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/ciFlatArrayKlass.hpp"
28 #include "ci/ciUtilities.inline.hpp"
29 #include "classfile/vmIntrinsics.hpp"
30 #include "compiler/compileBroker.hpp"
31 #include "compiler/compileLog.hpp"
32 #include "gc/shared/barrierSet.hpp"
33 #include "jfr/support/jfrIntrinsics.hpp"
34 #include "memory/resourceArea.hpp"
35 #include "oops/klass.inline.hpp"
36 #include "oops/objArrayKlass.hpp"
37 #include "opto/addnode.hpp"
38 #include "opto/arraycopynode.hpp"
39 #include "opto/c2compiler.hpp"
40 #include "opto/castnode.hpp"
41 #include "opto/cfgnode.hpp"
42 #include "opto/convertnode.hpp"
43 #include "opto/countbitsnode.hpp"
44 #include "opto/idealKit.hpp"
45 #include "opto/library_call.hpp"
46 #include "opto/mathexactnode.hpp"
47 #include "opto/mulnode.hpp"
48 #include "opto/narrowptrnode.hpp"
49 #include "opto/opaquenode.hpp"
50 #include "opto/parse.hpp"
51 #include "opto/runtime.hpp"
52 #include "opto/rootnode.hpp"
53 #include "opto/subnode.hpp"
54 #include "prims/jvmtiExport.hpp"
55 #include "prims/jvmtiThreadState.hpp"
56 #include "prims/unsafe.hpp"
57 #include "runtime/jniHandles.inline.hpp"
58 #include "runtime/objectMonitor.hpp"
59 #include "runtime/sharedRuntime.hpp"
60 #include "runtime/stubRoutines.hpp"
61 #include "utilities/macros.hpp"
62 #include "utilities/powerOfTwo.hpp"
63
64 //---------------------------make_vm_intrinsic----------------------------
65 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
66 vmIntrinsicID id = m->intrinsic_id();
67 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
68
69 if (!m->is_loaded()) {
70 // Do not attempt to inline unloaded methods.
71 return nullptr;
72 }
73
74 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
75 bool is_available = false;
76
77 {
78 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
79 // the compiler must transition to '_thread_in_vm' state because both
80 // methods access VM-internal data.
81 VM_ENTRY_MARK;
82 methodHandle mh(THREAD, m->get_Method());
83 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
84 if (is_available && is_virtual) {
85 is_available = vmIntrinsics::does_virtual_dispatch(id);
86 }
87 }
88
89 if (is_available) {
90 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
91 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
92 return new LibraryIntrinsic(m, is_virtual,
93 vmIntrinsics::predicates_needed(id),
94 vmIntrinsics::does_virtual_dispatch(id),
95 id);
96 } else {
97 return nullptr;
98 }
99 }
100
101 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
102 LibraryCallKit kit(jvms, this);
103 Compile* C = kit.C;
104 int nodes = C->unique();
105 #ifndef PRODUCT
106 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
107 char buf[1000];
108 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
109 tty->print_cr("Intrinsic %s", str);
110 }
111 #endif
112 ciMethod* callee = kit.callee();
113 const int bci = kit.bci();
114 #ifdef ASSERT
115 Node* ctrl = kit.control();
116 #endif
117 // Try to inline the intrinsic.
118 if (callee->check_intrinsic_candidate() &&
119 kit.try_to_inline(_last_predicate)) {
120 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
121 : "(intrinsic)";
122 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
123 if (C->print_intrinsics() || C->print_inlining()) {
124 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
125 }
126 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
127 if (C->log()) {
128 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
129 vmIntrinsics::name_at(intrinsic_id()),
130 (is_virtual() ? " virtual='1'" : ""),
131 C->unique() - nodes);
132 }
133 // Push the result from the inlined method onto the stack.
134 kit.push_result();
135 C->print_inlining_update(this);
136 return kit.transfer_exceptions_into_jvms();
137 }
138
139 // The intrinsic bailed out
140 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
141 if (jvms->has_method()) {
142 // Not a root compile.
143 const char* msg;
144 if (callee->intrinsic_candidate()) {
145 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
146 } else {
147 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
148 : "failed to inline (intrinsic), method not annotated";
149 }
150 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
151 if (C->print_intrinsics() || C->print_inlining()) {
152 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
153 }
154 } else {
155 // Root compile
156 ResourceMark rm;
157 stringStream msg_stream;
158 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
159 vmIntrinsics::name_at(intrinsic_id()),
160 is_virtual() ? " (virtual)" : "", bci);
161 const char *msg = msg_stream.freeze();
162 log_debug(jit, inlining)("%s", msg);
163 if (C->print_intrinsics() || C->print_inlining()) {
164 tty->print("%s", msg);
165 }
166 }
167 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
168 C->print_inlining_update(this);
169
170 return nullptr;
171 }
172
173 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
174 LibraryCallKit kit(jvms, this);
175 Compile* C = kit.C;
176 int nodes = C->unique();
177 _last_predicate = predicate;
178 #ifndef PRODUCT
179 assert(is_predicated() && predicate < predicates_count(), "sanity");
180 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
181 char buf[1000];
182 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
183 tty->print_cr("Predicate for intrinsic %s", str);
184 }
185 #endif
186 ciMethod* callee = kit.callee();
187 const int bci = kit.bci();
188
189 Node* slow_ctl = kit.try_to_predicate(predicate);
190 if (!kit.failing()) {
191 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
192 : "(intrinsic, predicate)";
193 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
194 if (C->print_intrinsics() || C->print_inlining()) {
195 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
196 }
197 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
198 if (C->log()) {
199 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
200 vmIntrinsics::name_at(intrinsic_id()),
201 (is_virtual() ? " virtual='1'" : ""),
202 C->unique() - nodes);
203 }
204 return slow_ctl; // Could be null if the check folds.
205 }
206
207 // The intrinsic bailed out
208 if (jvms->has_method()) {
209 // Not a root compile.
210 const char* msg = "failed to generate predicate for intrinsic";
211 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
212 if (C->print_intrinsics() || C->print_inlining()) {
213 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
214 }
215 } else {
216 // Root compile
217 ResourceMark rm;
218 stringStream msg_stream;
219 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
220 vmIntrinsics::name_at(intrinsic_id()),
221 is_virtual() ? " (virtual)" : "", bci);
222 const char *msg = msg_stream.freeze();
223 log_debug(jit, inlining)("%s", msg);
224 if (C->print_intrinsics() || C->print_inlining()) {
225 C->print_inlining_stream()->print("%s", msg);
226 }
227 }
228 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
229 return nullptr;
230 }
231
232 bool LibraryCallKit::try_to_inline(int predicate) {
233 // Handle symbolic names for otherwise undistinguished boolean switches:
234 const bool is_store = true;
235 const bool is_compress = true;
236 const bool is_static = true;
237 const bool is_volatile = true;
238
239 if (!jvms()->has_method()) {
240 // Root JVMState has a null method.
241 assert(map()->memory()->Opcode() == Op_Parm, "");
242 // Insert the memory aliasing node
243 set_all_memory(reset_memory());
244 }
245 assert(merged_memory(), "");
246
247 switch (intrinsic_id()) {
248 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
249 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
250 case vmIntrinsics::_getClass: return inline_native_getClass();
251
252 case vmIntrinsics::_ceil:
253 case vmIntrinsics::_floor:
254 case vmIntrinsics::_rint:
255 case vmIntrinsics::_dsin:
256 case vmIntrinsics::_dcos:
257 case vmIntrinsics::_dtan:
258 case vmIntrinsics::_dabs:
259 case vmIntrinsics::_fabs:
260 case vmIntrinsics::_iabs:
261 case vmIntrinsics::_labs:
262 case vmIntrinsics::_datan2:
263 case vmIntrinsics::_dsqrt:
264 case vmIntrinsics::_dsqrt_strict:
265 case vmIntrinsics::_dexp:
266 case vmIntrinsics::_dlog:
267 case vmIntrinsics::_dlog10:
268 case vmIntrinsics::_dpow:
269 case vmIntrinsics::_dcopySign:
270 case vmIntrinsics::_fcopySign:
271 case vmIntrinsics::_dsignum:
272 case vmIntrinsics::_roundF:
273 case vmIntrinsics::_roundD:
274 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
275
276 case vmIntrinsics::_notify:
277 case vmIntrinsics::_notifyAll:
278 return inline_notify(intrinsic_id());
279
280 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
281 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
282 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
283 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
284 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
285 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
286 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
287 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
288 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
289 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
290 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
291 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
292 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
293 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
294
295 case vmIntrinsics::_arraycopy: return inline_arraycopy();
296
297 case vmIntrinsics::_arraySort: return inline_array_sort();
298 case vmIntrinsics::_arrayPartition: return inline_array_partition();
299
300 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
301 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
302 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
303 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
304
305 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
306 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
307 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
308 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
309 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
310 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
311 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
312 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
313
314 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
315 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
316
317 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
318
319 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
320 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
321 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
322 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
323
324 case vmIntrinsics::_compressStringC:
325 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
326 case vmIntrinsics::_inflateStringC:
327 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
328
329 case vmIntrinsics::_makePrivateBuffer: return inline_unsafe_make_private_buffer();
330 case vmIntrinsics::_finishPrivateBuffer: return inline_unsafe_finish_private_buffer();
331 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
332 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
333 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
334 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
335 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
336 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
337 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
338 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
339 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
340 case vmIntrinsics::_getValue: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false, true);
341
342 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
343 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
344 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
345 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
346 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
347 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
348 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
349 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
350 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
351 case vmIntrinsics::_putValue: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false, true);
352
353 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
354 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
355 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
356 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
357 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
358 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
359 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
360 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
361 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
362
363 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
364 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
365 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
366 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
367 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
368 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
369 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
370 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
371 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
372
373 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
374 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
375 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
376 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
377
378 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
379 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
380 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
381 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
382
383 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
384 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
385 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
386 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
387 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
388 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
389 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
390 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
391 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
392
393 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
394 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
395 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
396 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
397 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
398 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
399 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
400 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
401 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
402
403 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
404 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
405 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
406 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
407 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
408 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
409 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
410 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
411 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
412
413 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
414 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
415 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
416 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
417 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
418 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
419 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
420 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
421 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
422
423 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
424 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
425 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
426 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
427 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
428
429 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
430 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
431 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
432 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
433 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
434 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
435 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
436 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
437 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
438 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
439 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
440 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
441 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
442 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
443 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
444 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
445 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
446 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
447 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
448 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
449
450 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
451 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
452 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
453 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
454 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
455 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
456 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
457 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
458 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
459 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
460 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
461 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
462 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
463 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
464 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
465
466 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
467 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
468 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
469 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
470
471 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
472 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
473 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
474 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
475 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
476
477 case vmIntrinsics::_loadFence:
478 case vmIntrinsics::_storeFence:
479 case vmIntrinsics::_storeStoreFence:
480 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
481
482 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
483
484 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
485 case vmIntrinsics::_currentThread: return inline_native_currentThread();
486 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
487
488 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
489 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
490
491 #if INCLUDE_JVMTI
492 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
493 "notifyJvmtiStart", true, false);
494 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
495 "notifyJvmtiEnd", false, true);
496 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
497 "notifyJvmtiMount", false, false);
498 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
499 "notifyJvmtiUnmount", false, false);
500 case vmIntrinsics::_notifyJvmtiVThreadHideFrames: return inline_native_notify_jvmti_hide();
501 #endif
502
503 #ifdef JFR_HAVE_INTRINSICS
504 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
505 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
506 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
507 #endif
508 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
509 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
510 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
511 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
512 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
513 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
514 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
515 case vmIntrinsics::_isFlatArray: return inline_unsafe_isFlatArray();
516 case vmIntrinsics::_getLength: return inline_native_getLength();
517 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
518 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
519 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
520 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
521 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
522 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
523 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
524
525 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
526 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
527 case vmIntrinsics::_newNullRestrictedArray: return inline_newNullRestrictedArray();
528
529 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
530
531 case vmIntrinsics::_isInstance:
532 case vmIntrinsics::_getModifiers:
533 case vmIntrinsics::_isInterface:
534 case vmIntrinsics::_isArray:
535 case vmIntrinsics::_isPrimitive:
536 case vmIntrinsics::_isHidden:
537 case vmIntrinsics::_getSuperclass:
538 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
539
540 case vmIntrinsics::_floatToRawIntBits:
541 case vmIntrinsics::_floatToIntBits:
542 case vmIntrinsics::_intBitsToFloat:
543 case vmIntrinsics::_doubleToRawLongBits:
544 case vmIntrinsics::_doubleToLongBits:
545 case vmIntrinsics::_longBitsToDouble:
546 case vmIntrinsics::_floatToFloat16:
547 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
548
549 case vmIntrinsics::_floatIsFinite:
550 case vmIntrinsics::_floatIsInfinite:
551 case vmIntrinsics::_doubleIsFinite:
552 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
553
554 case vmIntrinsics::_numberOfLeadingZeros_i:
555 case vmIntrinsics::_numberOfLeadingZeros_l:
556 case vmIntrinsics::_numberOfTrailingZeros_i:
557 case vmIntrinsics::_numberOfTrailingZeros_l:
558 case vmIntrinsics::_bitCount_i:
559 case vmIntrinsics::_bitCount_l:
560 case vmIntrinsics::_reverse_i:
561 case vmIntrinsics::_reverse_l:
562 case vmIntrinsics::_reverseBytes_i:
563 case vmIntrinsics::_reverseBytes_l:
564 case vmIntrinsics::_reverseBytes_s:
565 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
566
567 case vmIntrinsics::_compress_i:
568 case vmIntrinsics::_compress_l:
569 case vmIntrinsics::_expand_i:
570 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
571
572 case vmIntrinsics::_compareUnsigned_i:
573 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
574
575 case vmIntrinsics::_divideUnsigned_i:
576 case vmIntrinsics::_divideUnsigned_l:
577 case vmIntrinsics::_remainderUnsigned_i:
578 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
579
580 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
581
582 case vmIntrinsics::_Reference_get: return inline_reference_get();
583 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
584 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
585
586 case vmIntrinsics::_Class_cast: return inline_Class_cast();
587
588 case vmIntrinsics::_aescrypt_encryptBlock:
589 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
590
591 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
592 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
593 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
594
595 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
596 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
597 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
598
599 case vmIntrinsics::_counterMode_AESCrypt:
600 return inline_counterMode_AESCrypt(intrinsic_id());
601
602 case vmIntrinsics::_galoisCounterMode_AESCrypt:
603 return inline_galoisCounterMode_AESCrypt();
604
605 case vmIntrinsics::_md5_implCompress:
606 case vmIntrinsics::_sha_implCompress:
607 case vmIntrinsics::_sha2_implCompress:
608 case vmIntrinsics::_sha5_implCompress:
609 case vmIntrinsics::_sha3_implCompress:
610 return inline_digestBase_implCompress(intrinsic_id());
611
612 case vmIntrinsics::_digestBase_implCompressMB:
613 return inline_digestBase_implCompressMB(predicate);
614
615 case vmIntrinsics::_multiplyToLen:
616 return inline_multiplyToLen();
617
618 case vmIntrinsics::_squareToLen:
619 return inline_squareToLen();
620
621 case vmIntrinsics::_mulAdd:
622 return inline_mulAdd();
623
624 case vmIntrinsics::_montgomeryMultiply:
625 return inline_montgomeryMultiply();
626 case vmIntrinsics::_montgomerySquare:
627 return inline_montgomerySquare();
628
629 case vmIntrinsics::_bigIntegerRightShiftWorker:
630 return inline_bigIntegerShift(true);
631 case vmIntrinsics::_bigIntegerLeftShiftWorker:
632 return inline_bigIntegerShift(false);
633
634 case vmIntrinsics::_vectorizedMismatch:
635 return inline_vectorizedMismatch();
636
637 case vmIntrinsics::_ghash_processBlocks:
638 return inline_ghash_processBlocks();
639 case vmIntrinsics::_chacha20Block:
640 return inline_chacha20Block();
641 case vmIntrinsics::_base64_encodeBlock:
642 return inline_base64_encodeBlock();
643 case vmIntrinsics::_base64_decodeBlock:
644 return inline_base64_decodeBlock();
645 case vmIntrinsics::_poly1305_processBlocks:
646 return inline_poly1305_processBlocks();
647
648 case vmIntrinsics::_encodeISOArray:
649 case vmIntrinsics::_encodeByteISOArray:
650 return inline_encodeISOArray(false);
651 case vmIntrinsics::_encodeAsciiArray:
652 return inline_encodeISOArray(true);
653
654 case vmIntrinsics::_updateCRC32:
655 return inline_updateCRC32();
656 case vmIntrinsics::_updateBytesCRC32:
657 return inline_updateBytesCRC32();
658 case vmIntrinsics::_updateByteBufferCRC32:
659 return inline_updateByteBufferCRC32();
660
661 case vmIntrinsics::_updateBytesCRC32C:
662 return inline_updateBytesCRC32C();
663 case vmIntrinsics::_updateDirectByteBufferCRC32C:
664 return inline_updateDirectByteBufferCRC32C();
665
666 case vmIntrinsics::_updateBytesAdler32:
667 return inline_updateBytesAdler32();
668 case vmIntrinsics::_updateByteBufferAdler32:
669 return inline_updateByteBufferAdler32();
670
671 case vmIntrinsics::_profileBoolean:
672 return inline_profileBoolean();
673 case vmIntrinsics::_isCompileConstant:
674 return inline_isCompileConstant();
675
676 case vmIntrinsics::_countPositives:
677 return inline_countPositives();
678
679 case vmIntrinsics::_fmaD:
680 case vmIntrinsics::_fmaF:
681 return inline_fma(intrinsic_id());
682
683 case vmIntrinsics::_isDigit:
684 case vmIntrinsics::_isLowerCase:
685 case vmIntrinsics::_isUpperCase:
686 case vmIntrinsics::_isWhitespace:
687 return inline_character_compare(intrinsic_id());
688
689 case vmIntrinsics::_min:
690 case vmIntrinsics::_max:
691 case vmIntrinsics::_min_strict:
692 case vmIntrinsics::_max_strict:
693 return inline_min_max(intrinsic_id());
694
695 case vmIntrinsics::_maxF:
696 case vmIntrinsics::_minF:
697 case vmIntrinsics::_maxD:
698 case vmIntrinsics::_minD:
699 case vmIntrinsics::_maxF_strict:
700 case vmIntrinsics::_minF_strict:
701 case vmIntrinsics::_maxD_strict:
702 case vmIntrinsics::_minD_strict:
703 return inline_fp_min_max(intrinsic_id());
704
705 case vmIntrinsics::_VectorUnaryOp:
706 return inline_vector_nary_operation(1);
707 case vmIntrinsics::_VectorBinaryOp:
708 return inline_vector_nary_operation(2);
709 case vmIntrinsics::_VectorTernaryOp:
710 return inline_vector_nary_operation(3);
711 case vmIntrinsics::_VectorFromBitsCoerced:
712 return inline_vector_frombits_coerced();
713 case vmIntrinsics::_VectorShuffleIota:
714 return inline_vector_shuffle_iota();
715 case vmIntrinsics::_VectorMaskOp:
716 return inline_vector_mask_operation();
717 case vmIntrinsics::_VectorShuffleToVector:
718 return inline_vector_shuffle_to_vector();
719 case vmIntrinsics::_VectorLoadOp:
720 return inline_vector_mem_operation(/*is_store=*/false);
721 case vmIntrinsics::_VectorLoadMaskedOp:
722 return inline_vector_mem_masked_operation(/*is_store*/false);
723 case vmIntrinsics::_VectorStoreOp:
724 return inline_vector_mem_operation(/*is_store=*/true);
725 case vmIntrinsics::_VectorStoreMaskedOp:
726 return inline_vector_mem_masked_operation(/*is_store=*/true);
727 case vmIntrinsics::_VectorGatherOp:
728 return inline_vector_gather_scatter(/*is_scatter*/ false);
729 case vmIntrinsics::_VectorScatterOp:
730 return inline_vector_gather_scatter(/*is_scatter*/ true);
731 case vmIntrinsics::_VectorReductionCoerced:
732 return inline_vector_reduction();
733 case vmIntrinsics::_VectorTest:
734 return inline_vector_test();
735 case vmIntrinsics::_VectorBlend:
736 return inline_vector_blend();
737 case vmIntrinsics::_VectorRearrange:
738 return inline_vector_rearrange();
739 case vmIntrinsics::_VectorCompare:
740 return inline_vector_compare();
741 case vmIntrinsics::_VectorBroadcastInt:
742 return inline_vector_broadcast_int();
743 case vmIntrinsics::_VectorConvert:
744 return inline_vector_convert();
745 case vmIntrinsics::_VectorInsert:
746 return inline_vector_insert();
747 case vmIntrinsics::_VectorExtract:
748 return inline_vector_extract();
749 case vmIntrinsics::_VectorCompressExpand:
750 return inline_vector_compress_expand();
751 case vmIntrinsics::_IndexVector:
752 return inline_index_vector();
753 case vmIntrinsics::_IndexPartiallyInUpperRange:
754 return inline_index_partially_in_upper_range();
755
756 case vmIntrinsics::_getObjectSize:
757 return inline_getObjectSize();
758
759 case vmIntrinsics::_blackhole:
760 return inline_blackhole();
761
762 default:
763 // If you get here, it may be that someone has added a new intrinsic
764 // to the list in vmIntrinsics.hpp without implementing it here.
765 #ifndef PRODUCT
766 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
767 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
768 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
769 }
770 #endif
771 return false;
772 }
773 }
774
775 Node* LibraryCallKit::try_to_predicate(int predicate) {
776 if (!jvms()->has_method()) {
777 // Root JVMState has a null method.
778 assert(map()->memory()->Opcode() == Op_Parm, "");
779 // Insert the memory aliasing node
780 set_all_memory(reset_memory());
781 }
782 assert(merged_memory(), "");
783
784 switch (intrinsic_id()) {
785 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
786 return inline_cipherBlockChaining_AESCrypt_predicate(false);
787 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
788 return inline_cipherBlockChaining_AESCrypt_predicate(true);
789 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
790 return inline_electronicCodeBook_AESCrypt_predicate(false);
791 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
792 return inline_electronicCodeBook_AESCrypt_predicate(true);
793 case vmIntrinsics::_counterMode_AESCrypt:
794 return inline_counterMode_AESCrypt_predicate();
795 case vmIntrinsics::_digestBase_implCompressMB:
796 return inline_digestBase_implCompressMB_predicate(predicate);
797 case vmIntrinsics::_galoisCounterMode_AESCrypt:
798 return inline_galoisCounterMode_AESCrypt_predicate();
799
800 default:
801 // If you get here, it may be that someone has added a new intrinsic
802 // to the list in vmIntrinsics.hpp without implementing it here.
803 #ifndef PRODUCT
804 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
805 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
806 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
807 }
808 #endif
809 Node* slow_ctl = control();
810 set_control(top()); // No fast path intrinsic
811 return slow_ctl;
812 }
813 }
814
815 //------------------------------set_result-------------------------------
816 // Helper function for finishing intrinsics.
817 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
818 record_for_igvn(region);
819 set_control(_gvn.transform(region));
820 set_result( _gvn.transform(value));
821 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
822 }
823
824 //------------------------------generate_guard---------------------------
825 // Helper function for generating guarded fast-slow graph structures.
826 // The given 'test', if true, guards a slow path. If the test fails
827 // then a fast path can be taken. (We generally hope it fails.)
828 // In all cases, GraphKit::control() is updated to the fast path.
829 // The returned value represents the control for the slow path.
830 // The return value is never 'top'; it is either a valid control
831 // or null if it is obvious that the slow path can never be taken.
832 // Also, if region and the slow control are not null, the slow edge
833 // is appended to the region.
834 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
835 if (stopped()) {
836 // Already short circuited.
837 return nullptr;
838 }
839
840 // Build an if node and its projections.
841 // If test is true we take the slow path, which we assume is uncommon.
842 if (_gvn.type(test) == TypeInt::ZERO) {
843 // The slow branch is never taken. No need to build this guard.
844 return nullptr;
845 }
846
847 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
848
849 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
850 if (if_slow == top()) {
851 // The slow branch is never taken. No need to build this guard.
852 return nullptr;
853 }
854
855 if (region != nullptr)
856 region->add_req(if_slow);
857
858 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
859 set_control(if_fast);
860
861 return if_slow;
862 }
863
864 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
865 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
866 }
867 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
868 return generate_guard(test, region, PROB_FAIR);
869 }
870
871 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
872 Node* *pos_index) {
873 if (stopped())
874 return nullptr; // already stopped
875 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
876 return nullptr; // index is already adequately typed
877 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
878 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
879 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
880 if (is_neg != nullptr && pos_index != nullptr) {
881 // Emulate effect of Parse::adjust_map_after_if.
882 Node* ccast = new CastIINode(index, TypeInt::POS);
883 ccast->set_req(0, control());
884 (*pos_index) = _gvn.transform(ccast);
885 }
886 return is_neg;
887 }
888
889 // Make sure that 'position' is a valid limit index, in [0..length].
890 // There are two equivalent plans for checking this:
891 // A. (offset + copyLength) unsigned<= arrayLength
892 // B. offset <= (arrayLength - copyLength)
893 // We require that all of the values above, except for the sum and
894 // difference, are already known to be non-negative.
895 // Plan A is robust in the face of overflow, if offset and copyLength
896 // are both hugely positive.
897 //
898 // Plan B is less direct and intuitive, but it does not overflow at
899 // all, since the difference of two non-negatives is always
900 // representable. Whenever Java methods must perform the equivalent
901 // check they generally use Plan B instead of Plan A.
902 // For the moment we use Plan A.
903 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
904 Node* subseq_length,
905 Node* array_length,
906 RegionNode* region) {
907 if (stopped())
908 return nullptr; // already stopped
909 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
910 if (zero_offset && subseq_length->eqv_uncast(array_length))
911 return nullptr; // common case of whole-array copy
912 Node* last = subseq_length;
913 if (!zero_offset) // last += offset
914 last = _gvn.transform(new AddINode(last, offset));
915 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
916 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
917 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
918 return is_over;
919 }
920
921 // Emit range checks for the given String.value byte array
922 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
923 if (stopped()) {
924 return; // already stopped
925 }
926 RegionNode* bailout = new RegionNode(1);
927 record_for_igvn(bailout);
928 if (char_count) {
929 // Convert char count to byte count
930 count = _gvn.transform(new LShiftINode(count, intcon(1)));
931 }
932
933 // Offset and count must not be negative
934 generate_negative_guard(offset, bailout);
935 generate_negative_guard(count, bailout);
936 // Offset + count must not exceed length of array
937 generate_limit_guard(offset, count, load_array_length(array), bailout);
938
939 if (bailout->req() > 1) {
940 PreserveJVMState pjvms(this);
941 set_control(_gvn.transform(bailout));
942 uncommon_trap(Deoptimization::Reason_intrinsic,
943 Deoptimization::Action_maybe_recompile);
944 }
945 }
946
947 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
948 bool is_immutable) {
949 ciKlass* thread_klass = env()->Thread_klass();
950 const Type* thread_type
951 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
952
953 Node* thread = _gvn.transform(new ThreadLocalNode());
954 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
955 tls_output = thread;
956
957 Node* thread_obj_handle
958 = (is_immutable
959 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
960 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
961 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
962 thread_obj_handle = _gvn.transform(thread_obj_handle);
963
964 DecoratorSet decorators = IN_NATIVE;
965 if (is_immutable) {
966 decorators |= C2_IMMUTABLE_MEMORY;
967 }
968 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
969 }
970
971 //--------------------------generate_current_thread--------------------
972 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
973 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
974 /*is_immutable*/false);
975 }
976
977 //--------------------------generate_virtual_thread--------------------
978 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
979 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
980 !C->method()->changes_current_thread());
981 }
982
983 //------------------------------make_string_method_node------------------------
984 // Helper method for String intrinsic functions. This version is called with
985 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
986 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
987 // containing the lengths of str1 and str2.
988 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
989 Node* result = nullptr;
990 switch (opcode) {
991 case Op_StrIndexOf:
992 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
993 str1_start, cnt1, str2_start, cnt2, ae);
994 break;
995 case Op_StrComp:
996 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
997 str1_start, cnt1, str2_start, cnt2, ae);
998 break;
999 case Op_StrEquals:
1000 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1001 // Use the constant length if there is one because optimized match rule may exist.
1002 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1003 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1004 break;
1005 default:
1006 ShouldNotReachHere();
1007 return nullptr;
1008 }
1009
1010 // All these intrinsics have checks.
1011 C->set_has_split_ifs(true); // Has chance for split-if optimization
1012 clear_upper_avx();
1013
1014 return _gvn.transform(result);
1015 }
1016
1017 //------------------------------inline_string_compareTo------------------------
1018 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1019 Node* arg1 = argument(0);
1020 Node* arg2 = argument(1);
1021
1022 arg1 = must_be_not_null(arg1, true);
1023 arg2 = must_be_not_null(arg2, true);
1024
1025 // Get start addr and length of first argument
1026 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1027 Node* arg1_cnt = load_array_length(arg1);
1028
1029 // Get start addr and length of second argument
1030 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1031 Node* arg2_cnt = load_array_length(arg2);
1032
1033 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1034 set_result(result);
1035 return true;
1036 }
1037
1038 //------------------------------inline_string_equals------------------------
1039 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1040 Node* arg1 = argument(0);
1041 Node* arg2 = argument(1);
1042
1043 // paths (plus control) merge
1044 RegionNode* region = new RegionNode(3);
1045 Node* phi = new PhiNode(region, TypeInt::BOOL);
1046
1047 if (!stopped()) {
1048
1049 arg1 = must_be_not_null(arg1, true);
1050 arg2 = must_be_not_null(arg2, true);
1051
1052 // Get start addr and length of first argument
1053 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1054 Node* arg1_cnt = load_array_length(arg1);
1055
1056 // Get start addr and length of second argument
1057 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1058 Node* arg2_cnt = load_array_length(arg2);
1059
1060 // Check for arg1_cnt != arg2_cnt
1061 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1062 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1063 Node* if_ne = generate_slow_guard(bol, nullptr);
1064 if (if_ne != nullptr) {
1065 phi->init_req(2, intcon(0));
1066 region->init_req(2, if_ne);
1067 }
1068
1069 // Check for count == 0 is done by assembler code for StrEquals.
