1 /*
2 * Copyright (c) 1999, 2024, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "ci/ciUtilities.inline.hpp"
28 #include "classfile/vmIntrinsics.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "gc/shared/barrierSet.hpp"
32 #include "jfr/support/jfrIntrinsics.hpp"
33 #include "memory/resourceArea.hpp"
34 #include "oops/klass.inline.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "opto/addnode.hpp"
37 #include "opto/arraycopynode.hpp"
38 #include "opto/c2compiler.hpp"
39 #include "opto/castnode.hpp"
40 #include "opto/cfgnode.hpp"
41 #include "opto/convertnode.hpp"
42 #include "opto/countbitsnode.hpp"
43 #include "opto/idealKit.hpp"
44 #include "opto/library_call.hpp"
45 #include "opto/mathexactnode.hpp"
46 #include "opto/mulnode.hpp"
47 #include "opto/narrowptrnode.hpp"
48 #include "opto/opaquenode.hpp"
49 #include "opto/parse.hpp"
50 #include "opto/runtime.hpp"
51 #include "opto/rootnode.hpp"
52 #include "opto/subnode.hpp"
53 #include "prims/jvmtiExport.hpp"
54 #include "prims/jvmtiThreadState.hpp"
55 #include "prims/unsafe.hpp"
56 #include "runtime/jniHandles.inline.hpp"
57 #include "runtime/objectMonitor.hpp"
58 #include "runtime/sharedRuntime.hpp"
59 #include "runtime/stubRoutines.hpp"
60 #include "utilities/macros.hpp"
61 #include "utilities/powerOfTwo.hpp"
62
63 //---------------------------make_vm_intrinsic----------------------------
64 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
65 vmIntrinsicID id = m->intrinsic_id();
66 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
67
68 if (!m->is_loaded()) {
69 // Do not attempt to inline unloaded methods.
70 return nullptr;
71 }
72
73 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
74 bool is_available = false;
75
76 {
77 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
78 // the compiler must transition to '_thread_in_vm' state because both
79 // methods access VM-internal data.
80 VM_ENTRY_MARK;
81 methodHandle mh(THREAD, m->get_Method());
82 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
83 if (is_available && is_virtual) {
84 is_available = vmIntrinsics::does_virtual_dispatch(id);
85 }
86 }
87
88 if (is_available) {
89 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
90 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
91 return new LibraryIntrinsic(m, is_virtual,
92 vmIntrinsics::predicates_needed(id),
93 vmIntrinsics::does_virtual_dispatch(id),
94 id);
95 } else {
96 return nullptr;
97 }
98 }
99
100 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
101 LibraryCallKit kit(jvms, this);
102 Compile* C = kit.C;
103 int nodes = C->unique();
104 #ifndef PRODUCT
105 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
106 char buf[1000];
107 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
108 tty->print_cr("Intrinsic %s", str);
109 }
110 #endif
111 ciMethod* callee = kit.callee();
112 const int bci = kit.bci();
113 #ifdef ASSERT
114 Node* ctrl = kit.control();
115 #endif
116 // Try to inline the intrinsic.
117 if (callee->check_intrinsic_candidate() &&
118 kit.try_to_inline(_last_predicate)) {
119 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
120 : "(intrinsic)";
121 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
122 if (C->print_intrinsics() || C->print_inlining()) {
123 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
124 }
125 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
126 if (C->log()) {
127 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
128 vmIntrinsics::name_at(intrinsic_id()),
129 (is_virtual() ? " virtual='1'" : ""),
130 C->unique() - nodes);
131 }
132 // Push the result from the inlined method onto the stack.
133 kit.push_result();
134 C->print_inlining_update(this);
135 return kit.transfer_exceptions_into_jvms();
136 }
137
138 // The intrinsic bailed out
139 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
140 if (jvms->has_method()) {
141 // Not a root compile.
142 const char* msg;
143 if (callee->intrinsic_candidate()) {
144 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
145 } else {
146 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
147 : "failed to inline (intrinsic), method not annotated";
148 }
149 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
150 if (C->print_intrinsics() || C->print_inlining()) {
151 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
152 }
153 } else {
154 // Root compile
155 ResourceMark rm;
156 stringStream msg_stream;
157 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
158 vmIntrinsics::name_at(intrinsic_id()),
159 is_virtual() ? " (virtual)" : "", bci);
160 const char *msg = msg_stream.freeze();
161 log_debug(jit, inlining)("%s", msg);
162 if (C->print_intrinsics() || C->print_inlining()) {
163 tty->print("%s", msg);
164 }
165 }
166 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
167 C->print_inlining_update(this);
168
169 return nullptr;
170 }
171
172 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
173 LibraryCallKit kit(jvms, this);
174 Compile* C = kit.C;
175 int nodes = C->unique();
176 _last_predicate = predicate;
177 #ifndef PRODUCT
178 assert(is_predicated() && predicate < predicates_count(), "sanity");
179 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
180 char buf[1000];
181 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
182 tty->print_cr("Predicate for intrinsic %s", str);
183 }
184 #endif
185 ciMethod* callee = kit.callee();
186 const int bci = kit.bci();
187
188 Node* slow_ctl = kit.try_to_predicate(predicate);
189 if (!kit.failing()) {
190 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
191 : "(intrinsic, predicate)";
192 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
193 if (C->print_intrinsics() || C->print_inlining()) {
194 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
195 }
196 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
197 if (C->log()) {
198 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
199 vmIntrinsics::name_at(intrinsic_id()),
200 (is_virtual() ? " virtual='1'" : ""),
201 C->unique() - nodes);
202 }
203 return slow_ctl; // Could be null if the check folds.
204 }
205
206 // The intrinsic bailed out
207 if (jvms->has_method()) {
208 // Not a root compile.
209 const char* msg = "failed to generate predicate for intrinsic";
210 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
211 if (C->print_intrinsics() || C->print_inlining()) {
212 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
213 }
214 } else {
215 // Root compile
216 ResourceMark rm;
217 stringStream msg_stream;
218 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
219 vmIntrinsics::name_at(intrinsic_id()),
220 is_virtual() ? " (virtual)" : "", bci);
221 const char *msg = msg_stream.freeze();
222 log_debug(jit, inlining)("%s", msg);
223 if (C->print_intrinsics() || C->print_inlining()) {
224 C->print_inlining_stream()->print("%s", msg);
225 }
226 }
227 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
228 return nullptr;
229 }
230
231 bool LibraryCallKit::try_to_inline(int predicate) {
232 // Handle symbolic names for otherwise undistinguished boolean switches:
233 const bool is_store = true;
234 const bool is_compress = true;
235 const bool is_static = true;
236 const bool is_volatile = true;
237
238 if (!jvms()->has_method()) {
239 // Root JVMState has a null method.
240 assert(map()->memory()->Opcode() == Op_Parm, "");
241 // Insert the memory aliasing node
242 set_all_memory(reset_memory());
243 }
244 assert(merged_memory(), "");
245
246 switch (intrinsic_id()) {
247 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
248 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
249 case vmIntrinsics::_getClass: return inline_native_getClass();
250
251 case vmIntrinsics::_ceil:
252 case vmIntrinsics::_floor:
253 case vmIntrinsics::_rint:
254 case vmIntrinsics::_dsin:
255 case vmIntrinsics::_dcos:
256 case vmIntrinsics::_dtan:
257 case vmIntrinsics::_dtanh:
258 case vmIntrinsics::_dabs:
259 case vmIntrinsics::_fabs:
260 case vmIntrinsics::_iabs:
261 case vmIntrinsics::_labs:
262 case vmIntrinsics::_datan2:
263 case vmIntrinsics::_dsqrt:
264 case vmIntrinsics::_dsqrt_strict:
265 case vmIntrinsics::_dexp:
266 case vmIntrinsics::_dlog:
267 case vmIntrinsics::_dlog10:
268 case vmIntrinsics::_dpow:
269 case vmIntrinsics::_dcopySign:
270 case vmIntrinsics::_fcopySign:
271 case vmIntrinsics::_dsignum:
272 case vmIntrinsics::_roundF:
273 case vmIntrinsics::_roundD:
274 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
275
276 case vmIntrinsics::_notify:
277 case vmIntrinsics::_notifyAll:
278 return inline_notify(intrinsic_id());
279
280 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
281 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
282 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
283 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
284 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
285 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
286 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
287 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
288 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
289 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
290 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
291 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
292 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
293 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
294
295 case vmIntrinsics::_arraycopy: return inline_arraycopy();
296
297 case vmIntrinsics::_arraySort: return inline_array_sort();
298 case vmIntrinsics::_arrayPartition: return inline_array_partition();
299
300 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
301 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
302 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
303 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
304
305 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
306 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
307 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
308 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
309 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
310 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
311 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
312 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
313
314 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
315
316 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
317
318 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
319 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
320 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
321 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
322
323 case vmIntrinsics::_compressStringC:
324 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
325 case vmIntrinsics::_inflateStringC:
326 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
327
328 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
329 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
330 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
331 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
332 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
333 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
334 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
335 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
336 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
337
338 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
339 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
340 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
341 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
342 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
343 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
344 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
345 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
346 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
347
348 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
349 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
350 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
351 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
352 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
353 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
354 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
355 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
356 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
357
358 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
359 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
360 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
361 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
362 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
363 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
364 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
365 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
366 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
367
368 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
369 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
370 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
371 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
372
373 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
374 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
375 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
376 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
377
378 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
379 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
380 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
381 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
382 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
383 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
384 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
385 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
386 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
387
388 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
389 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
390 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
391 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
392 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
393 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
394 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
395 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
396 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
397
398 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
399 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
400 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
401 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
402 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
403 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
404 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
405 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
406 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
407
408 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
409 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
410 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
411 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
412 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
413 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
414 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
415 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
416 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
417
418 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
419 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
420 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
421 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
422 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
423
424 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
425 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
426 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
427 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
428 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
429 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
430 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
431 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
432 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
433 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
434 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
435 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
436 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
437 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
438 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
439 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
440 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
441 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
442 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
443 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
444
445 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
446 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
447 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
448 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
449 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
450 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
451 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
452 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
453 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
454 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
455 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
456 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
457 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
458 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
459 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
460
461 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
462 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
463 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
464 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
465
466 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
467 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
468 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
469 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
470 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
471
472 case vmIntrinsics::_loadFence:
473 case vmIntrinsics::_storeFence:
474 case vmIntrinsics::_storeStoreFence:
475 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
476
477 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
478
479 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
480 case vmIntrinsics::_currentThread: return inline_native_currentThread();
481 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
482
483 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
484 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
485
486 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
487 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
488
489 #if INCLUDE_JVMTI
490 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
491 "notifyJvmtiStart", true, false);
492 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
493 "notifyJvmtiEnd", false, true);
494 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
495 "notifyJvmtiMount", false, false);
496 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
497 "notifyJvmtiUnmount", false, false);
498 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
499 #endif
500
501 #ifdef JFR_HAVE_INTRINSICS
502 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
503 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
504 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
505 #endif
506 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
507 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
508 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
509 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
510 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
511 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
512 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
513 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
514 case vmIntrinsics::_getLength: return inline_native_getLength();
515 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
516 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
517 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
518 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
519 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
520 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
521 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
522
523 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
524 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
525
526 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
527
528 case vmIntrinsics::_isInstance:
529 case vmIntrinsics::_getModifiers:
530 case vmIntrinsics::_isInterface:
531 case vmIntrinsics::_isArray:
532 case vmIntrinsics::_isPrimitive:
533 case vmIntrinsics::_isHidden:
534 case vmIntrinsics::_getSuperclass:
535 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
536
537 case vmIntrinsics::_floatToRawIntBits:
538 case vmIntrinsics::_floatToIntBits:
539 case vmIntrinsics::_intBitsToFloat:
540 case vmIntrinsics::_doubleToRawLongBits:
541 case vmIntrinsics::_doubleToLongBits:
542 case vmIntrinsics::_longBitsToDouble:
543 case vmIntrinsics::_floatToFloat16:
544 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
545
546 case vmIntrinsics::_floatIsFinite:
547 case vmIntrinsics::_floatIsInfinite:
548 case vmIntrinsics::_doubleIsFinite:
549 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
550
551 case vmIntrinsics::_numberOfLeadingZeros_i:
552 case vmIntrinsics::_numberOfLeadingZeros_l:
553 case vmIntrinsics::_numberOfTrailingZeros_i:
554 case vmIntrinsics::_numberOfTrailingZeros_l:
555 case vmIntrinsics::_bitCount_i:
556 case vmIntrinsics::_bitCount_l:
557 case vmIntrinsics::_reverse_i:
558 case vmIntrinsics::_reverse_l:
559 case vmIntrinsics::_reverseBytes_i:
560 case vmIntrinsics::_reverseBytes_l:
561 case vmIntrinsics::_reverseBytes_s:
562 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
563
564 case vmIntrinsics::_compress_i:
565 case vmIntrinsics::_compress_l:
566 case vmIntrinsics::_expand_i:
567 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
568
569 case vmIntrinsics::_compareUnsigned_i:
570 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
571
572 case vmIntrinsics::_divideUnsigned_i:
573 case vmIntrinsics::_divideUnsigned_l:
574 case vmIntrinsics::_remainderUnsigned_i:
575 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
576
577 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
578
579 case vmIntrinsics::_Reference_get: return inline_reference_get();
580 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
581 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
582 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
583 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
584
585 case vmIntrinsics::_Class_cast: return inline_Class_cast();
586
587 case vmIntrinsics::_aescrypt_encryptBlock:
588 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
589
590 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
591 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
592 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
593
594 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
595 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
596 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
597
598 case vmIntrinsics::_counterMode_AESCrypt:
599 return inline_counterMode_AESCrypt(intrinsic_id());
600
601 case vmIntrinsics::_galoisCounterMode_AESCrypt:
602 return inline_galoisCounterMode_AESCrypt();
603
604 case vmIntrinsics::_md5_implCompress:
605 case vmIntrinsics::_sha_implCompress:
606 case vmIntrinsics::_sha2_implCompress:
607 case vmIntrinsics::_sha5_implCompress:
608 case vmIntrinsics::_sha3_implCompress:
609 return inline_digestBase_implCompress(intrinsic_id());
610
611 case vmIntrinsics::_digestBase_implCompressMB:
612 return inline_digestBase_implCompressMB(predicate);
613
614 case vmIntrinsics::_multiplyToLen:
615 return inline_multiplyToLen();
616
617 case vmIntrinsics::_squareToLen:
618 return inline_squareToLen();
619
620 case vmIntrinsics::_mulAdd:
621 return inline_mulAdd();
622
623 case vmIntrinsics::_montgomeryMultiply:
624 return inline_montgomeryMultiply();
625 case vmIntrinsics::_montgomerySquare:
626 return inline_montgomerySquare();
627
628 case vmIntrinsics::_bigIntegerRightShiftWorker:
629 return inline_bigIntegerShift(true);
630 case vmIntrinsics::_bigIntegerLeftShiftWorker:
631 return inline_bigIntegerShift(false);
632
633 case vmIntrinsics::_vectorizedMismatch:
634 return inline_vectorizedMismatch();
635
636 case vmIntrinsics::_ghash_processBlocks:
637 return inline_ghash_processBlocks();
638 case vmIntrinsics::_chacha20Block:
639 return inline_chacha20Block();
640 case vmIntrinsics::_base64_encodeBlock:
641 return inline_base64_encodeBlock();
642 case vmIntrinsics::_base64_decodeBlock:
643 return inline_base64_decodeBlock();
644 case vmIntrinsics::_poly1305_processBlocks:
645 return inline_poly1305_processBlocks();
646 case vmIntrinsics::_intpoly_montgomeryMult_P256:
647 return inline_intpoly_montgomeryMult_P256();
648 case vmIntrinsics::_intpoly_assign:
649 return inline_intpoly_assign();
650 case vmIntrinsics::_encodeISOArray:
651 case vmIntrinsics::_encodeByteISOArray:
652 return inline_encodeISOArray(false);
653 case vmIntrinsics::_encodeAsciiArray:
654 return inline_encodeISOArray(true);
655
656 case vmIntrinsics::_updateCRC32:
657 return inline_updateCRC32();
658 case vmIntrinsics::_updateBytesCRC32:
659 return inline_updateBytesCRC32();
660 case vmIntrinsics::_updateByteBufferCRC32:
661 return inline_updateByteBufferCRC32();
662
663 case vmIntrinsics::_updateBytesCRC32C:
664 return inline_updateBytesCRC32C();
665 case vmIntrinsics::_updateDirectByteBufferCRC32C:
666 return inline_updateDirectByteBufferCRC32C();
667
668 case vmIntrinsics::_updateBytesAdler32:
669 return inline_updateBytesAdler32();
670 case vmIntrinsics::_updateByteBufferAdler32:
671 return inline_updateByteBufferAdler32();
672
673 case vmIntrinsics::_profileBoolean:
674 return inline_profileBoolean();
675 case vmIntrinsics::_isCompileConstant:
676 return inline_isCompileConstant();
677
678 case vmIntrinsics::_countPositives:
679 return inline_countPositives();
680
681 case vmIntrinsics::_fmaD:
682 case vmIntrinsics::_fmaF:
683 return inline_fma(intrinsic_id());
684
685 case vmIntrinsics::_isDigit:
686 case vmIntrinsics::_isLowerCase:
687 case vmIntrinsics::_isUpperCase:
688 case vmIntrinsics::_isWhitespace:
689 return inline_character_compare(intrinsic_id());
690
691 case vmIntrinsics::_min:
692 case vmIntrinsics::_max:
693 case vmIntrinsics::_min_strict:
694 case vmIntrinsics::_max_strict:
695 return inline_min_max(intrinsic_id());
696
697 case vmIntrinsics::_maxF:
698 case vmIntrinsics::_minF:
699 case vmIntrinsics::_maxD:
700 case vmIntrinsics::_minD:
701 case vmIntrinsics::_maxF_strict:
702 case vmIntrinsics::_minF_strict:
703 case vmIntrinsics::_maxD_strict:
704 case vmIntrinsics::_minD_strict:
705 return inline_fp_min_max(intrinsic_id());
706
707 case vmIntrinsics::_VectorUnaryOp:
708 return inline_vector_nary_operation(1);
709 case vmIntrinsics::_VectorBinaryOp:
710 return inline_vector_nary_operation(2);
711 case vmIntrinsics::_VectorTernaryOp:
712 return inline_vector_nary_operation(3);
713 case vmIntrinsics::_VectorFromBitsCoerced:
714 return inline_vector_frombits_coerced();
715 case vmIntrinsics::_VectorShuffleIota:
716 return inline_vector_shuffle_iota();
717 case vmIntrinsics::_VectorMaskOp:
718 return inline_vector_mask_operation();
719 case vmIntrinsics::_VectorShuffleToVector:
720 return inline_vector_shuffle_to_vector();
721 case vmIntrinsics::_VectorWrapShuffleIndexes:
722 return inline_vector_wrap_shuffle_indexes();
723 case vmIntrinsics::_VectorLoadOp:
724 return inline_vector_mem_operation(/*is_store=*/false);
725 case vmIntrinsics::_VectorLoadMaskedOp:
726 return inline_vector_mem_masked_operation(/*is_store*/false);
727 case vmIntrinsics::_VectorStoreOp:
728 return inline_vector_mem_operation(/*is_store=*/true);
729 case vmIntrinsics::_VectorStoreMaskedOp:
730 return inline_vector_mem_masked_operation(/*is_store=*/true);
731 case vmIntrinsics::_VectorGatherOp:
732 return inline_vector_gather_scatter(/*is_scatter*/ false);
733 case vmIntrinsics::_VectorScatterOp:
734 return inline_vector_gather_scatter(/*is_scatter*/ true);
735 case vmIntrinsics::_VectorReductionCoerced:
736 return inline_vector_reduction();
737 case vmIntrinsics::_VectorTest:
738 return inline_vector_test();
739 case vmIntrinsics::_VectorBlend:
740 return inline_vector_blend();
741 case vmIntrinsics::_VectorRearrange:
742 return inline_vector_rearrange();
743 case vmIntrinsics::_VectorSelectFrom:
744 return inline_vector_select_from();
745 case vmIntrinsics::_VectorCompare:
746 return inline_vector_compare();
747 case vmIntrinsics::_VectorBroadcastInt:
748 return inline_vector_broadcast_int();
749 case vmIntrinsics::_VectorConvert:
750 return inline_vector_convert();
751 case vmIntrinsics::_VectorInsert:
752 return inline_vector_insert();
753 case vmIntrinsics::_VectorExtract:
754 return inline_vector_extract();
755 case vmIntrinsics::_VectorCompressExpand:
756 return inline_vector_compress_expand();
757 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
758 return inline_vector_select_from_two_vectors();
759 case vmIntrinsics::_IndexVector:
760 return inline_index_vector();
761 case vmIntrinsics::_IndexPartiallyInUpperRange:
762 return inline_index_partially_in_upper_range();
763
764 case vmIntrinsics::_getObjectSize:
765 return inline_getObjectSize();
766
767 case vmIntrinsics::_blackhole:
768 return inline_blackhole();
769
770 default:
771 // If you get here, it may be that someone has added a new intrinsic
772 // to the list in vmIntrinsics.hpp without implementing it here.
773 #ifndef PRODUCT
774 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
775 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
776 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
777 }
778 #endif
779 return false;
780 }
781 }
782
783 Node* LibraryCallKit::try_to_predicate(int predicate) {
784 if (!jvms()->has_method()) {
785 // Root JVMState has a null method.
786 assert(map()->memory()->Opcode() == Op_Parm, "");
787 // Insert the memory aliasing node
788 set_all_memory(reset_memory());
789 }
790 assert(merged_memory(), "");
791
792 switch (intrinsic_id()) {
793 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
794 return inline_cipherBlockChaining_AESCrypt_predicate(false);
795 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
796 return inline_cipherBlockChaining_AESCrypt_predicate(true);
797 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
798 return inline_electronicCodeBook_AESCrypt_predicate(false);
799 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
800 return inline_electronicCodeBook_AESCrypt_predicate(true);
801 case vmIntrinsics::_counterMode_AESCrypt:
802 return inline_counterMode_AESCrypt_predicate();
803 case vmIntrinsics::_digestBase_implCompressMB:
804 return inline_digestBase_implCompressMB_predicate(predicate);
805 case vmIntrinsics::_galoisCounterMode_AESCrypt:
806 return inline_galoisCounterMode_AESCrypt_predicate();
807
808 default:
809 // If you get here, it may be that someone has added a new intrinsic
810 // to the list in vmIntrinsics.hpp without implementing it here.
811 #ifndef PRODUCT
812 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
813 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
814 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
815 }
816 #endif
817 Node* slow_ctl = control();
818 set_control(top()); // No fast path intrinsic
819 return slow_ctl;
820 }
821 }
822
823 //------------------------------set_result-------------------------------
824 // Helper function for finishing intrinsics.
825 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
826 record_for_igvn(region);
827 set_control(_gvn.transform(region));
828 set_result( _gvn.transform(value));
829 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
830 }
831
832 //------------------------------generate_guard---------------------------
833 // Helper function for generating guarded fast-slow graph structures.
834 // The given 'test', if true, guards a slow path. If the test fails
835 // then a fast path can be taken. (We generally hope it fails.)
836 // In all cases, GraphKit::control() is updated to the fast path.
837 // The returned value represents the control for the slow path.
838 // The return value is never 'top'; it is either a valid control
839 // or null if it is obvious that the slow path can never be taken.
840 // Also, if region and the slow control are not null, the slow edge
841 // is appended to the region.
842 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
843 if (stopped()) {
844 // Already short circuited.
845 return nullptr;
846 }
847
848 // Build an if node and its projections.
849 // If test is true we take the slow path, which we assume is uncommon.
850 if (_gvn.type(test) == TypeInt::ZERO) {
851 // The slow branch is never taken. No need to build this guard.
852 return nullptr;
853 }
854
855 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
856
857 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
858 if (if_slow == top()) {
859 // The slow branch is never taken. No need to build this guard.
860 return nullptr;
861 }
862
863 if (region != nullptr)
864 region->add_req(if_slow);
865
866 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
867 set_control(if_fast);
868
869 return if_slow;
870 }
871
872 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
873 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
874 }
875 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
876 return generate_guard(test, region, PROB_FAIR);
877 }
878
879 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
880 Node* *pos_index) {
881 if (stopped())
882 return nullptr; // already stopped
883 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
884 return nullptr; // index is already adequately typed
885 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
886 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
887 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
888 if (is_neg != nullptr && pos_index != nullptr) {
889 // Emulate effect of Parse::adjust_map_after_if.
890 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
891 (*pos_index) = _gvn.transform(ccast);
892 }
893 return is_neg;
894 }
895
896 // Make sure that 'position' is a valid limit index, in [0..length].
897 // There are two equivalent plans for checking this:
898 // A. (offset + copyLength) unsigned<= arrayLength
899 // B. offset <= (arrayLength - copyLength)
900 // We require that all of the values above, except for the sum and
901 // difference, are already known to be non-negative.
902 // Plan A is robust in the face of overflow, if offset and copyLength
903 // are both hugely positive.
904 //
905 // Plan B is less direct and intuitive, but it does not overflow at
906 // all, since the difference of two non-negatives is always
907 // representable. Whenever Java methods must perform the equivalent
908 // check they generally use Plan B instead of Plan A.
909 // For the moment we use Plan A.
910 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
911 Node* subseq_length,
912 Node* array_length,
913 RegionNode* region) {
914 if (stopped())
915 return nullptr; // already stopped
916 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
917 if (zero_offset && subseq_length->eqv_uncast(array_length))
918 return nullptr; // common case of whole-array copy
919 Node* last = subseq_length;
920 if (!zero_offset) // last += offset
921 last = _gvn.transform(new AddINode(last, offset));
922 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
923 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
924 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
925 return is_over;
926 }
927
928 // Emit range checks for the given String.value byte array
929 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
930 if (stopped()) {
931 return; // already stopped
932 }
933 RegionNode* bailout = new RegionNode(1);
934 record_for_igvn(bailout);
935 if (char_count) {
936 // Convert char count to byte count
937 count = _gvn.transform(new LShiftINode(count, intcon(1)));
938 }
939
940 // Offset and count must not be negative
941 generate_negative_guard(offset, bailout);
942 generate_negative_guard(count, bailout);
943 // Offset + count must not exceed length of array
944 generate_limit_guard(offset, count, load_array_length(array), bailout);
945
946 if (bailout->req() > 1) {
947 PreserveJVMState pjvms(this);
948 set_control(_gvn.transform(bailout));
949 uncommon_trap(Deoptimization::Reason_intrinsic,
950 Deoptimization::Action_maybe_recompile);
951 }
952 }
953
954 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
955 bool is_immutable) {
956 ciKlass* thread_klass = env()->Thread_klass();
957 const Type* thread_type
958 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
959
960 Node* thread = _gvn.transform(new ThreadLocalNode());
961 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
962 tls_output = thread;
963
964 Node* thread_obj_handle
965 = (is_immutable
966 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
967 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
968 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
969 thread_obj_handle = _gvn.transform(thread_obj_handle);
970
971 DecoratorSet decorators = IN_NATIVE;
972 if (is_immutable) {
973 decorators |= C2_IMMUTABLE_MEMORY;
974 }
975 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
976 }
977
978 //--------------------------generate_current_thread--------------------
979 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
980 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
981 /*is_immutable*/false);
982 }
983
984 //--------------------------generate_virtual_thread--------------------
985 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
986 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
987 !C->method()->changes_current_thread());
988 }
989
990 //------------------------------make_string_method_node------------------------
991 // Helper method for String intrinsic functions. This version is called with
992 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
993 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
994 // containing the lengths of str1 and str2.
995 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
996 Node* result = nullptr;
997 switch (opcode) {
998 case Op_StrIndexOf:
999 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1000 str1_start, cnt1, str2_start, cnt2, ae);
1001 break;
1002 case Op_StrComp:
1003 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1004 str1_start, cnt1, str2_start, cnt2, ae);
1005 break;
1006 case Op_StrEquals:
1007 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1008 // Use the constant length if there is one because optimized match rule may exist.
1009 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1010 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1011 break;
1012 default:
1013 ShouldNotReachHere();
1014 return nullptr;
1015 }
1016
1017 // All these intrinsics have checks.
1018 C->set_has_split_ifs(true); // Has chance for split-if optimization
1019 clear_upper_avx();
1020
1021 return _gvn.transform(result);
1022 }
1023
1024 //------------------------------inline_string_compareTo------------------------
1025 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1026 Node* arg1 = argument(0);
1027 Node* arg2 = argument(1);
1028
1029 arg1 = must_be_not_null(arg1, true);
1030 arg2 = must_be_not_null(arg2, true);
1031
1032 // Get start addr and length of first argument
1033 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1034 Node* arg1_cnt = load_array_length(arg1);
1035
1036 // Get start addr and length of second argument
1037 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1038 Node* arg2_cnt = load_array_length(arg2);
1039
1040 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1041 set_result(result);
1042 return true;
1043 }
1044
1045 //------------------------------inline_string_equals------------------------
1046 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1047 Node* arg1 = argument(0);
1048 Node* arg2 = argument(1);
1049
1050 // paths (plus control) merge
1051 RegionNode* region = new RegionNode(3);
1052 Node* phi = new PhiNode(region, TypeInt::BOOL);
1053
1054 if (!stopped()) {
1055
1056 arg1 = must_be_not_null(arg1, true);
1057 arg2 = must_be_not_null(arg2, true);
1058
1059 // Get start addr and length of first argument
1060 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1061 Node* arg1_cnt = load_array_length(arg1);
1062
1063 // Get start addr and length of second argument
1064 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1065 Node* arg2_cnt = load_array_length(arg2);
1066
1067 // Check for arg1_cnt != arg2_cnt
1068 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1069 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1070 Node* if_ne = generate_slow_guard(bol, nullptr);
1071 if (if_ne != nullptr) {
1072 phi->init_req(2, intcon(0));
1073 region->init_req(2, if_ne);
1074 }
1075
1076 // Check for count == 0 is done by assembler code for StrEquals.
