1 /*
2 * Copyright (c) 1999, 2025, 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 "opto/vectornode.hpp"
54 #include "prims/jvmtiExport.hpp"
55 #include "prims/jvmtiThreadState.hpp"
56 #include "prims/unsafe.hpp"
57 #include "runtime/jniHandles.inline.hpp"
58 #include "runtime/objectMonitor.hpp"
59 #include "runtime/sharedRuntime.hpp"
60 #include "runtime/stubRoutines.hpp"
61 #include "utilities/macros.hpp"
62 #include "utilities/powerOfTwo.hpp"
63
64 //---------------------------make_vm_intrinsic----------------------------
65 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
66 vmIntrinsicID id = m->intrinsic_id();
67 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
68
69 if (!m->is_loaded()) {
70 // Do not attempt to inline unloaded methods.
71 return nullptr;
72 }
73
74 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
75 bool is_available = false;
76
77 {
78 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
79 // the compiler must transition to '_thread_in_vm' state because both
80 // methods access VM-internal data.
81 VM_ENTRY_MARK;
82 methodHandle mh(THREAD, m->get_Method());
83 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
84 if (is_available && is_virtual) {
85 is_available = vmIntrinsics::does_virtual_dispatch(id);
86 }
87 }
88
89 if (is_available) {
90 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
91 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
92 return new LibraryIntrinsic(m, is_virtual,
93 vmIntrinsics::predicates_needed(id),
94 vmIntrinsics::does_virtual_dispatch(id),
95 id);
96 } else {
97 return nullptr;
98 }
99 }
100
101 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
102 LibraryCallKit kit(jvms, this);
103 Compile* C = kit.C;
104 int nodes = C->unique();
105 #ifndef PRODUCT
106 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
107 char buf[1000];
108 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
109 tty->print_cr("Intrinsic %s", str);
110 }
111 #endif
112 ciMethod* callee = kit.callee();
113 const int bci = kit.bci();
114 #ifdef ASSERT
115 Node* ctrl = kit.control();
116 #endif
117 // Try to inline the intrinsic.
118 if (callee->check_intrinsic_candidate() &&
119 kit.try_to_inline(_last_predicate)) {
120 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
121 : "(intrinsic)";
122 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
123 if (C->print_intrinsics() || C->print_inlining()) {
124 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
125 }
126 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
127 if (C->log()) {
128 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
129 vmIntrinsics::name_at(intrinsic_id()),
130 (is_virtual() ? " virtual='1'" : ""),
131 C->unique() - nodes);
132 }
133 // Push the result from the inlined method onto the stack.
134 kit.push_result();
135 C->print_inlining_update(this);
136 return kit.transfer_exceptions_into_jvms();
137 }
138
139 // The intrinsic bailed out
140 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
141 if (jvms->has_method()) {
142 // Not a root compile.
143 const char* msg;
144 if (callee->intrinsic_candidate()) {
145 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
146 } else {
147 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
148 : "failed to inline (intrinsic), method not annotated";
149 }
150 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
151 if (C->print_intrinsics() || C->print_inlining()) {
152 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
153 }
154 } else {
155 // Root compile
156 ResourceMark rm;
157 stringStream msg_stream;
158 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
159 vmIntrinsics::name_at(intrinsic_id()),
160 is_virtual() ? " (virtual)" : "", bci);
161 const char *msg = msg_stream.freeze();
162 log_debug(jit, inlining)("%s", msg);
163 if (C->print_intrinsics() || C->print_inlining()) {
164 tty->print("%s", msg);
165 }
166 }
167 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
168 C->print_inlining_update(this);
169
170 return nullptr;
171 }
172
173 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
174 LibraryCallKit kit(jvms, this);
175 Compile* C = kit.C;
176 int nodes = C->unique();
177 _last_predicate = predicate;
178 #ifndef PRODUCT
179 assert(is_predicated() && predicate < predicates_count(), "sanity");
180 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
181 char buf[1000];
182 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
183 tty->print_cr("Predicate for intrinsic %s", str);
184 }
185 #endif
186 ciMethod* callee = kit.callee();
187 const int bci = kit.bci();
188
189 Node* slow_ctl = kit.try_to_predicate(predicate);
190 if (!kit.failing()) {
191 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
192 : "(intrinsic, predicate)";
193 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
194 if (C->print_intrinsics() || C->print_inlining()) {
195 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
196 }
197 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
198 if (C->log()) {
199 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
200 vmIntrinsics::name_at(intrinsic_id()),
201 (is_virtual() ? " virtual='1'" : ""),
202 C->unique() - nodes);
203 }
204 return slow_ctl; // Could be null if the check folds.
205 }
206
207 // The intrinsic bailed out
208 if (jvms->has_method()) {
209 // Not a root compile.
210 const char* msg = "failed to generate predicate for intrinsic";
211 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
212 if (C->print_intrinsics() || C->print_inlining()) {
213 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
214 }
215 } else {
216 // Root compile
217 ResourceMark rm;
218 stringStream msg_stream;
219 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
220 vmIntrinsics::name_at(intrinsic_id()),
221 is_virtual() ? " (virtual)" : "", bci);
222 const char *msg = msg_stream.freeze();
223 log_debug(jit, inlining)("%s", msg);
224 if (C->print_intrinsics() || C->print_inlining()) {
225 C->print_inlining_stream()->print("%s", msg);
226 }
227 }
228 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
229 return nullptr;
230 }
231
232 bool LibraryCallKit::try_to_inline(int predicate) {
233 // Handle symbolic names for otherwise undistinguished boolean switches:
234 const bool is_store = true;
235 const bool is_compress = true;
236 const bool is_static = true;
237 const bool is_volatile = true;
238
239 if (!jvms()->has_method()) {
240 // Root JVMState has a null method.
241 assert(map()->memory()->Opcode() == Op_Parm, "");
242 // Insert the memory aliasing node
243 set_all_memory(reset_memory());
244 }
245 assert(merged_memory(), "");
246
247 switch (intrinsic_id()) {
248 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
249 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
250 case vmIntrinsics::_getClass: return inline_native_getClass();
251
252 case vmIntrinsics::_ceil:
253 case vmIntrinsics::_floor:
254 case vmIntrinsics::_rint:
255 case vmIntrinsics::_dsin:
256 case vmIntrinsics::_dcos:
257 case vmIntrinsics::_dtan:
258 case vmIntrinsics::_dtanh:
259 case vmIntrinsics::_dabs:
260 case vmIntrinsics::_fabs:
261 case vmIntrinsics::_iabs:
262 case vmIntrinsics::_labs:
263 case vmIntrinsics::_datan2:
264 case vmIntrinsics::_dsqrt:
265 case vmIntrinsics::_dsqrt_strict:
266 case vmIntrinsics::_dexp:
267 case vmIntrinsics::_dlog:
268 case vmIntrinsics::_dlog10:
269 case vmIntrinsics::_dpow:
270 case vmIntrinsics::_dcopySign:
271 case vmIntrinsics::_fcopySign:
272 case vmIntrinsics::_dsignum:
273 case vmIntrinsics::_roundF:
274 case vmIntrinsics::_roundD:
275 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
276
277 case vmIntrinsics::_notify:
278 case vmIntrinsics::_notifyAll:
279 return inline_notify(intrinsic_id());
280
281 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
282 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
283 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
284 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
285 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
286 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
287 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
288 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
289 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
290 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
291 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
292 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
293 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
294 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
295
296 case vmIntrinsics::_arraycopy: return inline_arraycopy();
297
298 case vmIntrinsics::_arraySort: return inline_array_sort();
299 case vmIntrinsics::_arrayPartition: return inline_array_partition();
300
301 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
302 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
303 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
304 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
305
306 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
307 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
308 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
309 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
310 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
311 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
312 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
313 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
314
315 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
316
317 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
318
319 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
320 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
321 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
322 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
323
324 case vmIntrinsics::_compressStringC:
325 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
326 case vmIntrinsics::_inflateStringC:
327 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
328
329 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
330 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
331 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
332 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
333 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
334 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
335 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
336 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
337 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
338
339 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
340 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
341 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
342 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
343 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
344 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
345 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
346 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
347 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
348
349 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
350 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
351 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
352 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
353 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
354 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
355 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
356 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
357 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
358
359 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
360 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
361 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
362 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
363 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
364 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
365 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
366 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
367 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
368
369 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
370 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
371 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
372 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
373
374 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
375 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
376 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
377 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
378
379 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
380 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
381 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
382 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
383 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
384 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
385 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
386 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
387 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
388
389 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
390 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
391 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
392 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
393 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
394 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
395 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
396 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
397 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
398
399 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
400 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
401 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
402 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
403 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
404 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
405 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
406 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
407 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
408
409 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
410 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
411 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
412 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
413 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
414 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
415 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
416 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
417 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
418
419 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
420 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
421 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
422 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
423 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
424
425 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
426 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
427 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
428 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
429 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
430 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
431 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
432 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
433 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
434 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
435 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
436 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
437 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
438 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
439 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
440 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
441 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
442 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
443 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
444 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
445
446 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
447 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
448 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
449 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
450 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
451 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
452 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
453 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
454 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
455 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
456 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
457 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
458 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
459 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
460 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
461
462 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
463 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
464 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
465 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
466
467 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
468 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
469 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
470 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
471 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
472
473 case vmIntrinsics::_loadFence:
474 case vmIntrinsics::_storeFence:
475 case vmIntrinsics::_storeStoreFence:
476 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
477
478 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
479
480 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
481 case vmIntrinsics::_currentThread: return inline_native_currentThread();
482 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
483
484 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
485 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
486
487 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
488 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
489
490 #if INCLUDE_JVMTI
491 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
492 "notifyJvmtiStart", true, false);
493 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
494 "notifyJvmtiEnd", false, true);
495 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
496 "notifyJvmtiMount", false, false);
497 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
498 "notifyJvmtiUnmount", false, false);
499 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
500 #endif
501
502 #ifdef JFR_HAVE_INTRINSICS
503 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
504 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
505 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
506 #endif
507 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
508 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
509 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
510 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
511 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
512 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
513 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
514 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
515 case vmIntrinsics::_getLength: return inline_native_getLength();
516 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
517 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
518 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
519 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
520 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
521 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
522 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
523
524 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
525 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
526
527 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
528
529 case vmIntrinsics::_isInstance:
530 case vmIntrinsics::_getModifiers:
531 case vmIntrinsics::_isInterface:
532 case vmIntrinsics::_isArray:
533 case vmIntrinsics::_isPrimitive:
534 case vmIntrinsics::_isHidden:
535 case vmIntrinsics::_getSuperclass:
536 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
537
538 case vmIntrinsics::_floatToRawIntBits:
539 case vmIntrinsics::_floatToIntBits:
540 case vmIntrinsics::_intBitsToFloat:
541 case vmIntrinsics::_doubleToRawLongBits:
542 case vmIntrinsics::_doubleToLongBits:
543 case vmIntrinsics::_longBitsToDouble:
544 case vmIntrinsics::_floatToFloat16:
545 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
546
547 case vmIntrinsics::_floatIsFinite:
548 case vmIntrinsics::_floatIsInfinite:
549 case vmIntrinsics::_doubleIsFinite:
550 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
551
552 case vmIntrinsics::_numberOfLeadingZeros_i:
553 case vmIntrinsics::_numberOfLeadingZeros_l:
554 case vmIntrinsics::_numberOfTrailingZeros_i:
555 case vmIntrinsics::_numberOfTrailingZeros_l:
556 case vmIntrinsics::_bitCount_i:
557 case vmIntrinsics::_bitCount_l:
558 case vmIntrinsics::_reverse_i:
559 case vmIntrinsics::_reverse_l:
560 case vmIntrinsics::_reverseBytes_i:
561 case vmIntrinsics::_reverseBytes_l:
562 case vmIntrinsics::_reverseBytes_s:
563 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
564
565 case vmIntrinsics::_compress_i:
566 case vmIntrinsics::_compress_l:
567 case vmIntrinsics::_expand_i:
568 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
569
570 case vmIntrinsics::_compareUnsigned_i:
571 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
572
573 case vmIntrinsics::_divideUnsigned_i:
574 case vmIntrinsics::_divideUnsigned_l:
575 case vmIntrinsics::_remainderUnsigned_i:
576 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
577
578 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
579
580 case vmIntrinsics::_Reference_get: return inline_reference_get();
581 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
582 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
583 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
584 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
585
586 case vmIntrinsics::_Class_cast: return inline_Class_cast();
587
588 case vmIntrinsics::_aescrypt_encryptBlock:
589 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
590
591 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
592 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
593 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
594
595 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
596 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
597 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
598
599 case vmIntrinsics::_counterMode_AESCrypt:
600 return inline_counterMode_AESCrypt(intrinsic_id());
601
602 case vmIntrinsics::_galoisCounterMode_AESCrypt:
603 return inline_galoisCounterMode_AESCrypt();
604
605 case vmIntrinsics::_md5_implCompress:
606 case vmIntrinsics::_sha_implCompress:
607 case vmIntrinsics::_sha2_implCompress:
608 case vmIntrinsics::_sha5_implCompress:
609 case vmIntrinsics::_sha3_implCompress:
610 return inline_digestBase_implCompress(intrinsic_id());
611
612 case vmIntrinsics::_digestBase_implCompressMB:
613 return inline_digestBase_implCompressMB(predicate);
614
615 case vmIntrinsics::_multiplyToLen:
616 return inline_multiplyToLen();
617
618 case vmIntrinsics::_squareToLen:
619 return inline_squareToLen();
620
621 case vmIntrinsics::_mulAdd:
622 return inline_mulAdd();
623
624 case vmIntrinsics::_montgomeryMultiply:
625 return inline_montgomeryMultiply();
626 case vmIntrinsics::_montgomerySquare:
627 return inline_montgomerySquare();
628
629 case vmIntrinsics::_bigIntegerRightShiftWorker:
630 return inline_bigIntegerShift(true);
631 case vmIntrinsics::_bigIntegerLeftShiftWorker:
632 return inline_bigIntegerShift(false);
633
634 case vmIntrinsics::_vectorizedMismatch:
635 return inline_vectorizedMismatch();
636
637 case vmIntrinsics::_ghash_processBlocks:
638 return inline_ghash_processBlocks();
639 case vmIntrinsics::_chacha20Block:
640 return inline_chacha20Block();
641 case vmIntrinsics::_base64_encodeBlock:
642 return inline_base64_encodeBlock();
643 case vmIntrinsics::_base64_decodeBlock:
644 return inline_base64_decodeBlock();
645 case vmIntrinsics::_poly1305_processBlocks:
646 return inline_poly1305_processBlocks();
647 case vmIntrinsics::_intpoly_montgomeryMult_P256:
648 return inline_intpoly_montgomeryMult_P256();
649 case vmIntrinsics::_intpoly_assign:
650 return inline_intpoly_assign();
651 case vmIntrinsics::_encodeISOArray:
652 case vmIntrinsics::_encodeByteISOArray:
653 return inline_encodeISOArray(false);
654 case vmIntrinsics::_encodeAsciiArray:
655 return inline_encodeISOArray(true);
656
657 case vmIntrinsics::_updateCRC32:
658 return inline_updateCRC32();
659 case vmIntrinsics::_updateBytesCRC32:
660 return inline_updateBytesCRC32();
661 case vmIntrinsics::_updateByteBufferCRC32:
662 return inline_updateByteBufferCRC32();
663
664 case vmIntrinsics::_updateBytesCRC32C:
665 return inline_updateBytesCRC32C();
666 case vmIntrinsics::_updateDirectByteBufferCRC32C:
667 return inline_updateDirectByteBufferCRC32C();
668
669 case vmIntrinsics::_updateBytesAdler32:
670 return inline_updateBytesAdler32();
671 case vmIntrinsics::_updateByteBufferAdler32:
672 return inline_updateByteBufferAdler32();
673
674 case vmIntrinsics::_profileBoolean:
675 return inline_profileBoolean();
676 case vmIntrinsics::_isCompileConstant:
677 return inline_isCompileConstant();
678
679 case vmIntrinsics::_countPositives:
680 return inline_countPositives();
681
682 case vmIntrinsics::_fmaD:
683 case vmIntrinsics::_fmaF:
684 return inline_fma(intrinsic_id());
685
686 case vmIntrinsics::_isDigit:
687 case vmIntrinsics::_isLowerCase:
688 case vmIntrinsics::_isUpperCase:
689 case vmIntrinsics::_isWhitespace:
690 return inline_character_compare(intrinsic_id());
691
692 case vmIntrinsics::_min:
693 case vmIntrinsics::_max:
694 case vmIntrinsics::_min_strict:
695 case vmIntrinsics::_max_strict:
696 return inline_min_max(intrinsic_id());
697
698 case vmIntrinsics::_maxF:
699 case vmIntrinsics::_minF:
700 case vmIntrinsics::_maxD:
701 case vmIntrinsics::_minD:
702 case vmIntrinsics::_maxF_strict:
703 case vmIntrinsics::_minF_strict:
704 case vmIntrinsics::_maxD_strict:
705 case vmIntrinsics::_minD_strict:
706 return inline_fp_min_max(intrinsic_id());
707
708 case vmIntrinsics::_VectorUnaryOp:
709 return inline_vector_nary_operation(1);
710 case vmIntrinsics::_VectorBinaryOp:
711 return inline_vector_nary_operation(2);
712 case vmIntrinsics::_VectorTernaryOp:
713 return inline_vector_nary_operation(3);
714 case vmIntrinsics::_VectorFromBitsCoerced:
715 return inline_vector_frombits_coerced();
716 case vmIntrinsics::_VectorMaskOp:
717 return inline_vector_mask_operation();
718 case vmIntrinsics::_VectorLoadOp:
719 return inline_vector_mem_operation(/*is_store=*/false);
720 case vmIntrinsics::_VectorLoadMaskedOp:
721 return inline_vector_mem_masked_operation(/*is_store*/false);
722 case vmIntrinsics::_VectorStoreOp:
723 return inline_vector_mem_operation(/*is_store=*/true);
724 case vmIntrinsics::_VectorStoreMaskedOp:
725 return inline_vector_mem_masked_operation(/*is_store=*/true);
726 case vmIntrinsics::_VectorGatherOp:
727 return inline_vector_gather_scatter(/*is_scatter*/ false);
728 case vmIntrinsics::_VectorScatterOp:
729 return inline_vector_gather_scatter(/*is_scatter*/ true);
730 case vmIntrinsics::_VectorReductionCoerced:
731 return inline_vector_reduction();
732 case vmIntrinsics::_VectorTest:
733 return inline_vector_test();
734 case vmIntrinsics::_VectorBlend:
735 return inline_vector_blend();
736 case vmIntrinsics::_VectorRearrange:
737 return inline_vector_rearrange();
738 case vmIntrinsics::_VectorSelectFrom:
739 return inline_vector_select_from();
740 case vmIntrinsics::_VectorCompare:
741 return inline_vector_compare();
742 case vmIntrinsics::_VectorBroadcastInt:
743 return inline_vector_broadcast_int();
744 case vmIntrinsics::_VectorConvert:
745 return inline_vector_convert();
746 case vmIntrinsics::_VectorInsert:
747 return inline_vector_insert();
748 case vmIntrinsics::_VectorExtract:
749 return inline_vector_extract();
750 case vmIntrinsics::_VectorCompressExpand:
751 return inline_vector_compress_expand();
752 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
753 return inline_vector_select_from_two_vectors();
754 case vmIntrinsics::_IndexVector:
755 return inline_index_vector();
756 case vmIntrinsics::_IndexPartiallyInUpperRange:
757 return inline_index_partially_in_upper_range();
758
759 case vmIntrinsics::_getObjectSize:
760 return inline_getObjectSize();
761
762 case vmIntrinsics::_blackhole:
763 return inline_blackhole();
764
765 default:
766 // If you get here, it may be that someone has added a new intrinsic
767 // to the list in vmIntrinsics.hpp without implementing it here.
768 #ifndef PRODUCT
769 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
770 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
771 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
772 }
773 #endif
774 return false;
775 }
776 }
777
778 Node* LibraryCallKit::try_to_predicate(int predicate) {
779 if (!jvms()->has_method()) {
780 // Root JVMState has a null method.
781 assert(map()->memory()->Opcode() == Op_Parm, "");
782 // Insert the memory aliasing node
783 set_all_memory(reset_memory());
784 }
785 assert(merged_memory(), "");
786
787 switch (intrinsic_id()) {
788 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
789 return inline_cipherBlockChaining_AESCrypt_predicate(false);
790 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
791 return inline_cipherBlockChaining_AESCrypt_predicate(true);
792 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
793 return inline_electronicCodeBook_AESCrypt_predicate(false);
794 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
795 return inline_electronicCodeBook_AESCrypt_predicate(true);
796 case vmIntrinsics::_counterMode_AESCrypt:
797 return inline_counterMode_AESCrypt_predicate();
798 case vmIntrinsics::_digestBase_implCompressMB:
799 return inline_digestBase_implCompressMB_predicate(predicate);
800 case vmIntrinsics::_galoisCounterMode_AESCrypt:
801 return inline_galoisCounterMode_AESCrypt_predicate();
802
803 default:
804 // If you get here, it may be that someone has added a new intrinsic
805 // to the list in vmIntrinsics.hpp without implementing it here.
806 #ifndef PRODUCT
807 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
808 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
809 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
810 }
811 #endif
812 Node* slow_ctl = control();
813 set_control(top()); // No fast path intrinsic
814 return slow_ctl;
815 }
816 }
817
818 //------------------------------set_result-------------------------------
819 // Helper function for finishing intrinsics.
820 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
821 record_for_igvn(region);
822 set_control(_gvn.transform(region));
823 set_result( _gvn.transform(value));
824 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
825 }
826
827 //------------------------------generate_guard---------------------------
828 // Helper function for generating guarded fast-slow graph structures.
829 // The given 'test', if true, guards a slow path. If the test fails
830 // then a fast path can be taken. (We generally hope it fails.)
831 // In all cases, GraphKit::control() is updated to the fast path.
832 // The returned value represents the control for the slow path.
833 // The return value is never 'top'; it is either a valid control
834 // or null if it is obvious that the slow path can never be taken.
835 // Also, if region and the slow control are not null, the slow edge
836 // is appended to the region.
837 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
838 if (stopped()) {
839 // Already short circuited.
840 return nullptr;
841 }
842
843 // Build an if node and its projections.
844 // If test is true we take the slow path, which we assume is uncommon.
845 if (_gvn.type(test) == TypeInt::ZERO) {
846 // The slow branch is never taken. No need to build this guard.
847 return nullptr;
848 }
849
850 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
851
852 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
853 if (if_slow == top()) {
854 // The slow branch is never taken. No need to build this guard.
855 return nullptr;
856 }
857
858 if (region != nullptr)
859 region->add_req(if_slow);
860
861 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
862 set_control(if_fast);
863
864 return if_slow;
865 }
866
867 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
868 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
869 }
870 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
871 return generate_guard(test, region, PROB_FAIR);
872 }
873
874 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
875 Node* *pos_index) {
876 if (stopped())
877 return nullptr; // already stopped
878 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
879 return nullptr; // index is already adequately typed
880 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
881 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
882 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
883 if (is_neg != nullptr && pos_index != nullptr) {
884 // Emulate effect of Parse::adjust_map_after_if.
885 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
886 (*pos_index) = _gvn.transform(ccast);
887 }
888 return is_neg;
889 }
890
891 // Make sure that 'position' is a valid limit index, in [0..length].
892 // There are two equivalent plans for checking this:
893 // A. (offset + copyLength) unsigned<= arrayLength
894 // B. offset <= (arrayLength - copyLength)
895 // We require that all of the values above, except for the sum and
896 // difference, are already known to be non-negative.
897 // Plan A is robust in the face of overflow, if offset and copyLength
898 // are both hugely positive.
899 //
900 // Plan B is less direct and intuitive, but it does not overflow at
901 // all, since the difference of two non-negatives is always
902 // representable. Whenever Java methods must perform the equivalent
903 // check they generally use Plan B instead of Plan A.
904 // For the moment we use Plan A.
905 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
906 Node* subseq_length,
907 Node* array_length,
908 RegionNode* region) {
909 if (stopped())
910 return nullptr; // already stopped
911 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
912 if (zero_offset && subseq_length->eqv_uncast(array_length))
913 return nullptr; // common case of whole-array copy
914 Node* last = subseq_length;
915 if (!zero_offset) // last += offset
916 last = _gvn.transform(new AddINode(last, offset));
917 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
918 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
919 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
920 return is_over;
921 }
922
923 // Emit range checks for the given String.value byte array
924 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
925 if (stopped()) {
926 return; // already stopped
927 }
928 RegionNode* bailout = new RegionNode(1);
929 record_for_igvn(bailout);
930 if (char_count) {
931 // Convert char count to byte count
932 count = _gvn.transform(new LShiftINode(count, intcon(1)));
933 }
934
935 // Offset and count must not be negative
936 generate_negative_guard(offset, bailout);
937 generate_negative_guard(count, bailout);
938 // Offset + count must not exceed length of array
939 generate_limit_guard(offset, count, load_array_length(array), bailout);
940
941 if (bailout->req() > 1) {
942 PreserveJVMState pjvms(this);
943 set_control(_gvn.transform(bailout));
944 uncommon_trap(Deoptimization::Reason_intrinsic,
945 Deoptimization::Action_maybe_recompile);
946 }
947 }
948
949 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
950 bool is_immutable) {
951 ciKlass* thread_klass = env()->Thread_klass();
952 const Type* thread_type
953 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
954
955 Node* thread = _gvn.transform(new ThreadLocalNode());
956 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
957 tls_output = thread;
958
959 Node* thread_obj_handle
960 = (is_immutable
961 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
962 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
963 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
964 thread_obj_handle = _gvn.transform(thread_obj_handle);
965
966 DecoratorSet decorators = IN_NATIVE;
967 if (is_immutable) {
968 decorators |= C2_IMMUTABLE_MEMORY;
969 }
970 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
971 }
972
973 //--------------------------generate_current_thread--------------------
974 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
975 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
976 /*is_immutable*/false);
977 }
978
979 //--------------------------generate_virtual_thread--------------------
980 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
981 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
982 !C->method()->changes_current_thread());
983 }
984
985 //------------------------------make_string_method_node------------------------
986 // Helper method for String intrinsic functions. This version is called with
987 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
988 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
989 // containing the lengths of str1 and str2.
990 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
991 Node* result = nullptr;
992 switch (opcode) {
993 case Op_StrIndexOf:
994 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
995 str1_start, cnt1, str2_start, cnt2, ae);
996 break;
997 case Op_StrComp:
998 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
999 str1_start, cnt1, str2_start, cnt2, ae);
1000 break;
1001 case Op_StrEquals:
1002 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1003 // Use the constant length if there is one because optimized match rule may exist.
1004 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1005 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1006 break;
1007 default:
1008 ShouldNotReachHere();
1009 return nullptr;
1010 }
1011
1012 // All these intrinsics have checks.
1013 C->set_has_split_ifs(true); // Has chance for split-if optimization
1014 clear_upper_avx();
1015
1016 return _gvn.transform(result);
1017 }
1018
1019 //------------------------------inline_string_compareTo------------------------
1020 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1021 Node* arg1 = argument(0);
1022 Node* arg2 = argument(1);
1023
1024 arg1 = must_be_not_null(arg1, true);
1025 arg2 = must_be_not_null(arg2, true);
1026
1027 // Get start addr and length of first argument
1028 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1029 Node* arg1_cnt = load_array_length(arg1);
1030
1031 // Get start addr and length of second argument
1032 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1033 Node* arg2_cnt = load_array_length(arg2);
1034
1035 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1036 set_result(result);
1037 return true;
1038 }
1039
1040 //------------------------------inline_string_equals------------------------
1041 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1042 Node* arg1 = argument(0);
1043 Node* arg2 = argument(1);
1044
1045 // paths (plus control) merge
1046 RegionNode* region = new RegionNode(3);
1047 Node* phi = new PhiNode(region, TypeInt::BOOL);
1048
1049 if (!stopped()) {
1050
1051 arg1 = must_be_not_null(arg1, true);
1052 arg2 = must_be_not_null(arg2, true);
1053
1054 // Get start addr and length of first argument
1055 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1056 Node* arg1_cnt = load_array_length(arg1);
1057
1058 // Get start addr and length of second argument
1059 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1060 Node* arg2_cnt = load_array_length(arg2);
1061
1062 // Check for arg1_cnt != arg2_cnt
1063 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1064 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1065 Node* if_ne = generate_slow_guard(bol, nullptr);
1066 if (if_ne != nullptr) {
1067 phi->init_req(2, intcon(0));
1068 region->init_req(2, if_ne);
1069 }
1070
1071 // Check for count == 0 is done by assembler code for StrEquals.
