;
7481 }
7482
7483 //------------------------------inline_aescrypt_Block-----------------------
7484 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7485 address stubAddr = nullptr;
7486 const char *stubName;
7487 assert(UseAES, "need AES instruction support");
7488
7489 switch(id) {
7490 case vmIntrinsics::_aescrypt_encryptBlock:
7491 stubAddr = StubRoutines::aescrypt_encryptBlock();
7492 stubName = "aescrypt_encryptBlock";
7493 break;
7494 case vmIntrinsics::_aescrypt_decryptBlock:
7495 stubAddr = StubRoutines::aescrypt_decryptBlock();
7496 stubName = "aescrypt_decryptBlock";
7497 break;
7498 default:
7499 break;
7500 }
7501 if (stubAddr == nullptr) return false;
7502
7503 Node* aescrypt_object = argument(0);
7504 Node* src = argument(1);
7505 Node* src_offset = argument(2);
7506 Node* dest = argument(3);
7507 Node* dest_offset = argument(4);
7508
7509 src = must_be_not_null(src, true);
7510 dest = must_be_not_null(dest, true);
7511
7512 // (1) src and dest are arrays.
7513 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7514 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7515 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7516 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7517
7518 // for the quick and dirty code we will skip all the checks.
7519 // we are just trying to get the call to be generated.
7520 Node* src_start = src;
7521 Node* dest_start = dest;
7522 if (src_offset != nullptr || dest_offset != nullptr) {
7523 assert(src_offset != nullptr && dest_offset != nullptr, "");
7524 src_start = array_element_address(src, src_offset, T_BYTE);
7525 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7526 }
7527
7528 // now need to get the start of its expanded key array
7529 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7530 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7531 if (k_start == nullptr) return false;
7532
7533 // Call the stub.
7534 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7535 stubAddr, stubName, TypePtr::BOTTOM,
7536 src_start, dest_start, k_start);
7537
7538 return true;
7539 }
7540
7541 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7542 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7543 address stubAddr = nullptr;
7544 const char *stubName = nullptr;
7545
7546 assert(UseAES, "need AES instruction support");
7547
7548 switch(id) {
7549 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7550 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7551 stubName = "cipherBlockChaining_encryptAESCrypt";
7552 break;
7553 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7554 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7555 stubName = "cipherBlockChaining_decryptAESCrypt";
7556 break;
7557 default:
7558 break;
7559 }
7560 if (stubAddr == nullptr) return false;
7561
7562 Node* cipherBlockChaining_object = argument(0);
7563 Node* src = argument(1);
7564 Node* src_offset = argument(2);
7565 Node* len = argument(3);
7566 Node* dest = argument(4);
7567 Node* dest_offset = argument(5);
7568
7569 src = must_be_not_null(src, false);
7570 dest = must_be_not_null(dest, false);
7571
7572 // (1) src and dest are arrays.
7573 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7574 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7575 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7576 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7577
7578 // checks are the responsibility of the caller
7579 Node* src_start = src;
7580 Node* dest_start = dest;
7581 if (src_offset != nullptr || dest_offset != nullptr) {
7582 assert(src_offset != nullptr && dest_offset != nullptr, "");
7583 src_start = array_element_address(src, src_offset, T_BYTE);
7584 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7585 }
7586
7587 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7588 // (because of the predicated logic executed earlier).
7589 // so we cast it here safely.
7590 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7591
7592 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7593 if (embeddedCipherObj == nullptr) return false;
7594
7595 // cast it to what we know it will be at runtime
7596 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7597 assert(tinst != nullptr, "CBC obj is null");
7598 assert(tinst->is_loaded(), "CBC obj is not loaded");
7599 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7600 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7601
7602 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7603 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7604 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7605 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7606 aescrypt_object = _gvn.transform(aescrypt_object);
7607
7608 // we need to get the start of the aescrypt_object's expanded key array
7609 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7610 if (k_start == nullptr) return false;
7611
7612 // similarly, get the start address of the r vector
7613 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7614 if (objRvec == nullptr) return false;
7615 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7616
7617 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7618 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7619 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7620 stubAddr, stubName, TypePtr::BOTTOM,
7621 src_start, dest_start, k_start, r_start, len);
7622
7623 // return cipher length (int)
7624 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7625 set_result(retvalue);
7626 return true;
7627 }
7628
7629 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7630 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7631 address stubAddr = nullptr;
7632 const char *stubName = nullptr;
7633
7634 assert(UseAES, "need AES instruction support");
7635
7636 switch (id) {
7637 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7638 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7639 stubName = "electronicCodeBook_encryptAESCrypt";
7640 break;
7641 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7642 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7643 stubName = "electronicCodeBook_decryptAESCrypt";
7644 break;
7645 default:
7646 break;
7647 }
7648
7649 if (stubAddr == nullptr) return false;
7650
7651 Node* electronicCodeBook_object = argument(0);
7652 Node* src = argument(1);
7653 Node* src_offset = argument(2);
7654 Node* len = argument(3);
7655 Node* dest = argument(4);
7656 Node* dest_offset = argument(5);
7657
7658 // (1) src and dest are arrays.
7659 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7660 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7661 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7662 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7663
7664 // checks are the responsibility of the caller
7665 Node* src_start = src;
7666 Node* dest_start = dest;
7667 if (src_offset != nullptr || dest_offset != nullptr) {
7668 assert(src_offset != nullptr && dest_offset != nullptr, "");
7669 src_start = array_element_address(src, src_offset, T_BYTE);
7670 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7671 }
7672
7673 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7674 // (because of the predicated logic executed earlier).
7675 // so we cast it here safely.