1077
1078 if (!stopped()) {
1079 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1080 phi->init_req(1, equals);
1081 region->init_req(1, control());
1082 }
1083 }
1084
1085 // post merge
1086 set_control(_gvn.transform(region));
1087 record_for_igvn(region);
1088
1089 set_result(_gvn.transform(phi));
1090 return true;
1091 }
1092
1093 //------------------------------inline_array_equals----------------------------
1094 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1095 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1096 Node* arg1 = argument(0);
1097 Node* arg2 = argument(1);
1098
1099 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1100 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1101 clear_upper_avx();
1102
1103 return true;
1104 }
1105
1106
1107 //------------------------------inline_countPositives------------------------------
1108 bool LibraryCallKit::inline_countPositives() {
1109 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1110 return false;
1111 }
1112
1113 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1114 // no receiver since it is static method
1115 Node* ba = argument(0);
1116 Node* offset = argument(1);
1117 Node* len = argument(2);
1118
1119 ba = must_be_not_null(ba, true);
1120
1121 // Range checks
1122 generate_string_range_check(ba, offset, len, false);
1123 if (stopped()) {
1124 return true;
1125 }
1126 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1127 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1128 set_result(_gvn.transform(result));
1129 clear_upper_avx();
1130 return true;
1131 }
1132
1133 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1134 Node* index = argument(0);
1135 Node* length = bt == T_INT ? argument(1) : argument(2);
1136 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1137 return false;
1138 }
1139
1140 // check that length is positive
1141 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1142 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1143
1144 {
1145 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1146 uncommon_trap(Deoptimization::Reason_intrinsic,
1147 Deoptimization::Action_make_not_entrant);
1148 }
1149
1150 if (stopped()) {
1151 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1152 return true;
1153 }
1154
1155 // length is now known positive, add a cast node to make this explicit
1156 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1157 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1158 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1159 ConstraintCastNode::RegularDependency, bt);
1160 casted_length = _gvn.transform(casted_length);
1161 replace_in_map(length, casted_length);
1162 length = casted_length;
1163
1164 // Use an unsigned comparison for the range check itself
1165 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1166 BoolTest::mask btest = BoolTest::lt;
1167 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1168 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1169 _gvn.set_type(rc, rc->Value(&_gvn));
1170 if (!rc_bool->is_Con()) {
1171 record_for_igvn(rc);
1172 }
1173 set_control(_gvn.transform(new IfTrueNode(rc)));
1174 {
1175 PreserveJVMState pjvms(this);
1176 set_control(_gvn.transform(new IfFalseNode(rc)));
1177 uncommon_trap(Deoptimization::Reason_range_check,
1178 Deoptimization::Action_make_not_entrant);
1179 }
1180
1181 if (stopped()) {
1182 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1183 return true;
1184 }
1185
1186 // index is now known to be >= 0 and < length, cast it
1187 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1188 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1189 ConstraintCastNode::RegularDependency, bt);
1190 result = _gvn.transform(result);
1191 set_result(result);
1192 replace_in_map(index, result);
1193 return true;
1194 }
1195
1196 //------------------------------inline_string_indexOf------------------------
1197 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1198 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1199 return false;
1200 }
1201 Node* src = argument(0);
1202 Node* tgt = argument(1);
1203
1204 // Make the merge point
1205 RegionNode* result_rgn = new RegionNode(4);
1206 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1207
1208 src = must_be_not_null(src, true);
1209 tgt = must_be_not_null(tgt, true);
1210
1211 // Get start addr and length of source string
1212 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1213 Node* src_count = load_array_length(src);
1214
1215 // Get start addr and length of substring
1216 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1217 Node* tgt_count = load_array_length(tgt);
1218
1219 Node* result = nullptr;
1220 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1221
1222 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1223 // Divide src size by 2 if String is UTF16 encoded
1224 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1225 }
1226 if (ae == StrIntrinsicNode::UU) {
1227 // Divide substring size by 2 if String is UTF16 encoded
1228 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1229 }
1230
1231 if (call_opt_stub) {
1232 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1233 StubRoutines::_string_indexof_array[ae],
1234 "stringIndexOf", TypePtr::BOTTOM, src_start,
1235 src_count, tgt_start, tgt_count);
1236 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1237 } else {
1238 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1239 result_rgn, result_phi, ae);
1240 }
1241 if (result != nullptr) {
1242 result_phi->init_req(3, result);
1243 result_rgn->init_req(3, control());
1244 }
1245 set_control(_gvn.transform(result_rgn));
1246 record_for_igvn(result_rgn);
1247 set_result(_gvn.transform(result_phi));
1248
1249 return true;
1250 }
1251
1252 //-----------------------------inline_string_indexOfI-----------------------
1253 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1254 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1255 return false;
1256 }
1257 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1258 return false;
1259 }
1260
1261 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1262 Node* src = argument(0); // byte[]
1263 Node* src_count = argument(1); // char count
1264 Node* tgt = argument(2); // byte[]
1265 Node* tgt_count = argument(3); // char count
1266 Node* from_index = argument(4); // char index
1267
1268 src = must_be_not_null(src, true);
1269 tgt = must_be_not_null(tgt, true);
1270
1271 // Multiply byte array index by 2 if String is UTF16 encoded
1272 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1273 src_count = _gvn.transform(new SubINode(src_count, from_index));
1274 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1275 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1276
1277 // Range checks
1278 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1279 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1280 if (stopped()) {
1281 return true;
1282 }
1283
1284 RegionNode* region = new RegionNode(5);
1285 Node* phi = new PhiNode(region, TypeInt::INT);
1286 Node* result = nullptr;
1287
1288 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1289
1290 if (call_opt_stub) {
1291 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1292 StubRoutines::_string_indexof_array[ae],
1293 "stringIndexOf", TypePtr::BOTTOM, src_start,
1294 src_count, tgt_start, tgt_count);
1295 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1296 } else {
1297 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1298 region, phi, ae);
1299 }
1300 if (result != nullptr) {
1301 // The result is index relative to from_index if substring was found, -1 otherwise.
1302 // Generate code which will fold into cmove.
1303 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1304 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1305
1306 Node* if_lt = generate_slow_guard(bol, nullptr);
1307 if (if_lt != nullptr) {
1308 // result == -1
1309 phi->init_req(3, result);
1310 region->init_req(3, if_lt);
1311 }
1312 if (!stopped()) {
1313 result = _gvn.transform(new AddINode(result, from_index));
1314 phi->init_req(4, result);
1315 region->init_req(4, control());
1316 }
1317 }
1318
1319 set_control(_gvn.transform(region));
1320 record_for_igvn(region);
1321 set_result(_gvn.transform(phi));
1322 clear_upper_avx();
1323
1324 return true;
1325 }
1326
1327 // Create StrIndexOfNode with fast path checks
1328 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1329 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1330 // Check for substr count > string count
1331 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1332 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1333 Node* if_gt = generate_slow_guard(bol, nullptr);
1334 if (if_gt != nullptr) {
1335 phi->init_req(1, intcon(-1));
1336 region->init_req(1, if_gt);
1337 }
1338 if (!stopped()) {
1339 // Check for substr count == 0
1340 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1341 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1342 Node* if_zero = generate_slow_guard(bol, nullptr);
1343 if (if_zero != nullptr) {
1344 phi->init_req(2, intcon(0));
1345 region->init_req(2, if_zero);
1346 }
1347 }
1348 if (!stopped()) {
1349 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1350 }
1351 return nullptr;
1352 }
1353
1354 //-----------------------------inline_string_indexOfChar-----------------------
1355 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1356 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1357 return false;
1358 }
1359 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1360 return false;
1361 }
1362 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1363 Node* src = argument(0); // byte[]
1364 Node* int_ch = argument(1);
1365 Node* from_index = argument(2);
1366 Node* max = argument(3);
1367
1368 src = must_be_not_null(src, true);
1369
1370 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1371 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1372 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1373
1374 // Range checks
1375 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1376
1377 // Check for int_ch >= 0
1378 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1379 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1380 {
1381 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1382 uncommon_trap(Deoptimization::Reason_intrinsic,
1383 Deoptimization::Action_maybe_recompile);
1384 }
1385 if (stopped()) {
1386 return true;
1387 }
1388
1389 RegionNode* region = new RegionNode(3);
1390 Node* phi = new PhiNode(region, TypeInt::INT);
1391
1392 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1393 C->set_has_split_ifs(true); // Has chance for split-if optimization
1394 _gvn.transform(result);
1395
1396 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1397 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1398
1399 Node* if_lt = generate_slow_guard(bol, nullptr);
1400 if (if_lt != nullptr) {
1401 // result == -1
1402 phi->init_req(2, result);
1403 region->init_req(2, if_lt);
1404 }
1405 if (!stopped()) {
1406 result = _gvn.transform(new AddINode(result, from_index));
1407 phi->init_req(1, result);
1408 region->init_req(1, control());
1409 }
1410 set_control(_gvn.transform(region));
1411 record_for_igvn(region);
1412 set_result(_gvn.transform(phi));
1413 clear_upper_avx();
1414
1415 return true;
1416 }
1417 //---------------------------inline_string_copy---------------------
1418 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1419 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1420 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1421 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1422 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1423 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1424 bool LibraryCallKit::inline_string_copy(bool compress) {
1425 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1426 return false;
1427 }
1428 int nargs = 5; // 2 oops, 3 ints
1429 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1430
1431 Node* src = argument(0);
1432 Node* src_offset = argument(1);
1433 Node* dst = argument(2);
1434 Node* dst_offset = argument(3);
1435 Node* length = argument(4);
1436
1437 // Check for allocation before we add nodes that would confuse
1438 // tightly_coupled_allocation()
1439 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1440
1441 // Figure out the size and type of the elements we will be copying.
1442 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1443 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1444 if (src_type == nullptr || dst_type == nullptr) {
1445 return false;
1446 }
1447 BasicType src_elem = src_type->elem()->array_element_basic_type();
1448 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1449 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1450 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1451 "Unsupported array types for inline_string_copy");
1452
1453 src = must_be_not_null(src, true);
1454 dst = must_be_not_null(dst, true);
1455
1456 // Convert char[] offsets to byte[] offsets
1457 bool convert_src = (compress && src_elem == T_BYTE);
1458 bool convert_dst = (!compress && dst_elem == T_BYTE);
1459 if (convert_src) {
1460 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1461 } else if (convert_dst) {
1462 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1463 }
1464
1465 // Range checks
1466 generate_string_range_check(src, src_offset, length, convert_src);
1467 generate_string_range_check(dst, dst_offset, length, convert_dst);
1468 if (stopped()) {
1469 return true;
1470 }
1471
1472 Node* src_start = array_element_address(src, src_offset, src_elem);
1473 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1474 // 'src_start' points to src array + scaled offset
1475 // 'dst_start' points to dst array + scaled offset
1476 Node* count = nullptr;
1477 if (compress) {
1478 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1479 } else {
1480 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1481 }
1482
1483 if (alloc != nullptr) {
1484 if (alloc->maybe_set_complete(&_gvn)) {
1485 // "You break it, you buy it."
1486 InitializeNode* init = alloc->initialization();
1487 assert(init->is_complete(), "we just did this");
1488 init->set_complete_with_arraycopy();
1489 assert(dst->is_CheckCastPP(), "sanity");
1490 assert(dst->in(0)->in(0) == init, "dest pinned");
1491 }
1492 // Do not let stores that initialize this object be reordered with
1493 // a subsequent store that would make this object accessible by
1494 // other threads.
1495 // Record what AllocateNode this StoreStore protects so that
1496 // escape analysis can go from the MemBarStoreStoreNode to the
1497 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1498 // based on the escape status of the AllocateNode.
1499 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1500 }
1501 if (compress) {
1502 set_result(_gvn.transform(count));
1503 }
1504 clear_upper_avx();
1505
1506 return true;
1507 }
1508
1509 #ifdef _LP64
1510 #define XTOP ,top() /*additional argument*/
1511 #else //_LP64
1512 #define XTOP /*no additional argument*/
1513 #endif //_LP64
1514
1515 //------------------------inline_string_toBytesU--------------------------
1516 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1517 bool LibraryCallKit::inline_string_toBytesU() {
1518 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1519 return false;
1520 }
1521 // Get the arguments.
1522 Node* value = argument(0);
1523 Node* offset = argument(1);
1524 Node* length = argument(2);
1525
1526 Node* newcopy = nullptr;
1527
1528 // Set the original stack and the reexecute bit for the interpreter to reexecute
1529 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1530 { PreserveReexecuteState preexecs(this);
1531 jvms()->set_should_reexecute(true);
1532
1533 // Check if a null path was taken unconditionally.
1534 value = null_check(value);
1535
1536 RegionNode* bailout = new RegionNode(1);
1537 record_for_igvn(bailout);
1538
1539 // Range checks
1540 generate_negative_guard(offset, bailout);
1541 generate_negative_guard(length, bailout);
1542 generate_limit_guard(offset, length, load_array_length(value), bailout);
1543 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1544 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1545
1546 if (bailout->req() > 1) {
1547 PreserveJVMState pjvms(this);
1548 set_control(_gvn.transform(bailout));
1549 uncommon_trap(Deoptimization::Reason_intrinsic,
1550 Deoptimization::Action_maybe_recompile);
1551 }
1552 if (stopped()) {
1553 return true;
1554 }
1555
1556 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1557 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1558 newcopy = new_array(klass_node, size, 0); // no arguments to push
1559 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1560 guarantee(alloc != nullptr, "created above");
1561
1562 // Calculate starting addresses.
1563 Node* src_start = array_element_address(value, offset, T_CHAR);
1564 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1565
1566 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1567 const TypeInt* toffset = gvn().type(offset)->is_int();
1568 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1569
1570 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1571 const char* copyfunc_name = "arraycopy";
1572 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1573 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1574 OptoRuntime::fast_arraycopy_Type(),
1575 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1576 src_start, dst_start, ConvI2X(length) XTOP);
1577 // Do not let reads from the cloned object float above the arraycopy.
1578 if (alloc->maybe_set_complete(&_gvn)) {
1579 // "You break it, you buy it."
1580 InitializeNode* init = alloc->initialization();
1581 assert(init->is_complete(), "we just did this");
1582 init->set_complete_with_arraycopy();
1583 assert(newcopy->is_CheckCastPP(), "sanity");
1584 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1585 }
1586 // Do not let stores that initialize this object be reordered with
1587 // a subsequent store that would make this object accessible by
1588 // other threads.
1589 // Record what AllocateNode this StoreStore protects so that
1590 // escape analysis can go from the MemBarStoreStoreNode to the
1591 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1592 // based on the escape status of the AllocateNode.
1593 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1594 } // original reexecute is set back here
1595
1596 C->set_has_split_ifs(true); // Has chance for split-if optimization
1597 if (!stopped()) {
1598 set_result(newcopy);
1599 }
1600 clear_upper_avx();
1601
1602 return true;
1603 }
1604
1605 //------------------------inline_string_getCharsU--------------------------
1606 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1607 bool LibraryCallKit::inline_string_getCharsU() {
1608 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1609 return false;
1610 }
1611
1612 // Get the arguments.
1613 Node* src = argument(0);
1614 Node* src_begin = argument(1);
1615 Node* src_end = argument(2); // exclusive offset (i < src_end)
1616 Node* dst = argument(3);
1617 Node* dst_begin = argument(4);
1618
1619 // Check for allocation before we add nodes that would confuse
1620 // tightly_coupled_allocation()
1621 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1622
1623 // Check if a null path was taken unconditionally.
1624 src = null_check(src);
1625 dst = null_check(dst);
1626 if (stopped()) {
1627 return true;
1628 }
1629
1630 // Get length and convert char[] offset to byte[] offset
1631 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1632 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1633
1634 // Range checks
1635 generate_string_range_check(src, src_begin, length, true);
1636 generate_string_range_check(dst, dst_begin, length, false);
1637 if (stopped()) {
1638 return true;
1639 }
1640
1641 if (!stopped()) {
1642 // Calculate starting addresses.
1643 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1644 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1645
1646 // Check if array addresses are aligned to HeapWordSize
1647 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1648 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1649 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1650 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1651
1652 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1653 const char* copyfunc_name = "arraycopy";
1654 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1655 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1656 OptoRuntime::fast_arraycopy_Type(),
1657 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1658 src_start, dst_start, ConvI2X(length) XTOP);
1659 // Do not let reads from the cloned object float above the arraycopy.
1660 if (alloc != nullptr) {
1661 if (alloc->maybe_set_complete(&_gvn)) {
1662 // "You break it, you buy it."
1663 InitializeNode* init = alloc->initialization();
1664 assert(init->is_complete(), "we just did this");
1665 init->set_complete_with_arraycopy();
1666 assert(dst->is_CheckCastPP(), "sanity");
1667 assert(dst->in(0)->in(0) == init, "dest pinned");
1668 }
1669 // Do not let stores that initialize this object be reordered with
1670 // a subsequent store that would make this object accessible by
1671 // other threads.
1672 // Record what AllocateNode this StoreStore protects so that
1673 // escape analysis can go from the MemBarStoreStoreNode to the
1674 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1675 // based on the escape status of the AllocateNode.
1676 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1677 } else {
1678 insert_mem_bar(Op_MemBarCPUOrder);
1679 }
1680 }
1681
1682 C->set_has_split_ifs(true); // Has chance for split-if optimization
1683 return true;
1684 }
1685
1686 //----------------------inline_string_char_access----------------------------
1687 // Store/Load char to/from byte[] array.
1688 // static void StringUTF16.putChar(byte[] val, int index, int c)
1689 // static char StringUTF16.getChar(byte[] val, int index)
1690 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1691 Node* value = argument(0);
1692 Node* index = argument(1);
1693 Node* ch = is_store ? argument(2) : nullptr;
1694
1695 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1696 // correctly requires matched array shapes.
1697 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1698 "sanity: byte[] and char[] bases agree");
1699 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1700 "sanity: byte[] and char[] scales agree");
1701
1702 // Bail when getChar over constants is requested: constant folding would
1703 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1704 // Java method would constant fold nicely instead.
1705 if (!is_store && value->is_Con() && index->is_Con()) {
1706 return false;
1707 }
1708
1709 // Save state and restore on bailout
1710 uint old_sp = sp();
1711 SafePointNode* old_map = clone_map();
1712
1713 value = must_be_not_null(value, true);
1714
1715 Node* adr = array_element_address(value, index, T_CHAR);
1716 if (adr->is_top()) {
1717 set_map(old_map);
1718 set_sp(old_sp);
1719 return false;
1720 }
1721 destruct_map_clone(old_map);
1722 if (is_store) {
1723 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1724 } else {
1725 ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD | C2_UNKNOWN_CONTROL_LOAD);
1726 set_result(ch);
1727 }
1728 return true;
1729 }
1730
1731 //--------------------------round_double_node--------------------------------
1732 // Round a double node if necessary.
1733 Node* LibraryCallKit::round_double_node(Node* n) {
1734 if (Matcher::strict_fp_requires_explicit_rounding) {
1735 #ifdef IA32
1736 if (UseSSE < 2) {
1737 n = _gvn.transform(new RoundDoubleNode(nullptr, n));
1738 }
1739 #else
1740 Unimplemented();
1741 #endif // IA32
1742 }
1743 return n;
1744 }
1745
1746 //------------------------------inline_math-----------------------------------
1747 // public static double Math.abs(double)
1748 // public static double Math.sqrt(double)
1749 // public static double Math.log(double)
1750 // public static double Math.log10(double)
1751 // public static double Math.round(double)
1752 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1753 Node* arg = round_double_node(argument(0));
1754 Node* n = nullptr;
1755 switch (id) {
1756 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1757 case vmIntrinsics::_dsqrt:
1758 case vmIntrinsics::_dsqrt_strict:
1759 n = new SqrtDNode(C, control(), arg); break;
1760 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1761 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1762 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1763 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1764 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1765 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1766 default: fatal_unexpected_iid(id); break;
1767 }
1768 set_result(_gvn.transform(n));
1769 return true;
1770 }
1771
1772 //------------------------------inline_math-----------------------------------
1773 // public static float Math.abs(float)
1774 // public static int Math.abs(int)
1775 // public static long Math.abs(long)
1776 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1777 Node* arg = argument(0);
1778 Node* n = nullptr;
1779 switch (id) {
1780 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1781 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1782 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1783 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1784 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1785 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1786 default: fatal_unexpected_iid(id); break;
1787 }
1788 set_result(_gvn.transform(n));
1789 return true;
1790 }
1791
1792 //------------------------------runtime_math-----------------------------
1793 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1794 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1795 "must be (DD)D or (D)D type");
1796
1797 // Inputs
1798 Node* a = round_double_node(argument(0));
1799 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : nullptr;
1800
1801 const TypePtr* no_memory_effects = nullptr;
1802 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1803 no_memory_effects,
1804 a, top(), b, b ? top() : nullptr);
1805 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1806 #ifdef ASSERT
1807 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1808 assert(value_top == top(), "second value must be top");
1809 #endif
1810
1811 set_result(value);
1812 return true;
1813 }
1814
1815 //------------------------------inline_math_pow-----------------------------
1816 bool LibraryCallKit::inline_math_pow() {
1817 Node* exp = round_double_node(argument(2));
1818 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1819 if (d != nullptr) {
1820 if (d->getd() == 2.0) {
1821 // Special case: pow(x, 2.0) => x * x
1822 Node* base = round_double_node(argument(0));
1823 set_result(_gvn.transform(new MulDNode(base, base)));
1824 return true;
1825 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1826 // Special case: pow(x, 0.5) => sqrt(x)
1827 Node* base = round_double_node(argument(0));
1828 Node* zero = _gvn.zerocon(T_DOUBLE);
1829
1830 RegionNode* region = new RegionNode(3);
1831 Node* phi = new PhiNode(region, Type::DOUBLE);
1832
1833 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1834 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1835 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1836 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1837 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1838
1839 Node* if_pow = generate_slow_guard(test, nullptr);
1840 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1841 phi->init_req(1, value_sqrt);
1842 region->init_req(1, control());
1843
1844 if (if_pow != nullptr) {
1845 set_control(if_pow);
1846 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1847 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1848 const TypePtr* no_memory_effects = nullptr;
1849 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1850 no_memory_effects, base, top(), exp, top());
1851 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1852 #ifdef ASSERT
1853 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1854 assert(value_top == top(), "second value must be top");
1855 #endif
1856 phi->init_req(2, value_pow);
1857 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1858 }
1859
1860 C->set_has_split_ifs(true); // Has chance for split-if optimization
1861 set_control(_gvn.transform(region));
1862 record_for_igvn(region);
1863 set_result(_gvn.transform(phi));
1864
1865 return true;
1866 }
1867 }
1868
1869 return StubRoutines::dpow() != nullptr ?
1870 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1871 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1872 }
1873
1874 //------------------------------inline_math_native-----------------------------
1875 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1876 switch (id) {
1877 case vmIntrinsics::_dsin:
1878 return StubRoutines::dsin() != nullptr ?
1879 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1880 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1881 case vmIntrinsics::_dcos:
1882 return StubRoutines::dcos() != nullptr ?
1883 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1884 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1885 case vmIntrinsics::_dtan:
1886 return StubRoutines::dtan() != nullptr ?
1887 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1888 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1889 case vmIntrinsics::_dtanh:
1890 return StubRoutines::dtanh() != nullptr ?
1891 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1892 case vmIntrinsics::_dexp:
1893 return StubRoutines::dexp() != nullptr ?
1894 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1895 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1896 case vmIntrinsics::_dlog:
1897 return StubRoutines::dlog() != nullptr ?
1898 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1899 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1900 case vmIntrinsics::_dlog10:
1901 return StubRoutines::dlog10() != nullptr ?
1902 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1903 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1904
1905 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1906 case vmIntrinsics::_ceil:
1907 case vmIntrinsics::_floor:
1908 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1909
1910 case vmIntrinsics::_dsqrt:
1911 case vmIntrinsics::_dsqrt_strict:
1912 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1913 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1914 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1915 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1916 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1917
1918 case vmIntrinsics::_dpow: return inline_math_pow();
1919 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1920 case vmIntrinsics::_fcopySign: return inline_math(id);
1921 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1922 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1923 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1924
1925 // These intrinsics are not yet correctly implemented
1926 case vmIntrinsics::_datan2:
1927 return false;
1928
1929 default:
1930 fatal_unexpected_iid(id);
1931 return false;
1932 }
1933 }
1934
1935 //----------------------------inline_notify-----------------------------------*
1936 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1937 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1938 address func;
1939 if (id == vmIntrinsics::_notify) {
1940 func = OptoRuntime::monitor_notify_Java();
1941 } else {
1942 func = OptoRuntime::monitor_notifyAll_Java();
1943 }
1944 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1945 make_slow_call_ex(call, env()->Throwable_klass(), false);
1946 return true;
1947 }
1948
1949
1950 //----------------------------inline_min_max-----------------------------------
1951 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1952 set_result(generate_min_max(id, argument(0), argument(1)));
1953 return true;
1954 }
1955
1956 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1957 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1958 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1959 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1960 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1961
1962 {
1963 PreserveJVMState pjvms(this);
1964 PreserveReexecuteState preexecs(this);
1965 jvms()->set_should_reexecute(true);
1966
1967 set_control(slow_path);
1968 set_i_o(i_o());
1969
1970 uncommon_trap(Deoptimization::Reason_intrinsic,
1971 Deoptimization::Action_none);
1972 }
1973
1974 set_control(fast_path);
1975 set_result(math);
1976 }
1977
1978 template <typename OverflowOp>
1979 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1980 typedef typename OverflowOp::MathOp MathOp;
1981
1982 MathOp* mathOp = new MathOp(arg1, arg2);
1983 Node* operation = _gvn.transform( mathOp );
1984 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1985 inline_math_mathExact(operation, ofcheck);
1986 return true;
1987 }
1988
1989 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1990 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1991 }
1992
1993 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1994 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1995 }
1996
1997 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1998 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1999 }
2000
2001 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2002 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2003 }
2004
2005 bool LibraryCallKit::inline_math_negateExactI() {
2006 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2007 }
2008
2009 bool LibraryCallKit::inline_math_negateExactL() {
2010 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2011 }
2012
2013 bool LibraryCallKit::inline_math_multiplyExactI() {
2014 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2015 }
2016
2017 bool LibraryCallKit::inline_math_multiplyExactL() {
2018 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2019 }
2020
2021 bool LibraryCallKit::inline_math_multiplyHigh() {
2022 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2023 return true;
2024 }
2025
2026 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2027 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2028 return true;
2029 }
2030
2031 Node*
2032 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
2033 Node* result_val = nullptr;
2034 switch (id) {
2035 case vmIntrinsics::_min:
2036 case vmIntrinsics::_min_strict:
2037 result_val = _gvn.transform(new MinINode(x0, y0));
2038 break;
2039 case vmIntrinsics::_max:
2040 case vmIntrinsics::_max_strict:
2041 result_val = _gvn.transform(new MaxINode(x0, y0));
2042 break;
2043 default:
2044 fatal_unexpected_iid(id);
2045 break;
2046 }
2047 return result_val;
2048 }
2049
2050 inline int
2051 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2052 const TypePtr* base_type = TypePtr::NULL_PTR;
2053 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2054 if (base_type == nullptr) {
2055 // Unknown type.
2056 return Type::AnyPtr;
2057 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2058 // Since this is a null+long form, we have to switch to a rawptr.
2059 base = _gvn.transform(new CastX2PNode(offset));
2060 offset = MakeConX(0);
2061 return Type::RawPtr;
2062 } else if (base_type->base() == Type::RawPtr) {
2063 return Type::RawPtr;
2064 } else if (base_type->isa_oopptr()) {
2065 // Base is never null => always a heap address.
2066 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2067 return Type::OopPtr;
2068 }
2069 // Offset is small => always a heap address.
2070 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2071 if (offset_type != nullptr &&
2072 base_type->offset() == 0 && // (should always be?)
2073 offset_type->_lo >= 0 &&
2074 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2075 return Type::OopPtr;
2076 } else if (type == T_OBJECT) {
2077 // off heap access to an oop doesn't make any sense. Has to be on
2078 // heap.
2079 return Type::OopPtr;
2080 }
2081 // Otherwise, it might either be oop+off or null+addr.
2082 return Type::AnyPtr;
2083 } else {
2084 // No information:
2085 return Type::AnyPtr;
2086 }
2087 }
2088
2089 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2090 Node* uncasted_base = base;
2091 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2092 if (kind == Type::RawPtr) {
2093 return basic_plus_adr(top(), uncasted_base, offset);
2094 } else if (kind == Type::AnyPtr) {
2095 assert(base == uncasted_base, "unexpected base change");
2096 if (can_cast) {
2097 if (!_gvn.type(base)->speculative_maybe_null() &&
2098 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2099 // According to profiling, this access is always on
2100 // heap. Casting the base to not null and thus avoiding membars
2101 // around the access should allow better optimizations
2102 Node* null_ctl = top();
2103 base = null_check_oop(base, &null_ctl, true, true, true);
2104 assert(null_ctl->is_top(), "no null control here");
2105 return basic_plus_adr(base, offset);
2106 } else if (_gvn.type(base)->speculative_always_null() &&
2107 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2108 // According to profiling, this access is always off
2109 // heap.
2110 base = null_assert(base);
2111 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2112 offset = MakeConX(0);
2113 return basic_plus_adr(top(), raw_base, offset);
2114 }
2115 }
2116 // We don't know if it's an on heap or off heap access. Fall back
2117 // to raw memory access.
2118 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2119 return basic_plus_adr(top(), raw, offset);
2120 } else {
2121 assert(base == uncasted_base, "unexpected base change");
2122 // We know it's an on heap access so base can't be null
2123 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2124 base = must_be_not_null(base, true);
2125 }
2126 return basic_plus_adr(base, offset);
2127 }
2128 }
2129
2130 //--------------------------inline_number_methods-----------------------------
2131 // inline int Integer.numberOfLeadingZeros(int)
2132 // inline int Long.numberOfLeadingZeros(long)
2133 //
2134 // inline int Integer.numberOfTrailingZeros(int)
2135 // inline int Long.numberOfTrailingZeros(long)
2136 //
2137 // inline int Integer.bitCount(int)
2138 // inline int Long.bitCount(long)
2139 //
2140 // inline char Character.reverseBytes(char)
2141 // inline short Short.reverseBytes(short)
2142 // inline int Integer.reverseBytes(int)
2143 // inline long Long.reverseBytes(long)
2144 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2145 Node* arg = argument(0);
2146 Node* n = nullptr;
2147 switch (id) {
2148 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2149 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2150 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2151 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2152 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2153 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2154 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(nullptr, arg); break;
2155 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( nullptr, arg); break;
2156 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( nullptr, arg); break;
2157 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( nullptr, arg); break;
2158 case vmIntrinsics::_reverse_i: n = new ReverseINode(nullptr, arg); break;
2159 case vmIntrinsics::_reverse_l: n = new ReverseLNode(nullptr, arg); break;
2160 default: fatal_unexpected_iid(id); break;
2161 }
2162 set_result(_gvn.transform(n));
2163 return true;
2164 }
2165
2166 //--------------------------inline_bitshuffle_methods-----------------------------
2167 // inline int Integer.compress(int, int)
2168 // inline int Integer.expand(int, int)
2169 // inline long Long.compress(long, long)
2170 // inline long Long.expand(long, long)
2171 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2172 Node* n = nullptr;
2173 switch (id) {
2174 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2175 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2176 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2177 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2178 default: fatal_unexpected_iid(id); break;
2179 }
2180 set_result(_gvn.transform(n));
2181 return true;
2182 }
2183
2184 //--------------------------inline_number_methods-----------------------------
2185 // inline int Integer.compareUnsigned(int, int)
2186 // inline int Long.compareUnsigned(long, long)
2187 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2188 Node* arg1 = argument(0);
2189 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2190 Node* n = nullptr;
2191 switch (id) {
2192 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2193 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2194 default: fatal_unexpected_iid(id); break;
2195 }
2196 set_result(_gvn.transform(n));
2197 return true;
2198 }
2199
2200 //--------------------------inline_unsigned_divmod_methods-----------------------------
2201 // inline int Integer.divideUnsigned(int, int)
2202 // inline int Integer.remainderUnsigned(int, int)
2203 // inline long Long.divideUnsigned(long, long)
2204 // inline long Long.remainderUnsigned(long, long)
2205 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2206 Node* n = nullptr;
2207 switch (id) {
2208 case vmIntrinsics::_divideUnsigned_i: {
2209 zero_check_int(argument(1));
2210 // Compile-time detect of null-exception
2211 if (stopped()) {
2212 return true; // keep the graph constructed so far
2213 }
2214 n = new UDivINode(control(), argument(0), argument(1));
2215 break;
2216 }
2217 case vmIntrinsics::_divideUnsigned_l: {
2218 zero_check_long(argument(2));
2219 // Compile-time detect of null-exception
2220 if (stopped()) {
2221 return true; // keep the graph constructed so far
2222 }
2223 n = new UDivLNode(control(), argument(0), argument(2));
2224 break;
2225 }
2226 case vmIntrinsics::_remainderUnsigned_i: {
2227 zero_check_int(argument(1));
2228 // Compile-time detect of null-exception
2229 if (stopped()) {
2230 return true; // keep the graph constructed so far
2231 }
2232 n = new UModINode(control(), argument(0), argument(1));
2233 break;
2234 }
2235 case vmIntrinsics::_remainderUnsigned_l: {
2236 zero_check_long(argument(2));
2237 // Compile-time detect of null-exception
2238 if (stopped()) {
2239 return true; // keep the graph constructed so far
2240 }
2241 n = new UModLNode(control(), argument(0), argument(2));
2242 break;
2243 }
2244 default: fatal_unexpected_iid(id); break;
2245 }
2246 set_result(_gvn.transform(n));
2247 return true;
2248 }
2249
2250 //----------------------------inline_unsafe_access----------------------------
2251
2252 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2253 // Attempt to infer a sharper value type from the offset and base type.