1072
1073 if (!stopped()) {
1074 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1075 phi->init_req(1, equals);
1076 region->init_req(1, control());
1077 }
1078 }
1079
1080 // post merge
1081 set_control(_gvn.transform(region));
1082 record_for_igvn(region);
1083
1084 set_result(_gvn.transform(phi));
1085 return true;
1086 }
1087
1088 //------------------------------inline_array_equals----------------------------
1089 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1090 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1091 Node* arg1 = argument(0);
1092 Node* arg2 = argument(1);
1093
1094 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1095 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1096 clear_upper_avx();
1097
1098 return true;
1099 }
1100
1101
1102 //------------------------------inline_countPositives------------------------------
1103 bool LibraryCallKit::inline_countPositives() {
1104 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1105 return false;
1106 }
1107
1108 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1109 // no receiver since it is static method
1110 Node* ba = argument(0);
1111 Node* offset = argument(1);
1112 Node* len = argument(2);
1113
1114 ba = must_be_not_null(ba, true);
1115
1116 // Range checks
1117 generate_string_range_check(ba, offset, len, false);
1118 if (stopped()) {
1119 return true;
1120 }
1121 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1122 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1123 set_result(_gvn.transform(result));
1124 clear_upper_avx();
1125 return true;
1126 }
1127
1128 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1129 Node* index = argument(0);
1130 Node* length = bt == T_INT ? argument(1) : argument(2);
1131 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1132 return false;
1133 }
1134
1135 // check that length is positive
1136 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1137 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1138
1139 {
1140 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1141 uncommon_trap(Deoptimization::Reason_intrinsic,
1142 Deoptimization::Action_make_not_entrant);
1143 }
1144
1145 if (stopped()) {
1146 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1147 return true;
1148 }
1149
1150 // length is now known positive, add a cast node to make this explicit
1151 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1152 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1153 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1154 ConstraintCastNode::RegularDependency, bt);
1155 casted_length = _gvn.transform(casted_length);
1156 replace_in_map(length, casted_length);
1157 length = casted_length;
1158
1159 // Use an unsigned comparison for the range check itself
1160 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1161 BoolTest::mask btest = BoolTest::lt;
1162 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1163 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1164 _gvn.set_type(rc, rc->Value(&_gvn));
1165 if (!rc_bool->is_Con()) {
1166 record_for_igvn(rc);
1167 }
1168 set_control(_gvn.transform(new IfTrueNode(rc)));
1169 {
1170 PreserveJVMState pjvms(this);
1171 set_control(_gvn.transform(new IfFalseNode(rc)));
1172 uncommon_trap(Deoptimization::Reason_range_check,
1173 Deoptimization::Action_make_not_entrant);
1174 }
1175
1176 if (stopped()) {
1177 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1178 return true;
1179 }
1180
1181 // index is now known to be >= 0 and < length, cast it
1182 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1183 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1184 ConstraintCastNode::RegularDependency, bt);
1185 result = _gvn.transform(result);
1186 set_result(result);
1187 replace_in_map(index, result);
1188 return true;
1189 }
1190
1191 //------------------------------inline_string_indexOf------------------------
1192 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1193 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1194 return false;
1195 }
1196 Node* src = argument(0);
1197 Node* tgt = argument(1);
1198
1199 // Make the merge point
1200 RegionNode* result_rgn = new RegionNode(4);
1201 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1202
1203 src = must_be_not_null(src, true);
1204 tgt = must_be_not_null(tgt, true);
1205
1206 // Get start addr and length of source string
1207 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1208 Node* src_count = load_array_length(src);
1209
1210 // Get start addr and length of substring
1211 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1212 Node* tgt_count = load_array_length(tgt);
1213
1214 Node* result = nullptr;
1215 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1216
1217 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1218 // Divide src size by 2 if String is UTF16 encoded
1219 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1220 }
1221 if (ae == StrIntrinsicNode::UU) {
1222 // Divide substring size by 2 if String is UTF16 encoded
1223 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1224 }
1225
1226 if (call_opt_stub) {
1227 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1228 StubRoutines::_string_indexof_array[ae],
1229 "stringIndexOf", TypePtr::BOTTOM, src_start,
1230 src_count, tgt_start, tgt_count);
1231 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1232 } else {
1233 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1234 result_rgn, result_phi, ae);
1235 }
1236 if (result != nullptr) {
1237 result_phi->init_req(3, result);
1238 result_rgn->init_req(3, control());
1239 }
1240 set_control(_gvn.transform(result_rgn));
1241 record_for_igvn(result_rgn);
1242 set_result(_gvn.transform(result_phi));
1243
1244 return true;
1245 }
1246
1247 //-----------------------------inline_string_indexOfI-----------------------
1248 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1249 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1250 return false;
1251 }
1252 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1253 return false;
1254 }
1255
1256 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1257 Node* src = argument(0); // byte[]
1258 Node* src_count = argument(1); // char count
1259 Node* tgt = argument(2); // byte[]
1260 Node* tgt_count = argument(3); // char count
1261 Node* from_index = argument(4); // char index
1262
1263 src = must_be_not_null(src, true);
1264 tgt = must_be_not_null(tgt, true);
1265
1266 // Multiply byte array index by 2 if String is UTF16 encoded
1267 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1268 src_count = _gvn.transform(new SubINode(src_count, from_index));
1269 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1270 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1271
1272 // Range checks
1273 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1274 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1275 if (stopped()) {
1276 return true;
1277 }
1278
1279 RegionNode* region = new RegionNode(5);
1280 Node* phi = new PhiNode(region, TypeInt::INT);
1281 Node* result = nullptr;
1282
1283 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1284
1285 if (call_opt_stub) {
1286 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1287 StubRoutines::_string_indexof_array[ae],
1288 "stringIndexOf", TypePtr::BOTTOM, src_start,
1289 src_count, tgt_start, tgt_count);
1290 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1291 } else {
1292 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1293 region, phi, ae);
1294 }
1295 if (result != nullptr) {
1296 // The result is index relative to from_index if substring was found, -1 otherwise.
1297 // Generate code which will fold into cmove.
1298 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1299 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1300
1301 Node* if_lt = generate_slow_guard(bol, nullptr);
1302 if (if_lt != nullptr) {
1303 // result == -1
1304 phi->init_req(3, result);
1305 region->init_req(3, if_lt);
1306 }
1307 if (!stopped()) {
1308 result = _gvn.transform(new AddINode(result, from_index));
1309 phi->init_req(4, result);
1310 region->init_req(4, control());
1311 }
1312 }
1313
1314 set_control(_gvn.transform(region));
1315 record_for_igvn(region);
1316 set_result(_gvn.transform(phi));
1317 clear_upper_avx();
1318
1319 return true;
1320 }
1321
1322 // Create StrIndexOfNode with fast path checks
1323 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1324 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1325 // Check for substr count > string count
1326 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1327 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1328 Node* if_gt = generate_slow_guard(bol, nullptr);
1329 if (if_gt != nullptr) {
1330 phi->init_req(1, intcon(-1));
1331 region->init_req(1, if_gt);
1332 }
1333 if (!stopped()) {
1334 // Check for substr count == 0
1335 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1336 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1337 Node* if_zero = generate_slow_guard(bol, nullptr);
1338 if (if_zero != nullptr) {
1339 phi->init_req(2, intcon(0));
1340 region->init_req(2, if_zero);
1341 }
1342 }
1343 if (!stopped()) {
1344 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1345 }
1346 return nullptr;
1347 }
1348
1349 //-----------------------------inline_string_indexOfChar-----------------------
1350 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1351 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1352 return false;
1353 }
1354 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1355 return false;
1356 }
1357 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1358 Node* src = argument(0); // byte[]
1359 Node* int_ch = argument(1);
1360 Node* from_index = argument(2);
1361 Node* max = argument(3);
1362
1363 src = must_be_not_null(src, true);
1364
1365 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1366 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1367 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1368
1369 // Range checks
1370 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1371
1372 // Check for int_ch >= 0
1373 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1374 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1375 {
1376 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1377 uncommon_trap(Deoptimization::Reason_intrinsic,
1378 Deoptimization::Action_maybe_recompile);
1379 }
1380 if (stopped()) {
1381 return true;
1382 }
1383
1384 RegionNode* region = new RegionNode(3);
1385 Node* phi = new PhiNode(region, TypeInt::INT);
1386
1387 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1388 C->set_has_split_ifs(true); // Has chance for split-if optimization
1389 _gvn.transform(result);
1390
1391 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1392 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1393
1394 Node* if_lt = generate_slow_guard(bol, nullptr);
1395 if (if_lt != nullptr) {
1396 // result == -1
1397 phi->init_req(2, result);
1398 region->init_req(2, if_lt);
1399 }
1400 if (!stopped()) {
1401 result = _gvn.transform(new AddINode(result, from_index));
1402 phi->init_req(1, result);
1403 region->init_req(1, control());
1404 }
1405 set_control(_gvn.transform(region));
1406 record_for_igvn(region);
1407 set_result(_gvn.transform(phi));
1408 clear_upper_avx();
1409
1410 return true;
1411 }
1412 //---------------------------inline_string_copy---------------------
1413 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1414 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1415 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1416 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1417 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1418 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1419 bool LibraryCallKit::inline_string_copy(bool compress) {
1420 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1421 return false;
1422 }
1423 int nargs = 5; // 2 oops, 3 ints
1424 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1425
1426 Node* src = argument(0);
1427 Node* src_offset = argument(1);
1428 Node* dst = argument(2);
1429 Node* dst_offset = argument(3);
1430 Node* length = argument(4);
1431
1432 // Check for allocation before we add nodes that would confuse
1433 // tightly_coupled_allocation()
1434 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1435
1436 // Figure out the size and type of the elements we will be copying.
1437 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1438 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1439 if (src_type == nullptr || dst_type == nullptr) {
1440 return false;
1441 }
1442 BasicType src_elem = src_type->elem()->array_element_basic_type();
1443 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1444 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1445 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1446 "Unsupported array types for inline_string_copy");
1447
1448 src = must_be_not_null(src, true);
1449 dst = must_be_not_null(dst, true);
1450
1451 // Convert char[] offsets to byte[] offsets
1452 bool convert_src = (compress && src_elem == T_BYTE);
1453 bool convert_dst = (!compress && dst_elem == T_BYTE);
1454 if (convert_src) {
1455 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1456 } else if (convert_dst) {
1457 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1458 }
1459
1460 // Range checks
1461 generate_string_range_check(src, src_offset, length, convert_src);
1462 generate_string_range_check(dst, dst_offset, length, convert_dst);
1463 if (stopped()) {
1464 return true;
1465 }
1466
1467 Node* src_start = array_element_address(src, src_offset, src_elem);
1468 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1469 // 'src_start' points to src array + scaled offset
1470 // 'dst_start' points to dst array + scaled offset
1471 Node* count = nullptr;
1472 if (compress) {
1473 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1474 } else {
1475 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1476 }
1477
1478 if (alloc != nullptr) {
1479 if (alloc->maybe_set_complete(&_gvn)) {
1480 // "You break it, you buy it."
1481 InitializeNode* init = alloc->initialization();
1482 assert(init->is_complete(), "we just did this");
1483 init->set_complete_with_arraycopy();
1484 assert(dst->is_CheckCastPP(), "sanity");
1485 assert(dst->in(0)->in(0) == init, "dest pinned");
1486 }
1487 // Do not let stores that initialize this object be reordered with
1488 // a subsequent store that would make this object accessible by
1489 // other threads.
1490 // Record what AllocateNode this StoreStore protects so that
1491 // escape analysis can go from the MemBarStoreStoreNode to the
1492 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1493 // based on the escape status of the AllocateNode.
1494 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1495 }
1496 if (compress) {
1497 set_result(_gvn.transform(count));
1498 }
1499 clear_upper_avx();
1500
1501 return true;
1502 }
1503
1504 #ifdef _LP64
1505 #define XTOP ,top() /*additional argument*/
1506 #else //_LP64
1507 #define XTOP /*no additional argument*/
1508 #endif //_LP64
1509
1510 //------------------------inline_string_toBytesU--------------------------
1511 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1512 bool LibraryCallKit::inline_string_toBytesU() {
1513 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1514 return false;
1515 }
1516 // Get the arguments.
1517 Node* value = argument(0);
1518 Node* offset = argument(1);
1519 Node* length = argument(2);
1520
1521 Node* newcopy = nullptr;
1522
1523 // Set the original stack and the reexecute bit for the interpreter to reexecute
1524 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1525 { PreserveReexecuteState preexecs(this);
1526 jvms()->set_should_reexecute(true);
1527
1528 // Check if a null path was taken unconditionally.
1529 value = null_check(value);
1530
1531 RegionNode* bailout = new RegionNode(1);
1532 record_for_igvn(bailout);
1533
1534 // Range checks
1535 generate_negative_guard(offset, bailout);
1536 generate_negative_guard(length, bailout);
1537 generate_limit_guard(offset, length, load_array_length(value), bailout);
1538 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1539 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1540
1541 if (bailout->req() > 1) {
1542 PreserveJVMState pjvms(this);
1543 set_control(_gvn.transform(bailout));
1544 uncommon_trap(Deoptimization::Reason_intrinsic,
1545 Deoptimization::Action_maybe_recompile);
1546 }
1547 if (stopped()) {
1548 return true;
1549 }
1550
1551 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1552 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1553 newcopy = new_array(klass_node, size, 0); // no arguments to push
1554 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1555 guarantee(alloc != nullptr, "created above");
1556
1557 // Calculate starting addresses.
1558 Node* src_start = array_element_address(value, offset, T_CHAR);
1559 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1560
1561 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1562 const TypeInt* toffset = gvn().type(offset)->is_int();
1563 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1564
1565 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1566 const char* copyfunc_name = "arraycopy";
1567 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1568 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1569 OptoRuntime::fast_arraycopy_Type(),
1570 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1571 src_start, dst_start, ConvI2X(length) XTOP);
1572 // Do not let reads from the cloned object float above the arraycopy.
1573 if (alloc->maybe_set_complete(&_gvn)) {
1574 // "You break it, you buy it."
1575 InitializeNode* init = alloc->initialization();
1576 assert(init->is_complete(), "we just did this");
1577 init->set_complete_with_arraycopy();
1578 assert(newcopy->is_CheckCastPP(), "sanity");
1579 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1580 }
1581 // Do not let stores that initialize this object be reordered with
1582 // a subsequent store that would make this object accessible by
1583 // other threads.
1584 // Record what AllocateNode this StoreStore protects so that
1585 // escape analysis can go from the MemBarStoreStoreNode to the
1586 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1587 // based on the escape status of the AllocateNode.
1588 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1589 } // original reexecute is set back here
1590
1591 C->set_has_split_ifs(true); // Has chance for split-if optimization
1592 if (!stopped()) {
1593 set_result(newcopy);
1594 }
1595 clear_upper_avx();
1596
1597 return true;
1598 }
1599
1600 //------------------------inline_string_getCharsU--------------------------
1601 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1602 bool LibraryCallKit::inline_string_getCharsU() {
1603 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1604 return false;
1605 }
1606
1607 // Get the arguments.
1608 Node* src = argument(0);
1609 Node* src_begin = argument(1);
1610 Node* src_end = argument(2); // exclusive offset (i < src_end)
1611 Node* dst = argument(3);
1612 Node* dst_begin = argument(4);
1613
1614 // Check for allocation before we add nodes that would confuse
1615 // tightly_coupled_allocation()
1616 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1617
1618 // Check if a null path was taken unconditionally.
1619 src = null_check(src);
1620 dst = null_check(dst);
1621 if (stopped()) {
1622 return true;
1623 }
1624
1625 // Get length and convert char[] offset to byte[] offset
1626 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1627 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1628
1629 // Range checks
1630 generate_string_range_check(src, src_begin, length, true);
1631 generate_string_range_check(dst, dst_begin, length, false);
1632 if (stopped()) {
1633 return true;
1634 }
1635
1636 if (!stopped()) {
1637 // Calculate starting addresses.
1638 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1639 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1640
1641 // Check if array addresses are aligned to HeapWordSize
1642 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1643 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1644 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1645 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1646
1647 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1648 const char* copyfunc_name = "arraycopy";
1649 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1650 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1651 OptoRuntime::fast_arraycopy_Type(),
1652 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1653 src_start, dst_start, ConvI2X(length) XTOP);
1654 // Do not let reads from the cloned object float above the arraycopy.
1655 if (alloc != nullptr) {
1656 if (alloc->maybe_set_complete(&_gvn)) {
1657 // "You break it, you buy it."
1658 InitializeNode* init = alloc->initialization();
1659 assert(init->is_complete(), "we just did this");
1660 init->set_complete_with_arraycopy();
1661 assert(dst->is_CheckCastPP(), "sanity");
1662 assert(dst->in(0)->in(0) == init, "dest pinned");
1663 }
1664 // Do not let stores that initialize this object be reordered with
1665 // a subsequent store that would make this object accessible by
1666 // other threads.
1667 // Record what AllocateNode this StoreStore protects so that
1668 // escape analysis can go from the MemBarStoreStoreNode to the
1669 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1670 // based on the escape status of the AllocateNode.
1671 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1672 } else {
1673 insert_mem_bar(Op_MemBarCPUOrder);
1674 }
1675 }
1676
1677 C->set_has_split_ifs(true); // Has chance for split-if optimization
1678 return true;
1679 }
1680
1681 //----------------------inline_string_char_access----------------------------
1682 // Store/Load char to/from byte[] array.
1683 // static void StringUTF16.putChar(byte[] val, int index, int c)
1684 // static char StringUTF16.getChar(byte[] val, int index)
1685 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1686 Node* value = argument(0);
1687 Node* index = argument(1);
1688 Node* ch = is_store ? argument(2) : nullptr;
1689
1690 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1691 // correctly requires matched array shapes.
1692 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1693 "sanity: byte[] and char[] bases agree");
1694 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1695 "sanity: byte[] and char[] scales agree");
1696
1697 // Bail when getChar over constants is requested: constant folding would
1698 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1699 // Java method would constant fold nicely instead.
1700 if (!is_store && value->is_Con() && index->is_Con()) {
1701 return false;
1702 }
1703
1704 // Save state and restore on bailout
1705 uint old_sp = sp();
1706 SafePointNode* old_map = clone_map();
1707
1708 value = must_be_not_null(value, true);
1709
1710 Node* adr = array_element_address(value, index, T_CHAR);
1711 if (adr->is_top()) {
1712 set_map(old_map);
1713 set_sp(old_sp);
1714 return false;
1715 }
1716 destruct_map_clone(old_map);
1717 if (is_store) {
1718 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1719 } else {
1720 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);
1721 set_result(ch);
1722 }
1723 return true;
1724 }
1725
1726 //--------------------------round_double_node--------------------------------
1727 // Round a double node if necessary.
1728 Node* LibraryCallKit::round_double_node(Node* n) {
1729 if (Matcher::strict_fp_requires_explicit_rounding) {
1730 #ifdef IA32
1731 if (UseSSE < 2) {
1732 n = _gvn.transform(new RoundDoubleNode(nullptr, n));
1733 }
1734 #else
1735 Unimplemented();
1736 #endif // IA32
1737 }
1738 return n;
1739 }
1740
1741 //------------------------------inline_math-----------------------------------
1742 // public static double Math.abs(double)
1743 // public static double Math.sqrt(double)
1744 // public static double Math.log(double)
1745 // public static double Math.log10(double)
1746 // public static double Math.round(double)
1747 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1748 Node* arg = round_double_node(argument(0));
1749 Node* n = nullptr;
1750 switch (id) {
1751 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1752 case vmIntrinsics::_dsqrt:
1753 case vmIntrinsics::_dsqrt_strict:
1754 n = new SqrtDNode(C, control(), arg); break;
1755 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1756 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1757 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1758 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1759 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1760 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1761 default: fatal_unexpected_iid(id); break;
1762 }
1763 set_result(_gvn.transform(n));
1764 return true;
1765 }
1766
1767 //------------------------------inline_math-----------------------------------
1768 // public static float Math.abs(float)
1769 // public static int Math.abs(int)
1770 // public static long Math.abs(long)
1771 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1772 Node* arg = argument(0);
1773 Node* n = nullptr;
1774 switch (id) {
1775 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1776 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1777 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1778 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1779 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1780 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1781 default: fatal_unexpected_iid(id); break;
1782 }
1783 set_result(_gvn.transform(n));
1784 return true;
1785 }
1786
1787 //------------------------------runtime_math-----------------------------
1788 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1789 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1790 "must be (DD)D or (D)D type");
1791
1792 // Inputs
1793 Node* a = round_double_node(argument(0));
1794 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : nullptr;
1795
1796 const TypePtr* no_memory_effects = nullptr;
1797 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1798 no_memory_effects,
1799 a, top(), b, b ? top() : nullptr);
1800 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1801 #ifdef ASSERT
1802 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1803 assert(value_top == top(), "second value must be top");
1804 #endif
1805
1806 set_result(value);
1807 return true;
1808 }
1809
1810 //------------------------------inline_math_pow-----------------------------
1811 bool LibraryCallKit::inline_math_pow() {
1812 Node* exp = round_double_node(argument(2));
1813 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1814 if (d != nullptr) {
1815 if (d->getd() == 2.0) {
1816 // Special case: pow(x, 2.0) => x * x
1817 Node* base = round_double_node(argument(0));
1818 set_result(_gvn.transform(new MulDNode(base, base)));
1819 return true;
1820 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1821 // Special case: pow(x, 0.5) => sqrt(x)
1822 Node* base = round_double_node(argument(0));
1823 Node* zero = _gvn.zerocon(T_DOUBLE);
1824
1825 RegionNode* region = new RegionNode(3);
1826 Node* phi = new PhiNode(region, Type::DOUBLE);
1827
1828 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1829 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1830 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1831 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1832 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1833
1834 Node* if_pow = generate_slow_guard(test, nullptr);
1835 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1836 phi->init_req(1, value_sqrt);
1837 region->init_req(1, control());
1838
1839 if (if_pow != nullptr) {
1840 set_control(if_pow);
1841 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1842 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1843 const TypePtr* no_memory_effects = nullptr;
1844 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1845 no_memory_effects, base, top(), exp, top());
1846 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1847 #ifdef ASSERT
1848 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1849 assert(value_top == top(), "second value must be top");
1850 #endif
1851 phi->init_req(2, value_pow);
1852 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1853 }
1854
1855 C->set_has_split_ifs(true); // Has chance for split-if optimization
1856 set_control(_gvn.transform(region));
1857 record_for_igvn(region);
1858 set_result(_gvn.transform(phi));
1859
1860 return true;
1861 }
1862 }
1863
1864 return StubRoutines::dpow() != nullptr ?
1865 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1866 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1867 }
1868
1869 //------------------------------inline_math_native-----------------------------
1870 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1871 switch (id) {
1872 case vmIntrinsics::_dsin:
1873 return StubRoutines::dsin() != nullptr ?
1874 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1875 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1876 case vmIntrinsics::_dcos:
1877 return StubRoutines::dcos() != nullptr ?
1878 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1879 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1880 case vmIntrinsics::_dtan:
1881 return StubRoutines::dtan() != nullptr ?
1882 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1883 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1884 case vmIntrinsics::_dtanh:
1885 return StubRoutines::dtanh() != nullptr ?
1886 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1887 case vmIntrinsics::_dexp:
1888 return StubRoutines::dexp() != nullptr ?
1889 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1890 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1891 case vmIntrinsics::_dlog:
1892 return StubRoutines::dlog() != nullptr ?
1893 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1894 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1895 case vmIntrinsics::_dlog10:
1896 return StubRoutines::dlog10() != nullptr ?
1897 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1898 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1899
1900 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1901 case vmIntrinsics::_ceil:
1902 case vmIntrinsics::_floor:
1903 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1904
1905 case vmIntrinsics::_dsqrt:
1906 case vmIntrinsics::_dsqrt_strict:
1907 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1908 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1909 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1910 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1911 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1912
1913 case vmIntrinsics::_dpow: return inline_math_pow();
1914 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1915 case vmIntrinsics::_fcopySign: return inline_math(id);
1916 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1917 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1918 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1919
1920 // These intrinsics are not yet correctly implemented
1921 case vmIntrinsics::_datan2:
1922 return false;
1923
1924 default:
1925 fatal_unexpected_iid(id);
1926 return false;
1927 }
1928 }
1929
1930 //----------------------------inline_notify-----------------------------------*
1931 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1932 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1933 address func;
1934 if (id == vmIntrinsics::_notify) {
1935 func = OptoRuntime::monitor_notify_Java();
1936 } else {
1937 func = OptoRuntime::monitor_notifyAll_Java();
1938 }
1939 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1940 make_slow_call_ex(call, env()->Throwable_klass(), false);
1941 return true;
1942 }
1943
1944
1945 //----------------------------inline_min_max-----------------------------------
1946 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1947 set_result(generate_min_max(id, argument(0), argument(1)));
1948 return true;
1949 }
1950
1951 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1952 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1953 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1954 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1955 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1956
1957 {
1958 PreserveJVMState pjvms(this);
1959 PreserveReexecuteState preexecs(this);
1960 jvms()->set_should_reexecute(true);
1961
1962 set_control(slow_path);
1963 set_i_o(i_o());
1964
1965 uncommon_trap(Deoptimization::Reason_intrinsic,
1966 Deoptimization::Action_none);
1967 }
1968
1969 set_control(fast_path);
1970 set_result(math);
1971 }
1972
1973 template <typename OverflowOp>
1974 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1975 typedef typename OverflowOp::MathOp MathOp;
1976
1977 MathOp* mathOp = new MathOp(arg1, arg2);
1978 Node* operation = _gvn.transform( mathOp );
1979 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1980 inline_math_mathExact(operation, ofcheck);
1981 return true;
1982 }
1983
1984 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1985 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1986 }
1987
1988 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1989 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1990 }
1991
1992 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1993 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1994 }
1995
1996 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1997 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1998 }
1999
2000 bool LibraryCallKit::inline_math_negateExactI() {
2001 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2002 }
2003
2004 bool LibraryCallKit::inline_math_negateExactL() {
2005 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2006 }
2007
2008 bool LibraryCallKit::inline_math_multiplyExactI() {
2009 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2010 }
2011
2012 bool LibraryCallKit::inline_math_multiplyExactL() {
2013 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2014 }
2015
2016 bool LibraryCallKit::inline_math_multiplyHigh() {
2017 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2018 return true;
2019 }
2020
2021 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2022 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2023 return true;
2024 }
2025
2026 Node*
2027 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
2028 Node* result_val = nullptr;
2029 switch (id) {
2030 case vmIntrinsics::_min:
2031 case vmIntrinsics::_min_strict:
2032 result_val = _gvn.transform(new MinINode(x0, y0));
2033 break;
2034 case vmIntrinsics::_max:
2035 case vmIntrinsics::_max_strict:
2036 result_val = _gvn.transform(new MaxINode(x0, y0));
2037 break;
2038 default:
2039 fatal_unexpected_iid(id);
2040 break;
2041 }
2042 return result_val;
2043 }
2044
2045 inline int
2046 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2047 const TypePtr* base_type = TypePtr::NULL_PTR;
2048 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2049 if (base_type == nullptr) {
2050 // Unknown type.
2051 return Type::AnyPtr;
2052 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2053 // Since this is a null+long form, we have to switch to a rawptr.
2054 base = _gvn.transform(new CastX2PNode(offset));
2055 offset = MakeConX(0);
2056 return Type::RawPtr;
2057 } else if (base_type->base() == Type::RawPtr) {
2058 return Type::RawPtr;
2059 } else if (base_type->isa_oopptr()) {
2060 // Base is never null => always a heap address.
2061 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2062 return Type::OopPtr;
2063 }
2064 // Offset is small => always a heap address.
2065 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2066 if (offset_type != nullptr &&
2067 base_type->offset() == 0 && // (should always be?)
2068 offset_type->_lo >= 0 &&
2069 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2070 return Type::OopPtr;
2071 } else if (type == T_OBJECT) {
2072 // off heap access to an oop doesn't make any sense. Has to be on
2073 // heap.
2074 return Type::OopPtr;
2075 }
2076 // Otherwise, it might either be oop+off or null+addr.
2077 return Type::AnyPtr;
2078 } else {
2079 // No information:
2080 return Type::AnyPtr;
2081 }
2082 }
2083
2084 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2085 Node* uncasted_base = base;
2086 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2087 if (kind == Type::RawPtr) {
2088 return basic_plus_adr(top(), uncasted_base, offset);
2089 } else if (kind == Type::AnyPtr) {
2090 assert(base == uncasted_base, "unexpected base change");
2091 if (can_cast) {
2092 if (!_gvn.type(base)->speculative_maybe_null() &&
2093 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2094 // According to profiling, this access is always on
2095 // heap. Casting the base to not null and thus avoiding membars
2096 // around the access should allow better optimizations
2097 Node* null_ctl = top();
2098 base = null_check_oop(base, &null_ctl, true, true, true);
2099 assert(null_ctl->is_top(), "no null control here");
2100 return basic_plus_adr(base, offset);
2101 } else if (_gvn.type(base)->speculative_always_null() &&
2102 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2103 // According to profiling, this access is always off
2104 // heap.
2105 base = null_assert(base);
2106 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2107 offset = MakeConX(0);
2108 return basic_plus_adr(top(), raw_base, offset);
2109 }
2110 }
2111 // We don't know if it's an on heap or off heap access. Fall back
2112 // to raw memory access.
