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