1 /*
2 * Copyright (c) 1999, 2024, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "ci/ciUtilities.inline.hpp"
28 #include "classfile/vmIntrinsics.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "gc/shared/barrierSet.hpp"
32 #include "jfr/support/jfrIntrinsics.hpp"
33 #include "memory/resourceArea.hpp"
34 #include "oops/klass.inline.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "opto/addnode.hpp"
37 #include "opto/arraycopynode.hpp"
38 #include "opto/c2compiler.hpp"
39 #include "opto/castnode.hpp"
40 #include "opto/cfgnode.hpp"
41 #include "opto/convertnode.hpp"
42 #include "opto/countbitsnode.hpp"
43 #include "opto/idealKit.hpp"
44 #include "opto/library_call.hpp"
45 #include "opto/mathexactnode.hpp"
46 #include "opto/mulnode.hpp"
47 #include "opto/narrowptrnode.hpp"
48 #include "opto/opaquenode.hpp"
49 #include "opto/parse.hpp"
50 #include "opto/runtime.hpp"
51 #include "opto/rootnode.hpp"
52 #include "opto/subnode.hpp"
53 #include "prims/jvmtiExport.hpp"
54 #include "prims/jvmtiThreadState.hpp"
55 #include "prims/unsafe.hpp"
56 #include "runtime/jniHandles.inline.hpp"
57 #include "runtime/objectMonitor.hpp"
58 #include "runtime/sharedRuntime.hpp"
59 #include "runtime/stubRoutines.hpp"
60 #include "utilities/macros.hpp"
61 #include "utilities/powerOfTwo.hpp"
62
63 //---------------------------make_vm_intrinsic----------------------------
64 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
65 vmIntrinsicID id = m->intrinsic_id();
66 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
67
68 if (!m->is_loaded()) {
69 // Do not attempt to inline unloaded methods.
70 return nullptr;
71 }
72
73 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
74 bool is_available = false;
75
76 {
77 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
78 // the compiler must transition to '_thread_in_vm' state because both
79 // methods access VM-internal data.
80 VM_ENTRY_MARK;
81 methodHandle mh(THREAD, m->get_Method());
82 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
83 if (is_available && is_virtual) {
84 is_available = vmIntrinsics::does_virtual_dispatch(id);
85 }
86 }
87
88 if (is_available) {
89 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
90 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
91 return new LibraryIntrinsic(m, is_virtual,
92 vmIntrinsics::predicates_needed(id),
93 vmIntrinsics::does_virtual_dispatch(id),
94 id);
95 } else {
96 return nullptr;
97 }
98 }
99
100 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
101 LibraryCallKit kit(jvms, this);
102 Compile* C = kit.C;
103 int nodes = C->unique();
104 #ifndef PRODUCT
105 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
106 char buf[1000];
107 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
108 tty->print_cr("Intrinsic %s", str);
109 }
110 #endif
111 ciMethod* callee = kit.callee();
112 const int bci = kit.bci();
113 #ifdef ASSERT
114 Node* ctrl = kit.control();
115 #endif
116 // Try to inline the intrinsic.
117 if (callee->check_intrinsic_candidate() &&
118 kit.try_to_inline(_last_predicate)) {
119 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
120 : "(intrinsic)";
121 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
122 if (C->print_intrinsics() || C->print_inlining()) {
123 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
124 }
125 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
126 if (C->log()) {
127 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
128 vmIntrinsics::name_at(intrinsic_id()),
129 (is_virtual() ? " virtual='1'" : ""),
130 C->unique() - nodes);
131 }
132 // Push the result from the inlined method onto the stack.
133 kit.push_result();
134 C->print_inlining_update(this);
135 return kit.transfer_exceptions_into_jvms();
136 }
137
138 // The intrinsic bailed out
139 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
140 if (jvms->has_method()) {
141 // Not a root compile.
142 const char* msg;
143 if (callee->intrinsic_candidate()) {
144 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
145 } else {
146 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
147 : "failed to inline (intrinsic), method not annotated";
148 }
149 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
150 if (C->print_intrinsics() || C->print_inlining()) {
151 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
152 }
153 } else {
154 // Root compile
155 ResourceMark rm;
156 stringStream msg_stream;
157 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
158 vmIntrinsics::name_at(intrinsic_id()),
159 is_virtual() ? " (virtual)" : "", bci);
160 const char *msg = msg_stream.freeze();
161 log_debug(jit, inlining)("%s", msg);
162 if (C->print_intrinsics() || C->print_inlining()) {
163 tty->print("%s", msg);
164 }
165 }
166 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
167 C->print_inlining_update(this);
168
169 return nullptr;
170 }
171
172 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
173 LibraryCallKit kit(jvms, this);
174 Compile* C = kit.C;
175 int nodes = C->unique();
176 _last_predicate = predicate;
177 #ifndef PRODUCT
178 assert(is_predicated() && predicate < predicates_count(), "sanity");
179 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
180 char buf[1000];
181 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
182 tty->print_cr("Predicate for intrinsic %s", str);
183 }
184 #endif
185 ciMethod* callee = kit.callee();
186 const int bci = kit.bci();
187
188 Node* slow_ctl = kit.try_to_predicate(predicate);
189 if (!kit.failing()) {
190 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
191 : "(intrinsic, predicate)";
192 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
193 if (C->print_intrinsics() || C->print_inlining()) {
194 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
195 }
196 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
197 if (C->log()) {
198 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
199 vmIntrinsics::name_at(intrinsic_id()),
200 (is_virtual() ? " virtual='1'" : ""),
201 C->unique() - nodes);
202 }
203 return slow_ctl; // Could be null if the check folds.
204 }
205
206 // The intrinsic bailed out
207 if (jvms->has_method()) {
208 // Not a root compile.
209 const char* msg = "failed to generate predicate for intrinsic";
210 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
211 if (C->print_intrinsics() || C->print_inlining()) {
212 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
213 }
214 } else {
215 // Root compile
216 ResourceMark rm;
217 stringStream msg_stream;
218 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
219 vmIntrinsics::name_at(intrinsic_id()),
220 is_virtual() ? " (virtual)" : "", bci);
221 const char *msg = msg_stream.freeze();
222 log_debug(jit, inlining)("%s", msg);
223 if (C->print_intrinsics() || C->print_inlining()) {
224 C->print_inlining_stream()->print("%s", msg);
225 }
226 }
227 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
228 return nullptr;
229 }
230
231 bool LibraryCallKit::try_to_inline(int predicate) {
232 // Handle symbolic names for otherwise undistinguished boolean switches:
233 const bool is_store = true;
234 const bool is_compress = true;
235 const bool is_static = true;
236 const bool is_volatile = true;
237
238 if (!jvms()->has_method()) {
239 // Root JVMState has a null method.
240 assert(map()->memory()->Opcode() == Op_Parm, "");
241 // Insert the memory aliasing node
242 set_all_memory(reset_memory());
243 }
244 assert(merged_memory(), "");
245
246 switch (intrinsic_id()) {
247 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
248 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
249 case vmIntrinsics::_getClass: return inline_native_getClass();
250
251 case vmIntrinsics::_ceil:
252 case vmIntrinsics::_floor:
253 case vmIntrinsics::_rint:
254 case vmIntrinsics::_dsin:
255 case vmIntrinsics::_dcos:
256 case vmIntrinsics::_dtan:
257 case vmIntrinsics::_dtanh:
258 case vmIntrinsics::_dabs:
259 case vmIntrinsics::_fabs:
260 case vmIntrinsics::_iabs:
261 case vmIntrinsics::_labs:
262 case vmIntrinsics::_datan2:
263 case vmIntrinsics::_dsqrt:
264 case vmIntrinsics::_dsqrt_strict:
265 case vmIntrinsics::_dexp:
266 case vmIntrinsics::_dlog:
267 case vmIntrinsics::_dlog10:
268 case vmIntrinsics::_dpow:
269 case vmIntrinsics::_dcopySign:
270 case vmIntrinsics::_fcopySign:
271 case vmIntrinsics::_dsignum:
272 case vmIntrinsics::_roundF:
273 case vmIntrinsics::_roundD:
274 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
275
276 case vmIntrinsics::_notify:
277 case vmIntrinsics::_notifyAll:
278 return inline_notify(intrinsic_id());
279
280 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
281 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
282 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
283 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
284 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
285 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
286 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
287 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
288 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
289 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
290 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
291 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
292 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
293 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
294
295 case vmIntrinsics::_arraycopy: return inline_arraycopy();
296
297 case vmIntrinsics::_arraySort: return inline_array_sort();
298 case vmIntrinsics::_arrayPartition: return inline_array_partition();
299
300 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
301 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
302 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
303 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
304
305 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
306 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
307 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
308 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
309 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
310 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
311 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
312 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
313
314 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
315
316 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
317
318 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
319 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
320 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
321 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
322
323 case vmIntrinsics::_compressStringC:
324 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
325 case vmIntrinsics::_inflateStringC:
326 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
327
328 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
329 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
330 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
331 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
332 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
333 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
334 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
335 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
336 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
337
338 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
339 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
340 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
341 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
342 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
343 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
344 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
345 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
346 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
347
348 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
349 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
350 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
351 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
352 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
353 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
354 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
355 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
356 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
357
358 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
359 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
360 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
361 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
362 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
363 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
364 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
365 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
366 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
367
368 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
369 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
370 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
371 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
372
373 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
374 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
375 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
376 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
377
378 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
379 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
380 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
381 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
382 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
383 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
384 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
385 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
386 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
387
388 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
389 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
390 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
391 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
392 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
393 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
394 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
395 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
396 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
397
398 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
399 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
400 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
401 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
402 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
403 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
404 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
405 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
406 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
407
408 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
409 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
410 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
411 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
412 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
413 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
414 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
415 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
416 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
417
418 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
419 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
420 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
421 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
422 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
423
424 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
425 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
426 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
427 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
428 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
429 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
430 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
431 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
432 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
433 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
434 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
435 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
436 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
437 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
438 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
439 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
440 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
441 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
442 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
443 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
444
445 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
446 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
447 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
448 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
449 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
450 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
451 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
452 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
453 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
454 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
455 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
456 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
457 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
458 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
459 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
460
461 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
462 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
463 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
464 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
465
466 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
467 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
468 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
469 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
470 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
471
472 case vmIntrinsics::_loadFence:
473 case vmIntrinsics::_storeFence:
474 case vmIntrinsics::_storeStoreFence:
475 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
476
477 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
478
479 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
480 case vmIntrinsics::_currentThread: return inline_native_currentThread();
481 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
482
483 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
484 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
485
486 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
487 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
488
489 #if INCLUDE_JVMTI
490 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
491 "notifyJvmtiStart", true, false);
492 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
493 "notifyJvmtiEnd", false, true);
494 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
495 "notifyJvmtiMount", false, false);
496 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
497 "notifyJvmtiUnmount", false, false);
498 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
499 #endif
500
501 #ifdef JFR_HAVE_INTRINSICS
502 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
503 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
504 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
505 #endif
506 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
507 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
508 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
509 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
510 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
511 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
512 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
513 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
514 case vmIntrinsics::_getLength: return inline_native_getLength();
515 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
516 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
517 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
518 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
519 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
520 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
521 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
522
523 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
524 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
525
526 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
527
528 case vmIntrinsics::_isInstance:
529 case vmIntrinsics::_getModifiers:
530 case vmIntrinsics::_isInterface:
531 case vmIntrinsics::_isArray:
532 case vmIntrinsics::_isPrimitive:
533 case vmIntrinsics::_isHidden:
534 case vmIntrinsics::_getSuperclass:
535 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
536
537 case vmIntrinsics::_floatToRawIntBits:
538 case vmIntrinsics::_floatToIntBits:
539 case vmIntrinsics::_intBitsToFloat:
540 case vmIntrinsics::_doubleToRawLongBits:
541 case vmIntrinsics::_doubleToLongBits:
542 case vmIntrinsics::_longBitsToDouble:
543 case vmIntrinsics::_floatToFloat16:
544 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
545
546 case vmIntrinsics::_floatIsFinite:
547 case vmIntrinsics::_floatIsInfinite:
548 case vmIntrinsics::_doubleIsFinite:
549 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
550
551 case vmIntrinsics::_numberOfLeadingZeros_i:
552 case vmIntrinsics::_numberOfLeadingZeros_l:
553 case vmIntrinsics::_numberOfTrailingZeros_i:
554 case vmIntrinsics::_numberOfTrailingZeros_l:
555 case vmIntrinsics::_bitCount_i:
556 case vmIntrinsics::_bitCount_l:
557 case vmIntrinsics::_reverse_i:
558 case vmIntrinsics::_reverse_l:
559 case vmIntrinsics::_reverseBytes_i:
560 case vmIntrinsics::_reverseBytes_l:
561 case vmIntrinsics::_reverseBytes_s:
562 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
563
564 case vmIntrinsics::_compress_i:
565 case vmIntrinsics::_compress_l:
566 case vmIntrinsics::_expand_i:
567 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
568
569 case vmIntrinsics::_compareUnsigned_i:
570 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
571
572 case vmIntrinsics::_divideUnsigned_i:
573 case vmIntrinsics::_divideUnsigned_l:
574 case vmIntrinsics::_remainderUnsigned_i:
575 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
576
577 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
578
579 case vmIntrinsics::_Reference_get: return inline_reference_get();
580 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
581 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
582 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
583 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
584
585 case vmIntrinsics::_Class_cast: return inline_Class_cast();
586
587 case vmIntrinsics::_aescrypt_encryptBlock:
588 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
589
590 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
591 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
592 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
593
594 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
595 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
596 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
597
598 case vmIntrinsics::_counterMode_AESCrypt:
599 return inline_counterMode_AESCrypt(intrinsic_id());
600
601 case vmIntrinsics::_galoisCounterMode_AESCrypt:
602 return inline_galoisCounterMode_AESCrypt();
603
604 case vmIntrinsics::_md5_implCompress:
605 case vmIntrinsics::_sha_implCompress:
606 case vmIntrinsics::_sha2_implCompress:
607 case vmIntrinsics::_sha5_implCompress:
608 case vmIntrinsics::_sha3_implCompress:
609 return inline_digestBase_implCompress(intrinsic_id());
610
611 case vmIntrinsics::_digestBase_implCompressMB:
612 return inline_digestBase_implCompressMB(predicate);
613
614 case vmIntrinsics::_multiplyToLen:
615 return inline_multiplyToLen();
616
617 case vmIntrinsics::_squareToLen:
618 return inline_squareToLen();
619
620 case vmIntrinsics::_mulAdd:
621 return inline_mulAdd();
622
623 case vmIntrinsics::_montgomeryMultiply:
624 return inline_montgomeryMultiply();
625 case vmIntrinsics::_montgomerySquare:
626 return inline_montgomerySquare();
627
628 case vmIntrinsics::_bigIntegerRightShiftWorker:
629 return inline_bigIntegerShift(true);
630 case vmIntrinsics::_bigIntegerLeftShiftWorker:
631 return inline_bigIntegerShift(false);
632
633 case vmIntrinsics::_vectorizedMismatch:
634 return inline_vectorizedMismatch();
635
636 case vmIntrinsics::_ghash_processBlocks:
637 return inline_ghash_processBlocks();
638 case vmIntrinsics::_chacha20Block:
639 return inline_chacha20Block();
640 case vmIntrinsics::_base64_encodeBlock:
641 return inline_base64_encodeBlock();
642 case vmIntrinsics::_base64_decodeBlock:
643 return inline_base64_decodeBlock();
644 case vmIntrinsics::_poly1305_processBlocks:
645 return inline_poly1305_processBlocks();
646 case vmIntrinsics::_intpoly_montgomeryMult_P256:
647 return inline_intpoly_montgomeryMult_P256();
648 case vmIntrinsics::_intpoly_assign:
649 return inline_intpoly_assign();
650 case vmIntrinsics::_encodeISOArray:
651 case vmIntrinsics::_encodeByteISOArray:
652 return inline_encodeISOArray(false);
653 case vmIntrinsics::_encodeAsciiArray:
654 return inline_encodeISOArray(true);
655
656 case vmIntrinsics::_updateCRC32:
657 return inline_updateCRC32();
658 case vmIntrinsics::_updateBytesCRC32:
659 return inline_updateBytesCRC32();
660 case vmIntrinsics::_updateByteBufferCRC32:
661 return inline_updateByteBufferCRC32();
662
663 case vmIntrinsics::_updateBytesCRC32C:
664 return inline_updateBytesCRC32C();
665 case vmIntrinsics::_updateDirectByteBufferCRC32C:
666 return inline_updateDirectByteBufferCRC32C();
667
668 case vmIntrinsics::_updateBytesAdler32:
669 return inline_updateBytesAdler32();
670 case vmIntrinsics::_updateByteBufferAdler32:
671 return inline_updateByteBufferAdler32();
672
673 case vmIntrinsics::_profileBoolean:
674 return inline_profileBoolean();
675 case vmIntrinsics::_isCompileConstant:
676 return inline_isCompileConstant();
677
678 case vmIntrinsics::_countPositives:
679 return inline_countPositives();
680
681 case vmIntrinsics::_fmaD:
682 case vmIntrinsics::_fmaF:
683 return inline_fma(intrinsic_id());
684
685 case vmIntrinsics::_isDigit:
686 case vmIntrinsics::_isLowerCase:
687 case vmIntrinsics::_isUpperCase:
688 case vmIntrinsics::_isWhitespace:
689 return inline_character_compare(intrinsic_id());
690
691 case vmIntrinsics::_min:
692 case vmIntrinsics::_max:
693 case vmIntrinsics::_min_strict:
694 case vmIntrinsics::_max_strict:
695 return inline_min_max(intrinsic_id());
696
697 case vmIntrinsics::_maxF:
698 case vmIntrinsics::_minF:
699 case vmIntrinsics::_maxD:
700 case vmIntrinsics::_minD:
701 case vmIntrinsics::_maxF_strict:
702 case vmIntrinsics::_minF_strict:
703 case vmIntrinsics::_maxD_strict:
704 case vmIntrinsics::_minD_strict:
705 return inline_fp_min_max(intrinsic_id());
706
707 case vmIntrinsics::_VectorUnaryOp:
708 return inline_vector_nary_operation(1);
709 case vmIntrinsics::_VectorBinaryOp:
710 return inline_vector_nary_operation(2);
711 case vmIntrinsics::_VectorTernaryOp:
712 return inline_vector_nary_operation(3);
713 case vmIntrinsics::_VectorFromBitsCoerced:
714 return inline_vector_frombits_coerced();
715 case vmIntrinsics::_VectorShuffleIota:
716 return inline_vector_shuffle_iota();
717 case vmIntrinsics::_VectorMaskOp:
718 return inline_vector_mask_operation();
719 case vmIntrinsics::_VectorShuffleToVector:
720 return inline_vector_shuffle_to_vector();
721 case vmIntrinsics::_VectorWrapShuffleIndexes:
722 return inline_vector_wrap_shuffle_indexes();
723 case vmIntrinsics::_VectorLoadOp:
724 return inline_vector_mem_operation(/*is_store=*/false);
725 case vmIntrinsics::_VectorLoadMaskedOp:
726 return inline_vector_mem_masked_operation(/*is_store*/false);
727 case vmIntrinsics::_VectorStoreOp:
728 return inline_vector_mem_operation(/*is_store=*/true);
729 case vmIntrinsics::_VectorStoreMaskedOp:
730 return inline_vector_mem_masked_operation(/*is_store=*/true);
731 case vmIntrinsics::_VectorGatherOp:
732 return inline_vector_gather_scatter(/*is_scatter*/ false);
733 case vmIntrinsics::_VectorScatterOp:
734 return inline_vector_gather_scatter(/*is_scatter*/ true);
735 case vmIntrinsics::_VectorReductionCoerced:
736 return inline_vector_reduction();
737 case vmIntrinsics::_VectorTest:
738 return inline_vector_test();
739 case vmIntrinsics::_VectorBlend:
740 return inline_vector_blend();
741 case vmIntrinsics::_VectorRearrange:
742 return inline_vector_rearrange();
743 case vmIntrinsics::_VectorSelectFrom:
744 return inline_vector_select_from();
745 case vmIntrinsics::_VectorCompare:
746 return inline_vector_compare();
747 case vmIntrinsics::_VectorBroadcastInt:
748 return inline_vector_broadcast_int();
749 case vmIntrinsics::_VectorConvert:
750 return inline_vector_convert();
751 case vmIntrinsics::_VectorInsert:
752 return inline_vector_insert();
753 case vmIntrinsics::_VectorExtract:
754 return inline_vector_extract();
755 case vmIntrinsics::_VectorCompressExpand:
756 return inline_vector_compress_expand();
757 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
758 return inline_vector_select_from_two_vectors();
759 case vmIntrinsics::_IndexVector:
760 return inline_index_vector();
761 case vmIntrinsics::_IndexPartiallyInUpperRange:
762 return inline_index_partially_in_upper_range();
763
764 case vmIntrinsics::_getObjectSize:
765 return inline_getObjectSize();
766
767 case vmIntrinsics::_blackhole:
768 return inline_blackhole();
769
770 default:
771 // If you get here, it may be that someone has added a new intrinsic
772 // to the list in vmIntrinsics.hpp without implementing it here.
773 #ifndef PRODUCT
774 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
775 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
776 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
777 }
778 #endif
779 return false;
780 }
781 }
782
783 Node* LibraryCallKit::try_to_predicate(int predicate) {
784 if (!jvms()->has_method()) {
785 // Root JVMState has a null method.
786 assert(map()->memory()->Opcode() == Op_Parm, "");
787 // Insert the memory aliasing node
788 set_all_memory(reset_memory());
789 }
790 assert(merged_memory(), "");
791
792 switch (intrinsic_id()) {
793 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
794 return inline_cipherBlockChaining_AESCrypt_predicate(false);
795 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
796 return inline_cipherBlockChaining_AESCrypt_predicate(true);
797 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
798 return inline_electronicCodeBook_AESCrypt_predicate(false);
799 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
800 return inline_electronicCodeBook_AESCrypt_predicate(true);
801 case vmIntrinsics::_counterMode_AESCrypt:
802 return inline_counterMode_AESCrypt_predicate();
803 case vmIntrinsics::_digestBase_implCompressMB:
804 return inline_digestBase_implCompressMB_predicate(predicate);
805 case vmIntrinsics::_galoisCounterMode_AESCrypt:
806 return inline_galoisCounterMode_AESCrypt_predicate();
807
808 default:
809 // If you get here, it may be that someone has added a new intrinsic
810 // to the list in vmIntrinsics.hpp without implementing it here.
811 #ifndef PRODUCT
812 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
813 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
814 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
815 }
816 #endif
817 Node* slow_ctl = control();
818 set_control(top()); // No fast path intrinsic
819 return slow_ctl;
820 }
821 }
822
823 //------------------------------set_result-------------------------------
824 // Helper function for finishing intrinsics.
825 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
826 record_for_igvn(region);
827 set_control(_gvn.transform(region));
828 set_result( _gvn.transform(value));
829 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
830 }
831
832 //------------------------------generate_guard---------------------------
833 // Helper function for generating guarded fast-slow graph structures.
834 // The given 'test', if true, guards a slow path. If the test fails
835 // then a fast path can be taken. (We generally hope it fails.)
836 // In all cases, GraphKit::control() is updated to the fast path.
837 // The returned value represents the control for the slow path.
838 // The return value is never 'top'; it is either a valid control
839 // or null if it is obvious that the slow path can never be taken.
840 // Also, if region and the slow control are not null, the slow edge
841 // is appended to the region.
842 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
843 if (stopped()) {
844 // Already short circuited.
845 return nullptr;
846 }
847
848 // Build an if node and its projections.
849 // If test is true we take the slow path, which we assume is uncommon.
850 if (_gvn.type(test) == TypeInt::ZERO) {
851 // The slow branch is never taken. No need to build this guard.
852 return nullptr;
853 }
854
855 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
856
857 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
858 if (if_slow == top()) {
859 // The slow branch is never taken. No need to build this guard.
860 return nullptr;
861 }
862
863 if (region != nullptr)
864 region->add_req(if_slow);
865
866 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
867 set_control(if_fast);
868
869 return if_slow;
870 }
871
872 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
873 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
874 }
875 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
876 return generate_guard(test, region, PROB_FAIR);
877 }
878
879 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
880 Node* *pos_index) {
881 if (stopped())
882 return nullptr; // already stopped
883 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
884 return nullptr; // index is already adequately typed
885 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
886 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
887 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
888 if (is_neg != nullptr && pos_index != nullptr) {
889 // Emulate effect of Parse::adjust_map_after_if.
890 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
891 (*pos_index) = _gvn.transform(ccast);
892 }
893 return is_neg;
894 }
895
896 // Make sure that 'position' is a valid limit index, in [0..length].
897 // There are two equivalent plans for checking this:
898 // A. (offset + copyLength) unsigned<= arrayLength
899 // B. offset <= (arrayLength - copyLength)
900 // We require that all of the values above, except for the sum and
901 // difference, are already known to be non-negative.
902 // Plan A is robust in the face of overflow, if offset and copyLength
903 // are both hugely positive.
904 //
905 // Plan B is less direct and intuitive, but it does not overflow at
906 // all, since the difference of two non-negatives is always
907 // representable. Whenever Java methods must perform the equivalent
908 // check they generally use Plan B instead of Plan A.
909 // For the moment we use Plan A.
910 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
911 Node* subseq_length,
912 Node* array_length,
913 RegionNode* region) {
914 if (stopped())
915 return nullptr; // already stopped
916 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
917 if (zero_offset && subseq_length->eqv_uncast(array_length))
918 return nullptr; // common case of whole-array copy
919 Node* last = subseq_length;
920 if (!zero_offset) // last += offset
921 last = _gvn.transform(new AddINode(last, offset));
922 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
923 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
924 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
925 return is_over;
926 }
927
928 // Emit range checks for the given String.value byte array
929 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
930 if (stopped()) {
931 return; // already stopped
932 }
933 RegionNode* bailout = new RegionNode(1);
934 record_for_igvn(bailout);
935 if (char_count) {
936 // Convert char count to byte count
937 count = _gvn.transform(new LShiftINode(count, intcon(1)));
938 }
939
940 // Offset and count must not be negative
941 generate_negative_guard(offset, bailout);
942 generate_negative_guard(count, bailout);
943 // Offset + count must not exceed length of array
944 generate_limit_guard(offset, count, load_array_length(array), bailout);
945
946 if (bailout->req() > 1) {
947 PreserveJVMState pjvms(this);
948 set_control(_gvn.transform(bailout));
949 uncommon_trap(Deoptimization::Reason_intrinsic,
950 Deoptimization::Action_maybe_recompile);
951 }
952 }
953
954 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
955 bool is_immutable) {
956 ciKlass* thread_klass = env()->Thread_klass();
957 const Type* thread_type
958 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
959
960 Node* thread = _gvn.transform(new ThreadLocalNode());
961 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
962 tls_output = thread;
963
964 Node* thread_obj_handle
965 = (is_immutable
966 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
967 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
968 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
969 thread_obj_handle = _gvn.transform(thread_obj_handle);
970
971 DecoratorSet decorators = IN_NATIVE;
972 if (is_immutable) {
973 decorators |= C2_IMMUTABLE_MEMORY;
974 }
975 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
976 }
977
978 //--------------------------generate_current_thread--------------------
979 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
980 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
981 /*is_immutable*/false);
982 }
983
984 //--------------------------generate_virtual_thread--------------------
985 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
986 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
987 !C->method()->changes_current_thread());
988 }
989
990 //------------------------------make_string_method_node------------------------
991 // Helper method for String intrinsic functions. This version is called with
992 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
993 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
994 // containing the lengths of str1 and str2.
995 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
996 Node* result = nullptr;
997 switch (opcode) {
998 case Op_StrIndexOf:
999 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1000 str1_start, cnt1, str2_start, cnt2, ae);
1001 break;
1002 case Op_StrComp:
1003 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1004 str1_start, cnt1, str2_start, cnt2, ae);
1005 break;
1006 case Op_StrEquals:
1007 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1008 // Use the constant length if there is one because optimized match rule may exist.
1009 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1010 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1011 break;
1012 default:
1013 ShouldNotReachHere();
1014 return nullptr;
1015 }
1016
1017 // All these intrinsics have checks.
1018 C->set_has_split_ifs(true); // Has chance for split-if optimization
1019 clear_upper_avx();
1020
1021 return _gvn.transform(result);
1022 }
1023
1024 //------------------------------inline_string_compareTo------------------------
1025 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1026 Node* arg1 = argument(0);
1027 Node* arg2 = argument(1);
1028
1029 arg1 = must_be_not_null(arg1, true);
1030 arg2 = must_be_not_null(arg2, true);
1031
1032 // Get start addr and length of first argument
1033 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1034 Node* arg1_cnt = load_array_length(arg1);
1035
1036 // Get start addr and length of second argument
1037 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1038 Node* arg2_cnt = load_array_length(arg2);
1039
1040 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1041 set_result(result);
1042 return true;
1043 }
1044
1045 //------------------------------inline_string_equals------------------------
1046 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1047 Node* arg1 = argument(0);
1048 Node* arg2 = argument(1);
1049
1050 // paths (plus control) merge
1051 RegionNode* region = new RegionNode(3);
1052 Node* phi = new PhiNode(region, TypeInt::BOOL);
1053
1054 if (!stopped()) {
1055
1056 arg1 = must_be_not_null(arg1, true);
1057 arg2 = must_be_not_null(arg2, true);
1058
1059 // Get start addr and length of first argument
1060 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1061 Node* arg1_cnt = load_array_length(arg1);
1062
1063 // Get start addr and length of second argument
1064 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1065 Node* arg2_cnt = load_array_length(arg2);
1066
1067 // Check for arg1_cnt != arg2_cnt
1068 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1069 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1070 Node* if_ne = generate_slow_guard(bol, nullptr);
1071 if (if_ne != nullptr) {
1072 phi->init_req(2, intcon(0));
1073 region->init_req(2, if_ne);
1074 }
1075
1076 // Check for count == 0 is done by assembler code for StrEquals.