1070
1071 if (!stopped()) {
1072 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1073 phi->init_req(1, equals);
1074 region->init_req(1, control());
1075 }
1076 }
1077
1078 // post merge
1079 set_control(_gvn.transform(region));
1080 record_for_igvn(region);
1081
1082 set_result(_gvn.transform(phi));
1083 return true;
1084 }
1085
1086 //------------------------------inline_array_equals----------------------------
1087 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1088 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1089 Node* arg1 = argument(0);
1090 Node* arg2 = argument(1);
1091
1092 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1093 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1094 clear_upper_avx();
1095
1096 return true;
1097 }
1098
1099
1100 //------------------------------inline_countPositives------------------------------
1101 bool LibraryCallKit::inline_countPositives() {
1102 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1103 return false;
1104 }
1105
1106 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1107 // no receiver since it is static method
1108 Node* ba = argument(0);
1109 Node* offset = argument(1);
1110 Node* len = argument(2);
1111
1112 ba = must_be_not_null(ba, true);
1113
1114 // Range checks
1115 generate_string_range_check(ba, offset, len, false);
1116 if (stopped()) {
1117 return true;
1118 }
1119 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1120 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1121 set_result(_gvn.transform(result));
1122 clear_upper_avx();
1123 return true;
1124 }
1125
1126 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1127 Node* index = argument(0);
1128 Node* length = bt == T_INT ? argument(1) : argument(2);
1129 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1130 return false;
1131 }
1132
1133 // check that length is positive
1134 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1135 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1136
1137 {
1138 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1139 uncommon_trap(Deoptimization::Reason_intrinsic,
1140 Deoptimization::Action_make_not_entrant);
1141 }
1142
1143 if (stopped()) {
1144 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1145 return true;
1146 }
1147
1148 // length is now known positive, add a cast node to make this explicit
1149 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1150 Node* casted_length = ConstraintCastNode::make(control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), ConstraintCastNode::RegularDependency, bt);
1151 casted_length = _gvn.transform(casted_length);
1152 replace_in_map(length, casted_length);
1153 length = casted_length;
1154
1155 // Use an unsigned comparison for the range check itself
1156 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1157 BoolTest::mask btest = BoolTest::lt;
1158 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1159 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1160 _gvn.set_type(rc, rc->Value(&_gvn));
1161 if (!rc_bool->is_Con()) {
1162 record_for_igvn(rc);
1163 }
1164 set_control(_gvn.transform(new IfTrueNode(rc)));
1165 {
1166 PreserveJVMState pjvms(this);
1167 set_control(_gvn.transform(new IfFalseNode(rc)));
1168 uncommon_trap(Deoptimization::Reason_range_check,
1169 Deoptimization::Action_make_not_entrant);
1170 }
1171
1172 if (stopped()) {
1173 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1174 return true;
1175 }
1176
1177 // index is now known to be >= 0 and < length, cast it
1178 Node* result = ConstraintCastNode::make(control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), ConstraintCastNode::RegularDependency, bt);
1179 result = _gvn.transform(result);
1180 set_result(result);
1181 replace_in_map(index, result);
1182 return true;
1183 }
1184
1185 //------------------------------inline_string_indexOf------------------------
1186 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1187 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1188 return false;
1189 }
1190 Node* src = argument(0);
1191 Node* tgt = argument(1);
1192
1193 // Make the merge point
1194 RegionNode* result_rgn = new RegionNode(4);
1195 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1196
1197 src = must_be_not_null(src, true);
1198 tgt = must_be_not_null(tgt, true);
1199
1200 // Get start addr and length of source string
1201 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1202 Node* src_count = load_array_length(src);
1203
1204 // Get start addr and length of substring
1205 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1206 Node* tgt_count = load_array_length(tgt);
1207
1208 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1209 // Divide src size by 2 if String is UTF16 encoded
1210 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1211 }
1212 if (ae == StrIntrinsicNode::UU) {
1213 // Divide substring size by 2 if String is UTF16 encoded
1214 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1215 }
1216
1217 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1218 if (result != nullptr) {
1219 result_phi->init_req(3, result);
1220 result_rgn->init_req(3, control());
1221 }
1222 set_control(_gvn.transform(result_rgn));
1223 record_for_igvn(result_rgn);
1224 set_result(_gvn.transform(result_phi));
1225
1226 return true;
1227 }
1228
1229 //-----------------------------inline_string_indexOf-----------------------
1230 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1231 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1232 return false;
1233 }
1234 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1235 return false;
1236 }
1237 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1238 Node* src = argument(0); // byte[]
1239 Node* src_count = argument(1); // char count
1240 Node* tgt = argument(2); // byte[]
1241 Node* tgt_count = argument(3); // char count
1242 Node* from_index = argument(4); // char index
1243
1244 src = must_be_not_null(src, true);
1245 tgt = must_be_not_null(tgt, true);
1246
1247 // Multiply byte array index by 2 if String is UTF16 encoded
1248 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1249 src_count = _gvn.transform(new SubINode(src_count, from_index));
1250 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1251 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1252
1253 // Range checks
1254 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1255 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1256 if (stopped()) {
1257 return true;
1258 }
1259
1260 RegionNode* region = new RegionNode(5);
1261 Node* phi = new PhiNode(region, TypeInt::INT);
1262
1263 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1264 if (result != nullptr) {
1265 // The result is index relative to from_index if substring was found, -1 otherwise.
1266 // Generate code which will fold into cmove.
1267 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1268 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1269
1270 Node* if_lt = generate_slow_guard(bol, nullptr);
1271 if (if_lt != nullptr) {
1272 // result == -1
1273 phi->init_req(3, result);
1274 region->init_req(3, if_lt);
1275 }
1276 if (!stopped()) {
1277 result = _gvn.transform(new AddINode(result, from_index));
1278 phi->init_req(4, result);
1279 region->init_req(4, control());
1280 }
1281 }
1282
1283 set_control(_gvn.transform(region));
1284 record_for_igvn(region);
1285 set_result(_gvn.transform(phi));
1286 clear_upper_avx();
1287
1288 return true;
1289 }
1290
1291 // Create StrIndexOfNode with fast path checks
1292 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1293 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1294 // Check for substr count > string count
1295 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1296 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1297 Node* if_gt = generate_slow_guard(bol, nullptr);
1298 if (if_gt != nullptr) {
1299 phi->init_req(1, intcon(-1));
1300 region->init_req(1, if_gt);
1301 }
1302 if (!stopped()) {
1303 // Check for substr count == 0
1304 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1305 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1306 Node* if_zero = generate_slow_guard(bol, nullptr);
1307 if (if_zero != nullptr) {
1308 phi->init_req(2, intcon(0));
1309 region->init_req(2, if_zero);
1310 }
1311 }
1312 if (!stopped()) {
1313 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1314 }
1315 return nullptr;
1316 }
1317
1318 //-----------------------------inline_string_indexOfChar-----------------------
1319 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1320 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1321 return false;
1322 }
1323 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1324 return false;
1325 }
1326 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1327 Node* src = argument(0); // byte[]
1328 Node* int_ch = argument(1);
1329 Node* from_index = argument(2);
1330 Node* max = argument(3);
1331
1332 src = must_be_not_null(src, true);
1333
1334 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1335 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1336 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1337
1338 // Range checks
1339 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1340
1341 // Check for int_ch >= 0
1342 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1343 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1344 {
1345 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1346 uncommon_trap(Deoptimization::Reason_intrinsic,
1347 Deoptimization::Action_maybe_recompile);
1348 }
1349 if (stopped()) {
1350 return true;
1351 }
1352
1353 RegionNode* region = new RegionNode(3);
1354 Node* phi = new PhiNode(region, TypeInt::INT);
1355
1356 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1357 C->set_has_split_ifs(true); // Has chance for split-if optimization
1358 _gvn.transform(result);
1359
1360 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1361 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1362
1363 Node* if_lt = generate_slow_guard(bol, nullptr);
1364 if (if_lt != nullptr) {
1365 // result == -1
1366 phi->init_req(2, result);
1367 region->init_req(2, if_lt);
1368 }
1369 if (!stopped()) {
1370 result = _gvn.transform(new AddINode(result, from_index));
1371 phi->init_req(1, result);
1372 region->init_req(1, control());
1373 }
1374 set_control(_gvn.transform(region));
1375 record_for_igvn(region);
1376 set_result(_gvn.transform(phi));
1377 clear_upper_avx();
1378
1379 return true;
1380 }
1381 //---------------------------inline_string_copy---------------------
1382 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1383 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1384 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1385 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1386 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1387 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1388 bool LibraryCallKit::inline_string_copy(bool compress) {
1389 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1390 return false;
1391 }
1392 int nargs = 5; // 2 oops, 3 ints
1393 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1394
1395 Node* src = argument(0);
1396 Node* src_offset = argument(1);
1397 Node* dst = argument(2);
1398 Node* dst_offset = argument(3);
1399 Node* length = argument(4);
1400
1401 // Check for allocation before we add nodes that would confuse
1402 // tightly_coupled_allocation()
1403 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1404
1405 // Figure out the size and type of the elements we will be copying.
1406 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1407 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1408 if (src_type == nullptr || dst_type == nullptr) {
1409 return false;
1410 }
1411 BasicType src_elem = src_type->elem()->array_element_basic_type();
1412 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1413 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1414 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1415 "Unsupported array types for inline_string_copy");
1416
1417 src = must_be_not_null(src, true);
1418 dst = must_be_not_null(dst, true);
1419
1420 // Convert char[] offsets to byte[] offsets
1421 bool convert_src = (compress && src_elem == T_BYTE);
1422 bool convert_dst = (!compress && dst_elem == T_BYTE);
1423 if (convert_src) {
1424 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1425 } else if (convert_dst) {
1426 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1427 }
1428
1429 // Range checks
1430 generate_string_range_check(src, src_offset, length, convert_src);
1431 generate_string_range_check(dst, dst_offset, length, convert_dst);
1432 if (stopped()) {
1433 return true;
1434 }
1435
1436 Node* src_start = array_element_address(src, src_offset, src_elem);
1437 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1438 // 'src_start' points to src array + scaled offset
1439 // 'dst_start' points to dst array + scaled offset
1440 Node* count = nullptr;
1441 if (compress) {
1442 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1443 } else {
1444 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1445 }
1446
1447 if (alloc != nullptr) {
1448 if (alloc->maybe_set_complete(&_gvn)) {
1449 // "You break it, you buy it."
1450 InitializeNode* init = alloc->initialization();
1451 assert(init->is_complete(), "we just did this");
1452 init->set_complete_with_arraycopy();
1453 assert(dst->is_CheckCastPP(), "sanity");
1454 assert(dst->in(0)->in(0) == init, "dest pinned");
1455 }
1456 // Do not let stores that initialize this object be reordered with
1457 // a subsequent store that would make this object accessible by
1458 // other threads.
1459 // Record what AllocateNode this StoreStore protects so that
1460 // escape analysis can go from the MemBarStoreStoreNode to the
1461 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1462 // based on the escape status of the AllocateNode.
1463 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1464 }
1465 if (compress) {
1466 set_result(_gvn.transform(count));
1467 }
1468 clear_upper_avx();
1469
1470 return true;
1471 }
1472
1473 #ifdef _LP64
1474 #define XTOP ,top() /*additional argument*/
1475 #else //_LP64
1476 #define XTOP /*no additional argument*/
1477 #endif //_LP64
1478
1479 //------------------------inline_string_toBytesU--------------------------
1480 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1481 bool LibraryCallKit::inline_string_toBytesU() {
1482 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1483 return false;
1484 }
1485 // Get the arguments.
1486 Node* value = argument(0);
1487 Node* offset = argument(1);
1488 Node* length = argument(2);
1489
1490 Node* newcopy = nullptr;
1491
1492 // Set the original stack and the reexecute bit for the interpreter to reexecute
1493 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1494 { PreserveReexecuteState preexecs(this);
1495 jvms()->set_should_reexecute(true);
1496
1497 // Check if a null path was taken unconditionally.
1498 value = null_check(value);
1499
1500 RegionNode* bailout = new RegionNode(1);
1501 record_for_igvn(bailout);
1502
1503 // Range checks
1504 generate_negative_guard(offset, bailout);
1505 generate_negative_guard(length, bailout);
1506 generate_limit_guard(offset, length, load_array_length(value), bailout);
1507 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1508 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1509
1510 if (bailout->req() > 1) {
1511 PreserveJVMState pjvms(this);
1512 set_control(_gvn.transform(bailout));
1513 uncommon_trap(Deoptimization::Reason_intrinsic,
1514 Deoptimization::Action_maybe_recompile);
1515 }
1516 if (stopped()) {
1517 return true;
1518 }
1519
1520 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1521 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1522 newcopy = new_array(klass_node, size, 0); // no arguments to push
1523 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1524 guarantee(alloc != nullptr, "created above");
1525
1526 // Calculate starting addresses.
1527 Node* src_start = array_element_address(value, offset, T_CHAR);
1528 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1529
1530 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1531 const TypeInt* toffset = gvn().type(offset)->is_int();
1532 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1533
1534 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1535 const char* copyfunc_name = "arraycopy";
1536 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1537 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1538 OptoRuntime::fast_arraycopy_Type(),
1539 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1540 src_start, dst_start, ConvI2X(length) XTOP);
1541 // Do not let reads from the cloned object float above the arraycopy.
1542 if (alloc->maybe_set_complete(&_gvn)) {
1543 // "You break it, you buy it."
1544 InitializeNode* init = alloc->initialization();
1545 assert(init->is_complete(), "we just did this");
1546 init->set_complete_with_arraycopy();
1547 assert(newcopy->is_CheckCastPP(), "sanity");
1548 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1549 }
1550 // Do not let stores that initialize this object be reordered with
1551 // a subsequent store that would make this object accessible by
1552 // other threads.
1553 // Record what AllocateNode this StoreStore protects so that
1554 // escape analysis can go from the MemBarStoreStoreNode to the
1555 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1556 // based on the escape status of the AllocateNode.
1557 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1558 } // original reexecute is set back here
1559
1560 C->set_has_split_ifs(true); // Has chance for split-if optimization
1561 if (!stopped()) {
1562 set_result(newcopy);
1563 }
1564 clear_upper_avx();
1565
1566 return true;
1567 }
1568
1569 //------------------------inline_string_getCharsU--------------------------
1570 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1571 bool LibraryCallKit::inline_string_getCharsU() {
1572 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1573 return false;
1574 }
1575
1576 // Get the arguments.
1577 Node* src = argument(0);
1578 Node* src_begin = argument(1);
1579 Node* src_end = argument(2); // exclusive offset (i < src_end)
1580 Node* dst = argument(3);
1581 Node* dst_begin = argument(4);
1582
1583 // Check for allocation before we add nodes that would confuse
1584 // tightly_coupled_allocation()
1585 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1586
1587 // Check if a null path was taken unconditionally.
1588 src = null_check(src);
1589 dst = null_check(dst);
1590 if (stopped()) {
1591 return true;
1592 }
1593
1594 // Get length and convert char[] offset to byte[] offset
1595 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1596 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1597
1598 // Range checks
1599 generate_string_range_check(src, src_begin, length, true);
1600 generate_string_range_check(dst, dst_begin, length, false);
1601 if (stopped()) {
1602 return true;
1603 }
1604
1605 if (!stopped()) {
1606 // Calculate starting addresses.
1607 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1608 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1609
1610 // Check if array addresses are aligned to HeapWordSize
1611 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1612 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1613 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1614 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1615
1616 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1617 const char* copyfunc_name = "arraycopy";
1618 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1619 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1620 OptoRuntime::fast_arraycopy_Type(),
1621 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1622 src_start, dst_start, ConvI2X(length) XTOP);
1623 // Do not let reads from the cloned object float above the arraycopy.
1624 if (alloc != nullptr) {
1625 if (alloc->maybe_set_complete(&_gvn)) {
1626 // "You break it, you buy it."
1627 InitializeNode* init = alloc->initialization();
1628 assert(init->is_complete(), "we just did this");
1629 init->set_complete_with_arraycopy();
1630 assert(dst->is_CheckCastPP(), "sanity");
1631 assert(dst->in(0)->in(0) == init, "dest pinned");
1632 }
1633 // Do not let stores that initialize this object be reordered with
1634 // a subsequent store that would make this object accessible by
1635 // other threads.
1636 // Record what AllocateNode this StoreStore protects so that
1637 // escape analysis can go from the MemBarStoreStoreNode to the
1638 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1639 // based on the escape status of the AllocateNode.
1640 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1641 } else {
1642 insert_mem_bar(Op_MemBarCPUOrder);
1643 }
1644 }
1645
1646 C->set_has_split_ifs(true); // Has chance for split-if optimization
1647 return true;
1648 }
1649
1650 //----------------------inline_string_char_access----------------------------
1651 // Store/Load char to/from byte[] array.
1652 // static void StringUTF16.putChar(byte[] val, int index, int c)
1653 // static char StringUTF16.getChar(byte[] val, int index)
1654 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1655 Node* value = argument(0);
1656 Node* index = argument(1);
1657 Node* ch = is_store ? argument(2) : nullptr;
1658
1659 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1660 // correctly requires matched array shapes.
1661 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1662 "sanity: byte[] and char[] bases agree");
1663 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1664 "sanity: byte[] and char[] scales agree");
1665
1666 // Bail when getChar over constants is requested: constant folding would
1667 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1668 // Java method would constant fold nicely instead.
1669 if (!is_store && value->is_Con() && index->is_Con()) {
1670 return false;
1671 }
1672
1673 // Save state and restore on bailout
1674 uint old_sp = sp();
1675 SafePointNode* old_map = clone_map();
1676
1677 value = must_be_not_null(value, true);
1678
1679 Node* adr = array_element_address(value, index, T_CHAR);
1680 if (adr->is_top()) {
1681 set_map(old_map);
1682 set_sp(old_sp);
1683 return false;
1684 }
1685 destruct_map_clone(old_map);
1686 if (is_store) {
1687 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1688 } else {
1689 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);
1690 set_result(ch);
1691 }
1692 return true;
1693 }
1694
1695 //--------------------------round_double_node--------------------------------
1696 // Round a double node if necessary.
1697 Node* LibraryCallKit::round_double_node(Node* n) {
1698 if (Matcher::strict_fp_requires_explicit_rounding) {
1699 #ifdef IA32
1700 if (UseSSE < 2) {
1701 n = _gvn.transform(new RoundDoubleNode(nullptr, n));
1702 }
1703 #else
1704 Unimplemented();
1705 #endif // IA32
1706 }
1707 return n;
1708 }
1709
1710 //------------------------------inline_math-----------------------------------
1711 // public static double Math.abs(double)
1712 // public static double Math.sqrt(double)
1713 // public static double Math.log(double)
1714 // public static double Math.log10(double)
1715 // public static double Math.round(double)
1716 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1717 Node* arg = round_double_node(argument(0));
1718 Node* n = nullptr;
1719 switch (id) {
1720 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1721 case vmIntrinsics::_dsqrt:
1722 case vmIntrinsics::_dsqrt_strict:
1723 n = new SqrtDNode(C, control(), arg); break;
1724 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1725 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1726 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1727 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1728 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1729 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1730 default: fatal_unexpected_iid(id); break;
1731 }
1732 set_result(_gvn.transform(n));
1733 return true;
1734 }
1735
1736 //------------------------------inline_math-----------------------------------
1737 // public static float Math.abs(float)
1738 // public static int Math.abs(int)
1739 // public static long Math.abs(long)
1740 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1741 Node* arg = argument(0);
1742 Node* n = nullptr;
1743 switch (id) {
1744 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1745 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1746 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1747 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1748 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1749 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1750 default: fatal_unexpected_iid(id); break;
1751 }
1752 set_result(_gvn.transform(n));
1753 return true;
1754 }
1755
1756 //------------------------------runtime_math-----------------------------
1757 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1758 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1759 "must be (DD)D or (D)D type");
1760
1761 // Inputs
1762 Node* a = round_double_node(argument(0));
1763 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : nullptr;
1764
1765 const TypePtr* no_memory_effects = nullptr;
1766 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1767 no_memory_effects,
1768 a, top(), b, b ? top() : nullptr);
1769 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1770 #ifdef ASSERT
1771 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1772 assert(value_top == top(), "second value must be top");
1773 #endif
1774
1775 set_result(value);
1776 return true;
1777 }
1778
1779 //------------------------------inline_math_pow-----------------------------
1780 bool LibraryCallKit::inline_math_pow() {
1781 Node* exp = round_double_node(argument(2));
1782 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1783 if (d != nullptr) {
1784 if (d->getd() == 2.0) {
1785 // Special case: pow(x, 2.0) => x * x
1786 Node* base = round_double_node(argument(0));
1787 set_result(_gvn.transform(new MulDNode(base, base)));
1788 return true;
1789 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1790 // Special case: pow(x, 0.5) => sqrt(x)
1791 Node* base = round_double_node(argument(0));
1792 Node* zero = _gvn.zerocon(T_DOUBLE);
1793
1794 RegionNode* region = new RegionNode(3);
1795 Node* phi = new PhiNode(region, Type::DOUBLE);
1796
1797 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1798 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1799 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1800 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1801 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1802
1803 Node* if_pow = generate_slow_guard(test, nullptr);
1804 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1805 phi->init_req(1, value_sqrt);
1806 region->init_req(1, control());
1807
1808 if (if_pow != nullptr) {
1809 set_control(if_pow);
1810 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1811 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1812 const TypePtr* no_memory_effects = nullptr;
1813 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1814 no_memory_effects, base, top(), exp, top());
1815 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1816 #ifdef ASSERT
1817 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1818 assert(value_top == top(), "second value must be top");
1819 #endif
1820 phi->init_req(2, value_pow);
1821 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1822 }
1823
1824 C->set_has_split_ifs(true); // Has chance for split-if optimization
1825 set_control(_gvn.transform(region));
1826 record_for_igvn(region);
1827 set_result(_gvn.transform(phi));
1828
1829 return true;
1830 }
1831 }
1832
1833 return StubRoutines::dpow() != nullptr ?
1834 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1835 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1836 }
1837
1838 //------------------------------inline_math_native-----------------------------
1839 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1840 switch (id) {
1841 case vmIntrinsics::_dsin:
1842 return StubRoutines::dsin() != nullptr ?
1843 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1844 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1845 case vmIntrinsics::_dcos:
1846 return StubRoutines::dcos() != nullptr ?
1847 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1848 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1849 case vmIntrinsics::_dtan:
1850 return StubRoutines::dtan() != nullptr ?
1851 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1852 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1853 case vmIntrinsics::_dexp:
1854 return StubRoutines::dexp() != nullptr ?
1855 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1856 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1857 case vmIntrinsics::_dlog:
1858 return StubRoutines::dlog() != nullptr ?
1859 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1860 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1861 case vmIntrinsics::_dlog10:
1862 return StubRoutines::dlog10() != nullptr ?
1863 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1864 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1865
1866 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1867 case vmIntrinsics::_ceil:
1868 case vmIntrinsics::_floor:
1869 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1870
1871 case vmIntrinsics::_dsqrt:
1872 case vmIntrinsics::_dsqrt_strict:
1873 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1874 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1875 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1876 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1877 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1878
1879 case vmIntrinsics::_dpow: return inline_math_pow();
1880 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1881 case vmIntrinsics::_fcopySign: return inline_math(id);
1882 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1883 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1884 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1885
1886 // These intrinsics are not yet correctly implemented
1887 case vmIntrinsics::_datan2:
1888 return false;
1889
1890 default:
1891 fatal_unexpected_iid(id);
1892 return false;
1893 }
1894 }
1895
1896 //----------------------------inline_notify-----------------------------------*
1897 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1898 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1899 address func;
1900 if (id == vmIntrinsics::_notify) {
1901 func = OptoRuntime::monitor_notify_Java();
1902 } else {
1903 func = OptoRuntime::monitor_notifyAll_Java();
1904 }
1905 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1906 make_slow_call_ex(call, env()->Throwable_klass(), false);
1907 return true;
1908 }
1909
1910
1911 //----------------------------inline_min_max-----------------------------------
1912 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1913 set_result(generate_min_max(id, argument(0), argument(1)));
1914 return true;
1915 }
1916
1917 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1918 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1919 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1920 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1921 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1922
1923 {
1924 PreserveJVMState pjvms(this);
1925 PreserveReexecuteState preexecs(this);
1926 jvms()->set_should_reexecute(true);
1927
1928 set_control(slow_path);
1929 set_i_o(i_o());
1930
1931 uncommon_trap(Deoptimization::Reason_intrinsic,
1932 Deoptimization::Action_none);
1933 }
1934
1935 set_control(fast_path);
1936 set_result(math);
1937 }
1938
1939 template <typename OverflowOp>
1940 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1941 typedef typename OverflowOp::MathOp MathOp;
1942
1943 MathOp* mathOp = new MathOp(arg1, arg2);
1944 Node* operation = _gvn.transform( mathOp );
1945 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1946 inline_math_mathExact(operation, ofcheck);
1947 return true;
1948 }
1949
1950 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1951 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1952 }
1953
1954 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1955 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1956 }
1957
1958 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1959 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1960 }
1961
1962 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1963 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1964 }
1965
1966 bool LibraryCallKit::inline_math_negateExactI() {
1967 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1968 }
1969
1970 bool LibraryCallKit::inline_math_negateExactL() {
1971 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1972 }
1973
1974 bool LibraryCallKit::inline_math_multiplyExactI() {
1975 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1976 }
1977
1978 bool LibraryCallKit::inline_math_multiplyExactL() {
1979 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1980 }
1981
1982 bool LibraryCallKit::inline_math_multiplyHigh() {
1983 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
1984 return true;
1985 }
1986
1987 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
1988 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
1989 return true;
1990 }
1991
1992 Node*
1993 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1994 Node* result_val = nullptr;
1995 switch (id) {
1996 case vmIntrinsics::_min:
1997 case vmIntrinsics::_min_strict:
1998 result_val = _gvn.transform(new MinINode(x0, y0));
1999 break;
2000 case vmIntrinsics::_max:
2001 case vmIntrinsics::_max_strict:
2002 result_val = _gvn.transform(new MaxINode(x0, y0));
2003 break;
2004 default:
2005 fatal_unexpected_iid(id);
2006 break;
2007 }
2008 return result_val;
2009 }
2010
2011 inline int
2012 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2013 const TypePtr* base_type = TypePtr::NULL_PTR;
2014 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2015 if (base_type == nullptr) {
2016 // Unknown type.
2017 return Type::AnyPtr;
2018 } else if (base_type == TypePtr::NULL_PTR) {
2019 // Since this is a null+long form, we have to switch to a rawptr.
2020 base = _gvn.transform(new CastX2PNode(offset));
2021 offset = MakeConX(0);
2022 return Type::RawPtr;
2023 } else if (base_type->base() == Type::RawPtr) {
2024 return Type::RawPtr;
2025 } else if (base_type->isa_oopptr()) {
2026 // Base is never null => always a heap address.
2027 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2028 return Type::OopPtr;
2029 }
2030 // Offset is small => always a heap address.
2031 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2032 if (offset_type != nullptr &&
2033 base_type->offset() == 0 && // (should always be?)
2034 offset_type->_lo >= 0 &&
2035 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2036 return Type::OopPtr;
2037 } else if (type == T_OBJECT) {
2038 // off heap access to an oop doesn't make any sense. Has to be on
2039 // heap.
2040 return Type::OopPtr;
2041 }
2042 // Otherwise, it might either be oop+off or null+addr.
2043 return Type::AnyPtr;
2044 } else {
2045 // No information:
2046 return Type::AnyPtr;
2047 }
2048 }
2049
2050 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2051 Node* uncasted_base = base;
2052 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2053 if (kind == Type::RawPtr) {
2054 return basic_plus_adr(top(), uncasted_base, offset);
2055 } else if (kind == Type::AnyPtr) {
2056 assert(base == uncasted_base, "unexpected base change");
2057 if (can_cast) {
2058 if (!_gvn.type(base)->speculative_maybe_null() &&
2059 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2060 // According to profiling, this access is always on
2061 // heap. Casting the base to not null and thus avoiding membars
2062 // around the access should allow better optimizations
2063 Node* null_ctl = top();
2064 base = null_check_oop(base, &null_ctl, true, true, true);
2065 assert(null_ctl->is_top(), "no null control here");
2066 return basic_plus_adr(base, offset);
2067 } else if (_gvn.type(base)->speculative_always_null() &&
2068 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2069 // According to profiling, this access is always off
2070 // heap.
2071 base = null_assert(base);
2072 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2073 offset = MakeConX(0);
2074 return basic_plus_adr(top(), raw_base, offset);
2075 }
2076 }
2077 // We don't know if it's an on heap or off heap access. Fall back
2078 // to raw memory access.
2079 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2080 return basic_plus_adr(top(), raw, offset);
2081 } else {
2082 assert(base == uncasted_base, "unexpected base change");
2083 // We know it's an on heap access so base can't be null
2084 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2085 base = must_be_not_null(base, true);
2086 }
2087 return basic_plus_adr(base, offset);
2088 }
2089 }
2090
2091 //--------------------------inline_number_methods-----------------------------
2092 // inline int Integer.numberOfLeadingZeros(int)
2093 // inline int Long.numberOfLeadingZeros(long)
2094 //
2095 // inline int Integer.numberOfTrailingZeros(int)
2096 // inline int Long.numberOfTrailingZeros(long)
2097 //
2098 // inline int Integer.bitCount(int)
2099 // inline int Long.bitCount(long)
2100 //
2101 // inline char Character.reverseBytes(char)
2102 // inline short Short.reverseBytes(short)
2103 // inline int Integer.reverseBytes(int)
2104 // inline long Long.reverseBytes(long)
2105 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2106 Node* arg = argument(0);
2107 Node* n = nullptr;
2108 switch (id) {
2109 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2110 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2111 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2112 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2113 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2114 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2115 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2116 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2117 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2118 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2119 case vmIntrinsics::_reverse_i: n = new ReverseINode(0, arg); break;
2120 case vmIntrinsics::_reverse_l: n = new ReverseLNode(0, arg); break;
2121 default: fatal_unexpected_iid(id); break;
2122 }
2123 set_result(_gvn.transform(n));
2124 return true;
2125 }
2126
2127 //--------------------------inline_bitshuffle_methods-----------------------------
2128 // inline int Integer.compress(int, int)
2129 // inline int Integer.expand(int, int)
2130 // inline long Long.compress(long, long)
2131 // inline long Long.expand(long, long)
2132 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2133 Node* n = nullptr;
2134 switch (id) {
2135 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2136 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2137 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2138 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2139 default: fatal_unexpected_iid(id); break;
2140 }
2141 set_result(_gvn.transform(n));
2142 return true;
2143 }
2144
2145 //--------------------------inline_number_methods-----------------------------
2146 // inline int Integer.compareUnsigned(int, int)
2147 // inline int Long.compareUnsigned(long, long)
2148 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2149 Node* arg1 = argument(0);
2150 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2151 Node* n = nullptr;
2152 switch (id) {
2153 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2154 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2155 default: fatal_unexpected_iid(id); break;
2156 }
2157 set_result(_gvn.transform(n));
2158 return true;
2159 }
2160
2161 //--------------------------inline_unsigned_divmod_methods-----------------------------
2162 // inline int Integer.divideUnsigned(int, int)
2163 // inline int Integer.remainderUnsigned(int, int)
2164 // inline long Long.divideUnsigned(long, long)
2165 // inline long Long.remainderUnsigned(long, long)
2166 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2167 Node* n = nullptr;
2168 switch (id) {
2169 case vmIntrinsics::_divideUnsigned_i: {
2170 zero_check_int(argument(1));
2171 // Compile-time detect of null-exception
2172 if (stopped()) {
2173 return true; // keep the graph constructed so far
2174 }
2175 n = new UDivINode(control(), argument(0), argument(1));
2176 break;
2177 }
2178 case vmIntrinsics::_divideUnsigned_l: {
2179 zero_check_long(argument(2));
2180 // Compile-time detect of null-exception
2181 if (stopped()) {
2182 return true; // keep the graph constructed so far
2183 }
2184 n = new UDivLNode(control(), argument(0), argument(2));
2185 break;
2186 }
2187 case vmIntrinsics::_remainderUnsigned_i: {
2188 zero_check_int(argument(1));
2189 // Compile-time detect of null-exception
2190 if (stopped()) {
2191 return true; // keep the graph constructed so far
2192 }
2193 n = new UModINode(control(), argument(0), argument(1));
2194 break;
2195 }
2196 case vmIntrinsics::_remainderUnsigned_l: {
2197 zero_check_long(argument(2));
2198 // Compile-time detect of null-exception
2199 if (stopped()) {
2200 return true; // keep the graph constructed so far
2201 }
2202 n = new UModLNode(control(), argument(0), argument(2));
2203 break;
2204 }
2205 default: fatal_unexpected_iid(id); break;
2206 }
2207 set_result(_gvn.transform(n));
2208 return true;
2209 }
2210
2211 //----------------------------inline_unsafe_access----------------------------
2212
2213 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2214 // Attempt to infer a sharper value type from the offset and base type.
2215 ciKlass* sharpened_klass = nullptr;
2216 bool null_free = false;
2217
2218 // See if it is an instance field, with an object type.
2219 if (alias_type->field() != nullptr) {
2220 if (alias_type->field()->type()->is_klass()) {
2221 sharpened_klass = alias_type->field()->type()->as_klass();
2222 null_free = alias_type->field()->is_null_free();
2223 }
2224 }
2225
2226 const TypeOopPtr* result = nullptr;
2227 // See if it is a narrow oop array.
2228 if (adr_type->isa_aryptr()) {
2229 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2230 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2231 null_free = adr_type->is_aryptr()->is_null_free();
2232 if (elem_type != nullptr && elem_type->is_loaded()) {
2233 // Sharpen the value type.
2234 result = elem_type;
2235 }
2236 }
2237 }
2238
2239 // The sharpened class might be unloaded if there is no class loader
2240 // contraint in place.
2241 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2242 // Sharpen the value type.
2243 result = TypeOopPtr::make_from_klass(sharpened_klass);
2244 if (null_free) {
2245 result = result->join_speculative(TypePtr::NOTNULL)->is_oopptr();
2246 }
2247 }
2248 if (result != nullptr) {
2249 #ifndef PRODUCT
2250 if (C->print_intrinsics() || C->print_inlining()) {
2251 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2252 tty->print(" sharpened value: "); result->dump(); tty->cr();
2253 }
2254 #endif
2255 }
2256 return result;
2257 }
2258
2259 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2260 switch (kind) {
2261 case Relaxed:
2262 return MO_UNORDERED;
2263 case Opaque:
2264 return MO_RELAXED;
2265 case Acquire:
2266 return MO_ACQUIRE;
2267 case Release:
2268 return MO_RELEASE;
2269 case Volatile:
2270 return MO_SEQ_CST;
2271 default:
2272 ShouldNotReachHere();
2273 return 0;
2274 }
2275 }
2276
2277 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned, const bool is_flat) {
2278 if (callee()->is_static()) return false; // caller must have the capability!
2279 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2280 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2281 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2282 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2283
2284 if (is_reference_type(type)) {
2285 decorators |= ON_UNKNOWN_OOP_REF;
2286 }
2287
2288 if (unaligned) {
2289 decorators |= C2_UNALIGNED;
2290 }
2291
2292 #ifndef PRODUCT
2293 {
2294 ResourceMark rm;
2295 // Check the signatures.