2254 ciKlass* sharpened_klass = nullptr;
2255
2256 // See if it is an instance field, with an object type.
2257 if (alias_type->field() != nullptr) {
2258 if (alias_type->field()->type()->is_klass()) {
2259 sharpened_klass = alias_type->field()->type()->as_klass();
2260 }
2261 }
2262
2263 const TypeOopPtr* result = nullptr;
2264 // See if it is a narrow oop array.
2265 if (adr_type->isa_aryptr()) {
2266 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2267 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2268 if (elem_type != nullptr && elem_type->is_loaded()) {
2269 // Sharpen the value type.
2270 result = elem_type;
2271 }
2272 }
2273 }
2274
2275 // The sharpened class might be unloaded if there is no class loader
2276 // contraint in place.
2277 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2278 // Sharpen the value type.
2279 result = TypeOopPtr::make_from_klass(sharpened_klass);
2280 }
2281 if (result != nullptr) {
2282 #ifndef PRODUCT
2283 if (C->print_intrinsics() || C->print_inlining()) {
2284 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2285 tty->print(" sharpened value: "); result->dump(); tty->cr();
2286 }
2287 #endif
2288 }
2289 return result;
2290 }
2291
2292 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2293 switch (kind) {
2294 case Relaxed:
2295 return MO_UNORDERED;
2296 case Opaque:
2297 return MO_RELAXED;
2298 case Acquire:
2299 return MO_ACQUIRE;
2300 case Release:
2301 return MO_RELEASE;
2302 case Volatile:
2303 return MO_SEQ_CST;
2304 default:
2305 ShouldNotReachHere();
2306 return 0;
2307 }
2308 }
2309
2310 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2311 if (callee()->is_static()) return false; // caller must have the capability!
2312 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2313 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2314 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2315 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2316
2317 if (is_reference_type(type)) {
2318 decorators |= ON_UNKNOWN_OOP_REF;
2319 }
2320
2321 if (unaligned) {
2322 decorators |= C2_UNALIGNED;
2323 }
2324
2325 #ifndef PRODUCT
2326 {
2327 ResourceMark rm;
2328 // Check the signatures.
2329 ciSignature* sig = callee()->signature();
2330 #ifdef ASSERT
2331 if (!is_store) {
2332 // Object getReference(Object base, int/long offset), etc.
2333 BasicType rtype = sig->return_type()->basic_type();
2334 assert(rtype == type, "getter must return the expected value");
2335 assert(sig->count() == 2, "oop getter has 2 arguments");
2336 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2337 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2338 } else {
2339 // void putReference(Object base, int/long offset, Object x), etc.
2340 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2341 assert(sig->count() == 3, "oop putter has 3 arguments");
2342 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2343 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2344 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2345 assert(vtype == type, "putter must accept the expected value");
2346 }
2347 #endif // ASSERT
2348 }
2349 #endif //PRODUCT
2350
2351 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2352
2353 Node* receiver = argument(0); // type: oop
2354
2355 // Build address expression.
2356 Node* heap_base_oop = top();
2357
2358 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2359 Node* base = argument(1); // type: oop
2360 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2361 Node* offset = argument(2); // type: long
2362 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2363 // to be plain byte offsets, which are also the same as those accepted
2364 // by oopDesc::field_addr.
2365 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2366 "fieldOffset must be byte-scaled");
2367 // 32-bit machines ignore the high half!
2368 offset = ConvL2X(offset);
2369
2370 // Save state and restore on bailout
2371 uint old_sp = sp();
2372 SafePointNode* old_map = clone_map();
2373
2374 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2375 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2376
2377 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2378 if (type != T_OBJECT) {
2379 decorators |= IN_NATIVE; // off-heap primitive access
2380 } else {
2381 set_map(old_map);
2382 set_sp(old_sp);
2383 return false; // off-heap oop accesses are not supported
2384 }
2385 } else {
2386 heap_base_oop = base; // on-heap or mixed access
2387 }
2388
2389 // Can base be null? Otherwise, always on-heap access.
2390 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2391
2392 if (!can_access_non_heap) {
2393 decorators |= IN_HEAP;
2394 }
2395
2396 Node* val = is_store ? argument(4) : nullptr;
2397
2398 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2399 if (adr_type == TypePtr::NULL_PTR) {
2400 set_map(old_map);
2401 set_sp(old_sp);
2402 return false; // off-heap access with zero address
2403 }
2404
2405 // Try to categorize the address.
2406 Compile::AliasType* alias_type = C->alias_type(adr_type);
2407 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2408
2409 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2410 alias_type->adr_type() == TypeAryPtr::RANGE) {
2411 set_map(old_map);
2412 set_sp(old_sp);
2413 return false; // not supported
2414 }
2415
2416 bool mismatched = false;
2417 BasicType bt = alias_type->basic_type();
2418 if (bt != T_ILLEGAL) {
2419 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2420 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2421 // Alias type doesn't differentiate between byte[] and boolean[]).
2422 // Use address type to get the element type.
2423 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2424 }
2425 if (is_reference_type(bt, true)) {
2426 // accessing an array field with getReference is not a mismatch
2427 bt = T_OBJECT;
2428 }
2429 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2430 // Don't intrinsify mismatched object accesses
2431 set_map(old_map);
2432 set_sp(old_sp);
2433 return false;
2434 }
2435 mismatched = (bt != type);
2436 } else if (alias_type->adr_type()->isa_oopptr()) {
2437 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2438 }
2439
2440 destruct_map_clone(old_map);
2441 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2442
2443 if (mismatched) {
2444 decorators |= C2_MISMATCHED;
2445 }
2446
2447 // First guess at the value type.
2448 const Type *value_type = Type::get_const_basic_type(type);
2449
2450 // Figure out the memory ordering.
2451 decorators |= mo_decorator_for_access_kind(kind);
2452
2453 if (!is_store && type == T_OBJECT) {
2454 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2455 if (tjp != nullptr) {
2456 value_type = tjp;
2457 }
2458 }
2459
2460 receiver = null_check(receiver);
2461 if (stopped()) {
2462 return true;
2463 }
2464 // Heap pointers get a null-check from the interpreter,
2465 // as a courtesy. However, this is not guaranteed by Unsafe,
2466 // and it is not possible to fully distinguish unintended nulls
2467 // from intended ones in this API.
2468
2469 if (!is_store) {
2470 Node* p = nullptr;
2471 // Try to constant fold a load from a constant field
2472 ciField* field = alias_type->field();
2473 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) {
2474 // final or stable field
2475 p = make_constant_from_field(field, heap_base_oop);
2476 }
2477
2478 if (p == nullptr) { // Could not constant fold the load
2479 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2480 // Normalize the value returned by getBoolean in the following cases
2481 if (type == T_BOOLEAN &&
2482 (mismatched ||
2483 heap_base_oop == top() || // - heap_base_oop is null or
2484 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2485 // and the unsafe access is made to large offset
2486 // (i.e., larger than the maximum offset necessary for any
2487 // field access)
2488 ) {
2489 IdealKit ideal = IdealKit(this);
2490 #define __ ideal.
2491 IdealVariable normalized_result(ideal);
2492 __ declarations_done();
2493 __ set(normalized_result, p);
2494 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2495 __ set(normalized_result, ideal.ConI(1));
2496 ideal.end_if();
2497 final_sync(ideal);
2498 p = __ value(normalized_result);
2499 #undef __
2500 }
2501 }
2502 if (type == T_ADDRESS) {
2503 p = gvn().transform(new CastP2XNode(nullptr, p));
2504 p = ConvX2UL(p);
2505 }
2506 // The load node has the control of the preceding MemBarCPUOrder. All
2507 // following nodes will have the control of the MemBarCPUOrder inserted at
2508 // the end of this method. So, pushing the load onto the stack at a later
2509 // point is fine.
2510 set_result(p);
2511 } else {
2512 if (bt == T_ADDRESS) {
2513 // Repackage the long as a pointer.
2514 val = ConvL2X(val);
2515 val = gvn().transform(new CastX2PNode(val));
2516 }
2517 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2518 }
2519
2520 return true;
2521 }
2522
2523 //----------------------------inline_unsafe_load_store----------------------------
2524 // This method serves a couple of different customers (depending on LoadStoreKind):
2525 //
2526 // LS_cmp_swap:
2527 //
2528 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2529 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2530 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2531 //
2532 // LS_cmp_swap_weak:
2533 //
2534 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2535 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2536 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2537 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2538 //
2539 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2540 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2541 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2542 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2543 //
2544 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2545 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2546 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2547 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2548 //
2549 // LS_cmp_exchange:
2550 //
2551 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2552 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2553 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2554 //
2555 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2556 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2557 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2558 //
2559 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2560 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2561 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2562 //
2563 // LS_get_add:
2564 //
2565 // int getAndAddInt( Object o, long offset, int delta)
2566 // long getAndAddLong(Object o, long offset, long delta)
2567 //
2568 // LS_get_set:
2569 //
2570 // int getAndSet(Object o, long offset, int newValue)
2571 // long getAndSet(Object o, long offset, long newValue)
2572 // Object getAndSet(Object o, long offset, Object newValue)
2573 //
2574 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2575 // This basic scheme here is the same as inline_unsafe_access, but
2576 // differs in enough details that combining them would make the code
2577 // overly confusing. (This is a true fact! I originally combined
2578 // them, but even I was confused by it!) As much code/comments as
2579 // possible are retained from inline_unsafe_access though to make
2580 // the correspondences clearer. - dl
2581
2582 if (callee()->is_static()) return false; // caller must have the capability!
2583
2584 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2585 decorators |= mo_decorator_for_access_kind(access_kind);
2586
2587 #ifndef PRODUCT
2588 BasicType rtype;
2589 {
2590 ResourceMark rm;
2591 // Check the signatures.
2592 ciSignature* sig = callee()->signature();
2593 rtype = sig->return_type()->basic_type();
2594 switch(kind) {
2595 case LS_get_add:
2596 case LS_get_set: {
2597 // Check the signatures.
2598 #ifdef ASSERT
2599 assert(rtype == type, "get and set must return the expected type");
2600 assert(sig->count() == 3, "get and set has 3 arguments");
2601 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2602 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2603 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2604 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2605 #endif // ASSERT
2606 break;
2607 }
2608 case LS_cmp_swap:
2609 case LS_cmp_swap_weak: {
2610 // Check the signatures.
2611 #ifdef ASSERT
2612 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2613 assert(sig->count() == 4, "CAS has 4 arguments");
2614 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2615 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2616 #endif // ASSERT
2617 break;
2618 }
2619 case LS_cmp_exchange: {
2620 // Check the signatures.
2621 #ifdef ASSERT
2622 assert(rtype == type, "CAS must return the expected type");
2623 assert(sig->count() == 4, "CAS has 4 arguments");
2624 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2625 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2626 #endif // ASSERT
2627 break;
2628 }
2629 default:
2630 ShouldNotReachHere();
2631 }
2632 }
2633 #endif //PRODUCT
2634
2635 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2636
2637 // Get arguments:
2638 Node* receiver = nullptr;
2639 Node* base = nullptr;
2640 Node* offset = nullptr;
2641 Node* oldval = nullptr;
2642 Node* newval = nullptr;
2643 switch(kind) {
2644 case LS_cmp_swap:
2645 case LS_cmp_swap_weak:
2646 case LS_cmp_exchange: {
2647 const bool two_slot_type = type2size[type] == 2;
2648 receiver = argument(0); // type: oop
2649 base = argument(1); // type: oop
2650 offset = argument(2); // type: long
2651 oldval = argument(4); // type: oop, int, or long
2652 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2653 break;
2654 }
2655 case LS_get_add:
2656 case LS_get_set: {
2657 receiver = argument(0); // type: oop
2658 base = argument(1); // type: oop
2659 offset = argument(2); // type: long
2660 oldval = nullptr;
2661 newval = argument(4); // type: oop, int, or long
2662 break;
2663 }
2664 default:
2665 ShouldNotReachHere();
2666 }
2667
2668 // Build field offset expression.
2669 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2670 // to be plain byte offsets, which are also the same as those accepted
2671 // by oopDesc::field_addr.
2672 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2673 // 32-bit machines ignore the high half of long offsets
2674 offset = ConvL2X(offset);
2675 // Save state and restore on bailout
2676 uint old_sp = sp();
2677 SafePointNode* old_map = clone_map();
2678 Node* adr = make_unsafe_address(base, offset,type, false);
2679 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2680
2681 Compile::AliasType* alias_type = C->alias_type(adr_type);
2682 BasicType bt = alias_type->basic_type();
2683 if (bt != T_ILLEGAL &&
2684 (is_reference_type(bt) != (type == T_OBJECT))) {
2685 // Don't intrinsify mismatched object accesses.
2686 set_map(old_map);
2687 set_sp(old_sp);
2688 return false;
2689 }
2690
2691 destruct_map_clone(old_map);
2692
2693 // For CAS, unlike inline_unsafe_access, there seems no point in
2694 // trying to refine types. Just use the coarse types here.
2695 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2696 const Type *value_type = Type::get_const_basic_type(type);
2697
2698 switch (kind) {
2699 case LS_get_set:
2700 case LS_cmp_exchange: {
2701 if (type == T_OBJECT) {
2702 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2703 if (tjp != nullptr) {
2704 value_type = tjp;
2705 }
2706 }
2707 break;
2708 }
2709 case LS_cmp_swap:
2710 case LS_cmp_swap_weak:
2711 case LS_get_add:
2712 break;
2713 default:
2714 ShouldNotReachHere();
2715 }
2716
2717 // Null check receiver.
2718 receiver = null_check(receiver);
2719 if (stopped()) {
2720 return true;
2721 }
2722
2723 int alias_idx = C->get_alias_index(adr_type);
2724
2725 if (is_reference_type(type)) {
2726 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2727
2728 // Transformation of a value which could be null pointer (CastPP #null)
2729 // could be delayed during Parse (for example, in adjust_map_after_if()).
2730 // Execute transformation here to avoid barrier generation in such case.
2731 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2732 newval = _gvn.makecon(TypePtr::NULL_PTR);
2733
2734 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2735 // Refine the value to a null constant, when it is known to be null
2736 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2737 }
2738 }
2739
2740 Node* result = nullptr;
2741 switch (kind) {
2742 case LS_cmp_exchange: {
2743 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2744 oldval, newval, value_type, type, decorators);
2745 break;
2746 }
2747 case LS_cmp_swap_weak:
2748 decorators |= C2_WEAK_CMPXCHG;
2749 case LS_cmp_swap: {
2750 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2751 oldval, newval, value_type, type, decorators);
2752 break;
2753 }
2754 case LS_get_set: {
2755 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2756 newval, value_type, type, decorators);
2757 break;
2758 }
2759 case LS_get_add: {
2760 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2761 newval, value_type, type, decorators);
2762 break;
2763 }
2764 default:
2765 ShouldNotReachHere();
2766 }
2767
2768 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2769 set_result(result);
2770 return true;
2771 }
2772
2773 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2774 // Regardless of form, don't allow previous ld/st to move down,
2775 // then issue acquire, release, or volatile mem_bar.
2776 insert_mem_bar(Op_MemBarCPUOrder);
2777 switch(id) {
2778 case vmIntrinsics::_loadFence:
2779 insert_mem_bar(Op_LoadFence);
2780 return true;
2781 case vmIntrinsics::_storeFence:
2782 insert_mem_bar(Op_StoreFence);
2783 return true;
2784 case vmIntrinsics::_storeStoreFence:
2785 insert_mem_bar(Op_StoreStoreFence);
2786 return true;
2787 case vmIntrinsics::_fullFence:
2788 insert_mem_bar(Op_MemBarVolatile);
2789 return true;
2790 default:
2791 fatal_unexpected_iid(id);
2792 return false;
2793 }
2794 }
2795
2796 bool LibraryCallKit::inline_onspinwait() {
2797 insert_mem_bar(Op_OnSpinWait);
2798 return true;
2799 }
2800
2801 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2802 if (!kls->is_Con()) {
2803 return true;
2804 }
2805 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2806 if (klsptr == nullptr) {
2807 return true;
2808 }
2809 ciInstanceKlass* ik = klsptr->instance_klass();
2810 // don't need a guard for a klass that is already initialized
2811 return !ik->is_initialized();
2812 }
2813
2814 //----------------------------inline_unsafe_writeback0-------------------------
2815 // public native void Unsafe.writeback0(long address)
2816 bool LibraryCallKit::inline_unsafe_writeback0() {
2817 if (!Matcher::has_match_rule(Op_CacheWB)) {
2818 return false;
2819 }
2820 #ifndef PRODUCT
2821 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2822 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2823 ciSignature* sig = callee()->signature();
2824 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2825 #endif
2826 null_check_receiver(); // null-check, then ignore
2827 Node *addr = argument(1);
2828 addr = new CastX2PNode(addr);
2829 addr = _gvn.transform(addr);
2830 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2831 flush = _gvn.transform(flush);
2832 set_memory(flush, TypeRawPtr::BOTTOM);
2833 return true;
2834 }
2835
2836 //----------------------------inline_unsafe_writeback0-------------------------
2837 // public native void Unsafe.writeback0(long address)
2838 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2839 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2840 return false;
2841 }
2842 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2843 return false;
2844 }
2845 #ifndef PRODUCT
2846 assert(Matcher::has_match_rule(Op_CacheWB),
2847 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2848 : "found match rule for CacheWBPostSync but not CacheWB"));
2849
2850 #endif
2851 null_check_receiver(); // null-check, then ignore
2852 Node *sync;
2853 if (is_pre) {
2854 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2855 } else {
2856 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2857 }
2858 sync = _gvn.transform(sync);
2859 set_memory(sync, TypeRawPtr::BOTTOM);
2860 return true;
2861 }
2862
2863 //----------------------------inline_unsafe_allocate---------------------------
2864 // public native Object Unsafe.allocateInstance(Class<?> cls);
2865 bool LibraryCallKit::inline_unsafe_allocate() {
2866
2867 #if INCLUDE_JVMTI
2868 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
2869 return false;
2870 }
2871 #endif //INCLUDE_JVMTI
2872
2873 if (callee()->is_static()) return false; // caller must have the capability!
2874
2875 null_check_receiver(); // null-check, then ignore
2876 Node* cls = null_check(argument(1));
2877 if (stopped()) return true;
2878
2879 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
2880 kls = null_check(kls);
2881 if (stopped()) return true; // argument was like int.class
2882
2883 #if INCLUDE_JVMTI
2884 // Don't try to access new allocated obj in the intrinsic.
2885 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
2886 // Deoptimize and allocate in interpreter instead.
2887 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
2888 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
2889 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
2890 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
2891 {
2892 BuildCutout unless(this, tst, PROB_MAX);
2893 uncommon_trap(Deoptimization::Reason_intrinsic,
2894 Deoptimization::Action_make_not_entrant);
2895 }
2896 if (stopped()) {
2897 return true;
2898 }
2899 #endif //INCLUDE_JVMTI
2900
2901 Node* test = nullptr;
2902 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2903 // Note: The argument might still be an illegal value like
2904 // Serializable.class or Object[].class. The runtime will handle it.
2905 // But we must make an explicit check for initialization.
2906 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2907 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2908 // can generate code to load it as unsigned byte.
2909 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
2910 Node* bits = intcon(InstanceKlass::fully_initialized);
2911 test = _gvn.transform(new SubINode(inst, bits));
2912 // The 'test' is non-zero if we need to take a slow path.
2913 }
2914
2915 Node* obj = new_instance(kls, test);
2916 set_result(obj);
2917 return true;
2918 }
2919
2920 //------------------------inline_native_time_funcs--------------
2921 // inline code for System.currentTimeMillis() and System.nanoTime()
2922 // these have the same type and signature
2923 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2924 const TypeFunc* tf = OptoRuntime::void_long_Type();
2925 const TypePtr* no_memory_effects = nullptr;
2926 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2927 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2928 #ifdef ASSERT
2929 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2930 assert(value_top == top(), "second value must be top");
2931 #endif
2932 set_result(value);
2933 return true;
2934 }
2935
2936
2937 #if INCLUDE_JVMTI
2938
2939 // When notifications are disabled then just update the VTMS transition bit and return.
2940 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
2941 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
2942 if (!DoJVMTIVirtualThreadTransitions) {
2943 return true;
2944 }
2945 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
2946 IdealKit ideal(this);
2947
2948 Node* ONE = ideal.ConI(1);
2949 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
2950 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
2951 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
2952
2953 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
2954 sync_kit(ideal);
2955 // if notifyJvmti enabled then make a call to the given SharedRuntime function
2956 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
2957 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
2958 ideal.sync_kit(this);
2959 } ideal.else_(); {
2960 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
2961 Node* thread = ideal.thread();
2962 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
2963 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
2964
2965 sync_kit(ideal);
2966 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2967 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2968
2969 ideal.sync_kit(this);
2970 } ideal.end_if();
2971 final_sync(ideal);
2972
2973 return true;
2974 }
2975
2976 // Always update the is_disable_suspend bit.
2977 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
2978 if (!DoJVMTIVirtualThreadTransitions) {
2979 return true;
2980 }
2981 IdealKit ideal(this);
2982
2983 {
2984 // unconditionally update the is_disable_suspend bit in current JavaThread
2985 Node* thread = ideal.thread();
2986 Node* arg = _gvn.transform(argument(0)); // argument for notification
2987 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
2988 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
2989
2990 sync_kit(ideal);
2991 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2992 ideal.sync_kit(this);
2993 }
2994 final_sync(ideal);
2995
2996 return true;
2997 }
2998
2999 #endif // INCLUDE_JVMTI
3000
3001 #ifdef JFR_HAVE_INTRINSICS
3002
3003 /**
3004 * if oop->klass != null
3005 * // normal class
3006 * epoch = _epoch_state ? 2 : 1
3007 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3008 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3009 * }
3010 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3011 * else
3012 * // primitive class
3013 * if oop->array_klass != null
3014 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3015 * else
3016 * id = LAST_TYPE_ID + 1 // void class path
3017 * if (!signaled)
3018 * signaled = true
3019 */
3020 bool LibraryCallKit::inline_native_classID() {
3021 Node* cls = argument(0);
3022
3023 IdealKit ideal(this);
3024 #define __ ideal.
3025 IdealVariable result(ideal); __ declarations_done();
3026 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3027 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3028 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3029
3030
3031 __ if_then(kls, BoolTest::ne, null()); {
3032 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3033 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3034
3035 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3036 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3037 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3038 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3039 mask = _gvn.transform(new OrLNode(mask, epoch));
3040 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3041
3042 float unlikely = PROB_UNLIKELY(0.999);
3043 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3044 sync_kit(ideal);
3045 make_runtime_call(RC_LEAF,
3046 OptoRuntime::class_id_load_barrier_Type(),
3047 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3048 "class id load barrier",
3049 TypePtr::BOTTOM,
3050 kls);
3051 ideal.sync_kit(this);
3052 } __ end_if();
3053
3054 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3055 } __ else_(); {
3056 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3057 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3058 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3059 __ if_then(array_kls, BoolTest::ne, null()); {
3060 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3061 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3062 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3063 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3064 } __ else_(); {
3065 // void class case
3066 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3067 } __ end_if();
3068
3069 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3070 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3071 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3072 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3073 } __ end_if();
3074 } __ end_if();
3075
3076 final_sync(ideal);
3077 set_result(ideal.value(result));
3078 #undef __
3079 return true;
3080 }
3081
3082 //------------------------inline_native_jvm_commit------------------
3083 bool LibraryCallKit::inline_native_jvm_commit() {
3084 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3085
3086 // Save input memory and i_o state.
3087 Node* input_memory_state = reset_memory();
3088 set_all_memory(input_memory_state);
3089 Node* input_io_state = i_o();
3090
3091 // TLS.
3092 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3093 // Jfr java buffer.
3094 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3095 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3096 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3097
3098 // Load the current value of the notified field in the JfrThreadLocal.
3099 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3100 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3101
3102 // Test for notification.
3103 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3104 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3105 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3106
3107 // True branch, is notified.
3108 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3109 set_control(is_notified);
3110
3111 // Reset notified state.
3112 Node* notified_reset_memory = store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3113
3114 // 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.
3115 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3116 // Convert the machine-word to a long.
3117 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3118
3119 // False branch, not notified.
3120 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3121 set_control(not_notified);
3122 set_all_memory(input_memory_state);
3123
3124 // Arg is the next position as a long.
3125 Node* arg = argument(0);
3126 // Convert long to machine-word.
3127 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3128
3129 // Store the next_position to the underlying jfr java buffer.
3130 Node* commit_memory;
3131 #ifdef _LP64
3132 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_LONG, MemNode::release);
3133 #else
3134 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_INT, MemNode::release);
3135 #endif
3136
3137 // 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.
3138 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3139 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3140 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3141
3142 // And flags with lease constant.
3143 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3144
3145 // Branch on lease to conditionalize returning the leased java buffer.
3146 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3147 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3148 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3149
3150 // False branch, not a lease.
3151 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3152
3153 // True branch, is lease.
3154 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3155 set_control(is_lease);
3156
3157 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3158 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3159 OptoRuntime::void_void_Type(),
3160 SharedRuntime::jfr_return_lease(),
3161 "return_lease", TypePtr::BOTTOM);
3162 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3163
3164 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3165 record_for_igvn(lease_compare_rgn);
3166 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3167 record_for_igvn(lease_compare_mem);
3168 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3169 record_for_igvn(lease_compare_io);
3170 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3171 record_for_igvn(lease_result_value);
3172
3173 // Update control and phi nodes.
3174 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3175 lease_compare_rgn->init_req(_false_path, not_lease);
3176
3177 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3178 lease_compare_mem->init_req(_false_path, commit_memory);
3179
3180 lease_compare_io->init_req(_true_path, i_o());
3181 lease_compare_io->init_req(_false_path, input_io_state);
3182
3183 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3184 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3185
3186 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3187 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3188 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3189 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3190
3191 // Update control and phi nodes.
3192 result_rgn->init_req(_true_path, is_notified);
3193 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3194
3195 result_mem->init_req(_true_path, notified_reset_memory);
3196 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3197
3198 result_io->init_req(_true_path, input_io_state);
3199 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3200
3201 result_value->init_req(_true_path, current_pos);
3202 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3203
3204 // Set output state.
3205 set_control(_gvn.transform(result_rgn));
3206 set_all_memory(_gvn.transform(result_mem));
3207 set_i_o(_gvn.transform(result_io));
3208 set_result(result_rgn, result_value);
3209 return true;
3210 }
3211
3212 /*
3213 * The intrinsic is a model of this pseudo-code:
3214 *
3215 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3216 * jobject h_event_writer = tl->java_event_writer();
3217 * if (h_event_writer == nullptr) {
3218 * return nullptr;
3219 * }
3220 * oop threadObj = Thread::threadObj();
3221 * oop vthread = java_lang_Thread::vthread(threadObj);
3222 * traceid tid;
3223 * bool pinVirtualThread;
3224 * bool excluded;
3225 * if (vthread != threadObj) { // i.e. current thread is virtual
3226 * tid = java_lang_Thread::tid(vthread);
3227 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3228 * pinVirtualThread = VMContinuations;
3229 * excluded = vthread_epoch_raw & excluded_mask;
3230 * if (!excluded) {
3231 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3232 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3233 * if (vthread_epoch != current_epoch) {
3234 * write_checkpoint();
3235 * }
3236 * }
3237 * } else {
3238 * tid = java_lang_Thread::tid(threadObj);
3239 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3240 * pinVirtualThread = false;
3241 * excluded = thread_epoch_raw & excluded_mask;
3242 * }
3243 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3244 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3245 * if (tid_in_event_writer != tid) {
3246 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3247 * setField(event_writer, "excluded", excluded);
3248 * setField(event_writer, "threadID", tid);
3249 * }
3250 * return event_writer
3251 */
3252 bool LibraryCallKit::inline_native_getEventWriter() {
3253 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3254
3255 // Save input memory and i_o state.
3256 Node* input_memory_state = reset_memory();
3257 set_all_memory(input_memory_state);
3258 Node* input_io_state = i_o();
3259
3260 Node* excluded_mask = _gvn.intcon(32768);
3261 Node* epoch_mask = _gvn.intcon(32767);
3262
3263 // TLS
3264 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3265
3266 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3267 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3268
3269 // Load the eventwriter jobject handle.
3270 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3271
3272 // Null check the jobject handle.
3273 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3274 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3275 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3276
3277 // False path, jobj is null.
3278 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3279
3280 // True path, jobj is not null.
3281 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3282
3283 set_control(jobj_is_not_null);
3284
3285 // Load the threadObj for the CarrierThread.
3286 Node* threadObj = generate_current_thread(tls_ptr);
3287
3288 // Load the vthread.
3289 Node* vthread = generate_virtual_thread(tls_ptr);
3290
3291 // If vthread != threadObj, this is a virtual thread.
3292 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3293 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3294 IfNode* iff_vthread_not_equal_threadObj =
3295 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3296
3297 // False branch, fallback to threadObj.
3298 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3299 set_control(vthread_equal_threadObj);
3300
3301 // Load the tid field from the vthread object.
3302 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3303
3304 // Load the raw epoch value from the threadObj.
3305 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3306 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3307 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3308 TypeInt::CHAR, T_CHAR,
3309 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3310
3311 // Mask off the excluded information from the epoch.
3312 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3313
3314 // True branch, this is a virtual thread.
3315 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3316 set_control(vthread_not_equal_threadObj);
3317
3318 // Load the tid field from the vthread object.
3319 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3320
3321 // Continuation support determines if a virtual thread should be pinned.
3322 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3323 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3324
3325 // Load the raw epoch value from the vthread.
3326 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3327 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3328 TypeInt::CHAR, T_CHAR,
3329 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3330
3331 // Mask off the excluded information from the epoch.
3332 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3333
3334 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3335 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3336 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3337 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3338
3339 // False branch, vthread is excluded, no need to write epoch info.
3340 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3341
3342 // True branch, vthread is included, update epoch info.
3343 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3344 set_control(included);
3345
3346 // Get epoch value.
3347 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3348
3349 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3350 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3351 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3352
3353 // Compare the epoch in the vthread to the current epoch generation.
3354 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3355 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3356 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3357
3358 // False path, epoch is equal, checkpoint information is valid.
3359 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3360
3361 // True path, epoch is not equal, write a checkpoint for the vthread.
3362 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3363
3364 set_control(epoch_is_not_equal);
3365
3366 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3367 // The call also updates the native thread local thread id and the vthread with the current epoch.
3368 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3369 OptoRuntime::jfr_write_checkpoint_Type(),
3370 SharedRuntime::jfr_write_checkpoint(),
3371 "write_checkpoint", TypePtr::BOTTOM);
3372 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3373
3374 // vthread epoch != current epoch
3375 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3376 record_for_igvn(epoch_compare_rgn);
3377 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3378 record_for_igvn(epoch_compare_mem);
3379 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3380 record_for_igvn(epoch_compare_io);
3381
3382 // Update control and phi nodes.
3383 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3384 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3385 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3386 epoch_compare_mem->init_req(_false_path, input_memory_state);
3387 epoch_compare_io->init_req(_true_path, i_o());
3388 epoch_compare_io->init_req(_false_path, input_io_state);
3389
3390 // excluded != true
3391 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3392 record_for_igvn(exclude_compare_rgn);
3393 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3394 record_for_igvn(exclude_compare_mem);
3395 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3396 record_for_igvn(exclude_compare_io);
3397
3398 // Update control and phi nodes.