1077
1078 if (!stopped()) {
1079 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1080 phi->init_req(1, equals);
1081 region->init_req(1, control());
1082 }
1083 }
1084
1085 // post merge
1086 set_control(_gvn.transform(region));
1087 record_for_igvn(region);
1088
1089 set_result(_gvn.transform(phi));
1090 return true;
1091 }
1092
1093 //------------------------------inline_array_equals----------------------------
1094 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1095 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1096 Node* arg1 = argument(0);
1097 Node* arg2 = argument(1);
1098
1099 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1100 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1101 clear_upper_avx();
1102
1103 return true;
1104 }
1105
1106
1107 //------------------------------inline_countPositives------------------------------
1108 bool LibraryCallKit::inline_countPositives() {
1109 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1110 return false;
1111 }
1112
1113 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1114 // no receiver since it is static method
1115 Node* ba = argument(0);
1116 Node* offset = argument(1);
1117 Node* len = argument(2);
1118
1119 ba = must_be_not_null(ba, true);
1120
1121 // Range checks
1122 generate_string_range_check(ba, offset, len, false);
1123 if (stopped()) {
1124 return true;
1125 }
1126 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1127 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1128 set_result(_gvn.transform(result));
1129 clear_upper_avx();
1130 return true;
1131 }
1132
1133 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1134 Node* index = argument(0);
1135 Node* length = bt == T_INT ? argument(1) : argument(2);
1136 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1137 return false;
1138 }
1139
1140 // check that length is positive
1141 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1142 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1143
1144 {
1145 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1146 uncommon_trap(Deoptimization::Reason_intrinsic,
1147 Deoptimization::Action_make_not_entrant);
1148 }
1149
1150 if (stopped()) {
1151 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1152 return true;
1153 }
1154
1155 // length is now known positive, add a cast node to make this explicit
1156 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1157 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1158 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1159 ConstraintCastNode::RegularDependency, bt);
1160 casted_length = _gvn.transform(casted_length);
1161 replace_in_map(length, casted_length);
1162 length = casted_length;
1163
1164 // Use an unsigned comparison for the range check itself
1165 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1166 BoolTest::mask btest = BoolTest::lt;
1167 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1168 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1169 _gvn.set_type(rc, rc->Value(&_gvn));
1170 if (!rc_bool->is_Con()) {
1171 record_for_igvn(rc);
1172 }
1173 set_control(_gvn.transform(new IfTrueNode(rc)));
1174 {
1175 PreserveJVMState pjvms(this);
1176 set_control(_gvn.transform(new IfFalseNode(rc)));
1177 uncommon_trap(Deoptimization::Reason_range_check,
1178 Deoptimization::Action_make_not_entrant);
1179 }
1180
1181 if (stopped()) {
1182 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1183 return true;
1184 }
1185
1186 // index is now known to be >= 0 and < length, cast it
1187 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1188 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1189 ConstraintCastNode::RegularDependency, bt);
1190 result = _gvn.transform(result);
1191 set_result(result);
1192 replace_in_map(index, result);
1193 return true;
1194 }
1195
1196 //------------------------------inline_string_indexOf------------------------
1197 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1198 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1199 return false;
1200 }
1201 Node* src = argument(0);
1202 Node* tgt = argument(1);
1203
1204 // Make the merge point
1205 RegionNode* result_rgn = new RegionNode(4);
1206 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1207
1208 src = must_be_not_null(src, true);
1209 tgt = must_be_not_null(tgt, true);
1210
1211 // Get start addr and length of source string
1212 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1213 Node* src_count = load_array_length(src);
1214
1215 // Get start addr and length of substring
1216 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1217 Node* tgt_count = load_array_length(tgt);
1218
1219 Node* result = nullptr;
1220 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1221
1222 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1223 // Divide src size by 2 if String is UTF16 encoded
1224 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1225 }
1226 if (ae == StrIntrinsicNode::UU) {
1227 // Divide substring size by 2 if String is UTF16 encoded
1228 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1229 }
1230
1231 if (call_opt_stub) {
1232 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1233 StubRoutines::_string_indexof_array[ae],
1234 "stringIndexOf", TypePtr::BOTTOM, src_start,
1235 src_count, tgt_start, tgt_count);
1236 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1237 } else {
1238 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1239 result_rgn, result_phi, ae);
1240 }
1241 if (result != nullptr) {
1242 result_phi->init_req(3, result);
1243 result_rgn->init_req(3, control());
1244 }
1245 set_control(_gvn.transform(result_rgn));
1246 record_for_igvn(result_rgn);
1247 set_result(_gvn.transform(result_phi));
1248
1249 return true;
1250 }
1251
1252 //-----------------------------inline_string_indexOfI-----------------------
1253 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1254 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1255 return false;
1256 }
1257 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1258 return false;
1259 }
1260
1261 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1262 Node* src = argument(0); // byte[]
1263 Node* src_count = argument(1); // char count
1264 Node* tgt = argument(2); // byte[]
1265 Node* tgt_count = argument(3); // char count
1266 Node* from_index = argument(4); // char index
1267
1268 src = must_be_not_null(src, true);
1269 tgt = must_be_not_null(tgt, true);
1270
1271 // Multiply byte array index by 2 if String is UTF16 encoded
1272 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1273 src_count = _gvn.transform(new SubINode(src_count, from_index));
1274 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1275 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1276
1277 // Range checks
1278 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1279 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1280 if (stopped()) {
1281 return true;
1282 }
1283
1284 RegionNode* region = new RegionNode(5);
1285 Node* phi = new PhiNode(region, TypeInt::INT);
1286 Node* result = nullptr;
1287
1288 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1289
1290 if (call_opt_stub) {
1291 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1292 StubRoutines::_string_indexof_array[ae],
1293 "stringIndexOf", TypePtr::BOTTOM, src_start,
1294 src_count, tgt_start, tgt_count);
1295 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1296 } else {
1297 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1298 region, phi, ae);
1299 }
1300 if (result != nullptr) {
1301 // The result is index relative to from_index if substring was found, -1 otherwise.
1302 // Generate code which will fold into cmove.
1303 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1304 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1305
1306 Node* if_lt = generate_slow_guard(bol, nullptr);
1307 if (if_lt != nullptr) {
1308 // result == -1
1309 phi->init_req(3, result);
1310 region->init_req(3, if_lt);
1311 }
1312 if (!stopped()) {
1313 result = _gvn.transform(new AddINode(result, from_index));
1314 phi->init_req(4, result);
1315 region->init_req(4, control());
1316 }
1317 }
1318
1319 set_control(_gvn.transform(region));
1320 record_for_igvn(region);
1321 set_result(_gvn.transform(phi));
1322 clear_upper_avx();
1323
1324 return true;
1325 }
1326
1327 // Create StrIndexOfNode with fast path checks
1328 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1329 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1330 // Check for substr count > string count
1331 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1332 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1333 Node* if_gt = generate_slow_guard(bol, nullptr);
1334 if (if_gt != nullptr) {
1335 phi->init_req(1, intcon(-1));
1336 region->init_req(1, if_gt);
1337 }
1338 if (!stopped()) {
1339 // Check for substr count == 0
1340 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1341 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1342 Node* if_zero = generate_slow_guard(bol, nullptr);
1343 if (if_zero != nullptr) {
1344 phi->init_req(2, intcon(0));
1345 region->init_req(2, if_zero);
1346 }
1347 }
1348 if (!stopped()) {
1349 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1350 }
1351 return nullptr;
1352 }
1353
1354 //-----------------------------inline_string_indexOfChar-----------------------
1355 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1356 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1357 return false;
1358 }
1359 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1360 return false;
1361 }
1362 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1363 Node* src = argument(0); // byte[]
1364 Node* int_ch = argument(1);
1365 Node* from_index = argument(2);
1366 Node* max = argument(3);
1367
1368 src = must_be_not_null(src, true);
1369
1370 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1371 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1372 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1373
1374 // Range checks
1375 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1376
1377 // Check for int_ch >= 0
1378 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1379 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1380 {
1381 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1382 uncommon_trap(Deoptimization::Reason_intrinsic,
1383 Deoptimization::Action_maybe_recompile);
1384 }
1385 if (stopped()) {
1386 return true;
1387 }
1388
1389 RegionNode* region = new RegionNode(3);
1390 Node* phi = new PhiNode(region, TypeInt::INT);
1391
1392 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1393 C->set_has_split_ifs(true); // Has chance for split-if optimization
1394 _gvn.transform(result);
1395
1396 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1397 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1398
1399 Node* if_lt = generate_slow_guard(bol, nullptr);
1400 if (if_lt != nullptr) {
1401 // result == -1
1402 phi->init_req(2, result);
1403 region->init_req(2, if_lt);
1404 }
1405 if (!stopped()) {
1406 result = _gvn.transform(new AddINode(result, from_index));
1407 phi->init_req(1, result);
1408 region->init_req(1, control());
1409 }
1410 set_control(_gvn.transform(region));
1411 record_for_igvn(region);
1412 set_result(_gvn.transform(phi));
1413 clear_upper_avx();
1414
1415 return true;
1416 }
1417 //---------------------------inline_string_copy---------------------
1418 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1419 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1420 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1421 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1422 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1423 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1424 bool LibraryCallKit::inline_string_copy(bool compress) {
1425 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1426 return false;
1427 }
1428 int nargs = 5; // 2 oops, 3 ints
1429 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1430
1431 Node* src = argument(0);
1432 Node* src_offset = argument(1);
1433 Node* dst = argument(2);
1434 Node* dst_offset = argument(3);
1435 Node* length = argument(4);
1436
1437 // Check for allocation before we add nodes that would confuse
1438 // tightly_coupled_allocation()
1439 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1440
1441 // Figure out the size and type of the elements we will be copying.
1442 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1443 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1444 if (src_type == nullptr || dst_type == nullptr) {
1445 return false;
1446 }
1447 BasicType src_elem = src_type->elem()->array_element_basic_type();
1448 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1449 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1450 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1451 "Unsupported array types for inline_string_copy");
1452
1453 src = must_be_not_null(src, true);
1454 dst = must_be_not_null(dst, true);
1455
1456 // Convert char[] offsets to byte[] offsets
1457 bool convert_src = (compress && src_elem == T_BYTE);
1458 bool convert_dst = (!compress && dst_elem == T_BYTE);
1459 if (convert_src) {
1460 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1461 } else if (convert_dst) {
1462 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1463 }
1464
1465 // Range checks
1466 generate_string_range_check(src, src_offset, length, convert_src);
1467 generate_string_range_check(dst, dst_offset, length, convert_dst);
1468 if (stopped()) {
1469 return true;
1470 }
1471
1472 Node* src_start = array_element_address(src, src_offset, src_elem);
1473 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1474 // 'src_start' points to src array + scaled offset
1475 // 'dst_start' points to dst array + scaled offset
1476 Node* count = nullptr;
1477 if (compress) {
1478 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1479 } else {
1480 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1481 }
1482
1483 if (alloc != nullptr) {
1484 if (alloc->maybe_set_complete(&_gvn)) {
1485 // "You break it, you buy it."
1486 InitializeNode* init = alloc->initialization();
1487 assert(init->is_complete(), "we just did this");
1488 init->set_complete_with_arraycopy();
1489 assert(dst->is_CheckCastPP(), "sanity");
1490 assert(dst->in(0)->in(0) == init, "dest pinned");
1491 }
1492 // Do not let stores that initialize this object be reordered with
1493 // a subsequent store that would make this object accessible by
1494 // other threads.
1495 // Record what AllocateNode this StoreStore protects so that
1496 // escape analysis can go from the MemBarStoreStoreNode to the
1497 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1498 // based on the escape status of the AllocateNode.
1499 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1500 }
1501 if (compress) {
1502 set_result(_gvn.transform(count));
1503 }
1504 clear_upper_avx();
1505
1506 return true;
1507 }
1508
1509 #ifdef _LP64
1510 #define XTOP ,top() /*additional argument*/
1511 #else //_LP64
1512 #define XTOP /*no additional argument*/
1513 #endif //_LP64
1514
1515 //------------------------inline_string_toBytesU--------------------------
1516 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1517 bool LibraryCallKit::inline_string_toBytesU() {
1518 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1519 return false;
1520 }
1521 // Get the arguments.
1522 Node* value = argument(0);
1523 Node* offset = argument(1);
1524 Node* length = argument(2);
1525
1526 Node* newcopy = nullptr;
1527
1528 // Set the original stack and the reexecute bit for the interpreter to reexecute
1529 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1530 { PreserveReexecuteState preexecs(this);
1531 jvms()->set_should_reexecute(true);
1532
1533 // Check if a null path was taken unconditionally.
1534 value = null_check(value);
1535
1536 RegionNode* bailout = new RegionNode(1);
1537 record_for_igvn(bailout);
1538
1539 // Range checks
1540 generate_negative_guard(offset, bailout);
1541 generate_negative_guard(length, bailout);
1542 generate_limit_guard(offset, length, load_array_length(value), bailout);
1543 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1544 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1545
1546 if (bailout->req() > 1) {
1547 PreserveJVMState pjvms(this);
1548 set_control(_gvn.transform(bailout));
1549 uncommon_trap(Deoptimization::Reason_intrinsic,
1550 Deoptimization::Action_maybe_recompile);
1551 }
1552 if (stopped()) {
1553 return true;
1554 }
1555
1556 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1557 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1558 newcopy = new_array(klass_node, size, 0); // no arguments to push
1559 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1560 guarantee(alloc != nullptr, "created above");
1561
1562 // Calculate starting addresses.
1563 Node* src_start = array_element_address(value, offset, T_CHAR);
1564 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1565
1566 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1567 const TypeInt* toffset = gvn().type(offset)->is_int();
1568 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1569
1570 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1571 const char* copyfunc_name = "arraycopy";
1572 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1573 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1574 OptoRuntime::fast_arraycopy_Type(),
1575 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1576 src_start, dst_start, ConvI2X(length) XTOP);
1577 // Do not let reads from the cloned object float above the arraycopy.
1578 if (alloc->maybe_set_complete(&_gvn)) {
1579 // "You break it, you buy it."
1580 InitializeNode* init = alloc->initialization();
1581 assert(init->is_complete(), "we just did this");
1582 init->set_complete_with_arraycopy();
1583 assert(newcopy->is_CheckCastPP(), "sanity");
1584 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1585 }
1586 // Do not let stores that initialize this object be reordered with
1587 // a subsequent store that would make this object accessible by
1588 // other threads.
1589 // Record what AllocateNode this StoreStore protects so that
1590 // escape analysis can go from the MemBarStoreStoreNode to the
1591 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1592 // based on the escape status of the AllocateNode.
1593 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1594 } // original reexecute is set back here
1595
1596 C->set_has_split_ifs(true); // Has chance for split-if optimization
1597 if (!stopped()) {
1598 set_result(newcopy);
1599 }
1600 clear_upper_avx();
1601
1602 return true;
1603 }
1604
1605 //------------------------inline_string_getCharsU--------------------------
1606 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1607 bool LibraryCallKit::inline_string_getCharsU() {
1608 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1609 return false;
1610 }
1611
1612 // Get the arguments.
1613 Node* src = argument(0);
1614 Node* src_begin = argument(1);
1615 Node* src_end = argument(2); // exclusive offset (i < src_end)
1616 Node* dst = argument(3);
1617 Node* dst_begin = argument(4);
1618
1619 // Check for allocation before we add nodes that would confuse
1620 // tightly_coupled_allocation()
1621 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1622
1623 // Check if a null path was taken unconditionally.
1624 src = null_check(src);
1625 dst = null_check(dst);
1626 if (stopped()) {
1627 return true;
1628 }
1629
1630 // Get length and convert char[] offset to byte[] offset
1631 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1632 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1633
1634 // Range checks
1635 generate_string_range_check(src, src_begin, length, true);
1636 generate_string_range_check(dst, dst_begin, length, false);
1637 if (stopped()) {
1638 return true;
1639 }
1640
1641 if (!stopped()) {
1642 // Calculate starting addresses.
1643 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1644 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1645
1646 // Check if array addresses are aligned to HeapWordSize
1647 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1648 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1649 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1650 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1651
1652 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1653 const char* copyfunc_name = "arraycopy";
1654 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1655 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1656 OptoRuntime::fast_arraycopy_Type(),
1657 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1658 src_start, dst_start, ConvI2X(length) XTOP);
1659 // Do not let reads from the cloned object float above the arraycopy.
1660 if (alloc != nullptr) {
1661 if (alloc->maybe_set_complete(&_gvn)) {
1662 // "You break it, you buy it."
1663 InitializeNode* init = alloc->initialization();
1664 assert(init->is_complete(), "we just did this");
1665 init->set_complete_with_arraycopy();
1666 assert(dst->is_CheckCastPP(), "sanity");
1667 assert(dst->in(0)->in(0) == init, "dest pinned");
1668 }
1669 // Do not let stores that initialize this object be reordered with
1670 // a subsequent store that would make this object accessible by
1671 // other threads.
1672 // Record what AllocateNode this StoreStore protects so that
1673 // escape analysis can go from the MemBarStoreStoreNode to the
1674 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1675 // based on the escape status of the AllocateNode.
1676 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1677 } else {
1678 insert_mem_bar(Op_MemBarCPUOrder);
1679 }
1680 }
1681
1682 C->set_has_split_ifs(true); // Has chance for split-if optimization
1683 return true;
1684 }
1685
1686 //----------------------inline_string_char_access----------------------------
1687 // Store/Load char to/from byte[] array.
1688 // static void StringUTF16.putChar(byte[] val, int index, int c)
1689 // static char StringUTF16.getChar(byte[] val, int index)
1690 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1691 Node* value = argument(0);
1692 Node* index = argument(1);
1693 Node* ch = is_store ? argument(2) : nullptr;
1694
1695 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1696 // correctly requires matched array shapes.
1697 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1698 "sanity: byte[] and char[] bases agree");
1699 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1700 "sanity: byte[] and char[] scales agree");
1701
1702 // Bail when getChar over constants is requested: constant folding would
1703 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1704 // Java method would constant fold nicely instead.
1705 if (!is_store && value->is_Con() && index->is_Con()) {
1706 return false;
1707 }
1708
1709 // Save state and restore on bailout
1710 uint old_sp = sp();
1711 SafePointNode* old_map = clone_map();
1712
1713 value = must_be_not_null(value, true);
1714
1715 Node* adr = array_element_address(value, index, T_CHAR);
1716 if (adr->is_top()) {
1717 set_map(old_map);
1718 set_sp(old_sp);
1719 return false;
1720 }
1721 destruct_map_clone(old_map);
1722 if (is_store) {
1723 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1724 } else {
1725 ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD | C2_UNKNOWN_CONTROL_LOAD);
1726 set_result(ch);
1727 }
1728 return true;
1729 }
1730
1731 //--------------------------round_double_node--------------------------------
1732 // Round a double node if necessary.
1733 Node* LibraryCallKit::round_double_node(Node* n) {
1734 if (Matcher::strict_fp_requires_explicit_rounding) {
1735 #ifdef IA32
1736 if (UseSSE < 2) {
1737 n = _gvn.transform(new RoundDoubleNode(nullptr, n));
1738 }
1739 #else
1740 Unimplemented();
1741 #endif // IA32
1742 }
1743 return n;
1744 }
1745
1746 //------------------------------inline_math-----------------------------------
1747 // public static double Math.abs(double)
1748 // public static double Math.sqrt(double)
1749 // public static double Math.log(double)
1750 // public static double Math.log10(double)
1751 // public static double Math.round(double)
1752 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1753 Node* arg = round_double_node(argument(0));
1754 Node* n = nullptr;
1755 switch (id) {
1756 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1757 case vmIntrinsics::_dsqrt:
1758 case vmIntrinsics::_dsqrt_strict:
1759 n = new SqrtDNode(C, control(), arg); break;
1760 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1761 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1762 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1763 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1764 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1765 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1766 default: fatal_unexpected_iid(id); break;
1767 }
1768 set_result(_gvn.transform(n));
1769 return true;
1770 }
1771
1772 //------------------------------inline_math-----------------------------------
1773 // public static float Math.abs(float)
1774 // public static int Math.abs(int)
1775 // public static long Math.abs(long)
1776 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1777 Node* arg = argument(0);
1778 Node* n = nullptr;
1779 switch (id) {
1780 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1781 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1782 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1783 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1784 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1785 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1786 default: fatal_unexpected_iid(id); break;
1787 }
1788 set_result(_gvn.transform(n));
1789 return true;
1790 }
1791
1792 //------------------------------runtime_math-----------------------------
1793 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1794 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1795 "must be (DD)D or (D)D type");
1796
1797 // Inputs
1798 Node* a = round_double_node(argument(0));
1799 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : nullptr;
1800
1801 const TypePtr* no_memory_effects = nullptr;
1802 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1803 no_memory_effects,
1804 a, top(), b, b ? top() : nullptr);
1805 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1806 #ifdef ASSERT
1807 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1808 assert(value_top == top(), "second value must be top");
1809 #endif
1810
1811 set_result(value);
1812 return true;
1813 }
1814
1815 //------------------------------inline_math_pow-----------------------------
1816 bool LibraryCallKit::inline_math_pow() {
1817 Node* exp = round_double_node(argument(2));
1818 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1819 if (d != nullptr) {
1820 if (d->getd() == 2.0) {
1821 // Special case: pow(x, 2.0) => x * x
1822 Node* base = round_double_node(argument(0));
1823 set_result(_gvn.transform(new MulDNode(base, base)));
1824 return true;
1825 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1826 // Special case: pow(x, 0.5) => sqrt(x)
1827 Node* base = round_double_node(argument(0));
1828 Node* zero = _gvn.zerocon(T_DOUBLE);
1829
1830 RegionNode* region = new RegionNode(3);
1831 Node* phi = new PhiNode(region, Type::DOUBLE);
1832
1833 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1834 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1835 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1836 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1837 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1838
1839 Node* if_pow = generate_slow_guard(test, nullptr);
1840 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1841 phi->init_req(1, value_sqrt);
1842 region->init_req(1, control());
1843
1844 if (if_pow != nullptr) {
1845 set_control(if_pow);
1846 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1847 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1848 const TypePtr* no_memory_effects = nullptr;
1849 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1850 no_memory_effects, base, top(), exp, top());
1851 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1852 #ifdef ASSERT
1853 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1854 assert(value_top == top(), "second value must be top");
1855 #endif
1856 phi->init_req(2, value_pow);
1857 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1858 }
1859
1860 C->set_has_split_ifs(true); // Has chance for split-if optimization
1861 set_control(_gvn.transform(region));
1862 record_for_igvn(region);
1863 set_result(_gvn.transform(phi));
1864
1865 return true;
1866 }
1867 }
1868
1869 return StubRoutines::dpow() != nullptr ?
1870 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1871 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1872 }
1873
1874 //------------------------------inline_math_native-----------------------------
1875 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1876 switch (id) {
1877 case vmIntrinsics::_dsin:
1878 return StubRoutines::dsin() != nullptr ?
1879 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1880 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1881 case vmIntrinsics::_dcos:
1882 return StubRoutines::dcos() != nullptr ?
1883 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1884 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1885 case vmIntrinsics::_dtan:
1886 return StubRoutines::dtan() != nullptr ?
1887 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1888 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1889 case vmIntrinsics::_dtanh:
1890 return StubRoutines::dtanh() != nullptr ?
1891 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1892 case vmIntrinsics::_dexp:
1893 return StubRoutines::dexp() != nullptr ?
1894 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1895 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1896 case vmIntrinsics::_dlog:
1897 return StubRoutines::dlog() != nullptr ?
1898 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1899 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1900 case vmIntrinsics::_dlog10:
1901 return StubRoutines::dlog10() != nullptr ?
1902 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1903 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1904
1905 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1906 case vmIntrinsics::_ceil:
1907 case vmIntrinsics::_floor:
1908 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1909
1910 case vmIntrinsics::_dsqrt:
1911 case vmIntrinsics::_dsqrt_strict:
1912 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1913 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1914 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1915 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1916 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1917
1918 case vmIntrinsics::_dpow: return inline_math_pow();
1919 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1920 case vmIntrinsics::_fcopySign: return inline_math(id);
1921 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1922 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1923 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1924
1925 // These intrinsics are not yet correctly implemented
1926 case vmIntrinsics::_datan2:
1927 return false;
1928
1929 default:
1930 fatal_unexpected_iid(id);
1931 return false;
1932 }
1933 }
1934
1935 //----------------------------inline_notify-----------------------------------*
1936 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1937 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1938 address func;
1939 if (id == vmIntrinsics::_notify) {
1940 func = OptoRuntime::monitor_notify_Java();
1941 } else {
1942 func = OptoRuntime::monitor_notifyAll_Java();
1943 }
1944 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1945 make_slow_call_ex(call, env()->Throwable_klass(), false);
1946 return true;
1947 }
1948
1949
1950 //----------------------------inline_min_max-----------------------------------
1951 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1952 set_result(generate_min_max(id, argument(0), argument(1)));
1953 return true;
1954 }
1955
1956 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1957 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1958 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1959 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1960 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1961
1962 {
1963 PreserveJVMState pjvms(this);
1964 PreserveReexecuteState preexecs(this);
1965 jvms()->set_should_reexecute(true);
1966
1967 set_control(slow_path);
1968 set_i_o(i_o());
1969
1970 uncommon_trap(Deoptimization::Reason_intrinsic,
1971 Deoptimization::Action_none);
1972 }
1973
1974 set_control(fast_path);
1975 set_result(math);
1976 }
1977
1978 template <typename OverflowOp>
1979 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1980 typedef typename OverflowOp::MathOp MathOp;
1981
1982 MathOp* mathOp = new MathOp(arg1, arg2);
1983 Node* operation = _gvn.transform( mathOp );
1984 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1985 inline_math_mathExact(operation, ofcheck);
1986 return true;
1987 }
1988
1989 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1990 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1991 }
1992
1993 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1994 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1995 }
1996
1997 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1998 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1999 }
2000
2001 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2002 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2003 }
2004
2005 bool LibraryCallKit::inline_math_negateExactI() {
2006 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2007 }
2008
2009 bool LibraryCallKit::inline_math_negateExactL() {
2010 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2011 }
2012
2013 bool LibraryCallKit::inline_math_multiplyExactI() {
2014 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2015 }
2016
2017 bool LibraryCallKit::inline_math_multiplyExactL() {
2018 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2019 }
2020
2021 bool LibraryCallKit::inline_math_multiplyHigh() {
2022 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2023 return true;
2024 }
2025
2026 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2027 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2028 return true;
2029 }
2030
2031 Node*
2032 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
2033 Node* result_val = nullptr;
2034 switch (id) {
2035 case vmIntrinsics::_min:
2036 case vmIntrinsics::_min_strict:
2037 result_val = _gvn.transform(new MinINode(x0, y0));
2038 break;
2039 case vmIntrinsics::_max:
2040 case vmIntrinsics::_max_strict:
2041 result_val = _gvn.transform(new MaxINode(x0, y0));
2042 break;
2043 default:
2044 fatal_unexpected_iid(id);
2045 break;
2046 }
2047 return result_val;
2048 }
2049
2050 inline int
2051 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2052 const TypePtr* base_type = TypePtr::NULL_PTR;
2053 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2054 if (base_type == nullptr) {
2055 // Unknown type.
2056 return Type::AnyPtr;
2057 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2058 // Since this is a null+long form, we have to switch to a rawptr.
2059 base = _gvn.transform(new CastX2PNode(offset));
2060 offset = MakeConX(0);
2061 return Type::RawPtr;
2062 } else if (base_type->base() == Type::RawPtr) {
2063 return Type::RawPtr;
2064 } else if (base_type->isa_oopptr()) {
2065 // Base is never null => always a heap address.
2066 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2067 return Type::OopPtr;
2068 }
2069 // Offset is small => always a heap address.
2070 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2071 if (offset_type != nullptr &&
2072 base_type->offset() == 0 && // (should always be?)
2073 offset_type->_lo >= 0 &&
2074 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2075 return Type::OopPtr;
2076 } else if (type == T_OBJECT) {
2077 // off heap access to an oop doesn't make any sense. Has to be on
2078 // heap.
2079 return Type::OopPtr;
2080 }
2081 // Otherwise, it might either be oop+off or null+addr.
2082 return Type::AnyPtr;
2083 } else {
2084 // No information:
2085 return Type::AnyPtr;
2086 }
2087 }
2088
2089 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2090 Node* uncasted_base = base;
2091 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2092 if (kind == Type::RawPtr) {
2093 return basic_plus_adr(top(), uncasted_base, offset);
2094 } else if (kind == Type::AnyPtr) {
2095 assert(base == uncasted_base, "unexpected base change");
2096 if (can_cast) {
2097 if (!_gvn.type(base)->speculative_maybe_null() &&
2098 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2099 // According to profiling, this access is always on
2100 // heap. Casting the base to not null and thus avoiding membars
2101 // around the access should allow better optimizations
2102 Node* null_ctl = top();
2103 base = null_check_oop(base, &null_ctl, true, true, true);
2104 assert(null_ctl->is_top(), "no null control here");
2105 return basic_plus_adr(base, offset);
2106 } else if (_gvn.type(base)->speculative_always_null() &&
2107 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2108 // According to profiling, this access is always off
2109 // heap.
2110 base = null_assert(base);
2111 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2112 offset = MakeConX(0);
2113 return basic_plus_adr(top(), raw_base, offset);
2114 }
2115 }
2116 // We don't know if it's an on heap or off heap access. Fall back
2117 // to raw memory access.
2118 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2119 return basic_plus_adr(top(), raw, offset);
2120 } else {
2121 assert(base == uncasted_base, "unexpected base change");
2122 // We know it's an on heap access so base can't be null
2123 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2124 base = must_be_not_null(base, true);
2125 }
2126 return basic_plus_adr(base, offset);
2127 }
2128 }
2129
2130 //--------------------------inline_number_methods-----------------------------
2131 // inline int Integer.numberOfLeadingZeros(int)
2132 // inline int Long.numberOfLeadingZeros(long)
2133 //
2134 // inline int Integer.numberOfTrailingZeros(int)
2135 // inline int Long.numberOfTrailingZeros(long)
2136 //
2137 // inline int Integer.bitCount(int)
2138 // inline int Long.bitCount(long)
2139 //
2140 // inline char Character.reverseBytes(char)
2141 // inline short Short.reverseBytes(short)
2142 // inline int Integer.reverseBytes(int)
2143 // inline long Long.reverseBytes(long)
2144 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2145 Node* arg = argument(0);
2146 Node* n = nullptr;
2147 switch (id) {
2148 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2149 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2150 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2151 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2152 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2153 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2154 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(nullptr, arg); break;
2155 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( nullptr, arg); break;
2156 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( nullptr, arg); break;
2157 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( nullptr, arg); break;
2158 case vmIntrinsics::_reverse_i: n = new ReverseINode(nullptr, arg); break;
2159 case vmIntrinsics::_reverse_l: n = new ReverseLNode(nullptr, arg); break;
2160 default: fatal_unexpected_iid(id); break;
2161 }
2162 set_result(_gvn.transform(n));
2163 return true;
2164 }
2165
2166 //--------------------------inline_bitshuffle_methods-----------------------------
2167 // inline int Integer.compress(int, int)
2168 // inline int Integer.expand(int, int)
2169 // inline long Long.compress(long, long)
2170 // inline long Long.expand(long, long)
2171 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2172 Node* n = nullptr;
2173 switch (id) {
2174 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2175 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2176 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2177 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2178 default: fatal_unexpected_iid(id); break;
2179 }
2180 set_result(_gvn.transform(n));
2181 return true;
2182 }
2183
2184 //--------------------------inline_number_methods-----------------------------
2185 // inline int Integer.compareUnsigned(int, int)
2186 // inline int Long.compareUnsigned(long, long)
2187 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2188 Node* arg1 = argument(0);
2189 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2190 Node* n = nullptr;
2191 switch (id) {
2192 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2193 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2194 default: fatal_unexpected_iid(id); break;
2195 }
2196 set_result(_gvn.transform(n));
2197 return true;
2198 }
2199
2200 //--------------------------inline_unsigned_divmod_methods-----------------------------
2201 // inline int Integer.divideUnsigned(int, int)
2202 // inline int Integer.remainderUnsigned(int, int)
2203 // inline long Long.divideUnsigned(long, long)
2204 // inline long Long.remainderUnsigned(long, long)
2205 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2206 Node* n = nullptr;
2207 switch (id) {
2208 case vmIntrinsics::_divideUnsigned_i: {
2209 zero_check_int(argument(1));
2210 // Compile-time detect of null-exception
2211 if (stopped()) {
2212 return true; // keep the graph constructed so far
2213 }
2214 n = new UDivINode(control(), argument(0), argument(1));
2215 break;
2216 }
2217 case vmIntrinsics::_divideUnsigned_l: {
2218 zero_check_long(argument(2));
2219 // Compile-time detect of null-exception
2220 if (stopped()) {
2221 return true; // keep the graph constructed so far
2222 }
2223 n = new UDivLNode(control(), argument(0), argument(2));
2224 break;
2225 }
2226 case vmIntrinsics::_remainderUnsigned_i: {
2227 zero_check_int(argument(1));
2228 // Compile-time detect of null-exception
2229 if (stopped()) {
2230 return true; // keep the graph constructed so far
2231 }
2232 n = new UModINode(control(), argument(0), argument(1));
2233 break;
2234 }
2235 case vmIntrinsics::_remainderUnsigned_l: {
2236 zero_check_long(argument(2));
2237 // Compile-time detect of null-exception
2238 if (stopped()) {
2239 return true; // keep the graph constructed so far
2240 }
2241 n = new UModLNode(control(), argument(0), argument(2));
2242 break;
2243 }
2244 default: fatal_unexpected_iid(id); break;
2245 }
2246 set_result(_gvn.transform(n));
2247 return true;
2248 }
2249
2250 //----------------------------inline_unsafe_access----------------------------
2251
2252 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2253 // Attempt to infer a sharper value type from the offset and base type.