2296 ciSignature* sig = callee()->signature();
2297 #ifdef ASSERT
2298 if (!is_store) {
2299 // Object getReference(Object base, int/long offset), etc.
2300 BasicType rtype = sig->return_type()->basic_type();
2301 assert(rtype == type, "getter must return the expected value");
2302 assert(sig->count() == 2 || (is_flat && sig->count() == 3), "oop getter has 2 or 3 arguments");
2303 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2304 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2305 } else {
2306 // void putReference(Object base, int/long offset, Object x), etc.
2307 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2308 assert(sig->count() == 3 || (is_flat && sig->count() == 4), "oop putter has 3 arguments");
2309 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2310 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2311 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2312 assert(vtype == type, "putter must accept the expected value");
2313 }
2314 #endif // ASSERT
2315 }
2316 #endif //PRODUCT
2317
2318 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2319
2320 Node* receiver = argument(0); // type: oop
2321
2322 // Build address expression.
2323 Node* heap_base_oop = top();
2324
2325 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2326 Node* base = argument(1); // type: oop
2327 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2328 Node* offset = argument(2); // type: long
2329 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2330 // to be plain byte offsets, which are also the same as those accepted
2331 // by oopDesc::field_addr.
2332 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2333 "fieldOffset must be byte-scaled");
2334
2335 ciInlineKlass* inline_klass = nullptr;
2336 if (is_flat) {
2337 const TypeInstPtr* cls = _gvn.type(argument(4))->isa_instptr();
2338 if (cls == nullptr || cls->const_oop() == nullptr) {
2339 return false;
2340 }
2341 ciType* mirror_type = cls->const_oop()->as_instance()->java_mirror_type();
2342 if (!mirror_type->is_inlinetype()) {
2343 return false;
2344 }
2345 inline_klass = mirror_type->as_inline_klass();
2346 }
2347
2348 if (base->is_InlineType()) {
2349 InlineTypeNode* vt = base->as_InlineType();
2350 if (is_store) {
2351 if (!vt->is_allocated(&_gvn)) {
2352 return false;
2353 }
2354 base = vt->get_oop();
2355 } else {
2356 if (offset->is_Con()) {
2357 long off = find_long_con(offset, 0);
2358 ciInlineKlass* vk = vt->type()->inline_klass();
2359 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2360 return false;
2361 }
2362
2363 ciField* field = vk->get_non_flat_field_by_offset(off);
2364 if (field != nullptr) {
2365 BasicType bt = type2field[field->type()->basic_type()];
2366 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2367 bt = T_OBJECT;
2368 }
2369 if (bt == type && (!field->is_flat() || field->type() == inline_klass)) {
2370 Node* value = vt->field_value_by_offset(off, false);
2371 if (value->is_InlineType()) {
2372 value = value->as_InlineType()->adjust_scalarization_depth(this);
2373 }
2374 set_result(value);
2375 return true;
2376 }
2377 }
2378 }
2379 {
2380 // Re-execute the unsafe access if allocation triggers deoptimization.
2381 PreserveReexecuteState preexecs(this);
2382 jvms()->set_should_reexecute(true);
2383 vt = vt->buffer(this);
2384 }
2385 base = vt->get_oop();
2386 }
2387 }
2388
2389 // 32-bit machines ignore the high half!
2390 offset = ConvL2X(offset);
2391
2392 // Save state and restore on bailout
2393 uint old_sp = sp();
2394 SafePointNode* old_map = clone_map();
2395
2396 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2397
2398 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) {
2399 if (type != T_OBJECT && (inline_klass == nullptr || !inline_klass->has_object_fields())) {
2400 decorators |= IN_NATIVE; // off-heap primitive access
2401 } else {
2402 set_map(old_map);
2403 set_sp(old_sp);
2404 return false; // off-heap oop accesses are not supported
2405 }
2406 } else {
2407 heap_base_oop = base; // on-heap or mixed access
2408 }
2409
2410 // Can base be null? Otherwise, always on-heap access.
2411 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2412
2413 if (!can_access_non_heap) {
2414 decorators |= IN_HEAP;
2415 }
2416
2417 Node* val = is_store ? argument(4 + (is_flat ? 1 : 0)) : nullptr;
2418
2419 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2420 if (adr_type == TypePtr::NULL_PTR) {
2421 set_map(old_map);
2422 set_sp(old_sp);
2423 return false; // off-heap access with zero address
2424 }
2425
2426 // Try to categorize the address.
2427 Compile::AliasType* alias_type = C->alias_type(adr_type);
2428 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2429
2430 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2431 alias_type->adr_type() == TypeAryPtr::RANGE) {
2432 set_map(old_map);
2433 set_sp(old_sp);
2434 return false; // not supported
2435 }
2436
2437 bool mismatched = false;
2438 BasicType bt = T_ILLEGAL;
2439 ciField* field = nullptr;
2440 if (adr_type->isa_instptr()) {
2441 const TypeInstPtr* instptr = adr_type->is_instptr();
2442 ciInstanceKlass* k = instptr->instance_klass();
2443 int off = instptr->offset();
2444 if (instptr->const_oop() != nullptr &&
2445 k == ciEnv::current()->Class_klass() &&
2446 instptr->offset() >= (k->size_helper() * wordSize)) {
2447 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2448 field = k->get_field_by_offset(off, true);
2449 } else {
2450 field = k->get_non_flat_field_by_offset(off);
2451 }
2452 if (field != nullptr) {
2453 bt = type2field[field->type()->basic_type()];
2454 }
2455 assert(bt == alias_type->basic_type() || is_flat, "should match");
2456 } else {
2457 bt = alias_type->basic_type();
2458 }
2459
2460 if (bt != T_ILLEGAL) {
2461 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2462 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2463 // Alias type doesn't differentiate between byte[] and boolean[]).
2464 // Use address type to get the element type.
2465 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2466 }
2467 if (is_reference_type(bt, true)) {
2468 // accessing an array field with getReference is not a mismatch
2469 bt = T_OBJECT;
2470 }
2471 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2472 // Don't intrinsify mismatched object accesses
2473 set_map(old_map);
2474 set_sp(old_sp);
2475 return false;
2476 }
2477 mismatched = (bt != type);
2478 } else if (alias_type->adr_type()->isa_oopptr()) {
2479 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2480 }
2481
2482 if (is_flat) {
2483 if (adr_type->isa_instptr()) {
2484 if (field == nullptr || field->type() != inline_klass) {
2485 mismatched = true;
2486 }
2487 } else if (adr_type->isa_aryptr()) {
2488 const Type* elem = adr_type->is_aryptr()->elem();
2489 if (!adr_type->is_flat() || elem->inline_klass() != inline_klass) {
2490 mismatched = true;
2491 }
2492 } else {
2493 mismatched = true;
2494 }
2495 if (is_store) {
2496 const Type* val_t = _gvn.type(val);
2497 if (!val_t->is_inlinetypeptr() || val_t->inline_klass() != inline_klass) {
2498 set_map(old_map);
2499 set_sp(old_sp);
2500 return false;
2501 }
2502 }
2503 }
2504
2505 destruct_map_clone(old_map);
2506 assert(!mismatched || is_flat || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2507
2508 if (mismatched) {
2509 decorators |= C2_MISMATCHED;
2510 }
2511
2512 // First guess at the value type.
2513 const Type *value_type = Type::get_const_basic_type(type);
2514
2515 // Figure out the memory ordering.
2516 decorators |= mo_decorator_for_access_kind(kind);
2517
2518 if (!is_store) {
2519 if (type == T_OBJECT && !is_flat) {
2520 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2521 if (tjp != nullptr) {
2522 value_type = tjp;
2523 }
2524 }
2525 }
2526
2527 receiver = null_check(receiver);
2528 if (stopped()) {
2529 return true;
2530 }
2531 // Heap pointers get a null-check from the interpreter,
2532 // as a courtesy. However, this is not guaranteed by Unsafe,
2533 // and it is not possible to fully distinguish unintended nulls
2534 // from intended ones in this API.
2535
2536 if (!is_store) {
2537 Node* p = nullptr;
2538 // Try to constant fold a load from a constant field
2539
2540 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !field->is_flat() && !mismatched) {
2541 // final or stable field
2542 p = make_constant_from_field(field, heap_base_oop);
2543 }
2544
2545 if (p == nullptr) { // Could not constant fold the load
2546 if (is_flat) {
2547 if (adr_type->isa_instptr() && !mismatched) {
2548 ciInstanceKlass* holder = adr_type->is_instptr()->instance_klass();
2549 int offset = adr_type->is_instptr()->offset();
2550 p = InlineTypeNode::make_from_flat(this, inline_klass, base, base, holder, offset, decorators);
2551 } else {
2552 p = InlineTypeNode::make_from_flat(this, inline_klass, base, adr, nullptr, 0, decorators);
2553 }
2554 } else {
2555 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2556 const TypeOopPtr* ptr = value_type->make_oopptr();
2557 if (ptr != nullptr && ptr->is_inlinetypeptr()) {
2558 // Load a non-flattened inline type from memory
2559 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass(), !ptr->maybe_null());
2560 }
2561 }
2562 // Normalize the value returned by getBoolean in the following cases
2563 if (type == T_BOOLEAN &&
2564 (mismatched ||
2565 heap_base_oop == top() || // - heap_base_oop is null or
2566 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2567 // and the unsafe access is made to large offset
2568 // (i.e., larger than the maximum offset necessary for any
2569 // field access)
2570 ) {
2571 IdealKit ideal = IdealKit(this);
2572 #define __ ideal.
2573 IdealVariable normalized_result(ideal);
2574 __ declarations_done();
2575 __ set(normalized_result, p);
2576 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2577 __ set(normalized_result, ideal.ConI(1));
2578 ideal.end_if();
2579 final_sync(ideal);
2580 p = __ value(normalized_result);
2581 #undef __
2582 }
2583 }
2584 if (type == T_ADDRESS) {
2585 p = gvn().transform(new CastP2XNode(nullptr, p));
2586 p = ConvX2UL(p);
2587 }
2588 // The load node has the control of the preceding MemBarCPUOrder. All
2589 // following nodes will have the control of the MemBarCPUOrder inserted at
2590 // the end of this method. So, pushing the load onto the stack at a later
2591 // point is fine.
2592 set_result(p);
2593 } else {
2594 if (bt == T_ADDRESS) {
2595 // Repackage the long as a pointer.
2596 val = ConvL2X(val);
2597 val = gvn().transform(new CastX2PNode(val));
2598 }
2599 if (is_flat) {
2600 if (adr_type->isa_instptr() && !mismatched) {
2601 ciInstanceKlass* holder = adr_type->is_instptr()->instance_klass();
2602 int offset = adr_type->is_instptr()->offset();
2603 val->as_InlineType()->store_flat(this, base, base, holder, offset, decorators);
2604 } else {
2605 val->as_InlineType()->store_flat(this, base, adr, nullptr, 0, decorators);
2606 }
2607 } else {
2608 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2609 }
2610 }
2611
2612 if (argument(1)->is_InlineType() && is_store) {
2613 InlineTypeNode* value = InlineTypeNode::make_from_oop(this, base, _gvn.type(argument(1))->inline_klass());
2614 value = value->make_larval(this, false);
2615 replace_in_map(argument(1), value);
2616 }
2617
2618 return true;
2619 }
2620
2621 bool LibraryCallKit::inline_unsafe_make_private_buffer() {
2622 Node* receiver = argument(0);
2623 Node* value = argument(1);
2624 if (!value->is_InlineType()) {
2625 return false;
2626 }
2627
2628 receiver = null_check(receiver);
2629 if (stopped()) {
2630 return true;
2631 }
2632
2633 set_result(value->as_InlineType()->make_larval(this, true));
2634 return true;
2635 }
2636
2637 bool LibraryCallKit::inline_unsafe_finish_private_buffer() {
2638 Node* receiver = argument(0);
2639 Node* buffer = argument(1);
2640 if (!buffer->is_InlineType()) {
2641 return false;
2642 }
2643 InlineTypeNode* vt = buffer->as_InlineType();
2644 if (!vt->is_allocated(&_gvn)) {
2645 return false;
2646 }
2647 // TODO 8239003 Why is this needed?
2648 if (AllocateNode::Ideal_allocation(vt->get_oop()) == nullptr) {
2649 return false;
2650 }
2651
2652 receiver = null_check(receiver);
2653 if (stopped()) {
2654 return true;
2655 }
2656
2657 set_result(vt->finish_larval(this));
2658 return true;
2659 }
2660
2661 //----------------------------inline_unsafe_load_store----------------------------
2662 // This method serves a couple of different customers (depending on LoadStoreKind):
2663 //
2664 // LS_cmp_swap:
2665 //
2666 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2667 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2668 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2669 //
2670 // LS_cmp_swap_weak:
2671 //
2672 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2673 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2674 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2675 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2676 //
2677 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2678 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2679 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2680 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2681 //
2682 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2683 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2684 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2685 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2686 //
2687 // LS_cmp_exchange:
2688 //
2689 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2690 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2691 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2692 //
2693 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2694 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2695 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2696 //
2697 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2698 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2699 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2700 //
2701 // LS_get_add:
2702 //
2703 // int getAndAddInt( Object o, long offset, int delta)
2704 // long getAndAddLong(Object o, long offset, long delta)
2705 //
2706 // LS_get_set:
2707 //
2708 // int getAndSet(Object o, long offset, int newValue)
2709 // long getAndSet(Object o, long offset, long newValue)
2710 // Object getAndSet(Object o, long offset, Object newValue)
2711 //
2712 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2713 // This basic scheme here is the same as inline_unsafe_access, but
2714 // differs in enough details that combining them would make the code
2715 // overly confusing. (This is a true fact! I originally combined
2716 // them, but even I was confused by it!) As much code/comments as
2717 // possible are retained from inline_unsafe_access though to make
2718 // the correspondences clearer. - dl
2719
2720 if (callee()->is_static()) return false; // caller must have the capability!
2721
2722 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2723 decorators |= mo_decorator_for_access_kind(access_kind);
2724
2725 #ifndef PRODUCT
2726 BasicType rtype;
2727 {
2728 ResourceMark rm;
2729 // Check the signatures.
2730 ciSignature* sig = callee()->signature();
2731 rtype = sig->return_type()->basic_type();
2732 switch(kind) {
2733 case LS_get_add:
2734 case LS_get_set: {
2735 // Check the signatures.
2736 #ifdef ASSERT
2737 assert(rtype == type, "get and set must return the expected type");
2738 assert(sig->count() == 3, "get and set has 3 arguments");
2739 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2740 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2741 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2742 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2743 #endif // ASSERT
2744 break;
2745 }
2746 case LS_cmp_swap:
2747 case LS_cmp_swap_weak: {
2748 // Check the signatures.
2749 #ifdef ASSERT
2750 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2751 assert(sig->count() == 4, "CAS has 4 arguments");
2752 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2753 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2754 #endif // ASSERT
2755 break;
2756 }
2757 case LS_cmp_exchange: {
2758 // Check the signatures.
2759 #ifdef ASSERT
2760 assert(rtype == type, "CAS must return the expected type");
2761 assert(sig->count() == 4, "CAS has 4 arguments");
2762 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2763 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2764 #endif // ASSERT
2765 break;
2766 }
2767 default:
2768 ShouldNotReachHere();
2769 }
2770 }
2771 #endif //PRODUCT
2772
2773 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2774
2775 // Get arguments:
2776 Node* receiver = nullptr;
2777 Node* base = nullptr;
2778 Node* offset = nullptr;
2779 Node* oldval = nullptr;
2780 Node* newval = nullptr;
2781 switch(kind) {
2782 case LS_cmp_swap:
2783 case LS_cmp_swap_weak:
2784 case LS_cmp_exchange: {
2785 const bool two_slot_type = type2size[type] == 2;
2786 receiver = argument(0); // type: oop
2787 base = argument(1); // type: oop
2788 offset = argument(2); // type: long
2789 oldval = argument(4); // type: oop, int, or long
2790 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2791 break;
2792 }
2793 case LS_get_add:
2794 case LS_get_set: {
2795 receiver = argument(0); // type: oop
2796 base = argument(1); // type: oop
2797 offset = argument(2); // type: long
2798 oldval = nullptr;
2799 newval = argument(4); // type: oop, int, or long
2800 break;
2801 }
2802 default:
2803 ShouldNotReachHere();
2804 }
2805
2806 // Build field offset expression.
2807 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2808 // to be plain byte offsets, which are also the same as those accepted
2809 // by oopDesc::field_addr.
2810 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2811 // 32-bit machines ignore the high half of long offsets
2812 offset = ConvL2X(offset);
2813 // Save state and restore on bailout
2814 uint old_sp = sp();
2815 SafePointNode* old_map = clone_map();
2816 Node* adr = make_unsafe_address(base, offset,type, false);
2817 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2818
2819 Compile::AliasType* alias_type = C->alias_type(adr_type);
2820 BasicType bt = alias_type->basic_type();
2821 if (bt != T_ILLEGAL &&
2822 (is_reference_type(bt) != (type == T_OBJECT))) {
2823 // Don't intrinsify mismatched object accesses.
2824 set_map(old_map);
2825 set_sp(old_sp);
2826 return false;
2827 }
2828
2829 destruct_map_clone(old_map);
2830
2831 // For CAS, unlike inline_unsafe_access, there seems no point in
2832 // trying to refine types. Just use the coarse types here.
2833 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2834 const Type *value_type = Type::get_const_basic_type(type);
2835
2836 switch (kind) {
2837 case LS_get_set:
2838 case LS_cmp_exchange: {
2839 if (type == T_OBJECT) {
2840 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2841 if (tjp != nullptr) {
2842 value_type = tjp;
2843 }
2844 }
2845 break;
2846 }
2847 case LS_cmp_swap:
2848 case LS_cmp_swap_weak:
2849 case LS_get_add:
2850 break;
2851 default:
2852 ShouldNotReachHere();
2853 }
2854
2855 // Null check receiver.
2856 receiver = null_check(receiver);
2857 if (stopped()) {
2858 return true;
2859 }
2860
2861 int alias_idx = C->get_alias_index(adr_type);
2862
2863 if (is_reference_type(type)) {
2864 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2865
2866 if (oldval != nullptr && oldval->is_InlineType()) {
2867 // Re-execute the unsafe access if allocation triggers deoptimization.
2868 PreserveReexecuteState preexecs(this);
2869 jvms()->set_should_reexecute(true);
2870 oldval = oldval->as_InlineType()->buffer(this)->get_oop();
2871 }
2872 if (newval != nullptr && newval->is_InlineType()) {
2873 // Re-execute the unsafe access if allocation triggers deoptimization.
2874 PreserveReexecuteState preexecs(this);
2875 jvms()->set_should_reexecute(true);
2876 newval = newval->as_InlineType()->buffer(this)->get_oop();
2877 }
2878
2879 // Transformation of a value which could be null pointer (CastPP #null)
2880 // could be delayed during Parse (for example, in adjust_map_after_if()).
2881 // Execute transformation here to avoid barrier generation in such case.
2882 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2883 newval = _gvn.makecon(TypePtr::NULL_PTR);
2884
2885 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2886 // Refine the value to a null constant, when it is known to be null
2887 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2888 }
2889 }
2890
2891 Node* result = nullptr;
2892 switch (kind) {
2893 case LS_cmp_exchange: {
2894 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2895 oldval, newval, value_type, type, decorators);
2896 break;
2897 }
2898 case LS_cmp_swap_weak:
2899 decorators |= C2_WEAK_CMPXCHG;
2900 case LS_cmp_swap: {
2901 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2902 oldval, newval, value_type, type, decorators);
2903 break;
2904 }
2905 case LS_get_set: {
2906 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2907 newval, value_type, type, decorators);
2908 break;
2909 }
2910 case LS_get_add: {
2911 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2912 newval, value_type, type, decorators);
2913 break;
2914 }
2915 default:
2916 ShouldNotReachHere();
2917 }
2918
2919 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2920 set_result(result);
2921 return true;
2922 }
2923
2924 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2925 // Regardless of form, don't allow previous ld/st to move down,
2926 // then issue acquire, release, or volatile mem_bar.
2927 insert_mem_bar(Op_MemBarCPUOrder);
2928 switch(id) {
2929 case vmIntrinsics::_loadFence:
2930 insert_mem_bar(Op_LoadFence);
2931 return true;
2932 case vmIntrinsics::_storeFence:
2933 insert_mem_bar(Op_StoreFence);
2934 return true;
2935 case vmIntrinsics::_storeStoreFence:
2936 insert_mem_bar(Op_StoreStoreFence);
2937 return true;
2938 case vmIntrinsics::_fullFence:
2939 insert_mem_bar(Op_MemBarVolatile);
2940 return true;
2941 default:
2942 fatal_unexpected_iid(id);
2943 return false;
2944 }
2945 }
2946
2947 bool LibraryCallKit::inline_onspinwait() {
2948 insert_mem_bar(Op_OnSpinWait);
2949 return true;
2950 }
2951
2952 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2953 if (!kls->is_Con()) {
2954 return true;
2955 }
2956 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2957 if (klsptr == nullptr) {
2958 return true;
2959 }
2960 ciInstanceKlass* ik = klsptr->instance_klass();
2961 // don't need a guard for a klass that is already initialized
2962 return !ik->is_initialized();
2963 }
2964
2965 //----------------------------inline_unsafe_writeback0-------------------------
2966 // public native void Unsafe.writeback0(long address)
2967 bool LibraryCallKit::inline_unsafe_writeback0() {
2968 if (!Matcher::has_match_rule(Op_CacheWB)) {
2969 return false;
2970 }
2971 #ifndef PRODUCT
2972 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2973 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2974 ciSignature* sig = callee()->signature();
2975 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2976 #endif
2977 null_check_receiver(); // null-check, then ignore
2978 Node *addr = argument(1);
2979 addr = new CastX2PNode(addr);
2980 addr = _gvn.transform(addr);
2981 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2982 flush = _gvn.transform(flush);
2983 set_memory(flush, TypeRawPtr::BOTTOM);
2984 return true;
2985 }
2986
2987 //----------------------------inline_unsafe_writeback0-------------------------
2988 // public native void Unsafe.writeback0(long address)
2989 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2990 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2991 return false;
2992 }
2993 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2994 return false;
2995 }
2996 #ifndef PRODUCT
2997 assert(Matcher::has_match_rule(Op_CacheWB),
2998 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2999 : "found match rule for CacheWBPostSync but not CacheWB"));
3000
3001 #endif
3002 null_check_receiver(); // null-check, then ignore
3003 Node *sync;
3004 if (is_pre) {
3005 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3006 } else {
3007 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3008 }
3009 sync = _gvn.transform(sync);
3010 set_memory(sync, TypeRawPtr::BOTTOM);
3011 return true;
3012 }
3013
3014 //----------------------------inline_unsafe_allocate---------------------------
3015 // public native Object Unsafe.allocateInstance(Class<?> cls);
3016 bool LibraryCallKit::inline_unsafe_allocate() {
3017
3018 #if INCLUDE_JVMTI
3019 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3020 return false;
3021 }
3022 #endif //INCLUDE_JVMTI
3023
3024 if (callee()->is_static()) return false; // caller must have the capability!
3025
3026 null_check_receiver(); // null-check, then ignore
3027 Node* cls = null_check(argument(1));
3028 if (stopped()) return true;
3029
3030 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
3031 kls = null_check(kls);
3032 if (stopped()) return true; // argument was like int.class
3033
3034 #if INCLUDE_JVMTI
3035 // Don't try to access new allocated obj in the intrinsic.
3036 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
3037 // Deoptimize and allocate in interpreter instead.
3038 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
3039 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
3040 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
3041 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
3042 {
3043 BuildCutout unless(this, tst, PROB_MAX);
3044 uncommon_trap(Deoptimization::Reason_intrinsic,
3045 Deoptimization::Action_make_not_entrant);
3046 }
3047 if (stopped()) {
3048 return true;
3049 }
3050 #endif //INCLUDE_JVMTI
3051
3052 Node* test = nullptr;
3053 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3054 // Note: The argument might still be an illegal value like
3055 // Serializable.class or Object[].class. The runtime will handle it.
3056 // But we must make an explicit check for initialization.
3057 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3058 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3059 // can generate code to load it as unsigned byte.
3060 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
3061 Node* bits = intcon(InstanceKlass::fully_initialized);
3062 test = _gvn.transform(new SubINode(inst, bits));
3063 // The 'test' is non-zero if we need to take a slow path.
3064 }
3065 Node* obj = nullptr;
3066 const TypeInstKlassPtr* tkls = _gvn.type(kls)->isa_instklassptr();
3067 if (tkls != nullptr && tkls->instance_klass()->is_inlinetype()) {
3068 obj = InlineTypeNode::make_default(_gvn, tkls->instance_klass()->as_inline_klass())->buffer(this);
3069 } else {
3070 obj = new_instance(kls, test);
3071 }
3072 set_result(obj);
3073 return true;
3074 }
3075
3076 //------------------------inline_native_time_funcs--------------
3077 // inline code for System.currentTimeMillis() and System.nanoTime()
3078 // these have the same type and signature
3079 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3080 const TypeFunc* tf = OptoRuntime::void_long_Type();
3081 const TypePtr* no_memory_effects = nullptr;
3082 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3083 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3084 #ifdef ASSERT
3085 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3086 assert(value_top == top(), "second value must be top");
3087 #endif
3088 set_result(value);
3089 return true;
3090 }
3091
3092
3093 #if INCLUDE_JVMTI
3094
3095 // When notifications are disabled then just update the VTMS transition bit and return.
3096 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
3097 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
3098 if (!DoJVMTIVirtualThreadTransitions) {
3099 return true;
3100 }
3101 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3102 IdealKit ideal(this);
3103
3104 Node* ONE = ideal.ConI(1);
3105 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
3106 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
3107 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3108
3109 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
3110 sync_kit(ideal);
3111 // if notifyJvmti enabled then make a call to the given SharedRuntime function
3112 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
3113 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
3114 ideal.sync_kit(this);
3115 } ideal.else_(); {
3116 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
3117 Node* thread = ideal.thread();
3118 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
3119 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
3120 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3121
3122 sync_kit(ideal);
3123 access_store_at(nullptr, jt_addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3124 access_store_at(nullptr, vt_addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3125
3126 ideal.sync_kit(this);
3127 } ideal.end_if();
3128 final_sync(ideal);
3129
3130 return true;
3131 }
3132
3133 // Always update the temporary VTMS transition bit.
3134 bool LibraryCallKit::inline_native_notify_jvmti_hide() {
3135 if (!DoJVMTIVirtualThreadTransitions) {
3136 return true;
3137 }
3138 IdealKit ideal(this);
3139
3140 {
3141 // unconditionally update the temporary VTMS transition bit in current JavaThread
3142 Node* thread = ideal.thread();
3143 Node* hide = _gvn.transform(argument(1)); // hide argument for temporary VTMS transition notification
3144 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_tmp_VTMS_transition_offset()));
3145 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3146
3147 sync_kit(ideal);
3148 access_store_at(nullptr, addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3149 ideal.sync_kit(this);
3150 }
3151 final_sync(ideal);
3152
3153 return true;
3154 }
3155
3156 #endif // INCLUDE_JVMTI
3157
3158 #ifdef JFR_HAVE_INTRINSICS
3159
3160 /**
3161 * if oop->klass != null
3162 * // normal class
3163 * epoch = _epoch_state ? 2 : 1
3164 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3165 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3166 * }
3167 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3168 * else
3169 * // primitive class
3170 * if oop->array_klass != null
3171 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3172 * else
3173 * id = LAST_TYPE_ID + 1 // void class path
3174 * if (!signaled)
3175 * signaled = true
3176 */
3177 bool LibraryCallKit::inline_native_classID() {
3178 Node* cls = argument(0);
3179
3180 IdealKit ideal(this);
3181 #define __ ideal.
3182 IdealVariable result(ideal); __ declarations_done();
3183 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3184 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3185 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3186
3187
3188 __ if_then(kls, BoolTest::ne, null()); {
3189 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3190 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3191
3192 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3193 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3194 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3195 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3196 mask = _gvn.transform(new OrLNode(mask, epoch));
3197 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3198
3199 float unlikely = PROB_UNLIKELY(0.999);
3200 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3201 sync_kit(ideal);
3202 make_runtime_call(RC_LEAF,
3203 OptoRuntime::class_id_load_barrier_Type(),
3204 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3205 "class id load barrier",
3206 TypePtr::BOTTOM,
3207 kls);
3208 ideal.sync_kit(this);
3209 } __ end_if();
3210
3211 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3212 } __ else_(); {
3213 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3214 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3215 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3216 __ if_then(array_kls, BoolTest::ne, null()); {
3217 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3218 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3219 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3220 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3221 } __ else_(); {
3222 // void class case
3223 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3224 } __ end_if();
3225
3226 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3227 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3228 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3229 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3230 } __ end_if();
3231 } __ end_if();
3232
3233 final_sync(ideal);
3234 set_result(ideal.value(result));
3235 #undef __
3236 return true;
3237 }
3238
3239 //------------------------inline_native_jvm_commit------------------
3240 bool LibraryCallKit::inline_native_jvm_commit() {
3241 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3242
3243 // Save input memory and i_o state.
3244 Node* input_memory_state = reset_memory();
3245 set_all_memory(input_memory_state);
3246 Node* input_io_state = i_o();
3247
3248 // TLS.
3249 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3250 // Jfr java buffer.
3251 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3252 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3253 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3254
3255 // Load the current value of the notified field in the JfrThreadLocal.
3256 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3257 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3258
3259 // Test for notification.
3260 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3261 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3262 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3263
3264 // True branch, is notified.
3265 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3266 set_control(is_notified);
3267
3268 // Reset notified state.
3269 Node* notified_reset_memory = store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered);
3270
3271 // 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.
3272 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3273 // Convert the machine-word to a long.
3274 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3275
3276 // False branch, not notified.
3277 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3278 set_control(not_notified);
3279 set_all_memory(input_memory_state);
3280
3281 // Arg is the next position as a long.
3282 Node* arg = argument(0);
3283 // Convert long to machine-word.
3284 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3285
3286 // Store the next_position to the underlying jfr java buffer.
3287 Node* commit_memory;
3288 #ifdef _LP64
3289 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_LONG, Compile::AliasIdxRaw, MemNode::release);
3290 #else
3291 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_INT, Compile::AliasIdxRaw, MemNode::release);
3292 #endif
3293
3294 // 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.
3295 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3296 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3297 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3298
3299 // And flags with lease constant.
3300 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3301
3302 // Branch on lease to conditionalize returning the leased java buffer.
3303 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3304 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3305 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3306
3307 // False branch, not a lease.
3308 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3309
3310 // True branch, is lease.
3311 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3312 set_control(is_lease);
3313
3314 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3315 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3316 OptoRuntime::void_void_Type(),
3317 StubRoutines::jfr_return_lease(),
3318 "return_lease", TypePtr::BOTTOM);
3319 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3320
3321 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3322 record_for_igvn(lease_compare_rgn);
3323 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3324 record_for_igvn(lease_compare_mem);
3325 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3326 record_for_igvn(lease_compare_io);
3327 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3328 record_for_igvn(lease_result_value);
3329
3330 // Update control and phi nodes.
3331 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3332 lease_compare_rgn->init_req(_false_path, not_lease);
3333
3334 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3335 lease_compare_mem->init_req(_false_path, commit_memory);
3336
3337 lease_compare_io->init_req(_true_path, i_o());
3338 lease_compare_io->init_req(_false_path, input_io_state);
3339
3340 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3341 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3342
3343 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3344 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3345 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3346 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3347
3348 // Update control and phi nodes.
3349 result_rgn->init_req(_true_path, is_notified);
3350 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3351
3352 result_mem->init_req(_true_path, notified_reset_memory);
3353 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3354
3355 result_io->init_req(_true_path, input_io_state);
3356 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3357
3358 result_value->init_req(_true_path, current_pos);
3359 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3360
3361 // Set output state.
3362 set_control(_gvn.transform(result_rgn));
3363 set_all_memory(_gvn.transform(result_mem));
3364 set_i_o(_gvn.transform(result_io));
3365 set_result(result_rgn, result_value);
3366 return true;
3367 }
3368
3369 /*
3370 * The intrinsic is a model of this pseudo-code:
3371 *
3372 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3373 * jobject h_event_writer = tl->java_event_writer();
3374 * if (h_event_writer == nullptr) {
3375 * return nullptr;
3376 * }
3377 * oop threadObj = Thread::threadObj();
3378 * oop vthread = java_lang_Thread::vthread(threadObj);
3379 * traceid tid;
3380 * bool excluded;
3381 * if (vthread != threadObj) { // i.e. current thread is virtual
3382 * tid = java_lang_Thread::tid(vthread);
3383 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3384 * excluded = vthread_epoch_raw & excluded_mask;
3385 * if (!excluded) {
3386 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3387 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3388 * if (vthread_epoch != current_epoch) {
3389 * write_checkpoint();
3390 * }
3391 * }
3392 * } else {
3393 * tid = java_lang_Thread::tid(threadObj);
3394 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3395 * excluded = thread_epoch_raw & excluded_mask;
3396 * }
3397 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3398 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3399 * if (tid_in_event_writer != tid) {
3400 * setField(event_writer, "threadID", tid);
3401 * setField(event_writer, "excluded", excluded);
3402 * }
3403 * return event_writer
3404 */
3405 bool LibraryCallKit::inline_native_getEventWriter() {
3406 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3407
3408 // Save input memory and i_o state.