3399 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3400 exclude_compare_rgn->init_req(_false_path, excluded);
3401 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3402 exclude_compare_mem->init_req(_false_path, input_memory_state);
3403 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3404 exclude_compare_io->init_req(_false_path, input_io_state);
3405
3406 // vthread != threadObj
3407 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3408 record_for_igvn(vthread_compare_rgn);
3409 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3410 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3411 record_for_igvn(vthread_compare_io);
3412 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3413 record_for_igvn(tid);
3414 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3415 record_for_igvn(exclusion);
3416 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3417 record_for_igvn(pinVirtualThread);
3418
3419 // Update control and phi nodes.
3420 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3421 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3422 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3423 vthread_compare_mem->init_req(_false_path, input_memory_state);
3424 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3425 vthread_compare_io->init_req(_false_path, input_io_state);
3426 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3427 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3428 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3429 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3430 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3431 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3432
3433 // Update branch state.
3434 set_control(_gvn.transform(vthread_compare_rgn));
3435 set_all_memory(_gvn.transform(vthread_compare_mem));
3436 set_i_o(_gvn.transform(vthread_compare_io));
3437
3438 // Load the event writer oop by dereferencing the jobject handle.
3439 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3440 assert(klass_EventWriter->is_loaded(), "invariant");
3441 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3442 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3443 const TypeOopPtr* const xtype = aklass->as_instance_type();
3444 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3445 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3446
3447 // Load the current thread id from the event writer object.
3448 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3449 // Get the field offset to, conditionally, store an updated tid value later.
3450 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3451 // Get the field offset to, conditionally, store an updated exclusion value later.
3452 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3453 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3454 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3455
3456 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3457 record_for_igvn(event_writer_tid_compare_rgn);
3458 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3459 record_for_igvn(event_writer_tid_compare_mem);
3460 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3461 record_for_igvn(event_writer_tid_compare_io);
3462
3463 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3464 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3465 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3466 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3467
3468 // False path, tids are the same.
3469 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3470
3471 // True path, tid is not equal, need to update the tid in the event writer.
3472 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3473 record_for_igvn(tid_is_not_equal);
3474
3475 // Store the pin state to the event writer.
3476 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3477
3478 // Store the exclusion state to the event writer.
3479 store_to_memory(tid_is_not_equal, event_writer_excluded_field, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered);
3480
3481 // Store the tid to the event writer.
3482 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3483
3484 // Update control and phi nodes.
3485 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3486 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3487 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3488 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3489 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3490 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3491
3492 // Result of top level CFG, Memory, IO and Value.
3493 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3494 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3495 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3496 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3497
3498 // Result control.
3499 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3500 result_rgn->init_req(_false_path, jobj_is_null);
3501
3502 // Result memory.
3503 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3504 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3505
3506 // Result IO.
3507 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3508 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3509
3510 // Result value.
3511 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3512 result_value->init_req(_false_path, null()); // return null
3513
3514 // Set output state.
3515 set_control(_gvn.transform(result_rgn));
3516 set_all_memory(_gvn.transform(result_mem));
3517 set_i_o(_gvn.transform(result_io));
3518 set_result(result_rgn, result_value);
3519 return true;
3520 }
3521
3522 /*
3523 * The intrinsic is a model of this pseudo-code:
3524 *
3525 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3526 * if (carrierThread != thread) { // is virtual thread
3527 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3528 * bool excluded = vthread_epoch_raw & excluded_mask;
3529 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3530 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3531 * if (!excluded) {
3532 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3533 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3534 * }
3535 * Atomic::release_store(&tl->_vthread, true);
3536 * return;
3537 * }
3538 * Atomic::release_store(&tl->_vthread, false);
3539 */
3540 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3541 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3542
3543 Node* input_memory_state = reset_memory();
3544 set_all_memory(input_memory_state);
3545
3546 Node* excluded_mask = _gvn.intcon(32768);
3547 Node* epoch_mask = _gvn.intcon(32767);
3548
3549 Node* const carrierThread = generate_current_thread(jt);
3550 // If thread != carrierThread, this is a virtual thread.
3551 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3552 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3553 IfNode* iff_thread_not_equal_carrierThread =
3554 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3555
3556 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3557
3558 // False branch, is carrierThread.
3559 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3560 // Store release
3561 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3562
3563 set_all_memory(input_memory_state);
3564
3565 // True branch, is virtual thread.
3566 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3567 set_control(thread_not_equal_carrierThread);
3568
3569 // Load the raw epoch value from the vthread.
3570 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3571 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
3572 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3573
3574 // Mask off the excluded information from the epoch.
3575 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3576
3577 // Load the tid field from the thread.
3578 Node* tid = load_field_from_object(thread, "tid", "J");
3579
3580 // Store the vthread tid to the jfr thread local.
3581 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3582 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
3583
3584 // Branch is_excluded to conditionalize updating the epoch .
3585 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3586 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3587 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3588
3589 // True branch, vthread is excluded, no need to write epoch info.
3590 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3591 set_control(excluded);
3592 Node* vthread_is_excluded = _gvn.intcon(1);
3593
3594 // False branch, vthread is included, update epoch info.
3595 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3596 set_control(included);
3597 Node* vthread_is_included = _gvn.intcon(0);
3598
3599 // Get epoch value.
3600 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3601
3602 // Store the vthread epoch to the jfr thread local.
3603 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3604 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
3605
3606 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3607 record_for_igvn(excluded_rgn);
3608 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3609 record_for_igvn(excluded_mem);
3610 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3611 record_for_igvn(exclusion);
3612
3613 // Merge the excluded control and memory.
3614 excluded_rgn->init_req(_true_path, excluded);
3615 excluded_rgn->init_req(_false_path, included);
3616 excluded_mem->init_req(_true_path, tid_memory);
3617 excluded_mem->init_req(_false_path, included_memory);
3618 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3619 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3620
3621 // Set intermediate state.
3622 set_control(_gvn.transform(excluded_rgn));
3623 set_all_memory(excluded_mem);
3624
3625 // Store the vthread exclusion state to the jfr thread local.
3626 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3627 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
3628
3629 // Store release
3630 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
3631
3632 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3633 record_for_igvn(thread_compare_rgn);
3634 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3635 record_for_igvn(thread_compare_mem);
3636 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3637 record_for_igvn(vthread);
3638
3639 // Merge the thread_compare control and memory.
3640 thread_compare_rgn->init_req(_true_path, control());
3641 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3642 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3643 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3644
3645 // Set output state.
3646 set_control(_gvn.transform(thread_compare_rgn));
3647 set_all_memory(_gvn.transform(thread_compare_mem));
3648 }
3649
3650 #endif // JFR_HAVE_INTRINSICS
3651
3652 //------------------------inline_native_currentCarrierThread------------------
3653 bool LibraryCallKit::inline_native_currentCarrierThread() {
3654 Node* junk = nullptr;
3655 set_result(generate_current_thread(junk));
3656 return true;
3657 }
3658
3659 //------------------------inline_native_currentThread------------------
3660 bool LibraryCallKit::inline_native_currentThread() {
3661 Node* junk = nullptr;
3662 set_result(generate_virtual_thread(junk));
3663 return true;
3664 }
3665
3666 //------------------------inline_native_setVthread------------------
3667 bool LibraryCallKit::inline_native_setCurrentThread() {
3668 assert(C->method()->changes_current_thread(),
3669 "method changes current Thread but is not annotated ChangesCurrentThread");
3670 Node* arr = argument(1);
3671 Node* thread = _gvn.transform(new ThreadLocalNode());
3672 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3673 Node* thread_obj_handle
3674 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3675 thread_obj_handle = _gvn.transform(thread_obj_handle);
3676 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3677 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3678
3679 // Change the lock_id of the JavaThread
3680 Node* tid = load_field_from_object(arr, "tid", "J");
3681 Node* thread_id_offset = basic_plus_adr(thread, in_bytes(JavaThread::lock_id_offset()));
3682 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
3683
3684 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3685 return true;
3686 }
3687
3688 const Type* LibraryCallKit::scopedValueCache_type() {
3689 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3690 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3691 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3692
3693 // Because we create the scopedValue cache lazily we have to make the
3694 // type of the result BotPTR.
3695 bool xk = etype->klass_is_exact();
3696 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3697 return objects_type;
3698 }
3699
3700 Node* LibraryCallKit::scopedValueCache_helper() {
3701 Node* thread = _gvn.transform(new ThreadLocalNode());
3702 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3703 // We cannot use immutable_memory() because we might flip onto a
3704 // different carrier thread, at which point we'll need to use that
3705 // carrier thread's cache.
3706 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3707 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3708 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3709 }
3710
3711 //------------------------inline_native_scopedValueCache------------------
3712 bool LibraryCallKit::inline_native_scopedValueCache() {
3713 Node* cache_obj_handle = scopedValueCache_helper();
3714 const Type* objects_type = scopedValueCache_type();
3715 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3716
3717 return true;
3718 }
3719
3720 //------------------------inline_native_setScopedValueCache------------------
3721 bool LibraryCallKit::inline_native_setScopedValueCache() {
3722 Node* arr = argument(0);
3723 Node* cache_obj_handle = scopedValueCache_helper();
3724 const Type* objects_type = scopedValueCache_type();
3725
3726 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3727 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3728
3729 return true;
3730 }
3731
3732 //------------------------inline_native_Continuation_pin and unpin-----------
3733
3734 // Shared implementation routine for both pin and unpin.
3735 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
3736 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3737
3738 // Save input memory.
3739 Node* input_memory_state = reset_memory();
3740 set_all_memory(input_memory_state);
3741
3742 // TLS
3743 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3744 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
3745 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
3746
3747 // Null check the last continuation object.
3748 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
3749 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
3750 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3751
3752 // False path, last continuation is null.
3753 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
3754
3755 // True path, last continuation is not null.
3756 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
3757
3758 set_control(continuation_is_not_null);
3759
3760 // Load the pin count from the last continuation.
3761 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
3762 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
3763
3764 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
3765 Node* pin_count_rhs;
3766 if (unpin) {
3767 pin_count_rhs = _gvn.intcon(0);
3768 } else {
3769 pin_count_rhs = _gvn.intcon(UINT32_MAX);
3770 }
3771 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
3772 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
3773 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
3774
3775 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
3776 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
3777 set_control(valid_pin_count);
3778
3779 Node* next_pin_count;
3780 if (unpin) {
3781 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
3782 } else {
3783 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
3784 }
3785
3786 Node* updated_pin_count_memory = store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
3787
3788 // True branch, pin count over/underflow.
3789 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
3790 {
3791 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
3792 // which will throw IllegalStateException for pin count over/underflow.
3793 PreserveJVMState pjvms(this);
3794 set_control(pin_count_over_underflow);
3795 set_all_memory(input_memory_state);
3796 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
3797 Deoptimization::Action_none);
3798 assert(stopped(), "invariant");
3799 }
3800
3801 // Result of top level CFG and Memory.
3802 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3803 record_for_igvn(result_rgn);
3804 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3805 record_for_igvn(result_mem);
3806
3807 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
3808 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
3809 result_mem->init_req(_true_path, _gvn.transform(updated_pin_count_memory));
3810 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3811
3812 // Set output state.
3813 set_control(_gvn.transform(result_rgn));
3814 set_all_memory(_gvn.transform(result_mem));
3815
3816 return true;
3817 }
3818
3819 //---------------------------load_mirror_from_klass----------------------------
3820 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3821 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3822 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3823 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3824 // mirror = ((OopHandle)mirror)->resolve();
3825 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3826 }
3827
3828 //-----------------------load_klass_from_mirror_common-------------------------
3829 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3830 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3831 // and branch to the given path on the region.
3832 // If never_see_null, take an uncommon trap on null, so we can optimistically
3833 // compile for the non-null case.
3834 // If the region is null, force never_see_null = true.
3835 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3836 bool never_see_null,
3837 RegionNode* region,
3838 int null_path,
3839 int offset) {
3840 if (region == nullptr) never_see_null = true;
3841 Node* p = basic_plus_adr(mirror, offset);
3842 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3843 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3844 Node* null_ctl = top();
3845 kls = null_check_oop(kls, &null_ctl, never_see_null);
3846 if (region != nullptr) {
3847 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3848 region->init_req(null_path, null_ctl);
3849 } else {
3850 assert(null_ctl == top(), "no loose ends");
3851 }
3852 return kls;
3853 }
3854
3855 //--------------------(inline_native_Class_query helpers)---------------------
3856 // Use this for JVM_ACC_INTERFACE.
3857 // Fall through if (mods & mask) == bits, take the guard otherwise.
3858 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
3859 ByteSize offset, const Type* type, BasicType bt) {
3860 // Branch around if the given klass has the given modifier bit set.
3861 // Like generate_guard, adds a new path onto the region.
3862 Node* modp = basic_plus_adr(kls, in_bytes(offset));
3863 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
3864 Node* mask = intcon(modifier_mask);
3865 Node* bits = intcon(modifier_bits);
3866 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3867 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3868 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3869 return generate_fair_guard(bol, region);
3870 }
3871 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3872 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
3873 Klass::access_flags_offset(), TypeInt::INT, T_INT);
3874 }
3875
3876 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
3877 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3878 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
3879 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
3880 }
3881
3882 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3883 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
3884 }
3885
3886 //-------------------------inline_native_Class_query-------------------
3887 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3888 const Type* return_type = TypeInt::BOOL;
3889 Node* prim_return_value = top(); // what happens if it's a primitive class?
3890 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3891 bool expect_prim = false; // most of these guys expect to work on refs
3892
3893 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3894
3895 Node* mirror = argument(0);
3896 Node* obj = top();
3897
3898 switch (id) {
3899 case vmIntrinsics::_isInstance:
3900 // nothing is an instance of a primitive type
3901 prim_return_value = intcon(0);
3902 obj = argument(1);
3903 break;
3904 case vmIntrinsics::_getModifiers:
3905 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3906 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3907 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3908 break;
3909 case vmIntrinsics::_isInterface:
3910 prim_return_value = intcon(0);
3911 break;
3912 case vmIntrinsics::_isArray:
3913 prim_return_value = intcon(0);
3914 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3915 break;
3916 case vmIntrinsics::_isPrimitive:
3917 prim_return_value = intcon(1);
3918 expect_prim = true; // obviously
3919 break;
3920 case vmIntrinsics::_isHidden:
3921 prim_return_value = intcon(0);
3922 break;
3923 case vmIntrinsics::_getSuperclass:
3924 prim_return_value = null();
3925 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3926 break;
3927 case vmIntrinsics::_getClassAccessFlags:
3928 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3929 return_type = TypeInt::INT; // not bool! 6297094
3930 break;
3931 default:
3932 fatal_unexpected_iid(id);
3933 break;
3934 }
3935
3936 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3937 if (mirror_con == nullptr) return false; // cannot happen?
3938
3939 #ifndef PRODUCT
3940 if (C->print_intrinsics() || C->print_inlining()) {
3941 ciType* k = mirror_con->java_mirror_type();
3942 if (k) {
3943 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3944 k->print_name();
3945 tty->cr();
3946 }
3947 }
3948 #endif
3949
3950 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3951 RegionNode* region = new RegionNode(PATH_LIMIT);
3952 record_for_igvn(region);
3953 PhiNode* phi = new PhiNode(region, return_type);
3954
3955 // The mirror will never be null of Reflection.getClassAccessFlags, however
3956 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3957 // if it is. See bug 4774291.
3958
3959 // For Reflection.getClassAccessFlags(), the null check occurs in
3960 // the wrong place; see inline_unsafe_access(), above, for a similar
3961 // situation.
3962 mirror = null_check(mirror);
3963 // If mirror or obj is dead, only null-path is taken.
3964 if (stopped()) return true;
3965
3966 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3967
3968 // Now load the mirror's klass metaobject, and null-check it.
3969 // Side-effects region with the control path if the klass is null.
3970 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3971 // If kls is null, we have a primitive mirror.
3972 phi->init_req(_prim_path, prim_return_value);
3973 if (stopped()) { set_result(region, phi); return true; }
3974 bool safe_for_replace = (region->in(_prim_path) == top());
3975
3976 Node* p; // handy temp
3977 Node* null_ctl;
3978
3979 // Now that we have the non-null klass, we can perform the real query.
3980 // For constant classes, the query will constant-fold in LoadNode::Value.
3981 Node* query_value = top();
3982 switch (id) {
3983 case vmIntrinsics::_isInstance:
3984 // nothing is an instance of a primitive type
3985 query_value = gen_instanceof(obj, kls, safe_for_replace);
3986 break;
3987
3988 case vmIntrinsics::_getModifiers:
3989 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3990 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
3991 break;
3992
3993 case vmIntrinsics::_isInterface:
3994 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3995 if (generate_interface_guard(kls, region) != nullptr)
3996 // A guard was added. If the guard is taken, it was an interface.
3997 phi->add_req(intcon(1));
3998 // If we fall through, it's a plain class.
3999 query_value = intcon(0);
4000 break;
4001
4002 case vmIntrinsics::_isArray:
4003 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
4004 if (generate_array_guard(kls, region) != nullptr)
4005 // A guard was added. If the guard is taken, it was an array.
4006 phi->add_req(intcon(1));
4007 // If we fall through, it's a plain class.
4008 query_value = intcon(0);
4009 break;
4010
4011 case vmIntrinsics::_isPrimitive:
4012 query_value = intcon(0); // "normal" path produces false
4013 break;
4014
4015 case vmIntrinsics::_isHidden:
4016 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4017 if (generate_hidden_class_guard(kls, region) != nullptr)
4018 // A guard was added. If the guard is taken, it was an hidden class.
4019 phi->add_req(intcon(1));
4020 // If we fall through, it's a plain class.
4021 query_value = intcon(0);
4022 break;
4023
4024
4025 case vmIntrinsics::_getSuperclass:
4026 // The rules here are somewhat unfortunate, but we can still do better
4027 // with random logic than with a JNI call.
4028 // Interfaces store null or Object as _super, but must report null.
4029 // Arrays store an intermediate super as _super, but must report Object.
4030 // Other types can report the actual _super.
4031 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4032 if (generate_interface_guard(kls, region) != nullptr)
4033 // A guard was added. If the guard is taken, it was an interface.
4034 phi->add_req(null());
4035 if (generate_array_guard(kls, region) != nullptr)
4036 // A guard was added. If the guard is taken, it was an array.
4037 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4038 // If we fall through, it's a plain class. Get its _super.
4039 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4040 kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4041 null_ctl = top();
4042 kls = null_check_oop(kls, &null_ctl);
4043 if (null_ctl != top()) {
4044 // If the guard is taken, Object.superClass is null (both klass and mirror).
4045 region->add_req(null_ctl);
4046 phi ->add_req(null());
4047 }
4048 if (!stopped()) {
4049 query_value = load_mirror_from_klass(kls);
4050 }
4051 break;
4052
4053 case vmIntrinsics::_getClassAccessFlags:
4054 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
4055 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
4056 break;
4057
4058 default:
4059 fatal_unexpected_iid(id);
4060 break;
4061 }
4062
4063 // Fall-through is the normal case of a query to a real class.
4064 phi->init_req(1, query_value);
4065 region->init_req(1, control());
4066
4067 C->set_has_split_ifs(true); // Has chance for split-if optimization
4068 set_result(region, phi);
4069 return true;
4070 }
4071
4072 //-------------------------inline_Class_cast-------------------
4073 bool LibraryCallKit::inline_Class_cast() {
4074 Node* mirror = argument(0); // Class
4075 Node* obj = argument(1);
4076 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4077 if (mirror_con == nullptr) {
4078 return false; // dead path (mirror->is_top()).
4079 }
4080 if (obj == nullptr || obj->is_top()) {
4081 return false; // dead path
4082 }
4083 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4084
4085 // First, see if Class.cast() can be folded statically.
4086 // java_mirror_type() returns non-null for compile-time Class constants.
4087 ciType* tm = mirror_con->java_mirror_type();
4088 if (tm != nullptr && tm->is_klass() &&
4089 tp != nullptr) {
4090 if (!tp->is_loaded()) {
4091 // Don't use intrinsic when class is not loaded.
4092 return false;
4093 } else {
4094 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
4095 if (static_res == Compile::SSC_always_true) {
4096 // isInstance() is true - fold the code.
4097 set_result(obj);
4098 return true;
4099 } else if (static_res == Compile::SSC_always_false) {
4100 // Don't use intrinsic, have to throw ClassCastException.
4101 // If the reference is null, the non-intrinsic bytecode will
4102 // be optimized appropriately.
4103 return false;
4104 }
4105 }
4106 }
4107
4108 // Bailout intrinsic and do normal inlining if exception path is frequent.
4109 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4110 return false;
4111 }
4112
4113 // Generate dynamic checks.
4114 // Class.cast() is java implementation of _checkcast bytecode.
4115 // Do checkcast (Parse::do_checkcast()) optimizations here.
4116
4117 mirror = null_check(mirror);
4118 // If mirror is dead, only null-path is taken.
4119 if (stopped()) {
4120 return true;
4121 }
4122
4123 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive).
4124 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
4125 RegionNode* region = new RegionNode(PATH_LIMIT);
4126 record_for_igvn(region);
4127
4128 // Now load the mirror's klass metaobject, and null-check it.
4129 // If kls is null, we have a primitive mirror and
4130 // nothing is an instance of a primitive type.
4131 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4132
4133 Node* res = top();
4134 if (!stopped()) {
4135 Node* bad_type_ctrl = top();
4136 // Do checkcast optimizations.
4137 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4138 region->init_req(_bad_type_path, bad_type_ctrl);
4139 }
4140 if (region->in(_prim_path) != top() ||
4141 region->in(_bad_type_path) != top()) {
4142 // Let Interpreter throw ClassCastException.
4143 PreserveJVMState pjvms(this);
4144 set_control(_gvn.transform(region));
4145 uncommon_trap(Deoptimization::Reason_intrinsic,
4146 Deoptimization::Action_maybe_recompile);
4147 }
4148 if (!stopped()) {
4149 set_result(res);
4150 }
4151 return true;
4152 }
4153
4154
4155 //--------------------------inline_native_subtype_check------------------------
4156 // This intrinsic takes the JNI calls out of the heart of
4157 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4158 bool LibraryCallKit::inline_native_subtype_check() {
4159 // Pull both arguments off the stack.
4160 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4161 args[0] = argument(0);
4162 args[1] = argument(1);
4163 Node* klasses[2]; // corresponding Klasses: superk, subk
4164 klasses[0] = klasses[1] = top();
4165
4166 enum {
4167 // A full decision tree on {superc is prim, subc is prim}:
4168 _prim_0_path = 1, // {P,N} => false
4169 // {P,P} & superc!=subc => false
4170 _prim_same_path, // {P,P} & superc==subc => true
4171 _prim_1_path, // {N,P} => false
4172 _ref_subtype_path, // {N,N} & subtype check wins => true
4173 _both_ref_path, // {N,N} & subtype check loses => false
4174 PATH_LIMIT
4175 };
4176
4177 RegionNode* region = new RegionNode(PATH_LIMIT);
4178 Node* phi = new PhiNode(region, TypeInt::BOOL);
4179 record_for_igvn(region);
4180
4181 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4182 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4183 int class_klass_offset = java_lang_Class::klass_offset();
4184
4185 // First null-check both mirrors and load each mirror's klass metaobject.
4186 int which_arg;
4187 for (which_arg = 0; which_arg <= 1; which_arg++) {
4188 Node* arg = args[which_arg];
4189 arg = null_check(arg);
4190 if (stopped()) break;
4191 args[which_arg] = arg;
4192
4193 Node* p = basic_plus_adr(arg, class_klass_offset);
4194 Node* kls = LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, adr_type, kls_type);
4195 klasses[which_arg] = _gvn.transform(kls);
4196 }
4197
4198 // Having loaded both klasses, test each for null.
4199 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4200 for (which_arg = 0; which_arg <= 1; which_arg++) {
4201 Node* kls = klasses[which_arg];
4202 Node* null_ctl = top();
4203 kls = null_check_oop(kls, &null_ctl, never_see_null);
4204 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
4205 region->init_req(prim_path, null_ctl);
4206 if (stopped()) break;
4207 klasses[which_arg] = kls;
4208 }
4209
4210 if (!stopped()) {
4211 // now we have two reference types, in klasses[0..1]
4212 Node* subk = klasses[1]; // the argument to isAssignableFrom
4213 Node* superk = klasses[0]; // the receiver
4214 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4215 // now we have a successful reference subtype check
4216 region->set_req(_ref_subtype_path, control());
4217 }
4218
4219 // If both operands are primitive (both klasses null), then
4220 // we must return true when they are identical primitives.
4221 // It is convenient to test this after the first null klass check.
4222 set_control(region->in(_prim_0_path)); // go back to first null check
4223 if (!stopped()) {
4224 // Since superc is primitive, make a guard for the superc==subc case.
4225 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4226 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4227 generate_guard(bol_eq, region, PROB_FAIR);
4228 if (region->req() == PATH_LIMIT+1) {
4229 // A guard was added. If the added guard is taken, superc==subc.
4230 region->swap_edges(PATH_LIMIT, _prim_same_path);
4231 region->del_req(PATH_LIMIT);
4232 }
4233 region->set_req(_prim_0_path, control()); // Not equal after all.
4234 }
4235
4236 // these are the only paths that produce 'true':
4237 phi->set_req(_prim_same_path, intcon(1));
4238 phi->set_req(_ref_subtype_path, intcon(1));
4239
4240 // pull together the cases:
4241 assert(region->req() == PATH_LIMIT, "sane region");
4242 for (uint i = 1; i < region->req(); i++) {
4243 Node* ctl = region->in(i);
4244 if (ctl == nullptr || ctl == top()) {
4245 region->set_req(i, top());
4246 phi ->set_req(i, top());
4247 } else if (phi->in(i) == nullptr) {
4248 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4249 }
4250 }
4251
4252 set_control(_gvn.transform(region));
4253 set_result(_gvn.transform(phi));
4254 return true;
4255 }
4256
4257 //---------------------generate_array_guard_common------------------------
4258 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
4259 bool obj_array, bool not_array) {
4260
4261 if (stopped()) {
4262 return nullptr;
4263 }
4264
4265 // If obj_array/non_array==false/false:
4266 // Branch around if the given klass is in fact an array (either obj or prim).
4267 // If obj_array/non_array==false/true:
4268 // Branch around if the given klass is not an array klass of any kind.
4269 // If obj_array/non_array==true/true:
4270 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
4271 // If obj_array/non_array==true/false:
4272 // Branch around if the kls is an oop array (Object[] or subtype)
4273 //
4274 // Like generate_guard, adds a new path onto the region.
4275 jint layout_con = 0;
4276 Node* layout_val = get_layout_helper(kls, layout_con);
4277 if (layout_val == nullptr) {
4278 bool query = (obj_array
4279 ? Klass::layout_helper_is_objArray(layout_con)
4280 : Klass::layout_helper_is_array(layout_con));
4281 if (query == not_array) {
4282 return nullptr; // never a branch
4283 } else { // always a branch
4284 Node* always_branch = control();
4285 if (region != nullptr)
4286 region->add_req(always_branch);
4287 set_control(top());
4288 return always_branch;
4289 }
4290 }
4291 // Now test the correct condition.
4292 jint nval = (obj_array
4293 ? (jint)(Klass::_lh_array_tag_type_value
4294 << Klass::_lh_array_tag_shift)
4295 : Klass::_lh_neutral_value);
4296 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4297 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
4298 // invert the test if we are looking for a non-array
4299 if (not_array) btest = BoolTest(btest).negate();
4300 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4301 return generate_fair_guard(bol, region);
4302 }
4303
4304
4305 //-----------------------inline_native_newArray--------------------------
4306 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4307 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4308 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4309 Node* mirror;
4310 Node* count_val;
4311 if (uninitialized) {
4312 null_check_receiver();
4313 mirror = argument(1);
4314 count_val = argument(2);
4315 } else {
4316 mirror = argument(0);
4317 count_val = argument(1);
4318 }
4319
4320 mirror = null_check(mirror);
4321 // If mirror or obj is dead, only null-path is taken.
4322 if (stopped()) return true;
4323
4324 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4325 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4326 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4327 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4328 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4329
4330 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4331 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4332 result_reg, _slow_path);
4333 Node* normal_ctl = control();
4334 Node* no_array_ctl = result_reg->in(_slow_path);
4335
4336 // Generate code for the slow case. We make a call to newArray().
4337 set_control(no_array_ctl);
4338 if (!stopped()) {
4339 // Either the input type is void.class, or else the
4340 // array klass has not yet been cached. Either the
4341 // ensuing call will throw an exception, or else it
4342 // will cache the array klass for next time.
4343 PreserveJVMState pjvms(this);
4344 CallJavaNode* slow_call = nullptr;
4345 if (uninitialized) {
4346 // Generate optimized virtual call (holder class 'Unsafe' is final)
4347 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4348 } else {
4349 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4350 }
4351 Node* slow_result = set_results_for_java_call(slow_call);
4352 // this->control() comes from set_results_for_java_call
4353 result_reg->set_req(_slow_path, control());
4354 result_val->set_req(_slow_path, slow_result);
4355 result_io ->set_req(_slow_path, i_o());
4356 result_mem->set_req(_slow_path, reset_memory());
4357 }
4358
4359 set_control(normal_ctl);
4360 if (!stopped()) {
4361 // Normal case: The array type has been cached in the java.lang.Class.
4362 // The following call works fine even if the array type is polymorphic.
4363 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4364 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4365 result_reg->init_req(_normal_path, control());
4366 result_val->init_req(_normal_path, obj);
4367 result_io ->init_req(_normal_path, i_o());
4368 result_mem->init_req(_normal_path, reset_memory());
4369
4370 if (uninitialized) {
4371 // Mark the allocation so that zeroing is skipped
4372 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4373 alloc->maybe_set_complete(&_gvn);
4374 }
4375 }
4376
4377 // Return the combined state.
4378 set_i_o( _gvn.transform(result_io) );
4379 set_all_memory( _gvn.transform(result_mem));
4380
4381 C->set_has_split_ifs(true); // Has chance for split-if optimization
4382 set_result(result_reg, result_val);
4383 return true;
4384 }
4385
4386 //----------------------inline_native_getLength--------------------------
4387 // public static native int java.lang.reflect.Array.getLength(Object array);
4388 bool LibraryCallKit::inline_native_getLength() {
4389 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4390
4391 Node* array = null_check(argument(0));
4392 // If array is dead, only null-path is taken.
4393 if (stopped()) return true;
4394
4395 // Deoptimize if it is a non-array.
4396 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr);
4397
4398 if (non_array != nullptr) {
4399 PreserveJVMState pjvms(this);
4400 set_control(non_array);
4401 uncommon_trap(Deoptimization::Reason_intrinsic,
4402 Deoptimization::Action_maybe_recompile);
4403 }
4404
4405 // If control is dead, only non-array-path is taken.
4406 if (stopped()) return true;
4407
4408 // The works fine even if the array type is polymorphic.
4409 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4410 Node* result = load_array_length(array);
4411
4412 C->set_has_split_ifs(true); // Has chance for split-if optimization
4413 set_result(result);
4414 return true;
4415 }
4416
4417 //------------------------inline_array_copyOf----------------------------
4418 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4419 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4420 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4421 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4422
4423 // Get the arguments.