2254 ciKlass* sharpened_klass = nullptr;
2255
2256 // See if it is an instance field, with an object type.
2257 if (alias_type->field() != nullptr) {
2258 if (alias_type->field()->type()->is_klass()) {
2259 sharpened_klass = alias_type->field()->type()->as_klass();
2260 }
2261 }
2262
2263 const TypeOopPtr* result = nullptr;
2264 // See if it is a narrow oop array.
2265 if (adr_type->isa_aryptr()) {
2266 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2267 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2268 if (elem_type != nullptr && elem_type->is_loaded()) {
2269 // Sharpen the value type.
2270 result = elem_type;
2271 }
2272 }
2273 }
2274
2275 // The sharpened class might be unloaded if there is no class loader
2276 // contraint in place.
2277 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2278 // Sharpen the value type.
2279 result = TypeOopPtr::make_from_klass(sharpened_klass);
2280 }
2281 if (result != nullptr) {
2282 #ifndef PRODUCT
2283 if (C->print_intrinsics() || C->print_inlining()) {
2284 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2285 tty->print(" sharpened value: "); result->dump(); tty->cr();
2286 }
2287 #endif
2288 }
2289 return result;
2290 }
2291
2292 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2293 switch (kind) {
2294 case Relaxed:
2295 return MO_UNORDERED;
2296 case Opaque:
2297 return MO_RELAXED;
2298 case Acquire:
2299 return MO_ACQUIRE;
2300 case Release:
2301 return MO_RELEASE;
2302 case Volatile:
2303 return MO_SEQ_CST;
2304 default:
2305 ShouldNotReachHere();
2306 return 0;
2307 }
2308 }
2309
2310 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2311 if (callee()->is_static()) return false; // caller must have the capability!
2312 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2313 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2314 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2315 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2316
2317 if (is_reference_type(type)) {
2318 decorators |= ON_UNKNOWN_OOP_REF;
2319 }
2320
2321 if (unaligned) {
2322 decorators |= C2_UNALIGNED;
2323 }
2324
2325 #ifndef PRODUCT
2326 {
2327 ResourceMark rm;
2328 // Check the signatures.
2329 ciSignature* sig = callee()->signature();
2330 #ifdef ASSERT
2331 if (!is_store) {
2332 // Object getReference(Object base, int/long offset), etc.
2333 BasicType rtype = sig->return_type()->basic_type();
2334 assert(rtype == type, "getter must return the expected value");
2335 assert(sig->count() == 2, "oop getter has 2 arguments");
2336 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2337 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2338 } else {
2339 // void putReference(Object base, int/long offset, Object x), etc.
2340 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2341 assert(sig->count() == 3, "oop putter has 3 arguments");
2342 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2343 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2344 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2345 assert(vtype == type, "putter must accept the expected value");
2346 }
2347 #endif // ASSERT
2348 }
2349 #endif //PRODUCT
2350
2351 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2352
2353 Node* receiver = argument(0); // type: oop
2354
2355 // Build address expression.
2356 Node* heap_base_oop = top();
2357
2358 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2359 Node* base = argument(1); // type: oop
2360 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2361 Node* offset = argument(2); // type: long
2362 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2363 // to be plain byte offsets, which are also the same as those accepted
2364 // by oopDesc::field_addr.
2365 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2366 "fieldOffset must be byte-scaled");
2367 // 32-bit machines ignore the high half!
2368 offset = ConvL2X(offset);
2369
2370 // Save state and restore on bailout
2371 uint old_sp = sp();
2372 SafePointNode* old_map = clone_map();
2373
2374 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2375 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2376
2377 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2378 if (type != T_OBJECT) {
2379 decorators |= IN_NATIVE; // off-heap primitive access
2380 } else {
2381 set_map(old_map);
2382 set_sp(old_sp);
2383 return false; // off-heap oop accesses are not supported
2384 }
2385 } else {
2386 heap_base_oop = base; // on-heap or mixed access
2387 }
2388
2389 // Can base be null? Otherwise, always on-heap access.
2390 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2391
2392 if (!can_access_non_heap) {
2393 decorators |= IN_HEAP;
2394 }
2395
2396 Node* val = is_store ? argument(4) : nullptr;
2397
2398 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2399 if (adr_type == TypePtr::NULL_PTR) {
2400 set_map(old_map);
2401 set_sp(old_sp);
2402 return false; // off-heap access with zero address
2403 }
2404
2405 // Try to categorize the address.
2406 Compile::AliasType* alias_type = C->alias_type(adr_type);
2407 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2408
2409 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2410 alias_type->adr_type() == TypeAryPtr::RANGE) {
2411 set_map(old_map);
2412 set_sp(old_sp);
2413 return false; // not supported
2414 }
2415
2416 bool mismatched = false;
2417 BasicType bt = alias_type->basic_type();
2418 if (bt != T_ILLEGAL) {
2419 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2420 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2421 // Alias type doesn't differentiate between byte[] and boolean[]).
2422 // Use address type to get the element type.
2423 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2424 }
2425 if (is_reference_type(bt, true)) {
2426 // accessing an array field with getReference is not a mismatch
2427 bt = T_OBJECT;
2428 }
2429 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2430 // Don't intrinsify mismatched object accesses
2431 set_map(old_map);
2432 set_sp(old_sp);
2433 return false;
2434 }
2435 mismatched = (bt != type);
2436 } else if (alias_type->adr_type()->isa_oopptr()) {
2437 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2438 }
2439
2440 destruct_map_clone(old_map);
2441 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2442
2443 if (mismatched) {
2444 decorators |= C2_MISMATCHED;
2445 }
2446
2447 // First guess at the value type.
2448 const Type *value_type = Type::get_const_basic_type(type);
2449
2450 // Figure out the memory ordering.
2451 decorators |= mo_decorator_for_access_kind(kind);
2452
2453 if (!is_store && type == T_OBJECT) {
2454 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2455 if (tjp != nullptr) {
2456 value_type = tjp;
2457 }
2458 }
2459
2460 receiver = null_check(receiver);
2461 if (stopped()) {
2462 return true;
2463 }
2464 // Heap pointers get a null-check from the interpreter,
2465 // as a courtesy. However, this is not guaranteed by Unsafe,
2466 // and it is not possible to fully distinguish unintended nulls
2467 // from intended ones in this API.
2468
2469 if (!is_store) {
2470 Node* p = nullptr;
2471 // Try to constant fold a load from a constant field
2472 ciField* field = alias_type->field();
2473 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) {
2474 // final or stable field
2475 p = make_constant_from_field(field, heap_base_oop);
2476 }
2477
2478 if (p == nullptr) { // Could not constant fold the load
2479 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2480 // Normalize the value returned by getBoolean in the following cases
2481 if (type == T_BOOLEAN &&
2482 (mismatched ||
2483 heap_base_oop == top() || // - heap_base_oop is null or
2484 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2485 // and the unsafe access is made to large offset
2486 // (i.e., larger than the maximum offset necessary for any
2487 // field access)
2488 ) {
2489 IdealKit ideal = IdealKit(this);
2490 #define __ ideal.
2491 IdealVariable normalized_result(ideal);
2492 __ declarations_done();
2493 __ set(normalized_result, p);
2494 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2495 __ set(normalized_result, ideal.ConI(1));
2496 ideal.end_if();
2497 final_sync(ideal);
2498 p = __ value(normalized_result);
2499 #undef __
2500 }
2501 }
2502 if (type == T_ADDRESS) {
2503 p = gvn().transform(new CastP2XNode(nullptr, p));
2504 p = ConvX2UL(p);
2505 }
2506 // The load node has the control of the preceding MemBarCPUOrder. All
2507 // following nodes will have the control of the MemBarCPUOrder inserted at
2508 // the end of this method. So, pushing the load onto the stack at a later
2509 // point is fine.
2510 set_result(p);
2511 } else {
2512 if (bt == T_ADDRESS) {
2513 // Repackage the long as a pointer.
2514 val = ConvL2X(val);
2515 val = gvn().transform(new CastX2PNode(val));
2516 }
2517 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2518 }
2519
2520 return true;
2521 }
2522
2523 //----------------------------inline_unsafe_load_store----------------------------
2524 // This method serves a couple of different customers (depending on LoadStoreKind):
2525 //
2526 // LS_cmp_swap:
2527 //
2528 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2529 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2530 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2531 //
2532 // LS_cmp_swap_weak:
2533 //
2534 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2535 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2536 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2537 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2538 //
2539 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2540 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2541 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2542 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2543 //
2544 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2545 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2546 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2547 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2548 //
2549 // LS_cmp_exchange:
2550 //
2551 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2552 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2553 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2554 //
2555 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2556 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2557 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2558 //
2559 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2560 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2561 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2562 //
2563 // LS_get_add:
2564 //
2565 // int getAndAddInt( Object o, long offset, int delta)
2566 // long getAndAddLong(Object o, long offset, long delta)
2567 //
2568 // LS_get_set:
2569 //
2570 // int getAndSet(Object o, long offset, int newValue)
2571 // long getAndSet(Object o, long offset, long newValue)
2572 // Object getAndSet(Object o, long offset, Object newValue)
2573 //
2574 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2575 // This basic scheme here is the same as inline_unsafe_access, but
2576 // differs in enough details that combining them would make the code
2577 // overly confusing. (This is a true fact! I originally combined
2578 // them, but even I was confused by it!) As much code/comments as
2579 // possible are retained from inline_unsafe_access though to make
2580 // the correspondences clearer. - dl
2581
2582 if (callee()->is_static()) return false; // caller must have the capability!
2583
2584 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2585 decorators |= mo_decorator_for_access_kind(access_kind);
2586
2587 #ifndef PRODUCT
2588 BasicType rtype;
2589 {
2590 ResourceMark rm;
2591 // Check the signatures.
2592 ciSignature* sig = callee()->signature();
2593 rtype = sig->return_type()->basic_type();
2594 switch(kind) {
2595 case LS_get_add:
2596 case LS_get_set: {
2597 // Check the signatures.
2598 #ifdef ASSERT
2599 assert(rtype == type, "get and set must return the expected type");
2600 assert(sig->count() == 3, "get and set has 3 arguments");
2601 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2602 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2603 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2604 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2605 #endif // ASSERT
2606 break;
2607 }
2608 case LS_cmp_swap:
2609 case LS_cmp_swap_weak: {
2610 // Check the signatures.
2611 #ifdef ASSERT
2612 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2613 assert(sig->count() == 4, "CAS has 4 arguments");
2614 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2615 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2616 #endif // ASSERT
2617 break;
2618 }
2619 case LS_cmp_exchange: {
2620 // Check the signatures.
2621 #ifdef ASSERT
2622 assert(rtype == type, "CAS must return the expected type");
2623 assert(sig->count() == 4, "CAS has 4 arguments");
2624 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2625 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2626 #endif // ASSERT
2627 break;
2628 }
2629 default:
2630 ShouldNotReachHere();
2631 }
2632 }
2633 #endif //PRODUCT
2634
2635 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2636
2637 // Get arguments:
2638 Node* receiver = nullptr;
2639 Node* base = nullptr;
2640 Node* offset = nullptr;
2641 Node* oldval = nullptr;
2642 Node* newval = nullptr;
2643 switch(kind) {
2644 case LS_cmp_swap:
2645 case LS_cmp_swap_weak:
2646 case LS_cmp_exchange: {
2647 const bool two_slot_type = type2size[type] == 2;
2648 receiver = argument(0); // type: oop
2649 base = argument(1); // type: oop
2650 offset = argument(2); // type: long
2651 oldval = argument(4); // type: oop, int, or long
2652 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2653 break;
2654 }
2655 case LS_get_add:
2656 case LS_get_set: {
2657 receiver = argument(0); // type: oop
2658 base = argument(1); // type: oop
2659 offset = argument(2); // type: long
2660 oldval = nullptr;
2661 newval = argument(4); // type: oop, int, or long
2662 break;
2663 }
2664 default:
2665 ShouldNotReachHere();
2666 }
2667
2668 // Build field offset expression.
2669 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2670 // to be plain byte offsets, which are also the same as those accepted
2671 // by oopDesc::field_addr.
2672 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2673 // 32-bit machines ignore the high half of long offsets
2674 offset = ConvL2X(offset);
2675 // Save state and restore on bailout
2676 uint old_sp = sp();
2677 SafePointNode* old_map = clone_map();
2678 Node* adr = make_unsafe_address(base, offset,type, false);
2679 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2680
2681 Compile::AliasType* alias_type = C->alias_type(adr_type);
2682 BasicType bt = alias_type->basic_type();
2683 if (bt != T_ILLEGAL &&
2684 (is_reference_type(bt) != (type == T_OBJECT))) {
2685 // Don't intrinsify mismatched object accesses.
2686 set_map(old_map);
2687 set_sp(old_sp);
2688 return false;
2689 }
2690
2691 destruct_map_clone(old_map);
2692
2693 // For CAS, unlike inline_unsafe_access, there seems no point in
2694 // trying to refine types. Just use the coarse types here.
2695 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2696 const Type *value_type = Type::get_const_basic_type(type);
2697
2698 switch (kind) {
2699 case LS_get_set:
2700 case LS_cmp_exchange: {
2701 if (type == T_OBJECT) {
2702 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2703 if (tjp != nullptr) {
2704 value_type = tjp;
2705 }
2706 }
2707 break;
2708 }
2709 case LS_cmp_swap:
2710 case LS_cmp_swap_weak:
2711 case LS_get_add:
2712 break;
2713 default:
2714 ShouldNotReachHere();
2715 }
2716
2717 // Null check receiver.
2718 receiver = null_check(receiver);
2719 if (stopped()) {
2720 return true;
2721 }
2722
2723 int alias_idx = C->get_alias_index(adr_type);
2724
2725 if (is_reference_type(type)) {
2726 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2727
2728 // Transformation of a value which could be null pointer (CastPP #null)
2729 // could be delayed during Parse (for example, in adjust_map_after_if()).
2730 // Execute transformation here to avoid barrier generation in such case.
2731 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2732 newval = _gvn.makecon(TypePtr::NULL_PTR);
2733
2734 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2735 // Refine the value to a null constant, when it is known to be null
2736 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2737 }
2738 }
2739
2740 Node* result = nullptr;
2741 switch (kind) {
2742 case LS_cmp_exchange: {
2743 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2744 oldval, newval, value_type, type, decorators);
2745 break;
2746 }
2747 case LS_cmp_swap_weak:
2748 decorators |= C2_WEAK_CMPXCHG;
2749 case LS_cmp_swap: {
2750 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2751 oldval, newval, value_type, type, decorators);
2752 break;
2753 }
2754 case LS_get_set: {
2755 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2756 newval, value_type, type, decorators);
2757 break;
2758 }
2759 case LS_get_add: {
2760 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2761 newval, value_type, type, decorators);
2762 break;
2763 }
2764 default:
2765 ShouldNotReachHere();
2766 }
2767
2768 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2769 set_result(result);
2770 return true;
2771 }
2772
2773 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2774 // Regardless of form, don't allow previous ld/st to move down,
2775 // then issue acquire, release, or volatile mem_bar.
2776 insert_mem_bar(Op_MemBarCPUOrder);
2777 switch(id) {
2778 case vmIntrinsics::_loadFence:
2779 insert_mem_bar(Op_LoadFence);
2780 return true;
2781 case vmIntrinsics::_storeFence:
2782 insert_mem_bar(Op_StoreFence);
2783 return true;
2784 case vmIntrinsics::_storeStoreFence:
2785 insert_mem_bar(Op_StoreStoreFence);
2786 return true;
2787 case vmIntrinsics::_fullFence:
2788 insert_mem_bar(Op_MemBarVolatile);
2789 return true;
2790 default:
2791 fatal_unexpected_iid(id);
2792 return false;
2793 }
2794 }
2795
2796 bool LibraryCallKit::inline_onspinwait() {
2797 insert_mem_bar(Op_OnSpinWait);
2798 return true;
2799 }
2800
2801 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2802 if (!kls->is_Con()) {
2803 return true;
2804 }
2805 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2806 if (klsptr == nullptr) {
2807 return true;
2808 }
2809 ciInstanceKlass* ik = klsptr->instance_klass();
2810 // don't need a guard for a klass that is already initialized
2811 return !ik->is_initialized();
2812 }
2813
2814 //----------------------------inline_unsafe_writeback0-------------------------
2815 // public native void Unsafe.writeback0(long address)
2816 bool LibraryCallKit::inline_unsafe_writeback0() {
2817 if (!Matcher::has_match_rule(Op_CacheWB)) {
2818 return false;
2819 }
2820 #ifndef PRODUCT
2821 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2822 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2823 ciSignature* sig = callee()->signature();
2824 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2825 #endif
2826 null_check_receiver(); // null-check, then ignore
2827 Node *addr = argument(1);
2828 addr = new CastX2PNode(addr);
2829 addr = _gvn.transform(addr);
2830 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2831 flush = _gvn.transform(flush);
2832 set_memory(flush, TypeRawPtr::BOTTOM);
2833 return true;
2834 }
2835
2836 //----------------------------inline_unsafe_writeback0-------------------------
2837 // public native void Unsafe.writeback0(long address)
2838 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2839 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2840 return false;
2841 }
2842 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2843 return false;
2844 }
2845 #ifndef PRODUCT
2846 assert(Matcher::has_match_rule(Op_CacheWB),
2847 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2848 : "found match rule for CacheWBPostSync but not CacheWB"));
2849
2850 #endif
2851 null_check_receiver(); // null-check, then ignore
2852 Node *sync;
2853 if (is_pre) {
2854 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2855 } else {
2856 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2857 }
2858 sync = _gvn.transform(sync);
2859 set_memory(sync, TypeRawPtr::BOTTOM);
2860 return true;
2861 }
2862
2863 //----------------------------inline_unsafe_allocate---------------------------
2864 // public native Object Unsafe.allocateInstance(Class<?> cls);
2865 bool LibraryCallKit::inline_unsafe_allocate() {
2866
2867 #if INCLUDE_JVMTI
2868 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
2869 return false;
2870 }
2871 #endif //INCLUDE_JVMTI
2872
2873 if (callee()->is_static()) return false; // caller must have the capability!
2874
2875 null_check_receiver(); // null-check, then ignore
2876 Node* cls = null_check(argument(1));
2877 if (stopped()) return true;
2878
2879 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
2880 kls = null_check(kls);
2881 if (stopped()) return true; // argument was like int.class
2882
2883 #if INCLUDE_JVMTI
2884 // Don't try to access new allocated obj in the intrinsic.
2885 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
2886 // Deoptimize and allocate in interpreter instead.
2887 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
2888 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
2889 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
2890 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
2891 {
2892 BuildCutout unless(this, tst, PROB_MAX);
2893 uncommon_trap(Deoptimization::Reason_intrinsic,
2894 Deoptimization::Action_make_not_entrant);
2895 }
2896 if (stopped()) {
2897 return true;
2898 }
2899 #endif //INCLUDE_JVMTI
2900
2901 Node* test = nullptr;
2902 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2903 // Note: The argument might still be an illegal value like
2904 // Serializable.class or Object[].class. The runtime will handle it.
2905 // But we must make an explicit check for initialization.
2906 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2907 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2908 // can generate code to load it as unsigned byte.
2909 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
2910 Node* bits = intcon(InstanceKlass::fully_initialized);
2911 test = _gvn.transform(new SubINode(inst, bits));
2912 // The 'test' is non-zero if we need to take a slow path.
2913 }
2914
2915 Node* obj = new_instance(kls, test);
2916 set_result(obj);
2917 return true;
2918 }
2919
2920 //------------------------inline_native_time_funcs--------------
2921 // inline code for System.currentTimeMillis() and System.nanoTime()
2922 // these have the same type and signature
2923 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2924 const TypeFunc* tf = OptoRuntime::void_long_Type();
2925 const TypePtr* no_memory_effects = nullptr;
2926 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2927 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2928 #ifdef ASSERT
2929 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2930 assert(value_top == top(), "second value must be top");
2931 #endif
2932 set_result(value);
2933 return true;
2934 }
2935
2936
2937 #if INCLUDE_JVMTI
2938
2939 // When notifications are disabled then just update the VTMS transition bit and return.
2940 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
2941 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
2942 if (!DoJVMTIVirtualThreadTransitions) {
2943 return true;
2944 }
2945 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
2946 IdealKit ideal(this);
2947
2948 Node* ONE = ideal.ConI(1);
2949 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
2950 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
2951 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
2952
2953 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
2954 sync_kit(ideal);
2955 // if notifyJvmti enabled then make a call to the given SharedRuntime function
2956 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
2957 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
2958 ideal.sync_kit(this);
2959 } ideal.else_(); {
2960 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
2961 Node* thread = ideal.thread();
2962 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
2963 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
2964
2965 sync_kit(ideal);
2966 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2967 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2968
2969 ideal.sync_kit(this);
2970 } ideal.end_if();
2971 final_sync(ideal);
2972
2973 return true;
2974 }
2975
2976 // Always update the is_disable_suspend bit.
2977 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
2978 if (!DoJVMTIVirtualThreadTransitions) {
2979 return true;
2980 }
2981 IdealKit ideal(this);
2982
2983 {
2984 // unconditionally update the is_disable_suspend bit in current JavaThread
2985 Node* thread = ideal.thread();
2986 Node* arg = _gvn.transform(argument(0)); // argument for notification
2987 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
2988 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
2989
2990 sync_kit(ideal);
2991 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2992 ideal.sync_kit(this);
2993 }
2994 final_sync(ideal);
2995
2996 return true;
2997 }
2998
2999 #endif // INCLUDE_JVMTI
3000
3001 #ifdef JFR_HAVE_INTRINSICS
3002
3003 /**
3004 * if oop->klass != null
3005 * // normal class
3006 * epoch = _epoch_state ? 2 : 1
3007 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3008 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3009 * }
3010 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3011 * else
3012 * // primitive class
3013 * if oop->array_klass != null
3014 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3015 * else
3016 * id = LAST_TYPE_ID + 1 // void class path
3017 * if (!signaled)
3018 * signaled = true
3019 */
3020 bool LibraryCallKit::inline_native_classID() {
3021 Node* cls = argument(0);
3022
3023 IdealKit ideal(this);
3024 #define __ ideal.
3025 IdealVariable result(ideal); __ declarations_done();
3026 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3027 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3028 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3029
3030
3031 __ if_then(kls, BoolTest::ne, null()); {
3032 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3033 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3034
3035 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3036 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3037 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3038 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3039 mask = _gvn.transform(new OrLNode(mask, epoch));
3040 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3041
3042 float unlikely = PROB_UNLIKELY(0.999);
3043 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3044 sync_kit(ideal);
3045 make_runtime_call(RC_LEAF,
3046 OptoRuntime::class_id_load_barrier_Type(),
3047 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3048 "class id load barrier",
3049 TypePtr::BOTTOM,
3050 kls);
3051 ideal.sync_kit(this);
3052 } __ end_if();
3053
3054 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3055 } __ else_(); {
3056 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3057 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3058 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3059 __ if_then(array_kls, BoolTest::ne, null()); {
3060 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3061 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3062 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3063 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3064 } __ else_(); {
3065 // void class case
3066 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3067 } __ end_if();
3068
3069 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3070 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3071 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3072 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3073 } __ end_if();
3074 } __ end_if();
3075
3076 final_sync(ideal);
3077 set_result(ideal.value(result));
3078 #undef __
3079 return true;
3080 }
3081
3082 //------------------------inline_native_jvm_commit------------------
3083 bool LibraryCallKit::inline_native_jvm_commit() {
3084 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3085
3086 // Save input memory and i_o state.
3087 Node* input_memory_state = reset_memory();
3088 set_all_memory(input_memory_state);
3089 Node* input_io_state = i_o();
3090
3091 // TLS.
3092 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3093 // Jfr java buffer.
3094 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3095 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3096 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3097
3098 // Load the current value of the notified field in the JfrThreadLocal.
3099 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3100 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3101
3102 // Test for notification.
3103 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3104 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3105 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3106
3107 // True branch, is notified.
3108 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3109 set_control(is_notified);
3110
3111 // Reset notified state.
3112 Node* notified_reset_memory = store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3113
3114 // Iff notified, the return address of the commit method is the current position of the backing java buffer. This is used to reset the event writer.
3115 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3116 // Convert the machine-word to a long.
3117 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3118
3119 // False branch, not notified.
3120 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3121 set_control(not_notified);
3122 set_all_memory(input_memory_state);
3123
3124 // Arg is the next position as a long.
3125 Node* arg = argument(0);
3126 // Convert long to machine-word.
3127 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3128
3129 // Store the next_position to the underlying jfr java buffer.
3130 Node* commit_memory;
3131 #ifdef _LP64
3132 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_LONG, MemNode::release);
3133 #else
3134 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_INT, MemNode::release);
3135 #endif
3136
3137 // Now load the flags from off the java buffer and decide if the buffer is a lease. If so, it needs to be returned post-commit.
3138 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3139 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3140 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3141
3142 // And flags with lease constant.
3143 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3144
3145 // Branch on lease to conditionalize returning the leased java buffer.
3146 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3147 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3148 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3149
3150 // False branch, not a lease.
3151 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3152
3153 // True branch, is lease.
3154 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3155 set_control(is_lease);
3156
3157 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3158 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3159 OptoRuntime::void_void_Type(),
3160 SharedRuntime::jfr_return_lease(),
3161 "return_lease", TypePtr::BOTTOM);
3162 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3163
3164 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3165 record_for_igvn(lease_compare_rgn);
3166 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3167 record_for_igvn(lease_compare_mem);
3168 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3169 record_for_igvn(lease_compare_io);
3170 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3171 record_for_igvn(lease_result_value);
3172
3173 // Update control and phi nodes.
3174 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3175 lease_compare_rgn->init_req(_false_path, not_lease);
3176
3177 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3178 lease_compare_mem->init_req(_false_path, commit_memory);
3179
3180 lease_compare_io->init_req(_true_path, i_o());
3181 lease_compare_io->init_req(_false_path, input_io_state);
3182
3183 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3184 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3185
3186 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3187 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3188 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3189 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3190
3191 // Update control and phi nodes.
3192 result_rgn->init_req(_true_path, is_notified);
3193 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3194
3195 result_mem->init_req(_true_path, notified_reset_memory);
3196 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3197
3198 result_io->init_req(_true_path, input_io_state);
3199 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3200
3201 result_value->init_req(_true_path, current_pos);
3202 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3203
3204 // Set output state.
3205 set_control(_gvn.transform(result_rgn));
3206 set_all_memory(_gvn.transform(result_mem));
3207 set_i_o(_gvn.transform(result_io));
3208 set_result(result_rgn, result_value);
3209 return true;
3210 }
3211
3212 /*
3213 * The intrinsic is a model of this pseudo-code:
3214 *
3215 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3216 * jobject h_event_writer = tl->java_event_writer();
3217 * if (h_event_writer == nullptr) {
3218 * return nullptr;
3219 * }
3220 * oop threadObj = Thread::threadObj();
3221 * oop vthread = java_lang_Thread::vthread(threadObj);
3222 * traceid tid;
3223 * bool pinVirtualThread;
3224 * bool excluded;
3225 * if (vthread != threadObj) { // i.e. current thread is virtual
3226 * tid = java_lang_Thread::tid(vthread);
3227 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3228 * pinVirtualThread = VMContinuations;
3229 * excluded = vthread_epoch_raw & excluded_mask;
3230 * if (!excluded) {
3231 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3232 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3233 * if (vthread_epoch != current_epoch) {
3234 * write_checkpoint();
3235 * }
3236 * }
3237 * } else {
3238 * tid = java_lang_Thread::tid(threadObj);
3239 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3240 * pinVirtualThread = false;
3241 * excluded = thread_epoch_raw & excluded_mask;
3242 * }
3243 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3244 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3245 * if (tid_in_event_writer != tid) {
3246 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3247 * setField(event_writer, "excluded", excluded);
3248 * setField(event_writer, "threadID", tid);
3249 * }
3250 * return event_writer
3251 */
3252 bool LibraryCallKit::inline_native_getEventWriter() {
3253 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3254
3255 // Save input memory and i_o state.
3256 Node* input_memory_state = reset_memory();
3257 set_all_memory(input_memory_state);
3258 Node* input_io_state = i_o();
3259
3260 // The most significant bit of the u2 is used to denote thread exclusion
3261 Node* excluded_shift = _gvn.intcon(15);
3262 Node* excluded_mask = _gvn.intcon(1 << 15);
3263 // The epoch generation is the range [1-32767]
3264 Node* epoch_mask = _gvn.intcon(32767);
3265
3266 // TLS
3267 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3268
3269 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3270 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3271
3272 // Load the eventwriter jobject handle.
3273 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3274
3275 // Null check the jobject handle.
3276 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3277 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3278 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3279
3280 // False path, jobj is null.
3281 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3282
3283 // True path, jobj is not null.
3284 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3285
3286 set_control(jobj_is_not_null);
3287
3288 // Load the threadObj for the CarrierThread.
3289 Node* threadObj = generate_current_thread(tls_ptr);
3290
3291 // Load the vthread.
3292 Node* vthread = generate_virtual_thread(tls_ptr);
3293
3294 // If vthread != threadObj, this is a virtual thread.
3295 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3296 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3297 IfNode* iff_vthread_not_equal_threadObj =
3298 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3299
3300 // False branch, fallback to threadObj.
3301 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3302 set_control(vthread_equal_threadObj);
3303
3304 // Load the tid field from the vthread object.
3305 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3306
3307 // Load the raw epoch value from the threadObj.
3308 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3309 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3310 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3311 TypeInt::CHAR, T_CHAR,
3312 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3313
3314 // Mask off the excluded information from the epoch.
3315 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3316
3317 // True branch, this is a virtual thread.
3318 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3319 set_control(vthread_not_equal_threadObj);
3320
3321 // Load the tid field from the vthread object.
3322 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3323
3324 // Continuation support determines if a virtual thread should be pinned.
3325 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3326 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3327
3328 // Load the raw epoch value from the vthread.
3329 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3330 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3331 TypeInt::CHAR, T_CHAR,
3332 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3333
3334 // Mask off the excluded information from the epoch.
3335 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3336
3337 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3338 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3339 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3340 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3341
3342 // False branch, vthread is excluded, no need to write epoch info.
3343 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3344
3345 // True branch, vthread is included, update epoch info.
3346 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3347 set_control(included);
3348
3349 // Get epoch value.
3350 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3351
3352 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3353 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3354 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3355
3356 // Compare the epoch in the vthread to the current epoch generation.
3357 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3358 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3359 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3360
3361 // False path, epoch is equal, checkpoint information is valid.
3362 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3363
3364 // True path, epoch is not equal, write a checkpoint for the vthread.
3365 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3366
3367 set_control(epoch_is_not_equal);
3368
3369 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3370 // The call also updates the native thread local thread id and the vthread with the current epoch.
3371 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3372 OptoRuntime::jfr_write_checkpoint_Type(),
3373 SharedRuntime::jfr_write_checkpoint(),
3374 "write_checkpoint", TypePtr::BOTTOM);
3375 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3376
3377 // vthread epoch != current epoch
3378 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3379 record_for_igvn(epoch_compare_rgn);
3380 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3381 record_for_igvn(epoch_compare_mem);
3382 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3383 record_for_igvn(epoch_compare_io);
3384
3385 // Update control and phi nodes.