3409 Node* input_memory_state = reset_memory();
3410 set_all_memory(input_memory_state);
3411 Node* input_io_state = i_o();
3412
3413 Node* excluded_mask = _gvn.intcon(32768);
3414 Node* epoch_mask = _gvn.intcon(32767);
3415
3416 // TLS
3417 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3418
3419 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3420 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3421
3422 // Load the eventwriter jobject handle.
3423 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3424
3425 // Null check the jobject handle.
3426 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3427 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3428 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3429
3430 // False path, jobj is null.
3431 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3432
3433 // True path, jobj is not null.
3434 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3435
3436 set_control(jobj_is_not_null);
3437
3438 // Load the threadObj for the CarrierThread.
3439 Node* threadObj = generate_current_thread(tls_ptr);
3440
3441 // Load the vthread.
3442 Node* vthread = generate_virtual_thread(tls_ptr);
3443
3444 // If vthread != threadObj, this is a virtual thread.
3445 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3446 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3447 IfNode* iff_vthread_not_equal_threadObj =
3448 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3449
3450 // False branch, fallback to threadObj.
3451 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3452 set_control(vthread_equal_threadObj);
3453
3454 // Load the tid field from the vthread object.
3455 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3456
3457 // Load the raw epoch value from the threadObj.
3458 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3459 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3460 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3461
3462 // Mask off the excluded information from the epoch.
3463 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3464
3465 // True branch, this is a virtual thread.
3466 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3467 set_control(vthread_not_equal_threadObj);
3468
3469 // Load the tid field from the vthread object.
3470 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3471
3472 // Load the raw epoch value from the vthread.
3473 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3474 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3475 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3476
3477 // Mask off the excluded information from the epoch.
3478 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3479
3480 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3481 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3482 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3483 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3484
3485 // False branch, vthread is excluded, no need to write epoch info.
3486 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3487
3488 // True branch, vthread is included, update epoch info.
3489 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3490 set_control(included);
3491
3492 // Get epoch value.
3493 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3494
3495 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3496 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3497 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3498
3499 // Compare the epoch in the vthread to the current epoch generation.
3500 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3501 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3502 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3503
3504 // False path, epoch is equal, checkpoint information is valid.
3505 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3506
3507 // True path, epoch is not equal, write a checkpoint for the vthread.
3508 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3509
3510 set_control(epoch_is_not_equal);
3511
3512 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3513 // The call also updates the native thread local thread id and the vthread with the current epoch.
3514 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3515 OptoRuntime::jfr_write_checkpoint_Type(),
3516 StubRoutines::jfr_write_checkpoint(),
3517 "write_checkpoint", TypePtr::BOTTOM);
3518 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3519
3520 // vthread epoch != current epoch
3521 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3522 record_for_igvn(epoch_compare_rgn);
3523 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3524 record_for_igvn(epoch_compare_mem);
3525 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3526 record_for_igvn(epoch_compare_io);
3527
3528 // Update control and phi nodes.
3529 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3530 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3531 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3532 epoch_compare_mem->init_req(_false_path, input_memory_state);
3533 epoch_compare_io->init_req(_true_path, i_o());
3534 epoch_compare_io->init_req(_false_path, input_io_state);
3535
3536 // excluded != true
3537 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3538 record_for_igvn(exclude_compare_rgn);
3539 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3540 record_for_igvn(exclude_compare_mem);
3541 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3542 record_for_igvn(exclude_compare_io);
3543
3544 // Update control and phi nodes.
3545 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3546 exclude_compare_rgn->init_req(_false_path, excluded);
3547 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3548 exclude_compare_mem->init_req(_false_path, input_memory_state);
3549 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3550 exclude_compare_io->init_req(_false_path, input_io_state);
3551
3552 // vthread != threadObj
3553 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3554 record_for_igvn(vthread_compare_rgn);
3555 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3556 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3557 record_for_igvn(vthread_compare_io);
3558 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3559 record_for_igvn(tid);
3560 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3561 record_for_igvn(exclusion);
3562
3563 // Update control and phi nodes.
3564 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3565 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3566 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3567 vthread_compare_mem->init_req(_false_path, input_memory_state);
3568 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3569 vthread_compare_io->init_req(_false_path, input_io_state);
3570 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3571 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3572 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3573 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3574
3575 // Update branch state.
3576 set_control(_gvn.transform(vthread_compare_rgn));
3577 set_all_memory(_gvn.transform(vthread_compare_mem));
3578 set_i_o(_gvn.transform(vthread_compare_io));
3579
3580 // Load the event writer oop by dereferencing the jobject handle.
3581 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3582 assert(klass_EventWriter->is_loaded(), "invariant");
3583 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3584 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3585 const TypeOopPtr* const xtype = aklass->as_instance_type();
3586 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3587 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3588
3589 // Load the current thread id from the event writer object.
3590 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3591 // Get the field offset to, conditionally, store an updated tid value later.
3592 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3593 const TypePtr* event_writer_tid_field_type = _gvn.type(event_writer_tid_field)->isa_ptr();
3594 // Get the field offset to, conditionally, store an updated exclusion value later.
3595 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3596 const TypePtr* event_writer_excluded_field_type = _gvn.type(event_writer_excluded_field)->isa_ptr();
3597
3598 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3599 record_for_igvn(event_writer_tid_compare_rgn);
3600 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3601 record_for_igvn(event_writer_tid_compare_mem);
3602 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3603 record_for_igvn(event_writer_tid_compare_io);
3604
3605 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3606 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3607 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3608 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3609
3610 // False path, tids are the same.
3611 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3612
3613 // True path, tid is not equal, need to update the tid in the event writer.
3614 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3615 record_for_igvn(tid_is_not_equal);
3616
3617 // Store the exclusion state to the event writer.
3618 store_to_memory(tid_is_not_equal, event_writer_excluded_field, _gvn.transform(exclusion), T_BOOLEAN, event_writer_excluded_field_type, MemNode::unordered);
3619
3620 // Store the tid to the event writer.
3621 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, event_writer_tid_field_type, MemNode::unordered);
3622
3623 // Update control and phi nodes.
3624 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3625 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3626 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3627 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3628 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3629 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3630
3631 // Result of top level CFG, Memory, IO and Value.
3632 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3633 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3634 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3635 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3636
3637 // Result control.
3638 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3639 result_rgn->init_req(_false_path, jobj_is_null);
3640
3641 // Result memory.
3642 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3643 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3644
3645 // Result IO.
3646 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3647 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3648
3649 // Result value.
3650 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3651 result_value->init_req(_false_path, null()); // return null
3652
3653 // Set output state.
3654 set_control(_gvn.transform(result_rgn));
3655 set_all_memory(_gvn.transform(result_mem));
3656 set_i_o(_gvn.transform(result_io));
3657 set_result(result_rgn, result_value);
3658 return true;
3659 }
3660
3661 /*
3662 * The intrinsic is a model of this pseudo-code:
3663 *
3664 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3665 * if (carrierThread != thread) { // is virtual thread
3666 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3667 * bool excluded = vthread_epoch_raw & excluded_mask;
3668 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3669 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3670 * if (!excluded) {
3671 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3672 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3673 * }
3674 * Atomic::release_store(&tl->_vthread, true);
3675 * return;
3676 * }
3677 * Atomic::release_store(&tl->_vthread, false);
3678 */
3679 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3680 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3681
3682 Node* input_memory_state = reset_memory();
3683 set_all_memory(input_memory_state);
3684
3685 Node* excluded_mask = _gvn.intcon(32768);
3686 Node* epoch_mask = _gvn.intcon(32767);
3687
3688 Node* const carrierThread = generate_current_thread(jt);
3689 // If thread != carrierThread, this is a virtual thread.
3690 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3691 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3692 IfNode* iff_thread_not_equal_carrierThread =
3693 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3694
3695 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3696
3697 // False branch, is carrierThread.
3698 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3699 // Store release
3700 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3701
3702 set_all_memory(input_memory_state);
3703
3704 // True branch, is virtual thread.
3705 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3706 set_control(thread_not_equal_carrierThread);
3707
3708 // Load the raw epoch value from the vthread.
3709 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3710 Node* epoch_raw = access_load_at(thread, epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3711 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3712
3713 // Mask off the excluded information from the epoch.
3714 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3715
3716 // Load the tid field from the thread.
3717 Node* tid = load_field_from_object(thread, "tid", "J");
3718
3719 // Store the vthread tid to the jfr thread local.
3720 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3721 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, Compile::AliasIdxRaw, MemNode::unordered, true);
3722
3723 // Branch is_excluded to conditionalize updating the epoch .
3724 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3725 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3726 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3727
3728 // True branch, vthread is excluded, no need to write epoch info.
3729 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3730 set_control(excluded);
3731 Node* vthread_is_excluded = _gvn.intcon(1);
3732
3733 // False branch, vthread is included, update epoch info.
3734 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3735 set_control(included);
3736 Node* vthread_is_included = _gvn.intcon(0);
3737
3738 // Get epoch value.
3739 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3740
3741 // Store the vthread epoch to the jfr thread local.
3742 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3743 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, Compile::AliasIdxRaw, MemNode::unordered, true);
3744
3745 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3746 record_for_igvn(excluded_rgn);
3747 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3748 record_for_igvn(excluded_mem);
3749 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3750 record_for_igvn(exclusion);
3751
3752 // Merge the excluded control and memory.
3753 excluded_rgn->init_req(_true_path, excluded);
3754 excluded_rgn->init_req(_false_path, included);
3755 excluded_mem->init_req(_true_path, tid_memory);
3756 excluded_mem->init_req(_false_path, included_memory);
3757 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3758 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3759
3760 // Set intermediate state.
3761 set_control(_gvn.transform(excluded_rgn));
3762 set_all_memory(excluded_mem);
3763
3764 // Store the vthread exclusion state to the jfr thread local.
3765 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3766 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered, true);
3767
3768 // Store release
3769 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3770
3771 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3772 record_for_igvn(thread_compare_rgn);
3773 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3774 record_for_igvn(thread_compare_mem);
3775 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3776 record_for_igvn(vthread);
3777
3778 // Merge the thread_compare control and memory.
3779 thread_compare_rgn->init_req(_true_path, control());
3780 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3781 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3782 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3783
3784 // Set output state.
3785 set_control(_gvn.transform(thread_compare_rgn));
3786 set_all_memory(_gvn.transform(thread_compare_mem));
3787 }
3788
3789 #endif // JFR_HAVE_INTRINSICS
3790
3791 //------------------------inline_native_currentCarrierThread------------------
3792 bool LibraryCallKit::inline_native_currentCarrierThread() {
3793 Node* junk = nullptr;
3794 set_result(generate_current_thread(junk));
3795 return true;
3796 }
3797
3798 //------------------------inline_native_currentThread------------------
3799 bool LibraryCallKit::inline_native_currentThread() {
3800 Node* junk = nullptr;
3801 set_result(generate_virtual_thread(junk));
3802 return true;
3803 }
3804
3805 //------------------------inline_native_setVthread------------------
3806 bool LibraryCallKit::inline_native_setCurrentThread() {
3807 assert(C->method()->changes_current_thread(),
3808 "method changes current Thread but is not annotated ChangesCurrentThread");
3809 Node* arr = argument(1);
3810 Node* thread = _gvn.transform(new ThreadLocalNode());
3811 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3812 Node* thread_obj_handle
3813 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3814 thread_obj_handle = _gvn.transform(thread_obj_handle);
3815 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3816 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3817 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3818 return true;
3819 }
3820
3821 const Type* LibraryCallKit::scopedValueCache_type() {
3822 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3823 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3824 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS, /* stable= */ false, /* flat= */ false, /* not_flat= */ true, /* not_null_free= */ true);
3825
3826 // Because we create the scopedValue cache lazily we have to make the
3827 // type of the result BotPTR.
3828 bool xk = etype->klass_is_exact();
3829 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, TypeAryPtr::Offset(0));
3830 return objects_type;
3831 }
3832
3833 Node* LibraryCallKit::scopedValueCache_helper() {
3834 Node* thread = _gvn.transform(new ThreadLocalNode());
3835 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3836 // We cannot use immutable_memory() because we might flip onto a
3837 // different carrier thread, at which point we'll need to use that
3838 // carrier thread's cache.
3839 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3840 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3841 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3842 }
3843
3844 //------------------------inline_native_scopedValueCache------------------
3845 bool LibraryCallKit::inline_native_scopedValueCache() {
3846 Node* cache_obj_handle = scopedValueCache_helper();
3847 const Type* objects_type = scopedValueCache_type();
3848 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3849
3850 return true;
3851 }
3852
3853 //------------------------inline_native_setScopedValueCache------------------
3854 bool LibraryCallKit::inline_native_setScopedValueCache() {
3855 Node* arr = argument(0);
3856 Node* cache_obj_handle = scopedValueCache_helper();
3857 const Type* objects_type = scopedValueCache_type();
3858
3859 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3860 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3861
3862 return true;
3863 }
3864
3865 //-----------------------load_klass_from_mirror_common-------------------------
3866 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3867 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3868 // and branch to the given path on the region.
3869 // If never_see_null, take an uncommon trap on null, so we can optimistically
3870 // compile for the non-null case.
3871 // If the region is null, force never_see_null = true.
3872 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3873 bool never_see_null,
3874 RegionNode* region,
3875 int null_path,
3876 int offset) {
3877 if (region == nullptr) never_see_null = true;
3878 Node* p = basic_plus_adr(mirror, offset);
3879 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3880 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3881 Node* null_ctl = top();
3882 kls = null_check_oop(kls, &null_ctl, never_see_null);
3883 if (region != nullptr) {
3884 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3885 region->init_req(null_path, null_ctl);
3886 } else {
3887 assert(null_ctl == top(), "no loose ends");
3888 }
3889 return kls;
3890 }
3891
3892 //--------------------(inline_native_Class_query helpers)---------------------
3893 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3894 // Fall through if (mods & mask) == bits, take the guard otherwise.
3895 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3896 // Branch around if the given klass has the given modifier bit set.
3897 // Like generate_guard, adds a new path onto the region.
3898 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3899 Node* mods = make_load(nullptr, modp, TypeInt::INT, T_INT, MemNode::unordered);
3900 Node* mask = intcon(modifier_mask);
3901 Node* bits = intcon(modifier_bits);
3902 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3903 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3904 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3905 return generate_fair_guard(bol, region);
3906 }
3907
3908 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3909 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3910 }
3911 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3912 return generate_access_flags_guard(kls, JVM_ACC_IS_HIDDEN_CLASS, 0, region);
3913 }
3914
3915 //-------------------------inline_native_Class_query-------------------
3916 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3917 const Type* return_type = TypeInt::BOOL;
3918 Node* prim_return_value = top(); // what happens if it's a primitive class?
3919 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3920 bool expect_prim = false; // most of these guys expect to work on refs
3921
3922 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3923
3924 Node* mirror = argument(0);
3925 Node* obj = top();
3926
3927 switch (id) {
3928 case vmIntrinsics::_isInstance:
3929 // nothing is an instance of a primitive type
3930 prim_return_value = intcon(0);
3931 obj = argument(1);
3932 break;
3933 case vmIntrinsics::_getModifiers:
3934 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3935 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3936 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3937 break;
3938 case vmIntrinsics::_isInterface:
3939 prim_return_value = intcon(0);
3940 break;
3941 case vmIntrinsics::_isArray:
3942 prim_return_value = intcon(0);
3943 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3944 break;
3945 case vmIntrinsics::_isPrimitive:
3946 prim_return_value = intcon(1);
3947 expect_prim = true; // obviously
3948 break;
3949 case vmIntrinsics::_isHidden:
3950 prim_return_value = intcon(0);
3951 break;
3952 case vmIntrinsics::_getSuperclass:
3953 prim_return_value = null();
3954 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3955 break;
3956 case vmIntrinsics::_getClassAccessFlags:
3957 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3958 return_type = TypeInt::INT; // not bool! 6297094
3959 break;
3960 default:
3961 fatal_unexpected_iid(id);
3962 break;
3963 }
3964
3965 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3966 if (mirror_con == nullptr) return false; // cannot happen?
3967
3968 #ifndef PRODUCT
3969 if (C->print_intrinsics() || C->print_inlining()) {
3970 ciType* k = mirror_con->java_mirror_type();
3971 if (k) {
3972 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3973 k->print_name();
3974 tty->cr();
3975 }
3976 }
3977 #endif
3978
3979 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3980 RegionNode* region = new RegionNode(PATH_LIMIT);
3981 record_for_igvn(region);
3982 PhiNode* phi = new PhiNode(region, return_type);
3983
3984 // The mirror will never be null of Reflection.getClassAccessFlags, however
3985 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3986 // if it is. See bug 4774291.
3987
3988 // For Reflection.getClassAccessFlags(), the null check occurs in
3989 // the wrong place; see inline_unsafe_access(), above, for a similar
3990 // situation.
3991 mirror = null_check(mirror);
3992 // If mirror or obj is dead, only null-path is taken.
3993 if (stopped()) return true;
3994
3995 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3996
3997 // Now load the mirror's klass metaobject, and null-check it.
3998 // Side-effects region with the control path if the klass is null.
3999 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4000 // If kls is null, we have a primitive mirror.
4001 phi->init_req(_prim_path, prim_return_value);
4002 if (stopped()) { set_result(region, phi); return true; }
4003 bool safe_for_replace = (region->in(_prim_path) == top());
4004
4005 Node* p; // handy temp
4006 Node* null_ctl;
4007
4008 // Now that we have the non-null klass, we can perform the real query.
4009 // For constant classes, the query will constant-fold in LoadNode::Value.
4010 Node* query_value = top();
4011 switch (id) {
4012 case vmIntrinsics::_isInstance:
4013 // nothing is an instance of a primitive type
4014 query_value = gen_instanceof(obj, kls, safe_for_replace);
4015 break;
4016
4017 case vmIntrinsics::_getModifiers:
4018 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
4019 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
4020 break;
4021
4022 case vmIntrinsics::_isInterface:
4023 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4024 if (generate_interface_guard(kls, region) != nullptr)
4025 // A guard was added. If the guard is taken, it was an interface.
4026 phi->add_req(intcon(1));
4027 // If we fall through, it's a plain class.
4028 query_value = intcon(0);
4029 break;
4030
4031 case vmIntrinsics::_isArray:
4032 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
4033 if (generate_array_guard(kls, region) != nullptr)
4034 // A guard was added. If the guard is taken, it was an array.
4035 phi->add_req(intcon(1));
4036 // If we fall through, it's a plain class.
4037 query_value = intcon(0);
4038 break;
4039
4040 case vmIntrinsics::_isPrimitive:
4041 query_value = intcon(0); // "normal" path produces false
4042 break;
4043
4044 case vmIntrinsics::_isHidden:
4045 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4046 if (generate_hidden_class_guard(kls, region) != nullptr)
4047 // A guard was added. If the guard is taken, it was an hidden class.
4048 phi->add_req(intcon(1));
4049 // If we fall through, it's a plain class.
4050 query_value = intcon(0);
4051 break;
4052
4053
4054 case vmIntrinsics::_getSuperclass:
4055 // The rules here are somewhat unfortunate, but we can still do better
4056 // with random logic than with a JNI call.
4057 // Interfaces store null or Object as _super, but must report null.
4058 // Arrays store an intermediate super as _super, but must report Object.
4059 // Other types can report the actual _super.
4060 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4061 if (generate_interface_guard(kls, region) != nullptr)
4062 // A guard was added. If the guard is taken, it was an interface.
4063 phi->add_req(null());
4064 if (generate_array_guard(kls, region) != nullptr)
4065 // A guard was added. If the guard is taken, it was an array.
4066 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4067 // If we fall through, it's a plain class. Get its _super.
4068 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4069 kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4070 null_ctl = top();
4071 kls = null_check_oop(kls, &null_ctl);
4072 if (null_ctl != top()) {
4073 // If the guard is taken, Object.superClass is null (both klass and mirror).
4074 region->add_req(null_ctl);
4075 phi ->add_req(null());
4076 }
4077 if (!stopped()) {
4078 query_value = load_mirror_from_klass(kls);
4079 }
4080 break;
4081
4082 case vmIntrinsics::_getClassAccessFlags:
4083 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
4084 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
4085 break;
4086
4087 default:
4088 fatal_unexpected_iid(id);
4089 break;
4090 }
4091
4092 // Fall-through is the normal case of a query to a real class.
4093 phi->init_req(1, query_value);
4094 region->init_req(1, control());
4095
4096 C->set_has_split_ifs(true); // Has chance for split-if optimization
4097 set_result(region, phi);
4098 return true;
4099 }
4100
4101
4102 //-------------------------inline_Class_cast-------------------
4103 bool LibraryCallKit::inline_Class_cast() {
4104 Node* mirror = argument(0); // Class
4105 Node* obj = argument(1);
4106 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4107 if (mirror_con == nullptr) {
4108 return false; // dead path (mirror->is_top()).
4109 }
4110 if (obj == nullptr || obj->is_top()) {
4111 return false; // dead path
4112 }
4113 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4114
4115 // First, see if Class.cast() can be folded statically.
4116 // java_mirror_type() returns non-null for compile-time Class constants.
4117 ciType* tm = mirror_con->java_mirror_type();
4118 if (tm != nullptr && tm->is_klass() &&
4119 tp != nullptr) {
4120 if (!tp->is_loaded()) {
4121 // Don't use intrinsic when class is not loaded.
4122 return false;
4123 } else {
4124 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
4125 if (static_res == Compile::SSC_always_true) {
4126 // isInstance() is true - fold the code.
4127 set_result(obj);
4128 return true;
4129 } else if (static_res == Compile::SSC_always_false) {
4130 // Don't use intrinsic, have to throw ClassCastException.
4131 // If the reference is null, the non-intrinsic bytecode will
4132 // be optimized appropriately.
4133 return false;
4134 }
4135 }
4136 }
4137
4138 // Bailout intrinsic and do normal inlining if exception path is frequent.
4139 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4140 return false;
4141 }
4142
4143 // Generate dynamic checks.
4144 // Class.cast() is java implementation of _checkcast bytecode.
4145 // Do checkcast (Parse::do_checkcast()) optimizations here.
4146
4147 mirror = null_check(mirror);
4148 // If mirror is dead, only null-path is taken.
4149 if (stopped()) {
4150 return true;
4151 }
4152
4153 // Not-subtype or the mirror's klass ptr is nullptr (in case it is a primitive).
4154 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
4155 RegionNode* region = new RegionNode(PATH_LIMIT);
4156 record_for_igvn(region);
4157
4158 // Now load the mirror's klass metaobject, and null-check it.
4159 // If kls is null, we have a primitive mirror and
4160 // nothing is an instance of a primitive type.
4161 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4162
4163 Node* res = top();
4164 Node* io = i_o();
4165 Node* mem = merged_memory();
4166 if (!stopped()) {
4167
4168 Node* bad_type_ctrl = top();
4169 // Do checkcast optimizations.
4170 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4171 region->init_req(_bad_type_path, bad_type_ctrl);
4172 }
4173 if (region->in(_prim_path) != top() ||
4174 region->in(_bad_type_path) != top() ||
4175 region->in(_npe_path) != top()) {
4176 // Let Interpreter throw ClassCastException.
4177 PreserveJVMState pjvms(this);
4178 set_control(_gvn.transform(region));
4179 // Set IO and memory because gen_checkcast may override them when buffering inline types
4180 set_i_o(io);
4181 set_all_memory(mem);
4182 uncommon_trap(Deoptimization::Reason_intrinsic,
4183 Deoptimization::Action_maybe_recompile);
4184 }
4185 if (!stopped()) {
4186 set_result(res);
4187 }
4188 return true;
4189 }
4190
4191
4192 //--------------------------inline_native_subtype_check------------------------
4193 // This intrinsic takes the JNI calls out of the heart of
4194 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4195 bool LibraryCallKit::inline_native_subtype_check() {
4196 // Pull both arguments off the stack.
4197 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4198 args[0] = argument(0);
4199 args[1] = argument(1);
4200 Node* klasses[2]; // corresponding Klasses: superk, subk
4201 klasses[0] = klasses[1] = top();
4202
4203 enum {
4204 // A full decision tree on {superc is prim, subc is prim}:
4205 _prim_0_path = 1, // {P,N} => false
4206 // {P,P} & superc!=subc => false
4207 _prim_same_path, // {P,P} & superc==subc => true
4208 _prim_1_path, // {N,P} => false
4209 _ref_subtype_path, // {N,N} & subtype check wins => true
4210 _both_ref_path, // {N,N} & subtype check loses => false
4211 PATH_LIMIT
4212 };
4213
4214 RegionNode* region = new RegionNode(PATH_LIMIT);
4215 RegionNode* prim_region = new RegionNode(2);
4216 Node* phi = new PhiNode(region, TypeInt::BOOL);
4217 record_for_igvn(region);
4218 record_for_igvn(prim_region);
4219
4220 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4221 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4222 int class_klass_offset = java_lang_Class::klass_offset();
4223
4224 // First null-check both mirrors and load each mirror's klass metaobject.
4225 int which_arg;
4226 for (which_arg = 0; which_arg <= 1; which_arg++) {
4227 Node* arg = args[which_arg];
4228 arg = null_check(arg);
4229 if (stopped()) break;
4230 args[which_arg] = arg;
4231
4232 Node* p = basic_plus_adr(arg, class_klass_offset);
4233 Node* kls = LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, adr_type, kls_type);
4234 klasses[which_arg] = _gvn.transform(kls);
4235 }
4236
4237 // Having loaded both klasses, test each for null.
4238 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4239 for (which_arg = 0; which_arg <= 1; which_arg++) {
4240 Node* kls = klasses[which_arg];
4241 Node* null_ctl = top();
4242 kls = null_check_oop(kls, &null_ctl, never_see_null);
4243 if (which_arg == 0) {
4244 prim_region->init_req(1, null_ctl);
4245 } else {
4246 region->init_req(_prim_1_path, null_ctl);
4247 }
4248 if (stopped()) break;
4249 klasses[which_arg] = kls;
4250 }
4251
4252 if (!stopped()) {
4253 // now we have two reference types, in klasses[0..1]
4254 Node* subk = klasses[1]; // the argument to isAssignableFrom
4255 Node* superk = klasses[0]; // the receiver
4256 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4257 region->set_req(_ref_subtype_path, control());
4258 }
4259
4260 // If both operands are primitive (both klasses null), then
4261 // we must return true when they are identical primitives.
4262 // It is convenient to test this after the first null klass check.
4263 // This path is also used if superc is a value mirror.
4264 set_control(_gvn.transform(prim_region));
4265 if (!stopped()) {
4266 // Since superc is primitive, make a guard for the superc==subc case.
4267 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4268 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4269 generate_fair_guard(bol_eq, region);
4270 if (region->req() == PATH_LIMIT+1) {
4271 // A guard was added. If the added guard is taken, superc==subc.
4272 region->swap_edges(PATH_LIMIT, _prim_same_path);
4273 region->del_req(PATH_LIMIT);
4274 }
4275 region->set_req(_prim_0_path, control()); // Not equal after all.
4276 }
4277
4278 // these are the only paths that produce 'true':
4279 phi->set_req(_prim_same_path, intcon(1));
4280 phi->set_req(_ref_subtype_path, intcon(1));
4281
4282 // pull together the cases:
4283 assert(region->req() == PATH_LIMIT, "sane region");
4284 for (uint i = 1; i < region->req(); i++) {
4285 Node* ctl = region->in(i);
4286 if (ctl == nullptr || ctl == top()) {
4287 region->set_req(i, top());
4288 phi ->set_req(i, top());
4289 } else if (phi->in(i) == nullptr) {
4290 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4291 }
4292 }
4293
4294 set_control(_gvn.transform(region));
4295 set_result(_gvn.transform(phi));
4296 return true;
4297 }
4298
4299 //---------------------generate_array_guard_common------------------------
4300 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind) {
4301
4302 if (stopped()) {
4303 return nullptr;
4304 }
4305
4306 // Like generate_guard, adds a new path onto the region.
4307 jint layout_con = 0;
4308 Node* layout_val = get_layout_helper(kls, layout_con);
4309 if (layout_val == nullptr) {
4310 bool query = 0;
4311 switch(kind) {
4312 case ObjectArray: query = Klass::layout_helper_is_objArray(layout_con); break;
4313 case NonObjectArray: query = !Klass::layout_helper_is_objArray(layout_con); break;
4314 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
4315 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
4316 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
4317 default:
4318 ShouldNotReachHere();
4319 }
4320 if (!query) {
4321 return nullptr; // never a branch
4322 } else { // always a branch
4323 Node* always_branch = control();
4324 if (region != nullptr)
4325 region->add_req(always_branch);
4326 set_control(top());
4327 return always_branch;
4328 }
4329 }
4330 unsigned int value = 0;
4331 BoolTest::mask btest = BoolTest::illegal;
4332 switch(kind) {
4333 case ObjectArray:
4334 case NonObjectArray: {
4335 value = Klass::_lh_array_tag_obj_value;
4336 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4337 btest = (kind == ObjectArray) ? BoolTest::eq : BoolTest::ne;
4338 break;
4339 }
4340 case TypeArray: {
4341 value = Klass::_lh_array_tag_type_value;
4342 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4343 btest = BoolTest::eq;
4344 break;
4345 }
4346 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
4347 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
4348 default:
4349 ShouldNotReachHere();
4350 }
4351 // Now test the correct condition.
4352 jint nval = (jint)value;
4353 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4354 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4355 return generate_fair_guard(bol, region);
4356 }
4357
4358 //-----------------------inline_newNullRestrictedArray--------------------------
4359 // public static native Object[] newNullRestrictedArray(Class<?> componentType, int length);
4360 bool LibraryCallKit::inline_newNullRestrictedArray() {
4361 Node* componentType = argument(0);
4362 Node* length = argument(1);
4363
4364 const TypeInstPtr* tp = _gvn.type(componentType)->isa_instptr();
4365 if (tp != nullptr) {
4366 ciInstanceKlass* ik = tp->instance_klass();
4367 if (ik == C->env()->Class_klass()) {
4368 ciType* t = tp->java_mirror_type();
4369 if (t != nullptr && t->is_inlinetype()) {
4370 ciArrayKlass* array_klass = ciArrayKlass::make(t, true);
4371 if (array_klass->is_loaded() && array_klass->element_klass()->as_inline_klass()->is_initialized()) {
4372 const TypeAryKlassPtr* array_klass_type = TypeKlassPtr::make(array_klass, Type::trust_interfaces)->is_aryklassptr();
4373 array_klass_type = array_klass_type->cast_to_null_free();
4374 Node* obj = new_array(makecon(array_klass_type), length, 0, nullptr, false); // no arguments to push
4375 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4376 alloc->set_null_free();
4377 set_result(obj);
4378 assert(gvn().type(obj)->is_aryptr()->is_null_free(), "must be null-free");
4379 return true;
4380 }
4381 }
4382 }
4383 }
4384 return false;
4385 }
4386
4387 //-----------------------inline_native_newArray--------------------------
4388 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
4389 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4390 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4391 Node* mirror;
4392 Node* count_val;
4393 if (uninitialized) {
4394 null_check_receiver();
4395 mirror = argument(1);
4396 count_val = argument(2);
4397 } else {
4398 mirror = argument(0);
4399 count_val = argument(1);
4400 }
4401
4402 mirror = null_check(mirror);
4403 // If mirror or obj is dead, only null-path is taken.
4404 if (stopped()) return true;
4405
4406 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4407 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4408 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4409 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4410 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4411
4412 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4413 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4414 result_reg, _slow_path);
4415 Node* normal_ctl = control();
4416 Node* no_array_ctl = result_reg->in(_slow_path);
4417
4418 // Generate code for the slow case. We make a call to newArray().
4419 set_control(no_array_ctl);
4420 if (!stopped()) {
4421 // Either the input type is void.class, or else the
4422 // array klass has not yet been cached. Either the
4423 // ensuing call will throw an exception, or else it
4424 // will cache the array klass for next time.
4425 PreserveJVMState pjvms(this);
4426 CallJavaNode* slow_call = nullptr;
4427 if (uninitialized) {
4428 // Generate optimized virtual call (holder class 'Unsafe' is final)
4429 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4430 } else {
4431 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4432 }
4433 Node* slow_result = set_results_for_java_call(slow_call);
4434 // this->control() comes from set_results_for_java_call
4435 result_reg->set_req(_slow_path, control());
4436 result_val->set_req(_slow_path, slow_result);
4437 result_io ->set_req(_slow_path, i_o());
4438 result_mem->set_req(_slow_path, reset_memory());
4439 }
4440
4441 set_control(normal_ctl);
4442 if (!stopped()) {
4443 // Normal case: The array type has been cached in the java.lang.Class.