2113 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2114 return basic_plus_adr(top(), raw, offset);
2115 } else {
2116 assert(base == uncasted_base, "unexpected base change");
2117 // We know it's an on heap access so base can't be null
2118 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2119 base = must_be_not_null(base, true);
2120 }
2121 return basic_plus_adr(base, offset);
2122 }
2123 }
2124
2125 //--------------------------inline_number_methods-----------------------------
2126 // inline int Integer.numberOfLeadingZeros(int)
2127 // inline int Long.numberOfLeadingZeros(long)
2128 //
2129 // inline int Integer.numberOfTrailingZeros(int)
2130 // inline int Long.numberOfTrailingZeros(long)
2131 //
2132 // inline int Integer.bitCount(int)
2133 // inline int Long.bitCount(long)
2134 //
2135 // inline char Character.reverseBytes(char)
2136 // inline short Short.reverseBytes(short)
2137 // inline int Integer.reverseBytes(int)
2138 // inline long Long.reverseBytes(long)
2139 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2140 Node* arg = argument(0);
2141 Node* n = nullptr;
2142 switch (id) {
2143 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2144 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2145 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2146 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2147 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2148 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2149 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(nullptr, arg); break;
2150 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( nullptr, arg); break;
2151 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( nullptr, arg); break;
2152 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( nullptr, arg); break;
2153 case vmIntrinsics::_reverse_i: n = new ReverseINode(nullptr, arg); break;
2154 case vmIntrinsics::_reverse_l: n = new ReverseLNode(nullptr, arg); break;
2155 default: fatal_unexpected_iid(id); break;
2156 }
2157 set_result(_gvn.transform(n));
2158 return true;
2159 }
2160
2161 //--------------------------inline_bitshuffle_methods-----------------------------
2162 // inline int Integer.compress(int, int)
2163 // inline int Integer.expand(int, int)
2164 // inline long Long.compress(long, long)
2165 // inline long Long.expand(long, long)
2166 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2167 Node* n = nullptr;
2168 switch (id) {
2169 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2170 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2171 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2172 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2173 default: fatal_unexpected_iid(id); break;
2174 }
2175 set_result(_gvn.transform(n));
2176 return true;
2177 }
2178
2179 //--------------------------inline_number_methods-----------------------------
2180 // inline int Integer.compareUnsigned(int, int)
2181 // inline int Long.compareUnsigned(long, long)
2182 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2183 Node* arg1 = argument(0);
2184 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2185 Node* n = nullptr;
2186 switch (id) {
2187 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2188 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2189 default: fatal_unexpected_iid(id); break;
2190 }
2191 set_result(_gvn.transform(n));
2192 return true;
2193 }
2194
2195 //--------------------------inline_unsigned_divmod_methods-----------------------------
2196 // inline int Integer.divideUnsigned(int, int)
2197 // inline int Integer.remainderUnsigned(int, int)
2198 // inline long Long.divideUnsigned(long, long)
2199 // inline long Long.remainderUnsigned(long, long)
2200 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2201 Node* n = nullptr;
2202 switch (id) {
2203 case vmIntrinsics::_divideUnsigned_i: {
2204 zero_check_int(argument(1));
2205 // Compile-time detect of null-exception
2206 if (stopped()) {
2207 return true; // keep the graph constructed so far
2208 }
2209 n = new UDivINode(control(), argument(0), argument(1));
2210 break;
2211 }
2212 case vmIntrinsics::_divideUnsigned_l: {
2213 zero_check_long(argument(2));
2214 // Compile-time detect of null-exception
2215 if (stopped()) {
2216 return true; // keep the graph constructed so far
2217 }
2218 n = new UDivLNode(control(), argument(0), argument(2));
2219 break;
2220 }
2221 case vmIntrinsics::_remainderUnsigned_i: {
2222 zero_check_int(argument(1));
2223 // Compile-time detect of null-exception
2224 if (stopped()) {
2225 return true; // keep the graph constructed so far
2226 }
2227 n = new UModINode(control(), argument(0), argument(1));
2228 break;
2229 }
2230 case vmIntrinsics::_remainderUnsigned_l: {
2231 zero_check_long(argument(2));
2232 // Compile-time detect of null-exception
2233 if (stopped()) {
2234 return true; // keep the graph constructed so far
2235 }
2236 n = new UModLNode(control(), argument(0), argument(2));
2237 break;
2238 }
2239 default: fatal_unexpected_iid(id); break;
2240 }
2241 set_result(_gvn.transform(n));
2242 return true;
2243 }
2244
2245 //----------------------------inline_unsafe_access----------------------------
2246
2247 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2248 // Attempt to infer a sharper value type from the offset and base type.
2249 ciKlass* sharpened_klass = nullptr;
2250
2251 // See if it is an instance field, with an object type.
2252 if (alias_type->field() != nullptr) {
2253 if (alias_type->field()->type()->is_klass()) {
2254 sharpened_klass = alias_type->field()->type()->as_klass();
2255 }
2256 }
2257
2258 const TypeOopPtr* result = nullptr;
2259 // See if it is a narrow oop array.
2260 if (adr_type->isa_aryptr()) {
2261 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2262 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2263 if (elem_type != nullptr && elem_type->is_loaded()) {
2264 // Sharpen the value type.
2265 result = elem_type;
2266 }
2267 }
2268 }
2269
2270 // The sharpened class might be unloaded if there is no class loader
2271 // contraint in place.
2272 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2273 // Sharpen the value type.
2274 result = TypeOopPtr::make_from_klass(sharpened_klass);
2275 }
2276 if (result != nullptr) {
2277 #ifndef PRODUCT
2278 if (C->print_intrinsics() || C->print_inlining()) {
2279 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2280 tty->print(" sharpened value: "); result->dump(); tty->cr();
2281 }
2282 #endif
2283 }
2284 return result;
2285 }
2286
2287 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2288 switch (kind) {
2289 case Relaxed:
2290 return MO_UNORDERED;
2291 case Opaque:
2292 return MO_RELAXED;
2293 case Acquire:
2294 return MO_ACQUIRE;
2295 case Release:
2296 return MO_RELEASE;
2297 case Volatile:
2298 return MO_SEQ_CST;
2299 default:
2300 ShouldNotReachHere();
2301 return 0;
2302 }
2303 }
2304
2305 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2306 if (callee()->is_static()) return false; // caller must have the capability!
2307 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2308 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2309 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2310 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2311
2312 if (is_reference_type(type)) {
2313 decorators |= ON_UNKNOWN_OOP_REF;
2314 }
2315
2316 if (unaligned) {
2317 decorators |= C2_UNALIGNED;
2318 }
2319
2320 #ifndef PRODUCT
2321 {
2322 ResourceMark rm;
2323 // Check the signatures.
2324 ciSignature* sig = callee()->signature();
2325 #ifdef ASSERT
2326 if (!is_store) {
2327 // Object getReference(Object base, int/long offset), etc.
2328 BasicType rtype = sig->return_type()->basic_type();
2329 assert(rtype == type, "getter must return the expected value");
2330 assert(sig->count() == 2, "oop getter has 2 arguments");
2331 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2332 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2333 } else {
2334 // void putReference(Object base, int/long offset, Object x), etc.
2335 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2336 assert(sig->count() == 3, "oop putter has 3 arguments");
2337 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2338 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2339 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2340 assert(vtype == type, "putter must accept the expected value");
2341 }
2342 #endif // ASSERT
2343 }
2344 #endif //PRODUCT
2345
2346 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2347
2348 Node* receiver = argument(0); // type: oop
2349
2350 // Build address expression.
2351 Node* heap_base_oop = top();
2352
2353 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2354 Node* base = argument(1); // type: oop
2355 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2356 Node* offset = argument(2); // type: long
2357 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2358 // to be plain byte offsets, which are also the same as those accepted
2359 // by oopDesc::field_addr.
2360 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2361 "fieldOffset must be byte-scaled");
2362 // 32-bit machines ignore the high half!
2363 offset = ConvL2X(offset);
2364
2365 // Save state and restore on bailout
2366 uint old_sp = sp();
2367 SafePointNode* old_map = clone_map();
2368
2369 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2370 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2371
2372 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2373 if (type != T_OBJECT) {
2374 decorators |= IN_NATIVE; // off-heap primitive access
2375 } else {
2376 set_map(old_map);
2377 set_sp(old_sp);
2378 return false; // off-heap oop accesses are not supported
2379 }
2380 } else {
2381 heap_base_oop = base; // on-heap or mixed access
2382 }
2383
2384 // Can base be null? Otherwise, always on-heap access.
2385 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2386
2387 if (!can_access_non_heap) {
2388 decorators |= IN_HEAP;
2389 }
2390
2391 Node* val = is_store ? argument(4) : nullptr;
2392
2393 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2394 if (adr_type == TypePtr::NULL_PTR) {
2395 set_map(old_map);
2396 set_sp(old_sp);
2397 return false; // off-heap access with zero address
2398 }
2399
2400 // Try to categorize the address.
2401 Compile::AliasType* alias_type = C->alias_type(adr_type);
2402 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2403
2404 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2405 alias_type->adr_type() == TypeAryPtr::RANGE) {
2406 set_map(old_map);
2407 set_sp(old_sp);
2408 return false; // not supported
2409 }
2410
2411 bool mismatched = false;
2412 BasicType bt = alias_type->basic_type();
2413 if (bt != T_ILLEGAL) {
2414 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2415 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2416 // Alias type doesn't differentiate between byte[] and boolean[]).
2417 // Use address type to get the element type.
2418 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2419 }
2420 if (is_reference_type(bt, true)) {
2421 // accessing an array field with getReference is not a mismatch
2422 bt = T_OBJECT;
2423 }
2424 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2425 // Don't intrinsify mismatched object accesses
2426 set_map(old_map);
2427 set_sp(old_sp);
2428 return false;
2429 }
2430 mismatched = (bt != type);
2431 } else if (alias_type->adr_type()->isa_oopptr()) {
2432 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2433 }
2434
2435 destruct_map_clone(old_map);
2436 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2437
2438 if (mismatched) {
2439 decorators |= C2_MISMATCHED;
2440 }
2441
2442 // First guess at the value type.
2443 const Type *value_type = Type::get_const_basic_type(type);
2444
2445 // Figure out the memory ordering.
2446 decorators |= mo_decorator_for_access_kind(kind);
2447
2448 if (!is_store && type == T_OBJECT) {
2449 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2450 if (tjp != nullptr) {
2451 value_type = tjp;
2452 }
2453 }
2454
2455 receiver = null_check(receiver);
2456 if (stopped()) {
2457 return true;
2458 }
2459 // Heap pointers get a null-check from the interpreter,
2460 // as a courtesy. However, this is not guaranteed by Unsafe,
2461 // and it is not possible to fully distinguish unintended nulls
2462 // from intended ones in this API.
2463
2464 if (!is_store) {
2465 Node* p = nullptr;
2466 // Try to constant fold a load from a constant field
2467 ciField* field = alias_type->field();
2468 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) {
2469 // final or stable field
2470 p = make_constant_from_field(field, heap_base_oop);
2471 }
2472
2473 if (p == nullptr) { // Could not constant fold the load
2474 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2475 // Normalize the value returned by getBoolean in the following cases
2476 if (type == T_BOOLEAN &&
2477 (mismatched ||
2478 heap_base_oop == top() || // - heap_base_oop is null or
2479 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2480 // and the unsafe access is made to large offset
2481 // (i.e., larger than the maximum offset necessary for any
2482 // field access)
2483 ) {
2484 IdealKit ideal = IdealKit(this);
2485 #define __ ideal.
2486 IdealVariable normalized_result(ideal);
2487 __ declarations_done();
2488 __ set(normalized_result, p);
2489 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2490 __ set(normalized_result, ideal.ConI(1));
2491 ideal.end_if();
2492 final_sync(ideal);
2493 p = __ value(normalized_result);
2494 #undef __
2495 }
2496 }
2497 if (type == T_ADDRESS) {
2498 p = gvn().transform(new CastP2XNode(nullptr, p));
2499 p = ConvX2UL(p);
2500 }
2501 // The load node has the control of the preceding MemBarCPUOrder. All
2502 // following nodes will have the control of the MemBarCPUOrder inserted at
2503 // the end of this method. So, pushing the load onto the stack at a later
2504 // point is fine.
2505 set_result(p);
2506 } else {
2507 if (bt == T_ADDRESS) {
2508 // Repackage the long as a pointer.
2509 val = ConvL2X(val);
2510 val = gvn().transform(new CastX2PNode(val));
2511 }
2512 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2513 }
2514
2515 return true;
2516 }
2517
2518 //----------------------------inline_unsafe_load_store----------------------------
2519 // This method serves a couple of different customers (depending on LoadStoreKind):
2520 //
2521 // LS_cmp_swap:
2522 //
2523 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2524 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2525 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2526 //
2527 // LS_cmp_swap_weak:
2528 //
2529 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2530 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2531 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2532 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2533 //
2534 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2535 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2536 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2537 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2538 //
2539 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2540 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2541 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2542 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2543 //
2544 // LS_cmp_exchange:
2545 //
2546 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2547 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2548 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2549 //
2550 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2551 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2552 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2553 //
2554 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2555 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2556 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2557 //
2558 // LS_get_add:
2559 //
2560 // int getAndAddInt( Object o, long offset, int delta)
2561 // long getAndAddLong(Object o, long offset, long delta)
2562 //
2563 // LS_get_set:
2564 //
2565 // int getAndSet(Object o, long offset, int newValue)
2566 // long getAndSet(Object o, long offset, long newValue)
2567 // Object getAndSet(Object o, long offset, Object newValue)
2568 //
2569 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2570 // This basic scheme here is the same as inline_unsafe_access, but
2571 // differs in enough details that combining them would make the code
2572 // overly confusing. (This is a true fact! I originally combined
2573 // them, but even I was confused by it!) As much code/comments as
2574 // possible are retained from inline_unsafe_access though to make
2575 // the correspondences clearer. - dl
2576
2577 if (callee()->is_static()) return false; // caller must have the capability!
2578
2579 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2580 decorators |= mo_decorator_for_access_kind(access_kind);
2581
2582 #ifndef PRODUCT
2583 BasicType rtype;
2584 {
2585 ResourceMark rm;
2586 // Check the signatures.
2587 ciSignature* sig = callee()->signature();
2588 rtype = sig->return_type()->basic_type();
2589 switch(kind) {
2590 case LS_get_add:
2591 case LS_get_set: {
2592 // Check the signatures.
2593 #ifdef ASSERT
2594 assert(rtype == type, "get and set must return the expected type");
2595 assert(sig->count() == 3, "get and set has 3 arguments");
2596 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2597 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2598 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2599 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2600 #endif // ASSERT
2601 break;
2602 }
2603 case LS_cmp_swap:
2604 case LS_cmp_swap_weak: {
2605 // Check the signatures.
2606 #ifdef ASSERT
2607 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2608 assert(sig->count() == 4, "CAS has 4 arguments");
2609 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2610 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2611 #endif // ASSERT
2612 break;
2613 }
2614 case LS_cmp_exchange: {
2615 // Check the signatures.
2616 #ifdef ASSERT
2617 assert(rtype == type, "CAS must return the expected type");
2618 assert(sig->count() == 4, "CAS has 4 arguments");
2619 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2620 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2621 #endif // ASSERT
2622 break;
2623 }
2624 default:
2625 ShouldNotReachHere();
2626 }
2627 }
2628 #endif //PRODUCT
2629
2630 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2631
2632 // Get arguments:
2633 Node* receiver = nullptr;
2634 Node* base = nullptr;
2635 Node* offset = nullptr;
2636 Node* oldval = nullptr;
2637 Node* newval = nullptr;
2638 switch(kind) {
2639 case LS_cmp_swap:
2640 case LS_cmp_swap_weak:
2641 case LS_cmp_exchange: {
2642 const bool two_slot_type = type2size[type] == 2;
2643 receiver = argument(0); // type: oop
2644 base = argument(1); // type: oop
2645 offset = argument(2); // type: long
2646 oldval = argument(4); // type: oop, int, or long
2647 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2648 break;
2649 }
2650 case LS_get_add:
2651 case LS_get_set: {
2652 receiver = argument(0); // type: oop
2653 base = argument(1); // type: oop
2654 offset = argument(2); // type: long
2655 oldval = nullptr;
2656 newval = argument(4); // type: oop, int, or long
2657 break;
2658 }
2659 default:
2660 ShouldNotReachHere();
2661 }
2662
2663 // Build field offset expression.
2664 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2665 // to be plain byte offsets, which are also the same as those accepted
2666 // by oopDesc::field_addr.
2667 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2668 // 32-bit machines ignore the high half of long offsets
2669 offset = ConvL2X(offset);
2670 // Save state and restore on bailout
2671 uint old_sp = sp();
2672 SafePointNode* old_map = clone_map();
2673 Node* adr = make_unsafe_address(base, offset,type, false);
2674 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2675
2676 Compile::AliasType* alias_type = C->alias_type(adr_type);
2677 BasicType bt = alias_type->basic_type();
2678 if (bt != T_ILLEGAL &&
2679 (is_reference_type(bt) != (type == T_OBJECT))) {
2680 // Don't intrinsify mismatched object accesses.
2681 set_map(old_map);
2682 set_sp(old_sp);
2683 return false;
2684 }
2685
2686 destruct_map_clone(old_map);
2687
2688 // For CAS, unlike inline_unsafe_access, there seems no point in
2689 // trying to refine types. Just use the coarse types here.
2690 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2691 const Type *value_type = Type::get_const_basic_type(type);
2692
2693 switch (kind) {
2694 case LS_get_set:
2695 case LS_cmp_exchange: {
2696 if (type == T_OBJECT) {
2697 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2698 if (tjp != nullptr) {
2699 value_type = tjp;
2700 }
2701 }
2702 break;
2703 }
2704 case LS_cmp_swap:
2705 case LS_cmp_swap_weak:
2706 case LS_get_add:
2707 break;
2708 default:
2709 ShouldNotReachHere();
2710 }
2711
2712 // Null check receiver.
2713 receiver = null_check(receiver);
2714 if (stopped()) {
2715 return true;
2716 }
2717
2718 int alias_idx = C->get_alias_index(adr_type);
2719
2720 if (is_reference_type(type)) {
2721 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2722
2723 // Transformation of a value which could be null pointer (CastPP #null)
2724 // could be delayed during Parse (for example, in adjust_map_after_if()).
2725 // Execute transformation here to avoid barrier generation in such case.
2726 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2727 newval = _gvn.makecon(TypePtr::NULL_PTR);
2728
2729 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2730 // Refine the value to a null constant, when it is known to be null
2731 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2732 }
2733 }
2734
2735 Node* result = nullptr;
2736 switch (kind) {
2737 case LS_cmp_exchange: {
2738 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2739 oldval, newval, value_type, type, decorators);
2740 break;
2741 }
2742 case LS_cmp_swap_weak:
2743 decorators |= C2_WEAK_CMPXCHG;
2744 case LS_cmp_swap: {
2745 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2746 oldval, newval, value_type, type, decorators);
2747 break;
2748 }
2749 case LS_get_set: {
2750 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2751 newval, value_type, type, decorators);
2752 break;
2753 }
2754 case LS_get_add: {
2755 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2756 newval, value_type, type, decorators);
2757 break;
2758 }
2759 default:
2760 ShouldNotReachHere();
2761 }
2762
2763 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2764 set_result(result);
2765 return true;
2766 }
2767
2768 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2769 // Regardless of form, don't allow previous ld/st to move down,
2770 // then issue acquire, release, or volatile mem_bar.
2771 insert_mem_bar(Op_MemBarCPUOrder);
2772 switch(id) {
2773 case vmIntrinsics::_loadFence:
2774 insert_mem_bar(Op_LoadFence);
2775 return true;
2776 case vmIntrinsics::_storeFence:
2777 insert_mem_bar(Op_StoreFence);
2778 return true;
2779 case vmIntrinsics::_storeStoreFence:
2780 insert_mem_bar(Op_StoreStoreFence);
2781 return true;
2782 case vmIntrinsics::_fullFence:
2783 insert_mem_bar(Op_MemBarVolatile);
2784 return true;
2785 default:
2786 fatal_unexpected_iid(id);
2787 return false;
2788 }
2789 }
2790
2791 bool LibraryCallKit::inline_onspinwait() {
2792 insert_mem_bar(Op_OnSpinWait);
2793 return true;
2794 }
2795
2796 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2797 if (!kls->is_Con()) {
2798 return true;
2799 }
2800 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2801 if (klsptr == nullptr) {
2802 return true;
2803 }
2804 ciInstanceKlass* ik = klsptr->instance_klass();
2805 // don't need a guard for a klass that is already initialized
2806 return !ik->is_initialized();
2807 }
2808
2809 //----------------------------inline_unsafe_writeback0-------------------------
2810 // public native void Unsafe.writeback0(long address)
2811 bool LibraryCallKit::inline_unsafe_writeback0() {
2812 if (!Matcher::has_match_rule(Op_CacheWB)) {
2813 return false;
2814 }
2815 #ifndef PRODUCT
2816 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2817 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2818 ciSignature* sig = callee()->signature();
2819 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2820 #endif
2821 null_check_receiver(); // null-check, then ignore
2822 Node *addr = argument(1);
2823 addr = new CastX2PNode(addr);
2824 addr = _gvn.transform(addr);
2825 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2826 flush = _gvn.transform(flush);
2827 set_memory(flush, TypeRawPtr::BOTTOM);
2828 return true;
2829 }
2830
2831 //----------------------------inline_unsafe_writeback0-------------------------
2832 // public native void Unsafe.writeback0(long address)
2833 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2834 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2835 return false;
2836 }
2837 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2838 return false;
2839 }
2840 #ifndef PRODUCT
2841 assert(Matcher::has_match_rule(Op_CacheWB),
2842 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2843 : "found match rule for CacheWBPostSync but not CacheWB"));
2844
2845 #endif
2846 null_check_receiver(); // null-check, then ignore
2847 Node *sync;
2848 if (is_pre) {
2849 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2850 } else {
2851 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2852 }
2853 sync = _gvn.transform(sync);
2854 set_memory(sync, TypeRawPtr::BOTTOM);
2855 return true;
2856 }
2857
2858 //----------------------------inline_unsafe_allocate---------------------------
2859 // public native Object Unsafe.allocateInstance(Class<?> cls);
2860 bool LibraryCallKit::inline_unsafe_allocate() {
2861
2862 #if INCLUDE_JVMTI
2863 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
2864 return false;
2865 }
2866 #endif //INCLUDE_JVMTI
2867
2868 if (callee()->is_static()) return false; // caller must have the capability!
2869
2870 null_check_receiver(); // null-check, then ignore
2871 Node* cls = null_check(argument(1));
2872 if (stopped()) return true;
2873
2874 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
2875 kls = null_check(kls);
2876 if (stopped()) return true; // argument was like int.class
2877
2878 #if INCLUDE_JVMTI
2879 // Don't try to access new allocated obj in the intrinsic.
2880 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
2881 // Deoptimize and allocate in interpreter instead.
2882 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
2883 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
2884 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
2885 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
2886 {
2887 BuildCutout unless(this, tst, PROB_MAX);
2888 uncommon_trap(Deoptimization::Reason_intrinsic,
2889 Deoptimization::Action_make_not_entrant);
2890 }
2891 if (stopped()) {
2892 return true;
2893 }
2894 #endif //INCLUDE_JVMTI
2895
2896 Node* test = nullptr;
2897 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2898 // Note: The argument might still be an illegal value like
2899 // Serializable.class or Object[].class. The runtime will handle it.
2900 // But we must make an explicit check for initialization.
2901 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2902 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2903 // can generate code to load it as unsigned byte.
2904 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
2905 Node* bits = intcon(InstanceKlass::fully_initialized);
2906 test = _gvn.transform(new SubINode(inst, bits));
2907 // The 'test' is non-zero if we need to take a slow path.
2908 }
2909
2910 Node* obj = new_instance(kls, test);
2911 set_result(obj);
2912 return true;
2913 }
2914
2915 //------------------------inline_native_time_funcs--------------
2916 // inline code for System.currentTimeMillis() and System.nanoTime()
2917 // these have the same type and signature
2918 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2919 const TypeFunc* tf = OptoRuntime::void_long_Type();
2920 const TypePtr* no_memory_effects = nullptr;
2921 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2922 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2923 #ifdef ASSERT
2924 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2925 assert(value_top == top(), "second value must be top");
2926 #endif
2927 set_result(value);
2928 return true;
2929 }
2930
2931
2932 #if INCLUDE_JVMTI
2933
2934 // When notifications are disabled then just update the VTMS transition bit and return.
2935 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
2936 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
2937 if (!DoJVMTIVirtualThreadTransitions) {
2938 return true;
2939 }
2940 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
2941 IdealKit ideal(this);
2942
2943 Node* ONE = ideal.ConI(1);
2944 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
2945 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
2946 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
2947
2948 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
2949 sync_kit(ideal);
2950 // if notifyJvmti enabled then make a call to the given SharedRuntime function
2951 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
2952 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
2953 ideal.sync_kit(this);
2954 } ideal.else_(); {
2955 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
2956 Node* thread = ideal.thread();
2957 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
2958 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
2959
2960 sync_kit(ideal);
2961 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2962 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2963
2964 ideal.sync_kit(this);
2965 } ideal.end_if();
2966 final_sync(ideal);
2967
2968 return true;
2969 }
2970
2971 // Always update the is_disable_suspend bit.
2972 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
2973 if (!DoJVMTIVirtualThreadTransitions) {
2974 return true;
2975 }
2976 IdealKit ideal(this);
2977
2978 {
2979 // unconditionally update the is_disable_suspend bit in current JavaThread
2980 Node* thread = ideal.thread();
2981 Node* arg = _gvn.transform(argument(0)); // argument for notification
2982 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
2983 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
2984
2985 sync_kit(ideal);
2986 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2987 ideal.sync_kit(this);
2988 }
2989 final_sync(ideal);
2990
2991 return true;
2992 }
2993
2994 #endif // INCLUDE_JVMTI
2995
2996 #ifdef JFR_HAVE_INTRINSICS
2997
2998 /**
2999 * if oop->klass != null
3000 * // normal class
3001 * epoch = _epoch_state ? 2 : 1
3002 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3003 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3004 * }
3005 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3006 * else
3007 * // primitive class
3008 * if oop->array_klass != null
3009 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3010 * else
3011 * id = LAST_TYPE_ID + 1 // void class path
3012 * if (!signaled)
3013 * signaled = true
3014 */
3015 bool LibraryCallKit::inline_native_classID() {
3016 Node* cls = argument(0);
3017
3018 IdealKit ideal(this);
3019 #define __ ideal.
3020 IdealVariable result(ideal); __ declarations_done();
3021 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3022 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3023 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3024
3025
3026 __ if_then(kls, BoolTest::ne, null()); {
3027 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3028 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3029
3030 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3031 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3032 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3033 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3034 mask = _gvn.transform(new OrLNode(mask, epoch));
3035 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3036
3037 float unlikely = PROB_UNLIKELY(0.999);
3038 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3039 sync_kit(ideal);
3040 make_runtime_call(RC_LEAF,
3041 OptoRuntime::class_id_load_barrier_Type(),
3042 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3043 "class id load barrier",
3044 TypePtr::BOTTOM,
3045 kls);
3046 ideal.sync_kit(this);
3047 } __ end_if();
3048
3049 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3050 } __ else_(); {
3051 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3052 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3053 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3054 __ if_then(array_kls, BoolTest::ne, null()); {
3055 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3056 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3057 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3058 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3059 } __ else_(); {
3060 // void class case
3061 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3062 } __ end_if();
3063
3064 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3065 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3066 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3067 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3068 } __ end_if();
3069 } __ end_if();
3070
3071 final_sync(ideal);
3072 set_result(ideal.value(result));
3073 #undef __
3074 return true;
3075 }
3076
3077 //------------------------inline_native_jvm_commit------------------
3078 bool LibraryCallKit::inline_native_jvm_commit() {
3079 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3080
3081 // Save input memory and i_o state.
3082 Node* input_memory_state = reset_memory();
3083 set_all_memory(input_memory_state);
3084 Node* input_io_state = i_o();
3085
3086 // TLS.
3087 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3088 // Jfr java buffer.
3089 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3090 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3091 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3092
3093 // Load the current value of the notified field in the JfrThreadLocal.
3094 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3095 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3096
3097 // Test for notification.
3098 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3099 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3100 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3101
3102 // True branch, is notified.
3103 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3104 set_control(is_notified);
3105
3106 // Reset notified state.
3107 Node* notified_reset_memory = store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3108
3109 // 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.
3110 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3111 // Convert the machine-word to a long.
3112 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3113
3114 // False branch, not notified.
3115 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3116 set_control(not_notified);
3117 set_all_memory(input_memory_state);
3118
3119 // Arg is the next position as a long.
3120 Node* arg = argument(0);
3121 // Convert long to machine-word.
3122 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3123
3124 // Store the next_position to the underlying jfr java buffer.
3125 Node* commit_memory;
3126 #ifdef _LP64
3127 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_LONG, MemNode::release);
3128 #else
3129 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_INT, MemNode::release);
3130 #endif
3131
3132 // 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.
3133 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3134 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3135 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3136
3137 // And flags with lease constant.
3138 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3139
3140 // Branch on lease to conditionalize returning the leased java buffer.
3141 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3142 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3143 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3144
3145 // False branch, not a lease.
3146 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3147
3148 // True branch, is lease.
3149 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3150 set_control(is_lease);
3151
3152 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3153 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3154 OptoRuntime::void_void_Type(),
3155 SharedRuntime::jfr_return_lease(),
3156 "return_lease", TypePtr::BOTTOM);
3157 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3158
3159 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3160 record_for_igvn(lease_compare_rgn);
3161 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3162 record_for_igvn(lease_compare_mem);
3163 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3164 record_for_igvn(lease_compare_io);
3165 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3166 record_for_igvn(lease_result_value);
3167
3168 // Update control and phi nodes.
3169 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3170 lease_compare_rgn->init_req(_false_path, not_lease);
3171
3172 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3173 lease_compare_mem->init_req(_false_path, commit_memory);
3174
3175 lease_compare_io->init_req(_true_path, i_o());
3176 lease_compare_io->init_req(_false_path, input_io_state);
3177
3178 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3179 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3180
3181 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3182 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3183 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3184 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3185
3186 // Update control and phi nodes.
3187 result_rgn->init_req(_true_path, is_notified);
3188 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3189
3190 result_mem->init_req(_true_path, notified_reset_memory);
3191 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3192
3193 result_io->init_req(_true_path, input_io_state);
3194 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3195
3196 result_value->init_req(_true_path, current_pos);
3197 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3198
3199 // Set output state.
3200 set_control(_gvn.transform(result_rgn));
3201 set_all_memory(_gvn.transform(result_mem));
3202 set_i_o(_gvn.transform(result_io));
3203 set_result(result_rgn, result_value);
3204 return true;
3205 }
3206
3207 /*
3208 * The intrinsic is a model of this pseudo-code:
3209 *
3210 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3211 * jobject h_event_writer = tl->java_event_writer();
3212 * if (h_event_writer == nullptr) {
3213 * return nullptr;
3214 * }
3215 * oop threadObj = Thread::threadObj();
3216 * oop vthread = java_lang_Thread::vthread(threadObj);
3217 * traceid tid;
3218 * bool pinVirtualThread;
3219 * bool excluded;
3220 * if (vthread != threadObj) { // i.e. current thread is virtual
3221 * tid = java_lang_Thread::tid(vthread);
3222 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3223 * pinVirtualThread = VMContinuations;
3224 * excluded = vthread_epoch_raw & excluded_mask;
3225 * if (!excluded) {
3226 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3227 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3228 * if (vthread_epoch != current_epoch) {
3229 * write_checkpoint();
3230 * }
3231 * }
3232 * } else {
3233 * tid = java_lang_Thread::tid(threadObj);
3234 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3235 * pinVirtualThread = false;
3236 * excluded = thread_epoch_raw & excluded_mask;
3237 * }
3238 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3239 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3240 * if (tid_in_event_writer != tid) {
3241 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3242 * setField(event_writer, "excluded", excluded);
3243 * setField(event_writer, "threadID", tid);
3244 * }
3245 * return event_writer
3246 */
3247 bool LibraryCallKit::inline_native_getEventWriter() {
3248 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3249
3250 // Save input memory and i_o state.