7676 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7677
7678 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7679 if (embeddedCipherObj == nullptr) return false;
7680
7681 // cast it to what we know it will be at runtime
7682 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7683 assert(tinst != nullptr, "ECB obj is null");
7684 assert(tinst->is_loaded(), "ECB obj is not loaded");
7685 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7686 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7687
7688 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7689 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7690 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7691 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7692 aescrypt_object = _gvn.transform(aescrypt_object);
7693
7694 // we need to get the start of the aescrypt_object's expanded key array
7695 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7696 if (k_start == nullptr) return false;
7697
7698 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7699 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7700 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7701 stubAddr, stubName, TypePtr::BOTTOM,
7702 src_start, dest_start, k_start, len);
7703
7704 // return cipher length (int)
7705 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7706 set_result(retvalue);
7707 return true;
7708 }
7709
7710 //------------------------------inline_counterMode_AESCrypt-----------------------
7711 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7712 assert(UseAES, "need AES instruction support");
7713 if (!UseAESCTRIntrinsics) return false;
7714
7715 address stubAddr = nullptr;
7716 const char *stubName = nullptr;
7717 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7718 stubAddr = StubRoutines::counterMode_AESCrypt();
7719 stubName = "counterMode_AESCrypt";
7720 }
7721 if (stubAddr == nullptr) return false;
7722
7723 Node* counterMode_object = argument(0);
7724 Node* src = argument(1);
7725 Node* src_offset = argument(2);
7726 Node* len = argument(3);
7727 Node* dest = argument(4);
7728 Node* dest_offset = argument(5);
7729
7730 // (1) src and dest are arrays.
7731 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7732 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7733 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7734 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7735
7736 // checks are the responsibility of the caller
7737 Node* src_start = src;
7738 Node* dest_start = dest;
7739 if (src_offset != nullptr || dest_offset != nullptr) {
7740 assert(src_offset != nullptr && dest_offset != nullptr, "");
7741 src_start = array_element_address(src, src_offset, T_BYTE);
7742 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7743 }
7744
7745 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7746 // (because of the predicated logic executed earlier).
7747 // so we cast it here safely.
7748 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7749 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7750 if (embeddedCipherObj == nullptr) return false;
7751 // cast it to what we know it will be at runtime
7752 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7753 assert(tinst != nullptr, "CTR obj is null");
7754 assert(tinst->is_loaded(), "CTR obj is not loaded");
7755 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7756 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7757 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7758 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7759 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7760 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7761 aescrypt_object = _gvn.transform(aescrypt_object);
7762 // we need to get the start of the aescrypt_object's expanded key array
7763 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7764 if (k_start == nullptr) return false;
7765 // similarly, get the start address of the r vector
7766 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7767 if (obj_counter == nullptr) return false;
7768 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7769
7770 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7771 if (saved_encCounter == nullptr) return false;
7772 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7773 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7774
7775 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7776 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7777 OptoRuntime::counterMode_aescrypt_Type(),
7778 stubAddr, stubName, TypePtr::BOTTOM,
7779 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7780
7781 // return cipher length (int)
7782 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7783 set_result(retvalue);
7784 return true;
7785 }
7786
7787 //------------------------------get_key_start_from_aescrypt_object-----------------------
7788 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7789 #if defined(PPC64) || defined(S390)
7790 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7791 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7792 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7793 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7794 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7795 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7796 if (objSessionK == nullptr) {
7797 return (Node *) nullptr;
7798 }
7799 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7800 #else
7801 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7802 #endif // PPC64
7803 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7804 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7805
7806 // now have the array, need to get the start address of the K array
7807 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7808 return k_start;
7809 }
7810
7811 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7812 // Return node representing slow path of predicate check.
7813 // the pseudo code we want to emulate with this predicate is:
7814 // for encryption:
7815 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7816 // for decryption:
7817 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7818 // note cipher==plain is more conservative than the original java code but that's OK
7819 //
7820 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7821 // The receiver was checked for null already.
7822 Node* objCBC = argument(0);
7823
7824 Node* src = argument(1);
7825 Node* dest = argument(4);
7826
7827 // Load embeddedCipher field of CipherBlockChaining object.
7828 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7829
7830 // get AESCrypt klass for instanceOf check
7831 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7832 // will have same classloader as CipherBlockChaining object
7833 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7834 assert(tinst != nullptr, "CBCobj is null");
7835 assert(tinst->is_loaded(), "CBCobj is not loaded");
7836
7837 // we want to do an instanceof comparison against the AESCrypt class
7838 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7839 if (!klass_AESCrypt->is_loaded()) {
7840 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7841 Node* ctrl = control();
7842 set_control(top()); // no regular fast path
7843 return ctrl;
7844 }
7845
7846 src = must_be_not_null(src, true);
7847 dest = must_be_not_null(dest, true);
7848
7849 // Resolve oops to stable for CmpP below.
7850 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7851
7852 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7853 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7854 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7855
7856 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7857
7858 // for encryption, we are done
7859 if (!decrypting)
7860 return instof_false; // even if it is null
7861
7862 // for decryption, we need to add a further check to avoid
7863 // taking the intrinsic path when cipher and plain are the same
7864 // see the original java code for why.
7865 RegionNode* region = new RegionNode(3);
7866 region->init_req(1, instof_false);
7867
7868 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7869 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7870 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7871 region->init_req(2, src_dest_conjoint);
7872
7873 record_for_igvn(region);
7874 return _gvn.transform(region);
7875 }
7876
7877 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7878 // Return node representing slow path of predicate check.
7879 // the pseudo code we want to emulate with this predicate is:
7880 // for encryption:
7881 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7882 // for decryption:
7883 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7884 // note cipher==plain is more conservative than the original java code but that's OK
7885 //
7886 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7887 // The receiver was checked for null already.
7888 Node* objECB = argument(0);
7889
7890 // Load embeddedCipher field of ElectronicCodeBook object.