3386 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3387 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3388 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3389 epoch_compare_mem->init_req(_false_path, input_memory_state);
3390 epoch_compare_io->init_req(_true_path, i_o());
3391 epoch_compare_io->init_req(_false_path, input_io_state);
3392
3393 // excluded != true
3394 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3395 record_for_igvn(exclude_compare_rgn);
3396 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3397 record_for_igvn(exclude_compare_mem);
3398 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3399 record_for_igvn(exclude_compare_io);
3400
3401 // Update control and phi nodes.
3402 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3403 exclude_compare_rgn->init_req(_false_path, excluded);
3404 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3405 exclude_compare_mem->init_req(_false_path, input_memory_state);
3406 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3407 exclude_compare_io->init_req(_false_path, input_io_state);
3408
3409 // vthread != threadObj
3410 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3411 record_for_igvn(vthread_compare_rgn);
3412 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3413 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3414 record_for_igvn(vthread_compare_io);
3415 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3416 record_for_igvn(tid);
3417 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3418 record_for_igvn(exclusion);
3419 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3420 record_for_igvn(pinVirtualThread);
3421
3422 // Update control and phi nodes.
3423 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3424 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3425 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3426 vthread_compare_mem->init_req(_false_path, input_memory_state);
3427 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3428 vthread_compare_io->init_req(_false_path, input_io_state);
3429 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3430 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3431 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3432 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3433 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3434 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3435
3436 // Update branch state.
3437 set_control(_gvn.transform(vthread_compare_rgn));
3438 set_all_memory(_gvn.transform(vthread_compare_mem));
3439 set_i_o(_gvn.transform(vthread_compare_io));
3440
3441 // Load the event writer oop by dereferencing the jobject handle.
3442 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3443 assert(klass_EventWriter->is_loaded(), "invariant");
3444 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3445 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3446 const TypeOopPtr* const xtype = aklass->as_instance_type();
3447 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3448 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3449
3450 // Load the current thread id from the event writer object.
3451 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3452 // Get the field offset to, conditionally, store an updated tid value later.
3453 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3454 // Get the field offset to, conditionally, store an updated exclusion value later.
3455 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3456 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3457 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3458
3459 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3460 record_for_igvn(event_writer_tid_compare_rgn);
3461 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3462 record_for_igvn(event_writer_tid_compare_mem);
3463 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3464 record_for_igvn(event_writer_tid_compare_io);
3465
3466 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3467 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3468 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3469 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3470
3471 // False path, tids are the same.
3472 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3473
3474 // True path, tid is not equal, need to update the tid in the event writer.
3475 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3476 record_for_igvn(tid_is_not_equal);
3477
3478 // Store the pin state to the event writer.
3479 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3480
3481 // Store the exclusion state to the event writer.
3482 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3483 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3484
3485 // Store the tid to the event writer.
3486 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3487
3488 // Update control and phi nodes.
3489 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3490 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3491 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3492 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3493 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3494 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3495
3496 // Result of top level CFG, Memory, IO and Value.
3497 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3498 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3499 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3500 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3501
3502 // Result control.
3503 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3504 result_rgn->init_req(_false_path, jobj_is_null);
3505
3506 // Result memory.
3507 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3508 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3509
3510 // Result IO.
3511 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3512 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3513
3514 // Result value.
3515 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3516 result_value->init_req(_false_path, null()); // return null
3517
3518 // Set output state.
3519 set_control(_gvn.transform(result_rgn));
3520 set_all_memory(_gvn.transform(result_mem));
3521 set_i_o(_gvn.transform(result_io));
3522 set_result(result_rgn, result_value);
3523 return true;
3524 }
3525
3526 /*
3527 * The intrinsic is a model of this pseudo-code:
3528 *
3529 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3530 * if (carrierThread != thread) { // is virtual thread
3531 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3532 * bool excluded = vthread_epoch_raw & excluded_mask;
3533 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3534 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3535 * if (!excluded) {
3536 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3537 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3538 * }
3539 * Atomic::release_store(&tl->_vthread, true);
3540 * return;
3541 * }
3542 * Atomic::release_store(&tl->_vthread, false);
3543 */
3544 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3545 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3546
3547 Node* input_memory_state = reset_memory();
3548 set_all_memory(input_memory_state);
3549
3550 // The most significant bit of the u2 is used to denote thread exclusion
3551 Node* excluded_mask = _gvn.intcon(1 << 15);
3552 // The epoch generation is the range [1-32767]
3553 Node* epoch_mask = _gvn.intcon(32767);
3554
3555 Node* const carrierThread = generate_current_thread(jt);
3556 // If thread != carrierThread, this is a virtual thread.
3557 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3558 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3559 IfNode* iff_thread_not_equal_carrierThread =
3560 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3561
3562 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3563
3564 // False branch, is carrierThread.
3565 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3566 // Store release
3567 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3568
3569 set_all_memory(input_memory_state);
3570
3571 // True branch, is virtual thread.
3572 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3573 set_control(thread_not_equal_carrierThread);
3574
3575 // Load the raw epoch value from the vthread.
3576 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3577 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
3578 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3579
3580 // Mask off the excluded information from the epoch.
3581 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3582
3583 // Load the tid field from the thread.
3584 Node* tid = load_field_from_object(thread, "tid", "J");
3585
3586 // Store the vthread tid to the jfr thread local.
3587 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3588 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
3589
3590 // Branch is_excluded to conditionalize updating the epoch .
3591 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3592 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3593 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3594
3595 // True branch, vthread is excluded, no need to write epoch info.
3596 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3597 set_control(excluded);
3598 Node* vthread_is_excluded = _gvn.intcon(1);
3599
3600 // False branch, vthread is included, update epoch info.
3601 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3602 set_control(included);
3603 Node* vthread_is_included = _gvn.intcon(0);
3604
3605 // Get epoch value.
3606 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3607
3608 // Store the vthread epoch to the jfr thread local.
3609 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3610 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
3611
3612 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3613 record_for_igvn(excluded_rgn);
3614 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3615 record_for_igvn(excluded_mem);
3616 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3617 record_for_igvn(exclusion);
3618
3619 // Merge the excluded control and memory.
3620 excluded_rgn->init_req(_true_path, excluded);
3621 excluded_rgn->init_req(_false_path, included);
3622 excluded_mem->init_req(_true_path, tid_memory);
3623 excluded_mem->init_req(_false_path, included_memory);
3624 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3625 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3626
3627 // Set intermediate state.
3628 set_control(_gvn.transform(excluded_rgn));
3629 set_all_memory(excluded_mem);
3630
3631 // Store the vthread exclusion state to the jfr thread local.
3632 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3633 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
3634
3635 // Store release
3636 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
3637
3638 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3639 record_for_igvn(thread_compare_rgn);
3640 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3641 record_for_igvn(thread_compare_mem);
3642 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3643 record_for_igvn(vthread);
3644
3645 // Merge the thread_compare control and memory.
3646 thread_compare_rgn->init_req(_true_path, control());
3647 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3648 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3649 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3650
3651 // Set output state.
3652 set_control(_gvn.transform(thread_compare_rgn));
3653 set_all_memory(_gvn.transform(thread_compare_mem));
3654 }
3655
3656 #endif // JFR_HAVE_INTRINSICS
3657
3658 //------------------------inline_native_currentCarrierThread------------------
3659 bool LibraryCallKit::inline_native_currentCarrierThread() {
3660 Node* junk = nullptr;
3661 set_result(generate_current_thread(junk));
3662 return true;
3663 }
3664
3665 //------------------------inline_native_currentThread------------------
3666 bool LibraryCallKit::inline_native_currentThread() {
3667 Node* junk = nullptr;
3668 set_result(generate_virtual_thread(junk));
3669 return true;
3670 }
3671
3672 //------------------------inline_native_setVthread------------------
3673 bool LibraryCallKit::inline_native_setCurrentThread() {
3674 assert(C->method()->changes_current_thread(),
3675 "method changes current Thread but is not annotated ChangesCurrentThread");
3676 Node* arr = argument(1);
3677 Node* thread = _gvn.transform(new ThreadLocalNode());
3678 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3679 Node* thread_obj_handle
3680 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3681 thread_obj_handle = _gvn.transform(thread_obj_handle);
3682 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3683 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3684
3685 // Change the _monitor_owner_id of the JavaThread
3686 Node* tid = load_field_from_object(arr, "tid", "J");
3687 Node* monitor_owner_id_offset = basic_plus_adr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
3688 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
3689
3690 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3691 return true;
3692 }
3693
3694 const Type* LibraryCallKit::scopedValueCache_type() {
3695 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3696 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3697 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3698
3699 // Because we create the scopedValue cache lazily we have to make the
3700 // type of the result BotPTR.
3701 bool xk = etype->klass_is_exact();
3702 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3703 return objects_type;
3704 }
3705
3706 Node* LibraryCallKit::scopedValueCache_helper() {
3707 Node* thread = _gvn.transform(new ThreadLocalNode());
3708 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3709 // We cannot use immutable_memory() because we might flip onto a
3710 // different carrier thread, at which point we'll need to use that
3711 // carrier thread's cache.
3712 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3713 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3714 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3715 }
3716
3717 //------------------------inline_native_scopedValueCache------------------
3718 bool LibraryCallKit::inline_native_scopedValueCache() {
3719 Node* cache_obj_handle = scopedValueCache_helper();
3720 const Type* objects_type = scopedValueCache_type();
3721 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3722
3723 return true;
3724 }
3725
3726 //------------------------inline_native_setScopedValueCache------------------
3727 bool LibraryCallKit::inline_native_setScopedValueCache() {
3728 Node* arr = argument(0);
3729 Node* cache_obj_handle = scopedValueCache_helper();
3730 const Type* objects_type = scopedValueCache_type();
3731
3732 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3733 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3734
3735 return true;
3736 }
3737
3738 //------------------------inline_native_Continuation_pin and unpin-----------
3739
3740 // Shared implementation routine for both pin and unpin.
3741 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
3742 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3743
3744 // Save input memory.
3745 Node* input_memory_state = reset_memory();
3746 set_all_memory(input_memory_state);
3747
3748 // TLS
3749 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3750 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
3751 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
3752
3753 // Null check the last continuation object.
3754 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
3755 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
3756 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3757
3758 // False path, last continuation is null.
3759 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
3760
3761 // True path, last continuation is not null.
3762 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
3763
3764 set_control(continuation_is_not_null);
3765
3766 // Load the pin count from the last continuation.
3767 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
3768 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
3769
3770 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
3771 Node* pin_count_rhs;
3772 if (unpin) {
3773 pin_count_rhs = _gvn.intcon(0);
3774 } else {
3775 pin_count_rhs = _gvn.intcon(UINT32_MAX);
3776 }
3777 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
3778 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
3779 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
3780
3781 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
3782 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
3783 set_control(valid_pin_count);
3784
3785 Node* next_pin_count;
3786 if (unpin) {
3787 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
3788 } else {
3789 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
3790 }
3791
3792 Node* updated_pin_count_memory = store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
3793
3794 // True branch, pin count over/underflow.
3795 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
3796 {
3797 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
3798 // which will throw IllegalStateException for pin count over/underflow.
3799 PreserveJVMState pjvms(this);
3800 set_control(pin_count_over_underflow);
3801 set_all_memory(input_memory_state);
3802 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
3803 Deoptimization::Action_none);
3804 assert(stopped(), "invariant");
3805 }
3806
3807 // Result of top level CFG and Memory.
3808 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3809 record_for_igvn(result_rgn);
3810 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3811 record_for_igvn(result_mem);
3812
3813 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
3814 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
3815 result_mem->init_req(_true_path, _gvn.transform(updated_pin_count_memory));
3816 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3817
3818 // Set output state.
3819 set_control(_gvn.transform(result_rgn));
3820 set_all_memory(_gvn.transform(result_mem));
3821
3822 return true;
3823 }
3824
3825 //---------------------------load_mirror_from_klass----------------------------
3826 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3827 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3828 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3829 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3830 // mirror = ((OopHandle)mirror)->resolve();
3831 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3832 }
3833
3834 //-----------------------load_klass_from_mirror_common-------------------------
3835 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3836 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3837 // and branch to the given path on the region.
3838 // If never_see_null, take an uncommon trap on null, so we can optimistically
3839 // compile for the non-null case.
3840 // If the region is null, force never_see_null = true.
3841 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3842 bool never_see_null,
3843 RegionNode* region,
3844 int null_path,
3845 int offset) {
3846 if (region == nullptr) never_see_null = true;
3847 Node* p = basic_plus_adr(mirror, offset);
3848 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3849 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3850 Node* null_ctl = top();
3851 kls = null_check_oop(kls, &null_ctl, never_see_null);
3852 if (region != nullptr) {
3853 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3854 region->init_req(null_path, null_ctl);
3855 } else {
3856 assert(null_ctl == top(), "no loose ends");
3857 }
3858 return kls;
3859 }
3860
3861 //--------------------(inline_native_Class_query helpers)---------------------
3862 // Use this for JVM_ACC_INTERFACE.
3863 // Fall through if (mods & mask) == bits, take the guard otherwise.
3864 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
3865 ByteSize offset, const Type* type, BasicType bt) {
3866 // Branch around if the given klass has the given modifier bit set.
3867 // Like generate_guard, adds a new path onto the region.
3868 Node* modp = basic_plus_adr(kls, in_bytes(offset));
3869 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
3870 Node* mask = intcon(modifier_mask);
3871 Node* bits = intcon(modifier_bits);
3872 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3873 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3874 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3875 return generate_fair_guard(bol, region);
3876 }
3877 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3878 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
3879 Klass::access_flags_offset(), TypeInt::INT, T_INT);
3880 }
3881
3882 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
3883 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3884 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
3885 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
3886 }
3887
3888 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3889 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
3890 }
3891
3892 //-------------------------inline_native_Class_query-------------------
3893 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3894 const Type* return_type = TypeInt::BOOL;
3895 Node* prim_return_value = top(); // what happens if it's a primitive class?
3896 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3897 bool expect_prim = false; // most of these guys expect to work on refs
3898
3899 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3900
3901 Node* mirror = argument(0);
3902 Node* obj = top();
3903
3904 switch (id) {
3905 case vmIntrinsics::_isInstance:
3906 // nothing is an instance of a primitive type
3907 prim_return_value = intcon(0);
3908 obj = argument(1);
3909 break;
3910 case vmIntrinsics::_getModifiers:
3911 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3912 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3913 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3914 break;
3915 case vmIntrinsics::_isInterface:
3916 prim_return_value = intcon(0);
3917 break;
3918 case vmIntrinsics::_isArray:
3919 prim_return_value = intcon(0);
3920 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3921 break;
3922 case vmIntrinsics::_isPrimitive:
3923 prim_return_value = intcon(1);
3924 expect_prim = true; // obviously
3925 break;
3926 case vmIntrinsics::_isHidden:
3927 prim_return_value = intcon(0);
3928 break;
3929 case vmIntrinsics::_getSuperclass:
3930 prim_return_value = null();
3931 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3932 break;
3933 case vmIntrinsics::_getClassAccessFlags:
3934 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3935 return_type = TypeInt::INT; // not bool! 6297094
3936 break;
3937 default:
3938 fatal_unexpected_iid(id);
3939 break;
3940 }
3941
3942 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3943 if (mirror_con == nullptr) return false; // cannot happen?
3944
3945 #ifndef PRODUCT
3946 if (C->print_intrinsics() || C->print_inlining()) {
3947 ciType* k = mirror_con->java_mirror_type();
3948 if (k) {
3949 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3950 k->print_name();
3951 tty->cr();
3952 }
3953 }
3954 #endif
3955
3956 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3957 RegionNode* region = new RegionNode(PATH_LIMIT);
3958 record_for_igvn(region);
3959 PhiNode* phi = new PhiNode(region, return_type);
3960
3961 // The mirror will never be null of Reflection.getClassAccessFlags, however
3962 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3963 // if it is. See bug 4774291.
3964
3965 // For Reflection.getClassAccessFlags(), the null check occurs in
3966 // the wrong place; see inline_unsafe_access(), above, for a similar
3967 // situation.
3968 mirror = null_check(mirror);
3969 // If mirror or obj is dead, only null-path is taken.
3970 if (stopped()) return true;
3971
3972 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3973
3974 // Now load the mirror's klass metaobject, and null-check it.
3975 // Side-effects region with the control path if the klass is null.
3976 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3977 // If kls is null, we have a primitive mirror.
3978 phi->init_req(_prim_path, prim_return_value);
3979 if (stopped()) { set_result(region, phi); return true; }
3980 bool safe_for_replace = (region->in(_prim_path) == top());
3981
3982 Node* p; // handy temp
3983 Node* null_ctl;
3984
3985 // Now that we have the non-null klass, we can perform the real query.
3986 // For constant classes, the query will constant-fold in LoadNode::Value.
3987 Node* query_value = top();
3988 switch (id) {
3989 case vmIntrinsics::_isInstance:
3990 // nothing is an instance of a primitive type
3991 query_value = gen_instanceof(obj, kls, safe_for_replace);
3992 break;
3993
3994 case vmIntrinsics::_getModifiers:
3995 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3996 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
3997 break;
3998
3999 case vmIntrinsics::_isInterface:
4000 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4001 if (generate_interface_guard(kls, region) != nullptr)
4002 // A guard was added. If the guard is taken, it was an interface.
4003 phi->add_req(intcon(1));
4004 // If we fall through, it's a plain class.
4005 query_value = intcon(0);
4006 break;
4007
4008 case vmIntrinsics::_isArray:
4009 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
4010 if (generate_array_guard(kls, region) != nullptr)
4011 // A guard was added. If the guard is taken, it was an array.
4012 phi->add_req(intcon(1));
4013 // If we fall through, it's a plain class.
4014 query_value = intcon(0);
4015 break;
4016
4017 case vmIntrinsics::_isPrimitive:
4018 query_value = intcon(0); // "normal" path produces false
4019 break;
4020
4021 case vmIntrinsics::_isHidden:
4022 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4023 if (generate_hidden_class_guard(kls, region) != nullptr)
4024 // A guard was added. If the guard is taken, it was an hidden class.
4025 phi->add_req(intcon(1));
4026 // If we fall through, it's a plain class.
4027 query_value = intcon(0);
4028 break;
4029
4030
4031 case vmIntrinsics::_getSuperclass:
4032 // The rules here are somewhat unfortunate, but we can still do better
4033 // with random logic than with a JNI call.
4034 // Interfaces store null or Object as _super, but must report null.
4035 // Arrays store an intermediate super as _super, but must report Object.
4036 // Other types can report the actual _super.
4037 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4038 if (generate_interface_guard(kls, region) != nullptr)
4039 // A guard was added. If the guard is taken, it was an interface.
4040 phi->add_req(null());
4041 if (generate_array_guard(kls, region) != nullptr)
4042 // A guard was added. If the guard is taken, it was an array.
4043 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4044 // If we fall through, it's a plain class. Get its _super.
4045 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4046 kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4047 null_ctl = top();
4048 kls = null_check_oop(kls, &null_ctl);
4049 if (null_ctl != top()) {
4050 // If the guard is taken, Object.superClass is null (both klass and mirror).
4051 region->add_req(null_ctl);
4052 phi ->add_req(null());
4053 }
4054 if (!stopped()) {
4055 query_value = load_mirror_from_klass(kls);
4056 }
4057 break;
4058
4059 case vmIntrinsics::_getClassAccessFlags:
4060 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
4061 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
4062 break;
4063
4064 default:
4065 fatal_unexpected_iid(id);
4066 break;
4067 }
4068
4069 // Fall-through is the normal case of a query to a real class.
4070 phi->init_req(1, query_value);
4071 region->init_req(1, control());
4072
4073 C->set_has_split_ifs(true); // Has chance for split-if optimization
4074 set_result(region, phi);
4075 return true;
4076 }
4077
4078 //-------------------------inline_Class_cast-------------------
4079 bool LibraryCallKit::inline_Class_cast() {
4080 Node* mirror = argument(0); // Class
4081 Node* obj = argument(1);
4082 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4083 if (mirror_con == nullptr) {
4084 return false; // dead path (mirror->is_top()).
4085 }
4086 if (obj == nullptr || obj->is_top()) {
4087 return false; // dead path
4088 }
4089 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4090
4091 // First, see if Class.cast() can be folded statically.
4092 // java_mirror_type() returns non-null for compile-time Class constants.
4093 ciType* tm = mirror_con->java_mirror_type();
4094 if (tm != nullptr && tm->is_klass() &&
4095 tp != nullptr) {
4096 if (!tp->is_loaded()) {
4097 // Don't use intrinsic when class is not loaded.
4098 return false;
4099 } else {
4100 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
4101 if (static_res == Compile::SSC_always_true) {
4102 // isInstance() is true - fold the code.
4103 set_result(obj);
4104 return true;
4105 } else if (static_res == Compile::SSC_always_false) {
4106 // Don't use intrinsic, have to throw ClassCastException.
4107 // If the reference is null, the non-intrinsic bytecode will
4108 // be optimized appropriately.
4109 return false;
4110 }
4111 }
4112 }
4113
4114 // Bailout intrinsic and do normal inlining if exception path is frequent.
4115 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4116 return false;
4117 }
4118
4119 // Generate dynamic checks.
4120 // Class.cast() is java implementation of _checkcast bytecode.
4121 // Do checkcast (Parse::do_checkcast()) optimizations here.
4122
4123 mirror = null_check(mirror);
4124 // If mirror is dead, only null-path is taken.
4125 if (stopped()) {
4126 return true;
4127 }
4128
4129 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive).
4130 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
4131 RegionNode* region = new RegionNode(PATH_LIMIT);
4132 record_for_igvn(region);
4133
4134 // Now load the mirror's klass metaobject, and null-check it.
4135 // If kls is null, we have a primitive mirror and
4136 // nothing is an instance of a primitive type.
4137 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4138
4139 Node* res = top();
4140 if (!stopped()) {
4141 Node* bad_type_ctrl = top();
4142 // Do checkcast optimizations.
4143 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4144 region->init_req(_bad_type_path, bad_type_ctrl);
4145 }
4146 if (region->in(_prim_path) != top() ||
4147 region->in(_bad_type_path) != top()) {
4148 // Let Interpreter throw ClassCastException.
4149 PreserveJVMState pjvms(this);
4150 set_control(_gvn.transform(region));
4151 uncommon_trap(Deoptimization::Reason_intrinsic,
4152 Deoptimization::Action_maybe_recompile);
4153 }
4154 if (!stopped()) {
4155 set_result(res);
4156 }
4157 return true;
4158 }
4159
4160
4161 //--------------------------inline_native_subtype_check------------------------
4162 // This intrinsic takes the JNI calls out of the heart of
4163 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4164 bool LibraryCallKit::inline_native_subtype_check() {
4165 // Pull both arguments off the stack.
4166 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4167 args[0] = argument(0);
4168 args[1] = argument(1);
4169 Node* klasses[2]; // corresponding Klasses: superk, subk
4170 klasses[0] = klasses[1] = top();
4171
4172 enum {
4173 // A full decision tree on {superc is prim, subc is prim}:
4174 _prim_0_path = 1, // {P,N} => false
4175 // {P,P} & superc!=subc => false
4176 _prim_same_path, // {P,P} & superc==subc => true
4177 _prim_1_path, // {N,P} => false
4178 _ref_subtype_path, // {N,N} & subtype check wins => true
4179 _both_ref_path, // {N,N} & subtype check loses => false
4180 PATH_LIMIT
4181 };
4182
4183 RegionNode* region = new RegionNode(PATH_LIMIT);
4184 Node* phi = new PhiNode(region, TypeInt::BOOL);
4185 record_for_igvn(region);
4186
4187 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4188 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4189 int class_klass_offset = java_lang_Class::klass_offset();
4190
4191 // First null-check both mirrors and load each mirror's klass metaobject.
4192 int which_arg;
4193 for (which_arg = 0; which_arg <= 1; which_arg++) {
4194 Node* arg = args[which_arg];
4195 arg = null_check(arg);
4196 if (stopped()) break;
4197 args[which_arg] = arg;
4198
4199 Node* p = basic_plus_adr(arg, class_klass_offset);
4200 Node* kls = LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, adr_type, kls_type);
4201 klasses[which_arg] = _gvn.transform(kls);
4202 }
4203
4204 // Having loaded both klasses, test each for null.
4205 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4206 for (which_arg = 0; which_arg <= 1; which_arg++) {
4207 Node* kls = klasses[which_arg];
4208 Node* null_ctl = top();
4209 kls = null_check_oop(kls, &null_ctl, never_see_null);
4210 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
4211 region->init_req(prim_path, null_ctl);
4212 if (stopped()) break;
4213 klasses[which_arg] = kls;
4214 }
4215
4216 if (!stopped()) {
4217 // now we have two reference types, in klasses[0..1]
4218 Node* subk = klasses[1]; // the argument to isAssignableFrom
4219 Node* superk = klasses[0]; // the receiver
4220 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4221 // now we have a successful reference subtype check
4222 region->set_req(_ref_subtype_path, control());
4223 }
4224
4225 // If both operands are primitive (both klasses null), then
4226 // we must return true when they are identical primitives.
4227 // It is convenient to test this after the first null klass check.
4228 set_control(region->in(_prim_0_path)); // go back to first null check
4229 if (!stopped()) {
4230 // Since superc is primitive, make a guard for the superc==subc case.
4231 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4232 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4233 generate_guard(bol_eq, region, PROB_FAIR);
4234 if (region->req() == PATH_LIMIT+1) {
4235 // A guard was added. If the added guard is taken, superc==subc.
4236 region->swap_edges(PATH_LIMIT, _prim_same_path);
4237 region->del_req(PATH_LIMIT);
4238 }
4239 region->set_req(_prim_0_path, control()); // Not equal after all.
4240 }
4241
4242 // these are the only paths that produce 'true':
4243 phi->set_req(_prim_same_path, intcon(1));
4244 phi->set_req(_ref_subtype_path, intcon(1));
4245
4246 // pull together the cases:
4247 assert(region->req() == PATH_LIMIT, "sane region");
4248 for (uint i = 1; i < region->req(); i++) {
4249 Node* ctl = region->in(i);
4250 if (ctl == nullptr || ctl == top()) {
4251 region->set_req(i, top());
4252 phi ->set_req(i, top());
4253 } else if (phi->in(i) == nullptr) {
4254 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4255 }
4256 }
4257
4258 set_control(_gvn.transform(region));
4259 set_result(_gvn.transform(phi));
4260 return true;
4261 }
4262
4263 //---------------------generate_array_guard_common------------------------
4264 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
4265 bool obj_array, bool not_array) {
4266
4267 if (stopped()) {
4268 return nullptr;
4269 }
4270
4271 // If obj_array/non_array==false/false:
4272 // Branch around if the given klass is in fact an array (either obj or prim).
4273 // If obj_array/non_array==false/true:
4274 // Branch around if the given klass is not an array klass of any kind.
4275 // If obj_array/non_array==true/true:
4276 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
4277 // If obj_array/non_array==true/false:
4278 // Branch around if the kls is an oop array (Object[] or subtype)
4279 //
4280 // Like generate_guard, adds a new path onto the region.
4281 jint layout_con = 0;
4282 Node* layout_val = get_layout_helper(kls, layout_con);
4283 if (layout_val == nullptr) {
4284 bool query = (obj_array
4285 ? Klass::layout_helper_is_objArray(layout_con)
4286 : Klass::layout_helper_is_array(layout_con));
4287 if (query == not_array) {
4288 return nullptr; // never a branch
4289 } else { // always a branch
4290 Node* always_branch = control();
4291 if (region != nullptr)
4292 region->add_req(always_branch);
4293 set_control(top());
4294 return always_branch;
4295 }
4296 }
4297 // Now test the correct condition.
4298 jint nval = (obj_array
4299 ? (jint)(Klass::_lh_array_tag_type_value
4300 << Klass::_lh_array_tag_shift)
4301 : Klass::_lh_neutral_value);
4302 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4303 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
4304 // invert the test if we are looking for a non-array
4305 if (not_array) btest = BoolTest(btest).negate();
4306 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4307 return generate_fair_guard(bol, region);
4308 }
4309
4310
4311 //-----------------------inline_native_newArray--------------------------
4312 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4313 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4314 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4315 Node* mirror;
4316 Node* count_val;
4317 if (uninitialized) {
4318 null_check_receiver();
4319 mirror = argument(1);
4320 count_val = argument(2);
4321 } else {
4322 mirror = argument(0);
4323 count_val = argument(1);
4324 }
4325
4326 mirror = null_check(mirror);
4327 // If mirror or obj is dead, only null-path is taken.
4328 if (stopped()) return true;
4329
4330 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4331 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4332 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4333 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4334 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4335
4336 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4337 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4338 result_reg, _slow_path);
4339 Node* normal_ctl = control();
4340 Node* no_array_ctl = result_reg->in(_slow_path);
4341
4342 // Generate code for the slow case. We make a call to newArray().
4343 set_control(no_array_ctl);
4344 if (!stopped()) {
4345 // Either the input type is void.class, or else the
4346 // array klass has not yet been cached. Either the
4347 // ensuing call will throw an exception, or else it
4348 // will cache the array klass for next time.
4349 PreserveJVMState pjvms(this);
4350 CallJavaNode* slow_call = nullptr;
4351 if (uninitialized) {
4352 // Generate optimized virtual call (holder class 'Unsafe' is final)
4353 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4354 } else {
4355 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4356 }
4357 Node* slow_result = set_results_for_java_call(slow_call);
4358 // this->control() comes from set_results_for_java_call
4359 result_reg->set_req(_slow_path, control());
4360 result_val->set_req(_slow_path, slow_result);
4361 result_io ->set_req(_slow_path, i_o());
4362 result_mem->set_req(_slow_path, reset_memory());
4363 }
4364
4365 set_control(normal_ctl);
4366 if (!stopped()) {
4367 // Normal case: The array type has been cached in the java.lang.Class.
4368 // The following call works fine even if the array type is polymorphic.
4369 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4370 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4371 result_reg->init_req(_normal_path, control());
4372 result_val->init_req(_normal_path, obj);
4373 result_io ->init_req(_normal_path, i_o());
4374 result_mem->init_req(_normal_path, reset_memory());
4375
4376 if (uninitialized) {
4377 // Mark the allocation so that zeroing is skipped
4378 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4379 alloc->maybe_set_complete(&_gvn);
4380 }
4381 }
4382
4383 // Return the combined state.
4384 set_i_o( _gvn.transform(result_io) );
4385 set_all_memory( _gvn.transform(result_mem));
4386
4387 C->set_has_split_ifs(true); // Has chance for split-if optimization
4388 set_result(result_reg, result_val);
4389 return true;
4390 }
4391
4392 //----------------------inline_native_getLength--------------------------
4393 // public static native int java.lang.reflect.Array.getLength(Object array);
4394 bool LibraryCallKit::inline_native_getLength() {
4395 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4396
4397 Node* array = null_check(argument(0));
4398 // If array is dead, only null-path is taken.
4399 if (stopped()) return true;
4400
4401 // Deoptimize if it is a non-array.
4402 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr);
4403
4404 if (non_array != nullptr) {
4405 PreserveJVMState pjvms(this);
4406 set_control(non_array);
4407 uncommon_trap(Deoptimization::Reason_intrinsic,
4408 Deoptimization::Action_maybe_recompile);
4409 }
4410
4411 // If control is dead, only non-array-path is taken.
4412 if (stopped()) return true;
4413
4414 // The works fine even if the array type is polymorphic.
4415 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4416 Node* result = load_array_length(array);
4417
4418 C->set_has_split_ifs(true); // Has chance for split-if optimization
4419 set_result(result);
4420 return true;
4421 }
4422
4423 //------------------------inline_array_copyOf----------------------------
4424 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4425 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4426 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4427 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4428
4429 // Get the arguments.