4424 Node* original = argument(0);
4425 Node* start = is_copyOfRange? argument(1): intcon(0);
4426 Node* end = is_copyOfRange? argument(2): argument(1);
4427 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4428
4429 Node* newcopy = nullptr;
4430
4431 // Set the original stack and the reexecute bit for the interpreter to reexecute
4432 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4433 { PreserveReexecuteState preexecs(this);
4434 jvms()->set_should_reexecute(true);
4435
4436 array_type_mirror = null_check(array_type_mirror);
4437 original = null_check(original);
4438
4439 // Check if a null path was taken unconditionally.
4440 if (stopped()) return true;
4441
4442 Node* orig_length = load_array_length(original);
4443
4444 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4445 klass_node = null_check(klass_node);
4446
4447 RegionNode* bailout = new RegionNode(1);
4448 record_for_igvn(bailout);
4449
4450 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4451 // Bail out if that is so.
4452 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4453 if (not_objArray != nullptr) {
4454 // Improve the klass node's type from the new optimistic assumption:
4455 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4456 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4457 Node* cast = new CastPPNode(control(), klass_node, akls);
4458 klass_node = _gvn.transform(cast);
4459 }
4460
4461 // Bail out if either start or end is negative.
4462 generate_negative_guard(start, bailout, &start);
4463 generate_negative_guard(end, bailout, &end);
4464
4465 Node* length = end;
4466 if (_gvn.type(start) != TypeInt::ZERO) {
4467 length = _gvn.transform(new SubINode(end, start));
4468 }
4469
4470 // Bail out if length is negative (i.e., if start > end).
4471 // Without this the new_array would throw
4472 // NegativeArraySizeException but IllegalArgumentException is what
4473 // should be thrown
4474 generate_negative_guard(length, bailout, &length);
4475
4476 // Bail out if start is larger than the original length
4477 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4478 generate_negative_guard(orig_tail, bailout, &orig_tail);
4479
4480 if (bailout->req() > 1) {
4481 PreserveJVMState pjvms(this);
4482 set_control(_gvn.transform(bailout));
4483 uncommon_trap(Deoptimization::Reason_intrinsic,
4484 Deoptimization::Action_maybe_recompile);
4485 }
4486
4487 if (!stopped()) {
4488 // How many elements will we copy from the original?
4489 // The answer is MinI(orig_tail, length).
4490 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4491
4492 // Generate a direct call to the right arraycopy function(s).
4493 // We know the copy is disjoint but we might not know if the
4494 // oop stores need checking.
4495 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4496 // This will fail a store-check if x contains any non-nulls.
4497
4498 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4499 // loads/stores but it is legal only if we're sure the
4500 // Arrays.copyOf would succeed. So we need all input arguments
4501 // to the copyOf to be validated, including that the copy to the
4502 // new array won't trigger an ArrayStoreException. That subtype
4503 // check can be optimized if we know something on the type of
4504 // the input array from type speculation.
4505 if (_gvn.type(klass_node)->singleton()) {
4506 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4507 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4508
4509 int test = C->static_subtype_check(superk, subk);
4510 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4511 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4512 if (t_original->speculative_type() != nullptr) {
4513 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4514 }
4515 }
4516 }
4517
4518 bool validated = false;
4519 // Reason_class_check rather than Reason_intrinsic because we
4520 // want to intrinsify even if this traps.
4521 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4522 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4523
4524 if (not_subtype_ctrl != top()) {
4525 PreserveJVMState pjvms(this);
4526 set_control(not_subtype_ctrl);
4527 uncommon_trap(Deoptimization::Reason_class_check,
4528 Deoptimization::Action_make_not_entrant);
4529 assert(stopped(), "Should be stopped");
4530 }
4531 validated = true;
4532 }
4533
4534 if (!stopped()) {
4535 newcopy = new_array(klass_node, length, 0); // no arguments to push
4536
4537 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4538 load_object_klass(original), klass_node);
4539 if (!is_copyOfRange) {
4540 ac->set_copyof(validated);
4541 } else {
4542 ac->set_copyofrange(validated);
4543 }
4544 Node* n = _gvn.transform(ac);
4545 if (n == ac) {
4546 ac->connect_outputs(this);
4547 } else {
4548 assert(validated, "shouldn't transform if all arguments not validated");
4549 set_all_memory(n);
4550 }
4551 }
4552 }
4553 } // original reexecute is set back here
4554
4555 C->set_has_split_ifs(true); // Has chance for split-if optimization
4556 if (!stopped()) {
4557 set_result(newcopy);
4558 }
4559 return true;
4560 }
4561
4562
4563 //----------------------generate_virtual_guard---------------------------
4564 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4565 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4566 RegionNode* slow_region) {
4567 ciMethod* method = callee();
4568 int vtable_index = method->vtable_index();
4569 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4570 "bad index %d", vtable_index);
4571 // Get the Method* out of the appropriate vtable entry.
4572 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4573 vtable_index*vtableEntry::size_in_bytes() +
4574 in_bytes(vtableEntry::method_offset());
4575 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4576 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4577
4578 // Compare the target method with the expected method (e.g., Object.hashCode).
4579 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4580
4581 Node* native_call = makecon(native_call_addr);
4582 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4583 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4584
4585 return generate_slow_guard(test_native, slow_region);
4586 }
4587
4588 //-----------------------generate_method_call----------------------------
4589 // Use generate_method_call to make a slow-call to the real
4590 // method if the fast path fails. An alternative would be to
4591 // use a stub like OptoRuntime::slow_arraycopy_Java.
4592 // This only works for expanding the current library call,
4593 // not another intrinsic. (E.g., don't use this for making an
4594 // arraycopy call inside of the copyOf intrinsic.)
4595 CallJavaNode*
4596 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4597 // When compiling the intrinsic method itself, do not use this technique.
4598 guarantee(callee() != C->method(), "cannot make slow-call to self");
4599
4600 ciMethod* method = callee();
4601 // ensure the JVMS we have will be correct for this call
4602 guarantee(method_id == method->intrinsic_id(), "must match");
4603
4604 const TypeFunc* tf = TypeFunc::make(method);
4605 if (res_not_null) {
4606 assert(tf->return_type() == T_OBJECT, "");
4607 const TypeTuple* range = tf->range();
4608 const Type** fields = TypeTuple::fields(range->cnt());
4609 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4610 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4611 tf = TypeFunc::make(tf->domain(), new_range);
4612 }
4613 CallJavaNode* slow_call;
4614 if (is_static) {
4615 assert(!is_virtual, "");
4616 slow_call = new CallStaticJavaNode(C, tf,
4617 SharedRuntime::get_resolve_static_call_stub(), method);
4618 } else if (is_virtual) {
4619 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4620 int vtable_index = Method::invalid_vtable_index;
4621 if (UseInlineCaches) {
4622 // Suppress the vtable call
4623 } else {
4624 // hashCode and clone are not a miranda methods,
4625 // so the vtable index is fixed.
4626 // No need to use the linkResolver to get it.
4627 vtable_index = method->vtable_index();
4628 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4629 "bad index %d", vtable_index);
4630 }
4631 slow_call = new CallDynamicJavaNode(tf,
4632 SharedRuntime::get_resolve_virtual_call_stub(),
4633 method, vtable_index);
4634 } else { // neither virtual nor static: opt_virtual
4635 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4636 slow_call = new CallStaticJavaNode(C, tf,
4637 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4638 slow_call->set_optimized_virtual(true);
4639 }
4640 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4641 // To be able to issue a direct call (optimized virtual or virtual)
4642 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4643 // about the method being invoked should be attached to the call site to
4644 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4645 slow_call->set_override_symbolic_info(true);
4646 }
4647 set_arguments_for_java_call(slow_call);
4648 set_edges_for_java_call(slow_call);
4649 return slow_call;
4650 }
4651
4652
4653 /**
4654 * Build special case code for calls to hashCode on an object. This call may
4655 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4656 * slightly different code.
4657 */
4658 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4659 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4660 assert(!(is_virtual && is_static), "either virtual, special, or static");
4661
4662 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4663
4664 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4665 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4666 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4667 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4668 Node* obj = nullptr;
4669 if (!is_static) {
4670 // Check for hashing null object
4671 obj = null_check_receiver();
4672 if (stopped()) return true; // unconditionally null
4673 result_reg->init_req(_null_path, top());
4674 result_val->init_req(_null_path, top());
4675 } else {
4676 // Do a null check, and return zero if null.
4677 // System.identityHashCode(null) == 0
4678 obj = argument(0);
4679 Node* null_ctl = top();
4680 obj = null_check_oop(obj, &null_ctl);
4681 result_reg->init_req(_null_path, null_ctl);
4682 result_val->init_req(_null_path, _gvn.intcon(0));
4683 }
4684
4685 // Unconditionally null? Then return right away.
4686 if (stopped()) {
4687 set_control( result_reg->in(_null_path));
4688 if (!stopped())
4689 set_result(result_val->in(_null_path));
4690 return true;
4691 }
4692
4693 // We only go to the fast case code if we pass a number of guards. The
4694 // paths which do not pass are accumulated in the slow_region.
4695 RegionNode* slow_region = new RegionNode(1);
4696 record_for_igvn(slow_region);
4697
4698 // If this is a virtual call, we generate a funny guard. We pull out
4699 // the vtable entry corresponding to hashCode() from the target object.
4700 // If the target method which we are calling happens to be the native
4701 // Object hashCode() method, we pass the guard. We do not need this
4702 // guard for non-virtual calls -- the caller is known to be the native
4703 // Object hashCode().
4704 if (is_virtual) {
4705 // After null check, get the object's klass.
4706 Node* obj_klass = load_object_klass(obj);
4707 generate_virtual_guard(obj_klass, slow_region);
4708 }
4709
4710 // Get the header out of the object, use LoadMarkNode when available
4711 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4712 // The control of the load must be null. Otherwise, the load can move before
4713 // the null check after castPP removal.
4714 Node* no_ctrl = nullptr;
4715 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4716
4717 if (!UseObjectMonitorTable) {
4718 // Test the header to see if it is safe to read w.r.t. locking.
4719 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4720 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4721 if (LockingMode == LM_LIGHTWEIGHT) {
4722 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
4723 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
4724 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
4725
4726 generate_slow_guard(test_monitor, slow_region);
4727 } else {
4728 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4729 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val));
4730 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne));
4731
4732 generate_slow_guard(test_not_unlocked, slow_region);
4733 }
4734 }
4735
4736 // Get the hash value and check to see that it has been properly assigned.
4737 // We depend on hash_mask being at most 32 bits and avoid the use of
4738 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4739 // vm: see markWord.hpp.
4740 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4741 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4742 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4743 // This hack lets the hash bits live anywhere in the mark object now, as long
4744 // as the shift drops the relevant bits into the low 32 bits. Note that
4745 // Java spec says that HashCode is an int so there's no point in capturing
4746 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4747 hshifted_header = ConvX2I(hshifted_header);
4748 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4749
4750 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4751 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4752 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4753
4754 generate_slow_guard(test_assigned, slow_region);
4755
4756 Node* init_mem = reset_memory();
4757 // fill in the rest of the null path:
4758 result_io ->init_req(_null_path, i_o());
4759 result_mem->init_req(_null_path, init_mem);
4760
4761 result_val->init_req(_fast_path, hash_val);
4762 result_reg->init_req(_fast_path, control());
4763 result_io ->init_req(_fast_path, i_o());
4764 result_mem->init_req(_fast_path, init_mem);
4765
4766 // Generate code for the slow case. We make a call to hashCode().
4767 set_control(_gvn.transform(slow_region));
4768 if (!stopped()) {
4769 // No need for PreserveJVMState, because we're using up the present state.
4770 set_all_memory(init_mem);
4771 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4772 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4773 Node* slow_result = set_results_for_java_call(slow_call);
4774 // this->control() comes from set_results_for_java_call
4775 result_reg->init_req(_slow_path, control());
4776 result_val->init_req(_slow_path, slow_result);
4777 result_io ->set_req(_slow_path, i_o());
4778 result_mem ->set_req(_slow_path, reset_memory());
4779 }
4780
4781 // Return the combined state.
4782 set_i_o( _gvn.transform(result_io) );
4783 set_all_memory( _gvn.transform(result_mem));
4784
4785 set_result(result_reg, result_val);
4786 return true;
4787 }
4788
4789 //---------------------------inline_native_getClass----------------------------
4790 // public final native Class<?> java.lang.Object.getClass();
4791 //
4792 // Build special case code for calls to getClass on an object.
4793 bool LibraryCallKit::inline_native_getClass() {
4794 Node* obj = null_check_receiver();
4795 if (stopped()) return true;
4796 set_result(load_mirror_from_klass(load_object_klass(obj)));
4797 return true;
4798 }
4799
4800 //-----------------inline_native_Reflection_getCallerClass---------------------
4801 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4802 //
4803 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4804 //
4805 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4806 // in that it must skip particular security frames and checks for
4807 // caller sensitive methods.
4808 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4809 #ifndef PRODUCT
4810 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4811 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4812 }
4813 #endif
4814
4815 if (!jvms()->has_method()) {
4816 #ifndef PRODUCT
4817 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4818 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4819 }
4820 #endif
4821 return false;
4822 }
4823
4824 // Walk back up the JVM state to find the caller at the required
4825 // depth.
4826 JVMState* caller_jvms = jvms();
4827
4828 // Cf. JVM_GetCallerClass
4829 // NOTE: Start the loop at depth 1 because the current JVM state does
4830 // not include the Reflection.getCallerClass() frame.
4831 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4832 ciMethod* m = caller_jvms->method();
4833 switch (n) {
4834 case 0:
4835 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4836 break;
4837 case 1:
4838 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4839 if (!m->caller_sensitive()) {
4840 #ifndef PRODUCT
4841 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4842 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4843 }
4844 #endif
4845 return false; // bail-out; let JVM_GetCallerClass do the work
4846 }
4847 break;
4848 default:
4849 if (!m->is_ignored_by_security_stack_walk()) {
4850 // We have reached the desired frame; return the holder class.
4851 // Acquire method holder as java.lang.Class and push as constant.
4852 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4853 ciInstance* caller_mirror = caller_klass->java_mirror();
4854 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4855
4856 #ifndef PRODUCT
4857 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4858 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());
4859 tty->print_cr(" JVM state at this point:");
4860 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4861 ciMethod* m = jvms()->of_depth(i)->method();
4862 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4863 }
4864 }
4865 #endif
4866 return true;
4867 }
4868 break;
4869 }
4870 }
4871
4872 #ifndef PRODUCT
4873 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4874 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4875 tty->print_cr(" JVM state at this point:");
4876 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4877 ciMethod* m = jvms()->of_depth(i)->method();
4878 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4879 }
4880 }
4881 #endif
4882
4883 return false; // bail-out; let JVM_GetCallerClass do the work
4884 }
4885
4886 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4887 Node* arg = argument(0);
4888 Node* result = nullptr;
4889
4890 switch (id) {
4891 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4892 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4893 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4894 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4895 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
4896 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
4897
4898 case vmIntrinsics::_doubleToLongBits: {
4899 // two paths (plus control) merge in a wood
4900 RegionNode *r = new RegionNode(3);
4901 Node *phi = new PhiNode(r, TypeLong::LONG);
4902
4903 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4904 // Build the boolean node
4905 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4906
4907 // Branch either way.
4908 // NaN case is less traveled, which makes all the difference.
4909 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4910 Node *opt_isnan = _gvn.transform(ifisnan);
4911 assert( opt_isnan->is_If(), "Expect an IfNode");
4912 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4913 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4914
4915 set_control(iftrue);
4916
4917 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4918 Node *slow_result = longcon(nan_bits); // return NaN
4919 phi->init_req(1, _gvn.transform( slow_result ));
4920 r->init_req(1, iftrue);
4921
4922 // Else fall through
4923 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4924 set_control(iffalse);
4925
4926 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4927 r->init_req(2, iffalse);
4928
4929 // Post merge
4930 set_control(_gvn.transform(r));
4931 record_for_igvn(r);
4932
4933 C->set_has_split_ifs(true); // Has chance for split-if optimization
4934 result = phi;
4935 assert(result->bottom_type()->isa_long(), "must be");
4936 break;
4937 }
4938
4939 case vmIntrinsics::_floatToIntBits: {
4940 // two paths (plus control) merge in a wood
4941 RegionNode *r = new RegionNode(3);
4942 Node *phi = new PhiNode(r, TypeInt::INT);
4943
4944 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4945 // Build the boolean node
4946 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4947
4948 // Branch either way.
4949 // NaN case is less traveled, which makes all the difference.
4950 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4951 Node *opt_isnan = _gvn.transform(ifisnan);
4952 assert( opt_isnan->is_If(), "Expect an IfNode");
4953 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4954 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4955
4956 set_control(iftrue);
4957
4958 static const jint nan_bits = 0x7fc00000;
4959 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4960 phi->init_req(1, _gvn.transform( slow_result ));
4961 r->init_req(1, iftrue);
4962
4963 // Else fall through
4964 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4965 set_control(iffalse);
4966
4967 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4968 r->init_req(2, iffalse);
4969
4970 // Post merge
4971 set_control(_gvn.transform(r));
4972 record_for_igvn(r);
4973
4974 C->set_has_split_ifs(true); // Has chance for split-if optimization
4975 result = phi;
4976 assert(result->bottom_type()->isa_int(), "must be");
4977 break;
4978 }
4979
4980 default:
4981 fatal_unexpected_iid(id);
4982 break;
4983 }
4984 set_result(_gvn.transform(result));
4985 return true;
4986 }
4987
4988 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
4989 Node* arg = argument(0);
4990 Node* result = nullptr;
4991
4992 switch (id) {
4993 case vmIntrinsics::_floatIsInfinite:
4994 result = new IsInfiniteFNode(arg);
4995 break;
4996 case vmIntrinsics::_floatIsFinite:
4997 result = new IsFiniteFNode(arg);
4998 break;
4999 case vmIntrinsics::_doubleIsInfinite:
5000 result = new IsInfiniteDNode(arg);
5001 break;
5002 case vmIntrinsics::_doubleIsFinite:
5003 result = new IsFiniteDNode(arg);
5004 break;
5005 default:
5006 fatal_unexpected_iid(id);
5007 break;
5008 }
5009 set_result(_gvn.transform(result));
5010 return true;
5011 }
5012
5013 //----------------------inline_unsafe_copyMemory-------------------------
5014 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5015
5016 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5017 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5018 const Type* base_t = gvn.type(base);
5019
5020 bool in_native = (base_t == TypePtr::NULL_PTR);
5021 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5022 bool is_mixed = !in_heap && !in_native;
5023
5024 if (is_mixed) {
5025 return true; // mixed accesses can touch both on-heap and off-heap memory
5026 }
5027 if (in_heap) {
5028 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5029 if (!is_prim_array) {
5030 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5031 // there's not enough type information available to determine proper memory slice for it.
5032 return true;
5033 }
5034 }
5035 return false;
5036 }
5037
5038 bool LibraryCallKit::inline_unsafe_copyMemory() {
5039 if (callee()->is_static()) return false; // caller must have the capability!
5040 null_check_receiver(); // null-check receiver
5041 if (stopped()) return true;
5042
5043 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5044
5045 Node* src_base = argument(1); // type: oop
5046 Node* src_off = ConvL2X(argument(2)); // type: long
5047 Node* dst_base = argument(4); // type: oop
5048 Node* dst_off = ConvL2X(argument(5)); // type: long
5049 Node* size = ConvL2X(argument(7)); // type: long
5050
5051 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5052 "fieldOffset must be byte-scaled");
5053
5054 Node* src_addr = make_unsafe_address(src_base, src_off);
5055 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5056
5057 Node* thread = _gvn.transform(new ThreadLocalNode());
5058 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5059 BasicType doing_unsafe_access_bt = T_BYTE;
5060 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5061
5062 // update volatile field
5063 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5064
5065 int flags = RC_LEAF | RC_NO_FP;
5066
5067 const TypePtr* dst_type = TypePtr::BOTTOM;
5068
5069 // Adjust memory effects of the runtime call based on input values.
5070 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5071 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5072 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5073
5074 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5075 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5076 flags |= RC_NARROW_MEM; // narrow in memory
5077 }
5078 }
5079
5080 // Call it. Note that the length argument is not scaled.
5081 make_runtime_call(flags,
5082 OptoRuntime::fast_arraycopy_Type(),
5083 StubRoutines::unsafe_arraycopy(),
5084 "unsafe_arraycopy",
5085 dst_type,
5086 src_addr, dst_addr, size XTOP);
5087
5088 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5089
5090 return true;
5091 }
5092
5093 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5094 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5095 bool LibraryCallKit::inline_unsafe_setMemory() {
5096 if (callee()->is_static()) return false; // caller must have the capability!
5097 null_check_receiver(); // null-check receiver
5098 if (stopped()) return true;
5099
5100 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5101
5102 Node* dst_base = argument(1); // type: oop
5103 Node* dst_off = ConvL2X(argument(2)); // type: long
5104 Node* size = ConvL2X(argument(4)); // type: long
5105 Node* byte = argument(6); // type: byte
5106
5107 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5108 "fieldOffset must be byte-scaled");
5109
5110 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5111
5112 Node* thread = _gvn.transform(new ThreadLocalNode());
5113 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5114 BasicType doing_unsafe_access_bt = T_BYTE;
5115 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5116
5117 // update volatile field
5118 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5119
5120 int flags = RC_LEAF | RC_NO_FP;
5121
5122 const TypePtr* dst_type = TypePtr::BOTTOM;
5123
5124 // Adjust memory effects of the runtime call based on input values.
5125 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5126 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5127
5128 flags |= RC_NARROW_MEM; // narrow in memory
5129 }
5130
5131 // Call it. Note that the length argument is not scaled.
5132 make_runtime_call(flags,
5133 OptoRuntime::make_setmemory_Type(),
5134 StubRoutines::unsafe_setmemory(),
5135 "unsafe_setmemory",
5136 dst_type,
5137 dst_addr, size XTOP, byte);
5138
5139 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5140
5141 return true;
5142 }
5143
5144 #undef XTOP
5145
5146 //------------------------clone_coping-----------------------------------
5147 // Helper function for inline_native_clone.
5148 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5149 assert(obj_size != nullptr, "");
5150 Node* raw_obj = alloc_obj->in(1);
5151 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5152
5153 AllocateNode* alloc = nullptr;
5154 if (ReduceBulkZeroing &&
5155 // If we are implementing an array clone without knowing its source type
5156 // (can happen when compiling the array-guarded branch of a reflective
5157 // Object.clone() invocation), initialize the array within the allocation.
5158 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5159 // to a runtime clone call that assumes fully initialized source arrays.
5160 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5161 // We will be completely responsible for initializing this object -
5162 // mark Initialize node as complete.
5163 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5164 // The object was just allocated - there should be no any stores!
5165 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5166 // Mark as complete_with_arraycopy so that on AllocateNode
5167 // expansion, we know this AllocateNode is initialized by an array
5168 // copy and a StoreStore barrier exists after the array copy.
5169 alloc->initialization()->set_complete_with_arraycopy();
5170 }
5171
5172 Node* size = _gvn.transform(obj_size);
5173 access_clone(obj, alloc_obj, size, is_array);
5174
5175 // Do not let reads from the cloned object float above the arraycopy.
5176 if (alloc != nullptr) {
5177 // Do not let stores that initialize this object be reordered with
5178 // a subsequent store that would make this object accessible by
5179 // other threads.
5180 // Record what AllocateNode this StoreStore protects so that
5181 // escape analysis can go from the MemBarStoreStoreNode to the
5182 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5183 // based on the escape status of the AllocateNode.
5184 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5185 } else {
5186 insert_mem_bar(Op_MemBarCPUOrder);
5187 }
5188 }
5189
5190 //------------------------inline_native_clone----------------------------
5191 // protected native Object java.lang.Object.clone();
5192 //
5193 // Here are the simple edge cases:
5194 // null receiver => normal trap
5195 // virtual and clone was overridden => slow path to out-of-line clone
5196 // not cloneable or finalizer => slow path to out-of-line Object.clone
5197 //
5198 // The general case has two steps, allocation and copying.
5199 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5200 //
5201 // Copying also has two cases, oop arrays and everything else.
5202 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5203 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5204 //
5205 // These steps fold up nicely if and when the cloned object's klass
5206 // can be sharply typed as an object array, a type array, or an instance.
5207 //
5208 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5209 PhiNode* result_val;
5210
5211 // Set the reexecute bit for the interpreter to reexecute
5212 // the bytecode that invokes Object.clone if deoptimization happens.
5213 { PreserveReexecuteState preexecs(this);
5214 jvms()->set_should_reexecute(true);
5215
5216 Node* obj = null_check_receiver();
5217 if (stopped()) return true;
5218
5219 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5220
5221 // If we are going to clone an instance, we need its exact type to
5222 // know the number and types of fields to convert the clone to
5223 // loads/stores. Maybe a speculative type can help us.
5224 if (!obj_type->klass_is_exact() &&
5225 obj_type->speculative_type() != nullptr &&
5226 obj_type->speculative_type()->is_instance_klass()) {
5227 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5228 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5229 !spec_ik->has_injected_fields()) {
5230 if (!obj_type->isa_instptr() ||
5231 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5232 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5233 }
5234 }
5235 }
5236
5237 // Conservatively insert a memory barrier on all memory slices.
5238 // Do not let writes into the original float below the clone.
5239 insert_mem_bar(Op_MemBarCPUOrder);
5240
5241 // paths into result_reg:
5242 enum {
5243 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5244 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5245 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5246 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5247 PATH_LIMIT
5248 };
5249 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5250 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5251 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5252 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5253 record_for_igvn(result_reg);
5254
5255 Node* obj_klass = load_object_klass(obj);
5256 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr);
5257 if (array_ctl != nullptr) {
5258 // It's an array.
5259 PreserveJVMState pjvms(this);
5260 set_control(array_ctl);
5261 Node* obj_length = load_array_length(obj);
5262 Node* array_size = nullptr; // Size of the array without object alignment padding.
5263 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5264
5265 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5266 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5267 // If it is an oop array, it requires very special treatment,
5268 // because gc barriers are required when accessing the array.
5269 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5270 if (is_obja != nullptr) {
5271 PreserveJVMState pjvms2(this);
5272 set_control(is_obja);
5273 // Generate a direct call to the right arraycopy function(s).
5274 // Clones are always tightly coupled.
5275 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5276 ac->set_clone_oop_array();
5277 Node* n = _gvn.transform(ac);
5278 assert(n == ac, "cannot disappear");
5279 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5280
5281 result_reg->init_req(_objArray_path, control());
5282 result_val->init_req(_objArray_path, alloc_obj);
5283 result_i_o ->set_req(_objArray_path, i_o());
5284 result_mem ->set_req(_objArray_path, reset_memory());
5285 }
5286 }
5287 // Otherwise, there are no barriers to worry about.
5288 // (We can dispense with card marks if we know the allocation
5289 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5290 // causes the non-eden paths to take compensating steps to
5291 // simulate a fresh allocation, so that no further
5292 // card marks are required in compiled code to initialize
5293 // the object.)
5294
5295 if (!stopped()) {
5296 copy_to_clone(obj, alloc_obj, array_size, true);
5297
5298 // Present the results of the copy.
5299 result_reg->init_req(_array_path, control());
5300 result_val->init_req(_array_path, alloc_obj);
5301 result_i_o ->set_req(_array_path, i_o());
5302 result_mem ->set_req(_array_path, reset_memory());
5303 }
5304 }
5305
5306 // We only go to the instance fast case code if we pass a number of guards.
5307 // The paths which do not pass are accumulated in the slow_region.
5308 RegionNode* slow_region = new RegionNode(1);
5309 record_for_igvn(slow_region);
5310 if (!stopped()) {
5311 // It's an instance (we did array above). Make the slow-path tests.
5312 // If this is a virtual call, we generate a funny guard. We grab
5313 // the vtable entry corresponding to clone() from the target object.
5314 // If the target method which we are calling happens to be the
5315 // Object clone() method, we pass the guard. We do not need this
5316 // guard for non-virtual calls; the caller is known to be the native
5317 // Object clone().
5318 if (is_virtual) {
5319 generate_virtual_guard(obj_klass, slow_region);
5320 }
5321
5322 // The object must be easily cloneable and must not have a finalizer.
5323 // Both of these conditions may be checked in a single test.
5324 // We could optimize the test further, but we don't care.
5325 generate_misc_flags_guard(obj_klass,
5326 // Test both conditions:
5327 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5328 // Must be cloneable but not finalizer:
5329 KlassFlags::_misc_is_cloneable_fast,
5330 slow_region);
5331 }
5332
5333 if (!stopped()) {
5334 // It's an instance, and it passed the slow-path tests.
5335 PreserveJVMState pjvms(this);
5336 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5337 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5338 // is reexecuted if deoptimization occurs and there could be problems when merging
5339 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5340 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5341
5342 copy_to_clone(obj, alloc_obj, obj_size, false);
5343
5344 // Present the results of the slow call.
5345 result_reg->init_req(_instance_path, control());
5346 result_val->init_req(_instance_path, alloc_obj);
5347 result_i_o ->set_req(_instance_path, i_o());
5348 result_mem ->set_req(_instance_path, reset_memory());
5349 }
5350
5351 // Generate code for the slow case. We make a call to clone().
5352 set_control(_gvn.transform(slow_region));
5353 if (!stopped()) {
5354 PreserveJVMState pjvms(this);
5355 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5356 // We need to deoptimize on exception (see comment above)
5357 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5358 // this->control() comes from set_results_for_java_call
5359 result_reg->init_req(_slow_path, control());
5360 result_val->init_req(_slow_path, slow_result);
5361 result_i_o ->set_req(_slow_path, i_o());
5362 result_mem ->set_req(_slow_path, reset_memory());
5363 }
5364
5365 // Return the combined state.
5366 set_control( _gvn.transform(result_reg));
5367 set_i_o( _gvn.transform(result_i_o));
5368 set_all_memory( _gvn.transform(result_mem));
5369 } // original reexecute is set back here
5370
5371 set_result(_gvn.transform(result_val));
5372 return true;
5373 }
5374
5375 // If we have a tightly coupled allocation, the arraycopy may take care
5376 // of the array initialization. If one of the guards we insert between
5377 // the allocation and the arraycopy causes a deoptimization, an
5378 // uninitialized array will escape the compiled method. To prevent that
5379 // we set the JVM state for uncommon traps between the allocation and
5380 // the arraycopy to the state before the allocation so, in case of
5381 // deoptimization, we'll reexecute the allocation and the
5382 // initialization.
5383 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5384 if (alloc != nullptr) {
5385 ciMethod* trap_method = alloc->jvms()->method();
5386 int trap_bci = alloc->jvms()->bci();
5387
5388 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5389 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5390 // Make sure there's no store between the allocation and the
5391 // arraycopy otherwise visible side effects could be rexecuted
5392 // in case of deoptimization and cause incorrect execution.
5393 bool no_interfering_store = true;
5394 Node* mem = alloc->in(TypeFunc::Memory);
5395 if (mem->is_MergeMem()) {
5396 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5397 Node* n = mms.memory();
5398 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5399 assert(n->is_Store(), "what else?");
5400 no_interfering_store = false;
5401 break;
5402 }
5403 }
5404 } else {
5405 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5406 Node* n = mms.memory();
5407 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5408 assert(n->is_Store(), "what else?");
5409 no_interfering_store = false;
5410 break;
5411 }
5412 }
5413 }
5414
5415 if (no_interfering_store) {
5416 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5417
5418 JVMState* saved_jvms = jvms();
5419 saved_reexecute_sp = _reexecute_sp;
5420
5421 set_jvms(sfpt->jvms());
5422 _reexecute_sp = jvms()->sp();
5423
5424 return saved_jvms;
5425 }
5426 }
5427 }
5428 return nullptr;
5429 }
5430
5431 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5432 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5433 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5434 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5435 uint size = alloc->req();
5436 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5437 old_jvms->set_map(sfpt);
5438 for (uint i = 0; i < size; i++) {
5439 sfpt->init_req(i, alloc->in(i));
5440 }
5441 // re-push array length for deoptimization
5442 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5443 old_jvms->set_sp(old_jvms->sp()+1);
5444 old_jvms->set_monoff(old_jvms->monoff()+1);
5445 old_jvms->set_scloff(old_jvms->scloff()+1);
5446 old_jvms->set_endoff(old_jvms->endoff()+1);
5447 old_jvms->set_should_reexecute(true);
5448
5449 sfpt->set_i_o(map()->i_o());
5450 sfpt->set_memory(map()->memory());
5451 sfpt->set_control(map()->control());
5452 return sfpt;
5453 }
5454
5455 // In case of a deoptimization, we restart execution at the
5456 // allocation, allocating a new array. We would leave an uninitialized
5457 // array in the heap that GCs wouldn't expect. Move the allocation
5458 // after the traps so we don't allocate the array if we
5459 // deoptimize. This is possible because tightly_coupled_allocation()
5460 // guarantees there's no observer of the allocated array at this point
5461 // and the control flow is simple enough.