7891 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7892
7893 // get AESCrypt klass for instanceOf check
7894 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7895 // will have same classloader as ElectronicCodeBook object
7896 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7897 assert(tinst != nullptr, "ECBobj is null");
7898 assert(tinst->is_loaded(), "ECBobj is not loaded");
7899
7900 // we want to do an instanceof comparison against the AESCrypt class
7901 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7902 if (!klass_AESCrypt->is_loaded()) {
7903 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7904 Node* ctrl = control();
7905 set_control(top()); // no regular fast path
7906 return ctrl;
7907 }
7908 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7909
7910 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7911 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7912 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7913
7914 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7915
7916 // for encryption, we are done
7917 if (!decrypting)
7918 return instof_false; // even if it is null
7919
7920 // for decryption, we need to add a further check to avoid
7921 // taking the intrinsic path when cipher and plain are the same
7922 // see the original java code for why.
7923 RegionNode* region = new RegionNode(3);
7924 region->init_req(1, instof_false);
7925 Node* src = argument(1);
7926 Node* dest = argument(4);
7927 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7928 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7929 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7930 region->init_req(2, src_dest_conjoint);
7931
7932 record_for_igvn(region);
7933 return _gvn.transform(region);
7934 }
7935
7936 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7937 // Return node representing slow path of predicate check.
7938 // the pseudo code we want to emulate with this predicate is:
7939 // for encryption:
7940 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7941 // for decryption:
7942 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7943 // note cipher==plain is more conservative than the original java code but that's OK
7944 //
7945
7946 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7947 // The receiver was checked for null already.
7948 Node* objCTR = argument(0);
7949
7950 // Load embeddedCipher field of CipherBlockChaining object.
7951 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7952
7953 // get AESCrypt klass for instanceOf check
7954 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7955 // will have same classloader as CipherBlockChaining object
7956 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7957 assert(tinst != nullptr, "CTRobj is null");
7958 assert(tinst->is_loaded(), "CTRobj is not loaded");
7959
7960 // we want to do an instanceof comparison against the AESCrypt class
7961 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7962 if (!klass_AESCrypt->is_loaded()) {
7963 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7964 Node* ctrl = control();
7965 set_control(top()); // no regular fast path
7966 return ctrl;
7967 }
7968
7969 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7970 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7971 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7972 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7973 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7974
7975 return instof_false; // even if it is null
7976 }
7977
7978 //------------------------------inline_ghash_processBlocks
7979 bool LibraryCallKit::inline_ghash_processBlocks() {
7980 address stubAddr;
7981 const char *stubName;
7982 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7983
7984 stubAddr = StubRoutines::ghash_processBlocks();
7985 stubName = "ghash_processBlocks";
7986
7987 Node* data = argument(0);
7988 Node* offset = argument(1);
7989 Node* len = argument(2);
7990 Node* state = argument(3);
7991 Node* subkeyH = argument(4);
7992
7993 state = must_be_not_null(state, true);
7994 subkeyH = must_be_not_null(subkeyH, true);
7995 data = must_be_not_null(data, true);
7996
7997 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7998 assert(state_start, "state is null");
7999 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
8000 assert(subkeyH_start, "subkeyH is null");
8001 Node* data_start = array_element_address(data, offset, T_BYTE);
8002 assert(data_start, "data is null");
8003
8004 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
8005 OptoRuntime::ghash_processBlocks_Type(),
8006 stubAddr, stubName, TypePtr::BOTTOM,
8007 state_start, subkeyH_start, data_start, len);
8008 return true;
8009 }
8010
8011 //------------------------------inline_chacha20Block
8012 bool LibraryCallKit::inline_chacha20Block() {
8013 address stubAddr;
8014 const char *stubName;
8015 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
8016
8017 stubAddr = StubRoutines::chacha20Block();
8018 stubName = "chacha20Block";
8019
8020 Node* state = argument(0);
8021 Node* result = argument(1);
8022
8023 state = must_be_not_null(state, true);
8024 result = must_be_not_null(result, true);
8025
8026 Node* state_start = array_element_address(state, intcon(0), T_INT);
8027 assert(state_start, "state is null");
8028 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
8029 assert(result_start, "result is null");
8030
8031 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
8032 OptoRuntime::chacha20Block_Type(),
8033 stubAddr, stubName, TypePtr::BOTTOM,
8034 state_start, result_start);
8035 // return key stream length (int)
8036 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
8037 set_result(retvalue);
8038 return true;
8039 }
8040
8041 bool LibraryCallKit::inline_base64_encodeBlock() {
8042 address stubAddr;
8043 const char *stubName;
8044 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8045 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8046 stubAddr = StubRoutines::base64_encodeBlock();
8047 stubName = "encodeBlock";
8048
8049 if (!stubAddr) return false;
8050 Node* base64obj = argument(0);
8051 Node* src = argument(1);
8052 Node* offset = argument(2);
8053 Node* len = argument(3);
8054 Node* dest = argument(4);
8055 Node* dp = argument(5);
8056 Node* isURL = argument(6);
8057
8058 src = must_be_not_null(src, true);
8059 dest = must_be_not_null(dest, true);
8060
8061 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8062 assert(src_start, "source array is null");
8063 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8064 assert(dest_start, "destination array is null");
8065
8066 Node* base64 = make_runtime_call(RC_LEAF,
8067 OptoRuntime::base64_encodeBlock_Type(),
8068 stubAddr, stubName, TypePtr::BOTTOM,
8069 src_start, offset, len, dest_start, dp, isURL);
8070 return true;
8071 }
8072
8073 bool LibraryCallKit::inline_base64_decodeBlock() {
8074 address stubAddr;
8075 const char *stubName;
8076 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8077 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8078 stubAddr = StubRoutines::base64_decodeBlock();
8079 stubName = "decodeBlock";
8080
8081 if (!stubAddr) return false;
8082 Node* base64obj = argument(0);
8083 Node* src = argument(1);
8084 Node* src_offset = argument(2);
8085 Node* len = argument(3);
8086 Node* dest = argument(4);
8087 Node* dest_offset = argument(5);
8088 Node* isURL = argument(6);