3251 Node* input_memory_state = reset_memory();
3252 set_all_memory(input_memory_state);
3253 Node* input_io_state = i_o();
3254
3255 // The most significant bit of the u2 is used to denote thread exclusion
3256 Node* excluded_shift = _gvn.intcon(15);
3257 Node* excluded_mask = _gvn.intcon(1 << 15);
3258 // The epoch generation is the range [1-32767]
3259 Node* epoch_mask = _gvn.intcon(32767);
3260
3261 // TLS
3262 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3263
3264 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3265 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3266
3267 // Load the eventwriter jobject handle.
3268 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3269
3270 // Null check the jobject handle.
3271 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3272 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3273 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3274
3275 // False path, jobj is null.
3276 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3277
3278 // True path, jobj is not null.
3279 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3280
3281 set_control(jobj_is_not_null);
3282
3283 // Load the threadObj for the CarrierThread.
3284 Node* threadObj = generate_current_thread(tls_ptr);
3285
3286 // Load the vthread.
3287 Node* vthread = generate_virtual_thread(tls_ptr);
3288
3289 // If vthread != threadObj, this is a virtual thread.
3290 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3291 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3292 IfNode* iff_vthread_not_equal_threadObj =
3293 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3294
3295 // False branch, fallback to threadObj.
3296 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3297 set_control(vthread_equal_threadObj);
3298
3299 // Load the tid field from the vthread object.
3300 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3301
3302 // Load the raw epoch value from the threadObj.
3303 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3304 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3305 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3306 TypeInt::CHAR, T_CHAR,
3307 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3308
3309 // Mask off the excluded information from the epoch.
3310 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3311
3312 // True branch, this is a virtual thread.
3313 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3314 set_control(vthread_not_equal_threadObj);
3315
3316 // Load the tid field from the vthread object.
3317 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3318
3319 // Continuation support determines if a virtual thread should be pinned.
3320 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3321 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3322
3323 // Load the raw epoch value from the vthread.
3324 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3325 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3326 TypeInt::CHAR, T_CHAR,
3327 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3328
3329 // Mask off the excluded information from the epoch.
3330 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3331
3332 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3333 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3334 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3335 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3336
3337 // False branch, vthread is excluded, no need to write epoch info.
3338 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3339
3340 // True branch, vthread is included, update epoch info.
3341 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3342 set_control(included);
3343
3344 // Get epoch value.
3345 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3346
3347 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3348 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3349 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3350
3351 // Compare the epoch in the vthread to the current epoch generation.
3352 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3353 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3354 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3355
3356 // False path, epoch is equal, checkpoint information is valid.
3357 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3358
3359 // True path, epoch is not equal, write a checkpoint for the vthread.
3360 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3361
3362 set_control(epoch_is_not_equal);
3363
3364 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3365 // The call also updates the native thread local thread id and the vthread with the current epoch.
3366 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3367 OptoRuntime::jfr_write_checkpoint_Type(),
3368 SharedRuntime::jfr_write_checkpoint(),
3369 "write_checkpoint", TypePtr::BOTTOM);
3370 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3371
3372 // vthread epoch != current epoch
3373 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3374 record_for_igvn(epoch_compare_rgn);
3375 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3376 record_for_igvn(epoch_compare_mem);
3377 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3378 record_for_igvn(epoch_compare_io);
3379
3380 // Update control and phi nodes.
3381 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3382 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3383 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3384 epoch_compare_mem->init_req(_false_path, input_memory_state);
3385 epoch_compare_io->init_req(_true_path, i_o());
3386 epoch_compare_io->init_req(_false_path, input_io_state);
3387
3388 // excluded != true
3389 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3390 record_for_igvn(exclude_compare_rgn);
3391 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3392 record_for_igvn(exclude_compare_mem);
3393 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3394 record_for_igvn(exclude_compare_io);
3395
3396 // Update control and phi nodes.
3397 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3398 exclude_compare_rgn->init_req(_false_path, excluded);
3399 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3400 exclude_compare_mem->init_req(_false_path, input_memory_state);
3401 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3402 exclude_compare_io->init_req(_false_path, input_io_state);
3403
3404 // vthread != threadObj
3405 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3406 record_for_igvn(vthread_compare_rgn);
3407 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3408 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3409 record_for_igvn(vthread_compare_io);
3410 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3411 record_for_igvn(tid);
3412 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3413 record_for_igvn(exclusion);
3414 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3415 record_for_igvn(pinVirtualThread);
3416
3417 // Update control and phi nodes.
3418 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3419 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3420 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3421 vthread_compare_mem->init_req(_false_path, input_memory_state);
3422 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3423 vthread_compare_io->init_req(_false_path, input_io_state);
3424 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3425 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3426 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3427 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3428 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3429 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3430
3431 // Update branch state.
3432 set_control(_gvn.transform(vthread_compare_rgn));
3433 set_all_memory(_gvn.transform(vthread_compare_mem));
3434 set_i_o(_gvn.transform(vthread_compare_io));
3435
3436 // Load the event writer oop by dereferencing the jobject handle.
3437 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3438 assert(klass_EventWriter->is_loaded(), "invariant");
3439 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3440 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3441 const TypeOopPtr* const xtype = aklass->as_instance_type();
3442 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3443 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3444
3445 // Load the current thread id from the event writer object.
3446 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3447 // Get the field offset to, conditionally, store an updated tid value later.
3448 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3449 // Get the field offset to, conditionally, store an updated exclusion value later.
3450 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3451 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3452 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3453
3454 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3455 record_for_igvn(event_writer_tid_compare_rgn);
3456 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3457 record_for_igvn(event_writer_tid_compare_mem);
3458 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3459 record_for_igvn(event_writer_tid_compare_io);
3460
3461 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3462 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3463 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3464 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3465
3466 // False path, tids are the same.
3467 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3468
3469 // True path, tid is not equal, need to update the tid in the event writer.
3470 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3471 record_for_igvn(tid_is_not_equal);
3472
3473 // Store the pin state to the event writer.
3474 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3475
3476 // Store the exclusion state to the event writer.
3477 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3478 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3479
3480 // Store the tid to the event writer.
3481 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3482
3483 // Update control and phi nodes.
3484 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3485 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3486 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3487 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3488 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3489 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3490
3491 // Result of top level CFG, Memory, IO and Value.
3492 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3493 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3494 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3495 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3496
3497 // Result control.
3498 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3499 result_rgn->init_req(_false_path, jobj_is_null);
3500
3501 // Result memory.
3502 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3503 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3504
3505 // Result IO.
3506 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3507 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3508
3509 // Result value.
3510 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3511 result_value->init_req(_false_path, null()); // return null
3512
3513 // Set output state.
3514 set_control(_gvn.transform(result_rgn));
3515 set_all_memory(_gvn.transform(result_mem));
3516 set_i_o(_gvn.transform(result_io));
3517 set_result(result_rgn, result_value);
3518 return true;
3519 }
3520
3521 /*
3522 * The intrinsic is a model of this pseudo-code:
3523 *
3524 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3525 * if (carrierThread != thread) { // is virtual thread
3526 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3527 * bool excluded = vthread_epoch_raw & excluded_mask;
3528 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3529 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3530 * if (!excluded) {
3531 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3532 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3533 * }
3534 * Atomic::release_store(&tl->_vthread, true);
3535 * return;
3536 * }
3537 * Atomic::release_store(&tl->_vthread, false);
3538 */
3539 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3540 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3541
3542 Node* input_memory_state = reset_memory();
3543 set_all_memory(input_memory_state);
3544
3545 // The most significant bit of the u2 is used to denote thread exclusion
3546 Node* excluded_mask = _gvn.intcon(1 << 15);
3547 // The epoch generation is the range [1-32767]
3548 Node* epoch_mask = _gvn.intcon(32767);
3549
3550 Node* const carrierThread = generate_current_thread(jt);
3551 // If thread != carrierThread, this is a virtual thread.
3552 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3553 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3554 IfNode* iff_thread_not_equal_carrierThread =
3555 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3556
3557 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3558
3559 // False branch, is carrierThread.
3560 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3561 // Store release
3562 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3563
3564 set_all_memory(input_memory_state);
3565
3566 // True branch, is virtual thread.
3567 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3568 set_control(thread_not_equal_carrierThread);
3569
3570 // Load the raw epoch value from the vthread.
3571 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3572 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
3573 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3574
3575 // Mask off the excluded information from the epoch.
3576 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3577
3578 // Load the tid field from the thread.
3579 Node* tid = load_field_from_object(thread, "tid", "J");
3580
3581 // Store the vthread tid to the jfr thread local.
3582 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3583 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
3584
3585 // Branch is_excluded to conditionalize updating the epoch .
3586 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3587 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3588 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3589
3590 // True branch, vthread is excluded, no need to write epoch info.
3591 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3592 set_control(excluded);
3593 Node* vthread_is_excluded = _gvn.intcon(1);
3594
3595 // False branch, vthread is included, update epoch info.
3596 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3597 set_control(included);
3598 Node* vthread_is_included = _gvn.intcon(0);
3599
3600 // Get epoch value.
3601 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3602
3603 // Store the vthread epoch to the jfr thread local.
3604 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3605 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
3606
3607 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3608 record_for_igvn(excluded_rgn);
3609 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3610 record_for_igvn(excluded_mem);
3611 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3612 record_for_igvn(exclusion);
3613
3614 // Merge the excluded control and memory.
3615 excluded_rgn->init_req(_true_path, excluded);
3616 excluded_rgn->init_req(_false_path, included);
3617 excluded_mem->init_req(_true_path, tid_memory);
3618 excluded_mem->init_req(_false_path, included_memory);
3619 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3620 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3621
3622 // Set intermediate state.
3623 set_control(_gvn.transform(excluded_rgn));
3624 set_all_memory(excluded_mem);
3625
3626 // Store the vthread exclusion state to the jfr thread local.
3627 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3628 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
3629
3630 // Store release
3631 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
3632
3633 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3634 record_for_igvn(thread_compare_rgn);
3635 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3636 record_for_igvn(thread_compare_mem);
3637 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3638 record_for_igvn(vthread);
3639
3640 // Merge the thread_compare control and memory.
3641 thread_compare_rgn->init_req(_true_path, control());
3642 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3643 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3644 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3645
3646 // Set output state.
3647 set_control(_gvn.transform(thread_compare_rgn));
3648 set_all_memory(_gvn.transform(thread_compare_mem));
3649 }
3650
3651 #endif // JFR_HAVE_INTRINSICS
3652
3653 //------------------------inline_native_currentCarrierThread------------------
3654 bool LibraryCallKit::inline_native_currentCarrierThread() {
3655 Node* junk = nullptr;
3656 set_result(generate_current_thread(junk));
3657 return true;
3658 }
3659
3660 //------------------------inline_native_currentThread------------------
3661 bool LibraryCallKit::inline_native_currentThread() {
3662 Node* junk = nullptr;
3663 set_result(generate_virtual_thread(junk));
3664 return true;
3665 }
3666
3667 //------------------------inline_native_setVthread------------------
3668 bool LibraryCallKit::inline_native_setCurrentThread() {
3669 assert(C->method()->changes_current_thread(),
3670 "method changes current Thread but is not annotated ChangesCurrentThread");
3671 Node* arr = argument(1);
3672 Node* thread = _gvn.transform(new ThreadLocalNode());
3673 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3674 Node* thread_obj_handle
3675 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3676 thread_obj_handle = _gvn.transform(thread_obj_handle);
3677 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3678 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3679
3680 // Change the _monitor_owner_id of the JavaThread
3681 Node* tid = load_field_from_object(arr, "tid", "J");
3682 Node* monitor_owner_id_offset = basic_plus_adr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
3683 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
3684
3685 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3686 return true;
3687 }
3688
3689 const Type* LibraryCallKit::scopedValueCache_type() {
3690 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3691 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3692 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3693
3694 // Because we create the scopedValue cache lazily we have to make the
3695 // type of the result BotPTR.
3696 bool xk = etype->klass_is_exact();
3697 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3698 return objects_type;
3699 }
3700
3701 Node* LibraryCallKit::scopedValueCache_helper() {
3702 Node* thread = _gvn.transform(new ThreadLocalNode());
3703 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3704 // We cannot use immutable_memory() because we might flip onto a
3705 // different carrier thread, at which point we'll need to use that
3706 // carrier thread's cache.
3707 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3708 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3709 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3710 }
3711
3712 //------------------------inline_native_scopedValueCache------------------
3713 bool LibraryCallKit::inline_native_scopedValueCache() {
3714 Node* cache_obj_handle = scopedValueCache_helper();
3715 const Type* objects_type = scopedValueCache_type();
3716 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3717
3718 return true;
3719 }
3720
3721 //------------------------inline_native_setScopedValueCache------------------
3722 bool LibraryCallKit::inline_native_setScopedValueCache() {
3723 Node* arr = argument(0);
3724 Node* cache_obj_handle = scopedValueCache_helper();
3725 const Type* objects_type = scopedValueCache_type();
3726
3727 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3728 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3729
3730 return true;
3731 }
3732
3733 //------------------------inline_native_Continuation_pin and unpin-----------
3734
3735 // Shared implementation routine for both pin and unpin.
3736 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
3737 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3738
3739 // Save input memory.
3740 Node* input_memory_state = reset_memory();
3741 set_all_memory(input_memory_state);
3742
3743 // TLS
3744 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3745 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
3746 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
3747
3748 // Null check the last continuation object.
3749 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
3750 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
3751 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3752
3753 // False path, last continuation is null.
3754 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
3755
3756 // True path, last continuation is not null.
3757 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
3758
3759 set_control(continuation_is_not_null);
3760
3761 // Load the pin count from the last continuation.
3762 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
3763 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
3764
3765 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
3766 Node* pin_count_rhs;
3767 if (unpin) {
3768 pin_count_rhs = _gvn.intcon(0);
3769 } else {
3770 pin_count_rhs = _gvn.intcon(UINT32_MAX);
3771 }
3772 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
3773 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
3774 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
3775
3776 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
3777 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
3778 set_control(valid_pin_count);
3779
3780 Node* next_pin_count;
3781 if (unpin) {
3782 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
3783 } else {
3784 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
3785 }
3786
3787 Node* updated_pin_count_memory = store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
3788
3789 // True branch, pin count over/underflow.
3790 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
3791 {
3792 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
3793 // which will throw IllegalStateException for pin count over/underflow.
3794 PreserveJVMState pjvms(this);
3795 set_control(pin_count_over_underflow);
3796 set_all_memory(input_memory_state);
3797 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
3798 Deoptimization::Action_none);
3799 assert(stopped(), "invariant");
3800 }
3801
3802 // Result of top level CFG and Memory.
3803 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3804 record_for_igvn(result_rgn);
3805 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3806 record_for_igvn(result_mem);
3807
3808 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
3809 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
3810 result_mem->init_req(_true_path, _gvn.transform(updated_pin_count_memory));
3811 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3812
3813 // Set output state.
3814 set_control(_gvn.transform(result_rgn));
3815 set_all_memory(_gvn.transform(result_mem));
3816
3817 return true;
3818 }
3819
3820 //---------------------------load_mirror_from_klass----------------------------
3821 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3822 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3823 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3824 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3825 // mirror = ((OopHandle)mirror)->resolve();
3826 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3827 }
3828
3829 //-----------------------load_klass_from_mirror_common-------------------------
3830 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3831 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3832 // and branch to the given path on the region.
3833 // If never_see_null, take an uncommon trap on null, so we can optimistically
3834 // compile for the non-null case.
3835 // If the region is null, force never_see_null = true.
3836 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3837 bool never_see_null,
3838 RegionNode* region,
3839 int null_path,
3840 int offset) {
3841 if (region == nullptr) never_see_null = true;
3842 Node* p = basic_plus_adr(mirror, offset);
3843 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3844 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3845 Node* null_ctl = top();
3846 kls = null_check_oop(kls, &null_ctl, never_see_null);
3847 if (region != nullptr) {
3848 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3849 region->init_req(null_path, null_ctl);
3850 } else {
3851 assert(null_ctl == top(), "no loose ends");
3852 }
3853 return kls;
3854 }
3855
3856 //--------------------(inline_native_Class_query helpers)---------------------
3857 // Use this for JVM_ACC_INTERFACE.
3858 // Fall through if (mods & mask) == bits, take the guard otherwise.
3859 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
3860 ByteSize offset, const Type* type, BasicType bt) {
3861 // Branch around if the given klass has the given modifier bit set.
3862 // Like generate_guard, adds a new path onto the region.
3863 Node* modp = basic_plus_adr(kls, in_bytes(offset));
3864 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
3865 Node* mask = intcon(modifier_mask);
3866 Node* bits = intcon(modifier_bits);
3867 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3868 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3869 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3870 return generate_fair_guard(bol, region);
3871 }
3872 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3873 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
3874 Klass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
3875 }
3876
3877 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
3878 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3879 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
3880 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
3881 }
3882
3883 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3884 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
3885 }
3886
3887 //-------------------------inline_native_Class_query-------------------
3888 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3889 const Type* return_type = TypeInt::BOOL;
3890 Node* prim_return_value = top(); // what happens if it's a primitive class?
3891 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3892 bool expect_prim = false; // most of these guys expect to work on refs
3893
3894 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3895
3896 Node* mirror = argument(0);
3897 Node* obj = top();
3898
3899 switch (id) {
3900 case vmIntrinsics::_isInstance:
3901 // nothing is an instance of a primitive type
3902 prim_return_value = intcon(0);
3903 obj = argument(1);
3904 break;
3905 case vmIntrinsics::_getModifiers:
3906 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3907 return_type = TypeInt::CHAR;
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::CHAR;
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::CHAR, T_CHAR, 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::CHAR, T_CHAR, 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, Node** obj) {
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 Node* ctrl = generate_fair_guard(bol, region);
4302 Node* is_array_ctrl = not_array ? control() : ctrl;
4303 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4304 // Keep track of the fact that 'obj' is an array to prevent
4305 // array specific accesses from floating above the guard.
4306 *obj = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4307 }
4308 return ctrl;
4309 }
4310
4311
4312 //-----------------------inline_native_newArray--------------------------
4313 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4314 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4315 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4316 Node* mirror;
4317 Node* count_val;
4318 if (uninitialized) {
4319 null_check_receiver();
4320 mirror = argument(1);
4321 count_val = argument(2);
4322 } else {
4323 mirror = argument(0);
4324 count_val = argument(1);
4325 }
4326
4327 mirror = null_check(mirror);
4328 // If mirror or obj is dead, only null-path is taken.
4329 if (stopped()) return true;
4330
4331 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4332 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4333 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4334 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4335 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4336
4337 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4338 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4339 result_reg, _slow_path);
4340 Node* normal_ctl = control();
4341 Node* no_array_ctl = result_reg->in(_slow_path);
4342
4343 // Generate code for the slow case. We make a call to newArray().
4344 set_control(no_array_ctl);
4345 if (!stopped()) {
4346 // Either the input type is void.class, or else the
4347 // array klass has not yet been cached. Either the
4348 // ensuing call will throw an exception, or else it
4349 // will cache the array klass for next time.
4350 PreserveJVMState pjvms(this);
4351 CallJavaNode* slow_call = nullptr;
4352 if (uninitialized) {
4353 // Generate optimized virtual call (holder class 'Unsafe' is final)
4354 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4355 } else {
4356 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4357 }
4358 Node* slow_result = set_results_for_java_call(slow_call);
4359 // this->control() comes from set_results_for_java_call
4360 result_reg->set_req(_slow_path, control());
4361 result_val->set_req(_slow_path, slow_result);
4362 result_io ->set_req(_slow_path, i_o());
4363 result_mem->set_req(_slow_path, reset_memory());
4364 }
4365
4366 set_control(normal_ctl);
4367 if (!stopped()) {
4368 // Normal case: The array type has been cached in the java.lang.Class.
4369 // The following call works fine even if the array type is polymorphic.
4370 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4371 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4372 result_reg->init_req(_normal_path, control());
4373 result_val->init_req(_normal_path, obj);
4374 result_io ->init_req(_normal_path, i_o());
4375 result_mem->init_req(_normal_path, reset_memory());
4376
4377 if (uninitialized) {
4378 // Mark the allocation so that zeroing is skipped
4379 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4380 alloc->maybe_set_complete(&_gvn);
4381 }
4382 }
4383
4384 // Return the combined state.
4385 set_i_o( _gvn.transform(result_io) );
4386 set_all_memory( _gvn.transform(result_mem));
4387
4388 C->set_has_split_ifs(true); // Has chance for split-if optimization
4389 set_result(result_reg, result_val);
4390 return true;
4391 }
4392
4393 //----------------------inline_native_getLength--------------------------
4394 // public static native int java.lang.reflect.Array.getLength(Object array);
4395 bool LibraryCallKit::inline_native_getLength() {
4396 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4397
4398 Node* array = null_check(argument(0));
4399 // If array is dead, only null-path is taken.
4400 if (stopped()) return true;
4401
4402 // Deoptimize if it is a non-array.
4403 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
4404
4405 if (non_array != nullptr) {
4406 PreserveJVMState pjvms(this);
4407 set_control(non_array);
4408 uncommon_trap(Deoptimization::Reason_intrinsic,
4409 Deoptimization::Action_maybe_recompile);
4410 }
4411
4412 // If control is dead, only non-array-path is taken.
4413 if (stopped()) return true;
4414
4415 // The works fine even if the array type is polymorphic.
4416 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4417 Node* result = load_array_length(array);
4418
4419 C->set_has_split_ifs(true); // Has chance for split-if optimization
4420 set_result(result);
4421 return true;
4422 }
4423
4424 //------------------------inline_array_copyOf----------------------------
4425 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4426 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4427 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4428 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4429
4430 // Get the arguments.
4431 Node* original = argument(0);
4432 Node* start = is_copyOfRange? argument(1): intcon(0);
4433 Node* end = is_copyOfRange? argument(2): argument(1);
4434 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4435
4436 Node* newcopy = nullptr;
4437
4438 // Set the original stack and the reexecute bit for the interpreter to reexecute
4439 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4440 { PreserveReexecuteState preexecs(this);
4441 jvms()->set_should_reexecute(true);
4442
4443 array_type_mirror = null_check(array_type_mirror);
4444 original = null_check(original);
4445
4446 // Check if a null path was taken unconditionally.
4447 if (stopped()) return true;
4448
4449 Node* orig_length = load_array_length(original);
4450
4451 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4452 klass_node = null_check(klass_node);
4453
4454 RegionNode* bailout = new RegionNode(1);
4455 record_for_igvn(bailout);
4456
4457 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4458 // Bail out if that is so.
4459 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4460 if (not_objArray != nullptr) {
4461 // Improve the klass node's type from the new optimistic assumption:
4462 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4463 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4464 Node* cast = new CastPPNode(control(), klass_node, akls);
4465 klass_node = _gvn.transform(cast);
4466 }
4467
4468 // Bail out if either start or end is negative.
4469 generate_negative_guard(start, bailout, &start);
4470 generate_negative_guard(end, bailout, &end);
4471
4472 Node* length = end;
4473 if (_gvn.type(start) != TypeInt::ZERO) {
4474 length = _gvn.transform(new SubINode(end, start));
4475 }
4476
4477 // Bail out if length is negative (i.e., if start > end).
4478 // Without this the new_array would throw
4479 // NegativeArraySizeException but IllegalArgumentException is what
4480 // should be thrown
4481 generate_negative_guard(length, bailout, &length);
4482
4483 // Bail out if start is larger than the original length
4484 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4485 generate_negative_guard(orig_tail, bailout, &orig_tail);
4486
4487 if (bailout->req() > 1) {
4488 PreserveJVMState pjvms(this);
4489 set_control(_gvn.transform(bailout));
4490 uncommon_trap(Deoptimization::Reason_intrinsic,
4491 Deoptimization::Action_maybe_recompile);
4492 }
4493
4494 if (!stopped()) {
4495 // How many elements will we copy from the original?
4496 // The answer is MinI(orig_tail, length).
4497 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4498
4499 // Generate a direct call to the right arraycopy function(s).
4500 // We know the copy is disjoint but we might not know if the
4501 // oop stores need checking.
4502 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4503 // This will fail a store-check if x contains any non-nulls.
4504
4505 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4506 // loads/stores but it is legal only if we're sure the
4507 // Arrays.copyOf would succeed. So we need all input arguments
4508 // to the copyOf to be validated, including that the copy to the
4509 // new array won't trigger an ArrayStoreException. That subtype
4510 // check can be optimized if we know something on the type of
4511 // the input array from type speculation.
4512 if (_gvn.type(klass_node)->singleton()) {
4513 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4514 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4515
4516 int test = C->static_subtype_check(superk, subk);
4517 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4518 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4519 if (t_original->speculative_type() != nullptr) {
4520 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4521 }
4522 }
4523 }
4524
4525 bool validated = false;
4526 // Reason_class_check rather than Reason_intrinsic because we
4527 // want to intrinsify even if this traps.
4528 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4529 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4530
4531 if (not_subtype_ctrl != top()) {
4532 PreserveJVMState pjvms(this);
4533 set_control(not_subtype_ctrl);
4534 uncommon_trap(Deoptimization::Reason_class_check,
4535 Deoptimization::Action_make_not_entrant);
4536 assert(stopped(), "Should be stopped");
4537 }
4538 validated = true;
4539 }
4540
4541 if (!stopped()) {
4542 newcopy = new_array(klass_node, length, 0); // no arguments to push
4543
4544 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4545 load_object_klass(original), klass_node);
4546 if (!is_copyOfRange) {
4547 ac->set_copyof(validated);
4548 } else {
4549 ac->set_copyofrange(validated);
4550 }
4551 Node* n = _gvn.transform(ac);
4552 if (n == ac) {
4553 ac->connect_outputs(this);
4554 } else {
4555 assert(validated, "shouldn't transform if all arguments not validated");
4556 set_all_memory(n);
4557 }
4558 }
4559 }
4560 } // original reexecute is set back here
4561
4562 C->set_has_split_ifs(true); // Has chance for split-if optimization
4563 if (!stopped()) {
4564 set_result(newcopy);
4565 }
4566 return true;
4567 }
4568
4569
4570 //----------------------generate_virtual_guard---------------------------
4571 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4572 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4573 RegionNode* slow_region) {
4574 ciMethod* method = callee();
4575 int vtable_index = method->vtable_index();
4576 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4577 "bad index %d", vtable_index);
4578 // Get the Method* out of the appropriate vtable entry.
4579 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4580 vtable_index*vtableEntry::size_in_bytes() +
4581 in_bytes(vtableEntry::method_offset());
4582 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4583 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4584
4585 // Compare the target method with the expected method (e.g., Object.hashCode).
4586 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4587
4588 Node* native_call = makecon(native_call_addr);
4589 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4590 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4591
4592 return generate_slow_guard(test_native, slow_region);
4593 }
4594
4595 //-----------------------generate_method_call----------------------------
4596 // Use generate_method_call to make a slow-call to the real
4597 // method if the fast path fails. An alternative would be to
4598 // use a stub like OptoRuntime::slow_arraycopy_Java.
4599 // This only works for expanding the current library call,
4600 // not another intrinsic. (E.g., don't use this for making an
4601 // arraycopy call inside of the copyOf intrinsic.)
4602 CallJavaNode*
4603 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4604 // When compiling the intrinsic method itself, do not use this technique.
4605 guarantee(callee() != C->method(), "cannot make slow-call to self");
4606
4607 ciMethod* method = callee();
4608 // ensure the JVMS we have will be correct for this call
4609 guarantee(method_id == method->intrinsic_id(), "must match");
4610
4611 const TypeFunc* tf = TypeFunc::make(method);
4612 if (res_not_null) {
4613 assert(tf->return_type() == T_OBJECT, "");
4614 const TypeTuple* range = tf->range();
4615 const Type** fields = TypeTuple::fields(range->cnt());
4616 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4617 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4618 tf = TypeFunc::make(tf->domain(), new_range);
4619 }
4620 CallJavaNode* slow_call;
4621 if (is_static) {
4622 assert(!is_virtual, "");
4623 slow_call = new CallStaticJavaNode(C, tf,
4624 SharedRuntime::get_resolve_static_call_stub(), method);
4625 } else if (is_virtual) {
4626 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4627 int vtable_index = Method::invalid_vtable_index;
4628 if (UseInlineCaches) {
4629 // Suppress the vtable call
4630 } else {
4631 // hashCode and clone are not a miranda methods,
4632 // so the vtable index is fixed.
4633 // No need to use the linkResolver to get it.
4634 vtable_index = method->vtable_index();
4635 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4636 "bad index %d", vtable_index);
4637 }
4638 slow_call = new CallDynamicJavaNode(tf,
4639 SharedRuntime::get_resolve_virtual_call_stub(),
4640 method, vtable_index);
4641 } else { // neither virtual nor static: opt_virtual
4642 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4643 slow_call = new CallStaticJavaNode(C, tf,
4644 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4645 slow_call->set_optimized_virtual(true);
4646 }
4647 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4648 // To be able to issue a direct call (optimized virtual or virtual)
4649 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4650 // about the method being invoked should be attached to the call site to
4651 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4652 slow_call->set_override_symbolic_info(true);
4653 }
4654 set_arguments_for_java_call(slow_call);
4655 set_edges_for_java_call(slow_call);
4656 return slow_call;
4657 }
4658
4659
4660 /**
4661 * Build special case code for calls to hashCode on an object. This call may
4662 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4663 * slightly different code.