4430 Node* original = argument(0);
4431 Node* start = is_copyOfRange? argument(1): intcon(0);
4432 Node* end = is_copyOfRange? argument(2): argument(1);
4433 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4434
4435 Node* newcopy = nullptr;
4436
4437 // Set the original stack and the reexecute bit for the interpreter to reexecute
4438 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4439 { PreserveReexecuteState preexecs(this);
4440 jvms()->set_should_reexecute(true);
4441
4442 array_type_mirror = null_check(array_type_mirror);
4443 original = null_check(original);
4444
4445 // Check if a null path was taken unconditionally.
4446 if (stopped()) return true;
4447
4448 Node* orig_length = load_array_length(original);
4449
4450 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4451 klass_node = null_check(klass_node);
4452
4453 RegionNode* bailout = new RegionNode(1);
4454 record_for_igvn(bailout);
4455
4456 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4457 // Bail out if that is so.
4458 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4459 if (not_objArray != nullptr) {
4460 // Improve the klass node's type from the new optimistic assumption:
4461 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4462 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4463 Node* cast = new CastPPNode(control(), klass_node, akls);
4464 klass_node = _gvn.transform(cast);
4465 }
4466
4467 // Bail out if either start or end is negative.
4468 generate_negative_guard(start, bailout, &start);
4469 generate_negative_guard(end, bailout, &end);
4470
4471 Node* length = end;
4472 if (_gvn.type(start) != TypeInt::ZERO) {
4473 length = _gvn.transform(new SubINode(end, start));
4474 }
4475
4476 // Bail out if length is negative (i.e., if start > end).
4477 // Without this the new_array would throw
4478 // NegativeArraySizeException but IllegalArgumentException is what
4479 // should be thrown
4480 generate_negative_guard(length, bailout, &length);
4481
4482 // Bail out if start is larger than the original length
4483 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4484 generate_negative_guard(orig_tail, bailout, &orig_tail);
4485
4486 if (bailout->req() > 1) {
4487 PreserveJVMState pjvms(this);
4488 set_control(_gvn.transform(bailout));
4489 uncommon_trap(Deoptimization::Reason_intrinsic,
4490 Deoptimization::Action_maybe_recompile);
4491 }
4492
4493 if (!stopped()) {
4494 // How many elements will we copy from the original?
4495 // The answer is MinI(orig_tail, length).
4496 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4497
4498 // Generate a direct call to the right arraycopy function(s).
4499 // We know the copy is disjoint but we might not know if the
4500 // oop stores need checking.
4501 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4502 // This will fail a store-check if x contains any non-nulls.
4503
4504 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4505 // loads/stores but it is legal only if we're sure the
4506 // Arrays.copyOf would succeed. So we need all input arguments
4507 // to the copyOf to be validated, including that the copy to the
4508 // new array won't trigger an ArrayStoreException. That subtype
4509 // check can be optimized if we know something on the type of
4510 // the input array from type speculation.
4511 if (_gvn.type(klass_node)->singleton()) {
4512 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4513 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4514
4515 int test = C->static_subtype_check(superk, subk);
4516 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4517 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4518 if (t_original->speculative_type() != nullptr) {
4519 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4520 }
4521 }
4522 }
4523
4524 bool validated = false;
4525 // Reason_class_check rather than Reason_intrinsic because we
4526 // want to intrinsify even if this traps.
4527 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4528 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4529
4530 if (not_subtype_ctrl != top()) {
4531 PreserveJVMState pjvms(this);
4532 set_control(not_subtype_ctrl);
4533 uncommon_trap(Deoptimization::Reason_class_check,
4534 Deoptimization::Action_make_not_entrant);
4535 assert(stopped(), "Should be stopped");
4536 }
4537 validated = true;
4538 }
4539
4540 if (!stopped()) {
4541 newcopy = new_array(klass_node, length, 0); // no arguments to push
4542
4543 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4544 load_object_klass(original), klass_node);
4545 if (!is_copyOfRange) {
4546 ac->set_copyof(validated);
4547 } else {
4548 ac->set_copyofrange(validated);
4549 }
4550 Node* n = _gvn.transform(ac);
4551 if (n == ac) {
4552 ac->connect_outputs(this);
4553 } else {
4554 assert(validated, "shouldn't transform if all arguments not validated");
4555 set_all_memory(n);
4556 }
4557 }
4558 }
4559 } // original reexecute is set back here
4560
4561 C->set_has_split_ifs(true); // Has chance for split-if optimization
4562 if (!stopped()) {
4563 set_result(newcopy);
4564 }
4565 return true;
4566 }
4567
4568
4569 //----------------------generate_virtual_guard---------------------------
4570 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4571 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4572 RegionNode* slow_region) {
4573 ciMethod* method = callee();
4574 int vtable_index = method->vtable_index();
4575 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4576 "bad index %d", vtable_index);
4577 // Get the Method* out of the appropriate vtable entry.
4578 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4579 vtable_index*vtableEntry::size_in_bytes() +
4580 in_bytes(vtableEntry::method_offset());
4581 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4582 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4583
4584 // Compare the target method with the expected method (e.g., Object.hashCode).
4585 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4586
4587 Node* native_call = makecon(native_call_addr);
4588 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4589 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4590
4591 return generate_slow_guard(test_native, slow_region);
4592 }
4593
4594 //-----------------------generate_method_call----------------------------
4595 // Use generate_method_call to make a slow-call to the real
4596 // method if the fast path fails. An alternative would be to
4597 // use a stub like OptoRuntime::slow_arraycopy_Java.
4598 // This only works for expanding the current library call,
4599 // not another intrinsic. (E.g., don't use this for making an
4600 // arraycopy call inside of the copyOf intrinsic.)
4601 CallJavaNode*
4602 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4603 // When compiling the intrinsic method itself, do not use this technique.
4604 guarantee(callee() != C->method(), "cannot make slow-call to self");
4605
4606 ciMethod* method = callee();
4607 // ensure the JVMS we have will be correct for this call
4608 guarantee(method_id == method->intrinsic_id(), "must match");
4609
4610 const TypeFunc* tf = TypeFunc::make(method);
4611 if (res_not_null) {
4612 assert(tf->return_type() == T_OBJECT, "");
4613 const TypeTuple* range = tf->range();
4614 const Type** fields = TypeTuple::fields(range->cnt());
4615 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4616 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4617 tf = TypeFunc::make(tf->domain(), new_range);
4618 }
4619 CallJavaNode* slow_call;
4620 if (is_static) {
4621 assert(!is_virtual, "");
4622 slow_call = new CallStaticJavaNode(C, tf,
4623 SharedRuntime::get_resolve_static_call_stub(), method);
4624 } else if (is_virtual) {
4625 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4626 int vtable_index = Method::invalid_vtable_index;
4627 if (UseInlineCaches) {
4628 // Suppress the vtable call
4629 } else {
4630 // hashCode and clone are not a miranda methods,
4631 // so the vtable index is fixed.
4632 // No need to use the linkResolver to get it.
4633 vtable_index = method->vtable_index();
4634 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4635 "bad index %d", vtable_index);
4636 }
4637 slow_call = new CallDynamicJavaNode(tf,
4638 SharedRuntime::get_resolve_virtual_call_stub(),
4639 method, vtable_index);
4640 } else { // neither virtual nor static: opt_virtual
4641 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4642 slow_call = new CallStaticJavaNode(C, tf,
4643 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4644 slow_call->set_optimized_virtual(true);
4645 }
4646 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4647 // To be able to issue a direct call (optimized virtual or virtual)
4648 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4649 // about the method being invoked should be attached to the call site to
4650 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4651 slow_call->set_override_symbolic_info(true);
4652 }
4653 set_arguments_for_java_call(slow_call);
4654 set_edges_for_java_call(slow_call);
4655 return slow_call;
4656 }
4657
4658
4659 /**
4660 * Build special case code for calls to hashCode on an object. This call may
4661 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4662 * slightly different code.
4663 */
4664 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4665 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4666 assert(!(is_virtual && is_static), "either virtual, special, or static");
4667
4668 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4669
4670 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4671 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4672 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4673 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4674 Node* obj = nullptr;
4675 if (!is_static) {
4676 // Check for hashing null object
4677 obj = null_check_receiver();
4678 if (stopped()) return true; // unconditionally null
4679 result_reg->init_req(_null_path, top());
4680 result_val->init_req(_null_path, top());
4681 } else {
4682 // Do a null check, and return zero if null.
4683 // System.identityHashCode(null) == 0
4684 obj = argument(0);
4685 Node* null_ctl = top();
4686 obj = null_check_oop(obj, &null_ctl);
4687 result_reg->init_req(_null_path, null_ctl);
4688 result_val->init_req(_null_path, _gvn.intcon(0));
4689 }
4690
4691 // Unconditionally null? Then return right away.
4692 if (stopped()) {
4693 set_control( result_reg->in(_null_path));
4694 if (!stopped())
4695 set_result(result_val->in(_null_path));
4696 return true;
4697 }
4698
4699 // We only go to the fast case code if we pass a number of guards. The
4700 // paths which do not pass are accumulated in the slow_region.
4701 RegionNode* slow_region = new RegionNode(1);
4702 record_for_igvn(slow_region);
4703
4704 // If this is a virtual call, we generate a funny guard. We pull out
4705 // the vtable entry corresponding to hashCode() from the target object.
4706 // If the target method which we are calling happens to be the native
4707 // Object hashCode() method, we pass the guard. We do not need this
4708 // guard for non-virtual calls -- the caller is known to be the native
4709 // Object hashCode().
4710 if (is_virtual) {
4711 // After null check, get the object's klass.
4712 Node* obj_klass = load_object_klass(obj);
4713 generate_virtual_guard(obj_klass, slow_region);
4714 }
4715
4716 // Get the header out of the object, use LoadMarkNode when available
4717 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4718 // The control of the load must be null. Otherwise, the load can move before
4719 // the null check after castPP removal.
4720 Node* no_ctrl = nullptr;
4721 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4722
4723 if (!UseObjectMonitorTable) {
4724 // Test the header to see if it is safe to read w.r.t. locking.
4725 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4726 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4727 if (LockingMode == LM_LIGHTWEIGHT) {
4728 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
4729 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
4730 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
4731
4732 generate_slow_guard(test_monitor, slow_region);
4733 } else {
4734 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4735 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val));
4736 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne));
4737
4738 generate_slow_guard(test_not_unlocked, slow_region);
4739 }
4740 }
4741
4742 // Get the hash value and check to see that it has been properly assigned.
4743 // We depend on hash_mask being at most 32 bits and avoid the use of
4744 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4745 // vm: see markWord.hpp.
4746 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4747 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4748 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4749 // This hack lets the hash bits live anywhere in the mark object now, as long
4750 // as the shift drops the relevant bits into the low 32 bits. Note that
4751 // Java spec says that HashCode is an int so there's no point in capturing
4752 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4753 hshifted_header = ConvX2I(hshifted_header);
4754 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4755
4756 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4757 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4758 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4759
4760 generate_slow_guard(test_assigned, slow_region);
4761
4762 Node* init_mem = reset_memory();
4763 // fill in the rest of the null path:
4764 result_io ->init_req(_null_path, i_o());
4765 result_mem->init_req(_null_path, init_mem);
4766
4767 result_val->init_req(_fast_path, hash_val);
4768 result_reg->init_req(_fast_path, control());
4769 result_io ->init_req(_fast_path, i_o());
4770 result_mem->init_req(_fast_path, init_mem);
4771
4772 // Generate code for the slow case. We make a call to hashCode().
4773 set_control(_gvn.transform(slow_region));
4774 if (!stopped()) {
4775 // No need for PreserveJVMState, because we're using up the present state.
4776 set_all_memory(init_mem);
4777 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4778 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4779 Node* slow_result = set_results_for_java_call(slow_call);
4780 // this->control() comes from set_results_for_java_call
4781 result_reg->init_req(_slow_path, control());
4782 result_val->init_req(_slow_path, slow_result);
4783 result_io ->set_req(_slow_path, i_o());
4784 result_mem ->set_req(_slow_path, reset_memory());
4785 }
4786
4787 // Return the combined state.
4788 set_i_o( _gvn.transform(result_io) );
4789 set_all_memory( _gvn.transform(result_mem));
4790
4791 set_result(result_reg, result_val);
4792 return true;
4793 }
4794
4795 //---------------------------inline_native_getClass----------------------------
4796 // public final native Class<?> java.lang.Object.getClass();
4797 //
4798 // Build special case code for calls to getClass on an object.
4799 bool LibraryCallKit::inline_native_getClass() {
4800 Node* obj = null_check_receiver();
4801 if (stopped()) return true;
4802 set_result(load_mirror_from_klass(load_object_klass(obj)));
4803 return true;
4804 }
4805
4806 //-----------------inline_native_Reflection_getCallerClass---------------------
4807 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4808 //
4809 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4810 //
4811 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4812 // in that it must skip particular security frames and checks for
4813 // caller sensitive methods.
4814 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4815 #ifndef PRODUCT
4816 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4817 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4818 }
4819 #endif
4820
4821 if (!jvms()->has_method()) {
4822 #ifndef PRODUCT
4823 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4824 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4825 }
4826 #endif
4827 return false;
4828 }
4829
4830 // Walk back up the JVM state to find the caller at the required
4831 // depth.
4832 JVMState* caller_jvms = jvms();
4833
4834 // Cf. JVM_GetCallerClass
4835 // NOTE: Start the loop at depth 1 because the current JVM state does
4836 // not include the Reflection.getCallerClass() frame.
4837 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4838 ciMethod* m = caller_jvms->method();
4839 switch (n) {
4840 case 0:
4841 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4842 break;
4843 case 1:
4844 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4845 if (!m->caller_sensitive()) {
4846 #ifndef PRODUCT
4847 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4848 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4849 }
4850 #endif
4851 return false; // bail-out; let JVM_GetCallerClass do the work
4852 }
4853 break;
4854 default:
4855 if (!m->is_ignored_by_security_stack_walk()) {
4856 // We have reached the desired frame; return the holder class.
4857 // Acquire method holder as java.lang.Class and push as constant.
4858 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4859 ciInstance* caller_mirror = caller_klass->java_mirror();
4860 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4861
4862 #ifndef PRODUCT
4863 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4864 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());
4865 tty->print_cr(" JVM state at this point:");
4866 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4867 ciMethod* m = jvms()->of_depth(i)->method();
4868 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4869 }
4870 }
4871 #endif
4872 return true;
4873 }
4874 break;
4875 }
4876 }
4877
4878 #ifndef PRODUCT
4879 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4880 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4881 tty->print_cr(" JVM state at this point:");
4882 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4883 ciMethod* m = jvms()->of_depth(i)->method();
4884 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4885 }
4886 }
4887 #endif
4888
4889 return false; // bail-out; let JVM_GetCallerClass do the work
4890 }
4891
4892 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4893 Node* arg = argument(0);
4894 Node* result = nullptr;
4895
4896 switch (id) {
4897 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4898 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4899 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4900 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4901 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
4902 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
4903
4904 case vmIntrinsics::_doubleToLongBits: {
4905 // two paths (plus control) merge in a wood
4906 RegionNode *r = new RegionNode(3);
4907 Node *phi = new PhiNode(r, TypeLong::LONG);
4908
4909 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4910 // Build the boolean node
4911 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4912
4913 // Branch either way.
4914 // NaN case is less traveled, which makes all the difference.
4915 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4916 Node *opt_isnan = _gvn.transform(ifisnan);
4917 assert( opt_isnan->is_If(), "Expect an IfNode");
4918 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4919 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4920
4921 set_control(iftrue);
4922
4923 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4924 Node *slow_result = longcon(nan_bits); // return NaN
4925 phi->init_req(1, _gvn.transform( slow_result ));
4926 r->init_req(1, iftrue);
4927
4928 // Else fall through
4929 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4930 set_control(iffalse);
4931
4932 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4933 r->init_req(2, iffalse);
4934
4935 // Post merge
4936 set_control(_gvn.transform(r));
4937 record_for_igvn(r);
4938
4939 C->set_has_split_ifs(true); // Has chance for split-if optimization
4940 result = phi;
4941 assert(result->bottom_type()->isa_long(), "must be");
4942 break;
4943 }
4944
4945 case vmIntrinsics::_floatToIntBits: {
4946 // two paths (plus control) merge in a wood
4947 RegionNode *r = new RegionNode(3);
4948 Node *phi = new PhiNode(r, TypeInt::INT);
4949
4950 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4951 // Build the boolean node
4952 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4953
4954 // Branch either way.
4955 // NaN case is less traveled, which makes all the difference.
4956 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4957 Node *opt_isnan = _gvn.transform(ifisnan);
4958 assert( opt_isnan->is_If(), "Expect an IfNode");
4959 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4960 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4961
4962 set_control(iftrue);
4963
4964 static const jint nan_bits = 0x7fc00000;
4965 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4966 phi->init_req(1, _gvn.transform( slow_result ));
4967 r->init_req(1, iftrue);
4968
4969 // Else fall through
4970 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4971 set_control(iffalse);
4972
4973 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4974 r->init_req(2, iffalse);
4975
4976 // Post merge
4977 set_control(_gvn.transform(r));
4978 record_for_igvn(r);
4979
4980 C->set_has_split_ifs(true); // Has chance for split-if optimization
4981 result = phi;
4982 assert(result->bottom_type()->isa_int(), "must be");
4983 break;
4984 }
4985
4986 default:
4987 fatal_unexpected_iid(id);
4988 break;
4989 }
4990 set_result(_gvn.transform(result));
4991 return true;
4992 }
4993
4994 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
4995 Node* arg = argument(0);
4996 Node* result = nullptr;
4997
4998 switch (id) {
4999 case vmIntrinsics::_floatIsInfinite:
5000 result = new IsInfiniteFNode(arg);
5001 break;
5002 case vmIntrinsics::_floatIsFinite:
5003 result = new IsFiniteFNode(arg);
5004 break;
5005 case vmIntrinsics::_doubleIsInfinite:
5006 result = new IsInfiniteDNode(arg);
5007 break;
5008 case vmIntrinsics::_doubleIsFinite:
5009 result = new IsFiniteDNode(arg);
5010 break;
5011 default:
5012 fatal_unexpected_iid(id);
5013 break;
5014 }
5015 set_result(_gvn.transform(result));
5016 return true;
5017 }
5018
5019 //----------------------inline_unsafe_copyMemory-------------------------
5020 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5021
5022 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5023 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5024 const Type* base_t = gvn.type(base);
5025
5026 bool in_native = (base_t == TypePtr::NULL_PTR);
5027 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5028 bool is_mixed = !in_heap && !in_native;
5029
5030 if (is_mixed) {
5031 return true; // mixed accesses can touch both on-heap and off-heap memory
5032 }
5033 if (in_heap) {
5034 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5035 if (!is_prim_array) {
5036 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5037 // there's not enough type information available to determine proper memory slice for it.
5038 return true;
5039 }
5040 }
5041 return false;
5042 }
5043
5044 bool LibraryCallKit::inline_unsafe_copyMemory() {
5045 if (callee()->is_static()) return false; // caller must have the capability!
5046 null_check_receiver(); // null-check receiver
5047 if (stopped()) return true;
5048
5049 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5050
5051 Node* src_base = argument(1); // type: oop
5052 Node* src_off = ConvL2X(argument(2)); // type: long
5053 Node* dst_base = argument(4); // type: oop
5054 Node* dst_off = ConvL2X(argument(5)); // type: long
5055 Node* size = ConvL2X(argument(7)); // type: long
5056
5057 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5058 "fieldOffset must be byte-scaled");
5059
5060 Node* src_addr = make_unsafe_address(src_base, src_off);
5061 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5062
5063 Node* thread = _gvn.transform(new ThreadLocalNode());
5064 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5065 BasicType doing_unsafe_access_bt = T_BYTE;
5066 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5067
5068 // update volatile field
5069 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5070
5071 int flags = RC_LEAF | RC_NO_FP;
5072
5073 const TypePtr* dst_type = TypePtr::BOTTOM;
5074
5075 // Adjust memory effects of the runtime call based on input values.
5076 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5077 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5078 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5079
5080 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5081 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5082 flags |= RC_NARROW_MEM; // narrow in memory
5083 }
5084 }
5085
5086 // Call it. Note that the length argument is not scaled.
5087 make_runtime_call(flags,
5088 OptoRuntime::fast_arraycopy_Type(),
5089 StubRoutines::unsafe_arraycopy(),
5090 "unsafe_arraycopy",
5091 dst_type,
5092 src_addr, dst_addr, size XTOP);
5093
5094 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5095
5096 return true;
5097 }
5098
5099 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5100 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5101 bool LibraryCallKit::inline_unsafe_setMemory() {
5102 if (callee()->is_static()) return false; // caller must have the capability!
5103 null_check_receiver(); // null-check receiver
5104 if (stopped()) return true;
5105
5106 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5107
5108 Node* dst_base = argument(1); // type: oop
5109 Node* dst_off = ConvL2X(argument(2)); // type: long
5110 Node* size = ConvL2X(argument(4)); // type: long
5111 Node* byte = argument(6); // type: byte
5112
5113 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5114 "fieldOffset must be byte-scaled");
5115
5116 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5117
5118 Node* thread = _gvn.transform(new ThreadLocalNode());
5119 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5120 BasicType doing_unsafe_access_bt = T_BYTE;
5121 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5122
5123 // update volatile field
5124 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5125
5126 int flags = RC_LEAF | RC_NO_FP;
5127
5128 const TypePtr* dst_type = TypePtr::BOTTOM;
5129
5130 // Adjust memory effects of the runtime call based on input values.
5131 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5132 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5133
5134 flags |= RC_NARROW_MEM; // narrow in memory
5135 }
5136
5137 // Call it. Note that the length argument is not scaled.
5138 make_runtime_call(flags,
5139 OptoRuntime::make_setmemory_Type(),
5140 StubRoutines::unsafe_setmemory(),
5141 "unsafe_setmemory",
5142 dst_type,
5143 dst_addr, size XTOP, byte);
5144
5145 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5146
5147 return true;
5148 }
5149
5150 #undef XTOP
5151
5152 //------------------------clone_coping-----------------------------------
5153 // Helper function for inline_native_clone.
5154 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5155 assert(obj_size != nullptr, "");
5156 Node* raw_obj = alloc_obj->in(1);
5157 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5158
5159 AllocateNode* alloc = nullptr;
5160 if (ReduceBulkZeroing &&
5161 // If we are implementing an array clone without knowing its source type
5162 // (can happen when compiling the array-guarded branch of a reflective
5163 // Object.clone() invocation), initialize the array within the allocation.
5164 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5165 // to a runtime clone call that assumes fully initialized source arrays.
5166 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5167 // We will be completely responsible for initializing this object -
5168 // mark Initialize node as complete.
5169 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5170 // The object was just allocated - there should be no any stores!
5171 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5172 // Mark as complete_with_arraycopy so that on AllocateNode
5173 // expansion, we know this AllocateNode is initialized by an array
5174 // copy and a StoreStore barrier exists after the array copy.
5175 alloc->initialization()->set_complete_with_arraycopy();
5176 }
5177
5178 Node* size = _gvn.transform(obj_size);
5179 access_clone(obj, alloc_obj, size, is_array);
5180
5181 // Do not let reads from the cloned object float above the arraycopy.
5182 if (alloc != nullptr) {
5183 // Do not let stores that initialize this object be reordered with
5184 // a subsequent store that would make this object accessible by
5185 // other threads.
5186 // Record what AllocateNode this StoreStore protects so that
5187 // escape analysis can go from the MemBarStoreStoreNode to the
5188 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5189 // based on the escape status of the AllocateNode.
5190 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5191 } else {
5192 insert_mem_bar(Op_MemBarCPUOrder);
5193 }
5194 }
5195
5196 //------------------------inline_native_clone----------------------------
5197 // protected native Object java.lang.Object.clone();
5198 //
5199 // Here are the simple edge cases:
5200 // null receiver => normal trap
5201 // virtual and clone was overridden => slow path to out-of-line clone
5202 // not cloneable or finalizer => slow path to out-of-line Object.clone
5203 //
5204 // The general case has two steps, allocation and copying.
5205 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5206 //
5207 // Copying also has two cases, oop arrays and everything else.
5208 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5209 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5210 //
5211 // These steps fold up nicely if and when the cloned object's klass
5212 // can be sharply typed as an object array, a type array, or an instance.
5213 //
5214 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5215 PhiNode* result_val;
5216
5217 // Set the reexecute bit for the interpreter to reexecute
5218 // the bytecode that invokes Object.clone if deoptimization happens.
5219 { PreserveReexecuteState preexecs(this);
5220 jvms()->set_should_reexecute(true);
5221
5222 Node* obj = null_check_receiver();
5223 if (stopped()) return true;
5224
5225 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5226
5227 // If we are going to clone an instance, we need its exact type to
5228 // know the number and types of fields to convert the clone to
5229 // loads/stores. Maybe a speculative type can help us.
5230 if (!obj_type->klass_is_exact() &&
5231 obj_type->speculative_type() != nullptr &&
5232 obj_type->speculative_type()->is_instance_klass()) {
5233 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5234 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5235 !spec_ik->has_injected_fields()) {
5236 if (!obj_type->isa_instptr() ||
5237 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5238 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5239 }
5240 }
5241 }
5242
5243 // Conservatively insert a memory barrier on all memory slices.
5244 // Do not let writes into the original float below the clone.
5245 insert_mem_bar(Op_MemBarCPUOrder);
5246
5247 // paths into result_reg:
5248 enum {
5249 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5250 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5251 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5252 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5253 PATH_LIMIT
5254 };
5255 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5256 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5257 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5258 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5259 record_for_igvn(result_reg);
5260
5261 Node* obj_klass = load_object_klass(obj);
5262 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr);
5263 if (array_ctl != nullptr) {
5264 // It's an array.
5265 PreserveJVMState pjvms(this);
5266 set_control(array_ctl);
5267 Node* obj_length = load_array_length(obj);
5268 Node* array_size = nullptr; // Size of the array without object alignment padding.
5269 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5270
5271 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5272 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5273 // If it is an oop array, it requires very special treatment,
5274 // because gc barriers are required when accessing the array.
5275 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5276 if (is_obja != nullptr) {
5277 PreserveJVMState pjvms2(this);
5278 set_control(is_obja);
5279 // Generate a direct call to the right arraycopy function(s).
5280 // Clones are always tightly coupled.
5281 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5282 ac->set_clone_oop_array();
5283 Node* n = _gvn.transform(ac);
5284 assert(n == ac, "cannot disappear");
5285 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5286
5287 result_reg->init_req(_objArray_path, control());
5288 result_val->init_req(_objArray_path, alloc_obj);
5289 result_i_o ->set_req(_objArray_path, i_o());
5290 result_mem ->set_req(_objArray_path, reset_memory());
5291 }
5292 }
5293 // Otherwise, there are no barriers to worry about.
5294 // (We can dispense with card marks if we know the allocation
5295 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5296 // causes the non-eden paths to take compensating steps to
5297 // simulate a fresh allocation, so that no further
5298 // card marks are required in compiled code to initialize
5299 // the object.)
5300
5301 if (!stopped()) {
5302 copy_to_clone(obj, alloc_obj, array_size, true);
5303
5304 // Present the results of the copy.
5305 result_reg->init_req(_array_path, control());
5306 result_val->init_req(_array_path, alloc_obj);
5307 result_i_o ->set_req(_array_path, i_o());
5308 result_mem ->set_req(_array_path, reset_memory());
5309 }
5310 }
5311
5312 // We only go to the instance fast case code if we pass a number of guards.
5313 // The paths which do not pass are accumulated in the slow_region.
5314 RegionNode* slow_region = new RegionNode(1);
5315 record_for_igvn(slow_region);
5316 if (!stopped()) {
5317 // It's an instance (we did array above). Make the slow-path tests.
5318 // If this is a virtual call, we generate a funny guard. We grab
5319 // the vtable entry corresponding to clone() from the target object.
5320 // If the target method which we are calling happens to be the
5321 // Object clone() method, we pass the guard. We do not need this
5322 // guard for non-virtual calls; the caller is known to be the native
5323 // Object clone().
5324 if (is_virtual) {
5325 generate_virtual_guard(obj_klass, slow_region);
5326 }
5327
5328 // The object must be easily cloneable and must not have a finalizer.
5329 // Both of these conditions may be checked in a single test.
5330 // We could optimize the test further, but we don't care.
5331 generate_misc_flags_guard(obj_klass,
5332 // Test both conditions:
5333 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5334 // Must be cloneable but not finalizer:
5335 KlassFlags::_misc_is_cloneable_fast,
5336 slow_region);
5337 }
5338
5339 if (!stopped()) {
5340 // It's an instance, and it passed the slow-path tests.
5341 PreserveJVMState pjvms(this);
5342 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5343 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5344 // is reexecuted if deoptimization occurs and there could be problems when merging
5345 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5346 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5347
5348 copy_to_clone(obj, alloc_obj, obj_size, false);
5349
5350 // Present the results of the slow call.
5351 result_reg->init_req(_instance_path, control());
5352 result_val->init_req(_instance_path, alloc_obj);
5353 result_i_o ->set_req(_instance_path, i_o());
5354 result_mem ->set_req(_instance_path, reset_memory());
5355 }
5356
5357 // Generate code for the slow case. We make a call to clone().
5358 set_control(_gvn.transform(slow_region));
5359 if (!stopped()) {
5360 PreserveJVMState pjvms(this);
5361 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5362 // We need to deoptimize on exception (see comment above)
5363 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5364 // this->control() comes from set_results_for_java_call
5365 result_reg->init_req(_slow_path, control());
5366 result_val->init_req(_slow_path, slow_result);
5367 result_i_o ->set_req(_slow_path, i_o());
5368 result_mem ->set_req(_slow_path, reset_memory());
5369 }
5370
5371 // Return the combined state.
5372 set_control( _gvn.transform(result_reg));
5373 set_i_o( _gvn.transform(result_i_o));
5374 set_all_memory( _gvn.transform(result_mem));
5375 } // original reexecute is set back here
5376
5377 set_result(_gvn.transform(result_val));
5378 return true;
5379 }
5380
5381 // If we have a tightly coupled allocation, the arraycopy may take care
5382 // of the array initialization. If one of the guards we insert between
5383 // the allocation and the arraycopy causes a deoptimization, an
5384 // uninitialized array will escape the compiled method. To prevent that
5385 // we set the JVM state for uncommon traps between the allocation and
5386 // the arraycopy to the state before the allocation so, in case of
5387 // deoptimization, we'll reexecute the allocation and the
5388 // initialization.
5389 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5390 if (alloc != nullptr) {
5391 ciMethod* trap_method = alloc->jvms()->method();
5392 int trap_bci = alloc->jvms()->bci();
5393
5394 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5395 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5396 // Make sure there's no store between the allocation and the
5397 // arraycopy otherwise visible side effects could be rexecuted
5398 // in case of deoptimization and cause incorrect execution.