8089 Node* isMIME = argument(7);
8090
8091 src = must_be_not_null(src, true);
8092 dest = must_be_not_null(dest, true);
8093
8094 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8095 assert(src_start, "source array is null");
8096 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8097 assert(dest_start, "destination array is null");
8098
8099 Node* call = make_runtime_call(RC_LEAF,
8100 OptoRuntime::base64_decodeBlock_Type(),
8101 stubAddr, stubName, TypePtr::BOTTOM,
8102 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8103 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8104 set_result(result);
8105 return true;
8106 }
8107
8108 bool LibraryCallKit::inline_poly1305_processBlocks() {
8109 address stubAddr;
8110 const char *stubName;
8111 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8112 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8113 stubAddr = StubRoutines::poly1305_processBlocks();
8114 stubName = "poly1305_processBlocks";
8115
8116 if (!stubAddr) return false;
8117 null_check_receiver(); // null-check receiver
8118 if (stopped()) return true;
8119
8120 Node* input = argument(1);
8121 Node* input_offset = argument(2);
8122 Node* len = argument(3);
8123 Node* alimbs = argument(4);
8124 Node* rlimbs = argument(5);
8125
8126 input = must_be_not_null(input, true);
8127 alimbs = must_be_not_null(alimbs, true);
8128 rlimbs = must_be_not_null(rlimbs, true);
8129
8130 Node* input_start = array_element_address(input, input_offset, T_BYTE);
8131 assert(input_start, "input array is null");
8132 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
8133 assert(acc_start, "acc array is null");
8134 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
8135 assert(r_start, "r array is null");
8136
8137 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8138 OptoRuntime::poly1305_processBlocks_Type(),
8139 stubAddr, stubName, TypePtr::BOTTOM,
8140 input_start, len, acc_start, r_start);
8141 return true;
8142 }
8143
8144 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
8145 address stubAddr;
8146 const char *stubName;
8147 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8148 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
8149 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
8150 stubName = "intpoly_montgomeryMult_P256";
8151
8152 if (!stubAddr) return false;
8153 null_check_receiver(); // null-check receiver
8154 if (stopped()) return true;
8155
8156 Node* a = argument(1);
8157 Node* b = argument(2);
8158 Node* r = argument(3);
8159
8160 a = must_be_not_null(a, true);
8161 b = must_be_not_null(b, true);
8162 r = must_be_not_null(r, true);
8163
8164 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8165 assert(a_start, "a array is NULL");
8166 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8167 assert(b_start, "b array is NULL");
8168 Node* r_start = array_element_address(r, intcon(0), T_LONG);
8169 assert(r_start, "r array is NULL");
8170
8171 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8172 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
8173 stubAddr, stubName, TypePtr::BOTTOM,
8174 a_start, b_start, r_start);
8175 return true;
8176 }
8177
8178 bool LibraryCallKit::inline_intpoly_assign() {
8179 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8180 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
8181 const char *stubName = "intpoly_assign";
8182 address stubAddr = StubRoutines::intpoly_assign();
8183 if (!stubAddr) return false;
8184
8185 Node* set = argument(0);
8186 Node* a = argument(1);
8187 Node* b = argument(2);
8188 Node* arr_length = load_array_length(a);
8189
8190 a = must_be_not_null(a, true);
8191 b = must_be_not_null(b, true);
8192
8193 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8194 assert(a_start, "a array is NULL");
8195 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8196 assert(b_start, "b array is NULL");
8197
8198 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8199 OptoRuntime::intpoly_assign_Type(),
8200 stubAddr, stubName, TypePtr::BOTTOM,
8201 set, a_start, b_start, arr_length);
8202 return true;
8203 }
8204
8205 //------------------------------inline_digestBase_implCompress-----------------------
8206 //
8207 // Calculate MD5 for single-block byte[] array.
8208 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
8209 //
8210 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
8211 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
8212 //
8213 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
8214 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
8215 //
8216 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
8217 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
8218 //
8219 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
8220 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
8221 //
8222 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
8223 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
8224
8225 Node* digestBase_obj = argument(0);
8226 Node* src = argument(1); // type oop
8227 Node* ofs = argument(2); // type int
8228
8229 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8230 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8231 // failed array check
8232 return false;
8233 }
8234 // Figure out the size and type of the elements we will be copying.
8235 BasicType src_elem = src_type->elem()->array_element_basic_type();
8236 if (src_elem != T_BYTE) {
8237 return false;
8238 }
8239 // 'src_start' points to src array + offset
8240 src = must_be_not_null(src, true);
8241 Node* src_start = array_element_address(src, ofs, src_elem);
8242 Node* state = nullptr;
8243 Node* block_size = nullptr;
8244 address stubAddr;
8245 const char *stubName;
8246
8247 switch(id) {
8248 case vmIntrinsics::_md5_implCompress:
8249 assert(UseMD5Intrinsics, "need MD5 instruction support");
8250 state = get_state_from_digest_object(digestBase_obj, T_INT);
8251 stubAddr = StubRoutines::md5_implCompress();
8252 stubName = "md5_implCompress";
8253 break;
8254 case vmIntrinsics::_sha_implCompress:
8255 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
8256 state = get_state_from_digest_object(digestBase_obj, T_INT);
8257 stubAddr = StubRoutines::sha1_implCompress();
8258 stubName = "sha1_implCompress";
8259 break;
8260 case vmIntrinsics::_sha2_implCompress:
8261 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
8262 state = get_state_from_digest_object(digestBase_obj, T_INT);
8263 stubAddr = StubRoutines::sha256_implCompress();
8264 stubName = "sha256_implCompress";
8265 break;
8266 case vmIntrinsics::_sha5_implCompress:
8267 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
8268 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8269 stubAddr = StubRoutines::sha512_implCompress();
8270 stubName = "sha512_implCompress";
8271 break;
8272 case vmIntrinsics::_sha3_implCompress:
8273 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
8274 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8275 stubAddr = StubRoutines::sha3_implCompress();
8276 stubName = "sha3_implCompress";
8277 block_size = get_block_size_from_digest_object(digestBase_obj);
8278 if (block_size == nullptr) return false;
8279 break;
8280 default:
8281 fatal_unexpected_iid(id);
8282 return false;
8283 }
8284 if (state == nullptr) return false;
8285
8286 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
8287 if (stubAddr == nullptr) return false;
8288
8289 // Call the stub.