4444 // The following call works fine even if the array type is polymorphic.
4445 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4446 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4447 result_reg->init_req(_normal_path, control());
4448 result_val->init_req(_normal_path, obj);
4449 result_io ->init_req(_normal_path, i_o());
4450 result_mem->init_req(_normal_path, reset_memory());
4451
4452 if (uninitialized) {
4453 // Mark the allocation so that zeroing is skipped
4454 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4455 alloc->maybe_set_complete(&_gvn);
4456 }
4457 }
4458
4459 // Return the combined state.
4460 set_i_o( _gvn.transform(result_io) );
4461 set_all_memory( _gvn.transform(result_mem));
4462
4463 C->set_has_split_ifs(true); // Has chance for split-if optimization
4464 set_result(result_reg, result_val);
4465 return true;
4466 }
4467
4468 //----------------------inline_native_getLength--------------------------
4469 // public static native int java.lang.reflect.Array.getLength(Object array);
4470 bool LibraryCallKit::inline_native_getLength() {
4471 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4472
4473 Node* array = null_check(argument(0));
4474 // If array is dead, only null-path is taken.
4475 if (stopped()) return true;
4476
4477 // Deoptimize if it is a non-array.
4478 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr);
4479
4480 if (non_array != nullptr) {
4481 PreserveJVMState pjvms(this);
4482 set_control(non_array);
4483 uncommon_trap(Deoptimization::Reason_intrinsic,
4484 Deoptimization::Action_maybe_recompile);
4485 }
4486
4487 // If control is dead, only non-array-path is taken.
4488 if (stopped()) return true;
4489
4490 // The works fine even if the array type is polymorphic.
4491 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4492 Node* result = load_array_length(array);
4493
4494 C->set_has_split_ifs(true); // Has chance for split-if optimization
4495 set_result(result);
4496 return true;
4497 }
4498
4499 //------------------------inline_array_copyOf----------------------------
4500 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4501 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4502 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4503 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4504
4505 // Get the arguments.
4506 Node* original = argument(0);
4507 Node* start = is_copyOfRange? argument(1): intcon(0);
4508 Node* end = is_copyOfRange? argument(2): argument(1);
4509 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4510
4511 Node* newcopy = nullptr;
4512
4513 // Set the original stack and the reexecute bit for the interpreter to reexecute
4514 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4515 { PreserveReexecuteState preexecs(this);
4516 jvms()->set_should_reexecute(true);
4517
4518 array_type_mirror = null_check(array_type_mirror);
4519 original = null_check(original);
4520
4521 // Check if a null path was taken unconditionally.
4522 if (stopped()) return true;
4523
4524 Node* orig_length = load_array_length(original);
4525
4526 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4527 klass_node = null_check(klass_node);
4528
4529 RegionNode* bailout = new RegionNode(1);
4530 record_for_igvn(bailout);
4531
4532 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4533 // Bail out if that is so.
4534 // Inline type array may have object field that would require a
4535 // write barrier. Conservatively, go to slow path.
4536 // TODO 8251971: Optimize for the case when flat src/dst are later found
4537 // to not contain oops (i.e., move this check to the macro expansion phase).
4538 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4539 const TypeAryPtr* orig_t = _gvn.type(original)->isa_aryptr();
4540 const TypeKlassPtr* tklass = _gvn.type(klass_node)->is_klassptr();
4541 bool exclude_flat = UseFlatArray && bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing) &&
4542 // Can src array be flat and contain oops?
4543 (orig_t == nullptr || (!orig_t->is_not_flat() && (!orig_t->is_flat() || orig_t->elem()->inline_klass()->contains_oops()))) &&
4544 // Can dest array be flat and contain oops?
4545 tklass->can_be_inline_array() && (!tklass->is_flat() || tklass->is_aryklassptr()->elem()->is_instklassptr()->instance_klass()->as_inline_klass()->contains_oops());
4546 Node* not_objArray = exclude_flat ? generate_non_objArray_guard(klass_node, bailout) : generate_typeArray_guard(klass_node, bailout);
4547 if (not_objArray != nullptr) {
4548 // Improve the klass node's type from the new optimistic assumption:
4549 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4550 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0));
4551 Node* cast = new CastPPNode(klass_node, akls);
4552 cast->init_req(0, control());
4553 klass_node = _gvn.transform(cast);
4554 }
4555
4556 // Bail out if either start or end is negative.
4557 generate_negative_guard(start, bailout, &start);
4558 generate_negative_guard(end, bailout, &end);
4559
4560 Node* length = end;
4561 if (_gvn.type(start) != TypeInt::ZERO) {
4562 length = _gvn.transform(new SubINode(end, start));
4563 }
4564
4565 // Bail out if length is negative.
4566 // Without this the new_array would throw
4567 // NegativeArraySizeException but IllegalArgumentException is what
4568 // should be thrown
4569 generate_negative_guard(length, bailout, &length);
4570
4571 // Handle inline type arrays
4572 bool can_validate = !too_many_traps(Deoptimization::Reason_class_check);
4573 if (!stopped()) {
4574 orig_t = _gvn.type(original)->isa_aryptr();
4575 if (orig_t != nullptr && orig_t->is_flat()) {
4576 // Src is flat, check that dest is flat as well
4577 if (exclude_flat) {
4578 // Dest can't be flat, bail out
4579 bailout->add_req(control());
4580 set_control(top());
4581 } else {
4582 generate_fair_guard(flat_array_test(klass_node, /* flat = */ false), bailout);
4583 }
4584 } else if (UseFlatArray && (orig_t == nullptr || !orig_t->is_not_flat()) &&
4585 // If dest is flat, src must be flat as well (guaranteed by src <: dest check if validated).
4586 ((!tklass->is_flat() && tklass->can_be_inline_array()) || !can_validate)) {
4587 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
4588 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
4589 generate_fair_guard(flat_array_test(load_object_klass(original)), bailout);
4590 if (orig_t != nullptr) {
4591 orig_t = orig_t->cast_to_not_flat();
4592 original = _gvn.transform(new CheckCastPPNode(control(), original, orig_t));
4593 }
4594 }
4595 if (!can_validate) {
4596 // No validation. The subtype check emitted at macro expansion time will not go to the slow
4597 // path but call checkcast_arraycopy which can not handle flat/null-free inline type arrays.
4598 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat/null-free.
4599 generate_fair_guard(null_free_array_test(klass_node), bailout);
4600 }
4601 }
4602
4603 if (bailout->req() > 1) {
4604 PreserveJVMState pjvms(this);
4605 set_control(_gvn.transform(bailout));
4606 uncommon_trap(Deoptimization::Reason_intrinsic,
4607 Deoptimization::Action_maybe_recompile);
4608 }
4609
4610 if (!stopped()) {
4611 // How many elements will we copy from the original?
4612 // The answer is MinI(orig_length - start, length).
4613 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4614 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4615
4616 // Generate a direct call to the right arraycopy function(s).
4617 // We know the copy is disjoint but we might not know if the
4618 // oop stores need checking.
4619 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4620 // This will fail a store-check if x contains any non-nulls.
4621
4622 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4623 // loads/stores but it is legal only if we're sure the
4624 // Arrays.copyOf would succeed. So we need all input arguments
4625 // to the copyOf to be validated, including that the copy to the
4626 // new array won't trigger an ArrayStoreException. That subtype
4627 // check can be optimized if we know something on the type of
4628 // the input array from type speculation.
4629 if (_gvn.type(klass_node)->singleton()) {
4630 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4631 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4632
4633 int test = C->static_subtype_check(superk, subk);
4634 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4635 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4636 if (t_original->speculative_type() != nullptr) {
4637 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4638 }
4639 }
4640 }
4641
4642 bool validated = false;
4643 // Reason_class_check rather than Reason_intrinsic because we
4644 // want to intrinsify even if this traps.
4645 if (can_validate) {
4646 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4647
4648 if (not_subtype_ctrl != top()) {
4649 PreserveJVMState pjvms(this);
4650 set_control(not_subtype_ctrl);
4651 uncommon_trap(Deoptimization::Reason_class_check,
4652 Deoptimization::Action_make_not_entrant);
4653 assert(stopped(), "Should be stopped");
4654 }
4655 validated = true;
4656 }
4657
4658 if (!stopped()) {
4659 newcopy = new_array(klass_node, length, 0); // no arguments to push
4660
4661 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
4662 load_object_klass(original), klass_node);
4663 if (!is_copyOfRange) {
4664 ac->set_copyof(validated);
4665 } else {
4666 ac->set_copyofrange(validated);
4667 }
4668 Node* n = _gvn.transform(ac);
4669 if (n == ac) {
4670 ac->connect_outputs(this);
4671 } else {
4672 assert(validated, "shouldn't transform if all arguments not validated");
4673 set_all_memory(n);
4674 }
4675 }
4676 }
4677 } // original reexecute is set back here
4678
4679 C->set_has_split_ifs(true); // Has chance for split-if optimization
4680 if (!stopped()) {
4681 set_result(newcopy);
4682 }
4683 return true;
4684 }
4685
4686
4687 //----------------------generate_virtual_guard---------------------------
4688 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4689 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4690 RegionNode* slow_region) {
4691 ciMethod* method = callee();
4692 int vtable_index = method->vtable_index();
4693 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4694 "bad index %d", vtable_index);
4695 // Get the Method* out of the appropriate vtable entry.
4696 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4697 vtable_index*vtableEntry::size_in_bytes() +
4698 in_bytes(vtableEntry::method_offset());
4699 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4700 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4701
4702 // Compare the target method with the expected method (e.g., Object.hashCode).
4703 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4704
4705 Node* native_call = makecon(native_call_addr);
4706 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4707 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4708
4709 return generate_slow_guard(test_native, slow_region);
4710 }
4711
4712 //-----------------------generate_method_call----------------------------
4713 // Use generate_method_call to make a slow-call to the real
4714 // method if the fast path fails. An alternative would be to
4715 // use a stub like OptoRuntime::slow_arraycopy_Java.
4716 // This only works for expanding the current library call,
4717 // not another intrinsic. (E.g., don't use this for making an
4718 // arraycopy call inside of the copyOf intrinsic.)
4719 CallJavaNode*
4720 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4721 // When compiling the intrinsic method itself, do not use this technique.
4722 guarantee(callee() != C->method(), "cannot make slow-call to self");
4723
4724 ciMethod* method = callee();
4725 // ensure the JVMS we have will be correct for this call
4726 guarantee(method_id == method->intrinsic_id(), "must match");
4727
4728 const TypeFunc* tf = TypeFunc::make(method);
4729 if (res_not_null) {
4730 assert(tf->return_type() == T_OBJECT, "");
4731 const TypeTuple* range = tf->range_cc();
4732 const Type** fields = TypeTuple::fields(range->cnt());
4733 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4734 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4735 tf = TypeFunc::make(tf->domain_cc(), new_range);
4736 }
4737 CallJavaNode* slow_call;
4738 if (is_static) {
4739 assert(!is_virtual, "");
4740 slow_call = new CallStaticJavaNode(C, tf,
4741 SharedRuntime::get_resolve_static_call_stub(), method);
4742 } else if (is_virtual) {
4743 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4744 int vtable_index = Method::invalid_vtable_index;
4745 if (UseInlineCaches) {
4746 // Suppress the vtable call
4747 } else {
4748 // hashCode and clone are not a miranda methods,
4749 // so the vtable index is fixed.
4750 // No need to use the linkResolver to get it.
4751 vtable_index = method->vtable_index();
4752 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4753 "bad index %d", vtable_index);
4754 }
4755 slow_call = new CallDynamicJavaNode(tf,
4756 SharedRuntime::get_resolve_virtual_call_stub(),
4757 method, vtable_index);
4758 } else { // neither virtual nor static: opt_virtual
4759 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4760 slow_call = new CallStaticJavaNode(C, tf,
4761 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4762 slow_call->set_optimized_virtual(true);
4763 }
4764 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4765 // To be able to issue a direct call (optimized virtual or virtual)
4766 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4767 // about the method being invoked should be attached to the call site to
4768 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4769 slow_call->set_override_symbolic_info(true);
4770 }
4771 set_arguments_for_java_call(slow_call);
4772 set_edges_for_java_call(slow_call);
4773 return slow_call;
4774 }
4775
4776
4777 /**
4778 * Build special case code for calls to hashCode on an object. This call may
4779 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4780 * slightly different code.
4781 */
4782 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4783 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4784 assert(!(is_virtual && is_static), "either virtual, special, or static");
4785
4786 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4787
4788 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4789 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4790 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4791 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4792 Node* obj = argument(0);
4793
4794 if (gvn().type(obj)->is_inlinetypeptr()) {
4795 return false;
4796 }
4797
4798 if (!is_static) {
4799 // Check for hashing null object
4800 obj = null_check_receiver();
4801 if (stopped()) return true; // unconditionally null
4802 result_reg->init_req(_null_path, top());
4803 result_val->init_req(_null_path, top());
4804 } else {
4805 // Do a null check, and return zero if null.
4806 // System.identityHashCode(null) == 0
4807 Node* null_ctl = top();
4808 obj = null_check_oop(obj, &null_ctl);
4809 result_reg->init_req(_null_path, null_ctl);
4810 result_val->init_req(_null_path, _gvn.intcon(0));
4811 }
4812
4813 // Unconditionally null? Then return right away.
4814 if (stopped()) {
4815 set_control( result_reg->in(_null_path));
4816 if (!stopped())
4817 set_result(result_val->in(_null_path));
4818 return true;
4819 }
4820
4821 // We only go to the fast case code if we pass a number of guards. The
4822 // paths which do not pass are accumulated in the slow_region.
4823 RegionNode* slow_region = new RegionNode(1);
4824 record_for_igvn(slow_region);
4825
4826 // If this is a virtual call, we generate a funny guard. We pull out
4827 // the vtable entry corresponding to hashCode() from the target object.
4828 // If the target method which we are calling happens to be the native
4829 // Object hashCode() method, we pass the guard. We do not need this
4830 // guard for non-virtual calls -- the caller is known to be the native
4831 // Object hashCode().
4832 if (is_virtual) {
4833 // After null check, get the object's klass.
4834 Node* obj_klass = load_object_klass(obj);
4835 generate_virtual_guard(obj_klass, slow_region);
4836 }
4837
4838 // Get the header out of the object, use LoadMarkNode when available
4839 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4840 // The control of the load must be null. Otherwise, the load can move before
4841 // the null check after castPP removal.
4842 Node* no_ctrl = nullptr;
4843 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4844
4845 // Test the header to see if it is unlocked.
4846 // This also serves as guard against inline types
4847 Node *lock_mask = _gvn.MakeConX(markWord::inline_type_mask_in_place);
4848 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4849 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4850 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4851 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4852
4853 generate_slow_guard(test_unlocked, slow_region);
4854
4855 // Get the hash value and check to see that it has been properly assigned.
4856 // We depend on hash_mask being at most 32 bits and avoid the use of
4857 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4858 // vm: see markWord.hpp.
4859 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4860 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4861 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4862 // This hack lets the hash bits live anywhere in the mark object now, as long
4863 // as the shift drops the relevant bits into the low 32 bits. Note that
4864 // Java spec says that HashCode is an int so there's no point in capturing
4865 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4866 hshifted_header = ConvX2I(hshifted_header);
4867 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4868
4869 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4870 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4871 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4872
4873 generate_slow_guard(test_assigned, slow_region);
4874
4875 Node* init_mem = reset_memory();
4876 // fill in the rest of the null path:
4877 result_io ->init_req(_null_path, i_o());
4878 result_mem->init_req(_null_path, init_mem);
4879
4880 result_val->init_req(_fast_path, hash_val);
4881 result_reg->init_req(_fast_path, control());
4882 result_io ->init_req(_fast_path, i_o());
4883 result_mem->init_req(_fast_path, init_mem);
4884
4885 // Generate code for the slow case. We make a call to hashCode().
4886 set_control(_gvn.transform(slow_region));
4887 if (!stopped()) {
4888 // No need for PreserveJVMState, because we're using up the present state.
4889 set_all_memory(init_mem);
4890 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4891 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4892 Node* slow_result = set_results_for_java_call(slow_call);
4893 // this->control() comes from set_results_for_java_call
4894 result_reg->init_req(_slow_path, control());
4895 result_val->init_req(_slow_path, slow_result);
4896 result_io ->set_req(_slow_path, i_o());
4897 result_mem ->set_req(_slow_path, reset_memory());
4898 }
4899
4900 // Return the combined state.
4901 set_i_o( _gvn.transform(result_io) );
4902 set_all_memory( _gvn.transform(result_mem));
4903
4904 set_result(result_reg, result_val);
4905 return true;
4906 }
4907
4908 //---------------------------inline_native_getClass----------------------------
4909 // public final native Class<?> java.lang.Object.getClass();
4910 //
4911 // Build special case code for calls to getClass on an object.
4912 bool LibraryCallKit::inline_native_getClass() {
4913 Node* obj = argument(0);
4914 if (obj->is_InlineType()) {
4915 const Type* t = _gvn.type(obj);
4916 if (t->maybe_null()) {
4917 null_check(obj);
4918 }
4919 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror())));
4920 return true;
4921 }
4922 obj = null_check_receiver();
4923 if (stopped()) return true;
4924 set_result(load_mirror_from_klass(load_object_klass(obj)));
4925 return true;
4926 }
4927
4928 //-----------------inline_native_Reflection_getCallerClass---------------------
4929 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4930 //
4931 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4932 //
4933 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4934 // in that it must skip particular security frames and checks for
4935 // caller sensitive methods.
4936 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4937 #ifndef PRODUCT
4938 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4939 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4940 }
4941 #endif
4942
4943 if (!jvms()->has_method()) {
4944 #ifndef PRODUCT
4945 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4946 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4947 }
4948 #endif
4949 return false;
4950 }
4951
4952 // Walk back up the JVM state to find the caller at the required
4953 // depth.
4954 JVMState* caller_jvms = jvms();
4955
4956 // Cf. JVM_GetCallerClass
4957 // NOTE: Start the loop at depth 1 because the current JVM state does
4958 // not include the Reflection.getCallerClass() frame.
4959 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4960 ciMethod* m = caller_jvms->method();
4961 switch (n) {
4962 case 0:
4963 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4964 break;
4965 case 1:
4966 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4967 if (!m->caller_sensitive()) {
4968 #ifndef PRODUCT
4969 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4970 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4971 }
4972 #endif
4973 return false; // bail-out; let JVM_GetCallerClass do the work
4974 }
4975 break;
4976 default:
4977 if (!m->is_ignored_by_security_stack_walk()) {
4978 // We have reached the desired frame; return the holder class.
4979 // Acquire method holder as java.lang.Class and push as constant.
4980 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4981 ciInstance* caller_mirror = caller_klass->java_mirror();
4982 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4983
4984 #ifndef PRODUCT
4985 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4986 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());
4987 tty->print_cr(" JVM state at this point:");
4988 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4989 ciMethod* m = jvms()->of_depth(i)->method();
4990 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4991 }
4992 }
4993 #endif
4994 return true;
4995 }
4996 break;
4997 }
4998 }
4999
5000 #ifndef PRODUCT
5001 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5002 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
5003 tty->print_cr(" JVM state at this point:");
5004 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5005 ciMethod* m = jvms()->of_depth(i)->method();
5006 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5007 }
5008 }
5009 #endif
5010
5011 return false; // bail-out; let JVM_GetCallerClass do the work
5012 }
5013
5014 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
5015 Node* arg = argument(0);
5016 Node* result = nullptr;
5017
5018 switch (id) {
5019 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
5020 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
5021 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5022 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5023 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5024 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5025
5026 case vmIntrinsics::_doubleToLongBits: {
5027 // two paths (plus control) merge in a wood
5028 RegionNode *r = new RegionNode(3);
5029 Node *phi = new PhiNode(r, TypeLong::LONG);
5030
5031 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5032 // Build the boolean node
5033 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5034
5035 // Branch either way.
5036 // NaN case is less traveled, which makes all the difference.
5037 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5038 Node *opt_isnan = _gvn.transform(ifisnan);
5039 assert( opt_isnan->is_If(), "Expect an IfNode");
5040 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5041 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5042
5043 set_control(iftrue);
5044
5045 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5046 Node *slow_result = longcon(nan_bits); // return NaN
5047 phi->init_req(1, _gvn.transform( slow_result ));
5048 r->init_req(1, iftrue);
5049
5050 // Else fall through
5051 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5052 set_control(iffalse);
5053
5054 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5055 r->init_req(2, iffalse);
5056
5057 // Post merge
5058 set_control(_gvn.transform(r));
5059 record_for_igvn(r);
5060
5061 C->set_has_split_ifs(true); // Has chance for split-if optimization
5062 result = phi;
5063 assert(result->bottom_type()->isa_long(), "must be");
5064 break;
5065 }
5066
5067 case vmIntrinsics::_floatToIntBits: {
5068 // two paths (plus control) merge in a wood
5069 RegionNode *r = new RegionNode(3);
5070 Node *phi = new PhiNode(r, TypeInt::INT);
5071
5072 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5073 // Build the boolean node
5074 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5075
5076 // Branch either way.
5077 // NaN case is less traveled, which makes all the difference.
5078 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5079 Node *opt_isnan = _gvn.transform(ifisnan);
5080 assert( opt_isnan->is_If(), "Expect an IfNode");
5081 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5082 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5083
5084 set_control(iftrue);
5085
5086 static const jint nan_bits = 0x7fc00000;
5087 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5088 phi->init_req(1, _gvn.transform( slow_result ));
5089 r->init_req(1, iftrue);
5090
5091 // Else fall through
5092 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5093 set_control(iffalse);
5094
5095 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5096 r->init_req(2, iffalse);
5097
5098 // Post merge
5099 set_control(_gvn.transform(r));
5100 record_for_igvn(r);
5101
5102 C->set_has_split_ifs(true); // Has chance for split-if optimization
5103 result = phi;
5104 assert(result->bottom_type()->isa_int(), "must be");
5105 break;
5106 }
5107
5108 default:
5109 fatal_unexpected_iid(id);
5110 break;
5111 }
5112 set_result(_gvn.transform(result));
5113 return true;
5114 }
5115
5116 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5117 Node* arg = argument(0);
5118 Node* result = nullptr;
5119
5120 switch (id) {
5121 case vmIntrinsics::_floatIsInfinite:
5122 result = new IsInfiniteFNode(arg);
5123 break;
5124 case vmIntrinsics::_floatIsFinite:
5125 result = new IsFiniteFNode(arg);
5126 break;
5127 case vmIntrinsics::_doubleIsInfinite:
5128 result = new IsInfiniteDNode(arg);
5129 break;
5130 case vmIntrinsics::_doubleIsFinite:
5131 result = new IsFiniteDNode(arg);
5132 break;
5133 default:
5134 fatal_unexpected_iid(id);
5135 break;
5136 }
5137 set_result(_gvn.transform(result));
5138 return true;
5139 }
5140
5141 //----------------------inline_unsafe_copyMemory-------------------------
5142 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5143
5144 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5145 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5146 const Type* base_t = gvn.type(base);
5147
5148 bool in_native = (base_t == TypePtr::NULL_PTR);
5149 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5150 bool is_mixed = !in_heap && !in_native;
5151
5152 if (is_mixed) {
5153 return true; // mixed accesses can touch both on-heap and off-heap memory
5154 }
5155 if (in_heap) {
5156 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5157 if (!is_prim_array) {
5158 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5159 // there's not enough type information available to determine proper memory slice for it.
5160 return true;
5161 }
5162 }
5163 return false;
5164 }
5165
5166 bool LibraryCallKit::inline_unsafe_copyMemory() {
5167 if (callee()->is_static()) return false; // caller must have the capability!
5168 null_check_receiver(); // null-check receiver
5169 if (stopped()) return true;
5170
5171 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5172
5173 Node* src_base = argument(1); // type: oop
5174 Node* src_off = ConvL2X(argument(2)); // type: long
5175 Node* dst_base = argument(4); // type: oop
5176 Node* dst_off = ConvL2X(argument(5)); // type: long
5177 Node* size = ConvL2X(argument(7)); // type: long
5178
5179 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5180 "fieldOffset must be byte-scaled");
5181
5182 Node* src_addr = make_unsafe_address(src_base, src_off);
5183 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5184
5185 Node* thread = _gvn.transform(new ThreadLocalNode());
5186 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5187 BasicType doing_unsafe_access_bt = T_BYTE;
5188 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5189
5190 // update volatile field
5191 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5192
5193 int flags = RC_LEAF | RC_NO_FP;
5194
5195 const TypePtr* dst_type = TypePtr::BOTTOM;
5196
5197 // Adjust memory effects of the runtime call based on input values.
5198 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5199 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5200 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5201
5202 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5203 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5204 flags |= RC_NARROW_MEM; // narrow in memory
5205 }
5206 }
5207
5208 // Call it. Note that the length argument is not scaled.
5209 make_runtime_call(flags,
5210 OptoRuntime::fast_arraycopy_Type(),
5211 StubRoutines::unsafe_arraycopy(),
5212 "unsafe_arraycopy",
5213 dst_type,
5214 src_addr, dst_addr, size XTOP);
5215
5216 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5217
5218 return true;
5219 }
5220
5221 #undef XTOP
5222
5223 // TODO 8325106 Remove this and corresponding tests. Flatness is not a property of the Class anymore with JEP 401.
5224 //----------------------inline_unsafe_isFlatArray------------------------
5225 // public native boolean Unsafe.isFlatArray(Class<?> arrayClass);
5226 // This intrinsic exploits assumptions made by the native implementation
5227 // (arrayClass is neither null nor primitive) to avoid unnecessary null checks.
5228 bool LibraryCallKit::inline_unsafe_isFlatArray() {
5229 Node* cls = argument(1);
5230 Node* p = basic_plus_adr(cls, java_lang_Class::klass_offset());
5231 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p,
5232 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT));
5233 Node* result = flat_array_test(kls);
5234 set_result(result);
5235 return true;
5236 }
5237
5238 //------------------------clone_coping-----------------------------------
5239 // Helper function for inline_native_clone.
5240 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5241 assert(obj_size != nullptr, "");
5242 Node* raw_obj = alloc_obj->in(1);
5243 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5244
5245 AllocateNode* alloc = nullptr;
5246 if (ReduceBulkZeroing) {
5247 // We will be completely responsible for initializing this object -
5248 // mark Initialize node as complete.
5249 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5250 // The object was just allocated - there should be no any stores!
5251 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5252 // Mark as complete_with_arraycopy so that on AllocateNode
5253 // expansion, we know this AllocateNode is initialized by an array
5254 // copy and a StoreStore barrier exists after the array copy.
5255 alloc->initialization()->set_complete_with_arraycopy();
5256 }
5257
5258 Node* size = _gvn.transform(obj_size);
5259 access_clone(obj, alloc_obj, size, is_array);
5260
5261 // Do not let reads from the cloned object float above the arraycopy.
5262 if (alloc != nullptr) {
5263 // Do not let stores that initialize this object be reordered with
5264 // a subsequent store that would make this object accessible by
5265 // other threads.
5266 // Record what AllocateNode this StoreStore protects so that
5267 // escape analysis can go from the MemBarStoreStoreNode to the
5268 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5269 // based on the escape status of the AllocateNode.
5270 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5271 } else {
5272 insert_mem_bar(Op_MemBarCPUOrder);
5273 }
5274 }
5275
5276 //------------------------inline_native_clone----------------------------
5277 // protected native Object java.lang.Object.clone();
5278 //
5279 // Here are the simple edge cases:
5280 // null receiver => normal trap
5281 // virtual and clone was overridden => slow path to out-of-line clone
5282 // not cloneable or finalizer => slow path to out-of-line Object.clone
5283 //
5284 // The general case has two steps, allocation and copying.
5285 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5286 //
5287 // Copying also has two cases, oop arrays and everything else.
5288 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5289 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5290 //
5291 // These steps fold up nicely if and when the cloned object's klass
5292 // can be sharply typed as an object array, a type array, or an instance.
5293 //
5294 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5295 PhiNode* result_val;
5296
5297 // Set the reexecute bit for the interpreter to reexecute
5298 // the bytecode that invokes Object.clone if deoptimization happens.
5299 { PreserveReexecuteState preexecs(this);
5300 jvms()->set_should_reexecute(true);
5301
5302 Node* obj = argument(0);
5303 obj = null_check_receiver();
5304 if (stopped()) return true;
5305
5306 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5307
5308 // If we are going to clone an instance, we need its exact type to
5309 // know the number and types of fields to convert the clone to
5310 // loads/stores. Maybe a speculative type can help us.
5311 if (!obj_type->klass_is_exact() &&
5312 obj_type->speculative_type() != nullptr &&
5313 obj_type->speculative_type()->is_instance_klass() &&
5314 !obj_type->speculative_type()->is_inlinetype()) {
5315 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5316 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5317 !spec_ik->has_injected_fields()) {
5318 if (!obj_type->isa_instptr() ||
5319 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5320 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5321 }
5322 }
5323 }
5324
5325 // Conservatively insert a memory barrier on all memory slices.
5326 // Do not let writes into the original float below the clone.
5327 insert_mem_bar(Op_MemBarCPUOrder);
5328
5329 // paths into result_reg:
5330 enum {
5331 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5332 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5333 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5334 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5335 PATH_LIMIT
5336 };
5337 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5338 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5339 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5340 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5341 record_for_igvn(result_reg);
5342
5343 Node* obj_klass = load_object_klass(obj);
5344 // We only go to the fast case code if we pass a number of guards.
5345 // The paths which do not pass are accumulated in the slow_region.
5346 RegionNode* slow_region = new RegionNode(1);
5347 record_for_igvn(slow_region);
5348
5349 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr);
5350 if (array_ctl != nullptr) {
5351 // It's an array.
5352 PreserveJVMState pjvms(this);
5353 set_control(array_ctl);
5354
5355 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5356 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr();
5357 if (UseFlatArray && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) &&
5358 obj_type->can_be_inline_array() &&
5359 (ary_ptr == nullptr || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) {
5360 // Flat inline type array may have object field that would require a
5361 // write barrier. Conservatively, go to slow path.
5362 generate_fair_guard(flat_array_test(obj_klass), slow_region);
5363 }
5364
5365 if (!stopped()) {
5366 Node* obj_length = load_array_length(obj);
5367 Node* array_size = nullptr; // Size of the array without object alignment padding.
5368 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5369
5370 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5371 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5372 // If it is an oop array, it requires very special treatment,
5373 // because gc barriers are required when accessing the array.
5374 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5375 if (is_obja != nullptr) {
5376 PreserveJVMState pjvms2(this);
5377 set_control(is_obja);
5378 // Generate a direct call to the right arraycopy function(s).
5379 // Clones are always tightly coupled.
5380 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5381 ac->set_clone_oop_array();
5382 Node* n = _gvn.transform(ac);
5383 assert(n == ac, "cannot disappear");
5384 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5385
5386 result_reg->init_req(_objArray_path, control());
5387 result_val->init_req(_objArray_path, alloc_obj);
5388 result_i_o ->set_req(_objArray_path, i_o());
5389 result_mem ->set_req(_objArray_path, reset_memory());
5390 }
5391 }
5392 // Otherwise, there are no barriers to worry about.
5393 // (We can dispense with card marks if we know the allocation
5394 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5395 // causes the non-eden paths to take compensating steps to
5396 // simulate a fresh allocation, so that no further
5397 // card marks are required in compiled code to initialize
5398 // the object.)
5399
5400 if (!stopped()) {
5401 copy_to_clone(obj, alloc_obj, array_size, true);
5402
5403 // Present the results of the copy.
5404 result_reg->init_req(_array_path, control());
5405 result_val->init_req(_array_path, alloc_obj);
5406 result_i_o ->set_req(_array_path, i_o());
5407 result_mem ->set_req(_array_path, reset_memory());
5408 }
5409 }
5410 }
5411
5412 if (!stopped()) {
5413 // It's an instance (we did array above). Make the slow-path tests.
5414 // If this is a virtual call, we generate a funny guard. We grab
5415 // the vtable entry corresponding to clone() from the target object.
5416 // If the target method which we are calling happens to be the
5417 // Object clone() method, we pass the guard. We do not need this
5418 // guard for non-virtual calls; the caller is known to be the native
5419 // Object clone().
5420 if (is_virtual) {
5421 generate_virtual_guard(obj_klass, slow_region);
5422 }
5423
5424 // The object must be easily cloneable and must not have a finalizer.
5425 // Both of these conditions may be checked in a single test.
5426 // We could optimize the test further, but we don't care.