4664 */
4665 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4666 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4667 assert(!(is_virtual && is_static), "either virtual, special, or static");
4668
4669 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4670
4671 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4672 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4673 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4674 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4675 Node* obj = nullptr;
4676 if (!is_static) {
4677 // Check for hashing null object
4678 obj = null_check_receiver();
4679 if (stopped()) return true; // unconditionally null
4680 result_reg->init_req(_null_path, top());
4681 result_val->init_req(_null_path, top());
4682 } else {
4683 // Do a null check, and return zero if null.
4684 // System.identityHashCode(null) == 0
4685 obj = argument(0);
4686 Node* null_ctl = top();
4687 obj = null_check_oop(obj, &null_ctl);
4688 result_reg->init_req(_null_path, null_ctl);
4689 result_val->init_req(_null_path, _gvn.intcon(0));
4690 }
4691
4692 // Unconditionally null? Then return right away.
4693 if (stopped()) {
4694 set_control( result_reg->in(_null_path));
4695 if (!stopped())
4696 set_result(result_val->in(_null_path));
4697 return true;
4698 }
4699
4700 // We only go to the fast case code if we pass a number of guards. The
4701 // paths which do not pass are accumulated in the slow_region.
4702 RegionNode* slow_region = new RegionNode(1);
4703 record_for_igvn(slow_region);
4704
4705 // If this is a virtual call, we generate a funny guard. We pull out
4706 // the vtable entry corresponding to hashCode() from the target object.
4707 // If the target method which we are calling happens to be the native
4708 // Object hashCode() method, we pass the guard. We do not need this
4709 // guard for non-virtual calls -- the caller is known to be the native
4710 // Object hashCode().
4711 if (is_virtual) {
4712 // After null check, get the object's klass.
4713 Node* obj_klass = load_object_klass(obj);
4714 generate_virtual_guard(obj_klass, slow_region);
4715 }
4716
4717 // Get the header out of the object, use LoadMarkNode when available
4718 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4719 // The control of the load must be null. Otherwise, the load can move before
4720 // the null check after castPP removal.
4721 Node* no_ctrl = nullptr;
4722 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4723
4724 if (!UseObjectMonitorTable) {
4725 // Test the header to see if it is safe to read w.r.t. locking.
4726 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4727 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4728 if (LockingMode == LM_LIGHTWEIGHT) {
4729 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
4730 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
4731 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
4732
4733 generate_slow_guard(test_monitor, slow_region);
4734 } else {
4735 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4736 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val));
4737 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne));
4738
4739 generate_slow_guard(test_not_unlocked, slow_region);
4740 }
4741 }
4742
4743 // Get the hash value and check to see that it has been properly assigned.
4744 // We depend on hash_mask being at most 32 bits and avoid the use of
4745 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4746 // vm: see markWord.hpp.
4747 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4748 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4749 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4750 // This hack lets the hash bits live anywhere in the mark object now, as long
4751 // as the shift drops the relevant bits into the low 32 bits. Note that
4752 // Java spec says that HashCode is an int so there's no point in capturing
4753 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4754 hshifted_header = ConvX2I(hshifted_header);
4755 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4756
4757 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4758 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4759 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4760
4761 generate_slow_guard(test_assigned, slow_region);
4762
4763 Node* init_mem = reset_memory();
4764 // fill in the rest of the null path:
4765 result_io ->init_req(_null_path, i_o());
4766 result_mem->init_req(_null_path, init_mem);
4767
4768 result_val->init_req(_fast_path, hash_val);
4769 result_reg->init_req(_fast_path, control());
4770 result_io ->init_req(_fast_path, i_o());
4771 result_mem->init_req(_fast_path, init_mem);
4772
4773 // Generate code for the slow case. We make a call to hashCode().
4774 set_control(_gvn.transform(slow_region));
4775 if (!stopped()) {
4776 // No need for PreserveJVMState, because we're using up the present state.
4777 set_all_memory(init_mem);
4778 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4779 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4780 Node* slow_result = set_results_for_java_call(slow_call);
4781 // this->control() comes from set_results_for_java_call
4782 result_reg->init_req(_slow_path, control());
4783 result_val->init_req(_slow_path, slow_result);
4784 result_io ->set_req(_slow_path, i_o());
4785 result_mem ->set_req(_slow_path, reset_memory());
4786 }
4787
4788 // Return the combined state.
4789 set_i_o( _gvn.transform(result_io) );
4790 set_all_memory( _gvn.transform(result_mem));
4791
4792 set_result(result_reg, result_val);
4793 return true;
4794 }
4795
4796 //---------------------------inline_native_getClass----------------------------
4797 // public final native Class<?> java.lang.Object.getClass();
4798 //
4799 // Build special case code for calls to getClass on an object.
4800 bool LibraryCallKit::inline_native_getClass() {
4801 Node* obj = null_check_receiver();
4802 if (stopped()) return true;
4803 set_result(load_mirror_from_klass(load_object_klass(obj)));
4804 return true;
4805 }
4806
4807 //-----------------inline_native_Reflection_getCallerClass---------------------
4808 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4809 //
4810 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4811 //
4812 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4813 // in that it must skip particular security frames and checks for
4814 // caller sensitive methods.
4815 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4816 #ifndef PRODUCT
4817 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4818 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4819 }
4820 #endif
4821
4822 if (!jvms()->has_method()) {
4823 #ifndef PRODUCT
4824 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4825 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4826 }
4827 #endif
4828 return false;
4829 }
4830
4831 // Walk back up the JVM state to find the caller at the required
4832 // depth.
4833 JVMState* caller_jvms = jvms();
4834
4835 // Cf. JVM_GetCallerClass
4836 // NOTE: Start the loop at depth 1 because the current JVM state does
4837 // not include the Reflection.getCallerClass() frame.
4838 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4839 ciMethod* m = caller_jvms->method();
4840 switch (n) {
4841 case 0:
4842 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4843 break;
4844 case 1:
4845 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4846 if (!m->caller_sensitive()) {
4847 #ifndef PRODUCT
4848 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4849 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4850 }
4851 #endif
4852 return false; // bail-out; let JVM_GetCallerClass do the work
4853 }
4854 break;
4855 default:
4856 if (!m->is_ignored_by_security_stack_walk()) {
4857 // We have reached the desired frame; return the holder class.
4858 // Acquire method holder as java.lang.Class and push as constant.
4859 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4860 ciInstance* caller_mirror = caller_klass->java_mirror();
4861 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4862
4863 #ifndef PRODUCT
4864 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4865 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());
4866 tty->print_cr(" JVM state at this point:");
4867 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4868 ciMethod* m = jvms()->of_depth(i)->method();
4869 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4870 }
4871 }
4872 #endif
4873 return true;
4874 }
4875 break;
4876 }
4877 }
4878
4879 #ifndef PRODUCT
4880 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4881 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4882 tty->print_cr(" JVM state at this point:");
4883 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4884 ciMethod* m = jvms()->of_depth(i)->method();
4885 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4886 }
4887 }
4888 #endif
4889
4890 return false; // bail-out; let JVM_GetCallerClass do the work
4891 }
4892
4893 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4894 Node* arg = argument(0);
4895 Node* result = nullptr;
4896
4897 switch (id) {
4898 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4899 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4900 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4901 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4902 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
4903 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
4904
4905 case vmIntrinsics::_doubleToLongBits: {
4906 // two paths (plus control) merge in a wood
4907 RegionNode *r = new RegionNode(3);
4908 Node *phi = new PhiNode(r, TypeLong::LONG);
4909
4910 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4911 // Build the boolean node
4912 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4913
4914 // Branch either way.
4915 // NaN case is less traveled, which makes all the difference.
4916 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4917 Node *opt_isnan = _gvn.transform(ifisnan);
4918 assert( opt_isnan->is_If(), "Expect an IfNode");
4919 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4920 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4921
4922 set_control(iftrue);
4923
4924 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4925 Node *slow_result = longcon(nan_bits); // return NaN
4926 phi->init_req(1, _gvn.transform( slow_result ));
4927 r->init_req(1, iftrue);
4928
4929 // Else fall through
4930 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4931 set_control(iffalse);
4932
4933 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4934 r->init_req(2, iffalse);
4935
4936 // Post merge
4937 set_control(_gvn.transform(r));
4938 record_for_igvn(r);
4939
4940 C->set_has_split_ifs(true); // Has chance for split-if optimization
4941 result = phi;
4942 assert(result->bottom_type()->isa_long(), "must be");
4943 break;
4944 }
4945
4946 case vmIntrinsics::_floatToIntBits: {
4947 // two paths (plus control) merge in a wood
4948 RegionNode *r = new RegionNode(3);
4949 Node *phi = new PhiNode(r, TypeInt::INT);
4950
4951 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4952 // Build the boolean node
4953 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4954
4955 // Branch either way.
4956 // NaN case is less traveled, which makes all the difference.
4957 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4958 Node *opt_isnan = _gvn.transform(ifisnan);
4959 assert( opt_isnan->is_If(), "Expect an IfNode");
4960 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4961 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4962
4963 set_control(iftrue);
4964
4965 static const jint nan_bits = 0x7fc00000;
4966 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4967 phi->init_req(1, _gvn.transform( slow_result ));
4968 r->init_req(1, iftrue);
4969
4970 // Else fall through
4971 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4972 set_control(iffalse);
4973
4974 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4975 r->init_req(2, iffalse);
4976
4977 // Post merge
4978 set_control(_gvn.transform(r));
4979 record_for_igvn(r);
4980
4981 C->set_has_split_ifs(true); // Has chance for split-if optimization
4982 result = phi;
4983 assert(result->bottom_type()->isa_int(), "must be");
4984 break;
4985 }
4986
4987 default:
4988 fatal_unexpected_iid(id);
4989 break;
4990 }
4991 set_result(_gvn.transform(result));
4992 return true;
4993 }
4994
4995 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
4996 Node* arg = argument(0);
4997 Node* result = nullptr;
4998
4999 switch (id) {
5000 case vmIntrinsics::_floatIsInfinite:
5001 result = new IsInfiniteFNode(arg);
5002 break;
5003 case vmIntrinsics::_floatIsFinite:
5004 result = new IsFiniteFNode(arg);
5005 break;
5006 case vmIntrinsics::_doubleIsInfinite:
5007 result = new IsInfiniteDNode(arg);
5008 break;
5009 case vmIntrinsics::_doubleIsFinite:
5010 result = new IsFiniteDNode(arg);
5011 break;
5012 default:
5013 fatal_unexpected_iid(id);
5014 break;
5015 }
5016 set_result(_gvn.transform(result));
5017 return true;
5018 }
5019
5020 //----------------------inline_unsafe_copyMemory-------------------------
5021 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5022
5023 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5024 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5025 const Type* base_t = gvn.type(base);
5026
5027 bool in_native = (base_t == TypePtr::NULL_PTR);
5028 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5029 bool is_mixed = !in_heap && !in_native;
5030
5031 if (is_mixed) {
5032 return true; // mixed accesses can touch both on-heap and off-heap memory
5033 }
5034 if (in_heap) {
5035 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5036 if (!is_prim_array) {
5037 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5038 // there's not enough type information available to determine proper memory slice for it.
5039 return true;
5040 }
5041 }
5042 return false;
5043 }
5044
5045 bool LibraryCallKit::inline_unsafe_copyMemory() {
5046 if (callee()->is_static()) return false; // caller must have the capability!
5047 null_check_receiver(); // null-check receiver
5048 if (stopped()) return true;
5049
5050 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5051
5052 Node* src_base = argument(1); // type: oop
5053 Node* src_off = ConvL2X(argument(2)); // type: long
5054 Node* dst_base = argument(4); // type: oop
5055 Node* dst_off = ConvL2X(argument(5)); // type: long
5056 Node* size = ConvL2X(argument(7)); // type: long
5057
5058 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5059 "fieldOffset must be byte-scaled");
5060
5061 Node* src_addr = make_unsafe_address(src_base, src_off);
5062 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5063
5064 Node* thread = _gvn.transform(new ThreadLocalNode());
5065 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5066 BasicType doing_unsafe_access_bt = T_BYTE;
5067 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5068
5069 // update volatile field
5070 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5071
5072 int flags = RC_LEAF | RC_NO_FP;
5073
5074 const TypePtr* dst_type = TypePtr::BOTTOM;
5075
5076 // Adjust memory effects of the runtime call based on input values.
5077 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5078 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5079 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5080
5081 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5082 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5083 flags |= RC_NARROW_MEM; // narrow in memory
5084 }
5085 }
5086
5087 // Call it. Note that the length argument is not scaled.
5088 make_runtime_call(flags,
5089 OptoRuntime::fast_arraycopy_Type(),
5090 StubRoutines::unsafe_arraycopy(),
5091 "unsafe_arraycopy",
5092 dst_type,
5093 src_addr, dst_addr, size XTOP);
5094
5095 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5096
5097 return true;
5098 }
5099
5100 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5101 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5102 bool LibraryCallKit::inline_unsafe_setMemory() {
5103 if (callee()->is_static()) return false; // caller must have the capability!
5104 null_check_receiver(); // null-check receiver
5105 if (stopped()) return true;
5106
5107 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5108
5109 Node* dst_base = argument(1); // type: oop
5110 Node* dst_off = ConvL2X(argument(2)); // type: long
5111 Node* size = ConvL2X(argument(4)); // type: long
5112 Node* byte = argument(6); // type: byte
5113
5114 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5115 "fieldOffset must be byte-scaled");
5116
5117 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5118
5119 Node* thread = _gvn.transform(new ThreadLocalNode());
5120 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5121 BasicType doing_unsafe_access_bt = T_BYTE;
5122 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5123
5124 // update volatile field
5125 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5126
5127 int flags = RC_LEAF | RC_NO_FP;
5128
5129 const TypePtr* dst_type = TypePtr::BOTTOM;
5130
5131 // Adjust memory effects of the runtime call based on input values.
5132 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5133 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5134
5135 flags |= RC_NARROW_MEM; // narrow in memory
5136 }
5137
5138 // Call it. Note that the length argument is not scaled.
5139 make_runtime_call(flags,
5140 OptoRuntime::make_setmemory_Type(),
5141 StubRoutines::unsafe_setmemory(),
5142 "unsafe_setmemory",
5143 dst_type,
5144 dst_addr, size XTOP, byte);
5145
5146 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5147
5148 return true;
5149 }
5150
5151 #undef XTOP
5152
5153 //------------------------clone_coping-----------------------------------
5154 // Helper function for inline_native_clone.
5155 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5156 assert(obj_size != nullptr, "");
5157 Node* raw_obj = alloc_obj->in(1);
5158 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5159
5160 AllocateNode* alloc = nullptr;
5161 if (ReduceBulkZeroing &&
5162 // If we are implementing an array clone without knowing its source type
5163 // (can happen when compiling the array-guarded branch of a reflective
5164 // Object.clone() invocation), initialize the array within the allocation.
5165 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5166 // to a runtime clone call that assumes fully initialized source arrays.
5167 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5168 // We will be completely responsible for initializing this object -
5169 // mark Initialize node as complete.
5170 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5171 // The object was just allocated - there should be no any stores!
5172 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5173 // Mark as complete_with_arraycopy so that on AllocateNode
5174 // expansion, we know this AllocateNode is initialized by an array
5175 // copy and a StoreStore barrier exists after the array copy.
5176 alloc->initialization()->set_complete_with_arraycopy();
5177 }
5178
5179 Node* size = _gvn.transform(obj_size);
5180 access_clone(obj, alloc_obj, size, is_array);
5181
5182 // Do not let reads from the cloned object float above the arraycopy.
5183 if (alloc != nullptr) {
5184 // Do not let stores that initialize this object be reordered with
5185 // a subsequent store that would make this object accessible by
5186 // other threads.
5187 // Record what AllocateNode this StoreStore protects so that
5188 // escape analysis can go from the MemBarStoreStoreNode to the
5189 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5190 // based on the escape status of the AllocateNode.
5191 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5192 } else {
5193 insert_mem_bar(Op_MemBarCPUOrder);
5194 }
5195 }
5196
5197 //------------------------inline_native_clone----------------------------
5198 // protected native Object java.lang.Object.clone();
5199 //
5200 // Here are the simple edge cases:
5201 // null receiver => normal trap
5202 // virtual and clone was overridden => slow path to out-of-line clone
5203 // not cloneable or finalizer => slow path to out-of-line Object.clone
5204 //
5205 // The general case has two steps, allocation and copying.
5206 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5207 //
5208 // Copying also has two cases, oop arrays and everything else.
5209 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5210 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5211 //
5212 // These steps fold up nicely if and when the cloned object's klass
5213 // can be sharply typed as an object array, a type array, or an instance.
5214 //
5215 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5216 PhiNode* result_val;
5217
5218 // Set the reexecute bit for the interpreter to reexecute
5219 // the bytecode that invokes Object.clone if deoptimization happens.
5220 { PreserveReexecuteState preexecs(this);
5221 jvms()->set_should_reexecute(true);
5222
5223 Node* obj = null_check_receiver();
5224 if (stopped()) return true;
5225
5226 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5227
5228 // If we are going to clone an instance, we need its exact type to
5229 // know the number and types of fields to convert the clone to
5230 // loads/stores. Maybe a speculative type can help us.
5231 if (!obj_type->klass_is_exact() &&
5232 obj_type->speculative_type() != nullptr &&
5233 obj_type->speculative_type()->is_instance_klass()) {
5234 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5235 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5236 !spec_ik->has_injected_fields()) {
5237 if (!obj_type->isa_instptr() ||
5238 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5239 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5240 }
5241 }
5242 }
5243
5244 // Conservatively insert a memory barrier on all memory slices.
5245 // Do not let writes into the original float below the clone.
5246 insert_mem_bar(Op_MemBarCPUOrder);
5247
5248 // paths into result_reg:
5249 enum {
5250 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5251 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5252 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5253 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5254 PATH_LIMIT
5255 };
5256 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5257 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5258 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5259 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5260 record_for_igvn(result_reg);
5261
5262 Node* obj_klass = load_object_klass(obj);
5263 Node* array_obj = obj;
5264 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5265 if (array_ctl != nullptr) {
5266 // It's an array.
5267 PreserveJVMState pjvms(this);
5268 set_control(array_ctl);
5269 Node* obj_length = load_array_length(array_obj);
5270 Node* array_size = nullptr; // Size of the array without object alignment padding.
5271 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5272
5273 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5274 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5275 // If it is an oop array, it requires very special treatment,
5276 // because gc barriers are required when accessing the array.
5277 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5278 if (is_obja != nullptr) {
5279 PreserveJVMState pjvms2(this);
5280 set_control(is_obja);
5281 // Generate a direct call to the right arraycopy function(s).
5282 // Clones are always tightly coupled.
5283 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5284 ac->set_clone_oop_array();
5285 Node* n = _gvn.transform(ac);
5286 assert(n == ac, "cannot disappear");
5287 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5288
5289 result_reg->init_req(_objArray_path, control());
5290 result_val->init_req(_objArray_path, alloc_obj);
5291 result_i_o ->set_req(_objArray_path, i_o());
5292 result_mem ->set_req(_objArray_path, reset_memory());
5293 }
5294 }
5295 // Otherwise, there are no barriers to worry about.
5296 // (We can dispense with card marks if we know the allocation
5297 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5298 // causes the non-eden paths to take compensating steps to
5299 // simulate a fresh allocation, so that no further
5300 // card marks are required in compiled code to initialize
5301 // the object.)
5302
5303 if (!stopped()) {
5304 copy_to_clone(array_obj, alloc_obj, array_size, true);
5305
5306 // Present the results of the copy.
5307 result_reg->init_req(_array_path, control());
5308 result_val->init_req(_array_path, alloc_obj);
5309 result_i_o ->set_req(_array_path, i_o());
5310 result_mem ->set_req(_array_path, reset_memory());
5311 }
5312 }
5313
5314 // We only go to the instance fast case code if we pass a number of guards.
5315 // The paths which do not pass are accumulated in the slow_region.
5316 RegionNode* slow_region = new RegionNode(1);
5317 record_for_igvn(slow_region);
5318 if (!stopped()) {
5319 // It's an instance (we did array above). Make the slow-path tests.
5320 // If this is a virtual call, we generate a funny guard. We grab
5321 // the vtable entry corresponding to clone() from the target object.
5322 // If the target method which we are calling happens to be the
5323 // Object clone() method, we pass the guard. We do not need this
5324 // guard for non-virtual calls; the caller is known to be the native
5325 // Object clone().
5326 if (is_virtual) {
5327 generate_virtual_guard(obj_klass, slow_region);
5328 }
5329
5330 // The object must be easily cloneable and must not have a finalizer.
5331 // Both of these conditions may be checked in a single test.
5332 // We could optimize the test further, but we don't care.
5333 generate_misc_flags_guard(obj_klass,
5334 // Test both conditions:
5335 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5336 // Must be cloneable but not finalizer:
5337 KlassFlags::_misc_is_cloneable_fast,
5338 slow_region);
5339 }
5340
5341 if (!stopped()) {
5342 // It's an instance, and it passed the slow-path tests.
5343 PreserveJVMState pjvms(this);
5344 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5345 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5346 // is reexecuted if deoptimization occurs and there could be problems when merging
5347 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5348 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5349
5350 copy_to_clone(obj, alloc_obj, obj_size, false);
5351
5352 // Present the results of the slow call.
5353 result_reg->init_req(_instance_path, control());
5354 result_val->init_req(_instance_path, alloc_obj);
5355 result_i_o ->set_req(_instance_path, i_o());
5356 result_mem ->set_req(_instance_path, reset_memory());
5357 }
5358
5359 // Generate code for the slow case. We make a call to clone().
5360 set_control(_gvn.transform(slow_region));
5361 if (!stopped()) {
5362 PreserveJVMState pjvms(this);
5363 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5364 // We need to deoptimize on exception (see comment above)
5365 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5366 // this->control() comes from set_results_for_java_call
5367 result_reg->init_req(_slow_path, control());
5368 result_val->init_req(_slow_path, slow_result);
5369 result_i_o ->set_req(_slow_path, i_o());
5370 result_mem ->set_req(_slow_path, reset_memory());
5371 }
5372
5373 // Return the combined state.
5374 set_control( _gvn.transform(result_reg));
5375 set_i_o( _gvn.transform(result_i_o));
5376 set_all_memory( _gvn.transform(result_mem));
5377 } // original reexecute is set back here
5378
5379 set_result(_gvn.transform(result_val));
5380 return true;
5381 }
5382
5383 // If we have a tightly coupled allocation, the arraycopy may take care
5384 // of the array initialization. If one of the guards we insert between
5385 // the allocation and the arraycopy causes a deoptimization, an
5386 // uninitialized array will escape the compiled method. To prevent that
5387 // we set the JVM state for uncommon traps between the allocation and
5388 // the arraycopy to the state before the allocation so, in case of
5389 // deoptimization, we'll reexecute the allocation and the
5390 // initialization.
5391 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5392 if (alloc != nullptr) {
5393 ciMethod* trap_method = alloc->jvms()->method();
5394 int trap_bci = alloc->jvms()->bci();
5395
5396 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5397 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5398 // Make sure there's no store between the allocation and the
5399 // arraycopy otherwise visible side effects could be rexecuted
5400 // in case of deoptimization and cause incorrect execution.
5401 bool no_interfering_store = true;
5402 Node* mem = alloc->in(TypeFunc::Memory);
5403 if (mem->is_MergeMem()) {
5404 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5405 Node* n = mms.memory();
5406 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5407 assert(n->is_Store(), "what else?");
5408 no_interfering_store = false;
5409 break;
5410 }
5411 }
5412 } else {
5413 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5414 Node* n = mms.memory();
5415 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5416 assert(n->is_Store(), "what else?");
5417 no_interfering_store = false;
5418 break;
5419 }
5420 }
5421 }
5422
5423 if (no_interfering_store) {
5424 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5425
5426 JVMState* saved_jvms = jvms();
5427 saved_reexecute_sp = _reexecute_sp;
5428
5429 set_jvms(sfpt->jvms());
5430 _reexecute_sp = jvms()->sp();
5431
5432 return saved_jvms;
5433 }
5434 }
5435 }
5436 return nullptr;
5437 }
5438
5439 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5440 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5441 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5442 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5443 uint size = alloc->req();
5444 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5445 old_jvms->set_map(sfpt);
5446 for (uint i = 0; i < size; i++) {
5447 sfpt->init_req(i, alloc->in(i));
5448 }
5449 // re-push array length for deoptimization
5450 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5451 old_jvms->set_sp(old_jvms->sp()+1);
5452 old_jvms->set_monoff(old_jvms->monoff()+1);
5453 old_jvms->set_scloff(old_jvms->scloff()+1);
5454 old_jvms->set_endoff(old_jvms->endoff()+1);
5455 old_jvms->set_should_reexecute(true);
5456
5457 sfpt->set_i_o(map()->i_o());
5458 sfpt->set_memory(map()->memory());
5459 sfpt->set_control(map()->control());
5460 return sfpt;
5461 }
5462
5463 // In case of a deoptimization, we restart execution at the
5464 // allocation, allocating a new array. We would leave an uninitialized
5465 // array in the heap that GCs wouldn't expect. Move the allocation
5466 // after the traps so we don't allocate the array if we
5467 // deoptimize. This is possible because tightly_coupled_allocation()
5468 // guarantees there's no observer of the allocated array at this point
5469 // and the control flow is simple enough.
5470 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5471 int saved_reexecute_sp, uint new_idx) {
5472 if (saved_jvms_before_guards != nullptr && !stopped()) {
5473 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5474
5475 assert(alloc != nullptr, "only with a tightly coupled allocation");
5476 // restore JVM state to the state at the arraycopy
5477 saved_jvms_before_guards->map()->set_control(map()->control());
5478 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5479 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5480 // If we've improved the types of some nodes (null check) while
5481 // emitting the guards, propagate them to the current state
5482 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5483 set_jvms(saved_jvms_before_guards);
5484 _reexecute_sp = saved_reexecute_sp;
5485
5486 // Remove the allocation from above the guards
5487 CallProjections callprojs;
5488 alloc->extract_projections(&callprojs, true);
5489 InitializeNode* init = alloc->initialization();
5490 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5491 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5492 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5493
5494 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5495 // the allocation (i.e. is only valid if the allocation succeeds):
5496 // 1) replace CastIINode with AllocateArrayNode's length here
5497 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5498 //
5499 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5500 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5501 Node* init_control = init->proj_out(TypeFunc::Control);
5502 Node* alloc_length = alloc->Ideal_length();
5503 #ifdef ASSERT
5504 Node* prev_cast = nullptr;
5505 #endif
5506 for (uint i = 0; i < init_control->outcnt(); i++) {
5507 Node* init_out = init_control->raw_out(i);
5508 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5509 #ifdef ASSERT
5510 if (prev_cast == nullptr) {
5511 prev_cast = init_out;
5512 } else {
5513 if (prev_cast->cmp(*init_out) == false) {
5514 prev_cast->dump();
5515 init_out->dump();
5516 assert(false, "not equal CastIINode");
5517 }
5518 }
5519 #endif
5520 C->gvn_replace_by(init_out, alloc_length);
5521 }
5522 }
5523 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5524
5525 // move the allocation here (after the guards)
5526 _gvn.hash_delete(alloc);
5527 alloc->set_req(TypeFunc::Control, control());
5528 alloc->set_req(TypeFunc::I_O, i_o());
5529 Node *mem = reset_memory();
5530 set_all_memory(mem);
5531 alloc->set_req(TypeFunc::Memory, mem);
5532 set_control(init->proj_out_or_null(TypeFunc::Control));
5533 set_i_o(callprojs.fallthrough_ioproj);
5534
5535 // Update memory as done in GraphKit::set_output_for_allocation()
5536 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5537 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5538 if (ary_type->isa_aryptr() && length_type != nullptr) {
5539 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5540 }
5541 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5542 int elemidx = C->get_alias_index(telemref);
5543 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5544 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5545
5546 Node* allocx = _gvn.transform(alloc);
5547 assert(allocx == alloc, "where has the allocation gone?");
5548 assert(dest->is_CheckCastPP(), "not an allocation result?");
5549
5550 _gvn.hash_delete(dest);
5551 dest->set_req(0, control());
5552 Node* destx = _gvn.transform(dest);
5553 assert(destx == dest, "where has the allocation result gone?");
5554
5555 array_ideal_length(alloc, ary_type, true);
5556 }
5557 }
5558
5559 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5560 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5561 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5562 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5563 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5564 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5565 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5566 JVMState* saved_jvms_before_guards) {
5567 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5568 // There is at least one unrelated uncommon trap which needs to be replaced.
5569 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5570
5571 JVMState* saved_jvms = jvms();
5572 const int saved_reexecute_sp = _reexecute_sp;
5573 set_jvms(sfpt->jvms());
5574 _reexecute_sp = jvms()->sp();
5575
5576 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5577
5578 // Restore state
5579 set_jvms(saved_jvms);
5580 _reexecute_sp = saved_reexecute_sp;
5581 }
5582 }
5583
5584 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5585 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5586 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5587 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5588 while (if_proj->is_IfProj()) {
5589 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5590 if (uncommon_trap != nullptr) {
5591 create_new_uncommon_trap(uncommon_trap);
5592 }
5593 assert(if_proj->in(0)->is_If(), "must be If");
5594 if_proj = if_proj->in(0)->in(0);
5595 }
5596 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5597 "must have reached control projection of init node");
5598 }
5599
5600 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5601 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5602 assert(trap_request != 0, "no valid UCT trap request");
5603 PreserveJVMState pjvms(this);
5604 set_control(uncommon_trap_call->in(0));
5605 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5606 Deoptimization::trap_request_action(trap_request));
5607 assert(stopped(), "Should be stopped");
5608 _gvn.hash_delete(uncommon_trap_call);
5609 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5610 }
5611
5612 // Common checks for array sorting intrinsics arguments.