5399 bool no_interfering_store = true;
5400 Node* mem = alloc->in(TypeFunc::Memory);
5401 if (mem->is_MergeMem()) {
5402 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5403 Node* n = mms.memory();
5404 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5405 assert(n->is_Store(), "what else?");
5406 no_interfering_store = false;
5407 break;
5408 }
5409 }
5410 } else {
5411 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5412 Node* n = mms.memory();
5413 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5414 assert(n->is_Store(), "what else?");
5415 no_interfering_store = false;
5416 break;
5417 }
5418 }
5419 }
5420
5421 if (no_interfering_store) {
5422 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5423
5424 JVMState* saved_jvms = jvms();
5425 saved_reexecute_sp = _reexecute_sp;
5426
5427 set_jvms(sfpt->jvms());
5428 _reexecute_sp = jvms()->sp();
5429
5430 return saved_jvms;
5431 }
5432 }
5433 }
5434 return nullptr;
5435 }
5436
5437 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5438 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5439 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5440 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5441 uint size = alloc->req();
5442 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5443 old_jvms->set_map(sfpt);
5444 for (uint i = 0; i < size; i++) {
5445 sfpt->init_req(i, alloc->in(i));
5446 }
5447 // re-push array length for deoptimization
5448 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5449 old_jvms->set_sp(old_jvms->sp()+1);
5450 old_jvms->set_monoff(old_jvms->monoff()+1);
5451 old_jvms->set_scloff(old_jvms->scloff()+1);
5452 old_jvms->set_endoff(old_jvms->endoff()+1);
5453 old_jvms->set_should_reexecute(true);
5454
5455 sfpt->set_i_o(map()->i_o());
5456 sfpt->set_memory(map()->memory());
5457 sfpt->set_control(map()->control());
5458 return sfpt;
5459 }
5460
5461 // In case of a deoptimization, we restart execution at the
5462 // allocation, allocating a new array. We would leave an uninitialized
5463 // array in the heap that GCs wouldn't expect. Move the allocation
5464 // after the traps so we don't allocate the array if we
5465 // deoptimize. This is possible because tightly_coupled_allocation()
5466 // guarantees there's no observer of the allocated array at this point
5467 // and the control flow is simple enough.
5468 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5469 int saved_reexecute_sp, uint new_idx) {
5470 if (saved_jvms_before_guards != nullptr && !stopped()) {
5471 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5472
5473 assert(alloc != nullptr, "only with a tightly coupled allocation");
5474 // restore JVM state to the state at the arraycopy
5475 saved_jvms_before_guards->map()->set_control(map()->control());
5476 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5477 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5478 // If we've improved the types of some nodes (null check) while
5479 // emitting the guards, propagate them to the current state
5480 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5481 set_jvms(saved_jvms_before_guards);
5482 _reexecute_sp = saved_reexecute_sp;
5483
5484 // Remove the allocation from above the guards
5485 CallProjections callprojs;
5486 alloc->extract_projections(&callprojs, true);
5487 InitializeNode* init = alloc->initialization();
5488 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5489 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5490 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5491
5492 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5493 // the allocation (i.e. is only valid if the allocation succeeds):
5494 // 1) replace CastIINode with AllocateArrayNode's length here
5495 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5496 //
5497 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5498 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5499 Node* init_control = init->proj_out(TypeFunc::Control);
5500 Node* alloc_length = alloc->Ideal_length();
5501 #ifdef ASSERT
5502 Node* prev_cast = nullptr;
5503 #endif
5504 for (uint i = 0; i < init_control->outcnt(); i++) {
5505 Node* init_out = init_control->raw_out(i);
5506 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5507 #ifdef ASSERT
5508 if (prev_cast == nullptr) {
5509 prev_cast = init_out;
5510 } else {
5511 if (prev_cast->cmp(*init_out) == false) {
5512 prev_cast->dump();
5513 init_out->dump();
5514 assert(false, "not equal CastIINode");
5515 }
5516 }
5517 #endif
5518 C->gvn_replace_by(init_out, alloc_length);
5519 }
5520 }
5521 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5522
5523 // move the allocation here (after the guards)
5524 _gvn.hash_delete(alloc);
5525 alloc->set_req(TypeFunc::Control, control());
5526 alloc->set_req(TypeFunc::I_O, i_o());
5527 Node *mem = reset_memory();
5528 set_all_memory(mem);
5529 alloc->set_req(TypeFunc::Memory, mem);
5530 set_control(init->proj_out_or_null(TypeFunc::Control));
5531 set_i_o(callprojs.fallthrough_ioproj);
5532
5533 // Update memory as done in GraphKit::set_output_for_allocation()
5534 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5535 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5536 if (ary_type->isa_aryptr() && length_type != nullptr) {
5537 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5538 }
5539 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5540 int elemidx = C->get_alias_index(telemref);
5541 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5542 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5543
5544 Node* allocx = _gvn.transform(alloc);
5545 assert(allocx == alloc, "where has the allocation gone?");
5546 assert(dest->is_CheckCastPP(), "not an allocation result?");
5547
5548 _gvn.hash_delete(dest);
5549 dest->set_req(0, control());
5550 Node* destx = _gvn.transform(dest);
5551 assert(destx == dest, "where has the allocation result gone?");
5552
5553 array_ideal_length(alloc, ary_type, true);
5554 }
5555 }
5556
5557 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5558 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5559 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5560 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5561 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5562 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5563 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5564 JVMState* saved_jvms_before_guards) {
5565 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5566 // There is at least one unrelated uncommon trap which needs to be replaced.
5567 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5568
5569 JVMState* saved_jvms = jvms();
5570 const int saved_reexecute_sp = _reexecute_sp;
5571 set_jvms(sfpt->jvms());
5572 _reexecute_sp = jvms()->sp();
5573
5574 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5575
5576 // Restore state
5577 set_jvms(saved_jvms);
5578 _reexecute_sp = saved_reexecute_sp;
5579 }
5580 }
5581
5582 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5583 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5584 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5585 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5586 while (if_proj->is_IfProj()) {
5587 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5588 if (uncommon_trap != nullptr) {
5589 create_new_uncommon_trap(uncommon_trap);
5590 }
5591 assert(if_proj->in(0)->is_If(), "must be If");
5592 if_proj = if_proj->in(0)->in(0);
5593 }
5594 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5595 "must have reached control projection of init node");
5596 }
5597
5598 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5599 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5600 assert(trap_request != 0, "no valid UCT trap request");
5601 PreserveJVMState pjvms(this);
5602 set_control(uncommon_trap_call->in(0));
5603 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5604 Deoptimization::trap_request_action(trap_request));
5605 assert(stopped(), "Should be stopped");
5606 _gvn.hash_delete(uncommon_trap_call);
5607 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5608 }
5609
5610 // Common checks for array sorting intrinsics arguments.
5611 // Returns `true` if checks passed.
5612 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
5613 // check address of the class
5614 if (elementType == nullptr || elementType->is_top()) {
5615 return false; // dead path
5616 }
5617 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5618 if (elem_klass == nullptr) {
5619 return false; // dead path
5620 }
5621 // java_mirror_type() returns non-null for compile-time Class constants only
5622 ciType* elem_type = elem_klass->java_mirror_type();
5623 if (elem_type == nullptr) {
5624 return false;
5625 }
5626 bt = elem_type->basic_type();
5627 // Disable the intrinsic if the CPU does not support SIMD sort
5628 if (!Matcher::supports_simd_sort(bt)) {
5629 return false;
5630 }
5631 // check address of the array
5632 if (obj == nullptr || obj->is_top()) {
5633 return false; // dead path
5634 }
5635 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5636 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
5637 return false; // failed input validation
5638 }
5639 return true;
5640 }
5641
5642 //------------------------------inline_array_partition-----------------------
5643 bool LibraryCallKit::inline_array_partition() {
5644 address stubAddr = StubRoutines::select_array_partition_function();
5645 if (stubAddr == nullptr) {
5646 return false; // Intrinsic's stub is not implemented on this platform
5647 }
5648 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
5649
5650 // no receiver because it is a static method
5651 Node* elementType = argument(0);
5652 Node* obj = argument(1);
5653 Node* offset = argument(2); // long
5654 Node* fromIndex = argument(4);
5655 Node* toIndex = argument(5);
5656 Node* indexPivot1 = argument(6);
5657 Node* indexPivot2 = argument(7);
5658 // PartitionOperation: argument(8) is ignored
5659
5660 Node* pivotIndices = nullptr;
5661 BasicType bt = T_ILLEGAL;
5662
5663 if (!check_array_sort_arguments(elementType, obj, bt)) {
5664 return false;
5665 }
5666 null_check(obj);
5667 // If obj is dead, only null-path is taken.
5668 if (stopped()) {
5669 return true;
5670 }
5671 // Set the original stack and the reexecute bit for the interpreter to reexecute
5672 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
5673 { PreserveReexecuteState preexecs(this);
5674 jvms()->set_should_reexecute(true);
5675
5676 Node* obj_adr = make_unsafe_address(obj, offset);
5677
5678 // create the pivotIndices array of type int and size = 2
5679 Node* size = intcon(2);
5680 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
5681 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
5682 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
5683 guarantee(alloc != nullptr, "created above");
5684 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
5685
5686 // pass the basic type enum to the stub
5687 Node* elemType = intcon(bt);
5688
5689 // Call the stub
5690 const char *stubName = "array_partition_stub";
5691 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
5692 stubAddr, stubName, TypePtr::BOTTOM,
5693 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
5694 indexPivot1, indexPivot2);
5695
5696 } // original reexecute is set back here
5697
5698 if (!stopped()) {
5699 set_result(pivotIndices);
5700 }
5701
5702 return true;
5703 }
5704
5705
5706 //------------------------------inline_array_sort-----------------------
5707 bool LibraryCallKit::inline_array_sort() {
5708 address stubAddr = StubRoutines::select_arraysort_function();
5709 if (stubAddr == nullptr) {
5710 return false; // Intrinsic's stub is not implemented on this platform
5711 }
5712 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
5713
5714 // no receiver because it is a static method
5715 Node* elementType = argument(0);
5716 Node* obj = argument(1);
5717 Node* offset = argument(2); // long
5718 Node* fromIndex = argument(4);
5719 Node* toIndex = argument(5);
5720 // SortOperation: argument(6) is ignored
5721
5722 BasicType bt = T_ILLEGAL;
5723
5724 if (!check_array_sort_arguments(elementType, obj, bt)) {
5725 return false;
5726 }
5727 null_check(obj);
5728 // If obj is dead, only null-path is taken.
5729 if (stopped()) {
5730 return true;
5731 }
5732 Node* obj_adr = make_unsafe_address(obj, offset);
5733
5734 // pass the basic type enum to the stub
5735 Node* elemType = intcon(bt);
5736
5737 // Call the stub.
5738 const char *stubName = "arraysort_stub";
5739 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
5740 stubAddr, stubName, TypePtr::BOTTOM,
5741 obj_adr, elemType, fromIndex, toIndex);
5742
5743 return true;
5744 }
5745
5746
5747 //------------------------------inline_arraycopy-----------------------
5748 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5749 // Object dest, int destPos,
5750 // int length);
5751 bool LibraryCallKit::inline_arraycopy() {
5752 // Get the arguments.
5753 Node* src = argument(0); // type: oop
5754 Node* src_offset = argument(1); // type: int
5755 Node* dest = argument(2); // type: oop
5756 Node* dest_offset = argument(3); // type: int
5757 Node* length = argument(4); // type: int
5758
5759 uint new_idx = C->unique();
5760
5761 // Check for allocation before we add nodes that would confuse
5762 // tightly_coupled_allocation()
5763 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5764
5765 int saved_reexecute_sp = -1;
5766 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5767 // See arraycopy_restore_alloc_state() comment
5768 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5769 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5770 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5771 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5772
5773 // The following tests must be performed
5774 // (1) src and dest are arrays.
5775 // (2) src and dest arrays must have elements of the same BasicType
5776 // (3) src and dest must not be null.
5777 // (4) src_offset must not be negative.
5778 // (5) dest_offset must not be negative.
5779 // (6) length must not be negative.
5780 // (7) src_offset + length must not exceed length of src.
5781 // (8) dest_offset + length must not exceed length of dest.
5782 // (9) each element of an oop array must be assignable
5783
5784 // (3) src and dest must not be null.
5785 // always do this here because we need the JVM state for uncommon traps
5786 Node* null_ctl = top();
5787 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5788 assert(null_ctl->is_top(), "no null control here");
5789 dest = null_check(dest, T_ARRAY);
5790
5791 if (!can_emit_guards) {
5792 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5793 // guards but the arraycopy node could still take advantage of a
5794 // tightly allocated allocation. tightly_coupled_allocation() is
5795 // called again to make sure it takes the null check above into
5796 // account: the null check is mandatory and if it caused an
5797 // uncommon trap to be emitted then the allocation can't be
5798 // considered tightly coupled in this context.
5799 alloc = tightly_coupled_allocation(dest);
5800 }
5801
5802 bool validated = false;
5803
5804 const Type* src_type = _gvn.type(src);
5805 const Type* dest_type = _gvn.type(dest);
5806 const TypeAryPtr* top_src = src_type->isa_aryptr();
5807 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5808
5809 // Do we have the type of src?
5810 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5811 // Do we have the type of dest?
5812 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5813 // Is the type for src from speculation?
5814 bool src_spec = false;
5815 // Is the type for dest from speculation?
5816 bool dest_spec = false;
5817
5818 if ((!has_src || !has_dest) && can_emit_guards) {
5819 // We don't have sufficient type information, let's see if
5820 // speculative types can help. We need to have types for both src
5821 // and dest so that it pays off.
5822
5823 // Do we already have or could we have type information for src
5824 bool could_have_src = has_src;
5825 // Do we already have or could we have type information for dest
5826 bool could_have_dest = has_dest;
5827
5828 ciKlass* src_k = nullptr;
5829 if (!has_src) {
5830 src_k = src_type->speculative_type_not_null();
5831 if (src_k != nullptr && src_k->is_array_klass()) {
5832 could_have_src = true;
5833 }
5834 }
5835
5836 ciKlass* dest_k = nullptr;
5837 if (!has_dest) {
5838 dest_k = dest_type->speculative_type_not_null();
5839 if (dest_k != nullptr && dest_k->is_array_klass()) {
5840 could_have_dest = true;
5841 }
5842 }
5843
5844 if (could_have_src && could_have_dest) {
5845 // This is going to pay off so emit the required guards
5846 if (!has_src) {
5847 src = maybe_cast_profiled_obj(src, src_k, true);
5848 src_type = _gvn.type(src);
5849 top_src = src_type->isa_aryptr();
5850 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5851 src_spec = true;
5852 }
5853 if (!has_dest) {
5854 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5855 dest_type = _gvn.type(dest);
5856 top_dest = dest_type->isa_aryptr();
5857 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5858 dest_spec = true;
5859 }
5860 }
5861 }
5862
5863 if (has_src && has_dest && can_emit_guards) {
5864 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5865 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5866 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5867 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5868
5869 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5870 // If both arrays are object arrays then having the exact types
5871 // for both will remove the need for a subtype check at runtime
5872 // before the call and may make it possible to pick a faster copy
5873 // routine (without a subtype check on every element)
5874 // Do we have the exact type of src?
5875 bool could_have_src = src_spec;
5876 // Do we have the exact type of dest?
5877 bool could_have_dest = dest_spec;
5878 ciKlass* src_k = nullptr;
5879 ciKlass* dest_k = nullptr;
5880 if (!src_spec) {
5881 src_k = src_type->speculative_type_not_null();
5882 if (src_k != nullptr && src_k->is_array_klass()) {
5883 could_have_src = true;
5884 }
5885 }
5886 if (!dest_spec) {
5887 dest_k = dest_type->speculative_type_not_null();
5888 if (dest_k != nullptr && dest_k->is_array_klass()) {
5889 could_have_dest = true;
5890 }
5891 }
5892 if (could_have_src && could_have_dest) {
5893 // If we can have both exact types, emit the missing guards
5894 if (could_have_src && !src_spec) {
5895 src = maybe_cast_profiled_obj(src, src_k, true);
5896 }
5897 if (could_have_dest && !dest_spec) {
5898 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5899 }
5900 }
5901 }
5902 }
5903
5904 ciMethod* trap_method = method();
5905 int trap_bci = bci();
5906 if (saved_jvms_before_guards != nullptr) {
5907 trap_method = alloc->jvms()->method();
5908 trap_bci = alloc->jvms()->bci();
5909 }
5910
5911 bool negative_length_guard_generated = false;
5912
5913 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5914 can_emit_guards &&
5915 !src->is_top() && !dest->is_top()) {
5916 // validate arguments: enables transformation the ArrayCopyNode
5917 validated = true;
5918
5919 RegionNode* slow_region = new RegionNode(1);
5920 record_for_igvn(slow_region);
5921
5922 // (1) src and dest are arrays.
5923 generate_non_array_guard(load_object_klass(src), slow_region);
5924 generate_non_array_guard(load_object_klass(dest), slow_region);
5925
5926 // (2) src and dest arrays must have elements of the same BasicType
5927 // done at macro expansion or at Ideal transformation time
5928
5929 // (4) src_offset must not be negative.
5930 generate_negative_guard(src_offset, slow_region);
5931
5932 // (5) dest_offset must not be negative.
5933 generate_negative_guard(dest_offset, slow_region);
5934
5935 // (7) src_offset + length must not exceed length of src.
5936 generate_limit_guard(src_offset, length,
5937 load_array_length(src),
5938 slow_region);
5939
5940 // (8) dest_offset + length must not exceed length of dest.
5941 generate_limit_guard(dest_offset, length,
5942 load_array_length(dest),
5943 slow_region);
5944
5945 // (6) length must not be negative.
5946 // This is also checked in generate_arraycopy() during macro expansion, but
5947 // we also have to check it here for the case where the ArrayCopyNode will
5948 // be eliminated by Escape Analysis.
5949 if (EliminateAllocations) {
5950 generate_negative_guard(length, slow_region);
5951 negative_length_guard_generated = true;
5952 }
5953
5954 // (9) each element of an oop array must be assignable
5955 Node* dest_klass = load_object_klass(dest);
5956 if (src != dest) {
5957 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
5958
5959 if (not_subtype_ctrl != top()) {
5960 PreserveJVMState pjvms(this);
5961 set_control(not_subtype_ctrl);
5962 uncommon_trap(Deoptimization::Reason_intrinsic,
5963 Deoptimization::Action_make_not_entrant);
5964 assert(stopped(), "Should be stopped");
5965 }
5966 }
5967 {
5968 PreserveJVMState pjvms(this);
5969 set_control(_gvn.transform(slow_region));
5970 uncommon_trap(Deoptimization::Reason_intrinsic,
5971 Deoptimization::Action_make_not_entrant);
5972 assert(stopped(), "Should be stopped");
5973 }
5974
5975 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5976 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
5977 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5978 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
5979 }
5980
5981 if (stopped()) {
5982 return true;
5983 }
5984
5985 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
5986 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5987 // so the compiler has a chance to eliminate them: during macro expansion,
5988 // we have to set their control (CastPP nodes are eliminated).
5989 load_object_klass(src), load_object_klass(dest),
5990 load_array_length(src), load_array_length(dest));
5991
5992 ac->set_arraycopy(validated);
5993
5994 Node* n = _gvn.transform(ac);
5995 if (n == ac) {
5996 ac->connect_outputs(this);
5997 } else {
5998 assert(validated, "shouldn't transform if all arguments not validated");
5999 set_all_memory(n);
6000 }
6001 clear_upper_avx();
6002
6003
6004 return true;
6005 }
6006
6007
6008 // Helper function which determines if an arraycopy immediately follows
6009 // an allocation, with no intervening tests or other escapes for the object.
6010 AllocateArrayNode*
6011 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6012 if (stopped()) return nullptr; // no fast path
6013 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6014
6015 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6016 if (alloc == nullptr) return nullptr;
6017
6018 Node* rawmem = memory(Compile::AliasIdxRaw);
6019 // Is the allocation's memory state untouched?
6020 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6021 // Bail out if there have been raw-memory effects since the allocation.
6022 // (Example: There might have been a call or safepoint.)
6023 return nullptr;
6024 }
6025 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6026 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6027 return nullptr;
6028 }
6029
6030 // There must be no unexpected observers of this allocation.
6031 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6032 Node* obs = ptr->fast_out(i);
6033 if (obs != this->map()) {
6034 return nullptr;
6035 }
6036 }
6037
6038 // This arraycopy must unconditionally follow the allocation of the ptr.
6039 Node* alloc_ctl = ptr->in(0);
6040 Node* ctl = control();
6041 while (ctl != alloc_ctl) {
6042 // There may be guards which feed into the slow_region.
6043 // Any other control flow means that we might not get a chance
6044 // to finish initializing the allocated object.
6045 // Various low-level checks bottom out in uncommon traps. These
6046 // are considered safe since we've already checked above that
6047 // there is no unexpected observer of this allocation.
6048 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6049 assert(ctl->in(0)->is_If(), "must be If");
6050 ctl = ctl->in(0)->in(0);
6051 } else {
6052 return nullptr;
6053 }
6054 }
6055
6056 // If we get this far, we have an allocation which immediately
6057 // precedes the arraycopy, and we can take over zeroing the new object.
6058 // The arraycopy will finish the initialization, and provide
6059 // a new control state to which we will anchor the destination pointer.
6060
6061 return alloc;
6062 }
6063
6064 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6065 if (node->is_IfProj()) {
6066 Node* other_proj = node->as_IfProj()->other_if_proj();
6067 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6068 Node* obs = other_proj->fast_out(j);
6069 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6070 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6071 return obs->as_CallStaticJava();
6072 }
6073 }
6074 }
6075 return nullptr;
6076 }
6077
6078 //-------------inline_encodeISOArray-----------------------------------
6079 // encode char[] to byte[] in ISO_8859_1 or ASCII
6080 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6081 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6082 // no receiver since it is static method
6083 Node *src = argument(0);
6084 Node *src_offset = argument(1);
6085 Node *dst = argument(2);
6086 Node *dst_offset = argument(3);
6087 Node *length = argument(4);
6088
6089 src = must_be_not_null(src, true);
6090 dst = must_be_not_null(dst, true);
6091
6092 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6093 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6094 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6095 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6096 // failed array check
6097 return false;
6098 }
6099
6100 // Figure out the size and type of the elements we will be copying.
6101 BasicType src_elem = src_type->elem()->array_element_basic_type();
6102 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6103 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6104 return false;
6105 }
6106
6107 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6108 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6109 // 'src_start' points to src array + scaled offset
6110 // 'dst_start' points to dst array + scaled offset
6111
6112 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6113 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6114 enc = _gvn.transform(enc);
6115 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6116 set_memory(res_mem, mtype);
6117 set_result(enc);
6118 clear_upper_avx();
6119
6120 return true;
6121 }
6122
6123 //-------------inline_multiplyToLen-----------------------------------
6124 bool LibraryCallKit::inline_multiplyToLen() {
6125 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6126
6127 address stubAddr = StubRoutines::multiplyToLen();
6128 if (stubAddr == nullptr) {
6129 return false; // Intrinsic's stub is not implemented on this platform
6130 }
6131 const char* stubName = "multiplyToLen";
6132
6133 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6134
6135 // no receiver because it is a static method
6136 Node* x = argument(0);
6137 Node* xlen = argument(1);
6138 Node* y = argument(2);
6139 Node* ylen = argument(3);
6140 Node* z = argument(4);
6141
6142 x = must_be_not_null(x, true);
6143 y = must_be_not_null(y, true);
6144
6145 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6146 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6147 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6148 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6149 // failed array check
6150 return false;
6151 }
6152
6153 BasicType x_elem = x_type->elem()->array_element_basic_type();
6154 BasicType y_elem = y_type->elem()->array_element_basic_type();
6155 if (x_elem != T_INT || y_elem != T_INT) {
6156 return false;
6157 }
6158
6159 Node* x_start = array_element_address(x, intcon(0), x_elem);
6160 Node* y_start = array_element_address(y, intcon(0), y_elem);
6161 // 'x_start' points to x array + scaled xlen
6162 // 'y_start' points to y array + scaled ylen
6163
6164 Node* z_start = array_element_address(z, intcon(0), T_INT);
6165
6166 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6167 OptoRuntime::multiplyToLen_Type(),
6168 stubAddr, stubName, TypePtr::BOTTOM,
6169 x_start, xlen, y_start, ylen, z_start);
6170
6171 C->set_has_split_ifs(true); // Has chance for split-if optimization
6172 set_result(z);
6173 return true;
6174 }
6175
6176 //-------------inline_squareToLen------------------------------------
6177 bool LibraryCallKit::inline_squareToLen() {
6178 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6179
6180 address stubAddr = StubRoutines::squareToLen();
6181 if (stubAddr == nullptr) {
6182 return false; // Intrinsic's stub is not implemented on this platform
6183 }
6184 const char* stubName = "squareToLen";
6185
6186 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6187
6188 Node* x = argument(0);
6189 Node* len = argument(1);
6190 Node* z = argument(2);
6191 Node* zlen = argument(3);
6192
6193 x = must_be_not_null(x, true);
6194 z = must_be_not_null(z, true);
6195
6196 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6197 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6198 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6199 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6200 // failed array check
6201 return false;
6202 }
6203
6204 BasicType x_elem = x_type->elem()->array_element_basic_type();
6205 BasicType z_elem = z_type->elem()->array_element_basic_type();
6206 if (x_elem != T_INT || z_elem != T_INT) {
6207 return false;
6208 }
6209
6210
6211 Node* x_start = array_element_address(x, intcon(0), x_elem);
6212 Node* z_start = array_element_address(z, intcon(0), z_elem);
6213
6214 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6215 OptoRuntime::squareToLen_Type(),
6216 stubAddr, stubName, TypePtr::BOTTOM,
6217 x_start, len, z_start, zlen);
6218
6219 set_result(z);
6220 return true;
6221 }
6222
6223 //-------------inline_mulAdd------------------------------------------
6224 bool LibraryCallKit::inline_mulAdd() {
6225 assert(UseMulAddIntrinsic, "not implemented on this platform");
6226
6227 address stubAddr = StubRoutines::mulAdd();
6228 if (stubAddr == nullptr) {
6229 return false; // Intrinsic's stub is not implemented on this platform
6230 }
6231 const char* stubName = "mulAdd";
6232
6233 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6234
6235 Node* out = argument(0);
6236 Node* in = argument(1);
6237 Node* offset = argument(2);
6238 Node* len = argument(3);
6239 Node* k = argument(4);
6240
6241 in = must_be_not_null(in, true);
6242 out = must_be_not_null(out, true);
6243
6244 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6245 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6246 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6247 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6248 // failed array check
6249 return false;
6250 }
6251
6252 BasicType out_elem = out_type->elem()->array_element_basic_type();
6253 BasicType in_elem = in_type->elem()->array_element_basic_type();
6254 if (out_elem != T_INT || in_elem != T_INT) {
6255 return false;
6256 }
6257
6258 Node* outlen = load_array_length(out);
6259 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6260 Node* out_start = array_element_address(out, intcon(0), out_elem);
6261 Node* in_start = array_element_address(in, intcon(0), in_elem);
6262
6263 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6264 OptoRuntime::mulAdd_Type(),
6265 stubAddr, stubName, TypePtr::BOTTOM,
6266 out_start,in_start, new_offset, len, k);
6267 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6268 set_result(result);
6269 return true;
6270 }
6271
6272 //-------------inline_montgomeryMultiply-----------------------------------
6273 bool LibraryCallKit::inline_montgomeryMultiply() {
6274 address stubAddr = StubRoutines::montgomeryMultiply();
6275 if (stubAddr == nullptr) {
6276 return false; // Intrinsic's stub is not implemented on this platform
6277 }
6278
6279 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6280 const char* stubName = "montgomery_multiply";
6281
6282 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6283
6284 Node* a = argument(0);
6285 Node* b = argument(1);
6286 Node* n = argument(2);
6287 Node* len = argument(3);
6288 Node* inv = argument(4);
6289 Node* m = argument(6);
6290
6291 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6292 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6293 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6294 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6295 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6296 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6297 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6298 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6299 // failed array check
6300 return false;
6301 }
6302
6303 BasicType a_elem = a_type->elem()->array_element_basic_type();
6304 BasicType b_elem = b_type->elem()->array_element_basic_type();
6305 BasicType n_elem = n_type->elem()->array_element_basic_type();
6306 BasicType m_elem = m_type->elem()->array_element_basic_type();
6307 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6308 return false;
6309 }
6310
6311 // Make the call
6312 {
6313 Node* a_start = array_element_address(a, intcon(0), a_elem);
6314 Node* b_start = array_element_address(b, intcon(0), b_elem);
6315 Node* n_start = array_element_address(n, intcon(0), n_elem);
6316 Node* m_start = array_element_address(m, intcon(0), m_elem);
6317
6318 Node* call = make_runtime_call(RC_LEAF,
6319 OptoRuntime::montgomeryMultiply_Type(),
6320 stubAddr, stubName, TypePtr::BOTTOM,
6321 a_start, b_start, n_start, len, inv, top(),
6322 m_start);
6323 set_result(m);
6324 }
6325
6326 return true;
6327 }
6328
6329 bool LibraryCallKit::inline_montgomerySquare() {
6330 address stubAddr = StubRoutines::montgomerySquare();
6331 if (stubAddr == nullptr) {
6332 return false; // Intrinsic's stub is not implemented on this platform
6333 }
6334
6335 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6336 const char* stubName = "montgomery_square";
6337
6338 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6339
6340 Node* a = argument(0);
6341 Node* n = argument(1);
6342 Node* len = argument(2);
6343 Node* inv = argument(3);
6344 Node* m = argument(5);
6345
6346 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6347 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6348 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6349 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6350 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6351 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6352 // failed array check
6353 return false;
6354 }
6355
6356 BasicType a_elem = a_type->elem()->array_element_basic_type();
6357 BasicType n_elem = n_type->elem()->array_element_basic_type();
6358 BasicType m_elem = m_type->elem()->array_element_basic_type();
6359 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6360 return false;
6361 }
6362
6363 // Make the call
6364 {
6365 Node* a_start = array_element_address(a, intcon(0), a_elem);
6366 Node* n_start = array_element_address(n, intcon(0), n_elem);
6367 Node* m_start = array_element_address(m, intcon(0), m_elem);
6368
6369 Node* call = make_runtime_call(RC_LEAF,
6370 OptoRuntime::montgomerySquare_Type(),
6371 stubAddr, stubName, TypePtr::BOTTOM,
6372 a_start, n_start, len, inv, top(),
6373 m_start);
6374 set_result(m);
6375 }
6376
6377 return true;
6378 }
6379
6380 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6381 address stubAddr = nullptr;
6382 const char* stubName = nullptr;
6383
6384 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6385 if (stubAddr == nullptr) {
6386 return false; // Intrinsic's stub is not implemented on this platform
6387 }
6388
6389 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6390
6391 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6392
6393 Node* newArr = argument(0);
6394 Node* oldArr = argument(1);
6395 Node* newIdx = argument(2);
6396 Node* shiftCount = argument(3);
6397 Node* numIter = argument(4);
6398
6399 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6400 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6401 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6402 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6403 return false;
6404 }
6405
6406 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6407 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6408 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6409 return false;
6410 }
6411
6412 // Make the call
6413 {
6414 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6415 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6416
6417 Node* call = make_runtime_call(RC_LEAF,
6418 OptoRuntime::bigIntegerShift_Type(),
6419 stubAddr,
6420 stubName,
6421 TypePtr::BOTTOM,
6422 newArr_start,
6423 oldArr_start,
6424 newIdx,
6425 shiftCount,
6426 numIter);
6427 }
6428
6429 return true;
6430 }
6431
6432 //-------------inline_vectorizedMismatch------------------------------
6433 bool LibraryCallKit::inline_vectorizedMismatch() {
6434 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6435
6436 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6437 Node* obja = argument(0); // Object
6438 Node* aoffset = argument(1); // long
6439 Node* objb = argument(3); // Object
6440 Node* boffset = argument(4); // long
6441 Node* length = argument(6); // int
6442 Node* scale = argument(7); // int
6443
6444 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6445 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6446 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6447 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6448 scale == top()) {
6449 return false; // failed input validation
6450 }
6451
6452 Node* obja_adr = make_unsafe_address(obja, aoffset);
6453 Node* objb_adr = make_unsafe_address(objb, boffset);
6454
6455 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6456 //
6457 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6458 // if (length <= inline_limit) {
6459 // inline_path:
6460 // vmask = VectorMaskGen length
6461 // vload1 = LoadVectorMasked obja, vmask
6462 // vload2 = LoadVectorMasked objb, vmask
6463 // result1 = VectorCmpMasked vload1, vload2, vmask
6464 // } else {
6465 // call_stub_path:
6466 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6467 // }
6468 // exit_block:
6469 // return Phi(result1, result2);
6470 //
6471 enum { inline_path = 1, // input is small enough to process it all at once
6472 stub_path = 2, // input is too large; call into the VM
6473 PATH_LIMIT = 3
6474 };
6475
6476 Node* exit_block = new RegionNode(PATH_LIMIT);
6477 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6478 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6479
6480 Node* call_stub_path = control();
6481
6482 BasicType elem_bt = T_ILLEGAL;
6483
6484 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6485 if (scale_t->is_con()) {
6486 switch (scale_t->get_con()) {
6487 case 0: elem_bt = T_BYTE; break;
6488 case 1: elem_bt = T_SHORT; break;
6489 case 2: elem_bt = T_INT; break;
6490 case 3: elem_bt = T_LONG; break;
6491
6492 default: elem_bt = T_ILLEGAL; break; // not supported
6493 }
6494 }
6495
6496 int inline_limit = 0;
6497 bool do_partial_inline = false;
6498
6499 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6500 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6501 do_partial_inline = inline_limit >= 16;
6502 }
6503
6504 if (do_partial_inline) {
6505 assert(elem_bt != T_ILLEGAL, "sanity");
6506
6507 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6508 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6509 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6510
6511 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6512 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6513 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6514
6515 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6516
6517 if (!stopped()) {
6518 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6519
6520 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6521 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6522 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6523 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6524
6525 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6526 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6527 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6528 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6529
6530 exit_block->init_req(inline_path, control());
6531 memory_phi->init_req(inline_path, map()->memory());
6532 result_phi->init_req(inline_path, result);
6533
6534 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6535 clear_upper_avx();
6536 }
6537 }
6538 }
6539
6540 if (call_stub_path != nullptr) {
6541 set_control(call_stub_path);
6542
6543 Node* call = make_runtime_call(RC_LEAF,
6544 OptoRuntime::vectorizedMismatch_Type(),
6545 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6546 obja_adr, objb_adr, length, scale);
6547
6548 exit_block->init_req(stub_path, control());
6549 memory_phi->init_req(stub_path, map()->memory());
6550 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6551 }
6552
6553 exit_block = _gvn.transform(exit_block);
6554 memory_phi = _gvn.transform(memory_phi);
6555 result_phi = _gvn.transform(result_phi);
6556
6557 set_control(exit_block);
6558 set_all_memory(memory_phi);
6559 set_result(result_phi);
6560
6561 return true;
6562 }
6563
6564 //------------------------------inline_vectorizedHashcode----------------------------
6565 bool LibraryCallKit::inline_vectorizedHashCode() {
6566 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6567
6568 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6569 Node* array = argument(0);
6570 Node* offset = argument(1);
6571 Node* length = argument(2);
6572 Node* initialValue = argument(3);
6573 Node* basic_type = argument(4);
6574
6575 if (basic_type == top()) {
6576 return false; // failed input validation
6577 }
6578
6579 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6580 if (!basic_type_t->is_con()) {
6581 return false; // Only intrinsify if mode argument is constant
6582 }
6583
6584 array = must_be_not_null(array, true);
6585
6586 BasicType bt = (BasicType)basic_type_t->get_con();
6587
6588 // Resolve address of first element
6589 Node* array_start = array_element_address(array, offset, bt);
6590
6591 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6592 array_start, length, initialValue, basic_type)));
6593 clear_upper_avx();
6594
6595 return true;
6596 }
6597
6598 /**
6599 * Calculate CRC32 for byte.