5427 generate_access_flags_guard(obj_klass,
5428 // Test both conditions:
5429 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
5430 // Must be cloneable but not finalizer:
5431 JVM_ACC_IS_CLONEABLE_FAST,
5432 slow_region);
5433 }
5434
5435 if (!stopped()) {
5436 // It's an instance, and it passed the slow-path tests.
5437 PreserveJVMState pjvms(this);
5438 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5439 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5440 // is reexecuted if deoptimization occurs and there could be problems when merging
5441 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5442 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5443
5444 copy_to_clone(obj, alloc_obj, obj_size, false);
5445
5446 // Present the results of the slow call.
5447 result_reg->init_req(_instance_path, control());
5448 result_val->init_req(_instance_path, alloc_obj);
5449 result_i_o ->set_req(_instance_path, i_o());
5450 result_mem ->set_req(_instance_path, reset_memory());
5451 }
5452
5453 // Generate code for the slow case. We make a call to clone().
5454 set_control(_gvn.transform(slow_region));
5455 if (!stopped()) {
5456 PreserveJVMState pjvms(this);
5457 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5458 // We need to deoptimize on exception (see comment above)
5459 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5460 // this->control() comes from set_results_for_java_call
5461 result_reg->init_req(_slow_path, control());
5462 result_val->init_req(_slow_path, slow_result);
5463 result_i_o ->set_req(_slow_path, i_o());
5464 result_mem ->set_req(_slow_path, reset_memory());
5465 }
5466
5467 // Return the combined state.
5468 set_control( _gvn.transform(result_reg));
5469 set_i_o( _gvn.transform(result_i_o));
5470 set_all_memory( _gvn.transform(result_mem));
5471 } // original reexecute is set back here
5472
5473 set_result(_gvn.transform(result_val));
5474 return true;
5475 }
5476
5477 // If we have a tightly coupled allocation, the arraycopy may take care
5478 // of the array initialization. If one of the guards we insert between
5479 // the allocation and the arraycopy causes a deoptimization, an
5480 // uninitialized array will escape the compiled method. To prevent that
5481 // we set the JVM state for uncommon traps between the allocation and
5482 // the arraycopy to the state before the allocation so, in case of
5483 // deoptimization, we'll reexecute the allocation and the
5484 // initialization.
5485 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5486 if (alloc != nullptr) {
5487 ciMethod* trap_method = alloc->jvms()->method();
5488 int trap_bci = alloc->jvms()->bci();
5489
5490 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5491 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5492 // Make sure there's no store between the allocation and the
5493 // arraycopy otherwise visible side effects could be rexecuted
5494 // in case of deoptimization and cause incorrect execution.
5495 bool no_interfering_store = true;
5496 Node* mem = alloc->in(TypeFunc::Memory);
5497 if (mem->is_MergeMem()) {
5498 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5499 Node* n = mms.memory();
5500 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5501 assert(n->is_Store(), "what else?");
5502 no_interfering_store = false;
5503 break;
5504 }
5505 }
5506 } else {
5507 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5508 Node* n = mms.memory();
5509 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5510 assert(n->is_Store(), "what else?");
5511 no_interfering_store = false;
5512 break;
5513 }
5514 }
5515 }
5516
5517 if (no_interfering_store) {
5518 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5519
5520 JVMState* saved_jvms = jvms();
5521 saved_reexecute_sp = _reexecute_sp;
5522
5523 set_jvms(sfpt->jvms());
5524 _reexecute_sp = jvms()->sp();
5525
5526 return saved_jvms;
5527 }
5528 }
5529 }
5530 return nullptr;
5531 }
5532
5533 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5534 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5535 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5536 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5537 uint size = alloc->req();
5538 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5539 old_jvms->set_map(sfpt);
5540 for (uint i = 0; i < size; i++) {
5541 sfpt->init_req(i, alloc->in(i));
5542 }
5543 int adjustment = 1;
5544 // TODO 8325106 why can't we check via the type of the const klass node?
5545 if (alloc->is_null_free()) {
5546 // A null-free, tightly coupled array allocation can only come from LibraryCallKit::inline_newNullRestrictedArray
5547 // which requires both the component type and the array length on stack for re-execution. Re-create and push
5548 // the component type.
5549 ciArrayKlass* klass = alloc->in(AllocateNode::KlassNode)->bottom_type()->is_aryklassptr()->exact_klass()->as_array_klass();
5550 ciInstance* instance = klass->component_mirror_instance();
5551 const TypeInstPtr* t_instance = TypeInstPtr::make(instance);
5552 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), makecon(t_instance));
5553 adjustment++;
5554 }
5555 // re-push array length for deoptimization
5556 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment - 1, alloc->in(AllocateNode::ALength));
5557 old_jvms->set_sp(old_jvms->sp() + adjustment);
5558 old_jvms->set_monoff(old_jvms->monoff() + adjustment);
5559 old_jvms->set_scloff(old_jvms->scloff() + adjustment);
5560 old_jvms->set_endoff(old_jvms->endoff() + adjustment);
5561 old_jvms->set_should_reexecute(true);
5562
5563 sfpt->set_i_o(map()->i_o());
5564 sfpt->set_memory(map()->memory());
5565 sfpt->set_control(map()->control());
5566 return sfpt;
5567 }
5568
5569 // In case of a deoptimization, we restart execution at the
5570 // allocation, allocating a new array. We would leave an uninitialized
5571 // array in the heap that GCs wouldn't expect. Move the allocation
5572 // after the traps so we don't allocate the array if we
5573 // deoptimize. This is possible because tightly_coupled_allocation()
5574 // guarantees there's no observer of the allocated array at this point
5575 // and the control flow is simple enough.
5576 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5577 int saved_reexecute_sp, uint new_idx) {
5578 if (saved_jvms_before_guards != nullptr && !stopped()) {
5579 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5580
5581 assert(alloc != nullptr, "only with a tightly coupled allocation");
5582 // restore JVM state to the state at the arraycopy
5583 saved_jvms_before_guards->map()->set_control(map()->control());
5584 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5585 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5586 // If we've improved the types of some nodes (null check) while
5587 // emitting the guards, propagate them to the current state
5588 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5589 set_jvms(saved_jvms_before_guards);
5590 _reexecute_sp = saved_reexecute_sp;
5591
5592 // Remove the allocation from above the guards
5593 CallProjections* callprojs = alloc->extract_projections(true);
5594 InitializeNode* init = alloc->initialization();
5595 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5596 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5597 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5598
5599 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5600 // the allocation (i.e. is only valid if the allocation succeeds):
5601 // 1) replace CastIINode with AllocateArrayNode's length here
5602 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5603 //
5604 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5605 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5606 Node* init_control = init->proj_out(TypeFunc::Control);
5607 Node* alloc_length = alloc->Ideal_length();
5608 #ifdef ASSERT
5609 Node* prev_cast = nullptr;
5610 #endif
5611 for (uint i = 0; i < init_control->outcnt(); i++) {
5612 Node* init_out = init_control->raw_out(i);
5613 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5614 #ifdef ASSERT
5615 if (prev_cast == nullptr) {
5616 prev_cast = init_out;
5617 } else {
5618 if (prev_cast->cmp(*init_out) == false) {
5619 prev_cast->dump();
5620 init_out->dump();
5621 assert(false, "not equal CastIINode");
5622 }
5623 }
5624 #endif
5625 C->gvn_replace_by(init_out, alloc_length);
5626 }
5627 }
5628 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5629
5630 // move the allocation here (after the guards)
5631 _gvn.hash_delete(alloc);
5632 alloc->set_req(TypeFunc::Control, control());
5633 alloc->set_req(TypeFunc::I_O, i_o());
5634 Node *mem = reset_memory();
5635 set_all_memory(mem);
5636 alloc->set_req(TypeFunc::Memory, mem);
5637 set_control(init->proj_out_or_null(TypeFunc::Control));
5638 set_i_o(callprojs->fallthrough_ioproj);
5639
5640 // Update memory as done in GraphKit::set_output_for_allocation()
5641 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5642 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5643 if (ary_type->isa_aryptr() && length_type != nullptr) {
5644 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5645 }
5646 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5647 int elemidx = C->get_alias_index(telemref);
5648 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5649 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5650
5651 Node* allocx = _gvn.transform(alloc);
5652 assert(allocx == alloc, "where has the allocation gone?");
5653 assert(dest->is_CheckCastPP(), "not an allocation result?");
5654
5655 _gvn.hash_delete(dest);
5656 dest->set_req(0, control());
5657 Node* destx = _gvn.transform(dest);
5658 assert(destx == dest, "where has the allocation result gone?");
5659
5660 array_ideal_length(alloc, ary_type, true);
5661 }
5662 }
5663
5664 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5665 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5666 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5667 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5668 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5669 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5670 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5671 JVMState* saved_jvms_before_guards) {
5672 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5673 // There is at least one unrelated uncommon trap which needs to be replaced.
5674 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5675
5676 JVMState* saved_jvms = jvms();
5677 const int saved_reexecute_sp = _reexecute_sp;
5678 set_jvms(sfpt->jvms());
5679 _reexecute_sp = jvms()->sp();
5680
5681 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5682
5683 // Restore state
5684 set_jvms(saved_jvms);
5685 _reexecute_sp = saved_reexecute_sp;
5686 }
5687 }
5688
5689 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5690 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5691 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5692 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5693 while (if_proj->is_IfProj()) {
5694 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5695 if (uncommon_trap != nullptr) {
5696 create_new_uncommon_trap(uncommon_trap);
5697 }
5698 assert(if_proj->in(0)->is_If(), "must be If");
5699 if_proj = if_proj->in(0)->in(0);
5700 }
5701 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5702 "must have reached control projection of init node");
5703 }
5704
5705 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5706 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5707 assert(trap_request != 0, "no valid UCT trap request");
5708 PreserveJVMState pjvms(this);
5709 set_control(uncommon_trap_call->in(0));
5710 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5711 Deoptimization::trap_request_action(trap_request));
5712 assert(stopped(), "Should be stopped");
5713 _gvn.hash_delete(uncommon_trap_call);
5714 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5715 }
5716
5717 //------------------------------inline_array_partition-----------------------
5718 bool LibraryCallKit::inline_array_partition() {
5719
5720 Node* elementType = null_check(argument(0));
5721 Node* obj = argument(1);
5722 Node* offset = argument(2);
5723 Node* fromIndex = argument(4);
5724 Node* toIndex = argument(5);
5725 Node* indexPivot1 = argument(6);
5726 Node* indexPivot2 = argument(7);
5727
5728 Node* pivotIndices = nullptr;
5729
5730 // Set the original stack and the reexecute bit for the interpreter to reexecute
5731 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
5732 { PreserveReexecuteState preexecs(this);
5733 jvms()->set_should_reexecute(true);
5734
5735 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5736 ciType* elem_type = elem_klass->const_oop()->as_instance()->java_mirror_type();
5737 BasicType bt = elem_type->basic_type();
5738 address stubAddr = nullptr;
5739 stubAddr = StubRoutines::select_array_partition_function();
5740 // stub not loaded
5741 if (stubAddr == nullptr) {
5742 return false;
5743 }
5744 // get the address of the array
5745 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5746 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM ) {
5747 return false; // failed input validation
5748 }
5749 Node* obj_adr = make_unsafe_address(obj, offset);
5750
5751 // create the pivotIndices array of type int and size = 2
5752 Node* size = intcon(2);
5753 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
5754 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
5755 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
5756 guarantee(alloc != nullptr, "created above");
5757 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
5758
5759 // pass the basic type enum to the stub
5760 Node* elemType = intcon(bt);
5761
5762 // Call the stub
5763 const char *stubName = "array_partition_stub";
5764 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
5765 stubAddr, stubName, TypePtr::BOTTOM,
5766 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
5767 indexPivot1, indexPivot2);
5768
5769 } // original reexecute is set back here
5770
5771 if (!stopped()) {
5772 set_result(pivotIndices);
5773 }
5774
5775 return true;
5776 }
5777
5778
5779 //------------------------------inline_array_sort-----------------------
5780 bool LibraryCallKit::inline_array_sort() {
5781
5782 Node* elementType = null_check(argument(0));
5783 Node* obj = argument(1);
5784 Node* offset = argument(2);
5785 Node* fromIndex = argument(4);
5786 Node* toIndex = argument(5);
5787
5788 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5789 ciType* elem_type = elem_klass->const_oop()->as_instance()->java_mirror_type();
5790 BasicType bt = elem_type->basic_type();
5791 address stubAddr = nullptr;
5792 stubAddr = StubRoutines::select_arraysort_function();
5793 //stub not loaded
5794 if (stubAddr == nullptr) {
5795 return false;
5796 }
5797
5798 // get address of the array
5799 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5800 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM ) {
5801 return false; // failed input validation
5802 }
5803 Node* obj_adr = make_unsafe_address(obj, offset);
5804
5805 // pass the basic type enum to the stub
5806 Node* elemType = intcon(bt);
5807
5808 // Call the stub.
5809 const char *stubName = "arraysort_stub";
5810 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
5811 stubAddr, stubName, TypePtr::BOTTOM,
5812 obj_adr, elemType, fromIndex, toIndex);
5813
5814 return true;
5815 }
5816
5817
5818 //------------------------------inline_arraycopy-----------------------
5819 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5820 // Object dest, int destPos,
5821 // int length);
5822 bool LibraryCallKit::inline_arraycopy() {
5823 // Get the arguments.
5824 Node* src = argument(0); // type: oop
5825 Node* src_offset = argument(1); // type: int
5826 Node* dest = argument(2); // type: oop
5827 Node* dest_offset = argument(3); // type: int
5828 Node* length = argument(4); // type: int
5829
5830 uint new_idx = C->unique();
5831
5832 // Check for allocation before we add nodes that would confuse
5833 // tightly_coupled_allocation()
5834 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5835
5836 int saved_reexecute_sp = -1;
5837 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5838 // See arraycopy_restore_alloc_state() comment
5839 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5840 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5841 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5842 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5843
5844 // The following tests must be performed
5845 // (1) src and dest are arrays.
5846 // (2) src and dest arrays must have elements of the same BasicType
5847 // (3) src and dest must not be null.
5848 // (4) src_offset must not be negative.
5849 // (5) dest_offset must not be negative.
5850 // (6) length must not be negative.
5851 // (7) src_offset + length must not exceed length of src.
5852 // (8) dest_offset + length must not exceed length of dest.
5853 // (9) each element of an oop array must be assignable
5854
5855 // (3) src and dest must not be null.
5856 // always do this here because we need the JVM state for uncommon traps
5857 Node* null_ctl = top();
5858 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5859 assert(null_ctl->is_top(), "no null control here");
5860 dest = null_check(dest, T_ARRAY);
5861
5862 if (!can_emit_guards) {
5863 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5864 // guards but the arraycopy node could still take advantage of a
5865 // tightly allocated allocation. tightly_coupled_allocation() is
5866 // called again to make sure it takes the null check above into
5867 // account: the null check is mandatory and if it caused an
5868 // uncommon trap to be emitted then the allocation can't be
5869 // considered tightly coupled in this context.
5870 alloc = tightly_coupled_allocation(dest);
5871 }
5872
5873 bool validated = false;
5874
5875 const Type* src_type = _gvn.type(src);
5876 const Type* dest_type = _gvn.type(dest);
5877 const TypeAryPtr* top_src = src_type->isa_aryptr();
5878 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5879
5880 // Do we have the type of src?
5881 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5882 // Do we have the type of dest?
5883 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5884 // Is the type for src from speculation?
5885 bool src_spec = false;
5886 // Is the type for dest from speculation?
5887 bool dest_spec = false;
5888
5889 if ((!has_src || !has_dest) && can_emit_guards) {
5890 // We don't have sufficient type information, let's see if
5891 // speculative types can help. We need to have types for both src
5892 // and dest so that it pays off.
5893
5894 // Do we already have or could we have type information for src
5895 bool could_have_src = has_src;
5896 // Do we already have or could we have type information for dest
5897 bool could_have_dest = has_dest;
5898
5899 ciKlass* src_k = nullptr;
5900 if (!has_src) {
5901 src_k = src_type->speculative_type_not_null();
5902 if (src_k != nullptr && src_k->is_array_klass()) {
5903 could_have_src = true;
5904 }
5905 }
5906
5907 ciKlass* dest_k = nullptr;
5908 if (!has_dest) {
5909 dest_k = dest_type->speculative_type_not_null();
5910 if (dest_k != nullptr && dest_k->is_array_klass()) {
5911 could_have_dest = true;
5912 }
5913 }
5914
5915 if (could_have_src && could_have_dest) {
5916 // This is going to pay off so emit the required guards
5917 if (!has_src) {
5918 src = maybe_cast_profiled_obj(src, src_k, true);
5919 src_type = _gvn.type(src);
5920 top_src = src_type->isa_aryptr();
5921 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5922 src_spec = true;
5923 }
5924 if (!has_dest) {
5925 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5926 dest_type = _gvn.type(dest);
5927 top_dest = dest_type->isa_aryptr();
5928 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5929 dest_spec = true;
5930 }
5931 }
5932 }
5933
5934 if (has_src && has_dest && can_emit_guards) {
5935 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5936 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5937 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5938 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5939
5940 if (src_elem == dest_elem && top_src->is_flat() == top_dest->is_flat() && src_elem == T_OBJECT) {
5941 // If both arrays are object arrays then having the exact types
5942 // for both will remove the need for a subtype check at runtime
5943 // before the call and may make it possible to pick a faster copy
5944 // routine (without a subtype check on every element)
5945 // Do we have the exact type of src?
5946 bool could_have_src = src_spec;
5947 // Do we have the exact type of dest?
5948 bool could_have_dest = dest_spec;
5949 ciKlass* src_k = nullptr;
5950 ciKlass* dest_k = nullptr;
5951 if (!src_spec) {
5952 src_k = src_type->speculative_type_not_null();
5953 if (src_k != nullptr && src_k->is_array_klass()) {
5954 could_have_src = true;
5955 }
5956 }
5957 if (!dest_spec) {
5958 dest_k = dest_type->speculative_type_not_null();
5959 if (dest_k != nullptr && dest_k->is_array_klass()) {
5960 could_have_dest = true;
5961 }
5962 }
5963 if (could_have_src && could_have_dest) {
5964 // If we can have both exact types, emit the missing guards
5965 if (could_have_src && !src_spec) {
5966 src = maybe_cast_profiled_obj(src, src_k, true);
5967 src_type = _gvn.type(src);
5968 top_src = src_type->isa_aryptr();
5969 }
5970 if (could_have_dest && !dest_spec) {
5971 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5972 dest_type = _gvn.type(dest);
5973 top_dest = dest_type->isa_aryptr();
5974 }
5975 }
5976 }
5977 }
5978
5979 ciMethod* trap_method = method();
5980 int trap_bci = bci();
5981 if (saved_jvms_before_guards != nullptr) {
5982 trap_method = alloc->jvms()->method();
5983 trap_bci = alloc->jvms()->bci();
5984 }
5985
5986 bool negative_length_guard_generated = false;
5987
5988 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5989 can_emit_guards && !src->is_top() && !dest->is_top()) {
5990 // validate arguments: enables transformation the ArrayCopyNode
5991 validated = true;
5992
5993 RegionNode* slow_region = new RegionNode(1);
5994 record_for_igvn(slow_region);
5995
5996 // (1) src and dest are arrays.
5997 generate_non_array_guard(load_object_klass(src), slow_region);
5998 generate_non_array_guard(load_object_klass(dest), slow_region);
5999
6000 // (2) src and dest arrays must have elements of the same BasicType
6001 // done at macro expansion or at Ideal transformation time
6002
6003 // (4) src_offset must not be negative.
6004 generate_negative_guard(src_offset, slow_region);
6005
6006 // (5) dest_offset must not be negative.
6007 generate_negative_guard(dest_offset, slow_region);
6008
6009 // (7) src_offset + length must not exceed length of src.
6010 generate_limit_guard(src_offset, length,
6011 load_array_length(src),
6012 slow_region);
6013
6014 // (8) dest_offset + length must not exceed length of dest.
6015 generate_limit_guard(dest_offset, length,
6016 load_array_length(dest),
6017 slow_region);
6018
6019 // (6) length must not be negative.
6020 // This is also checked in generate_arraycopy() during macro expansion, but
6021 // we also have to check it here for the case where the ArrayCopyNode will
6022 // be eliminated by Escape Analysis.
6023 if (EliminateAllocations) {
6024 generate_negative_guard(length, slow_region);
6025 negative_length_guard_generated = true;
6026 }
6027
6028 // (9) each element of an oop array must be assignable
6029 Node* dest_klass = load_object_klass(dest);
6030 if (src != dest) {
6031 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6032 slow_region->add_req(not_subtype_ctrl);
6033 }
6034
6035 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
6036 const Type* toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6037 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6038 src_type = _gvn.type(src);
6039 top_src = src_type->isa_aryptr();
6040
6041 // Handle flat inline type arrays (null-free arrays are handled by the subtype check above)
6042 if (!stopped() && UseFlatArray) {
6043 // If dest is flat, src must be flat as well (guaranteed by src <: dest check). Handle flat src here.
6044 assert(top_dest == nullptr || !top_dest->is_flat() || top_src->is_flat(), "src array must be flat");
6045 if (top_src != nullptr && top_src->is_flat()) {
6046 // Src is flat, check that dest is flat as well
6047 if (top_dest != nullptr && !top_dest->is_flat()) {
6048 generate_fair_guard(flat_array_test(dest_klass, /* flat = */ false), slow_region);
6049 // Since dest is flat and src <: dest, dest must have the same type as src.
6050 top_dest = top_src->cast_to_exactness(false);
6051 assert(top_dest->is_flat(), "dest must be flat");
6052 dest = _gvn.transform(new CheckCastPPNode(control(), dest, top_dest));
6053 }
6054 } else if (top_src == nullptr || !top_src->is_not_flat()) {
6055 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
6056 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
6057 assert(top_dest == nullptr || !top_dest->is_flat(), "dest array must not be flat");
6058 generate_fair_guard(flat_array_test(src), slow_region);
6059 if (top_src != nullptr) {
6060 top_src = top_src->cast_to_not_flat();
6061 src = _gvn.transform(new CheckCastPPNode(control(), src, top_src));
6062 }
6063 }
6064 }
6065
6066 {
6067 PreserveJVMState pjvms(this);
6068 set_control(_gvn.transform(slow_region));
6069 uncommon_trap(Deoptimization::Reason_intrinsic,
6070 Deoptimization::Action_make_not_entrant);
6071 assert(stopped(), "Should be stopped");
6072 }
6073 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6074 }
6075
6076 if (stopped()) {
6077 return true;
6078 }
6079
6080 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6081 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6082 // so the compiler has a chance to eliminate them: during macro expansion,
6083 // we have to set their control (CastPP nodes are eliminated).
6084 load_object_klass(src), load_object_klass(dest),
6085 load_array_length(src), load_array_length(dest));
6086
6087 ac->set_arraycopy(validated);
6088
6089 Node* n = _gvn.transform(ac);
6090 if (n == ac) {
6091 ac->connect_outputs(this);
6092 } else {
6093 assert(validated, "shouldn't transform if all arguments not validated");
6094 set_all_memory(n);
6095 }
6096 clear_upper_avx();
6097
6098
6099 return true;
6100 }
6101
6102
6103 // Helper function which determines if an arraycopy immediately follows
6104 // an allocation, with no intervening tests or other escapes for the object.
6105 AllocateArrayNode*
6106 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6107 if (stopped()) return nullptr; // no fast path
6108 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6109
6110 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6111 if (alloc == nullptr) return nullptr;
6112
6113 Node* rawmem = memory(Compile::AliasIdxRaw);
6114 // Is the allocation's memory state untouched?
6115 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6116 // Bail out if there have been raw-memory effects since the allocation.
6117 // (Example: There might have been a call or safepoint.)
6118 return nullptr;
6119 }
6120 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6121 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6122 return nullptr;
6123 }
6124
6125 // There must be no unexpected observers of this allocation.
6126 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6127 Node* obs = ptr->fast_out(i);
6128 if (obs != this->map()) {
6129 return nullptr;
6130 }
6131 }
6132
6133 // This arraycopy must unconditionally follow the allocation of the ptr.
6134 Node* alloc_ctl = ptr->in(0);
6135 Node* ctl = control();
6136 while (ctl != alloc_ctl) {
6137 // There may be guards which feed into the slow_region.
6138 // Any other control flow means that we might not get a chance
6139 // to finish initializing the allocated object.
6140 // Various low-level checks bottom out in uncommon traps. These
6141 // are considered safe since we've already checked above that
6142 // there is no unexpected observer of this allocation.
6143 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6144 assert(ctl->in(0)->is_If(), "must be If");
6145 ctl = ctl->in(0)->in(0);
6146 } else {
6147 return nullptr;
6148 }
6149 }
6150
6151 // If we get this far, we have an allocation which immediately
6152 // precedes the arraycopy, and we can take over zeroing the new object.
6153 // The arraycopy will finish the initialization, and provide
6154 // a new control state to which we will anchor the destination pointer.
6155
6156 return alloc;
6157 }
6158
6159 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6160 if (node->is_IfProj()) {
6161 Node* other_proj = node->as_IfProj()->other_if_proj();
6162 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6163 Node* obs = other_proj->fast_out(j);
6164 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6165 (obs->as_CallStaticJava()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
6166 return obs->as_CallStaticJava();
6167 }
6168 }
6169 }
6170 return nullptr;
6171 }
6172
6173 //-------------inline_encodeISOArray-----------------------------------
6174 // encode char[] to byte[] in ISO_8859_1 or ASCII
6175 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6176 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6177 // no receiver since it is static method
6178 Node *src = argument(0);
6179 Node *src_offset = argument(1);
6180 Node *dst = argument(2);
6181 Node *dst_offset = argument(3);
6182 Node *length = argument(4);
6183
6184 src = must_be_not_null(src, true);
6185 dst = must_be_not_null(dst, true);
6186
6187 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6188 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6189 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6190 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6191 // failed array check
6192 return false;
6193 }
6194
6195 // Figure out the size and type of the elements we will be copying.
6196 BasicType src_elem = src_type->elem()->array_element_basic_type();
6197 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6198 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6199 return false;
6200 }
6201
6202 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6203 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6204 // 'src_start' points to src array + scaled offset
6205 // 'dst_start' points to dst array + scaled offset
6206
6207 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6208 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6209 enc = _gvn.transform(enc);
6210 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6211 set_memory(res_mem, mtype);
6212 set_result(enc);
6213 clear_upper_avx();
6214
6215 return true;
6216 }
6217
6218 //-------------inline_multiplyToLen-----------------------------------
6219 bool LibraryCallKit::inline_multiplyToLen() {
6220 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6221
6222 address stubAddr = StubRoutines::multiplyToLen();
6223 if (stubAddr == nullptr) {
6224 return false; // Intrinsic's stub is not implemented on this platform
6225 }
6226 const char* stubName = "multiplyToLen";
6227
6228 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6229
6230 // no receiver because it is a static method
6231 Node* x = argument(0);
6232 Node* xlen = argument(1);
6233 Node* y = argument(2);
6234 Node* ylen = argument(3);
6235 Node* z = argument(4);
6236
6237 x = must_be_not_null(x, true);
6238 y = must_be_not_null(y, true);
6239
6240 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6241 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6242 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6243 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6244 // failed array check
6245 return false;
6246 }
6247
6248 BasicType x_elem = x_type->elem()->array_element_basic_type();
6249 BasicType y_elem = y_type->elem()->array_element_basic_type();
6250 if (x_elem != T_INT || y_elem != T_INT) {
6251 return false;
6252 }
6253
6254 // Set the original stack and the reexecute bit for the interpreter to reexecute
6255 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
6256 // on the return from z array allocation in runtime.
6257 { PreserveReexecuteState preexecs(this);
6258 jvms()->set_should_reexecute(true);
6259
6260 Node* x_start = array_element_address(x, intcon(0), x_elem);
6261 Node* y_start = array_element_address(y, intcon(0), y_elem);
6262 // 'x_start' points to x array + scaled xlen
6263 // 'y_start' points to y array + scaled ylen
6264
6265 // Allocate the result array
6266 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
6267 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
6268 Node* klass_node = makecon(TypeKlassPtr::make(klass));
6269
6270 IdealKit ideal(this);
6271
6272 #define __ ideal.
6273 Node* one = __ ConI(1);
6274 Node* zero = __ ConI(0);
6275 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
6276 __ set(need_alloc, zero);
6277 __ set(z_alloc, z);
6278 __ if_then(z, BoolTest::eq, null()); {
6279 __ increment (need_alloc, one);
6280 } __ else_(); {
6281 // Update graphKit memory and control from IdealKit.
6282 sync_kit(ideal);
6283 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
6284 cast->init_req(0, control());
6285 _gvn.set_type(cast, cast->bottom_type());
6286 C->record_for_igvn(cast);
6287
6288 Node* zlen_arg = load_array_length(cast);
6289 // Update IdealKit memory and control from graphKit.
6290 __ sync_kit(this);
6291 __ if_then(zlen_arg, BoolTest::lt, zlen); {
6292 __ increment (need_alloc, one);
6293 } __ end_if();
6294 } __ end_if();
6295
6296 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
6297 // Update graphKit memory and control from IdealKit.
6298 sync_kit(ideal);
6299 Node * narr = new_array(klass_node, zlen, 1);
6300 // Update IdealKit memory and control from graphKit.
6301 __ sync_kit(this);
6302 __ set(z_alloc, narr);
6303 } __ end_if();
6304
6305 sync_kit(ideal);
6306 z = __ value(z_alloc);
6307 // Can't use TypeAryPtr::INTS which uses Bottom offset.
6308 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
6309 // Final sync IdealKit and GraphKit.
