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