8290 Node* call;
8291 if (block_size == nullptr) {
8292 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
8293 stubAddr, stubName, TypePtr::BOTTOM,
8294 src_start, state);
8295 } else {
8296 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
8297 stubAddr, stubName, TypePtr::BOTTOM,
8298 src_start, state, block_size);
8299 }
8300
8301 return true;
8302 }
8303
8304 //------------------------------inline_digestBase_implCompressMB-----------------------
8305 //
8306 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
8307 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
8308 //
8309 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
8310 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8311 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8312 assert((uint)predicate < 5, "sanity");
8313 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
8314
8315 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
8316 Node* src = argument(1); // byte[] array
8317 Node* ofs = argument(2); // type int
8318 Node* limit = argument(3); // type int
8319
8320 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8321 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8322 // failed array check
8323 return false;
8324 }
8325 // Figure out the size and type of the elements we will be copying.
8326 BasicType src_elem = src_type->elem()->array_element_basic_type();
8327 if (src_elem != T_BYTE) {
8328 return false;
8329 }
8330 // 'src_start' points to src array + offset
8331 src = must_be_not_null(src, false);
8332 Node* src_start = array_element_address(src, ofs, src_elem);
8333
8334 const char* klass_digestBase_name = nullptr;
8335 const char* stub_name = nullptr;
8336 address stub_addr = nullptr;
8337 BasicType elem_type = T_INT;
8338
8339 switch (predicate) {
8340 case 0:
8341 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
8342 klass_digestBase_name = "sun/security/provider/MD5";
8343 stub_name = "md5_implCompressMB";
8344 stub_addr = StubRoutines::md5_implCompressMB();
8345 }
8346 break;
8347 case 1:
8348 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
8349 klass_digestBase_name = "sun/security/provider/SHA";
8350 stub_name = "sha1_implCompressMB";
8351 stub_addr = StubRoutines::sha1_implCompressMB();
8352 }
8353 break;
8354 case 2:
8355 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
8356 klass_digestBase_name = "sun/security/provider/SHA2";
8357 stub_name = "sha256_implCompressMB";
8358 stub_addr = StubRoutines::sha256_implCompressMB();
8359 }
8360 break;
8361 case 3:
8362 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
8363 klass_digestBase_name = "sun/security/provider/SHA5";
8364 stub_name = "sha512_implCompressMB";
8365 stub_addr = StubRoutines::sha512_implCompressMB();
8366 elem_type = T_LONG;
8367 }
8368 break;
8369 case 4:
8370 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
8371 klass_digestBase_name = "sun/security/provider/SHA3";
8372 stub_name = "sha3_implCompressMB";
8373 stub_addr = StubRoutines::sha3_implCompressMB();
8374 elem_type = T_LONG;
8375 }
8376 break;
8377 default:
8378 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
8379 }
8380 if (klass_digestBase_name != nullptr) {
8381 assert(stub_addr != nullptr, "Stub is generated");
8382 if (stub_addr == nullptr) return false;
8383
8384 // get DigestBase klass to lookup for SHA klass
8385 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
8386 assert(tinst != nullptr, "digestBase_obj is not instance???");
8387 assert(tinst->is_loaded(), "DigestBase is not loaded");
8388
8389 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
8390 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
8391 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
8392 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
8393 }
8394 return false;
8395 }
8396
8397 //------------------------------inline_digestBase_implCompressMB-----------------------
8398 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
8399 BasicType elem_type, address stubAddr, const char *stubName,
8400 Node* src_start, Node* ofs, Node* limit) {
8401 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
8402 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8403 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
8404 digest_obj = _gvn.transform(digest_obj);
8405
8406 Node* state = get_state_from_digest_object(digest_obj, elem_type);
8407 if (state == nullptr) return false;
8408
8409 Node* block_size = nullptr;
8410 if (strcmp("sha3_implCompressMB", stubName) == 0) {
8411 block_size = get_block_size_from_digest_object(digest_obj);
8412 if (block_size == nullptr) return false;
8413 }
8414
8415 // Call the stub.
8416 Node* call;
8417 if (block_size == nullptr) {
8418 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8419 OptoRuntime::digestBase_implCompressMB_Type(false),
8420 stubAddr, stubName, TypePtr::BOTTOM,
8421 src_start, state, ofs, limit);
8422 } else {
8423 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8424 OptoRuntime::digestBase_implCompressMB_Type(true),
8425 stubAddr, stubName, TypePtr::BOTTOM,
8426 src_start, state, block_size, ofs, limit);
8427 }
8428
8429 // return ofs (int)
8430 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8431 set_result(result);
8432
8433 return true;
8434 }
8435
8436 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8437 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8438 assert(UseAES, "need AES instruction support");
8439 address stubAddr = nullptr;
8440 const char *stubName = nullptr;
8441 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8442 stubName = "galoisCounterMode_AESCrypt";
8443
8444 if (stubAddr == nullptr) return false;
8445
8446 Node* in = argument(0);
8447 Node* inOfs = argument(1);
8448 Node* len = argument(2);
8449 Node* ct = argument(3);
8450 Node* ctOfs = argument(4);
8451 Node* out = argument(5);
8452 Node* outOfs = argument(6);
8453 Node* gctr_object = argument(7);
8454 Node* ghash_object = argument(8);
8455
8456 // (1) in, ct and out are arrays.