5462 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5463 int saved_reexecute_sp, uint new_idx) {
5464 if (saved_jvms_before_guards != nullptr && !stopped()) {
5465 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5466
5467 assert(alloc != nullptr, "only with a tightly coupled allocation");
5468 // restore JVM state to the state at the arraycopy
5469 saved_jvms_before_guards->map()->set_control(map()->control());
5470 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5471 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5472 // If we've improved the types of some nodes (null check) while
5473 // emitting the guards, propagate them to the current state
5474 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5475 set_jvms(saved_jvms_before_guards);
5476 _reexecute_sp = saved_reexecute_sp;
5477
5478 // Remove the allocation from above the guards
5479 CallProjections callprojs;
5480 alloc->extract_projections(&callprojs, true);
5481 InitializeNode* init = alloc->initialization();
5482 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5483 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5484 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5485
5486 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5487 // the allocation (i.e. is only valid if the allocation succeeds):
5488 // 1) replace CastIINode with AllocateArrayNode's length here
5489 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5490 //
5491 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5492 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5493 Node* init_control = init->proj_out(TypeFunc::Control);
5494 Node* alloc_length = alloc->Ideal_length();
5495 #ifdef ASSERT
5496 Node* prev_cast = nullptr;
5497 #endif
5498 for (uint i = 0; i < init_control->outcnt(); i++) {
5499 Node* init_out = init_control->raw_out(i);
5500 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5501 #ifdef ASSERT
5502 if (prev_cast == nullptr) {
5503 prev_cast = init_out;
5504 } else {
5505 if (prev_cast->cmp(*init_out) == false) {
5506 prev_cast->dump();
5507 init_out->dump();
5508 assert(false, "not equal CastIINode");
5509 }
5510 }
5511 #endif
5512 C->gvn_replace_by(init_out, alloc_length);
5513 }
5514 }
5515 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5516
5517 // move the allocation here (after the guards)
5518 _gvn.hash_delete(alloc);
5519 alloc->set_req(TypeFunc::Control, control());
5520 alloc->set_req(TypeFunc::I_O, i_o());
5521 Node *mem = reset_memory();
5522 set_all_memory(mem);
5523 alloc->set_req(TypeFunc::Memory, mem);
5524 set_control(init->proj_out_or_null(TypeFunc::Control));
5525 set_i_o(callprojs.fallthrough_ioproj);
5526
5527 // Update memory as done in GraphKit::set_output_for_allocation()
5528 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5529 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5530 if (ary_type->isa_aryptr() && length_type != nullptr) {
5531 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5532 }
5533 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5534 int elemidx = C->get_alias_index(telemref);
5535 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5536 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5537
5538 Node* allocx = _gvn.transform(alloc);
5539 assert(allocx == alloc, "where has the allocation gone?");
5540 assert(dest->is_CheckCastPP(), "not an allocation result?");
5541
5542 _gvn.hash_delete(dest);
5543 dest->set_req(0, control());
5544 Node* destx = _gvn.transform(dest);
5545 assert(destx == dest, "where has the allocation result gone?");
5546
5547 array_ideal_length(alloc, ary_type, true);
5548 }
5549 }
5550
5551 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5552 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5553 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5554 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5555 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5556 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5557 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5558 JVMState* saved_jvms_before_guards) {
5559 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5560 // There is at least one unrelated uncommon trap which needs to be replaced.
5561 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5562
5563 JVMState* saved_jvms = jvms();
5564 const int saved_reexecute_sp = _reexecute_sp;
5565 set_jvms(sfpt->jvms());
5566 _reexecute_sp = jvms()->sp();
5567
5568 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5569
5570 // Restore state
5571 set_jvms(saved_jvms);
5572 _reexecute_sp = saved_reexecute_sp;
5573 }
5574 }
5575
5576 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5577 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5578 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5579 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5580 while (if_proj->is_IfProj()) {
5581 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5582 if (uncommon_trap != nullptr) {
5583 create_new_uncommon_trap(uncommon_trap);
5584 }
5585 assert(if_proj->in(0)->is_If(), "must be If");
5586 if_proj = if_proj->in(0)->in(0);
5587 }
5588 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5589 "must have reached control projection of init node");
5590 }
5591
5592 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5593 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5594 assert(trap_request != 0, "no valid UCT trap request");
5595 PreserveJVMState pjvms(this);
5596 set_control(uncommon_trap_call->in(0));
5597 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5598 Deoptimization::trap_request_action(trap_request));
5599 assert(stopped(), "Should be stopped");
5600 _gvn.hash_delete(uncommon_trap_call);
5601 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5602 }
5603
5604 // Common checks for array sorting intrinsics arguments.
5605 // Returns `true` if checks passed.
5606 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
5607 // check address of the class
5608 if (elementType == nullptr || elementType->is_top()) {
5609 return false; // dead path
5610 }
5611 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5612 if (elem_klass == nullptr) {
5613 return false; // dead path
5614 }
5615 // java_mirror_type() returns non-null for compile-time Class constants only
5616 ciType* elem_type = elem_klass->java_mirror_type();
5617 if (elem_type == nullptr) {
5618 return false;
5619 }
5620 bt = elem_type->basic_type();
5621 // Disable the intrinsic if the CPU does not support SIMD sort
5622 if (!Matcher::supports_simd_sort(bt)) {
5623 return false;
5624 }
5625 // check address of the array
5626 if (obj == nullptr || obj->is_top()) {
5627 return false; // dead path
5628 }
5629 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5630 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
5631 return false; // failed input validation
5632 }
5633 return true;
5634 }
5635
5636 //------------------------------inline_array_partition-----------------------
5637 bool LibraryCallKit::inline_array_partition() {
5638 address stubAddr = StubRoutines::select_array_partition_function();
5639 if (stubAddr == nullptr) {
5640 return false; // Intrinsic's stub is not implemented on this platform
5641 }
5642 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
5643
5644 // no receiver because it is a static method
5645 Node* elementType = argument(0);
5646 Node* obj = argument(1);
5647 Node* offset = argument(2); // long
5648 Node* fromIndex = argument(4);
5649 Node* toIndex = argument(5);
5650 Node* indexPivot1 = argument(6);
5651 Node* indexPivot2 = argument(7);
5652 // PartitionOperation: argument(8) is ignored
5653
5654 Node* pivotIndices = nullptr;
5655 BasicType bt = T_ILLEGAL;
5656
5657 if (!check_array_sort_arguments(elementType, obj, bt)) {
5658 return false;
5659 }
5660 null_check(obj);
5661 // If obj is dead, only null-path is taken.
5662 if (stopped()) {
5663 return true;
5664 }
5665 // Set the original stack and the reexecute bit for the interpreter to reexecute
5666 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
5667 { PreserveReexecuteState preexecs(this);
5668 jvms()->set_should_reexecute(true);
5669
5670 Node* obj_adr = make_unsafe_address(obj, offset);
5671
5672 // create the pivotIndices array of type int and size = 2
5673 Node* size = intcon(2);
5674 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
5675 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
5676 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
5677 guarantee(alloc != nullptr, "created above");
5678 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
5679
5680 // pass the basic type enum to the stub
5681 Node* elemType = intcon(bt);
5682
5683 // Call the stub
5684 const char *stubName = "array_partition_stub";
5685 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
5686 stubAddr, stubName, TypePtr::BOTTOM,
5687 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
5688 indexPivot1, indexPivot2);
5689
5690 } // original reexecute is set back here
5691
5692 if (!stopped()) {
5693 set_result(pivotIndices);
5694 }
5695
5696 return true;
5697 }
5698
5699
5700 //------------------------------inline_array_sort-----------------------
5701 bool LibraryCallKit::inline_array_sort() {
5702 address stubAddr = StubRoutines::select_arraysort_function();
5703 if (stubAddr == nullptr) {
5704 return false; // Intrinsic's stub is not implemented on this platform
5705 }
5706 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
5707
5708 // no receiver because it is a static method
5709 Node* elementType = argument(0);
5710 Node* obj = argument(1);
5711 Node* offset = argument(2); // long
5712 Node* fromIndex = argument(4);
5713 Node* toIndex = argument(5);
5714 // SortOperation: argument(6) is ignored
5715
5716 BasicType bt = T_ILLEGAL;
5717
5718 if (!check_array_sort_arguments(elementType, obj, bt)) {
5719 return false;
5720 }
5721 null_check(obj);
5722 // If obj is dead, only null-path is taken.
5723 if (stopped()) {
5724 return true;
5725 }
5726 Node* obj_adr = make_unsafe_address(obj, offset);
5727
5728 // pass the basic type enum to the stub
5729 Node* elemType = intcon(bt);
5730
5731 // Call the stub.
5732 const char *stubName = "arraysort_stub";
5733 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
5734 stubAddr, stubName, TypePtr::BOTTOM,
5735 obj_adr, elemType, fromIndex, toIndex);
5736
5737 return true;
5738 }
5739
5740
5741 //------------------------------inline_arraycopy-----------------------
5742 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5743 // Object dest, int destPos,
5744 // int length);
5745 bool LibraryCallKit::inline_arraycopy() {
5746 // Get the arguments.
5747 Node* src = argument(0); // type: oop
5748 Node* src_offset = argument(1); // type: int
5749 Node* dest = argument(2); // type: oop
5750 Node* dest_offset = argument(3); // type: int
5751 Node* length = argument(4); // type: int
5752
5753 uint new_idx = C->unique();
5754
5755 // Check for allocation before we add nodes that would confuse
5756 // tightly_coupled_allocation()
5757 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5758
5759 int saved_reexecute_sp = -1;
5760 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5761 // See arraycopy_restore_alloc_state() comment
5762 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5763 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5764 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5765 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5766
5767 // The following tests must be performed
5768 // (1) src and dest are arrays.
5769 // (2) src and dest arrays must have elements of the same BasicType
5770 // (3) src and dest must not be null.
5771 // (4) src_offset must not be negative.
5772 // (5) dest_offset must not be negative.
5773 // (6) length must not be negative.
5774 // (7) src_offset + length must not exceed length of src.
5775 // (8) dest_offset + length must not exceed length of dest.
5776 // (9) each element of an oop array must be assignable
5777
5778 // (3) src and dest must not be null.
5779 // always do this here because we need the JVM state for uncommon traps
5780 Node* null_ctl = top();
5781 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5782 assert(null_ctl->is_top(), "no null control here");
5783 dest = null_check(dest, T_ARRAY);
5784
5785 if (!can_emit_guards) {
5786 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5787 // guards but the arraycopy node could still take advantage of a
5788 // tightly allocated allocation. tightly_coupled_allocation() is
5789 // called again to make sure it takes the null check above into
5790 // account: the null check is mandatory and if it caused an
5791 // uncommon trap to be emitted then the allocation can't be
5792 // considered tightly coupled in this context.
5793 alloc = tightly_coupled_allocation(dest);
5794 }
5795
5796 bool validated = false;
5797
5798 const Type* src_type = _gvn.type(src);
5799 const Type* dest_type = _gvn.type(dest);
5800 const TypeAryPtr* top_src = src_type->isa_aryptr();
5801 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5802
5803 // Do we have the type of src?
5804 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5805 // Do we have the type of dest?
5806 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5807 // Is the type for src from speculation?
5808 bool src_spec = false;
5809 // Is the type for dest from speculation?
5810 bool dest_spec = false;
5811
5812 if ((!has_src || !has_dest) && can_emit_guards) {
5813 // We don't have sufficient type information, let's see if
5814 // speculative types can help. We need to have types for both src
5815 // and dest so that it pays off.
5816
5817 // Do we already have or could we have type information for src
5818 bool could_have_src = has_src;
5819 // Do we already have or could we have type information for dest
5820 bool could_have_dest = has_dest;
5821
5822 ciKlass* src_k = nullptr;
5823 if (!has_src) {
5824 src_k = src_type->speculative_type_not_null();
5825 if (src_k != nullptr && src_k->is_array_klass()) {
5826 could_have_src = true;
5827 }
5828 }
5829
5830 ciKlass* dest_k = nullptr;
5831 if (!has_dest) {
5832 dest_k = dest_type->speculative_type_not_null();
5833 if (dest_k != nullptr && dest_k->is_array_klass()) {
5834 could_have_dest = true;
5835 }
5836 }
5837
5838 if (could_have_src && could_have_dest) {
5839 // This is going to pay off so emit the required guards
5840 if (!has_src) {
5841 src = maybe_cast_profiled_obj(src, src_k, true);
5842 src_type = _gvn.type(src);
5843 top_src = src_type->isa_aryptr();
5844 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5845 src_spec = true;
5846 }
5847 if (!has_dest) {
5848 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5849 dest_type = _gvn.type(dest);
5850 top_dest = dest_type->isa_aryptr();
5851 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5852 dest_spec = true;
5853 }
5854 }
5855 }
5856
5857 if (has_src && has_dest && can_emit_guards) {
5858 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5859 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5860 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5861 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5862
5863 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5864 // If both arrays are object arrays then having the exact types
5865 // for both will remove the need for a subtype check at runtime
5866 // before the call and may make it possible to pick a faster copy
5867 // routine (without a subtype check on every element)
5868 // Do we have the exact type of src?
5869 bool could_have_src = src_spec;
5870 // Do we have the exact type of dest?
5871 bool could_have_dest = dest_spec;
5872 ciKlass* src_k = nullptr;
5873 ciKlass* dest_k = nullptr;
5874 if (!src_spec) {
5875 src_k = src_type->speculative_type_not_null();
5876 if (src_k != nullptr && src_k->is_array_klass()) {
5877 could_have_src = true;
5878 }
5879 }
5880 if (!dest_spec) {
5881 dest_k = dest_type->speculative_type_not_null();
5882 if (dest_k != nullptr && dest_k->is_array_klass()) {
5883 could_have_dest = true;
5884 }
5885 }
5886 if (could_have_src && could_have_dest) {
5887 // If we can have both exact types, emit the missing guards
5888 if (could_have_src && !src_spec) {
5889 src = maybe_cast_profiled_obj(src, src_k, true);
5890 }
5891 if (could_have_dest && !dest_spec) {
5892 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5893 }
5894 }
5895 }
5896 }
5897
5898 ciMethod* trap_method = method();
5899 int trap_bci = bci();
5900 if (saved_jvms_before_guards != nullptr) {
5901 trap_method = alloc->jvms()->method();
5902 trap_bci = alloc->jvms()->bci();
5903 }
5904
5905 bool negative_length_guard_generated = false;
5906
5907 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5908 can_emit_guards &&
5909 !src->is_top() && !dest->is_top()) {
5910 // validate arguments: enables transformation the ArrayCopyNode
5911 validated = true;
5912
5913 RegionNode* slow_region = new RegionNode(1);
5914 record_for_igvn(slow_region);
5915
5916 // (1) src and dest are arrays.
5917 generate_non_array_guard(load_object_klass(src), slow_region);
5918 generate_non_array_guard(load_object_klass(dest), slow_region);
5919
5920 // (2) src and dest arrays must have elements of the same BasicType
5921 // done at macro expansion or at Ideal transformation time
5922
5923 // (4) src_offset must not be negative.
5924 generate_negative_guard(src_offset, slow_region);
5925
5926 // (5) dest_offset must not be negative.
5927 generate_negative_guard(dest_offset, slow_region);
5928
5929 // (7) src_offset + length must not exceed length of src.
5930 generate_limit_guard(src_offset, length,
5931 load_array_length(src),
5932 slow_region);
5933
5934 // (8) dest_offset + length must not exceed length of dest.
5935 generate_limit_guard(dest_offset, length,
5936 load_array_length(dest),
5937 slow_region);
5938
5939 // (6) length must not be negative.
5940 // This is also checked in generate_arraycopy() during macro expansion, but
5941 // we also have to check it here for the case where the ArrayCopyNode will
5942 // be eliminated by Escape Analysis.
5943 if (EliminateAllocations) {
5944 generate_negative_guard(length, slow_region);
5945 negative_length_guard_generated = true;
5946 }
5947
5948 // (9) each element of an oop array must be assignable
5949 Node* dest_klass = load_object_klass(dest);
5950 if (src != dest) {
5951 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
5952
5953 if (not_subtype_ctrl != top()) {
5954 PreserveJVMState pjvms(this);
5955 set_control(not_subtype_ctrl);
5956 uncommon_trap(Deoptimization::Reason_intrinsic,
5957 Deoptimization::Action_make_not_entrant);
5958 assert(stopped(), "Should be stopped");
5959 }
5960 }
5961 {
5962 PreserveJVMState pjvms(this);
5963 set_control(_gvn.transform(slow_region));
5964 uncommon_trap(Deoptimization::Reason_intrinsic,
5965 Deoptimization::Action_make_not_entrant);
5966 assert(stopped(), "Should be stopped");
5967 }
5968
5969 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5970 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
5971 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5972 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
5973 }
5974
5975 if (stopped()) {
5976 return true;
5977 }
5978
5979 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
5980 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5981 // so the compiler has a chance to eliminate them: during macro expansion,
5982 // we have to set their control (CastPP nodes are eliminated).
5983 load_object_klass(src), load_object_klass(dest),
5984 load_array_length(src), load_array_length(dest));
5985
5986 ac->set_arraycopy(validated);
5987
5988 Node* n = _gvn.transform(ac);
5989 if (n == ac) {
5990 ac->connect_outputs(this);
5991 } else {
5992 assert(validated, "shouldn't transform if all arguments not validated");
5993 set_all_memory(n);
5994 }
5995 clear_upper_avx();
5996
5997
5998 return true;
5999 }
6000
6001
6002 // Helper function which determines if an arraycopy immediately follows
6003 // an allocation, with no intervening tests or other escapes for the object.
6004 AllocateArrayNode*
6005 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6006 if (stopped()) return nullptr; // no fast path
6007 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6008
6009 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6010 if (alloc == nullptr) return nullptr;
6011
6012 Node* rawmem = memory(Compile::AliasIdxRaw);
6013 // Is the allocation's memory state untouched?
6014 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6015 // Bail out if there have been raw-memory effects since the allocation.
6016 // (Example: There might have been a call or safepoint.)
6017 return nullptr;
6018 }
6019 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6020 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6021 return nullptr;
6022 }
6023
6024 // There must be no unexpected observers of this allocation.
6025 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6026 Node* obs = ptr->fast_out(i);
6027 if (obs != this->map()) {
6028 return nullptr;
6029 }
6030 }
6031
6032 // This arraycopy must unconditionally follow the allocation of the ptr.
6033 Node* alloc_ctl = ptr->in(0);
6034 Node* ctl = control();
6035 while (ctl != alloc_ctl) {
6036 // There may be guards which feed into the slow_region.
6037 // Any other control flow means that we might not get a chance
6038 // to finish initializing the allocated object.
6039 // Various low-level checks bottom out in uncommon traps. These
6040 // are considered safe since we've already checked above that
6041 // there is no unexpected observer of this allocation.
6042 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6043 assert(ctl->in(0)->is_If(), "must be If");
6044 ctl = ctl->in(0)->in(0);
6045 } else {
6046 return nullptr;
6047 }
6048 }
6049
6050 // If we get this far, we have an allocation which immediately
6051 // precedes the arraycopy, and we can take over zeroing the new object.
6052 // The arraycopy will finish the initialization, and provide
6053 // a new control state to which we will anchor the destination pointer.
6054
6055 return alloc;
6056 }
6057
6058 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6059 if (node->is_IfProj()) {
6060 Node* other_proj = node->as_IfProj()->other_if_proj();
6061 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6062 Node* obs = other_proj->fast_out(j);
6063 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6064 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6065 return obs->as_CallStaticJava();
6066 }
6067 }
6068 }
6069 return nullptr;
6070 }
6071
6072 //-------------inline_encodeISOArray-----------------------------------
6073 // encode char[] to byte[] in ISO_8859_1 or ASCII
6074 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6075 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6076 // no receiver since it is static method
6077 Node *src = argument(0);
6078 Node *src_offset = argument(1);
6079 Node *dst = argument(2);
6080 Node *dst_offset = argument(3);
6081 Node *length = argument(4);
6082
6083 src = must_be_not_null(src, true);
6084 dst = must_be_not_null(dst, true);
6085
6086 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6087 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6088 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6089 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6090 // failed array check
6091 return false;
6092 }
6093
6094 // Figure out the size and type of the elements we will be copying.
6095 BasicType src_elem = src_type->elem()->array_element_basic_type();
6096 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6097 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6098 return false;
6099 }
6100
6101 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6102 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6103 // 'src_start' points to src array + scaled offset
6104 // 'dst_start' points to dst array + scaled offset
6105
6106 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6107 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6108 enc = _gvn.transform(enc);
6109 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6110 set_memory(res_mem, mtype);
6111 set_result(enc);
6112 clear_upper_avx();
6113
6114 return true;
6115 }
6116
6117 //-------------inline_multiplyToLen-----------------------------------
6118 bool LibraryCallKit::inline_multiplyToLen() {
6119 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6120
6121 address stubAddr = StubRoutines::multiplyToLen();
6122 if (stubAddr == nullptr) {
6123 return false; // Intrinsic's stub is not implemented on this platform
6124 }
6125 const char* stubName = "multiplyToLen";
6126
6127 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6128
6129 // no receiver because it is a static method
6130 Node* x = argument(0);
6131 Node* xlen = argument(1);
6132 Node* y = argument(2);
6133 Node* ylen = argument(3);
6134 Node* z = argument(4);
6135
6136 x = must_be_not_null(x, true);
6137 y = must_be_not_null(y, true);
6138
6139 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6140 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6141 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6142 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6143 // failed array check
6144 return false;
6145 }
6146
6147 BasicType x_elem = x_type->elem()->array_element_basic_type();
6148 BasicType y_elem = y_type->elem()->array_element_basic_type();
6149 if (x_elem != T_INT || y_elem != T_INT) {
6150 return false;
6151 }
6152
6153 Node* x_start = array_element_address(x, intcon(0), x_elem);
6154 Node* y_start = array_element_address(y, intcon(0), y_elem);
6155 // 'x_start' points to x array + scaled xlen
6156 // 'y_start' points to y array + scaled ylen
6157
6158 Node* z_start = array_element_address(z, intcon(0), T_INT);
6159
6160 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6161 OptoRuntime::multiplyToLen_Type(),
6162 stubAddr, stubName, TypePtr::BOTTOM,
6163 x_start, xlen, y_start, ylen, z_start);
6164
6165 C->set_has_split_ifs(true); // Has chance for split-if optimization
6166 set_result(z);
6167 return true;
6168 }
6169
6170 //-------------inline_squareToLen------------------------------------
6171 bool LibraryCallKit::inline_squareToLen() {
6172 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6173
6174 address stubAddr = StubRoutines::squareToLen();
6175 if (stubAddr == nullptr) {
6176 return false; // Intrinsic's stub is not implemented on this platform
6177 }
6178 const char* stubName = "squareToLen";
6179
6180 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6181
6182 Node* x = argument(0);
6183 Node* len = argument(1);
6184 Node* z = argument(2);
6185 Node* zlen = argument(3);
6186
6187 x = must_be_not_null(x, true);
6188 z = must_be_not_null(z, true);
6189
6190 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6191 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6192 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6193 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6194 // failed array check
6195 return false;
6196 }
6197
6198 BasicType x_elem = x_type->elem()->array_element_basic_type();
6199 BasicType z_elem = z_type->elem()->array_element_basic_type();
6200 if (x_elem != T_INT || z_elem != T_INT) {
6201 return false;
6202 }
6203
6204
6205 Node* x_start = array_element_address(x, intcon(0), x_elem);
6206 Node* z_start = array_element_address(z, intcon(0), z_elem);
6207
6208 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6209 OptoRuntime::squareToLen_Type(),
6210 stubAddr, stubName, TypePtr::BOTTOM,
6211 x_start, len, z_start, zlen);
6212
6213 set_result(z);
6214 return true;
6215 }
6216
6217 //-------------inline_mulAdd------------------------------------------
6218 bool LibraryCallKit::inline_mulAdd() {
6219 assert(UseMulAddIntrinsic, "not implemented on this platform");
6220
6221 address stubAddr = StubRoutines::mulAdd();
6222 if (stubAddr == nullptr) {
6223 return false; // Intrinsic's stub is not implemented on this platform
6224 }
6225 const char* stubName = "mulAdd";
6226
6227 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6228
6229 Node* out = argument(0);
6230 Node* in = argument(1);
6231 Node* offset = argument(2);
6232 Node* len = argument(3);
6233 Node* k = argument(4);
6234
6235 in = must_be_not_null(in, true);
6236 out = must_be_not_null(out, true);
6237
6238 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6239 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6240 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6241 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6242 // failed array check
6243 return false;
6244 }
6245
6246 BasicType out_elem = out_type->elem()->array_element_basic_type();
6247 BasicType in_elem = in_type->elem()->array_element_basic_type();
6248 if (out_elem != T_INT || in_elem != T_INT) {
6249 return false;
6250 }
6251
6252 Node* outlen = load_array_length(out);
6253 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6254 Node* out_start = array_element_address(out, intcon(0), out_elem);
6255 Node* in_start = array_element_address(in, intcon(0), in_elem);
6256
6257 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6258 OptoRuntime::mulAdd_Type(),
6259 stubAddr, stubName, TypePtr::BOTTOM,
6260 out_start,in_start, new_offset, len, k);
6261 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6262 set_result(result);
6263 return true;
6264 }
6265
6266 //-------------inline_montgomeryMultiply-----------------------------------
6267 bool LibraryCallKit::inline_montgomeryMultiply() {
6268 address stubAddr = StubRoutines::montgomeryMultiply();
6269 if (stubAddr == nullptr) {
6270 return false; // Intrinsic's stub is not implemented on this platform
6271 }
6272
6273 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6274 const char* stubName = "montgomery_multiply";
6275
6276 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6277
6278 Node* a = argument(0);
6279 Node* b = argument(1);
6280 Node* n = argument(2);
6281 Node* len = argument(3);
6282 Node* inv = argument(4);
6283 Node* m = argument(6);
6284
6285 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6286 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6287 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6288 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6289 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6290 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6291 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6292 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6293 // failed array check
6294 return false;
6295 }
6296
6297 BasicType a_elem = a_type->elem()->array_element_basic_type();
6298 BasicType b_elem = b_type->elem()->array_element_basic_type();
6299 BasicType n_elem = n_type->elem()->array_element_basic_type();
6300 BasicType m_elem = m_type->elem()->array_element_basic_type();
6301 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6302 return false;
6303 }
6304
6305 // Make the call
6306 {
6307 Node* a_start = array_element_address(a, intcon(0), a_elem);
6308 Node* b_start = array_element_address(b, intcon(0), b_elem);
6309 Node* n_start = array_element_address(n, intcon(0), n_elem);
6310 Node* m_start = array_element_address(m, intcon(0), m_elem);
6311
6312 Node* call = make_runtime_call(RC_LEAF,
6313 OptoRuntime::montgomeryMultiply_Type(),
6314 stubAddr, stubName, TypePtr::BOTTOM,
6315 a_start, b_start, n_start, len, inv, top(),
6316 m_start);
6317 set_result(m);
6318 }
6319
6320 return true;
6321 }
6322
6323 bool LibraryCallKit::inline_montgomerySquare() {
6324 address stubAddr = StubRoutines::montgomerySquare();
6325 if (stubAddr == nullptr) {
6326 return false; // Intrinsic's stub is not implemented on this platform
6327 }
6328
6329 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6330 const char* stubName = "montgomery_square";
6331
6332 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6333
6334 Node* a = argument(0);
6335 Node* n = argument(1);
6336 Node* len = argument(2);
6337 Node* inv = argument(3);
6338 Node* m = argument(5);
6339
6340 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6341 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6342 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6343 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6344 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6345 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6346 // failed array check
6347 return false;
6348 }
6349
6350 BasicType a_elem = a_type->elem()->array_element_basic_type();
6351 BasicType n_elem = n_type->elem()->array_element_basic_type();
6352 BasicType m_elem = m_type->elem()->array_element_basic_type();
6353 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6354 return false;
6355 }
6356
6357 // Make the call
6358 {
6359 Node* a_start = array_element_address(a, intcon(0), a_elem);
6360 Node* n_start = array_element_address(n, intcon(0), n_elem);
6361 Node* m_start = array_element_address(m, intcon(0), m_elem);
6362
6363 Node* call = make_runtime_call(RC_LEAF,
6364 OptoRuntime::montgomerySquare_Type(),
6365 stubAddr, stubName, TypePtr::BOTTOM,
6366 a_start, n_start, len, inv, top(),
6367 m_start);
6368 set_result(m);
6369 }
6370
6371 return true;
6372 }
6373
6374 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6375 address stubAddr = nullptr;
6376 const char* stubName = nullptr;
6377
6378 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6379 if (stubAddr == nullptr) {
6380 return false; // Intrinsic's stub is not implemented on this platform
6381 }
6382
6383 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6384
6385 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6386
6387 Node* newArr = argument(0);
6388 Node* oldArr = argument(1);
6389 Node* newIdx = argument(2);
6390 Node* shiftCount = argument(3);
6391 Node* numIter = argument(4);
6392
6393 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6394 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6395 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6396 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6397 return false;
6398 }
6399
6400 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6401 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6402 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6403 return false;
6404 }
6405
6406 // Make the call
6407 {
6408 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6409 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6410
6411 Node* call = make_runtime_call(RC_LEAF,
6412 OptoRuntime::bigIntegerShift_Type(),
6413 stubAddr,
6414 stubName,
6415 TypePtr::BOTTOM,
6416 newArr_start,
6417 oldArr_start,
6418 newIdx,
6419 shiftCount,
6420 numIter);
6421 }
6422
6423 return true;
6424 }
6425
6426 //-------------inline_vectorizedMismatch------------------------------
6427 bool LibraryCallKit::inline_vectorizedMismatch() {
6428 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6429
6430 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6431 Node* obja = argument(0); // Object
6432 Node* aoffset = argument(1); // long
6433 Node* objb = argument(3); // Object
6434 Node* boffset = argument(4); // long
6435 Node* length = argument(6); // int
6436 Node* scale = argument(7); // int
6437
6438 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6439 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6440 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6441 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6442 scale == top()) {
6443 return false; // failed input validation
6444 }
6445
6446 Node* obja_adr = make_unsafe_address(obja, aoffset);
6447 Node* objb_adr = make_unsafe_address(objb, boffset);
6448
6449 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6450 //
6451 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6452 // if (length <= inline_limit) {
6453 // inline_path:
6454 // vmask = VectorMaskGen length
6455 // vload1 = LoadVectorMasked obja, vmask
6456 // vload2 = LoadVectorMasked objb, vmask
6457 // result1 = VectorCmpMasked vload1, vload2, vmask
6458 // } else {
6459 // call_stub_path:
6460 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6461 // }
6462 // exit_block:
6463 // return Phi(result1, result2);
6464 //
6465 enum { inline_path = 1, // input is small enough to process it all at once
6466 stub_path = 2, // input is too large; call into the VM
6467 PATH_LIMIT = 3
6468 };
6469
6470 Node* exit_block = new RegionNode(PATH_LIMIT);
6471 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6472 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6473
6474 Node* call_stub_path = control();
6475
6476 BasicType elem_bt = T_ILLEGAL;
6477
6478 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6479 if (scale_t->is_con()) {
6480 switch (scale_t->get_con()) {
6481 case 0: elem_bt = T_BYTE; break;
6482 case 1: elem_bt = T_SHORT; break;
6483 case 2: elem_bt = T_INT; break;
6484 case 3: elem_bt = T_LONG; break;
6485
6486 default: elem_bt = T_ILLEGAL; break; // not supported
6487 }
6488 }
6489
6490 int inline_limit = 0;
6491 bool do_partial_inline = false;
6492
6493 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6494 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6495 do_partial_inline = inline_limit >= 16;
6496 }
6497
6498 if (do_partial_inline) {
6499 assert(elem_bt != T_ILLEGAL, "sanity");
6500
6501 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6502 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6503 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6504
6505 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6506 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6507 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6508
6509 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6510
6511 if (!stopped()) {
6512 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6513
6514 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6515 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6516 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6517 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6518
6519 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6520 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6521 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6522 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6523
6524 exit_block->init_req(inline_path, control());
6525 memory_phi->init_req(inline_path, map()->memory());
6526 result_phi->init_req(inline_path, result);
6527
6528 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6529 clear_upper_avx();
6530 }
6531 }
6532 }
6533
6534 if (call_stub_path != nullptr) {
6535 set_control(call_stub_path);
6536
6537 Node* call = make_runtime_call(RC_LEAF,
6538 OptoRuntime::vectorizedMismatch_Type(),
6539 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6540 obja_adr, objb_adr, length, scale);
6541
6542 exit_block->init_req(stub_path, control());
6543 memory_phi->init_req(stub_path, map()->memory());
6544 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6545 }
6546
6547 exit_block = _gvn.transform(exit_block);
6548 memory_phi = _gvn.transform(memory_phi);
6549 result_phi = _gvn.transform(result_phi);
6550
6551 set_control(exit_block);
6552 set_all_memory(memory_phi);
6553 set_result(result_phi);
6554
6555 return true;
6556 }
6557
6558 //------------------------------inline_vectorizedHashcode----------------------------
6559 bool LibraryCallKit::inline_vectorizedHashCode() {
6560 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6561
6562 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6563 Node* array = argument(0);
6564 Node* offset = argument(1);
6565 Node* length = argument(2);
6566 Node* initialValue = argument(3);
6567 Node* basic_type = argument(4);
6568
6569 if (basic_type == top()) {
6570 return false; // failed input validation
6571 }
6572
6573 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6574 if (!basic_type_t->is_con()) {
6575 return false; // Only intrinsify if mode argument is constant
6576 }
6577
6578 array = must_be_not_null(array, true);
6579
6580 BasicType bt = (BasicType)basic_type_t->get_con();
6581
6582 // Resolve address of first element
6583 Node* array_start = array_element_address(array, offset, bt);
6584
6585 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6586 array_start, length, initialValue, basic_type)));
6587 clear_upper_avx();
6588
6589 return true;
6590 }
6591
6592 /**
6593 * Calculate CRC32 for byte.