5613 // Returns `true` if checks passed.
5614 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
5615 // check address of the class
5616 if (elementType == nullptr || elementType->is_top()) {
5617 return false; // dead path
5618 }
5619 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5620 if (elem_klass == nullptr) {
5621 return false; // dead path
5622 }
5623 // java_mirror_type() returns non-null for compile-time Class constants only
5624 ciType* elem_type = elem_klass->java_mirror_type();
5625 if (elem_type == nullptr) {
5626 return false;
5627 }
5628 bt = elem_type->basic_type();
5629 // Disable the intrinsic if the CPU does not support SIMD sort
5630 if (!Matcher::supports_simd_sort(bt)) {
5631 return false;
5632 }
5633 // check address of the array
5634 if (obj == nullptr || obj->is_top()) {
5635 return false; // dead path
5636 }
5637 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5638 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
5639 return false; // failed input validation
5640 }
5641 return true;
5642 }
5643
5644 //------------------------------inline_array_partition-----------------------
5645 bool LibraryCallKit::inline_array_partition() {
5646 address stubAddr = StubRoutines::select_array_partition_function();
5647 if (stubAddr == nullptr) {
5648 return false; // Intrinsic's stub is not implemented on this platform
5649 }
5650 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
5651
5652 // no receiver because it is a static method
5653 Node* elementType = argument(0);
5654 Node* obj = argument(1);
5655 Node* offset = argument(2); // long
5656 Node* fromIndex = argument(4);
5657 Node* toIndex = argument(5);
5658 Node* indexPivot1 = argument(6);
5659 Node* indexPivot2 = argument(7);
5660 // PartitionOperation: argument(8) is ignored
5661
5662 Node* pivotIndices = nullptr;
5663 BasicType bt = T_ILLEGAL;
5664
5665 if (!check_array_sort_arguments(elementType, obj, bt)) {
5666 return false;
5667 }
5668 null_check(obj);
5669 // If obj is dead, only null-path is taken.
5670 if (stopped()) {
5671 return true;
5672 }
5673 // Set the original stack and the reexecute bit for the interpreter to reexecute
5674 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
5675 { PreserveReexecuteState preexecs(this);
5676 jvms()->set_should_reexecute(true);
5677
5678 Node* obj_adr = make_unsafe_address(obj, offset);
5679
5680 // create the pivotIndices array of type int and size = 2
5681 Node* size = intcon(2);
5682 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
5683 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
5684 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
5685 guarantee(alloc != nullptr, "created above");
5686 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
5687
5688 // pass the basic type enum to the stub
5689 Node* elemType = intcon(bt);
5690
5691 // Call the stub
5692 const char *stubName = "array_partition_stub";
5693 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
5694 stubAddr, stubName, TypePtr::BOTTOM,
5695 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
5696 indexPivot1, indexPivot2);
5697
5698 } // original reexecute is set back here
5699
5700 if (!stopped()) {
5701 set_result(pivotIndices);
5702 }
5703
5704 return true;
5705 }
5706
5707
5708 //------------------------------inline_array_sort-----------------------
5709 bool LibraryCallKit::inline_array_sort() {
5710 address stubAddr = StubRoutines::select_arraysort_function();
5711 if (stubAddr == nullptr) {
5712 return false; // Intrinsic's stub is not implemented on this platform
5713 }
5714 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
5715
5716 // no receiver because it is a static method
5717 Node* elementType = argument(0);
5718 Node* obj = argument(1);
5719 Node* offset = argument(2); // long
5720 Node* fromIndex = argument(4);
5721 Node* toIndex = argument(5);
5722 // SortOperation: argument(6) is ignored
5723
5724 BasicType bt = T_ILLEGAL;
5725
5726 if (!check_array_sort_arguments(elementType, obj, bt)) {
5727 return false;
5728 }
5729 null_check(obj);
5730 // If obj is dead, only null-path is taken.
5731 if (stopped()) {
5732 return true;
5733 }
5734 Node* obj_adr = make_unsafe_address(obj, offset);
5735
5736 // pass the basic type enum to the stub
5737 Node* elemType = intcon(bt);
5738
5739 // Call the stub.
5740 const char *stubName = "arraysort_stub";
5741 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
5742 stubAddr, stubName, TypePtr::BOTTOM,
5743 obj_adr, elemType, fromIndex, toIndex);
5744
5745 return true;
5746 }
5747
5748
5749 //------------------------------inline_arraycopy-----------------------
5750 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5751 // Object dest, int destPos,
5752 // int length);
5753 bool LibraryCallKit::inline_arraycopy() {
5754 // Get the arguments.
5755 Node* src = argument(0); // type: oop
5756 Node* src_offset = argument(1); // type: int
5757 Node* dest = argument(2); // type: oop
5758 Node* dest_offset = argument(3); // type: int
5759 Node* length = argument(4); // type: int
5760
5761 uint new_idx = C->unique();
5762
5763 // Check for allocation before we add nodes that would confuse
5764 // tightly_coupled_allocation()
5765 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5766
5767 int saved_reexecute_sp = -1;
5768 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5769 // See arraycopy_restore_alloc_state() comment
5770 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5771 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5772 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5773 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5774
5775 // The following tests must be performed
5776 // (1) src and dest are arrays.
5777 // (2) src and dest arrays must have elements of the same BasicType
5778 // (3) src and dest must not be null.
5779 // (4) src_offset must not be negative.
5780 // (5) dest_offset must not be negative.
5781 // (6) length must not be negative.
5782 // (7) src_offset + length must not exceed length of src.
5783 // (8) dest_offset + length must not exceed length of dest.
5784 // (9) each element of an oop array must be assignable
5785
5786 // (3) src and dest must not be null.
5787 // always do this here because we need the JVM state for uncommon traps
5788 Node* null_ctl = top();
5789 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5790 assert(null_ctl->is_top(), "no null control here");
5791 dest = null_check(dest, T_ARRAY);
5792
5793 if (!can_emit_guards) {
5794 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5795 // guards but the arraycopy node could still take advantage of a
5796 // tightly allocated allocation. tightly_coupled_allocation() is
5797 // called again to make sure it takes the null check above into
5798 // account: the null check is mandatory and if it caused an
5799 // uncommon trap to be emitted then the allocation can't be
5800 // considered tightly coupled in this context.
5801 alloc = tightly_coupled_allocation(dest);
5802 }
5803
5804 bool validated = false;
5805
5806 const Type* src_type = _gvn.type(src);
5807 const Type* dest_type = _gvn.type(dest);
5808 const TypeAryPtr* top_src = src_type->isa_aryptr();
5809 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5810
5811 // Do we have the type of src?
5812 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5813 // Do we have the type of dest?
5814 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5815 // Is the type for src from speculation?
5816 bool src_spec = false;
5817 // Is the type for dest from speculation?
5818 bool dest_spec = false;
5819
5820 if ((!has_src || !has_dest) && can_emit_guards) {
5821 // We don't have sufficient type information, let's see if
5822 // speculative types can help. We need to have types for both src
5823 // and dest so that it pays off.
5824
5825 // Do we already have or could we have type information for src
5826 bool could_have_src = has_src;
5827 // Do we already have or could we have type information for dest
5828 bool could_have_dest = has_dest;
5829
5830 ciKlass* src_k = nullptr;
5831 if (!has_src) {
5832 src_k = src_type->speculative_type_not_null();
5833 if (src_k != nullptr && src_k->is_array_klass()) {
5834 could_have_src = true;
5835 }
5836 }
5837
5838 ciKlass* dest_k = nullptr;
5839 if (!has_dest) {
5840 dest_k = dest_type->speculative_type_not_null();
5841 if (dest_k != nullptr && dest_k->is_array_klass()) {
5842 could_have_dest = true;
5843 }
5844 }
5845
5846 if (could_have_src && could_have_dest) {
5847 // This is going to pay off so emit the required guards
5848 if (!has_src) {
5849 src = maybe_cast_profiled_obj(src, src_k, true);
5850 src_type = _gvn.type(src);
5851 top_src = src_type->isa_aryptr();
5852 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5853 src_spec = true;
5854 }
5855 if (!has_dest) {
5856 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5857 dest_type = _gvn.type(dest);
5858 top_dest = dest_type->isa_aryptr();
5859 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5860 dest_spec = true;
5861 }
5862 }
5863 }
5864
5865 if (has_src && has_dest && can_emit_guards) {
5866 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5867 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5868 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5869 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5870
5871 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5872 // If both arrays are object arrays then having the exact types
5873 // for both will remove the need for a subtype check at runtime
5874 // before the call and may make it possible to pick a faster copy
5875 // routine (without a subtype check on every element)
5876 // Do we have the exact type of src?
5877 bool could_have_src = src_spec;
5878 // Do we have the exact type of dest?
5879 bool could_have_dest = dest_spec;
5880 ciKlass* src_k = nullptr;
5881 ciKlass* dest_k = nullptr;
5882 if (!src_spec) {
5883 src_k = src_type->speculative_type_not_null();
5884 if (src_k != nullptr && src_k->is_array_klass()) {
5885 could_have_src = true;
5886 }
5887 }
5888 if (!dest_spec) {
5889 dest_k = dest_type->speculative_type_not_null();
5890 if (dest_k != nullptr && dest_k->is_array_klass()) {
5891 could_have_dest = true;
5892 }
5893 }
5894 if (could_have_src && could_have_dest) {
5895 // If we can have both exact types, emit the missing guards
5896 if (could_have_src && !src_spec) {
5897 src = maybe_cast_profiled_obj(src, src_k, true);
5898 }
5899 if (could_have_dest && !dest_spec) {
5900 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5901 }
5902 }
5903 }
5904 }
5905
5906 ciMethod* trap_method = method();
5907 int trap_bci = bci();
5908 if (saved_jvms_before_guards != nullptr) {
5909 trap_method = alloc->jvms()->method();
5910 trap_bci = alloc->jvms()->bci();
5911 }
5912
5913 bool negative_length_guard_generated = false;
5914
5915 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5916 can_emit_guards &&
5917 !src->is_top() && !dest->is_top()) {
5918 // validate arguments: enables transformation the ArrayCopyNode
5919 validated = true;
5920
5921 RegionNode* slow_region = new RegionNode(1);
5922 record_for_igvn(slow_region);
5923
5924 // (1) src and dest are arrays.
5925 generate_non_array_guard(load_object_klass(src), slow_region, &src);
5926 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
5927
5928 // (2) src and dest arrays must have elements of the same BasicType
5929 // done at macro expansion or at Ideal transformation time
5930
5931 // (4) src_offset must not be negative.
5932 generate_negative_guard(src_offset, slow_region);
5933
5934 // (5) dest_offset must not be negative.
5935 generate_negative_guard(dest_offset, slow_region);
5936
5937 // (7) src_offset + length must not exceed length of src.
5938 generate_limit_guard(src_offset, length,
5939 load_array_length(src),
5940 slow_region);
5941
5942 // (8) dest_offset + length must not exceed length of dest.
5943 generate_limit_guard(dest_offset, length,
5944 load_array_length(dest),
5945 slow_region);
5946
5947 // (6) length must not be negative.
5948 // This is also checked in generate_arraycopy() during macro expansion, but
5949 // we also have to check it here for the case where the ArrayCopyNode will
5950 // be eliminated by Escape Analysis.
5951 if (EliminateAllocations) {
5952 generate_negative_guard(length, slow_region);
5953 negative_length_guard_generated = true;
5954 }
5955
5956 // (9) each element of an oop array must be assignable
5957 Node* dest_klass = load_object_klass(dest);
5958 if (src != dest) {
5959 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
5960
5961 if (not_subtype_ctrl != top()) {
5962 PreserveJVMState pjvms(this);
5963 set_control(not_subtype_ctrl);
5964 uncommon_trap(Deoptimization::Reason_intrinsic,
5965 Deoptimization::Action_make_not_entrant);
5966 assert(stopped(), "Should be stopped");
5967 }
5968 }
5969 {
5970 PreserveJVMState pjvms(this);
5971 set_control(_gvn.transform(slow_region));
5972 uncommon_trap(Deoptimization::Reason_intrinsic,
5973 Deoptimization::Action_make_not_entrant);
5974 assert(stopped(), "Should be stopped");
5975 }
5976
5977 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5978 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
5979 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5980 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
5981 }
5982
5983 if (stopped()) {
5984 return true;
5985 }
5986
5987 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
5988 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5989 // so the compiler has a chance to eliminate them: during macro expansion,
5990 // we have to set their control (CastPP nodes are eliminated).
5991 load_object_klass(src), load_object_klass(dest),
5992 load_array_length(src), load_array_length(dest));
5993
5994 ac->set_arraycopy(validated);
5995
5996 Node* n = _gvn.transform(ac);
5997 if (n == ac) {
5998 ac->connect_outputs(this);
5999 } else {
6000 assert(validated, "shouldn't transform if all arguments not validated");
6001 set_all_memory(n);
6002 }
6003 clear_upper_avx();
6004
6005
6006 return true;
6007 }
6008
6009
6010 // Helper function which determines if an arraycopy immediately follows
6011 // an allocation, with no intervening tests or other escapes for the object.
6012 AllocateArrayNode*
6013 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6014 if (stopped()) return nullptr; // no fast path
6015 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6016
6017 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6018 if (alloc == nullptr) return nullptr;
6019
6020 Node* rawmem = memory(Compile::AliasIdxRaw);
6021 // Is the allocation's memory state untouched?
6022 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6023 // Bail out if there have been raw-memory effects since the allocation.
6024 // (Example: There might have been a call or safepoint.)
6025 return nullptr;
6026 }
6027 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6028 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6029 return nullptr;
6030 }
6031
6032 // There must be no unexpected observers of this allocation.
6033 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6034 Node* obs = ptr->fast_out(i);
6035 if (obs != this->map()) {
6036 return nullptr;
6037 }
6038 }
6039
6040 // This arraycopy must unconditionally follow the allocation of the ptr.
6041 Node* alloc_ctl = ptr->in(0);
6042 Node* ctl = control();
6043 while (ctl != alloc_ctl) {
6044 // There may be guards which feed into the slow_region.
6045 // Any other control flow means that we might not get a chance
6046 // to finish initializing the allocated object.
6047 // Various low-level checks bottom out in uncommon traps. These
6048 // are considered safe since we've already checked above that
6049 // there is no unexpected observer of this allocation.
6050 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6051 assert(ctl->in(0)->is_If(), "must be If");
6052 ctl = ctl->in(0)->in(0);
6053 } else {
6054 return nullptr;
6055 }
6056 }
6057
6058 // If we get this far, we have an allocation which immediately
6059 // precedes the arraycopy, and we can take over zeroing the new object.
6060 // The arraycopy will finish the initialization, and provide
6061 // a new control state to which we will anchor the destination pointer.
6062
6063 return alloc;
6064 }
6065
6066 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6067 if (node->is_IfProj()) {
6068 Node* other_proj = node->as_IfProj()->other_if_proj();
6069 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6070 Node* obs = other_proj->fast_out(j);
6071 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6072 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6073 return obs->as_CallStaticJava();
6074 }
6075 }
6076 }
6077 return nullptr;
6078 }
6079
6080 //-------------inline_encodeISOArray-----------------------------------
6081 // encode char[] to byte[] in ISO_8859_1 or ASCII
6082 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6083 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6084 // no receiver since it is static method
6085 Node *src = argument(0);
6086 Node *src_offset = argument(1);
6087 Node *dst = argument(2);
6088 Node *dst_offset = argument(3);
6089 Node *length = argument(4);
6090
6091 src = must_be_not_null(src, true);
6092 dst = must_be_not_null(dst, true);
6093
6094 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6095 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6096 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6097 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6098 // failed array check
6099 return false;
6100 }
6101
6102 // Figure out the size and type of the elements we will be copying.
6103 BasicType src_elem = src_type->elem()->array_element_basic_type();
6104 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6105 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6106 return false;
6107 }
6108
6109 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6110 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6111 // 'src_start' points to src array + scaled offset
6112 // 'dst_start' points to dst array + scaled offset
6113
6114 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6115 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6116 enc = _gvn.transform(enc);
6117 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6118 set_memory(res_mem, mtype);
6119 set_result(enc);
6120 clear_upper_avx();
6121
6122 return true;
6123 }
6124
6125 //-------------inline_multiplyToLen-----------------------------------
6126 bool LibraryCallKit::inline_multiplyToLen() {
6127 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6128
6129 address stubAddr = StubRoutines::multiplyToLen();
6130 if (stubAddr == nullptr) {
6131 return false; // Intrinsic's stub is not implemented on this platform
6132 }
6133 const char* stubName = "multiplyToLen";
6134
6135 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6136
6137 // no receiver because it is a static method
6138 Node* x = argument(0);
6139 Node* xlen = argument(1);
6140 Node* y = argument(2);
6141 Node* ylen = argument(3);
6142 Node* z = argument(4);
6143
6144 x = must_be_not_null(x, true);
6145 y = must_be_not_null(y, true);
6146
6147 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6148 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6149 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6150 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6151 // failed array check
6152 return false;
6153 }
6154
6155 BasicType x_elem = x_type->elem()->array_element_basic_type();
6156 BasicType y_elem = y_type->elem()->array_element_basic_type();
6157 if (x_elem != T_INT || y_elem != T_INT) {
6158 return false;
6159 }
6160
6161 Node* x_start = array_element_address(x, intcon(0), x_elem);
6162 Node* y_start = array_element_address(y, intcon(0), y_elem);
6163 // 'x_start' points to x array + scaled xlen
6164 // 'y_start' points to y array + scaled ylen
6165
6166 Node* z_start = array_element_address(z, intcon(0), T_INT);
6167
6168 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6169 OptoRuntime::multiplyToLen_Type(),
6170 stubAddr, stubName, TypePtr::BOTTOM,
6171 x_start, xlen, y_start, ylen, z_start);
6172
6173 C->set_has_split_ifs(true); // Has chance for split-if optimization
6174 set_result(z);
6175 return true;
6176 }
6177
6178 //-------------inline_squareToLen------------------------------------
6179 bool LibraryCallKit::inline_squareToLen() {
6180 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6181
6182 address stubAddr = StubRoutines::squareToLen();
6183 if (stubAddr == nullptr) {
6184 return false; // Intrinsic's stub is not implemented on this platform
6185 }
6186 const char* stubName = "squareToLen";
6187
6188 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6189
6190 Node* x = argument(0);
6191 Node* len = argument(1);
6192 Node* z = argument(2);
6193 Node* zlen = argument(3);
6194
6195 x = must_be_not_null(x, true);
6196 z = must_be_not_null(z, true);
6197
6198 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6199 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6200 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6201 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6202 // failed array check
6203 return false;
6204 }
6205
6206 BasicType x_elem = x_type->elem()->array_element_basic_type();
6207 BasicType z_elem = z_type->elem()->array_element_basic_type();
6208 if (x_elem != T_INT || z_elem != T_INT) {
6209 return false;
6210 }
6211
6212
6213 Node* x_start = array_element_address(x, intcon(0), x_elem);
6214 Node* z_start = array_element_address(z, intcon(0), z_elem);
6215
6216 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6217 OptoRuntime::squareToLen_Type(),
6218 stubAddr, stubName, TypePtr::BOTTOM,
6219 x_start, len, z_start, zlen);
6220
6221 set_result(z);
6222 return true;
6223 }
6224
6225 //-------------inline_mulAdd------------------------------------------
6226 bool LibraryCallKit::inline_mulAdd() {
6227 assert(UseMulAddIntrinsic, "not implemented on this platform");
6228
6229 address stubAddr = StubRoutines::mulAdd();
6230 if (stubAddr == nullptr) {
6231 return false; // Intrinsic's stub is not implemented on this platform
6232 }
6233 const char* stubName = "mulAdd";
6234
6235 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6236
6237 Node* out = argument(0);
6238 Node* in = argument(1);
6239 Node* offset = argument(2);
6240 Node* len = argument(3);
6241 Node* k = argument(4);
6242
6243 in = must_be_not_null(in, true);
6244 out = must_be_not_null(out, true);
6245
6246 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6247 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6248 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6249 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6250 // failed array check
6251 return false;
6252 }
6253
6254 BasicType out_elem = out_type->elem()->array_element_basic_type();
6255 BasicType in_elem = in_type->elem()->array_element_basic_type();
6256 if (out_elem != T_INT || in_elem != T_INT) {
6257 return false;
6258 }
6259
6260 Node* outlen = load_array_length(out);
6261 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6262 Node* out_start = array_element_address(out, intcon(0), out_elem);
6263 Node* in_start = array_element_address(in, intcon(0), in_elem);
6264
6265 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6266 OptoRuntime::mulAdd_Type(),
6267 stubAddr, stubName, TypePtr::BOTTOM,
6268 out_start,in_start, new_offset, len, k);
6269 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6270 set_result(result);
6271 return true;
6272 }
6273
6274 //-------------inline_montgomeryMultiply-----------------------------------
6275 bool LibraryCallKit::inline_montgomeryMultiply() {
6276 address stubAddr = StubRoutines::montgomeryMultiply();
6277 if (stubAddr == nullptr) {
6278 return false; // Intrinsic's stub is not implemented on this platform
6279 }
6280
6281 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6282 const char* stubName = "montgomery_multiply";
6283
6284 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6285
6286 Node* a = argument(0);
6287 Node* b = argument(1);
6288 Node* n = argument(2);
6289 Node* len = argument(3);
6290 Node* inv = argument(4);
6291 Node* m = argument(6);
6292
6293 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6294 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6295 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6296 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6297 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6298 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6299 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6300 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6301 // failed array check
6302 return false;
6303 }
6304
6305 BasicType a_elem = a_type->elem()->array_element_basic_type();
6306 BasicType b_elem = b_type->elem()->array_element_basic_type();
6307 BasicType n_elem = n_type->elem()->array_element_basic_type();
6308 BasicType m_elem = m_type->elem()->array_element_basic_type();
6309 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6310 return false;
6311 }
6312
6313 // Make the call
6314 {
6315 Node* a_start = array_element_address(a, intcon(0), a_elem);
6316 Node* b_start = array_element_address(b, intcon(0), b_elem);
6317 Node* n_start = array_element_address(n, intcon(0), n_elem);
6318 Node* m_start = array_element_address(m, intcon(0), m_elem);
6319
6320 Node* call = make_runtime_call(RC_LEAF,
6321 OptoRuntime::montgomeryMultiply_Type(),
6322 stubAddr, stubName, TypePtr::BOTTOM,
6323 a_start, b_start, n_start, len, inv, top(),
6324 m_start);
6325 set_result(m);
6326 }
6327
6328 return true;
6329 }
6330
6331 bool LibraryCallKit::inline_montgomerySquare() {
6332 address stubAddr = StubRoutines::montgomerySquare();
6333 if (stubAddr == nullptr) {
6334 return false; // Intrinsic's stub is not implemented on this platform
6335 }
6336
6337 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6338 const char* stubName = "montgomery_square";
6339
6340 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6341
6342 Node* a = argument(0);
6343 Node* n = argument(1);
6344 Node* len = argument(2);
6345 Node* inv = argument(3);
6346 Node* m = argument(5);
6347
6348 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6349 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6350 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6351 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6352 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6353 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6354 // failed array check
6355 return false;
6356 }
6357
6358 BasicType a_elem = a_type->elem()->array_element_basic_type();
6359 BasicType n_elem = n_type->elem()->array_element_basic_type();
6360 BasicType m_elem = m_type->elem()->array_element_basic_type();
6361 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6362 return false;
6363 }
6364
6365 // Make the call
6366 {
6367 Node* a_start = array_element_address(a, intcon(0), a_elem);
6368 Node* n_start = array_element_address(n, intcon(0), n_elem);
6369 Node* m_start = array_element_address(m, intcon(0), m_elem);
6370
6371 Node* call = make_runtime_call(RC_LEAF,
6372 OptoRuntime::montgomerySquare_Type(),
6373 stubAddr, stubName, TypePtr::BOTTOM,
6374 a_start, n_start, len, inv, top(),
6375 m_start);
6376 set_result(m);
6377 }
6378
6379 return true;
6380 }
6381
6382 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6383 address stubAddr = nullptr;
6384 const char* stubName = nullptr;
6385
6386 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6387 if (stubAddr == nullptr) {
6388 return false; // Intrinsic's stub is not implemented on this platform
6389 }
6390
6391 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6392
6393 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6394
6395 Node* newArr = argument(0);
6396 Node* oldArr = argument(1);
6397 Node* newIdx = argument(2);
6398 Node* shiftCount = argument(3);
6399 Node* numIter = argument(4);
6400
6401 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6402 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6403 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6404 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6405 return false;
6406 }
6407
6408 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6409 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6410 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6411 return false;
6412 }
6413
6414 // Make the call
6415 {
6416 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6417 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6418
6419 Node* call = make_runtime_call(RC_LEAF,
6420 OptoRuntime::bigIntegerShift_Type(),
6421 stubAddr,
6422 stubName,
6423 TypePtr::BOTTOM,
6424 newArr_start,
6425 oldArr_start,
6426 newIdx,
6427 shiftCount,
6428 numIter);
6429 }
6430
6431 return true;
6432 }
6433
6434 //-------------inline_vectorizedMismatch------------------------------
6435 bool LibraryCallKit::inline_vectorizedMismatch() {
6436 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6437
6438 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6439 Node* obja = argument(0); // Object
6440 Node* aoffset = argument(1); // long
6441 Node* objb = argument(3); // Object
6442 Node* boffset = argument(4); // long
6443 Node* length = argument(6); // int
6444 Node* scale = argument(7); // int
6445
6446 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6447 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6448 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6449 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6450 scale == top()) {
6451 return false; // failed input validation
6452 }
6453
6454 Node* obja_adr = make_unsafe_address(obja, aoffset);
6455 Node* objb_adr = make_unsafe_address(objb, boffset);
6456
6457 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6458 //
6459 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6460 // if (length <= inline_limit) {
6461 // inline_path:
6462 // vmask = VectorMaskGen length
6463 // vload1 = LoadVectorMasked obja, vmask
6464 // vload2 = LoadVectorMasked objb, vmask
6465 // result1 = VectorCmpMasked vload1, vload2, vmask
6466 // } else {
6467 // call_stub_path:
6468 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6469 // }
6470 // exit_block:
6471 // return Phi(result1, result2);
6472 //
6473 enum { inline_path = 1, // input is small enough to process it all at once
6474 stub_path = 2, // input is too large; call into the VM
6475 PATH_LIMIT = 3
6476 };
6477
6478 Node* exit_block = new RegionNode(PATH_LIMIT);
6479 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6480 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6481
6482 Node* call_stub_path = control();
6483
6484 BasicType elem_bt = T_ILLEGAL;
6485
6486 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6487 if (scale_t->is_con()) {
6488 switch (scale_t->get_con()) {
6489 case 0: elem_bt = T_BYTE; break;
6490 case 1: elem_bt = T_SHORT; break;
6491 case 2: elem_bt = T_INT; break;
6492 case 3: elem_bt = T_LONG; break;
6493
6494 default: elem_bt = T_ILLEGAL; break; // not supported
6495 }
6496 }
6497
6498 int inline_limit = 0;
6499 bool do_partial_inline = false;
6500
6501 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6502 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6503 do_partial_inline = inline_limit >= 16;
6504 }
6505
6506 if (do_partial_inline) {
6507 assert(elem_bt != T_ILLEGAL, "sanity");
6508
6509 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6510 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6511 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6512
6513 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6514 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6515 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6516
6517 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6518
6519 if (!stopped()) {
6520 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6521
6522 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6523 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6524 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6525 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6526
6527 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6528 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6529 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6530 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6531
6532 exit_block->init_req(inline_path, control());
6533 memory_phi->init_req(inline_path, map()->memory());
6534 result_phi->init_req(inline_path, result);
6535
6536 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6537 clear_upper_avx();
6538 }
6539 }
6540 }
6541
6542 if (call_stub_path != nullptr) {
6543 set_control(call_stub_path);
6544
6545 Node* call = make_runtime_call(RC_LEAF,
6546 OptoRuntime::vectorizedMismatch_Type(),
6547 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6548 obja_adr, objb_adr, length, scale);
6549
6550 exit_block->init_req(stub_path, control());
6551 memory_phi->init_req(stub_path, map()->memory());
6552 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6553 }
6554
6555 exit_block = _gvn.transform(exit_block);
6556 memory_phi = _gvn.transform(memory_phi);
6557 result_phi = _gvn.transform(result_phi);
6558
6559 set_control(exit_block);
6560 set_all_memory(memory_phi);
6561 set_result(result_phi);
6562
6563 return true;
6564 }
6565
6566 //------------------------------inline_vectorizedHashcode----------------------------
6567 bool LibraryCallKit::inline_vectorizedHashCode() {
6568 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6569
6570 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6571 Node* array = argument(0);
6572 Node* offset = argument(1);
6573 Node* length = argument(2);
6574 Node* initialValue = argument(3);
6575 Node* basic_type = argument(4);
6576
6577 if (basic_type == top()) {
6578 return false; // failed input validation
6579 }
6580
6581 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6582 if (!basic_type_t->is_con()) {
6583 return false; // Only intrinsify if mode argument is constant
6584 }
6585
6586 array = must_be_not_null(array, true);
6587
6588 BasicType bt = (BasicType)basic_type_t->get_con();
6589
6590 // Resolve address of first element
6591 Node* array_start = array_element_address(array, offset, bt);
6592
6593 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6594 array_start, length, initialValue, basic_type)));
6595 clear_upper_avx();
6596
6597 return true;
6598 }
6599
6600 /**
6601 * Calculate CRC32 for byte.