8457 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8458 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8459 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8460 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8461 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8462 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8463
8464 // checks are the responsibility of the caller
8465 Node* in_start = in;
8466 Node* ct_start = ct;
8467 Node* out_start = out;
8468 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8469 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8470 in_start = array_element_address(in, inOfs, T_BYTE);
8471 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8472 out_start = array_element_address(out, outOfs, T_BYTE);
8473 }
8474
8475 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8476 // (because of the predicated logic executed earlier).
8477 // so we cast it here safely.
8478 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8479 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8480 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8481 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8482 Node* state = load_field_from_object(ghash_object, "state", "[J");
8483
8484 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8485 return false;
8486 }
8487 // cast it to what we know it will be at runtime
8488 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8489 assert(tinst != nullptr, "GCTR obj is null");
8490 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8491 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8492 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8493 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8494 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8495 const TypeOopPtr* xtype = aklass->as_instance_type();
8496 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8497 aescrypt_object = _gvn.transform(aescrypt_object);
8498 // we need to get the start of the aescrypt_object's expanded key array
8499 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8500 if (k_start == nullptr) return false;
8501 // similarly, get the start address of the r vector
8502 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8503 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8504 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8505
8506
8507 // Call the stub, passing params
8508 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8509 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8510 stubAddr, stubName, TypePtr::BOTTOM,
8511 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8512
8513 // return cipher length (int)
8514 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8515 set_result(retvalue);
8516
8517 return true;
8518 }
8519
8520 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8521 // Return node representing slow path of predicate check.
8522 // the pseudo code we want to emulate with this predicate is:
8523 // for encryption:
8524 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8525 // for decryption:
8526 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8527 // note cipher==plain is more conservative than the original java code but that's OK
8528 //
8529
8530 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8531 // The receiver was checked for null already.
8532 Node* objGCTR = argument(7);
8533 // Load embeddedCipher field of GCTR object.
8534 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8535 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8536
8537 // get AESCrypt klass for instanceOf check
8538 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8539 // will have same classloader as CipherBlockChaining object
8540 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8541 assert(tinst != nullptr, "GCTR obj is null");
8542 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8543
8544 // we want to do an instanceof comparison against the AESCrypt class
8545 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8546 if (!klass_AESCrypt->is_loaded()) {
8547 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8548 Node* ctrl = control();
8549 set_control(top()); // no regular fast path
8550 return ctrl;
8551 }
8552
8553 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8554 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8555 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8556 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8557 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8558
8559 return instof_false; // even if it is null
8560 }
8561
8562 //------------------------------get_state_from_digest_object-----------------------
8563 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8564 const char* state_type;
8565 switch (elem_type) {
8566 case T_BYTE: state_type = "[B"; break;
8567 case T_INT: state_type = "[I"; break;
8568 case T_LONG: state_type = "[J"; break;
8569 default: ShouldNotReachHere();
8570 }
8571 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8572 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8573 if (digest_state == nullptr) return (Node *) nullptr;
8574
8575 // now have the array, need to get the start address of the state array
8576 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8577 return state;
8578 }
8579
8580 //------------------------------get_block_size_from_sha3_object----------------------------------
8581 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8582 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8583 assert (block_size != nullptr, "sanity");
8584 return block_size;
8585 }
8586
8587 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8588 // Return node representing slow path of predicate check.
8589 // the pseudo code we want to emulate with this predicate is:
8590 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8591 //
8592 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8593 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8594 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8595 assert((uint)predicate < 5, "sanity");
8596
8597 // The receiver was checked for null already.
8598 Node* digestBaseObj = argument(0);
8599
8600 // get DigestBase klass for instanceOf check
8601 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8602 assert(tinst != nullptr, "digestBaseObj is null");
8603 assert(tinst->is_loaded(), "DigestBase is not loaded");
8604
8605 const char* klass_name = nullptr;
8606 switch (predicate) {
8607 case 0:
8608 if (UseMD5Intrinsics) {
8609 // we want to do an instanceof comparison against the MD5 class
8610 klass_name = "sun/security/provider/MD5";
8611 }
8612 break;
8613 case 1:
8614 if (UseSHA1Intrinsics) {
8615 // we want to do an instanceof comparison against the SHA class
8616 klass_name = "sun/security/provider/SHA";
8617 }
8618 break;
8619 case 2:
8620 if (UseSHA256Intrinsics) {
8621 // we want to do an instanceof comparison against the SHA2 class
8622 klass_name = "sun/security/provider/SHA2";
8623 }
8624 break;
8625 case 3:
8626 if (UseSHA512Intrinsics) {
8627 // we want to do an instanceof comparison against the SHA5 class
8628 klass_name = "sun/security/provider/SHA5";
8629 }
8630 break;
8631 case 4:
8632 if (UseSHA3Intrinsics) {
8633 // we want to do an instanceof comparison against the SHA3 class
8634 klass_name = "sun/security/provider/SHA3";
8635 }
8636 break;
8637 default:
8638 fatal("unknown SHA intrinsic predicate: %d", predicate);
8639 }
8640
8641 ciKlass* klass = nullptr;
8642 if (klass_name != nullptr) {
8643 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8644 }
8645 if ((klass == nullptr) || !klass->is_loaded()) {
8646 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8647 Node* ctrl = control();
8648 set_control(top()); // no intrinsic path
8649 return ctrl;
8650 }
8651 ciInstanceKlass* instklass = klass->as_instance_klass();
8652
8653 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8654 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8655 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8656 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8657
8658 return instof_false; // even if it is null
8659 }
8660
8661 //-------------inline_fma-----------------------------------
8662 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8663 Node *a = nullptr;
8664 Node *b = nullptr;
8665 Node *c = nullptr;
8666 Node* result = nullptr;
8667 switch (id) {
8668 case vmIntrinsics::_fmaD:
8669 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8670 // no receiver since it is static method
8671 a = round_double_node(argument(0));
8672 b = round_double_node(argument(2));
8673 c = round_double_node(argument(4));
8674 result = _gvn.transform(new FmaDNode(control(), a, b, c));
8675 break;
8676 case vmIntrinsics::_fmaF:
8677 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8678 a = argument(0);
8679 b = argument(1);
8680 c = argument(2);
8681 result = _gvn.transform(new FmaFNode(control(), a, b, c));
8682 break;
8683 default:
8684 fatal_unexpected_iid(id); break;
8685 }
8686 set_result(result);
8687 return true;
8688 }
8689
8690 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8691 // argument(0) is receiver
8692 Node* codePoint = argument(1);
8693 Node* n = nullptr;
8694
8695 switch (id) {
8696 case vmIntrinsics::_isDigit :
8697 n = new DigitNode(control(), codePoint);
8698 break;
8699 case vmIntrinsics::_isLowerCase :
8700 n = new LowerCaseNode(control(), codePoint);
8701 break;
8702 case vmIntrinsics::_isUpperCase :
8703 n = new UpperCaseNode(control(), codePoint);
8704 break;
8705 case vmIntrinsics::_isWhitespace :
8706 n = new WhitespaceNode(control(), codePoint);
8707 break;
8708 default:
8709 fatal_unexpected_iid(id);
8710 }
8711
8712 set_result(_gvn.transform(n));
8713 return true;
8714 }
8715
8716 //------------------------------inline_fp_min_max------------------------------
8717 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
8718 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
8719
8720 // The intrinsic should be used only when the API branches aren't predictable,
8721 // the last one performing the most important comparison. The following heuristic
8722 // uses the branch statistics to eventually bail out if necessary.