6594 * int java.util.zip.CRC32.update(int crc, int b)
6595 */
6596 bool LibraryCallKit::inline_updateCRC32() {
6597 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6598 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6599 // no receiver since it is static method
6600 Node* crc = argument(0); // type: int
6601 Node* b = argument(1); // type: int
6602
6603 /*
6604 * int c = ~ crc;
6605 * b = timesXtoThe32[(b ^ c) & 0xFF];
6606 * b = b ^ (c >>> 8);
6607 * crc = ~b;
6608 */
6609
6610 Node* M1 = intcon(-1);
6611 crc = _gvn.transform(new XorINode(crc, M1));
6612 Node* result = _gvn.transform(new XorINode(crc, b));
6613 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6614
6615 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6616 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6617 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6618 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6619
6620 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6621 result = _gvn.transform(new XorINode(crc, result));
6622 result = _gvn.transform(new XorINode(result, M1));
6623 set_result(result);
6624 return true;
6625 }
6626
6627 /**
6628 * Calculate CRC32 for byte[] array.
6629 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6630 */
6631 bool LibraryCallKit::inline_updateBytesCRC32() {
6632 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6633 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6634 // no receiver since it is static method
6635 Node* crc = argument(0); // type: int
6636 Node* src = argument(1); // type: oop
6637 Node* offset = argument(2); // type: int
6638 Node* length = argument(3); // type: int
6639
6640 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6641 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6642 // failed array check
6643 return false;
6644 }
6645
6646 // Figure out the size and type of the elements we will be copying.
6647 BasicType src_elem = src_type->elem()->array_element_basic_type();
6648 if (src_elem != T_BYTE) {
6649 return false;
6650 }
6651
6652 // 'src_start' points to src array + scaled offset
6653 src = must_be_not_null(src, true);
6654 Node* src_start = array_element_address(src, offset, src_elem);
6655
6656 // We assume that range check is done by caller.
6657 // TODO: generate range check (offset+length < src.length) in debug VM.
6658
6659 // Call the stub.
6660 address stubAddr = StubRoutines::updateBytesCRC32();
6661 const char *stubName = "updateBytesCRC32";
6662
6663 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6664 stubAddr, stubName, TypePtr::BOTTOM,
6665 crc, src_start, length);
6666 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6667 set_result(result);
6668 return true;
6669 }
6670
6671 /**
6672 * Calculate CRC32 for ByteBuffer.
6673 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6674 */
6675 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6676 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6677 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6678 // no receiver since it is static method
6679 Node* crc = argument(0); // type: int
6680 Node* src = argument(1); // type: long
6681 Node* offset = argument(3); // type: int
6682 Node* length = argument(4); // type: int
6683
6684 src = ConvL2X(src); // adjust Java long to machine word
6685 Node* base = _gvn.transform(new CastX2PNode(src));
6686 offset = ConvI2X(offset);
6687
6688 // 'src_start' points to src array + scaled offset
6689 Node* src_start = basic_plus_adr(top(), base, offset);
6690
6691 // Call the stub.
6692 address stubAddr = StubRoutines::updateBytesCRC32();
6693 const char *stubName = "updateBytesCRC32";
6694
6695 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6696 stubAddr, stubName, TypePtr::BOTTOM,
6697 crc, src_start, length);
6698 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6699 set_result(result);
6700 return true;
6701 }
6702
6703 //------------------------------get_table_from_crc32c_class-----------------------
6704 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6705 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6706 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6707
6708 return table;
6709 }
6710
6711 //------------------------------inline_updateBytesCRC32C-----------------------
6712 //
6713 // Calculate CRC32C for byte[] array.
6714 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6715 //
6716 bool LibraryCallKit::inline_updateBytesCRC32C() {
6717 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6718 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6719 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6720 // no receiver since it is a static method
6721 Node* crc = argument(0); // type: int
6722 Node* src = argument(1); // type: oop
6723 Node* offset = argument(2); // type: int
6724 Node* end = argument(3); // type: int
6725
6726 Node* length = _gvn.transform(new SubINode(end, offset));
6727
6728 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6729 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6730 // failed array check
6731 return false;
6732 }
6733
6734 // Figure out the size and type of the elements we will be copying.
6735 BasicType src_elem = src_type->elem()->array_element_basic_type();
6736 if (src_elem != T_BYTE) {
6737 return false;
6738 }
6739
6740 // 'src_start' points to src array + scaled offset
6741 src = must_be_not_null(src, true);
6742 Node* src_start = array_element_address(src, offset, src_elem);
6743
6744 // static final int[] byteTable in class CRC32C
6745 Node* table = get_table_from_crc32c_class(callee()->holder());
6746 table = must_be_not_null(table, true);
6747 Node* table_start = array_element_address(table, intcon(0), T_INT);
6748
6749 // We assume that range check is done by caller.
6750 // TODO: generate range check (offset+length < src.length) in debug VM.
6751
6752 // Call the stub.
6753 address stubAddr = StubRoutines::updateBytesCRC32C();
6754 const char *stubName = "updateBytesCRC32C";
6755
6756 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6757 stubAddr, stubName, TypePtr::BOTTOM,
6758 crc, src_start, length, table_start);
6759 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6760 set_result(result);
6761 return true;
6762 }
6763
6764 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6765 //
6766 // Calculate CRC32C for DirectByteBuffer.
6767 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6768 //
6769 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6770 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6771 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6772 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6773 // no receiver since it is a static method
6774 Node* crc = argument(0); // type: int
6775 Node* src = argument(1); // type: long
6776 Node* offset = argument(3); // type: int
6777 Node* end = argument(4); // type: int
6778
6779 Node* length = _gvn.transform(new SubINode(end, offset));
6780
6781 src = ConvL2X(src); // adjust Java long to machine word
6782 Node* base = _gvn.transform(new CastX2PNode(src));
6783 offset = ConvI2X(offset);
6784
6785 // 'src_start' points to src array + scaled offset
6786 Node* src_start = basic_plus_adr(top(), base, offset);
6787
6788 // static final int[] byteTable in class CRC32C
6789 Node* table = get_table_from_crc32c_class(callee()->holder());
6790 table = must_be_not_null(table, true);
6791 Node* table_start = array_element_address(table, intcon(0), T_INT);
6792
6793 // Call the stub.
6794 address stubAddr = StubRoutines::updateBytesCRC32C();
6795 const char *stubName = "updateBytesCRC32C";
6796
6797 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6798 stubAddr, stubName, TypePtr::BOTTOM,
6799 crc, src_start, length, table_start);
6800 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6801 set_result(result);
6802 return true;
6803 }
6804
6805 //------------------------------inline_updateBytesAdler32----------------------
6806 //
6807 // Calculate Adler32 checksum for byte[] array.
6808 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6809 //
6810 bool LibraryCallKit::inline_updateBytesAdler32() {
6811 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6812 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6813 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6814 // no receiver since it is static method
6815 Node* crc = argument(0); // type: int
6816 Node* src = argument(1); // type: oop
6817 Node* offset = argument(2); // type: int
6818 Node* length = argument(3); // type: int
6819
6820 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6821 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6822 // failed array check
6823 return false;
6824 }
6825
6826 // Figure out the size and type of the elements we will be copying.
6827 BasicType src_elem = src_type->elem()->array_element_basic_type();
6828 if (src_elem != T_BYTE) {
6829 return false;
6830 }
6831
6832 // 'src_start' points to src array + scaled offset
6833 Node* src_start = array_element_address(src, offset, src_elem);
6834
6835 // We assume that range check is done by caller.
6836 // TODO: generate range check (offset+length < src.length) in debug VM.
6837
6838 // Call the stub.
6839 address stubAddr = StubRoutines::updateBytesAdler32();
6840 const char *stubName = "updateBytesAdler32";
6841
6842 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6843 stubAddr, stubName, TypePtr::BOTTOM,
6844 crc, src_start, length);
6845 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6846 set_result(result);
6847 return true;
6848 }
6849
6850 //------------------------------inline_updateByteBufferAdler32---------------
6851 //
6852 // Calculate Adler32 checksum for DirectByteBuffer.
6853 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6854 //
6855 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6856 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6857 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6858 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6859 // no receiver since it is static method
6860 Node* crc = argument(0); // type: int
6861 Node* src = argument(1); // type: long
6862 Node* offset = argument(3); // type: int
6863 Node* length = argument(4); // type: int
6864
6865 src = ConvL2X(src); // adjust Java long to machine word
6866 Node* base = _gvn.transform(new CastX2PNode(src));
6867 offset = ConvI2X(offset);
6868
6869 // 'src_start' points to src array + scaled offset
6870 Node* src_start = basic_plus_adr(top(), base, offset);
6871
6872 // Call the stub.
6873 address stubAddr = StubRoutines::updateBytesAdler32();
6874 const char *stubName = "updateBytesAdler32";
6875
6876 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6877 stubAddr, stubName, TypePtr::BOTTOM,
6878 crc, src_start, length);
6879
6880 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6881 set_result(result);
6882 return true;
6883 }
6884
6885 //----------------------------inline_reference_get----------------------------
6886 // public T java.lang.ref.Reference.get();
6887 bool LibraryCallKit::inline_reference_get() {
6888 const int referent_offset = java_lang_ref_Reference::referent_offset();
6889
6890 // Get the argument:
6891 Node* reference_obj = null_check_receiver();
6892 if (stopped()) return true;
6893
6894 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6895 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6896 decorators, /*is_static*/ false, nullptr);
6897 if (result == nullptr) return false;
6898
6899 // Add memory barrier to prevent commoning reads from this field
6900 // across safepoint since GC can change its value.
6901 insert_mem_bar(Op_MemBarCPUOrder);
6902
6903 set_result(result);
6904 return true;
6905 }
6906
6907 //----------------------------inline_reference_refersTo0----------------------------
6908 // bool java.lang.ref.Reference.refersTo0();
6909 // bool java.lang.ref.PhantomReference.refersTo0();
6910 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6911 // Get arguments:
6912 Node* reference_obj = null_check_receiver();
6913 Node* other_obj = argument(1);
6914 if (stopped()) return true;
6915
6916 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6917 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6918 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6919 decorators, /*is_static*/ false, nullptr);
6920 if (referent == nullptr) return false;
6921
6922 // Add memory barrier to prevent commoning reads from this field
6923 // across safepoint since GC can change its value.
6924 insert_mem_bar(Op_MemBarCPUOrder);
6925
6926 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
6927 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6928 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
6929
6930 RegionNode* region = new RegionNode(3);
6931 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
6932
6933 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
6934 region->init_req(1, if_true);
6935 phi->init_req(1, intcon(1));
6936
6937 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
6938 region->init_req(2, if_false);
6939 phi->init_req(2, intcon(0));
6940
6941 set_control(_gvn.transform(region));
6942 record_for_igvn(region);
6943 set_result(_gvn.transform(phi));
6944 return true;
6945 }
6946
6947 //----------------------------inline_reference_clear0----------------------------
6948 // void java.lang.ref.Reference.clear0();
6949 // void java.lang.ref.PhantomReference.clear0();
6950 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
6951 // This matches the implementation in JVM_ReferenceClear, see the comments there.
6952
6953 // Get arguments
6954 Node* reference_obj = null_check_receiver();
6955 if (stopped()) return true;
6956
6957 // Common access parameters
6958 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6959 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6960 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
6961 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
6962 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
6963
6964 Node* referent = access_load_at(reference_obj,
6965 referent_field_addr,
6966 referent_field_addr_type,
6967 val_type,
6968 T_OBJECT,
6969 decorators);
6970
6971 IdealKit ideal(this);
6972 #define __ ideal.
6973 __ if_then(referent, BoolTest::ne, null());
6974 sync_kit(ideal);
6975 access_store_at(reference_obj,
6976 referent_field_addr,
6977 referent_field_addr_type,
6978 null(),
6979 val_type,
6980 T_OBJECT,
6981 decorators);
6982 __ sync_kit(this);
6983 __ end_if();
6984 final_sync(ideal);
6985 #undef __
6986
6987 return true;
6988 }
6989
6990 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
6991 DecoratorSet decorators, bool is_static,
6992 ciInstanceKlass* fromKls) {
6993 if (fromKls == nullptr) {
6994 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6995 assert(tinst != nullptr, "obj is null");
6996 assert(tinst->is_loaded(), "obj is not loaded");
6997 fromKls = tinst->instance_klass();
6998 } else {
6999 assert(is_static, "only for static field access");
7000 }
7001 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7002 ciSymbol::make(fieldTypeString),
7003 is_static);
7004
7005 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7006 if (field == nullptr) return (Node *) nullptr;
7007
7008 if (is_static) {
7009 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7010 fromObj = makecon(tip);
7011 }
7012
7013 // Next code copied from Parse::do_get_xxx():
7014
7015 // Compute address and memory type.
7016 int offset = field->offset_in_bytes();
7017 bool is_vol = field->is_volatile();
7018 ciType* field_klass = field->type();
7019 assert(field_klass->is_loaded(), "should be loaded");
7020 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7021 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7022 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7023 "slice of address and input slice don't match");
7024 BasicType bt = field->layout_type();
7025
7026 // Build the resultant type of the load
7027 const Type *type;
7028 if (bt == T_OBJECT) {
7029 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7030 } else {
7031 type = Type::get_const_basic_type(bt);
7032 }
7033
7034 if (is_vol) {
7035 decorators |= MO_SEQ_CST;
7036 }
7037
7038 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7039 }
7040
7041 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7042 bool is_exact /* true */, bool is_static /* false */,
7043 ciInstanceKlass * fromKls /* nullptr */) {
7044 if (fromKls == nullptr) {
7045 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7046 assert(tinst != nullptr, "obj is null");
7047 assert(tinst->is_loaded(), "obj is not loaded");
7048 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7049 fromKls = tinst->instance_klass();
7050 }
7051 else {
7052 assert(is_static, "only for static field access");
7053 }
7054 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7055 ciSymbol::make(fieldTypeString),
7056 is_static);
7057
7058 assert(field != nullptr, "undefined field");
7059 assert(!field->is_volatile(), "not defined for volatile fields");
7060
7061 if (is_static) {
7062 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7063 fromObj = makecon(tip);
7064 }
7065
7066 // Next code copied from Parse::do_get_xxx():
7067
7068 // Compute address and memory type.
7069 int offset = field->offset_in_bytes();
7070 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7071
7072 return adr;
7073 }
7074
7075 //------------------------------inline_aescrypt_Block-----------------------
7076 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7077 address stubAddr = nullptr;
7078 const char *stubName;
7079 assert(UseAES, "need AES instruction support");
7080
7081 switch(id) {
7082 case vmIntrinsics::_aescrypt_encryptBlock:
7083 stubAddr = StubRoutines::aescrypt_encryptBlock();
7084 stubName = "aescrypt_encryptBlock";
7085 break;
7086 case vmIntrinsics::_aescrypt_decryptBlock:
7087 stubAddr = StubRoutines::aescrypt_decryptBlock();
7088 stubName = "aescrypt_decryptBlock";
7089 break;
7090 default:
7091 break;
7092 }
7093 if (stubAddr == nullptr) return false;
7094
7095 Node* aescrypt_object = argument(0);
7096 Node* src = argument(1);
7097 Node* src_offset = argument(2);
7098 Node* dest = argument(3);
7099 Node* dest_offset = argument(4);
7100
7101 src = must_be_not_null(src, true);
7102 dest = must_be_not_null(dest, true);
7103
7104 // (1) src and dest are arrays.
7105 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7106 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7107 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7108 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7109
7110 // for the quick and dirty code we will skip all the checks.
7111 // we are just trying to get the call to be generated.
7112 Node* src_start = src;
7113 Node* dest_start = dest;
7114 if (src_offset != nullptr || dest_offset != nullptr) {
7115 assert(src_offset != nullptr && dest_offset != nullptr, "");
7116 src_start = array_element_address(src, src_offset, T_BYTE);
7117 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7118 }
7119
7120 // now need to get the start of its expanded key array
7121 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7122 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7123 if (k_start == nullptr) return false;
7124
7125 // Call the stub.
7126 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7127 stubAddr, stubName, TypePtr::BOTTOM,
7128 src_start, dest_start, k_start);
7129
7130 return true;
7131 }
7132
7133 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7134 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7135 address stubAddr = nullptr;
7136 const char *stubName = nullptr;
7137
7138 assert(UseAES, "need AES instruction support");
7139
7140 switch(id) {
7141 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7142 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7143 stubName = "cipherBlockChaining_encryptAESCrypt";
7144 break;
7145 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7146 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7147 stubName = "cipherBlockChaining_decryptAESCrypt";
7148 break;
7149 default:
7150 break;
7151 }
7152 if (stubAddr == nullptr) return false;
7153
7154 Node* cipherBlockChaining_object = argument(0);
7155 Node* src = argument(1);
7156 Node* src_offset = argument(2);
7157 Node* len = argument(3);
7158 Node* dest = argument(4);
7159 Node* dest_offset = argument(5);
7160
7161 src = must_be_not_null(src, false);
7162 dest = must_be_not_null(dest, false);
7163
7164 // (1) src and dest are arrays.
7165 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7166 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7167 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7168 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7169
7170 // checks are the responsibility of the caller
7171 Node* src_start = src;
7172 Node* dest_start = dest;
7173 if (src_offset != nullptr || dest_offset != nullptr) {
7174 assert(src_offset != nullptr && dest_offset != nullptr, "");
7175 src_start = array_element_address(src, src_offset, T_BYTE);
7176 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7177 }
7178
7179 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7180 // (because of the predicated logic executed earlier).
7181 // so we cast it here safely.
7182 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7183
7184 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7185 if (embeddedCipherObj == nullptr) return false;
7186
7187 // cast it to what we know it will be at runtime
7188 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7189 assert(tinst != nullptr, "CBC obj is null");
7190 assert(tinst->is_loaded(), "CBC obj is not loaded");
7191 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7192 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7193
7194 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7195 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7196 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7197 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7198 aescrypt_object = _gvn.transform(aescrypt_object);
7199
7200 // we need to get the start of the aescrypt_object's expanded key array
7201 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7202 if (k_start == nullptr) return false;
7203
7204 // similarly, get the start address of the r vector
7205 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7206 if (objRvec == nullptr) return false;
7207 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7208
7209 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7210 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7211 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7212 stubAddr, stubName, TypePtr::BOTTOM,
7213 src_start, dest_start, k_start, r_start, len);
7214
7215 // return cipher length (int)
7216 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7217 set_result(retvalue);
7218 return true;
7219 }
7220
7221 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7222 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7223 address stubAddr = nullptr;
7224 const char *stubName = nullptr;
7225
7226 assert(UseAES, "need AES instruction support");
7227
7228 switch (id) {
7229 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7230 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7231 stubName = "electronicCodeBook_encryptAESCrypt";
7232 break;
7233 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7234 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7235 stubName = "electronicCodeBook_decryptAESCrypt";
7236 break;
7237 default:
7238 break;
7239 }
7240
7241 if (stubAddr == nullptr) return false;
7242
7243 Node* electronicCodeBook_object = argument(0);
7244 Node* src = argument(1);
7245 Node* src_offset = argument(2);
7246 Node* len = argument(3);
7247 Node* dest = argument(4);
7248 Node* dest_offset = argument(5);
7249
7250 // (1) src and dest are arrays.
7251 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7252 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7253 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7254 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7255
7256 // checks are the responsibility of the caller
7257 Node* src_start = src;
7258 Node* dest_start = dest;
7259 if (src_offset != nullptr || dest_offset != nullptr) {
7260 assert(src_offset != nullptr && dest_offset != nullptr, "");
7261 src_start = array_element_address(src, src_offset, T_BYTE);
7262 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7263 }
7264
7265 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7266 // (because of the predicated logic executed earlier).
7267 // so we cast it here safely.
7268 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7269
7270 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7271 if (embeddedCipherObj == nullptr) return false;
7272
7273 // cast it to what we know it will be at runtime
7274 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7275 assert(tinst != nullptr, "ECB obj is null");
7276 assert(tinst->is_loaded(), "ECB obj is not loaded");
7277 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7278 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7279
7280 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7281 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7282 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7283 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7284 aescrypt_object = _gvn.transform(aescrypt_object);
7285
7286 // we need to get the start of the aescrypt_object's expanded key array
7287 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7288 if (k_start == nullptr) return false;
7289
7290 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7291 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7292 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7293 stubAddr, stubName, TypePtr::BOTTOM,
7294 src_start, dest_start, k_start, len);
7295
7296 // return cipher length (int)
7297 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7298 set_result(retvalue);
7299 return true;
7300 }
7301
7302 //------------------------------inline_counterMode_AESCrypt-----------------------
7303 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7304 assert(UseAES, "need AES instruction support");
7305 if (!UseAESCTRIntrinsics) return false;
7306
7307 address stubAddr = nullptr;
7308 const char *stubName = nullptr;
7309 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7310 stubAddr = StubRoutines::counterMode_AESCrypt();
7311 stubName = "counterMode_AESCrypt";
7312 }
7313 if (stubAddr == nullptr) return false;
7314
7315 Node* counterMode_object = argument(0);
7316 Node* src = argument(1);
7317 Node* src_offset = argument(2);
7318 Node* len = argument(3);
7319 Node* dest = argument(4);
7320 Node* dest_offset = argument(5);
7321
7322 // (1) src and dest are arrays.
7323 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7324 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7325 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7326 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7327
7328 // checks are the responsibility of the caller
7329 Node* src_start = src;
7330 Node* dest_start = dest;
7331 if (src_offset != nullptr || dest_offset != nullptr) {
7332 assert(src_offset != nullptr && dest_offset != nullptr, "");
7333 src_start = array_element_address(src, src_offset, T_BYTE);
7334 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7335 }
7336
7337 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7338 // (because of the predicated logic executed earlier).
7339 // so we cast it here safely.
7340 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7341 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7342 if (embeddedCipherObj == nullptr) return false;
7343 // cast it to what we know it will be at runtime
7344 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7345 assert(tinst != nullptr, "CTR obj is null");
7346 assert(tinst->is_loaded(), "CTR obj is not loaded");
7347 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7348 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7349 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7350 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7351 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7352 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7353 aescrypt_object = _gvn.transform(aescrypt_object);
7354 // we need to get the start of the aescrypt_object's expanded key array
7355 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7356 if (k_start == nullptr) return false;
7357 // similarly, get the start address of the r vector
7358 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7359 if (obj_counter == nullptr) return false;
7360 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7361
7362 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7363 if (saved_encCounter == nullptr) return false;
7364 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7365 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7366
7367 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7368 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7369 OptoRuntime::counterMode_aescrypt_Type(),
7370 stubAddr, stubName, TypePtr::BOTTOM,
7371 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7372
7373 // return cipher length (int)
7374 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7375 set_result(retvalue);
7376 return true;
7377 }
7378
7379 //------------------------------get_key_start_from_aescrypt_object-----------------------
7380 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7381 #if defined(PPC64) || defined(S390) || defined(RISCV64)
7382 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7383 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7384 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7385 // The ppc64 and riscv64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7386 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7387 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7388 if (objSessionK == nullptr) {
7389 return (Node *) nullptr;
7390 }
7391 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7392 #else
7393 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7394 #endif // PPC64
7395 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7396 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7397
7398 // now have the array, need to get the start address of the K array
7399 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7400 return k_start;
7401 }
7402
7403 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7404 // Return node representing slow path of predicate check.
7405 // the pseudo code we want to emulate with this predicate is:
7406 // for encryption:
7407 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7408 // for decryption:
7409 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7410 // note cipher==plain is more conservative than the original java code but that's OK
7411 //
7412 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7413 // The receiver was checked for null already.
7414 Node* objCBC = argument(0);
7415
7416 Node* src = argument(1);
7417 Node* dest = argument(4);
7418
7419 // Load embeddedCipher field of CipherBlockChaining object.
7420 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7421
7422 // get AESCrypt klass for instanceOf check
7423 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7424 // will have same classloader as CipherBlockChaining object
7425 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7426 assert(tinst != nullptr, "CBCobj is null");
7427 assert(tinst->is_loaded(), "CBCobj is not loaded");
7428
7429 // we want to do an instanceof comparison against the AESCrypt class
7430 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7431 if (!klass_AESCrypt->is_loaded()) {
7432 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7433 Node* ctrl = control();
7434 set_control(top()); // no regular fast path
7435 return ctrl;
7436 }
7437
7438 src = must_be_not_null(src, true);
7439 dest = must_be_not_null(dest, true);
7440
7441 // Resolve oops to stable for CmpP below.
7442 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7443
7444 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7445 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7446 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7447
7448 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7449
7450 // for encryption, we are done
7451 if (!decrypting)
7452 return instof_false; // even if it is null
7453
7454 // for decryption, we need to add a further check to avoid
7455 // taking the intrinsic path when cipher and plain are the same
7456 // see the original java code for why.
7457 RegionNode* region = new RegionNode(3);
7458 region->init_req(1, instof_false);
7459
7460 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7461 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7462 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7463 region->init_req(2, src_dest_conjoint);
7464
7465 record_for_igvn(region);
7466 return _gvn.transform(region);
7467 }
7468
7469 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7470 // Return node representing slow path of predicate check.
7471 // the pseudo code we want to emulate with this predicate is:
7472 // for encryption:
7473 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7474 // for decryption:
7475 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7476 // note cipher==plain is more conservative than the original java code but that's OK
7477 //
7478 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7479 // The receiver was checked for null already.
7480 Node* objECB = argument(0);
7481
7482 // Load embeddedCipher field of ElectronicCodeBook object.
7483 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7484
7485 // get AESCrypt klass for instanceOf check
7486 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7487 // will have same classloader as ElectronicCodeBook object
7488 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7489 assert(tinst != nullptr, "ECBobj is null");
7490 assert(tinst->is_loaded(), "ECBobj is not loaded");
7491
7492 // we want to do an instanceof comparison against the AESCrypt class
7493 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7494 if (!klass_AESCrypt->is_loaded()) {
7495 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7496 Node* ctrl = control();
7497 set_control(top()); // no regular fast path
7498 return ctrl;
7499 }
7500 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7501
7502 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7503 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7504 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7505
7506 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7507
7508 // for encryption, we are done
7509 if (!decrypting)
7510 return instof_false; // even if it is null
7511
7512 // for decryption, we need to add a further check to avoid
7513 // taking the intrinsic path when cipher and plain are the same
7514 // see the original java code for why.
7515 RegionNode* region = new RegionNode(3);
7516 region->init_req(1, instof_false);
7517 Node* src = argument(1);
7518 Node* dest = argument(4);
7519 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7520 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7521 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7522 region->init_req(2, src_dest_conjoint);
7523
7524 record_for_igvn(region);
7525 return _gvn.transform(region);
7526 }
7527
7528 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7529 // Return node representing slow path of predicate check.
7530 // the pseudo code we want to emulate with this predicate is:
7531 // for encryption:
7532 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7533 // for decryption:
7534 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7535 // note cipher==plain is more conservative than the original java code but that's OK
7536 //
7537
7538 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7539 // The receiver was checked for null already.
7540 Node* objCTR = argument(0);
7541
7542 // Load embeddedCipher field of CipherBlockChaining object.