6600 * int java.util.zip.CRC32.update(int crc, int b)
6601 */
6602 bool LibraryCallKit::inline_updateCRC32() {
6603 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6604 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6605 // no receiver since it is static method
6606 Node* crc = argument(0); // type: int
6607 Node* b = argument(1); // type: int
6608
6609 /*
6610 * int c = ~ crc;
6611 * b = timesXtoThe32[(b ^ c) & 0xFF];
6612 * b = b ^ (c >>> 8);
6613 * crc = ~b;
6614 */
6615
6616 Node* M1 = intcon(-1);
6617 crc = _gvn.transform(new XorINode(crc, M1));
6618 Node* result = _gvn.transform(new XorINode(crc, b));
6619 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6620
6621 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6622 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6623 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6624 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6625
6626 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6627 result = _gvn.transform(new XorINode(crc, result));
6628 result = _gvn.transform(new XorINode(result, M1));
6629 set_result(result);
6630 return true;
6631 }
6632
6633 /**
6634 * Calculate CRC32 for byte[] array.
6635 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6636 */
6637 bool LibraryCallKit::inline_updateBytesCRC32() {
6638 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6639 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6640 // no receiver since it is static method
6641 Node* crc = argument(0); // type: int
6642 Node* src = argument(1); // type: oop
6643 Node* offset = argument(2); // type: int
6644 Node* length = argument(3); // type: int
6645
6646 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6647 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6648 // failed array check
6649 return false;
6650 }
6651
6652 // Figure out the size and type of the elements we will be copying.
6653 BasicType src_elem = src_type->elem()->array_element_basic_type();
6654 if (src_elem != T_BYTE) {
6655 return false;
6656 }
6657
6658 // 'src_start' points to src array + scaled offset
6659 src = must_be_not_null(src, true);
6660 Node* src_start = array_element_address(src, offset, src_elem);
6661
6662 // We assume that range check is done by caller.
6663 // TODO: generate range check (offset+length < src.length) in debug VM.
6664
6665 // Call the stub.
6666 address stubAddr = StubRoutines::updateBytesCRC32();
6667 const char *stubName = "updateBytesCRC32";
6668
6669 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6670 stubAddr, stubName, TypePtr::BOTTOM,
6671 crc, src_start, length);
6672 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6673 set_result(result);
6674 return true;
6675 }
6676
6677 /**
6678 * Calculate CRC32 for ByteBuffer.
6679 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6680 */
6681 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6682 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6683 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6684 // no receiver since it is static method
6685 Node* crc = argument(0); // type: int
6686 Node* src = argument(1); // type: long
6687 Node* offset = argument(3); // type: int
6688 Node* length = argument(4); // type: int
6689
6690 src = ConvL2X(src); // adjust Java long to machine word
6691 Node* base = _gvn.transform(new CastX2PNode(src));
6692 offset = ConvI2X(offset);
6693
6694 // 'src_start' points to src array + scaled offset
6695 Node* src_start = basic_plus_adr(top(), base, offset);
6696
6697 // Call the stub.
6698 address stubAddr = StubRoutines::updateBytesCRC32();
6699 const char *stubName = "updateBytesCRC32";
6700
6701 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6702 stubAddr, stubName, TypePtr::BOTTOM,
6703 crc, src_start, length);
6704 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6705 set_result(result);
6706 return true;
6707 }
6708
6709 //------------------------------get_table_from_crc32c_class-----------------------
6710 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6711 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6712 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6713
6714 return table;
6715 }
6716
6717 //------------------------------inline_updateBytesCRC32C-----------------------
6718 //
6719 // Calculate CRC32C for byte[] array.
6720 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6721 //
6722 bool LibraryCallKit::inline_updateBytesCRC32C() {
6723 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6724 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6725 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6726 // no receiver since it is a static method
6727 Node* crc = argument(0); // type: int
6728 Node* src = argument(1); // type: oop
6729 Node* offset = argument(2); // type: int
6730 Node* end = argument(3); // type: int
6731
6732 Node* length = _gvn.transform(new SubINode(end, offset));
6733
6734 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6735 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6736 // failed array check
6737 return false;
6738 }
6739
6740 // Figure out the size and type of the elements we will be copying.
6741 BasicType src_elem = src_type->elem()->array_element_basic_type();
6742 if (src_elem != T_BYTE) {
6743 return false;
6744 }
6745
6746 // 'src_start' points to src array + scaled offset
6747 src = must_be_not_null(src, true);
6748 Node* src_start = array_element_address(src, offset, src_elem);
6749
6750 // static final int[] byteTable in class CRC32C
6751 Node* table = get_table_from_crc32c_class(callee()->holder());
6752 table = must_be_not_null(table, true);
6753 Node* table_start = array_element_address(table, intcon(0), T_INT);
6754
6755 // We assume that range check is done by caller.
6756 // TODO: generate range check (offset+length < src.length) in debug VM.
6757
6758 // Call the stub.
6759 address stubAddr = StubRoutines::updateBytesCRC32C();
6760 const char *stubName = "updateBytesCRC32C";
6761
6762 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6763 stubAddr, stubName, TypePtr::BOTTOM,
6764 crc, src_start, length, table_start);
6765 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6766 set_result(result);
6767 return true;
6768 }
6769
6770 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6771 //
6772 // Calculate CRC32C for DirectByteBuffer.
6773 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6774 //
6775 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6776 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6777 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6778 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6779 // no receiver since it is a static method
6780 Node* crc = argument(0); // type: int
6781 Node* src = argument(1); // type: long
6782 Node* offset = argument(3); // type: int
6783 Node* end = argument(4); // type: int
6784
6785 Node* length = _gvn.transform(new SubINode(end, offset));
6786
6787 src = ConvL2X(src); // adjust Java long to machine word
6788 Node* base = _gvn.transform(new CastX2PNode(src));
6789 offset = ConvI2X(offset);
6790
6791 // 'src_start' points to src array + scaled offset
6792 Node* src_start = basic_plus_adr(top(), base, offset);
6793
6794 // static final int[] byteTable in class CRC32C
6795 Node* table = get_table_from_crc32c_class(callee()->holder());
6796 table = must_be_not_null(table, true);
6797 Node* table_start = array_element_address(table, intcon(0), T_INT);
6798
6799 // Call the stub.
6800 address stubAddr = StubRoutines::updateBytesCRC32C();
6801 const char *stubName = "updateBytesCRC32C";
6802
6803 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6804 stubAddr, stubName, TypePtr::BOTTOM,
6805 crc, src_start, length, table_start);
6806 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6807 set_result(result);
6808 return true;
6809 }
6810
6811 //------------------------------inline_updateBytesAdler32----------------------
6812 //
6813 // Calculate Adler32 checksum for byte[] array.
6814 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6815 //
6816 bool LibraryCallKit::inline_updateBytesAdler32() {
6817 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6818 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6819 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6820 // no receiver since it is static method
6821 Node* crc = argument(0); // type: int
6822 Node* src = argument(1); // type: oop
6823 Node* offset = argument(2); // type: int
6824 Node* length = argument(3); // type: int
6825
6826 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6827 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6828 // failed array check
6829 return false;
6830 }
6831
6832 // Figure out the size and type of the elements we will be copying.
6833 BasicType src_elem = src_type->elem()->array_element_basic_type();
6834 if (src_elem != T_BYTE) {
6835 return false;
6836 }
6837
6838 // 'src_start' points to src array + scaled offset
6839 Node* src_start = array_element_address(src, offset, src_elem);
6840
6841 // We assume that range check is done by caller.
6842 // TODO: generate range check (offset+length < src.length) in debug VM.
6843
6844 // Call the stub.
6845 address stubAddr = StubRoutines::updateBytesAdler32();
6846 const char *stubName = "updateBytesAdler32";
6847
6848 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6849 stubAddr, stubName, TypePtr::BOTTOM,
6850 crc, src_start, length);
6851 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6852 set_result(result);
6853 return true;
6854 }
6855
6856 //------------------------------inline_updateByteBufferAdler32---------------
6857 //
6858 // Calculate Adler32 checksum for DirectByteBuffer.
6859 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6860 //
6861 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6862 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6863 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6864 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6865 // no receiver since it is static method
6866 Node* crc = argument(0); // type: int
6867 Node* src = argument(1); // type: long
6868 Node* offset = argument(3); // type: int
6869 Node* length = argument(4); // type: int
6870
6871 src = ConvL2X(src); // adjust Java long to machine word
6872 Node* base = _gvn.transform(new CastX2PNode(src));
6873 offset = ConvI2X(offset);
6874
6875 // 'src_start' points to src array + scaled offset
6876 Node* src_start = basic_plus_adr(top(), base, offset);
6877
6878 // Call the stub.
6879 address stubAddr = StubRoutines::updateBytesAdler32();
6880 const char *stubName = "updateBytesAdler32";
6881
6882 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6883 stubAddr, stubName, TypePtr::BOTTOM,
6884 crc, src_start, length);
6885
6886 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6887 set_result(result);
6888 return true;
6889 }
6890
6891 //----------------------------inline_reference_get----------------------------
6892 // public T java.lang.ref.Reference.get();
6893 bool LibraryCallKit::inline_reference_get() {
6894 const int referent_offset = java_lang_ref_Reference::referent_offset();
6895
6896 // Get the argument:
6897 Node* reference_obj = null_check_receiver();
6898 if (stopped()) return true;
6899
6900 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6901 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6902 decorators, /*is_static*/ false, nullptr);
6903 if (result == nullptr) return false;
6904
6905 // Add memory barrier to prevent commoning reads from this field
6906 // across safepoint since GC can change its value.
6907 insert_mem_bar(Op_MemBarCPUOrder);
6908
6909 set_result(result);
6910 return true;
6911 }
6912
6913 //----------------------------inline_reference_refersTo0----------------------------
6914 // bool java.lang.ref.Reference.refersTo0();
6915 // bool java.lang.ref.PhantomReference.refersTo0();
6916 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6917 // Get arguments:
6918 Node* reference_obj = null_check_receiver();
6919 Node* other_obj = argument(1);
6920 if (stopped()) return true;
6921
6922 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6923 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6924 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6925 decorators, /*is_static*/ false, nullptr);
6926 if (referent == nullptr) return false;
6927
6928 // Add memory barrier to prevent commoning reads from this field
6929 // across safepoint since GC can change its value.
6930 insert_mem_bar(Op_MemBarCPUOrder);
6931
6932 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
6933 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6934 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
6935
6936 RegionNode* region = new RegionNode(3);
6937 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
6938
6939 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
6940 region->init_req(1, if_true);
6941 phi->init_req(1, intcon(1));
6942
6943 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
6944 region->init_req(2, if_false);
6945 phi->init_req(2, intcon(0));
6946
6947 set_control(_gvn.transform(region));
6948 record_for_igvn(region);
6949 set_result(_gvn.transform(phi));
6950 return true;
6951 }
6952
6953 //----------------------------inline_reference_clear0----------------------------
6954 // void java.lang.ref.Reference.clear0();
6955 // void java.lang.ref.PhantomReference.clear0();
6956 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
6957 // This matches the implementation in JVM_ReferenceClear, see the comments there.
6958
6959 // Get arguments
6960 Node* reference_obj = null_check_receiver();
6961 if (stopped()) return true;
6962
6963 // Common access parameters
6964 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6965 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6966 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
6967 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
6968 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
6969
6970 Node* referent = access_load_at(reference_obj,
6971 referent_field_addr,
6972 referent_field_addr_type,
6973 val_type,
6974 T_OBJECT,
6975 decorators);
6976
6977 IdealKit ideal(this);
6978 #define __ ideal.
6979 __ if_then(referent, BoolTest::ne, null());
6980 sync_kit(ideal);
6981 access_store_at(reference_obj,
6982 referent_field_addr,
6983 referent_field_addr_type,
6984 null(),
6985 val_type,
6986 T_OBJECT,
6987 decorators);
6988 __ sync_kit(this);
6989 __ end_if();
6990 final_sync(ideal);
6991 #undef __
6992
6993 return true;
6994 }
6995
6996 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
6997 DecoratorSet decorators, bool is_static,
6998 ciInstanceKlass* fromKls) {
6999 if (fromKls == nullptr) {
7000 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7001 assert(tinst != nullptr, "obj is null");
7002 assert(tinst->is_loaded(), "obj is not loaded");
7003 fromKls = tinst->instance_klass();
7004 } else {
7005 assert(is_static, "only for static field access");
7006 }
7007 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7008 ciSymbol::make(fieldTypeString),
7009 is_static);
7010
7011 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7012 if (field == nullptr) return (Node *) nullptr;
7013
7014 if (is_static) {
7015 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7016 fromObj = makecon(tip);
7017 }
7018
7019 // Next code copied from Parse::do_get_xxx():
7020
7021 // Compute address and memory type.
7022 int offset = field->offset_in_bytes();
7023 bool is_vol = field->is_volatile();
7024 ciType* field_klass = field->type();
7025 assert(field_klass->is_loaded(), "should be loaded");
7026 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7027 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7028 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7029 "slice of address and input slice don't match");
7030 BasicType bt = field->layout_type();
7031
7032 // Build the resultant type of the load
7033 const Type *type;
7034 if (bt == T_OBJECT) {
7035 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7036 } else {
7037 type = Type::get_const_basic_type(bt);
7038 }
7039
7040 if (is_vol) {
7041 decorators |= MO_SEQ_CST;
7042 }
7043
7044 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7045 }
7046
7047 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7048 bool is_exact /* true */, bool is_static /* false */,
7049 ciInstanceKlass * fromKls /* nullptr */) {
7050 if (fromKls == nullptr) {
7051 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7052 assert(tinst != nullptr, "obj is null");
7053 assert(tinst->is_loaded(), "obj is not loaded");
7054 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7055 fromKls = tinst->instance_klass();
7056 }
7057 else {
7058 assert(is_static, "only for static field access");
7059 }
7060 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7061 ciSymbol::make(fieldTypeString),
7062 is_static);
7063
7064 assert(field != nullptr, "undefined field");
7065 assert(!field->is_volatile(), "not defined for volatile fields");
7066
7067 if (is_static) {
7068 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7069 fromObj = makecon(tip);
7070 }
7071
7072 // Next code copied from Parse::do_get_xxx():
7073
7074 // Compute address and memory type.
7075 int offset = field->offset_in_bytes();
7076 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7077
7078 return adr;
7079 }
7080
7081 //------------------------------inline_aescrypt_Block-----------------------
7082 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7083 address stubAddr = nullptr;
7084 const char *stubName;
7085 assert(UseAES, "need AES instruction support");
7086
7087 switch(id) {
7088 case vmIntrinsics::_aescrypt_encryptBlock:
7089 stubAddr = StubRoutines::aescrypt_encryptBlock();
7090 stubName = "aescrypt_encryptBlock";
7091 break;
7092 case vmIntrinsics::_aescrypt_decryptBlock:
7093 stubAddr = StubRoutines::aescrypt_decryptBlock();
7094 stubName = "aescrypt_decryptBlock";
7095 break;
7096 default:
7097 break;
7098 }
7099 if (stubAddr == nullptr) return false;
7100
7101 Node* aescrypt_object = argument(0);
7102 Node* src = argument(1);
7103 Node* src_offset = argument(2);
7104 Node* dest = argument(3);
7105 Node* dest_offset = argument(4);
7106
7107 src = must_be_not_null(src, true);
7108 dest = must_be_not_null(dest, true);
7109
7110 // (1) src and dest are arrays.
7111 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7112 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7113 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7114 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7115
7116 // for the quick and dirty code we will skip all the checks.
7117 // we are just trying to get the call to be generated.
7118 Node* src_start = src;
7119 Node* dest_start = dest;
7120 if (src_offset != nullptr || dest_offset != nullptr) {
7121 assert(src_offset != nullptr && dest_offset != nullptr, "");
7122 src_start = array_element_address(src, src_offset, T_BYTE);
7123 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7124 }
7125
7126 // now need to get the start of its expanded key array
7127 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7128 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7129 if (k_start == nullptr) return false;
7130
7131 // Call the stub.
7132 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7133 stubAddr, stubName, TypePtr::BOTTOM,
7134 src_start, dest_start, k_start);
7135
7136 return true;
7137 }
7138
7139 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7140 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7141 address stubAddr = nullptr;
7142 const char *stubName = nullptr;
7143
7144 assert(UseAES, "need AES instruction support");
7145
7146 switch(id) {
7147 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7148 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7149 stubName = "cipherBlockChaining_encryptAESCrypt";
7150 break;
7151 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7152 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7153 stubName = "cipherBlockChaining_decryptAESCrypt";
7154 break;
7155 default:
7156 break;
7157 }
7158 if (stubAddr == nullptr) return false;
7159
7160 Node* cipherBlockChaining_object = argument(0);
7161 Node* src = argument(1);
7162 Node* src_offset = argument(2);
7163 Node* len = argument(3);
7164 Node* dest = argument(4);
7165 Node* dest_offset = argument(5);
7166
7167 src = must_be_not_null(src, false);
7168 dest = must_be_not_null(dest, false);
7169
7170 // (1) src and dest are arrays.
7171 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7172 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7173 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7174 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7175
7176 // checks are the responsibility of the caller
7177 Node* src_start = src;
7178 Node* dest_start = dest;
7179 if (src_offset != nullptr || dest_offset != nullptr) {
7180 assert(src_offset != nullptr && dest_offset != nullptr, "");
7181 src_start = array_element_address(src, src_offset, T_BYTE);
7182 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7183 }
7184
7185 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7186 // (because of the predicated logic executed earlier).
7187 // so we cast it here safely.
7188 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7189
7190 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7191 if (embeddedCipherObj == nullptr) return false;
7192
7193 // cast it to what we know it will be at runtime
7194 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7195 assert(tinst != nullptr, "CBC obj is null");
7196 assert(tinst->is_loaded(), "CBC obj is not loaded");
7197 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7198 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7199
7200 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7201 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7202 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7203 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7204 aescrypt_object = _gvn.transform(aescrypt_object);
7205
7206 // we need to get the start of the aescrypt_object's expanded key array
7207 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7208 if (k_start == nullptr) return false;
7209
7210 // similarly, get the start address of the r vector
7211 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7212 if (objRvec == nullptr) return false;
7213 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7214
7215 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7216 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7217 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7218 stubAddr, stubName, TypePtr::BOTTOM,
7219 src_start, dest_start, k_start, r_start, len);
7220
7221 // return cipher length (int)
7222 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7223 set_result(retvalue);
7224 return true;
7225 }
7226
7227 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7228 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7229 address stubAddr = nullptr;
7230 const char *stubName = nullptr;
7231
7232 assert(UseAES, "need AES instruction support");
7233
7234 switch (id) {
7235 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7236 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7237 stubName = "electronicCodeBook_encryptAESCrypt";
7238 break;
7239 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7240 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7241 stubName = "electronicCodeBook_decryptAESCrypt";
7242 break;
7243 default:
7244 break;
7245 }
7246
7247 if (stubAddr == nullptr) return false;
7248
7249 Node* electronicCodeBook_object = argument(0);
7250 Node* src = argument(1);
7251 Node* src_offset = argument(2);
7252 Node* len = argument(3);
7253 Node* dest = argument(4);
7254 Node* dest_offset = argument(5);
7255
7256 // (1) src and dest are arrays.
7257 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7258 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7259 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7260 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7261
7262 // checks are the responsibility of the caller
7263 Node* src_start = src;
7264 Node* dest_start = dest;
7265 if (src_offset != nullptr || dest_offset != nullptr) {
7266 assert(src_offset != nullptr && dest_offset != nullptr, "");
7267 src_start = array_element_address(src, src_offset, T_BYTE);
7268 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7269 }
7270
7271 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7272 // (because of the predicated logic executed earlier).
7273 // so we cast it here safely.
7274 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7275
7276 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7277 if (embeddedCipherObj == nullptr) return false;
7278
7279 // cast it to what we know it will be at runtime
7280 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7281 assert(tinst != nullptr, "ECB obj is null");
7282 assert(tinst->is_loaded(), "ECB obj is not loaded");
7283 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7284 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7285
7286 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7287 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7288 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7289 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7290 aescrypt_object = _gvn.transform(aescrypt_object);
7291
7292 // we need to get the start of the aescrypt_object's expanded key array
7293 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7294 if (k_start == nullptr) return false;
7295
7296 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7297 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7298 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7299 stubAddr, stubName, TypePtr::BOTTOM,
7300 src_start, dest_start, k_start, len);
7301
7302 // return cipher length (int)
7303 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7304 set_result(retvalue);
7305 return true;
7306 }
7307
7308 //------------------------------inline_counterMode_AESCrypt-----------------------
7309 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7310 assert(UseAES, "need AES instruction support");
7311 if (!UseAESCTRIntrinsics) return false;
7312
7313 address stubAddr = nullptr;
7314 const char *stubName = nullptr;
7315 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7316 stubAddr = StubRoutines::counterMode_AESCrypt();
7317 stubName = "counterMode_AESCrypt";
7318 }
7319 if (stubAddr == nullptr) return false;
7320
7321 Node* counterMode_object = argument(0);
7322 Node* src = argument(1);
7323 Node* src_offset = argument(2);
7324 Node* len = argument(3);
7325 Node* dest = argument(4);
7326 Node* dest_offset = argument(5);
7327
7328 // (1) src and dest are arrays.
7329 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7330 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7331 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7332 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7333
7334 // checks are the responsibility of the caller
7335 Node* src_start = src;
7336 Node* dest_start = dest;
7337 if (src_offset != nullptr || dest_offset != nullptr) {
7338 assert(src_offset != nullptr && dest_offset != nullptr, "");
7339 src_start = array_element_address(src, src_offset, T_BYTE);
7340 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7341 }
7342
7343 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7344 // (because of the predicated logic executed earlier).
7345 // so we cast it here safely.
7346 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7347 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7348 if (embeddedCipherObj == nullptr) return false;
7349 // cast it to what we know it will be at runtime
7350 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7351 assert(tinst != nullptr, "CTR obj is null");
7352 assert(tinst->is_loaded(), "CTR obj is not loaded");
7353 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7354 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7355 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7356 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7357 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7358 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7359 aescrypt_object = _gvn.transform(aescrypt_object);
7360 // we need to get the start of the aescrypt_object's expanded key array
7361 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7362 if (k_start == nullptr) return false;
7363 // similarly, get the start address of the r vector
7364 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7365 if (obj_counter == nullptr) return false;
7366 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7367
7368 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7369 if (saved_encCounter == nullptr) return false;
7370 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7371 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7372
7373 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7374 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7375 OptoRuntime::counterMode_aescrypt_Type(),
7376 stubAddr, stubName, TypePtr::BOTTOM,
7377 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7378
7379 // return cipher length (int)
7380 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7381 set_result(retvalue);
7382 return true;
7383 }
7384
7385 //------------------------------get_key_start_from_aescrypt_object-----------------------
7386 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7387 #if defined(PPC64) || defined(S390) || defined(RISCV64)
7388 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7389 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7390 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7391 // The ppc64 and riscv64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7392 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7393 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7394 if (objSessionK == nullptr) {
7395 return (Node *) nullptr;
7396 }
7397 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7398 #else
7399 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7400 #endif // PPC64
7401 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7402 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7403
7404 // now have the array, need to get the start address of the K array
7405 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7406 return k_start;
7407 }
7408
7409 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7410 // Return node representing slow path of predicate check.
7411 // the pseudo code we want to emulate with this predicate is:
7412 // for encryption:
7413 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7414 // for decryption:
7415 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7416 // note cipher==plain is more conservative than the original java code but that's OK
7417 //
7418 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7419 // The receiver was checked for null already.
7420 Node* objCBC = argument(0);
7421
7422 Node* src = argument(1);
7423 Node* dest = argument(4);
7424
7425 // Load embeddedCipher field of CipherBlockChaining object.
7426 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7427
7428 // get AESCrypt klass for instanceOf check
7429 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7430 // will have same classloader as CipherBlockChaining object
7431 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7432 assert(tinst != nullptr, "CBCobj is null");
7433 assert(tinst->is_loaded(), "CBCobj is not loaded");
7434
7435 // we want to do an instanceof comparison against the AESCrypt class
7436 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7437 if (!klass_AESCrypt->is_loaded()) {
7438 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7439 Node* ctrl = control();
7440 set_control(top()); // no regular fast path
7441 return ctrl;
7442 }
7443
7444 src = must_be_not_null(src, true);
7445 dest = must_be_not_null(dest, true);
7446
7447 // Resolve oops to stable for CmpP below.
7448 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7449
7450 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7451 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7452 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7453
7454 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7455
7456 // for encryption, we are done
7457 if (!decrypting)
7458 return instof_false; // even if it is null
7459
7460 // for decryption, we need to add a further check to avoid
7461 // taking the intrinsic path when cipher and plain are the same
7462 // see the original java code for why.
7463 RegionNode* region = new RegionNode(3);
7464 region->init_req(1, instof_false);
7465
7466 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7467 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7468 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7469 region->init_req(2, src_dest_conjoint);
7470
7471 record_for_igvn(region);
7472 return _gvn.transform(region);
7473 }
7474
7475 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7476 // Return node representing slow path of predicate check.
7477 // the pseudo code we want to emulate with this predicate is:
7478 // for encryption:
7479 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7480 // for decryption:
7481 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7482 // note cipher==plain is more conservative than the original java code but that's OK
7483 //
7484 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7485 // The receiver was checked for null already.
7486 Node* objECB = argument(0);
7487
7488 // Load embeddedCipher field of ElectronicCodeBook object.
7489 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7490
7491 // get AESCrypt klass for instanceOf check
7492 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7493 // will have same classloader as ElectronicCodeBook object
7494 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7495 assert(tinst != nullptr, "ECBobj is null");
7496 assert(tinst->is_loaded(), "ECBobj is not loaded");
7497
7498 // we want to do an instanceof comparison against the AESCrypt class
7499 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7500 if (!klass_AESCrypt->is_loaded()) {
7501 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7502 Node* ctrl = control();
7503 set_control(top()); // no regular fast path
7504 return ctrl;
7505 }
7506 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7507
7508 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7509 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7510 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7511
7512 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7513
7514 // for encryption, we are done
7515 if (!decrypting)
7516 return instof_false; // even if it is null
7517
7518 // for decryption, we need to add a further check to avoid
7519 // taking the intrinsic path when cipher and plain are the same
7520 // see the original java code for why.
7521 RegionNode* region = new RegionNode(3);
7522 region->init_req(1, instof_false);
7523 Node* src = argument(1);
7524 Node* dest = argument(4);
7525 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7526 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7527 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7528 region->init_req(2, src_dest_conjoint);
7529
7530 record_for_igvn(region);
7531 return _gvn.transform(region);
7532 }
7533
7534 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7535 // Return node representing slow path of predicate check.
7536 // the pseudo code we want to emulate with this predicate is:
7537 // for encryption:
7538 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7539 // for decryption:
7540 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7541 // note cipher==plain is more conservative than the original java code but that's OK
7542 //
7543
7544 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7545 // The receiver was checked for null already.
7546 Node* objCTR = argument(0);
7547
7548 // Load embeddedCipher field of CipherBlockChaining object.