8723
8724 ciMethodData *md = callee()->method_data();
8725
8726 if ( md != nullptr && md->is_mature() && md->invocation_count() > 0 ) {
8727 ciCallProfile cp = caller()->call_profile_at_bci(bci());
8728
8729 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
8730 // Bail out if the call-site didn't contribute enough to the statistics.
8731 return false;
8732 }
8733
8734 uint taken = 0, not_taken = 0;
8735
8736 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
8737 if (p->is_BranchData()) {
8738 taken = ((ciBranchData*)p)->taken();
8739 not_taken = ((ciBranchData*)p)->not_taken();
8740 }
8741 }
8742
8743 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
8744 balance = balance < 0 ? -balance : balance;
8745 if ( balance > 0.2 ) {
8746 // Bail out if the most important branch is predictable enough.
8747 return false;
8748 }
8749 }
8750 */
8751
8752 Node *a = nullptr;
8753 Node *b = nullptr;
8754 Node *n = nullptr;
8755 switch (id) {
8756 case vmIntrinsics::_maxF:
8757 case vmIntrinsics::_minF:
8758 case vmIntrinsics::_maxF_strict:
8759 case vmIntrinsics::_minF_strict:
8760 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
8761 a = argument(0);
8762 b = argument(1);
8763 break;
8764 case vmIntrinsics::_maxD:
8765 case vmIntrinsics::_minD:
8766 case vmIntrinsics::_maxD_strict:
8767 case vmIntrinsics::_minD_strict:
8768 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
8769 a = round_double_node(argument(0));
8770 b = round_double_node(argument(2));
8771 break;
8772 default:
8773 fatal_unexpected_iid(id);
8774 break;
8775 }
8776 switch (id) {
8777 case vmIntrinsics::_maxF:
8778 case vmIntrinsics::_maxF_strict:
8779 n = new MaxFNode(a, b);
8780 break;
8781 case vmIntrinsics::_minF:
8782 case vmIntrinsics::_minF_strict:
8783 n = new MinFNode(a, b);
8784 break;
8785 case vmIntrinsics::_maxD:
8786 case vmIntrinsics::_maxD_strict:
8787 n = new MaxDNode(a, b);
8788 break;
8789 case vmIntrinsics::_minD:
8790 case vmIntrinsics::_minD_strict:
8791 n = new MinDNode(a, b);
8792 break;
8793 default:
8794 fatal_unexpected_iid(id);
8795 break;
8796 }
8797 set_result(_gvn.transform(n));
8798 return true;
8799 }
8800
8801 bool LibraryCallKit::inline_profileBoolean() {
8802 Node* counts = argument(1);
8803 const TypeAryPtr* ary = nullptr;
8804 ciArray* aobj = nullptr;
8805 if (counts->is_Con()
8806 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8807 && (aobj = ary->const_oop()->as_array()) != nullptr
8808 && (aobj->length() == 2)) {
8809 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8810 jint false_cnt = aobj->element_value(0).as_int();
8811 jint true_cnt = aobj->element_value(1).as_int();
8812
8813 if (C->log() != nullptr) {
8814 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8815 false_cnt, true_cnt);
8816 }
8817
8818 if (false_cnt + true_cnt == 0) {
8819 // According to profile, never executed.
8820 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8821 Deoptimization::Action_reinterpret);
8822 return true;
8823 }
8824
8825 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8826 // is a number of each value occurrences.
8827 Node* result = argument(0);
8828 if (false_cnt == 0 || true_cnt == 0) {
8829 // According to profile, one value has been never seen.
8830 int expected_val = (false_cnt == 0) ? 1 : 0;
8831
8832 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8833 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8834
8835 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8836 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8837 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8838
8839 { // Slow path: uncommon trap for never seen value and then reexecute
8840 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8841 // the value has been seen at least once.
8842 PreserveJVMState pjvms(this);
8843 PreserveReexecuteState preexecs(this);
8844 jvms()->set_should_reexecute(true);
8845
8846 set_control(slow_path);
8847 set_i_o(i_o());
8848
8849 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8850 Deoptimization::Action_reinterpret);
8851 }
8852 // The guard for never seen value enables sharpening of the result and
8853 // returning a constant. It allows to eliminate branches on the same value
8854 // later on.