6310 final_sync(ideal);
6311 #undef __
6312
6313 Node* z_start = array_element_address(z, intcon(0), T_INT);
6314
6315 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6316 OptoRuntime::multiplyToLen_Type(),
6317 stubAddr, stubName, TypePtr::BOTTOM,
6318 x_start, xlen, y_start, ylen, z_start, zlen);
6319 } // original reexecute is set back here
6320
6321 C->set_has_split_ifs(true); // Has chance for split-if optimization
6322 set_result(z);
6323 return true;
6324 }
6325
6326 //-------------inline_squareToLen------------------------------------
6327 bool LibraryCallKit::inline_squareToLen() {
6328 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6329
6330 address stubAddr = StubRoutines::squareToLen();
6331 if (stubAddr == nullptr) {
6332 return false; // Intrinsic's stub is not implemented on this platform
6333 }
6334 const char* stubName = "squareToLen";
6335
6336 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6337
6338 Node* x = argument(0);
6339 Node* len = argument(1);
6340 Node* z = argument(2);
6341 Node* zlen = argument(3);
6342
6343 x = must_be_not_null(x, true);
6344 z = must_be_not_null(z, true);
6345
6346 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6347 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6348 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6349 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6350 // failed array check
6351 return false;
6352 }
6353
6354 BasicType x_elem = x_type->elem()->array_element_basic_type();
6355 BasicType z_elem = z_type->elem()->array_element_basic_type();
6356 if (x_elem != T_INT || z_elem != T_INT) {
6357 return false;
6358 }
6359
6360
6361 Node* x_start = array_element_address(x, intcon(0), x_elem);
6362 Node* z_start = array_element_address(z, intcon(0), z_elem);
6363
6364 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6365 OptoRuntime::squareToLen_Type(),
6366 stubAddr, stubName, TypePtr::BOTTOM,
6367 x_start, len, z_start, zlen);
6368
6369 set_result(z);
6370 return true;
6371 }
6372
6373 //-------------inline_mulAdd------------------------------------------
6374 bool LibraryCallKit::inline_mulAdd() {
6375 assert(UseMulAddIntrinsic, "not implemented on this platform");
6376
6377 address stubAddr = StubRoutines::mulAdd();
6378 if (stubAddr == nullptr) {
6379 return false; // Intrinsic's stub is not implemented on this platform
6380 }
6381 const char* stubName = "mulAdd";
6382
6383 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6384
6385 Node* out = argument(0);
6386 Node* in = argument(1);
6387 Node* offset = argument(2);
6388 Node* len = argument(3);
6389 Node* k = argument(4);
6390
6391 in = must_be_not_null(in, true);
6392 out = must_be_not_null(out, true);
6393
6394 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6395 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6396 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6397 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6398 // failed array check
6399 return false;
6400 }
6401
6402 BasicType out_elem = out_type->elem()->array_element_basic_type();
6403 BasicType in_elem = in_type->elem()->array_element_basic_type();
6404 if (out_elem != T_INT || in_elem != T_INT) {
6405 return false;
6406 }
6407
6408 Node* outlen = load_array_length(out);
6409 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6410 Node* out_start = array_element_address(out, intcon(0), out_elem);
6411 Node* in_start = array_element_address(in, intcon(0), in_elem);
6412
6413 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6414 OptoRuntime::mulAdd_Type(),
6415 stubAddr, stubName, TypePtr::BOTTOM,
6416 out_start,in_start, new_offset, len, k);
6417 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6418 set_result(result);
6419 return true;
6420 }
6421
6422 //-------------inline_montgomeryMultiply-----------------------------------
6423 bool LibraryCallKit::inline_montgomeryMultiply() {
6424 address stubAddr = StubRoutines::montgomeryMultiply();
6425 if (stubAddr == nullptr) {
6426 return false; // Intrinsic's stub is not implemented on this platform
6427 }
6428
6429 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6430 const char* stubName = "montgomery_multiply";
6431
6432 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6433
6434 Node* a = argument(0);
6435 Node* b = argument(1);
6436 Node* n = argument(2);
6437 Node* len = argument(3);
6438 Node* inv = argument(4);
6439 Node* m = argument(6);
6440
6441 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6442 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6443 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6444 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6445 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6446 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6447 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6448 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6449 // failed array check
6450 return false;
6451 }
6452
6453 BasicType a_elem = a_type->elem()->array_element_basic_type();
6454 BasicType b_elem = b_type->elem()->array_element_basic_type();
6455 BasicType n_elem = n_type->elem()->array_element_basic_type();
6456 BasicType m_elem = m_type->elem()->array_element_basic_type();
6457 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6458 return false;
6459 }
6460
6461 // Make the call
6462 {
6463 Node* a_start = array_element_address(a, intcon(0), a_elem);
6464 Node* b_start = array_element_address(b, intcon(0), b_elem);
6465 Node* n_start = array_element_address(n, intcon(0), n_elem);
6466 Node* m_start = array_element_address(m, intcon(0), m_elem);
6467
6468 Node* call = make_runtime_call(RC_LEAF,
6469 OptoRuntime::montgomeryMultiply_Type(),
6470 stubAddr, stubName, TypePtr::BOTTOM,
6471 a_start, b_start, n_start, len, inv, top(),
6472 m_start);
6473 set_result(m);
6474 }
6475
6476 return true;
6477 }
6478
6479 bool LibraryCallKit::inline_montgomerySquare() {
6480 address stubAddr = StubRoutines::montgomerySquare();
6481 if (stubAddr == nullptr) {
6482 return false; // Intrinsic's stub is not implemented on this platform
6483 }
6484
6485 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6486 const char* stubName = "montgomery_square";
6487
6488 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6489
6490 Node* a = argument(0);
6491 Node* n = argument(1);
6492 Node* len = argument(2);
6493 Node* inv = argument(3);
6494 Node* m = argument(5);
6495
6496 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6497 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6498 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6499 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6500 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6501 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6502 // failed array check
6503 return false;
6504 }
6505
6506 BasicType a_elem = a_type->elem()->array_element_basic_type();
6507 BasicType n_elem = n_type->elem()->array_element_basic_type();
6508 BasicType m_elem = m_type->elem()->array_element_basic_type();
6509 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6510 return false;
6511 }
6512
6513 // Make the call
6514 {
6515 Node* a_start = array_element_address(a, intcon(0), a_elem);
6516 Node* n_start = array_element_address(n, intcon(0), n_elem);
6517 Node* m_start = array_element_address(m, intcon(0), m_elem);
6518
6519 Node* call = make_runtime_call(RC_LEAF,
6520 OptoRuntime::montgomerySquare_Type(),
6521 stubAddr, stubName, TypePtr::BOTTOM,
6522 a_start, n_start, len, inv, top(),
6523 m_start);
6524 set_result(m);
6525 }
6526
6527 return true;
6528 }
6529
6530 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6531 address stubAddr = nullptr;
6532 const char* stubName = nullptr;
6533
6534 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6535 if (stubAddr == nullptr) {
6536 return false; // Intrinsic's stub is not implemented on this platform
6537 }
6538
6539 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6540
6541 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6542
6543 Node* newArr = argument(0);
6544 Node* oldArr = argument(1);
6545 Node* newIdx = argument(2);
6546 Node* shiftCount = argument(3);
6547 Node* numIter = argument(4);
6548
6549 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6550 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6551 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6552 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6553 return false;
6554 }
6555
6556 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6557 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6558 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6559 return false;
6560 }
6561
6562 // Make the call
6563 {
6564 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6565 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6566
6567 Node* call = make_runtime_call(RC_LEAF,
6568 OptoRuntime::bigIntegerShift_Type(),
6569 stubAddr,
6570 stubName,
6571 TypePtr::BOTTOM,
6572 newArr_start,
6573 oldArr_start,
6574 newIdx,
6575 shiftCount,
6576 numIter);
6577 }
6578
6579 return true;
6580 }
6581
6582 //-------------inline_vectorizedMismatch------------------------------
6583 bool LibraryCallKit::inline_vectorizedMismatch() {
6584 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6585
6586 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6587 Node* obja = argument(0); // Object
6588 Node* aoffset = argument(1); // long
6589 Node* objb = argument(3); // Object
6590 Node* boffset = argument(4); // long
6591 Node* length = argument(6); // int
6592 Node* scale = argument(7); // int
6593
6594 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6595 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6596 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6597 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6598 scale == top()) {
6599 return false; // failed input validation
6600 }
6601
6602 Node* obja_adr = make_unsafe_address(obja, aoffset);
6603 Node* objb_adr = make_unsafe_address(objb, boffset);
6604
6605 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6606 //
6607 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6608 // if (length <= inline_limit) {
6609 // inline_path:
6610 // vmask = VectorMaskGen length
6611 // vload1 = LoadVectorMasked obja, vmask
6612 // vload2 = LoadVectorMasked objb, vmask
6613 // result1 = VectorCmpMasked vload1, vload2, vmask
6614 // } else {
6615 // call_stub_path:
6616 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6617 // }
6618 // exit_block:
6619 // return Phi(result1, result2);
6620 //
6621 enum { inline_path = 1, // input is small enough to process it all at once
6622 stub_path = 2, // input is too large; call into the VM
6623 PATH_LIMIT = 3
6624 };
6625
6626 Node* exit_block = new RegionNode(PATH_LIMIT);
6627 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6628 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6629
6630 Node* call_stub_path = control();
6631
6632 BasicType elem_bt = T_ILLEGAL;
6633
6634 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6635 if (scale_t->is_con()) {
6636 switch (scale_t->get_con()) {
6637 case 0: elem_bt = T_BYTE; break;
6638 case 1: elem_bt = T_SHORT; break;
6639 case 2: elem_bt = T_INT; break;
6640 case 3: elem_bt = T_LONG; break;
6641
6642 default: elem_bt = T_ILLEGAL; break; // not supported
6643 }
6644 }
6645
6646 int inline_limit = 0;
6647 bool do_partial_inline = false;
6648
6649 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6650 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6651 do_partial_inline = inline_limit >= 16;
6652 }
6653
6654 if (do_partial_inline) {
6655 assert(elem_bt != T_ILLEGAL, "sanity");
6656
6657 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6658 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6659 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6660
6661 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6662 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6663 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6664
6665 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6666
6667 if (!stopped()) {
6668 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6669
6670 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6671 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6672 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6673 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6674
6675 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6676 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6677 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6678 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6679
6680 exit_block->init_req(inline_path, control());
6681 memory_phi->init_req(inline_path, map()->memory());
6682 result_phi->init_req(inline_path, result);
6683
6684 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6685 clear_upper_avx();
6686 }
6687 }
6688 }
6689
6690 if (call_stub_path != nullptr) {
6691 set_control(call_stub_path);
6692
6693 Node* call = make_runtime_call(RC_LEAF,
6694 OptoRuntime::vectorizedMismatch_Type(),
6695 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6696 obja_adr, objb_adr, length, scale);
6697
6698 exit_block->init_req(stub_path, control());
6699 memory_phi->init_req(stub_path, map()->memory());
6700 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6701 }
6702
6703 exit_block = _gvn.transform(exit_block);
6704 memory_phi = _gvn.transform(memory_phi);
6705 result_phi = _gvn.transform(result_phi);
6706
6707 set_control(exit_block);
6708 set_all_memory(memory_phi);
6709 set_result(result_phi);
6710
6711 return true;
6712 }
6713
6714 //------------------------------inline_vectorizedHashcode----------------------------
6715 bool LibraryCallKit::inline_vectorizedHashCode() {
6716 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6717
6718 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6719 Node* array = argument(0);
6720 Node* offset = argument(1);
6721 Node* length = argument(2);
6722 Node* initialValue = argument(3);
6723 Node* basic_type = argument(4);
6724
6725 if (basic_type == top()) {
6726 return false; // failed input validation
6727 }
6728
6729 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6730 if (!basic_type_t->is_con()) {
6731 return false; // Only intrinsify if mode argument is constant
6732 }
6733
6734 array = must_be_not_null(array, true);
6735
6736 BasicType bt = (BasicType)basic_type_t->get_con();
6737
6738 // Resolve address of first element
6739 Node* array_start = array_element_address(array, offset, bt);
6740
6741 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6742 array_start, length, initialValue, basic_type)));
6743 clear_upper_avx();
6744
6745 return true;
6746 }
6747
6748 /**
6749 * Calculate CRC32 for byte.
6750 * int java.util.zip.CRC32.update(int crc, int b)
6751 */
6752 bool LibraryCallKit::inline_updateCRC32() {
6753 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6754 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6755 // no receiver since it is static method
6756 Node* crc = argument(0); // type: int
6757 Node* b = argument(1); // type: int
6758
6759 /*
6760 * int c = ~ crc;
6761 * b = timesXtoThe32[(b ^ c) & 0xFF];
6762 * b = b ^ (c >>> 8);
6763 * crc = ~b;
6764 */
6765
6766 Node* M1 = intcon(-1);
6767 crc = _gvn.transform(new XorINode(crc, M1));
6768 Node* result = _gvn.transform(new XorINode(crc, b));
6769 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6770
6771 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6772 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6773 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6774 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6775
6776 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6777 result = _gvn.transform(new XorINode(crc, result));
6778 result = _gvn.transform(new XorINode(result, M1));
6779 set_result(result);
6780 return true;
6781 }
6782
6783 /**
6784 * Calculate CRC32 for byte[] array.
6785 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6786 */
6787 bool LibraryCallKit::inline_updateBytesCRC32() {
6788 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6789 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6790 // no receiver since it is static method
6791 Node* crc = argument(0); // type: int
6792 Node* src = argument(1); // type: oop
6793 Node* offset = argument(2); // type: int
6794 Node* length = argument(3); // type: int
6795
6796 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6797 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6798 // failed array check
6799 return false;
6800 }
6801
6802 // Figure out the size and type of the elements we will be copying.
6803 BasicType src_elem = src_type->elem()->array_element_basic_type();
6804 if (src_elem != T_BYTE) {
6805 return false;
6806 }
6807
6808 // 'src_start' points to src array + scaled offset
6809 src = must_be_not_null(src, true);
6810 Node* src_start = array_element_address(src, offset, src_elem);
6811
6812 // We assume that range check is done by caller.
6813 // TODO: generate range check (offset+length < src.length) in debug VM.
6814
6815 // Call the stub.
6816 address stubAddr = StubRoutines::updateBytesCRC32();
6817 const char *stubName = "updateBytesCRC32";
6818
6819 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6820 stubAddr, stubName, TypePtr::BOTTOM,
6821 crc, src_start, length);
6822 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6823 set_result(result);
6824 return true;
6825 }
6826
6827 /**
6828 * Calculate CRC32 for ByteBuffer.
6829 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6830 */
6831 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6832 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6833 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6834 // no receiver since it is static method
6835 Node* crc = argument(0); // type: int
6836 Node* src = argument(1); // type: long
6837 Node* offset = argument(3); // type: int
6838 Node* length = argument(4); // type: int
6839
6840 src = ConvL2X(src); // adjust Java long to machine word
6841 Node* base = _gvn.transform(new CastX2PNode(src));
6842 offset = ConvI2X(offset);
6843
6844 // 'src_start' points to src array + scaled offset
6845 Node* src_start = basic_plus_adr(top(), base, offset);
6846
6847 // Call the stub.
6848 address stubAddr = StubRoutines::updateBytesCRC32();
6849 const char *stubName = "updateBytesCRC32";
6850
6851 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6852 stubAddr, stubName, TypePtr::BOTTOM,
6853 crc, src_start, length);
6854 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6855 set_result(result);
6856 return true;
6857 }
6858
6859 //------------------------------get_table_from_crc32c_class-----------------------
6860 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6861 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6862 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6863
6864 return table;
6865 }
6866
6867 //------------------------------inline_updateBytesCRC32C-----------------------
6868 //
6869 // Calculate CRC32C for byte[] array.
6870 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6871 //
6872 bool LibraryCallKit::inline_updateBytesCRC32C() {
6873 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6874 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6875 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6876 // no receiver since it is a static method
6877 Node* crc = argument(0); // type: int
6878 Node* src = argument(1); // type: oop
6879 Node* offset = argument(2); // type: int
6880 Node* end = argument(3); // type: int
6881
6882 Node* length = _gvn.transform(new SubINode(end, offset));
6883
6884 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6885 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6886 // failed array check
6887 return false;
6888 }
6889
6890 // Figure out the size and type of the elements we will be copying.
6891 BasicType src_elem = src_type->elem()->array_element_basic_type();
6892 if (src_elem != T_BYTE) {
6893 return false;
6894 }
6895
6896 // 'src_start' points to src array + scaled offset
6897 src = must_be_not_null(src, true);
6898 Node* src_start = array_element_address(src, offset, src_elem);
6899
6900 // static final int[] byteTable in class CRC32C
6901 Node* table = get_table_from_crc32c_class(callee()->holder());
6902 table = must_be_not_null(table, true);
6903 Node* table_start = array_element_address(table, intcon(0), T_INT);
6904
6905 // We assume that range check is done by caller.
6906 // TODO: generate range check (offset+length < src.length) in debug VM.
6907
6908 // Call the stub.
6909 address stubAddr = StubRoutines::updateBytesCRC32C();
6910 const char *stubName = "updateBytesCRC32C";
6911
6912 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6913 stubAddr, stubName, TypePtr::BOTTOM,
6914 crc, src_start, length, table_start);
6915 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6916 set_result(result);
6917 return true;
6918 }
6919
6920 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6921 //
6922 // Calculate CRC32C for DirectByteBuffer.
6923 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6924 //
6925 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6926 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6927 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6928 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6929 // no receiver since it is a static method
6930 Node* crc = argument(0); // type: int
6931 Node* src = argument(1); // type: long
6932 Node* offset = argument(3); // type: int
6933 Node* end = argument(4); // type: int
6934
6935 Node* length = _gvn.transform(new SubINode(end, offset));
6936
6937 src = ConvL2X(src); // adjust Java long to machine word
6938 Node* base = _gvn.transform(new CastX2PNode(src));
6939 offset = ConvI2X(offset);
6940
6941 // 'src_start' points to src array + scaled offset
6942 Node* src_start = basic_plus_adr(top(), base, offset);
6943
6944 // static final int[] byteTable in class CRC32C
6945 Node* table = get_table_from_crc32c_class(callee()->holder());
6946 table = must_be_not_null(table, true);
6947 Node* table_start = array_element_address(table, intcon(0), T_INT);
6948
6949 // Call the stub.
6950 address stubAddr = StubRoutines::updateBytesCRC32C();
6951 const char *stubName = "updateBytesCRC32C";
6952
6953 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6954 stubAddr, stubName, TypePtr::BOTTOM,
6955 crc, src_start, length, table_start);
6956 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6957 set_result(result);
6958 return true;
6959 }
6960
6961 //------------------------------inline_updateBytesAdler32----------------------
6962 //
6963 // Calculate Adler32 checksum for byte[] array.
6964 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6965 //
6966 bool LibraryCallKit::inline_updateBytesAdler32() {
6967 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6968 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6969 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6970 // no receiver since it is static method
6971 Node* crc = argument(0); // type: int
6972 Node* src = argument(1); // type: oop
6973 Node* offset = argument(2); // type: int
6974 Node* length = argument(3); // type: int
6975
6976 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6977 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6978 // failed array check
6979 return false;
6980 }
6981
6982 // Figure out the size and type of the elements we will be copying.
6983 BasicType src_elem = src_type->elem()->array_element_basic_type();
6984 if (src_elem != T_BYTE) {
6985 return false;
6986 }
6987
6988 // 'src_start' points to src array + scaled offset
6989 Node* src_start = array_element_address(src, offset, src_elem);
6990
6991 // We assume that range check is done by caller.
6992 // TODO: generate range check (offset+length < src.length) in debug VM.
6993
6994 // Call the stub.
6995 address stubAddr = StubRoutines::updateBytesAdler32();
6996 const char *stubName = "updateBytesAdler32";
6997
6998 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6999 stubAddr, stubName, TypePtr::BOTTOM,
7000 crc, src_start, length);
7001 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7002 set_result(result);
7003 return true;
7004 }
7005
7006 //------------------------------inline_updateByteBufferAdler32---------------
7007 //
7008 // Calculate Adler32 checksum for DirectByteBuffer.
7009 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
7010 //
7011 bool LibraryCallKit::inline_updateByteBufferAdler32() {
7012 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7013 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7014 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7015 // no receiver since it is static method
7016 Node* crc = argument(0); // type: int
7017 Node* src = argument(1); // type: long
7018 Node* offset = argument(3); // type: int
7019 Node* length = argument(4); // type: int
7020
7021 src = ConvL2X(src); // adjust Java long to machine word
7022 Node* base = _gvn.transform(new CastX2PNode(src));
7023 offset = ConvI2X(offset);
7024
7025 // 'src_start' points to src array + scaled offset
7026 Node* src_start = basic_plus_adr(top(), base, offset);
7027
7028 // Call the stub.
7029 address stubAddr = StubRoutines::updateBytesAdler32();
7030 const char *stubName = "updateBytesAdler32";
7031
7032 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7033 stubAddr, stubName, TypePtr::BOTTOM,
7034 crc, src_start, length);
7035
7036 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7037 set_result(result);
7038 return true;
7039 }
7040
7041 //----------------------------inline_reference_get----------------------------
7042 // public T java.lang.ref.Reference.get();
7043 bool LibraryCallKit::inline_reference_get() {
7044 const int referent_offset = java_lang_ref_Reference::referent_offset();
7045
7046 // Get the argument:
7047 Node* reference_obj = null_check_receiver();
7048 if (stopped()) return true;
7049
7050 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7051 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7052 decorators, /*is_static*/ false, nullptr);
7053 if (result == nullptr) return false;
7054
7055 // Add memory barrier to prevent commoning reads from this field
7056 // across safepoint since GC can change its value.
7057 insert_mem_bar(Op_MemBarCPUOrder);
7058
7059 set_result(result);
7060 return true;
7061 }
7062
7063 //----------------------------inline_reference_refersTo0----------------------------
7064 // bool java.lang.ref.Reference.refersTo0();
7065 // bool java.lang.ref.PhantomReference.refersTo0();
7066 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7067 // Get arguments:
7068 Node* reference_obj = null_check_receiver();
7069 Node* other_obj = argument(1);
7070 if (stopped()) return true;
7071
7072 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7073 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7074 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7075 decorators, /*is_static*/ false, nullptr);
7076 if (referent == nullptr) return false;
7077
7078 // Add memory barrier to prevent commoning reads from this field
7079 // across safepoint since GC can change its value.
7080 insert_mem_bar(Op_MemBarCPUOrder);
7081
7082 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7083 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7084 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7085
7086 RegionNode* region = new RegionNode(3);
7087 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7088
7089 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7090 region->init_req(1, if_true);
7091 phi->init_req(1, intcon(1));
7092
7093 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7094 region->init_req(2, if_false);
7095 phi->init_req(2, intcon(0));
7096
7097 set_control(_gvn.transform(region));
7098 record_for_igvn(region);
7099 set_result(_gvn.transform(phi));
7100 return true;
7101 }
7102
7103
7104 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7105 DecoratorSet decorators, bool is_static,
7106 ciInstanceKlass* fromKls) {
7107 if (fromKls == nullptr) {
7108 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7109 assert(tinst != nullptr, "obj is null");
7110 assert(tinst->is_loaded(), "obj is not loaded");
7111 fromKls = tinst->instance_klass();
7112 } else {
7113 assert(is_static, "only for static field access");
7114 }
7115 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7116 ciSymbol::make(fieldTypeString),
7117 is_static);
7118
7119 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7120 if (field == nullptr) return (Node *) nullptr;
7121
7122 if (is_static) {
7123 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7124 fromObj = makecon(tip);
7125 }
7126
7127 // Next code copied from Parse::do_get_xxx():
7128
7129 // Compute address and memory type.
7130 int offset = field->offset_in_bytes();
7131 bool is_vol = field->is_volatile();
7132 ciType* field_klass = field->type();
7133 assert(field_klass->is_loaded(), "should be loaded");
7134 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7135 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7136 BasicType bt = field->layout_type();
7137
7138 // Build the resultant type of the load
7139 const Type *type;
7140 if (bt == T_OBJECT) {
7141 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7142 } else {
7143 type = Type::get_const_basic_type(bt);
7144 }
7145
7146 if (is_vol) {
7147 decorators |= MO_SEQ_CST;
7148 }
7149
7150 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7151 }
7152
7153 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7154 bool is_exact /* true */, bool is_static /* false */,
7155 ciInstanceKlass * fromKls /* nullptr */) {
7156 if (fromKls == nullptr) {
7157 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7158 assert(tinst != nullptr, "obj is null");
7159 assert(tinst->is_loaded(), "obj is not loaded");
7160 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7161 fromKls = tinst->instance_klass();
7162 }
7163 else {
7164 assert(is_static, "only for static field access");
7165 }
7166 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7167 ciSymbol::make(fieldTypeString),
7168 is_static);
7169
7170 assert(field != nullptr, "undefined field");
7171 assert(!field->is_volatile(), "not defined for volatile fields");
7172
7173 if (is_static) {
7174 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7175 fromObj = makecon(tip);
7176 }
7177
7178 // Next code copied from Parse::do_get_xxx():
7179
7180 // Compute address and memory type.
7181 int offset = field->offset_in_bytes();
7182 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7183
7184 return adr;
7185 }
7186
7187 //------------------------------inline_aescrypt_Block-----------------------
7188 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7189 address stubAddr = nullptr;
7190 const char *stubName;
7191 assert(UseAES, "need AES instruction support");
7192
7193 switch(id) {
7194 case vmIntrinsics::_aescrypt_encryptBlock:
7195 stubAddr = StubRoutines::aescrypt_encryptBlock();
7196 stubName = "aescrypt_encryptBlock";
7197 break;
7198 case vmIntrinsics::_aescrypt_decryptBlock:
7199 stubAddr = StubRoutines::aescrypt_decryptBlock();
7200 stubName = "aescrypt_decryptBlock";
7201 break;
7202 default:
7203 break;
7204 }
7205 if (stubAddr == nullptr) return false;
7206
7207 Node* aescrypt_object = argument(0);
7208 Node* src = argument(1);
7209 Node* src_offset = argument(2);
7210 Node* dest = argument(3);
7211 Node* dest_offset = argument(4);
7212
7213 src = must_be_not_null(src, true);
7214 dest = must_be_not_null(dest, true);
7215
7216 // (1) src and dest are arrays.
7217 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7218 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7219 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7220 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7221
7222 // for the quick and dirty code we will skip all the checks.
7223 // we are just trying to get the call to be generated.
7224 Node* src_start = src;
7225 Node* dest_start = dest;
7226 if (src_offset != nullptr || dest_offset != nullptr) {
7227 assert(src_offset != nullptr && dest_offset != nullptr, "");
7228 src_start = array_element_address(src, src_offset, T_BYTE);
7229 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7230 }
7231
7232 // now need to get the start of its expanded key array
7233 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7234 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7235 if (k_start == nullptr) return false;
7236
7237 // Call the stub.
7238 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7239 stubAddr, stubName, TypePtr::BOTTOM,
7240 src_start, dest_start, k_start);
7241
7242 return true;
7243 }
7244
7245 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7246 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7247 address stubAddr = nullptr;
7248 const char *stubName = nullptr;
7249
7250 assert(UseAES, "need AES instruction support");
7251
7252 switch(id) {
7253 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7254 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7255 stubName = "cipherBlockChaining_encryptAESCrypt";
7256 break;
7257 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7258 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7259 stubName = "cipherBlockChaining_decryptAESCrypt";
7260 break;
7261 default:
7262 break;
7263 }
7264 if (stubAddr == nullptr) return false;
7265
7266 Node* cipherBlockChaining_object = argument(0);
7267 Node* src = argument(1);
7268 Node* src_offset = argument(2);
7269 Node* len = argument(3);
7270 Node* dest = argument(4);
7271 Node* dest_offset = argument(5);
7272
7273 src = must_be_not_null(src, false);
7274 dest = must_be_not_null(dest, false);
7275
7276 // (1) src and dest are arrays.
7277 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7278 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7279 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7280 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7281
7282 // checks are the responsibility of the caller
7283 Node* src_start = src;
7284 Node* dest_start = dest;
7285 if (src_offset != nullptr || dest_offset != nullptr) {
7286 assert(src_offset != nullptr && dest_offset != nullptr, "");
7287 src_start = array_element_address(src, src_offset, T_BYTE);
7288 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7289 }
7290
7291 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7292 // (because of the predicated logic executed earlier).
7293 // so we cast it here safely.
7294 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7295
7296 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7297 if (embeddedCipherObj == nullptr) return false;
7298
7299 // cast it to what we know it will be at runtime
7300 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7301 assert(tinst != nullptr, "CBC obj is null");
7302 assert(tinst->is_loaded(), "CBC obj is not loaded");
7303 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7304 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7305
7306 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7307 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7308 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7309 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7310 aescrypt_object = _gvn.transform(aescrypt_object);
7311
7312 // we need to get the start of the aescrypt_object's expanded key array
7313 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7314 if (k_start == nullptr) return false;
7315
7316 // similarly, get the start address of the r vector
7317 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7318 if (objRvec == nullptr) return false;
7319 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7320
7321 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7322 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7323 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7324 stubAddr, stubName, TypePtr::BOTTOM,
7325 src_start, dest_start, k_start, r_start, len);
7326
7327 // return cipher length (int)
7328 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7329 set_result(retvalue);
7330 return true;
7331 }
7332
7333 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7334 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7335 address stubAddr = nullptr;
7336 const char *stubName = nullptr;
7337
7338 assert(UseAES, "need AES instruction support");
7339
7340 switch (id) {
7341 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7342 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7343 stubName = "electronicCodeBook_encryptAESCrypt";
7344 break;
7345 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7346 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7347 stubName = "electronicCodeBook_decryptAESCrypt";
7348 break;
7349 default:
7350 break;
7351 }
7352
7353 if (stubAddr == nullptr) return false;
7354
7355 Node* electronicCodeBook_object = argument(0);
7356 Node* src = argument(1);
7357 Node* src_offset = argument(2);
7358 Node* len = argument(3);
7359 Node* dest = argument(4);
7360 Node* dest_offset = argument(5);
7361
7362 // (1) src and dest are arrays.
7363 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7364 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7365 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7366 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7367
7368 // checks are the responsibility of the caller
7369 Node* src_start = src;
7370 Node* dest_start = dest;
7371 if (src_offset != nullptr || dest_offset != nullptr) {
7372 assert(src_offset != nullptr && dest_offset != nullptr, "");
7373 src_start = array_element_address(src, src_offset, T_BYTE);
7374 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7375 }
7376
7377 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7378 // (because of the predicated logic executed earlier).
7379 // so we cast it here safely.
7380 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7381
7382 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7383 if (embeddedCipherObj == nullptr) return false;
7384
7385 // cast it to what we know it will be at runtime
7386 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7387 assert(tinst != nullptr, "ECB obj is null");
7388 assert(tinst->is_loaded(), "ECB obj is not loaded");
7389 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7390 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7391
7392 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7393 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7394 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7395 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7396 aescrypt_object = _gvn.transform(aescrypt_object);
7397
7398 // we need to get the start of the aescrypt_object's expanded key array
7399 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7400 if (k_start == nullptr) return false;
7401
7402 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7403 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7404 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7405 stubAddr, stubName, TypePtr::BOTTOM,
7406 src_start, dest_start, k_start, len);
7407
7408 // return cipher length (int)
7409 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7410 set_result(retvalue);
7411 return true;
7412 }
7413
7414 //------------------------------inline_counterMode_AESCrypt-----------------------
7415 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7416 assert(UseAES, "need AES instruction support");
7417 if (!UseAESCTRIntrinsics) return false;
7418
7419 address stubAddr = nullptr;
7420 const char *stubName = nullptr;
7421 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7422 stubAddr = StubRoutines::counterMode_AESCrypt();
7423 stubName = "counterMode_AESCrypt";
7424 }
7425 if (stubAddr == nullptr) return false;
7426
7427 Node* counterMode_object = argument(0);
7428 Node* src = argument(1);
7429 Node* src_offset = argument(2);
7430 Node* len = argument(3);
7431 Node* dest = argument(4);
7432 Node* dest_offset = argument(5);
7433
7434 // (1) src and dest are arrays.
7435 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7436 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7437 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7438 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7439
7440 // checks are the responsibility of the caller
7441 Node* src_start = src;
7442 Node* dest_start = dest;
7443 if (src_offset != nullptr || dest_offset != nullptr) {
7444 assert(src_offset != nullptr && dest_offset != nullptr, "");
7445 src_start = array_element_address(src, src_offset, T_BYTE);
7446 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7447 }
7448
7449 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7450 // (because of the predicated logic executed earlier).
7451 // so we cast it here safely.
7452 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7453 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7454 if (embeddedCipherObj == nullptr) return false;
7455 // cast it to what we know it will be at runtime
7456 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7457 assert(tinst != nullptr, "CTR obj is null");
7458 assert(tinst->is_loaded(), "CTR obj is not loaded");
7459 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7460 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7461 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7462 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7463 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7464 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7465 aescrypt_object = _gvn.transform(aescrypt_object);
7466 // we need to get the start of the aescrypt_object's expanded key array
7467 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7468 if (k_start == nullptr) return false;
7469 // similarly, get the start address of the r vector
7470 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7471 if (obj_counter == nullptr) return false;
7472 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7473
7474 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7475 if (saved_encCounter == nullptr) return false;
7476 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7477 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7478
7479 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7480 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7481 OptoRuntime::counterMode_aescrypt_Type(),
7482 stubAddr, stubName, TypePtr::BOTTOM,
7483 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7484
7485 // return cipher length (int)
7486 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7487 set_result(retvalue);
7488 return true;
7489 }
7490
7491 //------------------------------get_key_start_from_aescrypt_object-----------------------
7492 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7493 #if defined(PPC64) || defined(S390)
7494 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7495 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7496 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7497 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7498 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7499 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7500 if (objSessionK == nullptr) {
7501 return (Node *) nullptr;
7502 }
7503 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7504 #else
7505 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7506 #endif // PPC64
7507 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7508 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7509
7510 // now have the array, need to get the start address of the K array
7511 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7512 return k_start;
7513 }
7514
7515 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7516 // Return node representing slow path of predicate check.
7517 // the pseudo code we want to emulate with this predicate is:
7518 // for encryption:
7519 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7520 // for decryption:
7521 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7522 // note cipher==plain is more conservative than the original java code but that's OK
7523 //
7524 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7525 // The receiver was checked for null already.
7526 Node* objCBC = argument(0);
7527
7528 Node* src = argument(1);
7529 Node* dest = argument(4);
7530
7531 // Load embeddedCipher field of CipherBlockChaining object.
7532 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7533
7534 // get AESCrypt klass for instanceOf check
7535 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7536 // will have same classloader as CipherBlockChaining object
7537 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7538 assert(tinst != nullptr, "CBCobj is null");
7539 assert(tinst->is_loaded(), "CBCobj is not loaded");
7540
7541 // we want to do an instanceof comparison against the AESCrypt class
7542 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7543 if (!klass_AESCrypt->is_loaded()) {
7544 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7545 Node* ctrl = control();
7546 set_control(top()); // no regular fast path
7547 return ctrl;
7548 }
7549
7550 src = must_be_not_null(src, true);
7551 dest = must_be_not_null(dest, true);
7552
7553 // Resolve oops to stable for CmpP below.
7554 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7555
7556 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7557 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7558 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7559
7560 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7561
7562 // for encryption, we are done
7563 if (!decrypting)
7564 return instof_false; // even if it is null
7565
7566 // for decryption, we need to add a further check to avoid
7567 // taking the intrinsic path when cipher and plain are the same
7568 // see the original java code for why.
7569 RegionNode* region = new RegionNode(3);
7570 region->init_req(1, instof_false);
7571
7572 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7573 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7574 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7575 region->init_req(2, src_dest_conjoint);
7576
7577 record_for_igvn(region);
7578 return _gvn.transform(region);
7579 }
7580
7581 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7582 // Return node representing slow path of predicate check.
7583 // the pseudo code we want to emulate with this predicate is:
7584 // for encryption:
7585 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7586 // for decryption:
7587 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7588 // note cipher==plain is more conservative than the original java code but that's OK
7589 //
7590 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7591 // The receiver was checked for null already.