7543 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7544
7545 // get AESCrypt klass for instanceOf check
7546 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7547 // will have same classloader as CipherBlockChaining object
7548 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7549 assert(tinst != nullptr, "CTRobj is null");
7550 assert(tinst->is_loaded(), "CTRobj is not loaded");
7551
7552 // we want to do an instanceof comparison against the AESCrypt class
7553 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7554 if (!klass_AESCrypt->is_loaded()) {
7555 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7556 Node* ctrl = control();
7557 set_control(top()); // no regular fast path
7558 return ctrl;
7559 }
7560
7561 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7562 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7563 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7564 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7565 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7566
7567 return instof_false; // even if it is null
7568 }
7569
7570 //------------------------------inline_ghash_processBlocks
7571 bool LibraryCallKit::inline_ghash_processBlocks() {
7572 address stubAddr;
7573 const char *stubName;
7574 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7575
7576 stubAddr = StubRoutines::ghash_processBlocks();
7577 stubName = "ghash_processBlocks";
7578
7579 Node* data = argument(0);
7580 Node* offset = argument(1);
7581 Node* len = argument(2);
7582 Node* state = argument(3);
7583 Node* subkeyH = argument(4);
7584
7585 state = must_be_not_null(state, true);
7586 subkeyH = must_be_not_null(subkeyH, true);
7587 data = must_be_not_null(data, true);
7588
7589 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7590 assert(state_start, "state is null");
7591 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7592 assert(subkeyH_start, "subkeyH is null");
7593 Node* data_start = array_element_address(data, offset, T_BYTE);
7594 assert(data_start, "data is null");
7595
7596 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7597 OptoRuntime::ghash_processBlocks_Type(),
7598 stubAddr, stubName, TypePtr::BOTTOM,
7599 state_start, subkeyH_start, data_start, len);
7600 return true;
7601 }
7602
7603 //------------------------------inline_chacha20Block
7604 bool LibraryCallKit::inline_chacha20Block() {
7605 address stubAddr;
7606 const char *stubName;
7607 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7608
7609 stubAddr = StubRoutines::chacha20Block();
7610 stubName = "chacha20Block";
7611
7612 Node* state = argument(0);
7613 Node* result = argument(1);
7614
7615 state = must_be_not_null(state, true);
7616 result = must_be_not_null(result, true);
7617
7618 Node* state_start = array_element_address(state, intcon(0), T_INT);
7619 assert(state_start, "state is null");
7620 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7621 assert(result_start, "result is null");
7622
7623 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7624 OptoRuntime::chacha20Block_Type(),
7625 stubAddr, stubName, TypePtr::BOTTOM,
7626 state_start, result_start);
7627 // return key stream length (int)
7628 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7629 set_result(retvalue);
7630 return true;
7631 }
7632
7633 bool LibraryCallKit::inline_base64_encodeBlock() {
7634 address stubAddr;
7635 const char *stubName;
7636 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7637 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
7638 stubAddr = StubRoutines::base64_encodeBlock();
7639 stubName = "encodeBlock";
7640
7641 if (!stubAddr) return false;
7642 Node* base64obj = argument(0);
7643 Node* src = argument(1);
7644 Node* offset = argument(2);
7645 Node* len = argument(3);
7646 Node* dest = argument(4);
7647 Node* dp = argument(5);
7648 Node* isURL = argument(6);
7649
7650 src = must_be_not_null(src, true);
7651 dest = must_be_not_null(dest, true);
7652
7653 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7654 assert(src_start, "source array is null");
7655 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7656 assert(dest_start, "destination array is null");
7657
7658 Node* base64 = make_runtime_call(RC_LEAF,
7659 OptoRuntime::base64_encodeBlock_Type(),
7660 stubAddr, stubName, TypePtr::BOTTOM,
7661 src_start, offset, len, dest_start, dp, isURL);
7662 return true;
7663 }
7664
7665 bool LibraryCallKit::inline_base64_decodeBlock() {
7666 address stubAddr;
7667 const char *stubName;
7668 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7669 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
7670 stubAddr = StubRoutines::base64_decodeBlock();
7671 stubName = "decodeBlock";
7672
7673 if (!stubAddr) return false;
7674 Node* base64obj = argument(0);
7675 Node* src = argument(1);
7676 Node* src_offset = argument(2);
7677 Node* len = argument(3);
7678 Node* dest = argument(4);
7679 Node* dest_offset = argument(5);
7680 Node* isURL = argument(6);
7681 Node* isMIME = argument(7);
7682
7683 src = must_be_not_null(src, true);
7684 dest = must_be_not_null(dest, true);
7685
7686 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7687 assert(src_start, "source array is null");
7688 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7689 assert(dest_start, "destination array is null");
7690
7691 Node* call = make_runtime_call(RC_LEAF,
7692 OptoRuntime::base64_decodeBlock_Type(),
7693 stubAddr, stubName, TypePtr::BOTTOM,
7694 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
7695 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7696 set_result(result);
7697 return true;
7698 }
7699
7700 bool LibraryCallKit::inline_poly1305_processBlocks() {
7701 address stubAddr;
7702 const char *stubName;
7703 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
7704 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
7705 stubAddr = StubRoutines::poly1305_processBlocks();
7706 stubName = "poly1305_processBlocks";
7707
7708 if (!stubAddr) return false;
7709 null_check_receiver(); // null-check receiver
7710 if (stopped()) return true;
7711
7712 Node* input = argument(1);
7713 Node* input_offset = argument(2);
7714 Node* len = argument(3);
7715 Node* alimbs = argument(4);
7716 Node* rlimbs = argument(5);
7717
7718 input = must_be_not_null(input, true);
7719 alimbs = must_be_not_null(alimbs, true);
7720 rlimbs = must_be_not_null(rlimbs, true);
7721
7722 Node* input_start = array_element_address(input, input_offset, T_BYTE);
7723 assert(input_start, "input array is null");
7724 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
7725 assert(acc_start, "acc array is null");
7726 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
7727 assert(r_start, "r array is null");
7728
7729 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7730 OptoRuntime::poly1305_processBlocks_Type(),
7731 stubAddr, stubName, TypePtr::BOTTOM,
7732 input_start, len, acc_start, r_start);
7733 return true;
7734 }
7735
7736 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
7737 address stubAddr;
7738 const char *stubName;
7739 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
7740 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
7741 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
7742 stubName = "intpoly_montgomeryMult_P256";
7743
7744 if (!stubAddr) return false;
7745 null_check_receiver(); // null-check receiver
7746 if (stopped()) return true;
7747
7748 Node* a = argument(1);
7749 Node* b = argument(2);
7750 Node* r = argument(3);
7751
7752 a = must_be_not_null(a, true);
7753 b = must_be_not_null(b, true);
7754 r = must_be_not_null(r, true);
7755
7756 Node* a_start = array_element_address(a, intcon(0), T_LONG);
7757 assert(a_start, "a array is null");
7758 Node* b_start = array_element_address(b, intcon(0), T_LONG);
7759 assert(b_start, "b array is null");
7760 Node* r_start = array_element_address(r, 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::intpoly_montgomeryMult_P256_Type(),
7765 stubAddr, stubName, TypePtr::BOTTOM,
7766 a_start, b_start, r_start);
7767 return true;
7768 }
7769
7770 bool LibraryCallKit::inline_intpoly_assign() {
7771 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
7772 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
7773 const char *stubName = "intpoly_assign";
7774 address stubAddr = StubRoutines::intpoly_assign();
7775 if (!stubAddr) return false;
7776
7777 Node* set = argument(0);
7778 Node* a = argument(1);
7779 Node* b = argument(2);
7780 Node* arr_length = load_array_length(a);
7781
7782 a = must_be_not_null(a, true);
7783 b = must_be_not_null(b, true);
7784
7785 Node* a_start = array_element_address(a, intcon(0), T_LONG);
7786 assert(a_start, "a array is null");
7787 Node* b_start = array_element_address(b, intcon(0), T_LONG);
7788 assert(b_start, "b array is null");
7789
7790 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7791 OptoRuntime::intpoly_assign_Type(),
7792 stubAddr, stubName, TypePtr::BOTTOM,
7793 set, a_start, b_start, arr_length);
7794 return true;
7795 }
7796
7797 //------------------------------inline_digestBase_implCompress-----------------------
7798 //
7799 // Calculate MD5 for single-block byte[] array.
7800 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
7801 //
7802 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
7803 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
7804 //
7805 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
7806 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
7807 //
7808 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
7809 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
7810 //
7811 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
7812 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
7813 //
7814 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
7815 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
7816
7817 Node* digestBase_obj = argument(0);
7818 Node* src = argument(1); // type oop
7819 Node* ofs = argument(2); // type int
7820
7821 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7822 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7823 // failed array check
7824 return false;
7825 }
7826 // Figure out the size and type of the elements we will be copying.
7827 BasicType src_elem = src_type->elem()->array_element_basic_type();
7828 if (src_elem != T_BYTE) {
7829 return false;
7830 }
7831 // 'src_start' points to src array + offset
7832 src = must_be_not_null(src, true);
7833 Node* src_start = array_element_address(src, ofs, src_elem);
7834 Node* state = nullptr;
7835 Node* block_size = nullptr;
7836 address stubAddr;
7837 const char *stubName;
7838
7839 switch(id) {
7840 case vmIntrinsics::_md5_implCompress:
7841 assert(UseMD5Intrinsics, "need MD5 instruction support");
7842 state = get_state_from_digest_object(digestBase_obj, T_INT);
7843 stubAddr = StubRoutines::md5_implCompress();
7844 stubName = "md5_implCompress";
7845 break;
7846 case vmIntrinsics::_sha_implCompress:
7847 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
7848 state = get_state_from_digest_object(digestBase_obj, T_INT);
7849 stubAddr = StubRoutines::sha1_implCompress();
7850 stubName = "sha1_implCompress";
7851 break;
7852 case vmIntrinsics::_sha2_implCompress:
7853 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
7854 state = get_state_from_digest_object(digestBase_obj, T_INT);
7855 stubAddr = StubRoutines::sha256_implCompress();
7856 stubName = "sha256_implCompress";
7857 break;
7858 case vmIntrinsics::_sha5_implCompress:
7859 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
7860 state = get_state_from_digest_object(digestBase_obj, T_LONG);
7861 stubAddr = StubRoutines::sha512_implCompress();
7862 stubName = "sha512_implCompress";
7863 break;
7864 case vmIntrinsics::_sha3_implCompress:
7865 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
7866 state = get_state_from_digest_object(digestBase_obj, T_LONG);
7867 stubAddr = StubRoutines::sha3_implCompress();
7868 stubName = "sha3_implCompress";
7869 block_size = get_block_size_from_digest_object(digestBase_obj);
7870 if (block_size == nullptr) return false;
7871 break;
7872 default:
7873 fatal_unexpected_iid(id);
7874 return false;
7875 }
7876 if (state == nullptr) return false;
7877
7878 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
7879 if (stubAddr == nullptr) return false;
7880
7881 // Call the stub.
7882 Node* call;
7883 if (block_size == nullptr) {
7884 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
7885 stubAddr, stubName, TypePtr::BOTTOM,
7886 src_start, state);
7887 } else {
7888 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
7889 stubAddr, stubName, TypePtr::BOTTOM,
7890 src_start, state, block_size);
7891 }
7892
7893 return true;
7894 }
7895
7896 //------------------------------inline_digestBase_implCompressMB-----------------------
7897 //
7898 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
7899 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
7900 //
7901 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
7902 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7903 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7904 assert((uint)predicate < 5, "sanity");
7905 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
7906
7907 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
7908 Node* src = argument(1); // byte[] array
7909 Node* ofs = argument(2); // type int
7910 Node* limit = argument(3); // type int
7911
7912 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7913 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7914 // failed array check
7915 return false;
7916 }
7917 // Figure out the size and type of the elements we will be copying.
7918 BasicType src_elem = src_type->elem()->array_element_basic_type();
7919 if (src_elem != T_BYTE) {
7920 return false;
7921 }
7922 // 'src_start' points to src array + offset
7923 src = must_be_not_null(src, false);
7924 Node* src_start = array_element_address(src, ofs, src_elem);
7925
7926 const char* klass_digestBase_name = nullptr;
7927 const char* stub_name = nullptr;
7928 address stub_addr = nullptr;
7929 BasicType elem_type = T_INT;
7930
7931 switch (predicate) {
7932 case 0:
7933 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
7934 klass_digestBase_name = "sun/security/provider/MD5";
7935 stub_name = "md5_implCompressMB";
7936 stub_addr = StubRoutines::md5_implCompressMB();
7937 }
7938 break;
7939 case 1:
7940 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
7941 klass_digestBase_name = "sun/security/provider/SHA";
7942 stub_name = "sha1_implCompressMB";
7943 stub_addr = StubRoutines::sha1_implCompressMB();
7944 }
7945 break;
7946 case 2:
7947 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
7948 klass_digestBase_name = "sun/security/provider/SHA2";
7949 stub_name = "sha256_implCompressMB";
7950 stub_addr = StubRoutines::sha256_implCompressMB();
7951 }
7952 break;
7953 case 3:
7954 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
7955 klass_digestBase_name = "sun/security/provider/SHA5";
7956 stub_name = "sha512_implCompressMB";
7957 stub_addr = StubRoutines::sha512_implCompressMB();
7958 elem_type = T_LONG;
7959 }
7960 break;
7961 case 4:
7962 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
7963 klass_digestBase_name = "sun/security/provider/SHA3";
7964 stub_name = "sha3_implCompressMB";
7965 stub_addr = StubRoutines::sha3_implCompressMB();
7966 elem_type = T_LONG;
7967 }
7968 break;
7969 default:
7970 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
7971 }
7972 if (klass_digestBase_name != nullptr) {
7973 assert(stub_addr != nullptr, "Stub is generated");
7974 if (stub_addr == nullptr) return false;
7975
7976 // get DigestBase klass to lookup for SHA klass
7977 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
7978 assert(tinst != nullptr, "digestBase_obj is not instance???");
7979 assert(tinst->is_loaded(), "DigestBase is not loaded");
7980
7981 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
7982 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
7983 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
7984 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
7985 }
7986 return false;
7987 }
7988
7989 //------------------------------inline_digestBase_implCompressMB-----------------------
7990 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
7991 BasicType elem_type, address stubAddr, const char *stubName,
7992 Node* src_start, Node* ofs, Node* limit) {
7993 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
7994 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7995 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
7996 digest_obj = _gvn.transform(digest_obj);
7997
7998 Node* state = get_state_from_digest_object(digest_obj, elem_type);
7999 if (state == nullptr) return false;
8000
8001 Node* block_size = nullptr;
8002 if (strcmp("sha3_implCompressMB", stubName) == 0) {
8003 block_size = get_block_size_from_digest_object(digest_obj);
8004 if (block_size == nullptr) return false;
8005 }
8006
8007 // Call the stub.
8008 Node* call;
8009 if (block_size == nullptr) {
8010 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8011 OptoRuntime::digestBase_implCompressMB_Type(false),
8012 stubAddr, stubName, TypePtr::BOTTOM,
8013 src_start, state, ofs, limit);
8014 } else {
8015 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8016 OptoRuntime::digestBase_implCompressMB_Type(true),
8017 stubAddr, stubName, TypePtr::BOTTOM,
8018 src_start, state, block_size, ofs, limit);
8019 }
8020
8021 // return ofs (int)
8022 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8023 set_result(result);
8024
8025 return true;
8026 }
8027
8028 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8029 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8030 assert(UseAES, "need AES instruction support");
8031 address stubAddr = nullptr;
8032 const char *stubName = nullptr;
8033 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8034 stubName = "galoisCounterMode_AESCrypt";
8035
8036 if (stubAddr == nullptr) return false;
8037
8038 Node* in = argument(0);
8039 Node* inOfs = argument(1);
8040 Node* len = argument(2);
8041 Node* ct = argument(3);
8042 Node* ctOfs = argument(4);
8043 Node* out = argument(5);
8044 Node* outOfs = argument(6);
8045 Node* gctr_object = argument(7);
8046 Node* ghash_object = argument(8);
8047
8048 // (1) in, ct and out are arrays.
8049 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8050 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8051 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8052 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8053 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8054 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8055
8056 // checks are the responsibility of the caller
8057 Node* in_start = in;
8058 Node* ct_start = ct;
8059 Node* out_start = out;
8060 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8061 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8062 in_start = array_element_address(in, inOfs, T_BYTE);
8063 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8064 out_start = array_element_address(out, outOfs, T_BYTE);
8065 }
8066
8067 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8068 // (because of the predicated logic executed earlier).
8069 // so we cast it here safely.
8070 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8071 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8072 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8073 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8074 Node* state = load_field_from_object(ghash_object, "state", "[J");
8075
8076 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8077 return false;
8078 }
8079 // cast it to what we know it will be at runtime
8080 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8081 assert(tinst != nullptr, "GCTR obj is null");
8082 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8083 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8084 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8085 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8086 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8087 const TypeOopPtr* xtype = aklass->as_instance_type();
8088 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8089 aescrypt_object = _gvn.transform(aescrypt_object);
8090 // we need to get the start of the aescrypt_object's expanded key array
8091 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8092 if (k_start == nullptr) return false;
8093 // similarly, get the start address of the r vector
8094 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8095 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8096 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8097
8098
8099 // Call the stub, passing params
8100 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8101 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8102 stubAddr, stubName, TypePtr::BOTTOM,
8103 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8104
8105 // return cipher length (int)
8106 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8107 set_result(retvalue);
8108
8109 return true;
8110 }
8111
8112 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8113 // Return node representing slow path of predicate check.
8114 // the pseudo code we want to emulate with this predicate is:
8115 // for encryption:
8116 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8117 // for decryption:
8118 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8119 // note cipher==plain is more conservative than the original java code but that's OK
8120 //
8121
8122 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8123 // The receiver was checked for null already.
8124 Node* objGCTR = argument(7);
8125 // Load embeddedCipher field of GCTR object.
8126 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8127 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8128
8129 // get AESCrypt klass for instanceOf check
8130 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8131 // will have same classloader as CipherBlockChaining object
8132 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8133 assert(tinst != nullptr, "GCTR obj is null");
8134 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8135
8136 // we want to do an instanceof comparison against the AESCrypt class
8137 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8138 if (!klass_AESCrypt->is_loaded()) {
8139 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8140 Node* ctrl = control();
8141 set_control(top()); // no regular fast path
8142 return ctrl;
8143 }
8144
8145 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8146 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8147 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8148 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8149 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8150
8151 return instof_false; // even if it is null
8152 }
8153
8154 //------------------------------get_state_from_digest_object-----------------------
8155 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8156 const char* state_type;
8157 switch (elem_type) {
8158 case T_BYTE: state_type = "[B"; break;
8159 case T_INT: state_type = "[I"; break;
8160 case T_LONG: state_type = "[J"; break;
8161 default: ShouldNotReachHere();
8162 }
8163 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8164 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8165 if (digest_state == nullptr) return (Node *) nullptr;
8166
8167 // now have the array, need to get the start address of the state array
8168 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8169 return state;
8170 }
8171
8172 //------------------------------get_block_size_from_sha3_object----------------------------------
8173 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8174 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8175 assert (block_size != nullptr, "sanity");
8176 return block_size;
8177 }
8178
8179 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8180 // Return node representing slow path of predicate check.
8181 // the pseudo code we want to emulate with this predicate is:
8182 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8183 //
8184 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8185 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8186 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8187 assert((uint)predicate < 5, "sanity");
8188
8189 // The receiver was checked for null already.
8190 Node* digestBaseObj = argument(0);
8191
8192 // get DigestBase klass for instanceOf check
8193 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8194 assert(tinst != nullptr, "digestBaseObj is null");
8195 assert(tinst->is_loaded(), "DigestBase is not loaded");
8196
8197 const char* klass_name = nullptr;
8198 switch (predicate) {
8199 case 0:
8200 if (UseMD5Intrinsics) {
8201 // we want to do an instanceof comparison against the MD5 class
8202 klass_name = "sun/security/provider/MD5";
8203 }
8204 break;
8205 case 1:
8206 if (UseSHA1Intrinsics) {
8207 // we want to do an instanceof comparison against the SHA class
8208 klass_name = "sun/security/provider/SHA";
8209 }
8210 break;
8211 case 2:
8212 if (UseSHA256Intrinsics) {
8213 // we want to do an instanceof comparison against the SHA2 class
8214 klass_name = "sun/security/provider/SHA2";
8215 }
8216 break;
8217 case 3:
8218 if (UseSHA512Intrinsics) {
8219 // we want to do an instanceof comparison against the SHA5 class
8220 klass_name = "sun/security/provider/SHA5";
8221 }
8222 break;
8223 case 4:
8224 if (UseSHA3Intrinsics) {
8225 // we want to do an instanceof comparison against the SHA3 class
8226 klass_name = "sun/security/provider/SHA3";
8227 }
8228 break;
8229 default:
8230 fatal("unknown SHA intrinsic predicate: %d", predicate);
8231 }
8232
8233 ciKlass* klass = nullptr;
8234 if (klass_name != nullptr) {
8235 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8236 }
8237 if ((klass == nullptr) || !klass->is_loaded()) {
8238 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8239 Node* ctrl = control();
8240 set_control(top()); // no intrinsic path
8241 return ctrl;
8242 }
8243 ciInstanceKlass* instklass = klass->as_instance_klass();
8244
8245 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8246 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8247 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8248 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8249
8250 return instof_false; // even if it is null
8251 }
8252
8253 //-------------inline_fma-----------------------------------
8254 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8255 Node *a = nullptr;
8256 Node *b = nullptr;
8257 Node *c = nullptr;
8258 Node* result = nullptr;
8259 switch (id) {
8260 case vmIntrinsics::_fmaD:
8261 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8262 // no receiver since it is static method
8263 a = round_double_node(argument(0));
8264 b = round_double_node(argument(2));
8265 c = round_double_node(argument(4));
8266 result = _gvn.transform(new FmaDNode(control(), a, b, c));
8267 break;
8268 case vmIntrinsics::_fmaF:
8269 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8270 a = argument(0);
8271 b = argument(1);
8272 c = argument(2);
8273 result = _gvn.transform(new FmaFNode(control(), a, b, c));
8274 break;
8275 default:
8276 fatal_unexpected_iid(id); break;
8277 }
8278 set_result(result);
8279 return true;
8280 }
8281
8282 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8283 // argument(0) is receiver
8284 Node* codePoint = argument(1);
8285 Node* n = nullptr;
8286
8287 switch (id) {
8288 case vmIntrinsics::_isDigit :
8289 n = new DigitNode(control(), codePoint);
8290 break;
8291 case vmIntrinsics::_isLowerCase :
8292 n = new LowerCaseNode(control(), codePoint);
8293 break;
8294 case vmIntrinsics::_isUpperCase :
8295 n = new UpperCaseNode(control(), codePoint);
8296 break;
8297 case vmIntrinsics::_isWhitespace :
8298 n = new WhitespaceNode(control(), codePoint);
8299 break;
8300 default:
8301 fatal_unexpected_iid(id);
8302 }
8303
8304 set_result(_gvn.transform(n));
8305 return true;
8306 }
8307
8308 //------------------------------inline_fp_min_max------------------------------
8309 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
8310 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
8311
8312 // The intrinsic should be used only when the API branches aren't predictable,
8313 // the last one performing the most important comparison. The following heuristic
8314 // uses the branch statistics to eventually bail out if necessary.
8315
8316 ciMethodData *md = callee()->method_data();
8317
8318 if ( md != nullptr && md->is_mature() && md->invocation_count() > 0 ) {
8319 ciCallProfile cp = caller()->call_profile_at_bci(bci());
8320
8321 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
8322 // Bail out if the call-site didn't contribute enough to the statistics.
8323 return false;
8324 }
8325
8326 uint taken = 0, not_taken = 0;
8327
8328 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
8329 if (p->is_BranchData()) {
8330 taken = ((ciBranchData*)p)->taken();
8331 not_taken = ((ciBranchData*)p)->not_taken();
8332 }
8333 }
8334
8335 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
8336 balance = balance < 0 ? -balance : balance;
8337 if ( balance > 0.2 ) {
8338 // Bail out if the most important branch is predictable enough.
8339 return false;
8340 }
8341 }
8342 */
8343
8344 Node *a = nullptr;
8345 Node *b = nullptr;
8346 Node *n = nullptr;
8347 switch (id) {
8348 case vmIntrinsics::_maxF:
8349 case vmIntrinsics::_minF:
8350 case vmIntrinsics::_maxF_strict:
8351 case vmIntrinsics::_minF_strict:
8352 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
8353 a = argument(0);
8354 b = argument(1);
8355 break;
8356 case vmIntrinsics::_maxD:
8357 case vmIntrinsics::_minD:
8358 case vmIntrinsics::_maxD_strict:
8359 case vmIntrinsics::_minD_strict:
8360 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
8361 a = round_double_node(argument(0));
8362 b = round_double_node(argument(2));
8363 break;
8364 default:
8365 fatal_unexpected_iid(id);
8366 break;
8367 }
8368 switch (id) {
8369 case vmIntrinsics::_maxF:
8370 case vmIntrinsics::_maxF_strict:
8371 n = new MaxFNode(a, b);
8372 break;
8373 case vmIntrinsics::_minF:
8374 case vmIntrinsics::_minF_strict:
8375 n = new MinFNode(a, b);
8376 break;
8377 case vmIntrinsics::_maxD:
8378 case vmIntrinsics::_maxD_strict:
8379 n = new MaxDNode(a, b);
8380 break;
8381 case vmIntrinsics::_minD:
8382 case vmIntrinsics::_minD_strict:
8383 n = new MinDNode(a, b);
8384 break;
8385 default:
8386 fatal_unexpected_iid(id);
8387 break;
8388 }
8389 set_result(_gvn.transform(n));
8390 return true;
8391 }
8392
8393 bool LibraryCallKit::inline_profileBoolean() {
8394 Node* counts = argument(1);
8395 const TypeAryPtr* ary = nullptr;
8396 ciArray* aobj = nullptr;
8397 if (counts->is_Con()
8398 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8399 && (aobj = ary->const_oop()->as_array()) != nullptr
8400 && (aobj->length() == 2)) {
8401 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8402 jint false_cnt = aobj->element_value(0).as_int();
8403 jint true_cnt = aobj->element_value(1).as_int();
8404
8405 if (C->log() != nullptr) {
8406 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8407 false_cnt, true_cnt);
8408 }
8409
8410 if (false_cnt + true_cnt == 0) {
8411 // According to profile, never executed.
8412 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8413 Deoptimization::Action_reinterpret);
8414 return true;
8415 }
8416
8417 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8418 // is a number of each value occurrences.
8419 Node* result = argument(0);
8420 if (false_cnt == 0 || true_cnt == 0) {
8421 // According to profile, one value has been never seen.
8422 int expected_val = (false_cnt == 0) ? 1 : 0;
8423
8424 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8425 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8426
8427 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8428 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8429 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8430
8431 { // Slow path: uncommon trap for never seen value and then reexecute
8432 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8433 // the value has been seen at least once.
8434 PreserveJVMState pjvms(this);
8435 PreserveReexecuteState preexecs(this);
8436 jvms()->set_should_reexecute(true);
8437
8438 set_control(slow_path);
8439 set_i_o(i_o());
8440
8441 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8442 Deoptimization::Action_reinterpret);
8443 }
8444 // The guard for never seen value enables sharpening of the result and
8445 // returning a constant. It allows to eliminate branches on the same value
8446 // later on.
8447 set_control(fast_path);
8448 result = intcon(expected_val);
8449 }
8450 // Stop profiling.
8451 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8452 // By replacing method body with profile data (represented as ProfileBooleanNode
8453 // on IR level) we effectively disable profiling.
8454 // It enables full speed execution once optimized code is generated.
8455 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8456 C->record_for_igvn(profile);
8457 set_result(profile);
8458 return true;
8459 } else {
8460 // Continue profiling.
8461 // Profile data isn't available at the moment. So, execute method's bytecode version.
8462 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8463 // is compiled and counters aren't available since corresponding MethodHandle
8464 // isn't a compile-time constant.
8465 return false;
8466 }
8467 }
8468
8469 bool LibraryCallKit::inline_isCompileConstant() {
8470 Node* n = argument(0);
8471 set_result(n->is_Con() ? intcon(1) : intcon(0));
8472 return true;
8473 }
8474
8475 //------------------------------- inline_getObjectSize --------------------------------------
8476 //
8477 // Calculate the runtime size of the object/array.
8478 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8479 //
8480 bool LibraryCallKit::inline_getObjectSize() {
8481 Node* obj = argument(3);
8482 Node* klass_node = load_object_klass(obj);
8483
8484 jint layout_con = Klass::_lh_neutral_value;
8485 Node* layout_val = get_layout_helper(klass_node, layout_con);
8486 int layout_is_con = (layout_val == nullptr);
8487
8488 if (layout_is_con) {
8489 // Layout helper is constant, can figure out things at compile time.
8490
8491 if (Klass::layout_helper_is_instance(layout_con)) {
8492 // Instance case: layout_con contains the size itself.
8493 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8494 set_result(size);
8495 } else {
8496 // Array case: size is round(header + element_size*arraylength).
8497 // Since arraylength is different for every array instance, we have to
8498 // compute the whole thing at runtime.
8499
8500 Node* arr_length = load_array_length(obj);
8501
8502 int round_mask = MinObjAlignmentInBytes - 1;
8503 int hsize = Klass::layout_helper_header_size(layout_con);
8504 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8505
8506 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8507 round_mask = 0; // strength-reduce it if it goes away completely
8508 }
8509 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8510 Node* header_size = intcon(hsize + round_mask);
8511
8512 Node* lengthx = ConvI2X(arr_length);
8513 Node* headerx = ConvI2X(header_size);
8514
8515 Node* abody = lengthx;
8516 if (eshift != 0) {
8517 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8518 }
8519 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8520 if (round_mask != 0) {
8521 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8522 }
8523 size = ConvX2L(size);
8524 set_result(size);
8525 }
8526 } else {
8527 // Layout helper is not constant, need to test for array-ness at runtime.
8528
8529 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8530 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8531 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8532 record_for_igvn(result_reg);
8533
8534 Node* array_ctl = generate_array_guard(klass_node, nullptr);
8535 if (array_ctl != nullptr) {
8536 // Array case: size is round(header + element_size*arraylength).
8537 // Since arraylength is different for every array instance, we have to
8538 // compute the whole thing at runtime.
8539
8540 PreserveJVMState pjvms(this);
8541 set_control(array_ctl);
8542 Node* arr_length = load_array_length(obj);
8543
8544 int round_mask = MinObjAlignmentInBytes - 1;
8545 Node* mask = intcon(round_mask);
8546
8547 Node* hss = intcon(Klass::_lh_header_size_shift);
8548 Node* hsm = intcon(Klass::_lh_header_size_mask);
8549 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8550 header_size = _gvn.transform(new AndINode(header_size, hsm));
8551 header_size = _gvn.transform(new AddINode(header_size, mask));
8552
8553 // There is no need to mask or shift this value.
8554 // The semantics of LShiftINode include an implicit mask to 0x1F.
8555 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8556 Node* elem_shift = layout_val;
8557
8558 Node* lengthx = ConvI2X(arr_length);
8559 Node* headerx = ConvI2X(header_size);
8560
8561 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
8562 Node* size = _gvn.transform(new AddXNode(headerx, abody));
8563 if (round_mask != 0) {
8564 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
8565 }
8566 size = ConvX2L(size);
8567
8568 result_reg->init_req(_array_path, control());
8569 result_val->init_req(_array_path, size);
8570 }
8571
8572 if (!stopped()) {
8573 // Instance case: the layout helper gives us instance size almost directly,
8574 // but we need to mask out the _lh_instance_slow_path_bit.
8575 Node* size = ConvI2X(layout_val);
8576 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
8577 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
8578 size = _gvn.transform(new AndXNode(size, mask));
8579 size = ConvX2L(size);
8580
8581 result_reg->init_req(_instance_path, control());
8582 result_val->init_req(_instance_path, size);
8583 }
8584
8585 set_result(result_reg, result_val);
8586 }
8587
8588 return true;
8589 }
8590
8591 //------------------------------- inline_blackhole --------------------------------------
8592 //
8593 // Make sure all arguments to this node are alive.
8594 // This matches methods that were requested to be blackholed through compile commands.
8595 //
8596 bool LibraryCallKit::inline_blackhole() {
8597 assert(callee()->is_static(), "Should have been checked before: only static methods here");
8598 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
8599 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
8600
8601 // Blackhole node pinches only the control, not memory. This allows
8602 // the blackhole to be pinned in the loop that computes blackholed
8603 // values, but have no other side effects, like breaking the optimizations
8604 // across the blackhole.
8605
8606 Node* bh = _gvn.transform(new BlackholeNode(control()));
8607 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
8608
8609 // Bind call arguments as blackhole arguments to keep them alive
8610 uint nargs = callee()->arg_size();
8611 for (uint i = 0; i < nargs; i++) {
8612 bh->add_req(argument(i));
8613 }
8614
8615 return true;
8616 }