7549 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7550
7551 // get AESCrypt klass for instanceOf check
7552 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7553 // will have same classloader as CipherBlockChaining object
7554 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7555 assert(tinst != nullptr, "CTRobj is null");
7556 assert(tinst->is_loaded(), "CTRobj is not loaded");
7557
7558 // we want to do an instanceof comparison against the AESCrypt class
7559 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7560 if (!klass_AESCrypt->is_loaded()) {
7561 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7562 Node* ctrl = control();
7563 set_control(top()); // no regular fast path
7564 return ctrl;
7565 }
7566
7567 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7568 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7569 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7570 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7571 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7572
7573 return instof_false; // even if it is null
7574 }
7575
7576 //------------------------------inline_ghash_processBlocks
7577 bool LibraryCallKit::inline_ghash_processBlocks() {
7578 address stubAddr;
7579 const char *stubName;
7580 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7581
7582 stubAddr = StubRoutines::ghash_processBlocks();
7583 stubName = "ghash_processBlocks";
7584
7585 Node* data = argument(0);
7586 Node* offset = argument(1);
7587 Node* len = argument(2);
7588 Node* state = argument(3);
7589 Node* subkeyH = argument(4);
7590
7591 state = must_be_not_null(state, true);
7592 subkeyH = must_be_not_null(subkeyH, true);
7593 data = must_be_not_null(data, true);
7594
7595 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7596 assert(state_start, "state is null");
7597 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7598 assert(subkeyH_start, "subkeyH is null");
7599 Node* data_start = array_element_address(data, offset, T_BYTE);
7600 assert(data_start, "data is null");
7601
7602 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7603 OptoRuntime::ghash_processBlocks_Type(),
7604 stubAddr, stubName, TypePtr::BOTTOM,
7605 state_start, subkeyH_start, data_start, len);
7606 return true;
7607 }
7608
7609 //------------------------------inline_chacha20Block
7610 bool LibraryCallKit::inline_chacha20Block() {
7611 address stubAddr;
7612 const char *stubName;
7613 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7614
7615 stubAddr = StubRoutines::chacha20Block();
7616 stubName = "chacha20Block";
7617
7618 Node* state = argument(0);
7619 Node* result = argument(1);
7620
7621 state = must_be_not_null(state, true);
7622 result = must_be_not_null(result, true);
7623
7624 Node* state_start = array_element_address(state, intcon(0), T_INT);
7625 assert(state_start, "state is null");
7626 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7627 assert(result_start, "result is null");
7628
7629 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7630 OptoRuntime::chacha20Block_Type(),
7631 stubAddr, stubName, TypePtr::BOTTOM,
7632 state_start, result_start);
7633 // return key stream length (int)
7634 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7635 set_result(retvalue);
7636 return true;
7637 }
7638
7639 bool LibraryCallKit::inline_base64_encodeBlock() {
7640 address stubAddr;
7641 const char *stubName;
7642 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7643 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
7644 stubAddr = StubRoutines::base64_encodeBlock();
7645 stubName = "encodeBlock";
7646
7647 if (!stubAddr) return false;
7648 Node* base64obj = argument(0);
7649 Node* src = argument(1);
7650 Node* offset = argument(2);
7651 Node* len = argument(3);
7652 Node* dest = argument(4);
7653 Node* dp = argument(5);
7654 Node* isURL = argument(6);
7655
7656 src = must_be_not_null(src, true);
7657 dest = must_be_not_null(dest, true);
7658
7659 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7660 assert(src_start, "source array is null");
7661 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7662 assert(dest_start, "destination array is null");
7663
7664 Node* base64 = make_runtime_call(RC_LEAF,
7665 OptoRuntime::base64_encodeBlock_Type(),
7666 stubAddr, stubName, TypePtr::BOTTOM,
7667 src_start, offset, len, dest_start, dp, isURL);
7668 return true;
7669 }
7670
7671 bool LibraryCallKit::inline_base64_decodeBlock() {
7672 address stubAddr;
7673 const char *stubName;
7674 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7675 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
7676 stubAddr = StubRoutines::base64_decodeBlock();
7677 stubName = "decodeBlock";
7678
7679 if (!stubAddr) return false;
7680 Node* base64obj = argument(0);
7681 Node* src = argument(1);
7682 Node* src_offset = argument(2);
7683 Node* len = argument(3);
7684 Node* dest = argument(4);
7685 Node* dest_offset = argument(5);
7686 Node* isURL = argument(6);
7687 Node* isMIME = argument(7);
7688
7689 src = must_be_not_null(src, true);
7690 dest = must_be_not_null(dest, true);
7691
7692 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7693 assert(src_start, "source array is null");
7694 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7695 assert(dest_start, "destination array is null");
7696
7697 Node* call = make_runtime_call(RC_LEAF,
7698 OptoRuntime::base64_decodeBlock_Type(),
7699 stubAddr, stubName, TypePtr::BOTTOM,
7700 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
7701 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7702 set_result(result);
7703 return true;
7704 }
7705
7706 bool LibraryCallKit::inline_poly1305_processBlocks() {
7707 address stubAddr;
7708 const char *stubName;
7709 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
7710 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
7711 stubAddr = StubRoutines::poly1305_processBlocks();
7712 stubName = "poly1305_processBlocks";
7713
7714 if (!stubAddr) return false;
7715 null_check_receiver(); // null-check receiver
7716 if (stopped()) return true;
7717
7718 Node* input = argument(1);
7719 Node* input_offset = argument(2);
7720 Node* len = argument(3);
7721 Node* alimbs = argument(4);
7722 Node* rlimbs = argument(5);
7723
7724 input = must_be_not_null(input, true);
7725 alimbs = must_be_not_null(alimbs, true);
7726 rlimbs = must_be_not_null(rlimbs, true);
7727
7728 Node* input_start = array_element_address(input, input_offset, T_BYTE);
7729 assert(input_start, "input array is null");
7730 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
7731 assert(acc_start, "acc array is null");
7732 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
7733 assert(r_start, "r array is null");
7734
7735 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7736 OptoRuntime::poly1305_processBlocks_Type(),
7737 stubAddr, stubName, TypePtr::BOTTOM,
7738 input_start, len, acc_start, r_start);
7739 return true;
7740 }
7741
7742 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
7743 address stubAddr;
7744 const char *stubName;
7745 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
7746 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
7747 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
7748 stubName = "intpoly_montgomeryMult_P256";
7749
7750 if (!stubAddr) return false;
7751 null_check_receiver(); // null-check receiver
7752 if (stopped()) return true;
7753
7754 Node* a = argument(1);
7755 Node* b = argument(2);
7756 Node* r = argument(3);
7757
7758 a = must_be_not_null(a, true);
7759 b = must_be_not_null(b, true);
7760 r = must_be_not_null(r, true);
7761
7762 Node* a_start = array_element_address(a, intcon(0), T_LONG);
7763 assert(a_start, "a array is null");
7764 Node* b_start = array_element_address(b, intcon(0), T_LONG);
7765 assert(b_start, "b array is null");
7766 Node* r_start = array_element_address(r, intcon(0), T_LONG);
7767 assert(r_start, "r array is null");
7768
7769 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7770 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
7771 stubAddr, stubName, TypePtr::BOTTOM,
7772 a_start, b_start, r_start);
7773 return true;
7774 }
7775
7776 bool LibraryCallKit::inline_intpoly_assign() {
7777 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
7778 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
7779 const char *stubName = "intpoly_assign";
7780 address stubAddr = StubRoutines::intpoly_assign();
7781 if (!stubAddr) return false;
7782
7783 Node* set = argument(0);
7784 Node* a = argument(1);
7785 Node* b = argument(2);
7786 Node* arr_length = load_array_length(a);
7787
7788 a = must_be_not_null(a, true);
7789 b = must_be_not_null(b, true);
7790
7791 Node* a_start = array_element_address(a, intcon(0), T_LONG);
7792 assert(a_start, "a array is null");
7793 Node* b_start = array_element_address(b, intcon(0), T_LONG);
7794 assert(b_start, "b array is null");
7795
7796 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7797 OptoRuntime::intpoly_assign_Type(),
7798 stubAddr, stubName, TypePtr::BOTTOM,
7799 set, a_start, b_start, arr_length);
7800 return true;
7801 }
7802
7803 //------------------------------inline_digestBase_implCompress-----------------------
7804 //
7805 // Calculate MD5 for single-block byte[] array.
7806 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
7807 //
7808 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
7809 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
7810 //
7811 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
7812 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
7813 //
7814 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
7815 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
7816 //
7817 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
7818 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
7819 //
7820 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
7821 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
7822
7823 Node* digestBase_obj = argument(0);
7824 Node* src = argument(1); // type oop
7825 Node* ofs = argument(2); // type int
7826
7827 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7828 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7829 // failed array check
7830 return false;
7831 }
7832 // Figure out the size and type of the elements we will be copying.
7833 BasicType src_elem = src_type->elem()->array_element_basic_type();
7834 if (src_elem != T_BYTE) {
7835 return false;
7836 }
7837 // 'src_start' points to src array + offset
7838 src = must_be_not_null(src, true);
7839 Node* src_start = array_element_address(src, ofs, src_elem);
7840 Node* state = nullptr;
7841 Node* block_size = nullptr;
7842 address stubAddr;
7843 const char *stubName;
7844
7845 switch(id) {
7846 case vmIntrinsics::_md5_implCompress:
7847 assert(UseMD5Intrinsics, "need MD5 instruction support");
7848 state = get_state_from_digest_object(digestBase_obj, T_INT);
7849 stubAddr = StubRoutines::md5_implCompress();
7850 stubName = "md5_implCompress";
7851 break;
7852 case vmIntrinsics::_sha_implCompress:
7853 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
7854 state = get_state_from_digest_object(digestBase_obj, T_INT);
7855 stubAddr = StubRoutines::sha1_implCompress();
7856 stubName = "sha1_implCompress";
7857 break;
7858 case vmIntrinsics::_sha2_implCompress:
7859 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
7860 state = get_state_from_digest_object(digestBase_obj, T_INT);
7861 stubAddr = StubRoutines::sha256_implCompress();
7862 stubName = "sha256_implCompress";
7863 break;
7864 case vmIntrinsics::_sha5_implCompress:
7865 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
7866 state = get_state_from_digest_object(digestBase_obj, T_LONG);
7867 stubAddr = StubRoutines::sha512_implCompress();
7868 stubName = "sha512_implCompress";
7869 break;
7870 case vmIntrinsics::_sha3_implCompress:
7871 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
7872 state = get_state_from_digest_object(digestBase_obj, T_LONG);
7873 stubAddr = StubRoutines::sha3_implCompress();
7874 stubName = "sha3_implCompress";
7875 block_size = get_block_size_from_digest_object(digestBase_obj);
7876 if (block_size == nullptr) return false;
7877 break;
7878 default:
7879 fatal_unexpected_iid(id);
7880 return false;
7881 }
7882 if (state == nullptr) return false;
7883
7884 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
7885 if (stubAddr == nullptr) return false;
7886
7887 // Call the stub.
7888 Node* call;
7889 if (block_size == nullptr) {
7890 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
7891 stubAddr, stubName, TypePtr::BOTTOM,
7892 src_start, state);
7893 } else {
7894 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
7895 stubAddr, stubName, TypePtr::BOTTOM,
7896 src_start, state, block_size);
7897 }
7898
7899 return true;
7900 }
7901
7902 //------------------------------inline_digestBase_implCompressMB-----------------------
7903 //
7904 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
7905 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
7906 //
7907 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
7908 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7909 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7910 assert((uint)predicate < 5, "sanity");
7911 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
7912
7913 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
7914 Node* src = argument(1); // byte[] array
7915 Node* ofs = argument(2); // type int
7916 Node* limit = argument(3); // type int
7917
7918 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7919 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7920 // failed array check
7921 return false;
7922 }
7923 // Figure out the size and type of the elements we will be copying.
7924 BasicType src_elem = src_type->elem()->array_element_basic_type();
7925 if (src_elem != T_BYTE) {
7926 return false;
7927 }
7928 // 'src_start' points to src array + offset
7929 src = must_be_not_null(src, false);
7930 Node* src_start = array_element_address(src, ofs, src_elem);
7931
7932 const char* klass_digestBase_name = nullptr;
7933 const char* stub_name = nullptr;
7934 address stub_addr = nullptr;
7935 BasicType elem_type = T_INT;
7936
7937 switch (predicate) {
7938 case 0:
7939 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
7940 klass_digestBase_name = "sun/security/provider/MD5";
7941 stub_name = "md5_implCompressMB";
7942 stub_addr = StubRoutines::md5_implCompressMB();
7943 }
7944 break;
7945 case 1:
7946 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
7947 klass_digestBase_name = "sun/security/provider/SHA";
7948 stub_name = "sha1_implCompressMB";
7949 stub_addr = StubRoutines::sha1_implCompressMB();
7950 }
7951 break;
7952 case 2:
7953 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
7954 klass_digestBase_name = "sun/security/provider/SHA2";
7955 stub_name = "sha256_implCompressMB";
7956 stub_addr = StubRoutines::sha256_implCompressMB();
7957 }
7958 break;
7959 case 3:
7960 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
7961 klass_digestBase_name = "sun/security/provider/SHA5";
7962 stub_name = "sha512_implCompressMB";
7963 stub_addr = StubRoutines::sha512_implCompressMB();
7964 elem_type = T_LONG;
7965 }
7966 break;
7967 case 4:
7968 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
7969 klass_digestBase_name = "sun/security/provider/SHA3";
7970 stub_name = "sha3_implCompressMB";
7971 stub_addr = StubRoutines::sha3_implCompressMB();
7972 elem_type = T_LONG;
7973 }
7974 break;
7975 default:
7976 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
7977 }
7978 if (klass_digestBase_name != nullptr) {
7979 assert(stub_addr != nullptr, "Stub is generated");
7980 if (stub_addr == nullptr) return false;
7981
7982 // get DigestBase klass to lookup for SHA klass
7983 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
7984 assert(tinst != nullptr, "digestBase_obj is not instance???");
7985 assert(tinst->is_loaded(), "DigestBase is not loaded");
7986
7987 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
7988 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
7989 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
7990 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
7991 }
7992 return false;
7993 }
7994
7995 //------------------------------inline_digestBase_implCompressMB-----------------------
7996 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
7997 BasicType elem_type, address stubAddr, const char *stubName,
7998 Node* src_start, Node* ofs, Node* limit) {
7999 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
8000 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8001 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
8002 digest_obj = _gvn.transform(digest_obj);
8003
8004 Node* state = get_state_from_digest_object(digest_obj, elem_type);
8005 if (state == nullptr) return false;
8006
8007 Node* block_size = nullptr;
8008 if (strcmp("sha3_implCompressMB", stubName) == 0) {
8009 block_size = get_block_size_from_digest_object(digest_obj);
8010 if (block_size == nullptr) return false;
8011 }
8012
8013 // Call the stub.
8014 Node* call;
8015 if (block_size == nullptr) {
8016 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8017 OptoRuntime::digestBase_implCompressMB_Type(false),
8018 stubAddr, stubName, TypePtr::BOTTOM,
8019 src_start, state, ofs, limit);
8020 } else {
8021 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8022 OptoRuntime::digestBase_implCompressMB_Type(true),
8023 stubAddr, stubName, TypePtr::BOTTOM,
8024 src_start, state, block_size, ofs, limit);
8025 }
8026
8027 // return ofs (int)
8028 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8029 set_result(result);
8030
8031 return true;
8032 }
8033
8034 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8035 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8036 assert(UseAES, "need AES instruction support");
8037 address stubAddr = nullptr;
8038 const char *stubName = nullptr;
8039 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8040 stubName = "galoisCounterMode_AESCrypt";
8041
8042 if (stubAddr == nullptr) return false;
8043
8044 Node* in = argument(0);
8045 Node* inOfs = argument(1);
8046 Node* len = argument(2);
8047 Node* ct = argument(3);
8048 Node* ctOfs = argument(4);
8049 Node* out = argument(5);
8050 Node* outOfs = argument(6);
8051 Node* gctr_object = argument(7);
8052 Node* ghash_object = argument(8);
8053
8054 // (1) in, ct and out are arrays.
8055 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8056 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8057 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8058 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8059 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8060 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8061
8062 // checks are the responsibility of the caller
8063 Node* in_start = in;
8064 Node* ct_start = ct;
8065 Node* out_start = out;
8066 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8067 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8068 in_start = array_element_address(in, inOfs, T_BYTE);
8069 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8070 out_start = array_element_address(out, outOfs, T_BYTE);
8071 }
8072
8073 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8074 // (because of the predicated logic executed earlier).
8075 // so we cast it here safely.
8076 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8077 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8078 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8079 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8080 Node* state = load_field_from_object(ghash_object, "state", "[J");
8081
8082 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8083 return false;
8084 }
8085 // cast it to what we know it will be at runtime
8086 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8087 assert(tinst != nullptr, "GCTR obj is null");
8088 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8089 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8090 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8091 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8092 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8093 const TypeOopPtr* xtype = aklass->as_instance_type();
8094 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8095 aescrypt_object = _gvn.transform(aescrypt_object);
8096 // we need to get the start of the aescrypt_object's expanded key array
8097 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8098 if (k_start == nullptr) return false;
8099 // similarly, get the start address of the r vector
8100 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8101 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8102 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8103
8104
8105 // Call the stub, passing params
8106 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8107 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8108 stubAddr, stubName, TypePtr::BOTTOM,
8109 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8110
8111 // return cipher length (int)
8112 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8113 set_result(retvalue);
8114
8115 return true;
8116 }
8117
8118 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8119 // Return node representing slow path of predicate check.
8120 // the pseudo code we want to emulate with this predicate is:
8121 // for encryption:
8122 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8123 // for decryption:
8124 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8125 // note cipher==plain is more conservative than the original java code but that's OK
8126 //
8127
8128 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8129 // The receiver was checked for null already.
8130 Node* objGCTR = argument(7);
8131 // Load embeddedCipher field of GCTR object.
8132 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8133 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8134
8135 // get AESCrypt klass for instanceOf check
8136 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8137 // will have same classloader as CipherBlockChaining object
8138 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8139 assert(tinst != nullptr, "GCTR obj is null");
8140 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8141
8142 // we want to do an instanceof comparison against the AESCrypt class
8143 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8144 if (!klass_AESCrypt->is_loaded()) {
8145 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8146 Node* ctrl = control();
8147 set_control(top()); // no regular fast path
8148 return ctrl;
8149 }
8150
8151 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8152 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8153 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8154 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8155 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8156
8157 return instof_false; // even if it is null
8158 }
8159
8160 //------------------------------get_state_from_digest_object-----------------------
8161 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8162 const char* state_type;
8163 switch (elem_type) {
8164 case T_BYTE: state_type = "[B"; break;
8165 case T_INT: state_type = "[I"; break;
8166 case T_LONG: state_type = "[J"; break;
8167 default: ShouldNotReachHere();
8168 }
8169 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8170 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8171 if (digest_state == nullptr) return (Node *) nullptr;
8172
8173 // now have the array, need to get the start address of the state array
8174 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8175 return state;
8176 }
8177
8178 //------------------------------get_block_size_from_sha3_object----------------------------------
8179 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8180 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8181 assert (block_size != nullptr, "sanity");
8182 return block_size;
8183 }
8184
8185 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8186 // Return node representing slow path of predicate check.
8187 // the pseudo code we want to emulate with this predicate is:
8188 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8189 //
8190 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8191 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8192 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8193 assert((uint)predicate < 5, "sanity");
8194
8195 // The receiver was checked for null already.
8196 Node* digestBaseObj = argument(0);
8197
8198 // get DigestBase klass for instanceOf check
8199 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8200 assert(tinst != nullptr, "digestBaseObj is null");
8201 assert(tinst->is_loaded(), "DigestBase is not loaded");
8202
8203 const char* klass_name = nullptr;
8204 switch (predicate) {
8205 case 0:
8206 if (UseMD5Intrinsics) {
8207 // we want to do an instanceof comparison against the MD5 class
8208 klass_name = "sun/security/provider/MD5";
8209 }
8210 break;
8211 case 1:
8212 if (UseSHA1Intrinsics) {
8213 // we want to do an instanceof comparison against the SHA class
8214 klass_name = "sun/security/provider/SHA";
8215 }
8216 break;
8217 case 2:
8218 if (UseSHA256Intrinsics) {
8219 // we want to do an instanceof comparison against the SHA2 class
8220 klass_name = "sun/security/provider/SHA2";
8221 }
8222 break;
8223 case 3:
8224 if (UseSHA512Intrinsics) {
8225 // we want to do an instanceof comparison against the SHA5 class
8226 klass_name = "sun/security/provider/SHA5";
8227 }
8228 break;
8229 case 4:
8230 if (UseSHA3Intrinsics) {
8231 // we want to do an instanceof comparison against the SHA3 class
8232 klass_name = "sun/security/provider/SHA3";
8233 }
8234 break;
8235 default:
8236 fatal("unknown SHA intrinsic predicate: %d", predicate);
8237 }
8238
8239 ciKlass* klass = nullptr;
8240 if (klass_name != nullptr) {
8241 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8242 }
8243 if ((klass == nullptr) || !klass->is_loaded()) {
8244 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8245 Node* ctrl = control();
8246 set_control(top()); // no intrinsic path
8247 return ctrl;
8248 }
8249 ciInstanceKlass* instklass = klass->as_instance_klass();
8250
8251 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8252 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8253 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8254 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8255
8256 return instof_false; // even if it is null
8257 }
8258
8259 //-------------inline_fma-----------------------------------
8260 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8261 Node *a = nullptr;
8262 Node *b = nullptr;
8263 Node *c = nullptr;
8264 Node* result = nullptr;
8265 switch (id) {
8266 case vmIntrinsics::_fmaD:
8267 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8268 // no receiver since it is static method
8269 a = round_double_node(argument(0));
8270 b = round_double_node(argument(2));
8271 c = round_double_node(argument(4));
8272 result = _gvn.transform(new FmaDNode(control(), a, b, c));
8273 break;
8274 case vmIntrinsics::_fmaF:
8275 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8276 a = argument(0);
8277 b = argument(1);
8278 c = argument(2);
8279 result = _gvn.transform(new FmaFNode(control(), a, b, c));
8280 break;
8281 default:
8282 fatal_unexpected_iid(id); break;
8283 }
8284 set_result(result);
8285 return true;
8286 }
8287
8288 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8289 // argument(0) is receiver
8290 Node* codePoint = argument(1);
8291 Node* n = nullptr;
8292
8293 switch (id) {
8294 case vmIntrinsics::_isDigit :
8295 n = new DigitNode(control(), codePoint);
8296 break;
8297 case vmIntrinsics::_isLowerCase :
8298 n = new LowerCaseNode(control(), codePoint);
8299 break;
8300 case vmIntrinsics::_isUpperCase :
8301 n = new UpperCaseNode(control(), codePoint);
8302 break;
8303 case vmIntrinsics::_isWhitespace :
8304 n = new WhitespaceNode(control(), codePoint);
8305 break;
8306 default:
8307 fatal_unexpected_iid(id);
8308 }
8309
8310 set_result(_gvn.transform(n));
8311 return true;
8312 }
8313
8314 //------------------------------inline_fp_min_max------------------------------
8315 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
8316 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
8317
8318 // The intrinsic should be used only when the API branches aren't predictable,
8319 // the last one performing the most important comparison. The following heuristic
8320 // uses the branch statistics to eventually bail out if necessary.
8321
8322 ciMethodData *md = callee()->method_data();
8323
8324 if ( md != nullptr && md->is_mature() && md->invocation_count() > 0 ) {
8325 ciCallProfile cp = caller()->call_profile_at_bci(bci());
8326
8327 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
8328 // Bail out if the call-site didn't contribute enough to the statistics.
8329 return false;
8330 }
8331
8332 uint taken = 0, not_taken = 0;
8333
8334 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
8335 if (p->is_BranchData()) {
8336 taken = ((ciBranchData*)p)->taken();
8337 not_taken = ((ciBranchData*)p)->not_taken();
8338 }
8339 }
8340
8341 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
8342 balance = balance < 0 ? -balance : balance;
8343 if ( balance > 0.2 ) {
8344 // Bail out if the most important branch is predictable enough.
8345 return false;
8346 }
8347 }
8348 */
8349
8350 Node *a = nullptr;
8351 Node *b = nullptr;
8352 Node *n = nullptr;
8353 switch (id) {
8354 case vmIntrinsics::_maxF:
8355 case vmIntrinsics::_minF:
8356 case vmIntrinsics::_maxF_strict:
8357 case vmIntrinsics::_minF_strict:
8358 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
8359 a = argument(0);
8360 b = argument(1);
8361 break;
8362 case vmIntrinsics::_maxD:
8363 case vmIntrinsics::_minD:
8364 case vmIntrinsics::_maxD_strict:
8365 case vmIntrinsics::_minD_strict:
8366 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
8367 a = round_double_node(argument(0));
8368 b = round_double_node(argument(2));
8369 break;
8370 default:
8371 fatal_unexpected_iid(id);
8372 break;
8373 }
8374 switch (id) {
8375 case vmIntrinsics::_maxF:
8376 case vmIntrinsics::_maxF_strict:
8377 n = new MaxFNode(a, b);
8378 break;
8379 case vmIntrinsics::_minF:
8380 case vmIntrinsics::_minF_strict:
8381 n = new MinFNode(a, b);
8382 break;
8383 case vmIntrinsics::_maxD:
8384 case vmIntrinsics::_maxD_strict:
8385 n = new MaxDNode(a, b);
8386 break;
8387 case vmIntrinsics::_minD:
8388 case vmIntrinsics::_minD_strict:
8389 n = new MinDNode(a, b);
8390 break;
8391 default:
8392 fatal_unexpected_iid(id);
8393 break;
8394 }
8395 set_result(_gvn.transform(n));
8396 return true;
8397 }
8398
8399 bool LibraryCallKit::inline_profileBoolean() {
8400 Node* counts = argument(1);
8401 const TypeAryPtr* ary = nullptr;
8402 ciArray* aobj = nullptr;
8403 if (counts->is_Con()
8404 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8405 && (aobj = ary->const_oop()->as_array()) != nullptr
8406 && (aobj->length() == 2)) {
8407 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8408 jint false_cnt = aobj->element_value(0).as_int();
8409 jint true_cnt = aobj->element_value(1).as_int();
8410
8411 if (C->log() != nullptr) {
8412 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8413 false_cnt, true_cnt);
8414 }
8415
8416 if (false_cnt + true_cnt == 0) {
8417 // According to profile, never executed.
8418 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8419 Deoptimization::Action_reinterpret);
8420 return true;
8421 }
8422
8423 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8424 // is a number of each value occurrences.
8425 Node* result = argument(0);
8426 if (false_cnt == 0 || true_cnt == 0) {
8427 // According to profile, one value has been never seen.
8428 int expected_val = (false_cnt == 0) ? 1 : 0;
8429
8430 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8431 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8432
8433 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8434 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8435 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8436
8437 { // Slow path: uncommon trap for never seen value and then reexecute
8438 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8439 // the value has been seen at least once.
8440 PreserveJVMState pjvms(this);
8441 PreserveReexecuteState preexecs(this);
8442 jvms()->set_should_reexecute(true);
8443
8444 set_control(slow_path);
8445 set_i_o(i_o());
8446
8447 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8448 Deoptimization::Action_reinterpret);
8449 }
8450 // The guard for never seen value enables sharpening of the result and
8451 // returning a constant. It allows to eliminate branches on the same value
8452 // later on.
8453 set_control(fast_path);
8454 result = intcon(expected_val);
8455 }
8456 // Stop profiling.
8457 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8458 // By replacing method body with profile data (represented as ProfileBooleanNode
8459 // on IR level) we effectively disable profiling.
8460 // It enables full speed execution once optimized code is generated.
8461 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8462 C->record_for_igvn(profile);
8463 set_result(profile);
8464 return true;
8465 } else {
8466 // Continue profiling.
8467 // Profile data isn't available at the moment. So, execute method's bytecode version.
8468 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8469 // is compiled and counters aren't available since corresponding MethodHandle
8470 // isn't a compile-time constant.
8471 return false;
8472 }
8473 }
8474
8475 bool LibraryCallKit::inline_isCompileConstant() {
8476 Node* n = argument(0);
8477 set_result(n->is_Con() ? intcon(1) : intcon(0));
8478 return true;
8479 }
8480
8481 //------------------------------- inline_getObjectSize --------------------------------------
8482 //
8483 // Calculate the runtime size of the object/array.
8484 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8485 //
8486 bool LibraryCallKit::inline_getObjectSize() {
8487 Node* obj = argument(3);
8488 Node* klass_node = load_object_klass(obj);
8489
8490 jint layout_con = Klass::_lh_neutral_value;
8491 Node* layout_val = get_layout_helper(klass_node, layout_con);
8492 int layout_is_con = (layout_val == nullptr);
8493
8494 if (layout_is_con) {
8495 // Layout helper is constant, can figure out things at compile time.
8496
8497 if (Klass::layout_helper_is_instance(layout_con)) {
8498 // Instance case: layout_con contains the size itself.
8499 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8500 set_result(size);
8501 } else {
8502 // Array case: size is round(header + element_size*arraylength).
8503 // Since arraylength is different for every array instance, we have to
8504 // compute the whole thing at runtime.
8505
8506 Node* arr_length = load_array_length(obj);
8507
8508 int round_mask = MinObjAlignmentInBytes - 1;
8509 int hsize = Klass::layout_helper_header_size(layout_con);
8510 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8511
8512 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8513 round_mask = 0; // strength-reduce it if it goes away completely
8514 }
8515 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8516 Node* header_size = intcon(hsize + round_mask);
8517
8518 Node* lengthx = ConvI2X(arr_length);
8519 Node* headerx = ConvI2X(header_size);
8520
8521 Node* abody = lengthx;
8522 if (eshift != 0) {
8523 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8524 }
8525 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8526 if (round_mask != 0) {
8527 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8528 }
8529 size = ConvX2L(size);
8530 set_result(size);
8531 }
8532 } else {
8533 // Layout helper is not constant, need to test for array-ness at runtime.
8534
8535 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8536 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8537 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8538 record_for_igvn(result_reg);
8539
8540 Node* array_ctl = generate_array_guard(klass_node, nullptr);
8541 if (array_ctl != nullptr) {
8542 // Array case: size is round(header + element_size*arraylength).
8543 // Since arraylength is different for every array instance, we have to
8544 // compute the whole thing at runtime.
8545
8546 PreserveJVMState pjvms(this);
8547 set_control(array_ctl);
8548 Node* arr_length = load_array_length(obj);
8549
8550 int round_mask = MinObjAlignmentInBytes - 1;
8551 Node* mask = intcon(round_mask);
8552
8553 Node* hss = intcon(Klass::_lh_header_size_shift);
8554 Node* hsm = intcon(Klass::_lh_header_size_mask);
8555 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8556 header_size = _gvn.transform(new AndINode(header_size, hsm));
8557 header_size = _gvn.transform(new AddINode(header_size, mask));
8558
8559 // There is no need to mask or shift this value.
8560 // The semantics of LShiftINode include an implicit mask to 0x1F.
8561 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8562 Node* elem_shift = layout_val;
8563
8564 Node* lengthx = ConvI2X(arr_length);
8565 Node* headerx = ConvI2X(header_size);
8566
8567 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
8568 Node* size = _gvn.transform(new AddXNode(headerx, abody));
8569 if (round_mask != 0) {
8570 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
8571 }
8572 size = ConvX2L(size);
8573
8574 result_reg->init_req(_array_path, control());
8575 result_val->init_req(_array_path, size);
8576 }
8577
8578 if (!stopped()) {
8579 // Instance case: the layout helper gives us instance size almost directly,
8580 // but we need to mask out the _lh_instance_slow_path_bit.
8581 Node* size = ConvI2X(layout_val);
8582 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
8583 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
8584 size = _gvn.transform(new AndXNode(size, mask));
8585 size = ConvX2L(size);
8586
8587 result_reg->init_req(_instance_path, control());
8588 result_val->init_req(_instance_path, size);
8589 }
8590
8591 set_result(result_reg, result_val);
8592 }
8593
8594 return true;
8595 }
8596
8597 //------------------------------- inline_blackhole --------------------------------------
8598 //
8599 // Make sure all arguments to this node are alive.
8600 // This matches methods that were requested to be blackholed through compile commands.
8601 //
8602 bool LibraryCallKit::inline_blackhole() {
8603 assert(callee()->is_static(), "Should have been checked before: only static methods here");
8604 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
8605 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
8606
8607 // Blackhole node pinches only the control, not memory. This allows
8608 // the blackhole to be pinned in the loop that computes blackholed
8609 // values, but have no other side effects, like breaking the optimizations
8610 // across the blackhole.
8611
8612 Node* bh = _gvn.transform(new BlackholeNode(control()));
8613 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
8614
8615 // Bind call arguments as blackhole arguments to keep them alive
8616 uint nargs = callee()->arg_size();
8617 for (uint i = 0; i < nargs; i++) {
8618 bh->add_req(argument(i));
8619 }
8620
8621 return true;
8622 }