7592 Node* objECB = argument(0);
7593
7594 // Load embeddedCipher field of ElectronicCodeBook object.
7595 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7596
7597 // get AESCrypt klass for instanceOf check
7598 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7599 // will have same classloader as ElectronicCodeBook object
7600 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7601 assert(tinst != nullptr, "ECBobj is null");
7602 assert(tinst->is_loaded(), "ECBobj is not loaded");
7603
7604 // we want to do an instanceof comparison against the AESCrypt class
7605 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7606 if (!klass_AESCrypt->is_loaded()) {
7607 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7608 Node* ctrl = control();
7609 set_control(top()); // no regular fast path
7610 return ctrl;
7611 }
7612 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7613
7614 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7615 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7616 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7617
7618 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7619
7620 // for encryption, we are done
7621 if (!decrypting)
7622 return instof_false; // even if it is null
7623
7624 // for decryption, we need to add a further check to avoid
7625 // taking the intrinsic path when cipher and plain are the same
7626 // see the original java code for why.
7627 RegionNode* region = new RegionNode(3);
7628 region->init_req(1, instof_false);
7629 Node* src = argument(1);
7630 Node* dest = argument(4);
7631 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7632 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7633 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7634 region->init_req(2, src_dest_conjoint);
7635
7636 record_for_igvn(region);
7637 return _gvn.transform(region);
7638 }
7639
7640 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7641 // Return node representing slow path of predicate check.
7642 // the pseudo code we want to emulate with this predicate is:
7643 // for encryption:
7644 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7645 // for decryption:
7646 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7647 // note cipher==plain is more conservative than the original java code but that's OK
7648 //
7649
7650 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7651 // The receiver was checked for null already.
7652 Node* objCTR = argument(0);
7653
7654 // Load embeddedCipher field of CipherBlockChaining object.
7655 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7656
7657 // get AESCrypt klass for instanceOf check
7658 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7659 // will have same classloader as CipherBlockChaining object
7660 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7661 assert(tinst != nullptr, "CTRobj is null");
7662 assert(tinst->is_loaded(), "CTRobj is not loaded");
7663
7664 // we want to do an instanceof comparison against the AESCrypt class
7665 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7666 if (!klass_AESCrypt->is_loaded()) {
7667 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7668 Node* ctrl = control();
7669 set_control(top()); // no regular fast path
7670 return ctrl;
7671 }
7672
7673 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7674 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7675 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7676 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7677 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7678
7679 return instof_false; // even if it is null
7680 }
7681
7682 //------------------------------inline_ghash_processBlocks
7683 bool LibraryCallKit::inline_ghash_processBlocks() {
7684 address stubAddr;
7685 const char *stubName;
7686 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7687
7688 stubAddr = StubRoutines::ghash_processBlocks();
7689 stubName = "ghash_processBlocks";
7690
7691 Node* data = argument(0);
7692 Node* offset = argument(1);
7693 Node* len = argument(2);
7694 Node* state = argument(3);
7695 Node* subkeyH = argument(4);
7696
7697 state = must_be_not_null(state, true);
7698 subkeyH = must_be_not_null(subkeyH, true);
7699 data = must_be_not_null(data, true);
7700
7701 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7702 assert(state_start, "state is null");
7703 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7704 assert(subkeyH_start, "subkeyH is null");
7705 Node* data_start = array_element_address(data, offset, T_BYTE);
7706 assert(data_start, "data is null");
7707
7708 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7709 OptoRuntime::ghash_processBlocks_Type(),
7710 stubAddr, stubName, TypePtr::BOTTOM,
7711 state_start, subkeyH_start, data_start, len);
7712 return true;
7713 }
7714
7715 //------------------------------inline_chacha20Block
7716 bool LibraryCallKit::inline_chacha20Block() {
7717 address stubAddr;
7718 const char *stubName;
7719 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7720
7721 stubAddr = StubRoutines::chacha20Block();
7722 stubName = "chacha20Block";
7723
7724 Node* state = argument(0);
7725 Node* result = argument(1);
7726
7727 state = must_be_not_null(state, true);
7728 result = must_be_not_null(result, true);
7729
7730 Node* state_start = array_element_address(state, intcon(0), T_INT);
7731 assert(state_start, "state is null");
7732 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7733 assert(result_start, "result is null");
7734
7735 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7736 OptoRuntime::chacha20Block_Type(),
7737 stubAddr, stubName, TypePtr::BOTTOM,
7738 state_start, result_start);
7739 // return key stream length (int)
7740 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7741 set_result(retvalue);
7742 return true;
7743 }
7744
7745 bool LibraryCallKit::inline_base64_encodeBlock() {
7746 address stubAddr;
7747 const char *stubName;
7748 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7749 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
7750 stubAddr = StubRoutines::base64_encodeBlock();
7751 stubName = "encodeBlock";
7752
7753 if (!stubAddr) return false;
7754 Node* base64obj = argument(0);
7755 Node* src = argument(1);
7756 Node* offset = argument(2);
7757 Node* len = argument(3);
7758 Node* dest = argument(4);
7759 Node* dp = argument(5);
7760 Node* isURL = argument(6);
7761
7762 src = must_be_not_null(src, true);
7763 dest = must_be_not_null(dest, true);
7764
7765 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7766 assert(src_start, "source array is null");
7767 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7768 assert(dest_start, "destination array is null");
7769
7770 Node* base64 = make_runtime_call(RC_LEAF,
7771 OptoRuntime::base64_encodeBlock_Type(),
7772 stubAddr, stubName, TypePtr::BOTTOM,
7773 src_start, offset, len, dest_start, dp, isURL);
7774 return true;
7775 }
7776
7777 bool LibraryCallKit::inline_base64_decodeBlock() {
7778 address stubAddr;
7779 const char *stubName;
7780 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7781 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
7782 stubAddr = StubRoutines::base64_decodeBlock();
7783 stubName = "decodeBlock";
7784
7785 if (!stubAddr) return false;
7786 Node* base64obj = argument(0);
7787 Node* src = argument(1);
7788 Node* src_offset = argument(2);
7789 Node* len = argument(3);
7790 Node* dest = argument(4);
7791 Node* dest_offset = argument(5);
7792 Node* isURL = argument(6);
7793 Node* isMIME = argument(7);
7794
7795 src = must_be_not_null(src, true);
7796 dest = must_be_not_null(dest, true);
7797
7798 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7799 assert(src_start, "source array is null");
7800 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7801 assert(dest_start, "destination array is null");
7802
7803 Node* call = make_runtime_call(RC_LEAF,
7804 OptoRuntime::base64_decodeBlock_Type(),
7805 stubAddr, stubName, TypePtr::BOTTOM,
7806 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
7807 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7808 set_result(result);
7809 return true;
7810 }
7811
7812 bool LibraryCallKit::inline_poly1305_processBlocks() {
7813 address stubAddr;
7814 const char *stubName;
7815 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
7816 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
7817 stubAddr = StubRoutines::poly1305_processBlocks();
7818 stubName = "poly1305_processBlocks";
7819
7820 if (!stubAddr) return false;
7821 null_check_receiver(); // null-check receiver
7822 if (stopped()) return true;
7823
7824 Node* input = argument(1);
7825 Node* input_offset = argument(2);
7826 Node* len = argument(3);
7827 Node* alimbs = argument(4);
7828 Node* rlimbs = argument(5);
7829
7830 input = must_be_not_null(input, true);
7831 alimbs = must_be_not_null(alimbs, true);
7832 rlimbs = must_be_not_null(rlimbs, true);
7833
7834 Node* input_start = array_element_address(input, input_offset, T_BYTE);
7835 assert(input_start, "input array is null");
7836 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
7837 assert(acc_start, "acc array is null");
7838 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
7839 assert(r_start, "r array is null");
7840
7841 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7842 OptoRuntime::poly1305_processBlocks_Type(),
7843 stubAddr, stubName, TypePtr::BOTTOM,
7844 input_start, len, acc_start, r_start);
7845 return true;
7846 }
7847
7848 //------------------------------inline_digestBase_implCompress-----------------------
7849 //
7850 // Calculate MD5 for single-block byte[] array.
7851 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
7852 //
7853 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
7854 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
7855 //
7856 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
7857 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
7858 //
7859 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
7860 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
7861 //
7862 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
7863 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
7864 //
7865 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
7866 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
7867
7868 Node* digestBase_obj = argument(0);
7869 Node* src = argument(1); // type oop
7870 Node* ofs = argument(2); // type int
7871
7872 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7873 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7874 // failed array check
7875 return false;
7876 }
7877 // Figure out the size and type of the elements we will be copying.
7878 BasicType src_elem = src_type->elem()->array_element_basic_type();
7879 if (src_elem != T_BYTE) {
7880 return false;
7881 }
7882 // 'src_start' points to src array + offset
7883 src = must_be_not_null(src, true);
7884 Node* src_start = array_element_address(src, ofs, src_elem);
7885 Node* state = nullptr;
7886 Node* block_size = nullptr;
7887 address stubAddr;
7888 const char *stubName;
7889
7890 switch(id) {
7891 case vmIntrinsics::_md5_implCompress:
7892 assert(UseMD5Intrinsics, "need MD5 instruction support");
7893 state = get_state_from_digest_object(digestBase_obj, T_INT);
7894 stubAddr = StubRoutines::md5_implCompress();
7895 stubName = "md5_implCompress";
7896 break;
7897 case vmIntrinsics::_sha_implCompress:
7898 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
7899 state = get_state_from_digest_object(digestBase_obj, T_INT);
7900 stubAddr = StubRoutines::sha1_implCompress();
7901 stubName = "sha1_implCompress";
7902 break;
7903 case vmIntrinsics::_sha2_implCompress:
7904 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
7905 state = get_state_from_digest_object(digestBase_obj, T_INT);
7906 stubAddr = StubRoutines::sha256_implCompress();
7907 stubName = "sha256_implCompress";
7908 break;
7909 case vmIntrinsics::_sha5_implCompress:
7910 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
7911 state = get_state_from_digest_object(digestBase_obj, T_LONG);
7912 stubAddr = StubRoutines::sha512_implCompress();
7913 stubName = "sha512_implCompress";
7914 break;
7915 case vmIntrinsics::_sha3_implCompress:
7916 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
7917 state = get_state_from_digest_object(digestBase_obj, T_BYTE);
7918 stubAddr = StubRoutines::sha3_implCompress();
7919 stubName = "sha3_implCompress";
7920 block_size = get_block_size_from_digest_object(digestBase_obj);
7921 if (block_size == nullptr) return false;
7922 break;
7923 default:
7924 fatal_unexpected_iid(id);
7925 return false;
7926 }
7927 if (state == nullptr) return false;
7928
7929 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
7930 if (stubAddr == nullptr) return false;
7931
7932 // Call the stub.
7933 Node* call;
7934 if (block_size == nullptr) {
7935 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
7936 stubAddr, stubName, TypePtr::BOTTOM,
7937 src_start, state);
7938 } else {
7939 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
7940 stubAddr, stubName, TypePtr::BOTTOM,
7941 src_start, state, block_size);
7942 }
7943
7944 return true;
7945 }
7946
7947 //------------------------------inline_digestBase_implCompressMB-----------------------
7948 //
7949 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
7950 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
7951 //
7952 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
7953 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7954 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7955 assert((uint)predicate < 5, "sanity");
7956 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
7957
7958 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
7959 Node* src = argument(1); // byte[] array
7960 Node* ofs = argument(2); // type int
7961 Node* limit = argument(3); // type int
7962
7963 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7964 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7965 // failed array check
7966 return false;
7967 }
7968 // Figure out the size and type of the elements we will be copying.
7969 BasicType src_elem = src_type->elem()->array_element_basic_type();
7970 if (src_elem != T_BYTE) {
7971 return false;
7972 }
7973 // 'src_start' points to src array + offset
7974 src = must_be_not_null(src, false);
7975 Node* src_start = array_element_address(src, ofs, src_elem);
7976
7977 const char* klass_digestBase_name = nullptr;
7978 const char* stub_name = nullptr;
7979 address stub_addr = nullptr;
7980 BasicType elem_type = T_INT;
7981
7982 switch (predicate) {
7983 case 0:
7984 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
7985 klass_digestBase_name = "sun/security/provider/MD5";
7986 stub_name = "md5_implCompressMB";
7987 stub_addr = StubRoutines::md5_implCompressMB();
7988 }
7989 break;
7990 case 1:
7991 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
7992 klass_digestBase_name = "sun/security/provider/SHA";
7993 stub_name = "sha1_implCompressMB";
7994 stub_addr = StubRoutines::sha1_implCompressMB();
7995 }
7996 break;
7997 case 2:
7998 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
7999 klass_digestBase_name = "sun/security/provider/SHA2";
8000 stub_name = "sha256_implCompressMB";
8001 stub_addr = StubRoutines::sha256_implCompressMB();
8002 }
8003 break;
8004 case 3:
8005 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
8006 klass_digestBase_name = "sun/security/provider/SHA5";
8007 stub_name = "sha512_implCompressMB";
8008 stub_addr = StubRoutines::sha512_implCompressMB();
8009 elem_type = T_LONG;
8010 }
8011 break;
8012 case 4:
8013 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
8014 klass_digestBase_name = "sun/security/provider/SHA3";
8015 stub_name = "sha3_implCompressMB";
8016 stub_addr = StubRoutines::sha3_implCompressMB();
8017 elem_type = T_BYTE;
8018 }
8019 break;
8020 default:
8021 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
8022 }
8023 if (klass_digestBase_name != nullptr) {
8024 assert(stub_addr != nullptr, "Stub is generated");
8025 if (stub_addr == nullptr) return false;
8026
8027 // get DigestBase klass to lookup for SHA klass
8028 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
8029 assert(tinst != nullptr, "digestBase_obj is not instance???");
8030 assert(tinst->is_loaded(), "DigestBase is not loaded");
8031
8032 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
8033 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
8034 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
8035 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
8036 }
8037 return false;
8038 }
8039
8040 //------------------------------inline_digestBase_implCompressMB-----------------------
8041 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
8042 BasicType elem_type, address stubAddr, const char *stubName,
8043 Node* src_start, Node* ofs, Node* limit) {
8044 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
8045 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8046 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
8047 digest_obj = _gvn.transform(digest_obj);
8048
8049 Node* state = get_state_from_digest_object(digest_obj, elem_type);
8050 if (state == nullptr) return false;
8051
8052 Node* block_size = nullptr;
8053 if (strcmp("sha3_implCompressMB", stubName) == 0) {
8054 block_size = get_block_size_from_digest_object(digest_obj);
8055 if (block_size == nullptr) return false;
8056 }
8057
8058 // Call the stub.
8059 Node* call;
8060 if (block_size == nullptr) {
8061 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8062 OptoRuntime::digestBase_implCompressMB_Type(false),
8063 stubAddr, stubName, TypePtr::BOTTOM,
8064 src_start, state, ofs, limit);
8065 } else {
8066 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8067 OptoRuntime::digestBase_implCompressMB_Type(true),
8068 stubAddr, stubName, TypePtr::BOTTOM,
8069 src_start, state, block_size, ofs, limit);
8070 }
8071
8072 // return ofs (int)
8073 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8074 set_result(result);
8075
8076 return true;
8077 }
8078
8079 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8080 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8081 assert(UseAES, "need AES instruction support");
8082 address stubAddr = nullptr;
8083 const char *stubName = nullptr;
8084 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8085 stubName = "galoisCounterMode_AESCrypt";
8086
8087 if (stubAddr == nullptr) return false;
8088
8089 Node* in = argument(0);
8090 Node* inOfs = argument(1);
8091 Node* len = argument(2);
8092 Node* ct = argument(3);
8093 Node* ctOfs = argument(4);
8094 Node* out = argument(5);
8095 Node* outOfs = argument(6);
8096 Node* gctr_object = argument(7);
8097 Node* ghash_object = argument(8);
8098
8099 // (1) in, ct and out are arrays.
8100 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8101 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8102 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8103 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8104 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8105 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8106
8107 // checks are the responsibility of the caller
8108 Node* in_start = in;
8109 Node* ct_start = ct;
8110 Node* out_start = out;
8111 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8112 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8113 in_start = array_element_address(in, inOfs, T_BYTE);
8114 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8115 out_start = array_element_address(out, outOfs, T_BYTE);
8116 }
8117
8118 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8119 // (because of the predicated logic executed earlier).
8120 // so we cast it here safely.
8121 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8122 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8123 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8124 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8125 Node* state = load_field_from_object(ghash_object, "state", "[J");
8126
8127 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8128 return false;
8129 }
8130 // cast it to what we know it will be at runtime
8131 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8132 assert(tinst != nullptr, "GCTR obj is null");
8133 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8134 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8135 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8136 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8137 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8138 const TypeOopPtr* xtype = aklass->as_instance_type();
8139 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8140 aescrypt_object = _gvn.transform(aescrypt_object);
8141 // we need to get the start of the aescrypt_object's expanded key array
8142 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8143 if (k_start == nullptr) return false;
8144 // similarly, get the start address of the r vector
8145 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8146 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8147 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8148
8149
8150 // Call the stub, passing params
8151 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8152 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8153 stubAddr, stubName, TypePtr::BOTTOM,
8154 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8155
8156 // return cipher length (int)
8157 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8158 set_result(retvalue);
8159
8160 return true;
8161 }
8162
8163 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8164 // Return node representing slow path of predicate check.
8165 // the pseudo code we want to emulate with this predicate is:
8166 // for encryption:
8167 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8168 // for decryption:
8169 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8170 // note cipher==plain is more conservative than the original java code but that's OK
8171 //
8172
8173 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8174 // The receiver was checked for null already.
8175 Node* objGCTR = argument(7);
8176 // Load embeddedCipher field of GCTR object.
8177 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8178 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8179
8180 // get AESCrypt klass for instanceOf check
8181 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8182 // will have same classloader as CipherBlockChaining object
8183 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8184 assert(tinst != nullptr, "GCTR obj is null");
8185 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8186
8187 // we want to do an instanceof comparison against the AESCrypt class
8188 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8189 if (!klass_AESCrypt->is_loaded()) {
8190 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8191 Node* ctrl = control();
8192 set_control(top()); // no regular fast path
8193 return ctrl;
8194 }
8195
8196 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8197 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8198 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8199 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8200 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8201
8202 return instof_false; // even if it is null
8203 }
8204
8205 //------------------------------get_state_from_digest_object-----------------------
8206 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8207 const char* state_type;
8208 switch (elem_type) {
8209 case T_BYTE: state_type = "[B"; break;
8210 case T_INT: state_type = "[I"; break;
8211 case T_LONG: state_type = "[J"; break;
8212 default: ShouldNotReachHere();
8213 }
8214 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8215 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8216 if (digest_state == nullptr) return (Node *) nullptr;
8217
8218 // now have the array, need to get the start address of the state array
8219 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8220 return state;
8221 }
8222
8223 //------------------------------get_block_size_from_sha3_object----------------------------------
8224 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8225 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8226 assert (block_size != nullptr, "sanity");
8227 return block_size;
8228 }
8229
8230 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8231 // Return node representing slow path of predicate check.
8232 // the pseudo code we want to emulate with this predicate is:
8233 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8234 //
8235 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8236 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8237 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8238 assert((uint)predicate < 5, "sanity");
8239
8240 // The receiver was checked for null already.
8241 Node* digestBaseObj = argument(0);
8242
8243 // get DigestBase klass for instanceOf check
8244 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8245 assert(tinst != nullptr, "digestBaseObj is null");
8246 assert(tinst->is_loaded(), "DigestBase is not loaded");
8247
8248 const char* klass_name = nullptr;
8249 switch (predicate) {
8250 case 0:
8251 if (UseMD5Intrinsics) {
8252 // we want to do an instanceof comparison against the MD5 class
8253 klass_name = "sun/security/provider/MD5";
8254 }
8255 break;
8256 case 1:
8257 if (UseSHA1Intrinsics) {
8258 // we want to do an instanceof comparison against the SHA class
8259 klass_name = "sun/security/provider/SHA";
8260 }
8261 break;
8262 case 2:
8263 if (UseSHA256Intrinsics) {
8264 // we want to do an instanceof comparison against the SHA2 class
8265 klass_name = "sun/security/provider/SHA2";
8266 }
8267 break;
8268 case 3:
8269 if (UseSHA512Intrinsics) {
8270 // we want to do an instanceof comparison against the SHA5 class
8271 klass_name = "sun/security/provider/SHA5";
8272 }
8273 break;
8274 case 4:
8275 if (UseSHA3Intrinsics) {
8276 // we want to do an instanceof comparison against the SHA3 class
8277 klass_name = "sun/security/provider/SHA3";
8278 }
8279 break;
8280 default:
8281 fatal("unknown SHA intrinsic predicate: %d", predicate);
8282 }
8283
8284 ciKlass* klass = nullptr;
8285 if (klass_name != nullptr) {
8286 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8287 }
8288 if ((klass == nullptr) || !klass->is_loaded()) {
8289 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8290 Node* ctrl = control();
8291 set_control(top()); // no intrinsic path
8292 return ctrl;
8293 }
8294 ciInstanceKlass* instklass = klass->as_instance_klass();
8295
8296 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8297 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8298 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8299 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8300
8301 return instof_false; // even if it is null
8302 }
8303
8304 //-------------inline_fma-----------------------------------
8305 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8306 Node *a = nullptr;
8307 Node *b = nullptr;
8308 Node *c = nullptr;
8309 Node* result = nullptr;
8310 switch (id) {
8311 case vmIntrinsics::_fmaD:
8312 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8313 // no receiver since it is static method
8314 a = round_double_node(argument(0));
8315 b = round_double_node(argument(2));
8316 c = round_double_node(argument(4));
8317 result = _gvn.transform(new FmaDNode(control(), a, b, c));
8318 break;
8319 case vmIntrinsics::_fmaF:
8320 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8321 a = argument(0);
8322 b = argument(1);
8323 c = argument(2);
8324 result = _gvn.transform(new FmaFNode(control(), a, b, c));
8325 break;
8326 default:
8327 fatal_unexpected_iid(id); break;
8328 }
8329 set_result(result);
8330 return true;
8331 }
8332
8333 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8334 // argument(0) is receiver
8335 Node* codePoint = argument(1);
8336 Node* n = nullptr;
8337
8338 switch (id) {
8339 case vmIntrinsics::_isDigit :
8340 n = new DigitNode(control(), codePoint);
8341 break;
8342 case vmIntrinsics::_isLowerCase :
8343 n = new LowerCaseNode(control(), codePoint);
8344 break;
8345 case vmIntrinsics::_isUpperCase :
8346 n = new UpperCaseNode(control(), codePoint);
8347 break;
8348 case vmIntrinsics::_isWhitespace :
8349 n = new WhitespaceNode(control(), codePoint);
8350 break;
8351 default:
8352 fatal_unexpected_iid(id);
8353 }
8354
8355 set_result(_gvn.transform(n));
8356 return true;
8357 }
8358
8359 //------------------------------inline_fp_min_max------------------------------
8360 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
8361 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
8362
8363 // The intrinsic should be used only when the API branches aren't predictable,
8364 // the last one performing the most important comparison. The following heuristic
8365 // uses the branch statistics to eventually bail out if necessary.
8366
8367 ciMethodData *md = callee()->method_data();
8368
8369 if ( md != nullptr && md->is_mature() && md->invocation_count() > 0 ) {
8370 ciCallProfile cp = caller()->call_profile_at_bci(bci());
8371
8372 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
8373 // Bail out if the call-site didn't contribute enough to the statistics.
8374 return false;
8375 }
8376
8377 uint taken = 0, not_taken = 0;
8378
8379 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
8380 if (p->is_BranchData()) {
8381 taken = ((ciBranchData*)p)->taken();
8382 not_taken = ((ciBranchData*)p)->not_taken();
8383 }
8384 }
8385
8386 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
8387 balance = balance < 0 ? -balance : balance;
8388 if ( balance > 0.2 ) {
8389 // Bail out if the most important branch is predictable enough.
8390 return false;
8391 }
8392 }
8393 */
8394
8395 Node *a = nullptr;
8396 Node *b = nullptr;
8397 Node *n = nullptr;
8398 switch (id) {
8399 case vmIntrinsics::_maxF:
8400 case vmIntrinsics::_minF:
8401 case vmIntrinsics::_maxF_strict:
8402 case vmIntrinsics::_minF_strict:
8403 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
8404 a = argument(0);
8405 b = argument(1);
8406 break;
8407 case vmIntrinsics::_maxD:
8408 case vmIntrinsics::_minD:
8409 case vmIntrinsics::_maxD_strict:
8410 case vmIntrinsics::_minD_strict:
8411 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
8412 a = round_double_node(argument(0));
8413 b = round_double_node(argument(2));
8414 break;
8415 default:
8416 fatal_unexpected_iid(id);
8417 break;
8418 }
8419 switch (id) {
8420 case vmIntrinsics::_maxF:
8421 case vmIntrinsics::_maxF_strict:
8422 n = new MaxFNode(a, b);
8423 break;
8424 case vmIntrinsics::_minF:
8425 case vmIntrinsics::_minF_strict:
8426 n = new MinFNode(a, b);
8427 break;
8428 case vmIntrinsics::_maxD:
8429 case vmIntrinsics::_maxD_strict:
8430 n = new MaxDNode(a, b);
8431 break;
8432 case vmIntrinsics::_minD:
8433 case vmIntrinsics::_minD_strict:
8434 n = new MinDNode(a, b);
8435 break;
8436 default:
8437 fatal_unexpected_iid(id);
8438 break;
8439 }
8440 set_result(_gvn.transform(n));
8441 return true;
8442 }
8443
8444 bool LibraryCallKit::inline_profileBoolean() {
8445 Node* counts = argument(1);
8446 const TypeAryPtr* ary = nullptr;
8447 ciArray* aobj = nullptr;
8448 if (counts->is_Con()
8449 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8450 && (aobj = ary->const_oop()->as_array()) != nullptr
8451 && (aobj->length() == 2)) {
8452 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8453 jint false_cnt = aobj->element_value(0).as_int();
8454 jint true_cnt = aobj->element_value(1).as_int();
8455
8456 if (C->log() != nullptr) {
8457 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8458 false_cnt, true_cnt);
8459 }
8460
8461 if (false_cnt + true_cnt == 0) {
8462 // According to profile, never executed.
8463 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8464 Deoptimization::Action_reinterpret);
8465 return true;
8466 }
8467
8468 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8469 // is a number of each value occurrences.
8470 Node* result = argument(0);
8471 if (false_cnt == 0 || true_cnt == 0) {
8472 // According to profile, one value has been never seen.
8473 int expected_val = (false_cnt == 0) ? 1 : 0;
8474
8475 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8476 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8477
8478 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8479 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8480 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8481
8482 { // Slow path: uncommon trap for never seen value and then reexecute
8483 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8484 // the value has been seen at least once.
8485 PreserveJVMState pjvms(this);
8486 PreserveReexecuteState preexecs(this);
8487 jvms()->set_should_reexecute(true);
8488
8489 set_control(slow_path);
8490 set_i_o(i_o());
8491
8492 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8493 Deoptimization::Action_reinterpret);
8494 }
8495 // The guard for never seen value enables sharpening of the result and
8496 // returning a constant. It allows to eliminate branches on the same value
8497 // later on.
8498 set_control(fast_path);
8499 result = intcon(expected_val);
8500 }
8501 // Stop profiling.
8502 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8503 // By replacing method body with profile data (represented as ProfileBooleanNode
8504 // on IR level) we effectively disable profiling.
8505 // It enables full speed execution once optimized code is generated.
8506 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8507 C->record_for_igvn(profile);
8508 set_result(profile);
8509 return true;
8510 } else {
8511 // Continue profiling.
8512 // Profile data isn't available at the moment. So, execute method's bytecode version.
8513 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8514 // is compiled and counters aren't available since corresponding MethodHandle
8515 // isn't a compile-time constant.
8516 return false;
8517 }
8518 }
8519
8520 bool LibraryCallKit::inline_isCompileConstant() {
8521 Node* n = argument(0);
8522 set_result(n->is_Con() ? intcon(1) : intcon(0));
8523 return true;
8524 }
8525
8526 //------------------------------- inline_getObjectSize --------------------------------------
8527 //
8528 // Calculate the runtime size of the object/array.
8529 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8530 //
8531 bool LibraryCallKit::inline_getObjectSize() {
8532 Node* obj = argument(3);
8533 Node* klass_node = load_object_klass(obj);
8534
8535 jint layout_con = Klass::_lh_neutral_value;
8536 Node* layout_val = get_layout_helper(klass_node, layout_con);
8537 int layout_is_con = (layout_val == nullptr);
8538
8539 if (layout_is_con) {
8540 // Layout helper is constant, can figure out things at compile time.
8541
8542 if (Klass::layout_helper_is_instance(layout_con)) {
8543 // Instance case: layout_con contains the size itself.
8544 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8545 set_result(size);
8546 } else {
8547 // Array case: size is round(header + element_size*arraylength).
8548 // Since arraylength is different for every array instance, we have to
8549 // compute the whole thing at runtime.
8550
8551 Node* arr_length = load_array_length(obj);
8552
8553 int round_mask = MinObjAlignmentInBytes - 1;
8554 int hsize = Klass::layout_helper_header_size(layout_con);
8555 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8556
8557 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8558 round_mask = 0; // strength-reduce it if it goes away completely
8559 }
8560 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8561 Node* header_size = intcon(hsize + round_mask);
8562
8563 Node* lengthx = ConvI2X(arr_length);
8564 Node* headerx = ConvI2X(header_size);
8565
8566 Node* abody = lengthx;
8567 if (eshift != 0) {
8568 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8569 }
8570 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8571 if (round_mask != 0) {
8572 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8573 }
8574 size = ConvX2L(size);
8575 set_result(size);
8576 }
8577 } else {
8578 // Layout helper is not constant, need to test for array-ness at runtime.
8579
8580 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8581 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8582 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8583 record_for_igvn(result_reg);
8584
8585 Node* array_ctl = generate_array_guard(klass_node, nullptr);
8586 if (array_ctl != nullptr) {
8587 // Array case: size is round(header + element_size*arraylength).
8588 // Since arraylength is different for every array instance, we have to
8589 // compute the whole thing at runtime.
8590
8591 PreserveJVMState pjvms(this);
8592 set_control(array_ctl);
8593 Node* arr_length = load_array_length(obj);
8594
8595 int round_mask = MinObjAlignmentInBytes - 1;
8596 Node* mask = intcon(round_mask);
8597
8598 Node* hss = intcon(Klass::_lh_header_size_shift);
8599 Node* hsm = intcon(Klass::_lh_header_size_mask);
8600 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8601 header_size = _gvn.transform(new AndINode(header_size, hsm));
8602 header_size = _gvn.transform(new AddINode(header_size, mask));
8603
8604 // There is no need to mask or shift this value.
8605 // The semantics of LShiftINode include an implicit mask to 0x1F.
8606 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8607 Node* elem_shift = layout_val;
8608
8609 Node* lengthx = ConvI2X(arr_length);
8610 Node* headerx = ConvI2X(header_size);
8611
8612 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
8613 Node* size = _gvn.transform(new AddXNode(headerx, abody));
8614 if (round_mask != 0) {
8615 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
8616 }
8617 size = ConvX2L(size);
8618
8619 result_reg->init_req(_array_path, control());
8620 result_val->init_req(_array_path, size);
8621 }
8622
8623 if (!stopped()) {
8624 // Instance case: the layout helper gives us instance size almost directly,
8625 // but we need to mask out the _lh_instance_slow_path_bit.
8626 Node* size = ConvI2X(layout_val);
8627 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
8628 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
8629 size = _gvn.transform(new AndXNode(size, mask));
8630 size = ConvX2L(size);
8631
8632 result_reg->init_req(_instance_path, control());
8633 result_val->init_req(_instance_path, size);
8634 }
8635
8636 set_result(result_reg, result_val);
8637 }
8638
8639 return true;
8640 }
8641
8642 //------------------------------- inline_blackhole --------------------------------------
8643 //
8644 // Make sure all arguments to this node are alive.
8645 // This matches methods that were requested to be blackholed through compile commands.
8646 //
8647 bool LibraryCallKit::inline_blackhole() {
8648 assert(callee()->is_static(), "Should have been checked before: only static methods here");
8649 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
8650 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
8651
8652 // Blackhole node pinches only the control, not memory. This allows
8653 // the blackhole to be pinned in the loop that computes blackholed
8654 // values, but have no other side effects, like breaking the optimizations
8655 // across the blackhole.
8656
8657 Node* bh = _gvn.transform(new BlackholeNode(control()));
8658 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
8659
8660 // Bind call arguments as blackhole arguments to keep them alive
8661 uint nargs = callee()->arg_size();
8662 for (uint i = 0; i < nargs; i++) {
8663 bh->add_req(argument(i));
8664 }
8665
8666 return true;
8667 }