6602 * int java.util.zip.CRC32.update(int crc, int b)
6603 */
6604 bool LibraryCallKit::inline_updateCRC32() {
6605 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6606 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6607 // no receiver since it is static method
6608 Node* crc = argument(0); // type: int
6609 Node* b = argument(1); // type: int
6610
6611 /*
6612 * int c = ~ crc;
6613 * b = timesXtoThe32[(b ^ c) & 0xFF];
6614 * b = b ^ (c >>> 8);
6615 * crc = ~b;
6616 */
6617
6618 Node* M1 = intcon(-1);
6619 crc = _gvn.transform(new XorINode(crc, M1));
6620 Node* result = _gvn.transform(new XorINode(crc, b));
6621 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6622
6623 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6624 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6625 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6626 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6627
6628 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6629 result = _gvn.transform(new XorINode(crc, result));
6630 result = _gvn.transform(new XorINode(result, M1));
6631 set_result(result);
6632 return true;
6633 }
6634
6635 /**
6636 * Calculate CRC32 for byte[] array.
6637 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6638 */
6639 bool LibraryCallKit::inline_updateBytesCRC32() {
6640 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6641 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6642 // no receiver since it is static method
6643 Node* crc = argument(0); // type: int
6644 Node* src = argument(1); // type: oop
6645 Node* offset = argument(2); // type: int
6646 Node* length = argument(3); // type: int
6647
6648 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6649 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6650 // failed array check
6651 return false;
6652 }
6653
6654 // Figure out the size and type of the elements we will be copying.
6655 BasicType src_elem = src_type->elem()->array_element_basic_type();
6656 if (src_elem != T_BYTE) {
6657 return false;
6658 }
6659
6660 // 'src_start' points to src array + scaled offset
6661 src = must_be_not_null(src, true);
6662 Node* src_start = array_element_address(src, offset, src_elem);
6663
6664 // We assume that range check is done by caller.
6665 // TODO: generate range check (offset+length < src.length) in debug VM.
6666
6667 // Call the stub.
6668 address stubAddr = StubRoutines::updateBytesCRC32();
6669 const char *stubName = "updateBytesCRC32";
6670
6671 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6672 stubAddr, stubName, TypePtr::BOTTOM,
6673 crc, src_start, length);
6674 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6675 set_result(result);
6676 return true;
6677 }
6678
6679 /**
6680 * Calculate CRC32 for ByteBuffer.
6681 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6682 */
6683 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6684 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6685 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6686 // no receiver since it is static method
6687 Node* crc = argument(0); // type: int
6688 Node* src = argument(1); // type: long
6689 Node* offset = argument(3); // type: int
6690 Node* length = argument(4); // type: int
6691
6692 src = ConvL2X(src); // adjust Java long to machine word
6693 Node* base = _gvn.transform(new CastX2PNode(src));
6694 offset = ConvI2X(offset);
6695
6696 // 'src_start' points to src array + scaled offset
6697 Node* src_start = basic_plus_adr(top(), base, offset);
6698
6699 // Call the stub.
6700 address stubAddr = StubRoutines::updateBytesCRC32();
6701 const char *stubName = "updateBytesCRC32";
6702
6703 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6704 stubAddr, stubName, TypePtr::BOTTOM,
6705 crc, src_start, length);
6706 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6707 set_result(result);
6708 return true;
6709 }
6710
6711 //------------------------------get_table_from_crc32c_class-----------------------
6712 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6713 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6714 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6715
6716 return table;
6717 }
6718
6719 //------------------------------inline_updateBytesCRC32C-----------------------
6720 //
6721 // Calculate CRC32C for byte[] array.
6722 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6723 //
6724 bool LibraryCallKit::inline_updateBytesCRC32C() {
6725 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6726 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6727 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6728 // no receiver since it is a static method
6729 Node* crc = argument(0); // type: int
6730 Node* src = argument(1); // type: oop
6731 Node* offset = argument(2); // type: int
6732 Node* end = argument(3); // type: int
6733
6734 Node* length = _gvn.transform(new SubINode(end, offset));
6735
6736 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6737 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6738 // failed array check
6739 return false;
6740 }
6741
6742 // Figure out the size and type of the elements we will be copying.
6743 BasicType src_elem = src_type->elem()->array_element_basic_type();
6744 if (src_elem != T_BYTE) {
6745 return false;
6746 }
6747
6748 // 'src_start' points to src array + scaled offset
6749 src = must_be_not_null(src, true);
6750 Node* src_start = array_element_address(src, offset, src_elem);
6751
6752 // static final int[] byteTable in class CRC32C
6753 Node* table = get_table_from_crc32c_class(callee()->holder());
6754 table = must_be_not_null(table, true);
6755 Node* table_start = array_element_address(table, intcon(0), T_INT);
6756
6757 // We assume that range check is done by caller.
6758 // TODO: generate range check (offset+length < src.length) in debug VM.
6759
6760 // Call the stub.
6761 address stubAddr = StubRoutines::updateBytesCRC32C();
6762 const char *stubName = "updateBytesCRC32C";
6763
6764 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6765 stubAddr, stubName, TypePtr::BOTTOM,
6766 crc, src_start, length, table_start);
6767 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6768 set_result(result);
6769 return true;
6770 }
6771
6772 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6773 //
6774 // Calculate CRC32C for DirectByteBuffer.
6775 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6776 //
6777 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6778 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6779 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6780 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6781 // no receiver since it is a static method
6782 Node* crc = argument(0); // type: int
6783 Node* src = argument(1); // type: long
6784 Node* offset = argument(3); // type: int
6785 Node* end = argument(4); // type: int
6786
6787 Node* length = _gvn.transform(new SubINode(end, offset));
6788
6789 src = ConvL2X(src); // adjust Java long to machine word
6790 Node* base = _gvn.transform(new CastX2PNode(src));
6791 offset = ConvI2X(offset);
6792
6793 // 'src_start' points to src array + scaled offset
6794 Node* src_start = basic_plus_adr(top(), base, offset);
6795
6796 // static final int[] byteTable in class CRC32C
6797 Node* table = get_table_from_crc32c_class(callee()->holder());
6798 table = must_be_not_null(table, true);
6799 Node* table_start = array_element_address(table, intcon(0), T_INT);
6800
6801 // Call the stub.
6802 address stubAddr = StubRoutines::updateBytesCRC32C();
6803 const char *stubName = "updateBytesCRC32C";
6804
6805 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6806 stubAddr, stubName, TypePtr::BOTTOM,
6807 crc, src_start, length, table_start);
6808 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6809 set_result(result);
6810 return true;
6811 }
6812
6813 //------------------------------inline_updateBytesAdler32----------------------
6814 //
6815 // Calculate Adler32 checksum for byte[] array.
6816 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6817 //
6818 bool LibraryCallKit::inline_updateBytesAdler32() {
6819 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6820 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6821 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6822 // no receiver since it is static method
6823 Node* crc = argument(0); // type: int
6824 Node* src = argument(1); // type: oop
6825 Node* offset = argument(2); // type: int
6826 Node* length = argument(3); // type: int
6827
6828 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6829 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6830 // failed array check
6831 return false;
6832 }
6833
6834 // Figure out the size and type of the elements we will be copying.
6835 BasicType src_elem = src_type->elem()->array_element_basic_type();
6836 if (src_elem != T_BYTE) {
6837 return false;
6838 }
6839
6840 // 'src_start' points to src array + scaled offset
6841 Node* src_start = array_element_address(src, offset, src_elem);
6842
6843 // We assume that range check is done by caller.
6844 // TODO: generate range check (offset+length < src.length) in debug VM.
6845
6846 // Call the stub.
6847 address stubAddr = StubRoutines::updateBytesAdler32();
6848 const char *stubName = "updateBytesAdler32";
6849
6850 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6851 stubAddr, stubName, TypePtr::BOTTOM,
6852 crc, src_start, length);
6853 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6854 set_result(result);
6855 return true;
6856 }
6857
6858 //------------------------------inline_updateByteBufferAdler32---------------
6859 //
6860 // Calculate Adler32 checksum for DirectByteBuffer.
6861 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6862 //
6863 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6864 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6865 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6866 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6867 // no receiver since it is static method
6868 Node* crc = argument(0); // type: int
6869 Node* src = argument(1); // type: long
6870 Node* offset = argument(3); // type: int
6871 Node* length = argument(4); // type: int
6872
6873 src = ConvL2X(src); // adjust Java long to machine word
6874 Node* base = _gvn.transform(new CastX2PNode(src));
6875 offset = ConvI2X(offset);
6876
6877 // 'src_start' points to src array + scaled offset
6878 Node* src_start = basic_plus_adr(top(), base, offset);
6879
6880 // Call the stub.
6881 address stubAddr = StubRoutines::updateBytesAdler32();
6882 const char *stubName = "updateBytesAdler32";
6883
6884 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6885 stubAddr, stubName, TypePtr::BOTTOM,
6886 crc, src_start, length);
6887
6888 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6889 set_result(result);
6890 return true;
6891 }
6892
6893 //----------------------------inline_reference_get----------------------------
6894 // public T java.lang.ref.Reference.get();
6895 bool LibraryCallKit::inline_reference_get() {
6896 const int referent_offset = java_lang_ref_Reference::referent_offset();
6897
6898 // Get the argument:
6899 Node* reference_obj = null_check_receiver();
6900 if (stopped()) return true;
6901
6902 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6903 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6904 decorators, /*is_static*/ false, nullptr);
6905 if (result == nullptr) return false;
6906
6907 // Add memory barrier to prevent commoning reads from this field
6908 // across safepoint since GC can change its value.
6909 insert_mem_bar(Op_MemBarCPUOrder);
6910
6911 set_result(result);
6912 return true;
6913 }
6914
6915 //----------------------------inline_reference_refersTo0----------------------------
6916 // bool java.lang.ref.Reference.refersTo0();
6917 // bool java.lang.ref.PhantomReference.refersTo0();
6918 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6919 // Get arguments:
6920 Node* reference_obj = null_check_receiver();
6921 Node* other_obj = argument(1);
6922 if (stopped()) return true;
6923
6924 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6925 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6926 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6927 decorators, /*is_static*/ false, nullptr);
6928 if (referent == nullptr) return false;
6929
6930 // Add memory barrier to prevent commoning reads from this field
6931 // across safepoint since GC can change its value.
6932 insert_mem_bar(Op_MemBarCPUOrder);
6933
6934 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
6935 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6936 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
6937
6938 RegionNode* region = new RegionNode(3);
6939 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
6940
6941 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
6942 region->init_req(1, if_true);
6943 phi->init_req(1, intcon(1));
6944
6945 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
6946 region->init_req(2, if_false);
6947 phi->init_req(2, intcon(0));
6948
6949 set_control(_gvn.transform(region));
6950 record_for_igvn(region);
6951 set_result(_gvn.transform(phi));
6952 return true;
6953 }
6954
6955 //----------------------------inline_reference_clear0----------------------------
6956 // void java.lang.ref.Reference.clear0();
6957 // void java.lang.ref.PhantomReference.clear0();
6958 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
6959 // This matches the implementation in JVM_ReferenceClear, see the comments there.
6960
6961 // Get arguments
6962 Node* reference_obj = null_check_receiver();
6963 if (stopped()) return true;
6964
6965 // Common access parameters
6966 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6967 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6968 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
6969 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
6970 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
6971
6972 Node* referent = access_load_at(reference_obj,
6973 referent_field_addr,
6974 referent_field_addr_type,
6975 val_type,
6976 T_OBJECT,
6977 decorators);
6978
6979 IdealKit ideal(this);
6980 #define __ ideal.
6981 __ if_then(referent, BoolTest::ne, null());
6982 sync_kit(ideal);
6983 access_store_at(reference_obj,
6984 referent_field_addr,
6985 referent_field_addr_type,
6986 null(),
6987 val_type,
6988 T_OBJECT,
6989 decorators);
6990 __ sync_kit(this);
6991 __ end_if();
6992 final_sync(ideal);
6993 #undef __
6994
6995 return true;
6996 }
6997
6998 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
6999 DecoratorSet decorators, bool is_static,
7000 ciInstanceKlass* fromKls) {
7001 if (fromKls == nullptr) {
7002 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7003 assert(tinst != nullptr, "obj is null");
7004 assert(tinst->is_loaded(), "obj is not loaded");
7005 fromKls = tinst->instance_klass();
7006 } else {
7007 assert(is_static, "only for static field access");
7008 }
7009 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7010 ciSymbol::make(fieldTypeString),
7011 is_static);
7012
7013 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7014 if (field == nullptr) return (Node *) nullptr;
7015
7016 if (is_static) {
7017 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7018 fromObj = makecon(tip);
7019 }
7020
7021 // Next code copied from Parse::do_get_xxx():
7022
7023 // Compute address and memory type.
7024 int offset = field->offset_in_bytes();
7025 bool is_vol = field->is_volatile();
7026 ciType* field_klass = field->type();
7027 assert(field_klass->is_loaded(), "should be loaded");
7028 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7029 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7030 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7031 "slice of address and input slice don't match");
7032 BasicType bt = field->layout_type();
7033
7034 // Build the resultant type of the load
7035 const Type *type;
7036 if (bt == T_OBJECT) {
7037 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7038 } else {
7039 type = Type::get_const_basic_type(bt);
7040 }
7041
7042 if (is_vol) {
7043 decorators |= MO_SEQ_CST;
7044 }
7045
7046 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7047 }
7048
7049 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7050 bool is_exact /* true */, bool is_static /* false */,
7051 ciInstanceKlass * fromKls /* nullptr */) {
7052 if (fromKls == nullptr) {
7053 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7054 assert(tinst != nullptr, "obj is null");
7055 assert(tinst->is_loaded(), "obj is not loaded");
7056 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7057 fromKls = tinst->instance_klass();
7058 }
7059 else {
7060 assert(is_static, "only for static field access");
7061 }
7062 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7063 ciSymbol::make(fieldTypeString),
7064 is_static);
7065
7066 assert(field != nullptr, "undefined field");
7067 assert(!field->is_volatile(), "not defined for volatile fields");
7068
7069 if (is_static) {
7070 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7071 fromObj = makecon(tip);
7072 }
7073
7074 // Next code copied from Parse::do_get_xxx():
7075
7076 // Compute address and memory type.
7077 int offset = field->offset_in_bytes();
7078 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7079
7080 return adr;
7081 }
7082
7083 //------------------------------inline_aescrypt_Block-----------------------
7084 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7085 address stubAddr = nullptr;
7086 const char *stubName;
7087 assert(UseAES, "need AES instruction support");
7088
7089 switch(id) {
7090 case vmIntrinsics::_aescrypt_encryptBlock:
7091 stubAddr = StubRoutines::aescrypt_encryptBlock();
7092 stubName = "aescrypt_encryptBlock";
7093 break;
7094 case vmIntrinsics::_aescrypt_decryptBlock:
7095 stubAddr = StubRoutines::aescrypt_decryptBlock();
7096 stubName = "aescrypt_decryptBlock";
7097 break;
7098 default:
7099 break;
7100 }
7101 if (stubAddr == nullptr) return false;
7102
7103 Node* aescrypt_object = argument(0);
7104 Node* src = argument(1);
7105 Node* src_offset = argument(2);
7106 Node* dest = argument(3);
7107 Node* dest_offset = argument(4);
7108
7109 src = must_be_not_null(src, true);
7110 dest = must_be_not_null(dest, true);
7111
7112 // (1) src and dest are arrays.
7113 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7114 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7115 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7116 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7117
7118 // for the quick and dirty code we will skip all the checks.
7119 // we are just trying to get the call to be generated.
7120 Node* src_start = src;
7121 Node* dest_start = dest;
7122 if (src_offset != nullptr || dest_offset != nullptr) {
7123 assert(src_offset != nullptr && dest_offset != nullptr, "");
7124 src_start = array_element_address(src, src_offset, T_BYTE);
7125 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7126 }
7127
7128 // now need to get the start of its expanded key array
7129 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7130 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7131 if (k_start == nullptr) return false;
7132
7133 // Call the stub.
7134 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7135 stubAddr, stubName, TypePtr::BOTTOM,
7136 src_start, dest_start, k_start);
7137
7138 return true;
7139 }
7140
7141 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7142 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7143 address stubAddr = nullptr;
7144 const char *stubName = nullptr;
7145
7146 assert(UseAES, "need AES instruction support");
7147
7148 switch(id) {
7149 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7150 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7151 stubName = "cipherBlockChaining_encryptAESCrypt";
7152 break;
7153 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7154 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7155 stubName = "cipherBlockChaining_decryptAESCrypt";
7156 break;
7157 default:
7158 break;
7159 }
7160 if (stubAddr == nullptr) return false;
7161
7162 Node* cipherBlockChaining_object = argument(0);
7163 Node* src = argument(1);
7164 Node* src_offset = argument(2);
7165 Node* len = argument(3);
7166 Node* dest = argument(4);
7167 Node* dest_offset = argument(5);
7168
7169 src = must_be_not_null(src, false);
7170 dest = must_be_not_null(dest, false);
7171
7172 // (1) src and dest are arrays.
7173 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7174 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7175 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7176 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7177
7178 // checks are the responsibility of the caller
7179 Node* src_start = src;
7180 Node* dest_start = dest;
7181 if (src_offset != nullptr || dest_offset != nullptr) {
7182 assert(src_offset != nullptr && dest_offset != nullptr, "");
7183 src_start = array_element_address(src, src_offset, T_BYTE);
7184 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7185 }
7186
7187 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7188 // (because of the predicated logic executed earlier).
7189 // so we cast it here safely.
7190 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7191
7192 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7193 if (embeddedCipherObj == nullptr) return false;
7194
7195 // cast it to what we know it will be at runtime
7196 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7197 assert(tinst != nullptr, "CBC obj is null");
7198 assert(tinst->is_loaded(), "CBC obj is not loaded");
7199 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7200 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7201
7202 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7203 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7204 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7205 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7206 aescrypt_object = _gvn.transform(aescrypt_object);
7207
7208 // we need to get the start of the aescrypt_object's expanded key array
7209 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7210 if (k_start == nullptr) return false;
7211
7212 // similarly, get the start address of the r vector
7213 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7214 if (objRvec == nullptr) return false;
7215 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7216
7217 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7218 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7219 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7220 stubAddr, stubName, TypePtr::BOTTOM,
7221 src_start, dest_start, k_start, r_start, len);
7222
7223 // return cipher length (int)
7224 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7225 set_result(retvalue);
7226 return true;
7227 }
7228
7229 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7230 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7231 address stubAddr = nullptr;
7232 const char *stubName = nullptr;
7233
7234 assert(UseAES, "need AES instruction support");
7235
7236 switch (id) {
7237 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7238 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7239 stubName = "electronicCodeBook_encryptAESCrypt";
7240 break;
7241 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7242 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7243 stubName = "electronicCodeBook_decryptAESCrypt";
7244 break;
7245 default:
7246 break;
7247 }
7248
7249 if (stubAddr == nullptr) return false;
7250
7251 Node* electronicCodeBook_object = argument(0);
7252 Node* src = argument(1);
7253 Node* src_offset = argument(2);
7254 Node* len = argument(3);
7255 Node* dest = argument(4);
7256 Node* dest_offset = argument(5);
7257
7258 // (1) src and dest are arrays.
7259 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7260 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7261 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7262 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7263
7264 // checks are the responsibility of the caller
7265 Node* src_start = src;
7266 Node* dest_start = dest;
7267 if (src_offset != nullptr || dest_offset != nullptr) {
7268 assert(src_offset != nullptr && dest_offset != nullptr, "");
7269 src_start = array_element_address(src, src_offset, T_BYTE);
7270 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7271 }
7272
7273 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7274 // (because of the predicated logic executed earlier).
7275 // so we cast it here safely.
7276 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7277
7278 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7279 if (embeddedCipherObj == nullptr) return false;
7280
7281 // cast it to what we know it will be at runtime
7282 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7283 assert(tinst != nullptr, "ECB obj is null");
7284 assert(tinst->is_loaded(), "ECB obj is not loaded");
7285 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7286 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7287
7288 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7289 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7290 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7291 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7292 aescrypt_object = _gvn.transform(aescrypt_object);
7293
7294 // we need to get the start of the aescrypt_object's expanded key array
7295 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7296 if (k_start == nullptr) return false;
7297
7298 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7299 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7300 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7301 stubAddr, stubName, TypePtr::BOTTOM,
7302 src_start, dest_start, k_start, len);
7303
7304 // return cipher length (int)
7305 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7306 set_result(retvalue);
7307 return true;
7308 }
7309
7310 //------------------------------inline_counterMode_AESCrypt-----------------------
7311 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7312 assert(UseAES, "need AES instruction support");
7313 if (!UseAESCTRIntrinsics) return false;
7314
7315 address stubAddr = nullptr;
7316 const char *stubName = nullptr;
7317 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7318 stubAddr = StubRoutines::counterMode_AESCrypt();
7319 stubName = "counterMode_AESCrypt";
7320 }
7321 if (stubAddr == nullptr) return false;
7322
7323 Node* counterMode_object = argument(0);
7324 Node* src = argument(1);
7325 Node* src_offset = argument(2);
7326 Node* len = argument(3);
7327 Node* dest = argument(4);
7328 Node* dest_offset = argument(5);
7329
7330 // (1) src and dest are arrays.
7331 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7332 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7333 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7334 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7335
7336 // checks are the responsibility of the caller
7337 Node* src_start = src;
7338 Node* dest_start = dest;
7339 if (src_offset != nullptr || dest_offset != nullptr) {
7340 assert(src_offset != nullptr && dest_offset != nullptr, "");
7341 src_start = array_element_address(src, src_offset, T_BYTE);
7342 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7343 }
7344
7345 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7346 // (because of the predicated logic executed earlier).
7347 // so we cast it here safely.
7348 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7349 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7350 if (embeddedCipherObj == nullptr) return false;
7351 // cast it to what we know it will be at runtime
7352 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7353 assert(tinst != nullptr, "CTR obj is null");
7354 assert(tinst->is_loaded(), "CTR obj is not loaded");
7355 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7356 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7357 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7358 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7359 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7360 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7361 aescrypt_object = _gvn.transform(aescrypt_object);
7362 // we need to get the start of the aescrypt_object's expanded key array
7363 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7364 if (k_start == nullptr) return false;
7365 // similarly, get the start address of the r vector
7366 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7367 if (obj_counter == nullptr) return false;
7368 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7369
7370 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7371 if (saved_encCounter == nullptr) return false;
7372 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7373 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7374
7375 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7376 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7377 OptoRuntime::counterMode_aescrypt_Type(),
7378 stubAddr, stubName, TypePtr::BOTTOM,
7379 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7380
7381 // return cipher length (int)
7382 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7383 set_result(retvalue);
7384 return true;
7385 }
7386
7387 //------------------------------get_key_start_from_aescrypt_object-----------------------
7388 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7389 #if defined(PPC64) || defined(S390) || defined(RISCV64)
7390 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7391 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7392 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7393 // The ppc64 and riscv64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7394 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7395 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7396 if (objSessionK == nullptr) {
7397 return (Node *) nullptr;
7398 }
7399 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7400 #else
7401 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7402 #endif // PPC64
7403 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7404 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7405
7406 // now have the array, need to get the start address of the K array
7407 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7408 return k_start;
7409 }
7410
7411 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7412 // Return node representing slow path of predicate check.
7413 // the pseudo code we want to emulate with this predicate is:
7414 // for encryption:
7415 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7416 // for decryption:
7417 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7418 // note cipher==plain is more conservative than the original java code but that's OK
7419 //
7420 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7421 // The receiver was checked for null already.
7422 Node* objCBC = argument(0);
7423
7424 Node* src = argument(1);
7425 Node* dest = argument(4);
7426
7427 // Load embeddedCipher field of CipherBlockChaining object.
7428 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7429
7430 // get AESCrypt klass for instanceOf check
7431 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7432 // will have same classloader as CipherBlockChaining object
7433 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7434 assert(tinst != nullptr, "CBCobj is null");
7435 assert(tinst->is_loaded(), "CBCobj is not loaded");
7436
7437 // we want to do an instanceof comparison against the AESCrypt class
7438 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7439 if (!klass_AESCrypt->is_loaded()) {
7440 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7441 Node* ctrl = control();
7442 set_control(top()); // no regular fast path
7443 return ctrl;
7444 }
7445
7446 src = must_be_not_null(src, true);
7447 dest = must_be_not_null(dest, true);
7448
7449 // Resolve oops to stable for CmpP below.
7450 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7451
7452 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7453 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7454 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7455
7456 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7457
7458 // for encryption, we are done
7459 if (!decrypting)
7460 return instof_false; // even if it is null
7461
7462 // for decryption, we need to add a further check to avoid
7463 // taking the intrinsic path when cipher and plain are the same
7464 // see the original java code for why.
7465 RegionNode* region = new RegionNode(3);
7466 region->init_req(1, instof_false);
7467
7468 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7469 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7470 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7471 region->init_req(2, src_dest_conjoint);
7472
7473 record_for_igvn(region);
7474 return _gvn.transform(region);
7475 }
7476
7477 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7478 // Return node representing slow path of predicate check.
7479 // the pseudo code we want to emulate with this predicate is:
7480 // for encryption:
7481 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7482 // for decryption:
7483 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7484 // note cipher==plain is more conservative than the original java code but that's OK
7485 //
7486 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7487 // The receiver was checked for null already.
7488 Node* objECB = argument(0);
7489
7490 // Load embeddedCipher field of ElectronicCodeBook object.
7491 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7492
7493 // get AESCrypt klass for instanceOf check
7494 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7495 // will have same classloader as ElectronicCodeBook object
7496 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7497 assert(tinst != nullptr, "ECBobj is null");
7498 assert(tinst->is_loaded(), "ECBobj is not loaded");
7499
7500 // we want to do an instanceof comparison against the AESCrypt class
7501 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7502 if (!klass_AESCrypt->is_loaded()) {
7503 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7504 Node* ctrl = control();
7505 set_control(top()); // no regular fast path
7506 return ctrl;
7507 }
7508 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7509
7510 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7511 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7512 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7513
7514 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7515
7516 // for encryption, we are done
7517 if (!decrypting)
7518 return instof_false; // even if it is null
7519
7520 // for decryption, we need to add a further check to avoid
7521 // taking the intrinsic path when cipher and plain are the same
7522 // see the original java code for why.
7523 RegionNode* region = new RegionNode(3);
7524 region->init_req(1, instof_false);
7525 Node* src = argument(1);
7526 Node* dest = argument(4);
7527 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7528 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7529 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7530 region->init_req(2, src_dest_conjoint);
7531
7532 record_for_igvn(region);
7533 return _gvn.transform(region);
7534 }
7535
7536 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7537 // Return node representing slow path of predicate check.
7538 // the pseudo code we want to emulate with this predicate is:
7539 // for encryption:
7540 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7541 // for decryption:
7542 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7543 // note cipher==plain is more conservative than the original java code but that's OK
7544 //
7545
7546 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7547 // The receiver was checked for null already.
7548 Node* objCTR = argument(0);
7549
7550 // Load embeddedCipher field of CipherBlockChaining object.