8855 set_control(fast_path);
8856 result = intcon(expected_val);
8857 }
8858 // Stop profiling.
8859 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8860 // By replacing method body with profile data (represented as ProfileBooleanNode
8861 // on IR level) we effectively disable profiling.
8862 // It enables full speed execution once optimized code is generated.
8863 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8864 C->record_for_igvn(profile);
8865 set_result(profile);
8866 return true;
8867 } else {
8868 // Continue profiling.
8869 // Profile data isn't available at the moment. So, execute method's bytecode version.
8870 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8871 // is compiled and counters aren't available since corresponding MethodHandle
8872 // isn't a compile-time constant.
8873 return false;
8874 }
8875 }
8876
8877 bool LibraryCallKit::inline_isCompileConstant() {
8878 Node* n = argument(0);
8879 set_result(n->is_Con() ? intcon(1) : intcon(0));
8880 return true;
8881 }
8882
8883 //------------------------------- inline_getObjectSize --------------------------------------
8884 //
8885 // Calculate the runtime size of the object/array.
8886 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8887 //
8888 bool LibraryCallKit::inline_getObjectSize() {
8889 Node* obj = argument(3);
8890 Node* klass_node = load_object_klass(obj);
8891
8892 jint layout_con = Klass::_lh_neutral_value;
8893 Node* layout_val = get_layout_helper(klass_node, layout_con);
8894 int layout_is_con = (layout_val == nullptr);
8895
8896 if (layout_is_con) {
8897 // Layout helper is constant, can figure out things at compile time.
8898
8899 if (Klass::layout_helper_is_instance(layout_con)) {
8900 // Instance case: layout_con contains the size itself.
8901 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8902 set_result(size);
8903 } else {
8904 // Array case: size is round(header + element_size*arraylength).
8905 // Since arraylength is different for every array instance, we have to
8906 // compute the whole thing at runtime.
8907
8908 Node* arr_length = load_array_length(obj);
8909
8910 int round_mask = MinObjAlignmentInBytes - 1;
8911 int hsize = Klass::layout_helper_header_size(layout_con);
8912 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8913
8914 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8915 round_mask = 0; // strength-reduce it if it goes away completely
8916 }
8917 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8918 Node* header_size = intcon(hsize + round_mask);
8919
8920 Node* lengthx = ConvI2X(arr_length);
8921 Node* headerx = ConvI2X(header_size);
8922
8923 Node* abody = lengthx;
8924 if (eshift != 0) {
8925 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8926 }
8927 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8928 if (round_mask != 0) {
8929 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8930 }
8931 size = ConvX2L(size);
8932 set_result(size);
8933 }
8934 } else {
8935 // Layout helper is not constant, need to test for array-ness at runtime.
8936
8937 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8938 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8939 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8940 record_for_igvn(result_reg);
8941
8942 Node* array_ctl = generate_array_guard(klass_node, nullptr);
8943 if (array_ctl != nullptr) {
8944 // Array case: size is round(header + element_size*arraylength).
8945 // Since arraylength is different for every array instance, we have to
8946 // compute the whole thing at runtime.
8947
8948 PreserveJVMState pjvms(this);
8949 set_control(array_ctl);
8950 Node* arr_length = load_array_length(obj);
8951
8952 int round_mask = MinObjAlignmentInBytes - 1;
8953 Node* mask = intcon(round_mask);
8954
8955 Node* hss = intcon(Klass::_lh_header_size_shift);
8956 Node* hsm = intcon(Klass::_lh_header_size_mask);
8957 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8958 header_size = _gvn.transform(new AndINode(header_size, hsm));
8959 header_size = _gvn.transform(new AddINode(header_size, mask));
8960
8961 // There is no need to mask or shift this value.
8962 // The semantics of LShiftINode include an implicit mask to 0x1F.
8963 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8964 Node* elem_shift = layout_val;
8965
8966 Node* lengthx = ConvI2X(arr_length);
8967 Node* headerx = ConvI2X(header_size);
8968
8969 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
8970 Node* size = _gvn.transform(new AddXNode(headerx, abody));
8971 if (round_mask != 0) {
8972 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
8973 }
8974 size = ConvX2L(size);
8975
8976 result_reg->init_req(_array_path, control());
8977 result_val->init_req(_array_path, size);
8978 }
8979
8980 if (!stopped()) {
8981 // Instance case: the layout helper gives us instance size almost directly,
8982 // but we need to mask out the _lh_instance_slow_path_bit.
8983 Node* size = ConvI2X(layout_val);
8984 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
8985 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
8986 size = _gvn.transform(new AndXNode(size, mask));
8987 size = ConvX2L(size);
8988
8989 result_reg->init_req(_instance_path, control());
8990 result_val->init_req(_instance_path, size);
8991 }
8992
8993 set_result(result_reg, result_val);
8994 }
8995
8996 return true;
8997 }
8998
8999 //------------------------------- inline_blackhole --------------------------------------
9000 //
9001 // Make sure all arguments to this node are alive.
9002 // This matches methods that were requested to be blackholed through compile commands.
9003 //
9004 bool LibraryCallKit::inline_blackhole() {
9005 assert(callee()->is_static(), "Should have been checked before: only static methods here");
9006 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
9007 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
9008
9009 // Blackhole node pinches only the control, not memory. This allows
9010 // the blackhole to be pinned in the loop that computes blackholed
9011 // values, but have no other side effects, like breaking the optimizations
9012 // across the blackhole.
9013
9014 Node* bh = _gvn.transform(new BlackholeNode(control()));
9015 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
9016
9017 // Bind call arguments as blackhole arguments to keep them alive
9018 uint nargs = callee()->arg_size();
9019 for (uint i = 0; i < nargs; i++) {
9020 bh->add_req(argument(i));
9021 }
9022
9023 return true;
9024 }