7551 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7552
7553 // get AESCrypt klass for instanceOf check
7554 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7555 // will have same classloader as CipherBlockChaining object
7556 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7557 assert(tinst != nullptr, "CTRobj is null");
7558 assert(tinst->is_loaded(), "CTRobj is not loaded");
7559
7560 // we want to do an instanceof comparison against the AESCrypt class
7561 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7562 if (!klass_AESCrypt->is_loaded()) {
7563 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7564 Node* ctrl = control();
7565 set_control(top()); // no regular fast path
7566 return ctrl;
7567 }
7568
7569 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7570 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7571 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7572 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7573 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7574
7575 return instof_false; // even if it is null
7576 }
7577
7578 //------------------------------inline_ghash_processBlocks
7579 bool LibraryCallKit::inline_ghash_processBlocks() {
7580 address stubAddr;
7581 const char *stubName;
7582 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7583
7584 stubAddr = StubRoutines::ghash_processBlocks();
7585 stubName = "ghash_processBlocks";
7586
7587 Node* data = argument(0);
7588 Node* offset = argument(1);
7589 Node* len = argument(2);
7590 Node* state = argument(3);
7591 Node* subkeyH = argument(4);
7592
7593 state = must_be_not_null(state, true);
7594 subkeyH = must_be_not_null(subkeyH, true);
7595 data = must_be_not_null(data, true);
7596
7597 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7598 assert(state_start, "state is null");
7599 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7600 assert(subkeyH_start, "subkeyH is null");
7601 Node* data_start = array_element_address(data, offset, T_BYTE);
7602 assert(data_start, "data is null");
7603
7604 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7605 OptoRuntime::ghash_processBlocks_Type(),
7606 stubAddr, stubName, TypePtr::BOTTOM,
7607 state_start, subkeyH_start, data_start, len);
7608 return true;
7609 }
7610
7611 //------------------------------inline_chacha20Block
7612 bool LibraryCallKit::inline_chacha20Block() {
7613 address stubAddr;
7614 const char *stubName;
7615 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7616
7617 stubAddr = StubRoutines::chacha20Block();
7618 stubName = "chacha20Block";
7619
7620 Node* state = argument(0);
7621 Node* result = argument(1);
7622
7623 state = must_be_not_null(state, true);
7624 result = must_be_not_null(result, true);
7625
7626 Node* state_start = array_element_address(state, intcon(0), T_INT);
7627 assert(state_start, "state is null");
7628 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7629 assert(result_start, "result is null");
7630
7631 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7632 OptoRuntime::chacha20Block_Type(),
7633 stubAddr, stubName, TypePtr::BOTTOM,
7634 state_start, result_start);
7635 // return key stream length (int)
7636 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7637 set_result(retvalue);
7638 return true;
7639 }
7640
7641 bool LibraryCallKit::inline_base64_encodeBlock() {
7642 address stubAddr;
7643 const char *stubName;
7644 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7645 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
7646 stubAddr = StubRoutines::base64_encodeBlock();
7647 stubName = "encodeBlock";
7648
7649 if (!stubAddr) return false;
7650 Node* base64obj = argument(0);
7651 Node* src = argument(1);
7652 Node* offset = argument(2);
7653 Node* len = argument(3);
7654 Node* dest = argument(4);
7655 Node* dp = argument(5);
7656 Node* isURL = argument(6);
7657
7658 src = must_be_not_null(src, true);
7659 dest = must_be_not_null(dest, true);
7660
7661 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7662 assert(src_start, "source array is null");
7663 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7664 assert(dest_start, "destination array is null");
7665
7666 Node* base64 = make_runtime_call(RC_LEAF,
7667 OptoRuntime::base64_encodeBlock_Type(),
7668 stubAddr, stubName, TypePtr::BOTTOM,
7669 src_start, offset, len, dest_start, dp, isURL);
7670 return true;
7671 }
7672
7673 bool LibraryCallKit::inline_base64_decodeBlock() {
7674 address stubAddr;
7675 const char *stubName;
7676 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7677 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
7678 stubAddr = StubRoutines::base64_decodeBlock();
7679 stubName = "decodeBlock";
7680
7681 if (!stubAddr) return false;
7682 Node* base64obj = argument(0);
7683 Node* src = argument(1);
7684 Node* src_offset = argument(2);
7685 Node* len = argument(3);
7686 Node* dest = argument(4);
7687 Node* dest_offset = argument(5);
7688 Node* isURL = argument(6);
7689 Node* isMIME = argument(7);
7690
7691 src = must_be_not_null(src, true);
7692 dest = must_be_not_null(dest, true);
7693
7694 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7695 assert(src_start, "source array is null");
7696 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7697 assert(dest_start, "destination array is null");
7698
7699 Node* call = make_runtime_call(RC_LEAF,
7700 OptoRuntime::base64_decodeBlock_Type(),
7701 stubAddr, stubName, TypePtr::BOTTOM,
7702 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
7703 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7704 set_result(result);
7705 return true;
7706 }
7707
7708 bool LibraryCallKit::inline_poly1305_processBlocks() {
7709 address stubAddr;
7710 const char *stubName;
7711 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
7712 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
7713 stubAddr = StubRoutines::poly1305_processBlocks();
7714 stubName = "poly1305_processBlocks";
7715
7716 if (!stubAddr) return false;
7717 null_check_receiver(); // null-check receiver
7718 if (stopped()) return true;
7719
7720 Node* input = argument(1);
7721 Node* input_offset = argument(2);
7722 Node* len = argument(3);
7723 Node* alimbs = argument(4);
7724 Node* rlimbs = argument(5);
7725
7726 input = must_be_not_null(input, true);
7727 alimbs = must_be_not_null(alimbs, true);
7728 rlimbs = must_be_not_null(rlimbs, true);
7729
7730 Node* input_start = array_element_address(input, input_offset, T_BYTE);
7731 assert(input_start, "input array is null");
7732 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
7733 assert(acc_start, "acc array is null");
7734 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
7735 assert(r_start, "r array is null");
7736
7737 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7738 OptoRuntime::poly1305_processBlocks_Type(),
7739 stubAddr, stubName, TypePtr::BOTTOM,
7740 input_start, len, acc_start, r_start);
7741 return true;
7742 }
7743
7744 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
7745 address stubAddr;
7746 const char *stubName;
7747 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
7748 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
7749 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
7750 stubName = "intpoly_montgomeryMult_P256";
7751
7752 if (!stubAddr) return false;
7753 null_check_receiver(); // null-check receiver
7754 if (stopped()) return true;
7755
7756 Node* a = argument(1);
7757 Node* b = argument(2);
7758 Node* r = argument(3);
7759
7760 a = must_be_not_null(a, true);
7761 b = must_be_not_null(b, true);
7762 r = must_be_not_null(r, true);
7763
7764 Node* a_start = array_element_address(a, intcon(0), T_LONG);
7765 assert(a_start, "a array is null");
7766 Node* b_start = array_element_address(b, intcon(0), T_LONG);
7767 assert(b_start, "b array is null");
7768 Node* r_start = array_element_address(r, intcon(0), T_LONG);
7769 assert(r_start, "r array is null");
7770
7771 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7772 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
7773 stubAddr, stubName, TypePtr::BOTTOM,
7774 a_start, b_start, r_start);
7775 return true;
7776 }
7777
7778 bool LibraryCallKit::inline_intpoly_assign() {
7779 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
7780 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
7781 const char *stubName = "intpoly_assign";
7782 address stubAddr = StubRoutines::intpoly_assign();
7783 if (!stubAddr) return false;
7784
7785 Node* set = argument(0);
7786 Node* a = argument(1);
7787 Node* b = argument(2);
7788 Node* arr_length = load_array_length(a);
7789
7790 a = must_be_not_null(a, true);
7791 b = must_be_not_null(b, true);
7792
7793 Node* a_start = array_element_address(a, intcon(0), T_LONG);
7794 assert(a_start, "a array is null");
7795 Node* b_start = array_element_address(b, intcon(0), T_LONG);
7796 assert(b_start, "b array is null");
7797
7798 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7799 OptoRuntime::intpoly_assign_Type(),
7800 stubAddr, stubName, TypePtr::BOTTOM,
7801 set, a_start, b_start, arr_length);
7802 return true;
7803 }
7804
7805 //------------------------------inline_digestBase_implCompress-----------------------
7806 //
7807 // Calculate MD5 for single-block byte[] array.
7808 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
7809 //
7810 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
7811 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
7812 //
7813 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
7814 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
7815 //
7816 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
7817 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
7818 //
7819 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
7820 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
7821 //
7822 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
7823 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
7824
7825 Node* digestBase_obj = argument(0);
7826 Node* src = argument(1); // type oop
7827 Node* ofs = argument(2); // type int
7828
7829 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7830 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7831 // failed array check
7832 return false;
7833 }
7834 // Figure out the size and type of the elements we will be copying.
7835 BasicType src_elem = src_type->elem()->array_element_basic_type();
7836 if (src_elem != T_BYTE) {
7837 return false;
7838 }
7839 // 'src_start' points to src array + offset
7840 src = must_be_not_null(src, true);
7841 Node* src_start = array_element_address(src, ofs, src_elem);
7842 Node* state = nullptr;
7843 Node* block_size = nullptr;
7844 address stubAddr;
7845 const char *stubName;
7846
7847 switch(id) {
7848 case vmIntrinsics::_md5_implCompress:
7849 assert(UseMD5Intrinsics, "need MD5 instruction support");
7850 state = get_state_from_digest_object(digestBase_obj, T_INT);
7851 stubAddr = StubRoutines::md5_implCompress();
7852 stubName = "md5_implCompress";
7853 break;
7854 case vmIntrinsics::_sha_implCompress:
7855 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
7856 state = get_state_from_digest_object(digestBase_obj, T_INT);
7857 stubAddr = StubRoutines::sha1_implCompress();
7858 stubName = "sha1_implCompress";
7859 break;
7860 case vmIntrinsics::_sha2_implCompress:
7861 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
7862 state = get_state_from_digest_object(digestBase_obj, T_INT);
7863 stubAddr = StubRoutines::sha256_implCompress();
7864 stubName = "sha256_implCompress";
7865 break;
7866 case vmIntrinsics::_sha5_implCompress:
7867 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
7868 state = get_state_from_digest_object(digestBase_obj, T_LONG);
7869 stubAddr = StubRoutines::sha512_implCompress();
7870 stubName = "sha512_implCompress";
7871 break;
7872 case vmIntrinsics::_sha3_implCompress:
7873 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
7874 state = get_state_from_digest_object(digestBase_obj, T_LONG);
7875 stubAddr = StubRoutines::sha3_implCompress();
7876 stubName = "sha3_implCompress";
7877 block_size = get_block_size_from_digest_object(digestBase_obj);
7878 if (block_size == nullptr) return false;
7879 break;
7880 default:
7881 fatal_unexpected_iid(id);
7882 return false;
7883 }
7884 if (state == nullptr) return false;
7885
7886 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
7887 if (stubAddr == nullptr) return false;
7888
7889 // Call the stub.
7890 Node* call;
7891 if (block_size == nullptr) {
7892 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
7893 stubAddr, stubName, TypePtr::BOTTOM,
7894 src_start, state);
7895 } else {
7896 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
7897 stubAddr, stubName, TypePtr::BOTTOM,
7898 src_start, state, block_size);
7899 }
7900
7901 return true;
7902 }
7903
7904 //------------------------------inline_digestBase_implCompressMB-----------------------
7905 //
7906 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
7907 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
7908 //
7909 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
7910 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7911 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7912 assert((uint)predicate < 5, "sanity");
7913 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
7914
7915 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
7916 Node* src = argument(1); // byte[] array
7917 Node* ofs = argument(2); // type int
7918 Node* limit = argument(3); // type int
7919
7920 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7921 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7922 // failed array check
7923 return false;
7924 }
7925 // Figure out the size and type of the elements we will be copying.
7926 BasicType src_elem = src_type->elem()->array_element_basic_type();
7927 if (src_elem != T_BYTE) {
7928 return false;
7929 }
7930 // 'src_start' points to src array + offset
7931 src = must_be_not_null(src, false);
7932 Node* src_start = array_element_address(src, ofs, src_elem);
7933
7934 const char* klass_digestBase_name = nullptr;
7935 const char* stub_name = nullptr;
7936 address stub_addr = nullptr;
7937 BasicType elem_type = T_INT;
7938
7939 switch (predicate) {
7940 case 0:
7941 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
7942 klass_digestBase_name = "sun/security/provider/MD5";
7943 stub_name = "md5_implCompressMB";
7944 stub_addr = StubRoutines::md5_implCompressMB();
7945 }
7946 break;
7947 case 1:
7948 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
7949 klass_digestBase_name = "sun/security/provider/SHA";
7950 stub_name = "sha1_implCompressMB";
7951 stub_addr = StubRoutines::sha1_implCompressMB();
7952 }
7953 break;
7954 case 2:
7955 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
7956 klass_digestBase_name = "sun/security/provider/SHA2";
7957 stub_name = "sha256_implCompressMB";
7958 stub_addr = StubRoutines::sha256_implCompressMB();
7959 }
7960 break;
7961 case 3:
7962 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
7963 klass_digestBase_name = "sun/security/provider/SHA5";
7964 stub_name = "sha512_implCompressMB";
7965 stub_addr = StubRoutines::sha512_implCompressMB();
7966 elem_type = T_LONG;
7967 }
7968 break;
7969 case 4:
7970 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
7971 klass_digestBase_name = "sun/security/provider/SHA3";
7972 stub_name = "sha3_implCompressMB";
7973 stub_addr = StubRoutines::sha3_implCompressMB();
7974 elem_type = T_LONG;
7975 }
7976 break;
7977 default:
7978 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
7979 }
7980 if (klass_digestBase_name != nullptr) {
7981 assert(stub_addr != nullptr, "Stub is generated");
7982 if (stub_addr == nullptr) return false;
7983
7984 // get DigestBase klass to lookup for SHA klass
7985 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
7986 assert(tinst != nullptr, "digestBase_obj is not instance???");
7987 assert(tinst->is_loaded(), "DigestBase is not loaded");
7988
7989 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
7990 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
7991 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
7992 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
7993 }
7994 return false;
7995 }
7996
7997 //------------------------------inline_digestBase_implCompressMB-----------------------
7998 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
7999 BasicType elem_type, address stubAddr, const char *stubName,
8000 Node* src_start, Node* ofs, Node* limit) {
8001 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
8002 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8003 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
8004 digest_obj = _gvn.transform(digest_obj);
8005
8006 Node* state = get_state_from_digest_object(digest_obj, elem_type);
8007 if (state == nullptr) return false;
8008
8009 Node* block_size = nullptr;
8010 if (strcmp("sha3_implCompressMB", stubName) == 0) {
8011 block_size = get_block_size_from_digest_object(digest_obj);
8012 if (block_size == nullptr) return false;
8013 }
8014
8015 // Call the stub.
8016 Node* call;
8017 if (block_size == nullptr) {
8018 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8019 OptoRuntime::digestBase_implCompressMB_Type(false),
8020 stubAddr, stubName, TypePtr::BOTTOM,
8021 src_start, state, ofs, limit);
8022 } else {
8023 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8024 OptoRuntime::digestBase_implCompressMB_Type(true),
8025 stubAddr, stubName, TypePtr::BOTTOM,
8026 src_start, state, block_size, ofs, limit);
8027 }
8028
8029 // return ofs (int)
8030 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8031 set_result(result);
8032
8033 return true;
8034 }
8035
8036 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8037 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8038 assert(UseAES, "need AES instruction support");
8039 address stubAddr = nullptr;
8040 const char *stubName = nullptr;
8041 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8042 stubName = "galoisCounterMode_AESCrypt";
8043
8044 if (stubAddr == nullptr) return false;
8045
8046 Node* in = argument(0);
8047 Node* inOfs = argument(1);
8048 Node* len = argument(2);
8049 Node* ct = argument(3);
8050 Node* ctOfs = argument(4);
8051 Node* out = argument(5);
8052 Node* outOfs = argument(6);
8053 Node* gctr_object = argument(7);
8054 Node* ghash_object = argument(8);
8055
8056 // (1) in, ct and out are arrays.
8057 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8058 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8059 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8060 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8061 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8062 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8063
8064 // checks are the responsibility of the caller
8065 Node* in_start = in;
8066 Node* ct_start = ct;
8067 Node* out_start = out;
8068 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8069 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8070 in_start = array_element_address(in, inOfs, T_BYTE);
8071 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8072 out_start = array_element_address(out, outOfs, T_BYTE);
8073 }
8074
8075 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8076 // (because of the predicated logic executed earlier).
8077 // so we cast it here safely.
8078 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8079 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8080 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8081 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8082 Node* state = load_field_from_object(ghash_object, "state", "[J");
8083
8084 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8085 return false;
8086 }
8087 // cast it to what we know it will be at runtime
8088 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8089 assert(tinst != nullptr, "GCTR obj is null");
8090 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8091 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8092 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8093 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8094 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8095 const TypeOopPtr* xtype = aklass->as_instance_type();
8096 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8097 aescrypt_object = _gvn.transform(aescrypt_object);
8098 // we need to get the start of the aescrypt_object's expanded key array
8099 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8100 if (k_start == nullptr) return false;
8101 // similarly, get the start address of the r vector
8102 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8103 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8104 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8105
8106
8107 // Call the stub, passing params
8108 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8109 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8110 stubAddr, stubName, TypePtr::BOTTOM,
8111 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8112
8113 // return cipher length (int)
8114 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8115 set_result(retvalue);
8116
8117 return true;
8118 }
8119
8120 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8121 // Return node representing slow path of predicate check.
8122 // the pseudo code we want to emulate with this predicate is:
8123 // for encryption:
8124 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8125 // for decryption:
8126 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8127 // note cipher==plain is more conservative than the original java code but that's OK
8128 //
8129
8130 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8131 // The receiver was checked for null already.
8132 Node* objGCTR = argument(7);
8133 // Load embeddedCipher field of GCTR object.
8134 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8135 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8136
8137 // get AESCrypt klass for instanceOf check
8138 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8139 // will have same classloader as CipherBlockChaining object
8140 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8141 assert(tinst != nullptr, "GCTR obj is null");
8142 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8143
8144 // we want to do an instanceof comparison against the AESCrypt class
8145 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8146 if (!klass_AESCrypt->is_loaded()) {
8147 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8148 Node* ctrl = control();
8149 set_control(top()); // no regular fast path
8150 return ctrl;
8151 }
8152
8153 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8154 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8155 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8156 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8157 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8158
8159 return instof_false; // even if it is null
8160 }
8161
8162 //------------------------------get_state_from_digest_object-----------------------
8163 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8164 const char* state_type;
8165 switch (elem_type) {
8166 case T_BYTE: state_type = "[B"; break;
8167 case T_INT: state_type = "[I"; break;
8168 case T_LONG: state_type = "[J"; break;
8169 default: ShouldNotReachHere();
8170 }
8171 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8172 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8173 if (digest_state == nullptr) return (Node *) nullptr;
8174
8175 // now have the array, need to get the start address of the state array
8176 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8177 return state;
8178 }
8179
8180 //------------------------------get_block_size_from_sha3_object----------------------------------
8181 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8182 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8183 assert (block_size != nullptr, "sanity");
8184 return block_size;
8185 }
8186
8187 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8188 // Return node representing slow path of predicate check.
8189 // the pseudo code we want to emulate with this predicate is:
8190 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8191 //
8192 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8193 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8194 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8195 assert((uint)predicate < 5, "sanity");
8196
8197 // The receiver was checked for null already.
8198 Node* digestBaseObj = argument(0);
8199
8200 // get DigestBase klass for instanceOf check
8201 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8202 assert(tinst != nullptr, "digestBaseObj is null");
8203 assert(tinst->is_loaded(), "DigestBase is not loaded");
8204
8205 const char* klass_name = nullptr;
8206 switch (predicate) {
8207 case 0:
8208 if (UseMD5Intrinsics) {
8209 // we want to do an instanceof comparison against the MD5 class
8210 klass_name = "sun/security/provider/MD5";
8211 }
8212 break;
8213 case 1:
8214 if (UseSHA1Intrinsics) {
8215 // we want to do an instanceof comparison against the SHA class
8216 klass_name = "sun/security/provider/SHA";
8217 }
8218 break;
8219 case 2:
8220 if (UseSHA256Intrinsics) {
8221 // we want to do an instanceof comparison against the SHA2 class
8222 klass_name = "sun/security/provider/SHA2";
8223 }
8224 break;
8225 case 3:
8226 if (UseSHA512Intrinsics) {
8227 // we want to do an instanceof comparison against the SHA5 class
8228 klass_name = "sun/security/provider/SHA5";
8229 }
8230 break;
8231 case 4:
8232 if (UseSHA3Intrinsics) {
8233 // we want to do an instanceof comparison against the SHA3 class
8234 klass_name = "sun/security/provider/SHA3";
8235 }
8236 break;
8237 default:
8238 fatal("unknown SHA intrinsic predicate: %d", predicate);
8239 }
8240
8241 ciKlass* klass = nullptr;
8242 if (klass_name != nullptr) {
8243 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8244 }
8245 if ((klass == nullptr) || !klass->is_loaded()) {
8246 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8247 Node* ctrl = control();
8248 set_control(top()); // no intrinsic path
8249 return ctrl;
8250 }
8251 ciInstanceKlass* instklass = klass->as_instance_klass();
8252
8253 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8254 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8255 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8256 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8257
8258 return instof_false; // even if it is null
8259 }
8260
8261 //-------------inline_fma-----------------------------------
8262 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8263 Node *a = nullptr;
8264 Node *b = nullptr;
8265 Node *c = nullptr;
8266 Node* result = nullptr;
8267 switch (id) {
8268 case vmIntrinsics::_fmaD:
8269 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8270 // no receiver since it is static method
8271 a = round_double_node(argument(0));
8272 b = round_double_node(argument(2));
8273 c = round_double_node(argument(4));
8274 result = _gvn.transform(new FmaDNode(control(), a, b, c));
8275 break;
8276 case vmIntrinsics::_fmaF:
8277 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8278 a = argument(0);
8279 b = argument(1);
8280 c = argument(2);
8281 result = _gvn.transform(new FmaFNode(control(), a, b, c));
8282 break;
8283 default:
8284 fatal_unexpected_iid(id); break;
8285 }
8286 set_result(result);
8287 return true;
8288 }
8289
8290 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8291 // argument(0) is receiver
8292 Node* codePoint = argument(1);
8293 Node* n = nullptr;
8294
8295 switch (id) {
8296 case vmIntrinsics::_isDigit :
8297 n = new DigitNode(control(), codePoint);
8298 break;
8299 case vmIntrinsics::_isLowerCase :
8300 n = new LowerCaseNode(control(), codePoint);
8301 break;
8302 case vmIntrinsics::_isUpperCase :
8303 n = new UpperCaseNode(control(), codePoint);
8304 break;
8305 case vmIntrinsics::_isWhitespace :
8306 n = new WhitespaceNode(control(), codePoint);
8307 break;
8308 default:
8309 fatal_unexpected_iid(id);
8310 }
8311
8312 set_result(_gvn.transform(n));
8313 return true;
8314 }
8315
8316 //------------------------------inline_fp_min_max------------------------------
8317 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
8318 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
8319
8320 // The intrinsic should be used only when the API branches aren't predictable,
8321 // the last one performing the most important comparison. The following heuristic
8322 // uses the branch statistics to eventually bail out if necessary.
8323
8324 ciMethodData *md = callee()->method_data();
8325
8326 if ( md != nullptr && md->is_mature() && md->invocation_count() > 0 ) {
8327 ciCallProfile cp = caller()->call_profile_at_bci(bci());
8328
8329 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
8330 // Bail out if the call-site didn't contribute enough to the statistics.
8331 return false;
8332 }
8333
8334 uint taken = 0, not_taken = 0;
8335
8336 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
8337 if (p->is_BranchData()) {
8338 taken = ((ciBranchData*)p)->taken();
8339 not_taken = ((ciBranchData*)p)->not_taken();
8340 }
8341 }
8342
8343 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
8344 balance = balance < 0 ? -balance : balance;
8345 if ( balance > 0.2 ) {
8346 // Bail out if the most important branch is predictable enough.
8347 return false;
8348 }
8349 }
8350 */
8351
8352 Node *a = nullptr;
8353 Node *b = nullptr;
8354 Node *n = nullptr;
8355 switch (id) {
8356 case vmIntrinsics::_maxF:
8357 case vmIntrinsics::_minF:
8358 case vmIntrinsics::_maxF_strict:
8359 case vmIntrinsics::_minF_strict:
8360 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
8361 a = argument(0);
8362 b = argument(1);
8363 break;
8364 case vmIntrinsics::_maxD:
8365 case vmIntrinsics::_minD:
8366 case vmIntrinsics::_maxD_strict:
8367 case vmIntrinsics::_minD_strict:
8368 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
8369 a = round_double_node(argument(0));
8370 b = round_double_node(argument(2));
8371 break;
8372 default:
8373 fatal_unexpected_iid(id);
8374 break;
8375 }
8376 switch (id) {
8377 case vmIntrinsics::_maxF:
8378 case vmIntrinsics::_maxF_strict:
8379 n = new MaxFNode(a, b);
8380 break;
8381 case vmIntrinsics::_minF:
8382 case vmIntrinsics::_minF_strict:
8383 n = new MinFNode(a, b);
8384 break;
8385 case vmIntrinsics::_maxD:
8386 case vmIntrinsics::_maxD_strict:
8387 n = new MaxDNode(a, b);
8388 break;
8389 case vmIntrinsics::_minD:
8390 case vmIntrinsics::_minD_strict:
8391 n = new MinDNode(a, b);
8392 break;
8393 default:
8394 fatal_unexpected_iid(id);
8395 break;
8396 }
8397 set_result(_gvn.transform(n));
8398 return true;
8399 }
8400
8401 bool LibraryCallKit::inline_profileBoolean() {
8402 Node* counts = argument(1);
8403 const TypeAryPtr* ary = nullptr;
8404 ciArray* aobj = nullptr;
8405 if (counts->is_Con()
8406 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8407 && (aobj = ary->const_oop()->as_array()) != nullptr
8408 && (aobj->length() == 2)) {
8409 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8410 jint false_cnt = aobj->element_value(0).as_int();
8411 jint true_cnt = aobj->element_value(1).as_int();
8412
8413 if (C->log() != nullptr) {
8414 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8415 false_cnt, true_cnt);
8416 }
8417
8418 if (false_cnt + true_cnt == 0) {
8419 // According to profile, never executed.
8420 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8421 Deoptimization::Action_reinterpret);
8422 return true;
8423 }
8424
8425 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8426 // is a number of each value occurrences.
8427 Node* result = argument(0);
8428 if (false_cnt == 0 || true_cnt == 0) {
8429 // According to profile, one value has been never seen.
8430 int expected_val = (false_cnt == 0) ? 1 : 0;
8431
8432 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8433 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8434
8435 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8436 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8437 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8438
8439 { // Slow path: uncommon trap for never seen value and then reexecute
8440 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8441 // the value has been seen at least once.
8442 PreserveJVMState pjvms(this);
8443 PreserveReexecuteState preexecs(this);
8444 jvms()->set_should_reexecute(true);
8445
8446 set_control(slow_path);
8447 set_i_o(i_o());
8448
8449 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8450 Deoptimization::Action_reinterpret);
8451 }
8452 // The guard for never seen value enables sharpening of the result and
8453 // returning a constant. It allows to eliminate branches on the same value
8454 // later on.
8455 set_control(fast_path);
8456 result = intcon(expected_val);
8457 }
8458 // Stop profiling.
8459 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8460 // By replacing method body with profile data (represented as ProfileBooleanNode
8461 // on IR level) we effectively disable profiling.
8462 // It enables full speed execution once optimized code is generated.
8463 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8464 C->record_for_igvn(profile);
8465 set_result(profile);
8466 return true;
8467 } else {
8468 // Continue profiling.
8469 // Profile data isn't available at the moment. So, execute method's bytecode version.
8470 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8471 // is compiled and counters aren't available since corresponding MethodHandle
8472 // isn't a compile-time constant.
8473 return false;
8474 }
8475 }
8476
8477 bool LibraryCallKit::inline_isCompileConstant() {
8478 Node* n = argument(0);
8479 set_result(n->is_Con() ? intcon(1) : intcon(0));
8480 return true;
8481 }
8482
8483 //------------------------------- inline_getObjectSize --------------------------------------
8484 //
8485 // Calculate the runtime size of the object/array.
8486 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8487 //
8488 bool LibraryCallKit::inline_getObjectSize() {
8489 Node* obj = argument(3);
8490 Node* klass_node = load_object_klass(obj);
8491
8492 jint layout_con = Klass::_lh_neutral_value;
8493 Node* layout_val = get_layout_helper(klass_node, layout_con);
8494 int layout_is_con = (layout_val == nullptr);
8495
8496 if (layout_is_con) {
8497 // Layout helper is constant, can figure out things at compile time.
8498
8499 if (Klass::layout_helper_is_instance(layout_con)) {
8500 // Instance case: layout_con contains the size itself.
8501 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8502 set_result(size);
8503 } else {
8504 // Array case: size is round(header + element_size*arraylength).
8505 // Since arraylength is different for every array instance, we have to
8506 // compute the whole thing at runtime.
8507
8508 Node* arr_length = load_array_length(obj);
8509
8510 int round_mask = MinObjAlignmentInBytes - 1;
8511 int hsize = Klass::layout_helper_header_size(layout_con);
8512 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8513
8514 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8515 round_mask = 0; // strength-reduce it if it goes away completely
8516 }
8517 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8518 Node* header_size = intcon(hsize + round_mask);
8519
8520 Node* lengthx = ConvI2X(arr_length);
8521 Node* headerx = ConvI2X(header_size);
8522
8523 Node* abody = lengthx;
8524 if (eshift != 0) {
8525 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8526 }
8527 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8528 if (round_mask != 0) {
8529 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8530 }
8531 size = ConvX2L(size);
8532 set_result(size);
8533 }
8534 } else {
8535 // Layout helper is not constant, need to test for array-ness at runtime.
8536
8537 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8538 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8539 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8540 record_for_igvn(result_reg);
8541
8542 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
8543 if (array_ctl != nullptr) {
8544 // Array case: size is round(header + element_size*arraylength).
8545 // Since arraylength is different for every array instance, we have to
8546 // compute the whole thing at runtime.
8547
8548 PreserveJVMState pjvms(this);
8549 set_control(array_ctl);
8550 Node* arr_length = load_array_length(obj);
8551
8552 int round_mask = MinObjAlignmentInBytes - 1;
8553 Node* mask = intcon(round_mask);
8554
8555 Node* hss = intcon(Klass::_lh_header_size_shift);
8556 Node* hsm = intcon(Klass::_lh_header_size_mask);
8557 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8558 header_size = _gvn.transform(new AndINode(header_size, hsm));
8559 header_size = _gvn.transform(new AddINode(header_size, mask));
8560
8561 // There is no need to mask or shift this value.
8562 // The semantics of LShiftINode include an implicit mask to 0x1F.
8563 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8564 Node* elem_shift = layout_val;
8565
8566 Node* lengthx = ConvI2X(arr_length);
8567 Node* headerx = ConvI2X(header_size);
8568
8569 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
8570 Node* size = _gvn.transform(new AddXNode(headerx, abody));
8571 if (round_mask != 0) {
8572 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
8573 }
8574 size = ConvX2L(size);
8575
8576 result_reg->init_req(_array_path, control());
8577 result_val->init_req(_array_path, size);
8578 }
8579
8580 if (!stopped()) {
8581 // Instance case: the layout helper gives us instance size almost directly,
8582 // but we need to mask out the _lh_instance_slow_path_bit.
8583 Node* size = ConvI2X(layout_val);
8584 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
8585 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
8586 size = _gvn.transform(new AndXNode(size, mask));
8587 size = ConvX2L(size);
8588
8589 result_reg->init_req(_instance_path, control());
8590 result_val->init_req(_instance_path, size);
8591 }
8592
8593 set_result(result_reg, result_val);
8594 }
8595
8596 return true;
8597 }
8598
8599 //------------------------------- inline_blackhole --------------------------------------
8600 //
8601 // Make sure all arguments to this node are alive.
8602 // This matches methods that were requested to be blackholed through compile commands.
8603 //
8604 bool LibraryCallKit::inline_blackhole() {
8605 assert(callee()->is_static(), "Should have been checked before: only static methods here");
8606 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
8607 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
8608
8609 // Blackhole node pinches only the control, not memory. This allows
8610 // the blackhole to be pinned in the loop that computes blackholed
8611 // values, but have no other side effects, like breaking the optimizations
8612 // across the blackhole.
8613
8614 Node* bh = _gvn.transform(new BlackholeNode(control()));
8615 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
8616
8617 // Bind call arguments as blackhole arguments to keep them alive
8618 uint nargs = callee()->arg_size();
8619 for (uint i = 0; i < nargs; i++) {
8620 bh->add_req(argument(i));
8621 }
8622
8623 return true;
8624 }