1 /*
2 * Copyright (c) 1999, 2024, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "ci/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::_dtanh:
259 case vmIntrinsics::_dabs:
260 case vmIntrinsics::_fabs:
261 case vmIntrinsics::_iabs:
262 case vmIntrinsics::_labs:
263 case vmIntrinsics::_datan2:
264 case vmIntrinsics::_dsqrt:
265 case vmIntrinsics::_dsqrt_strict:
266 case vmIntrinsics::_dexp:
267 case vmIntrinsics::_dlog:
268 case vmIntrinsics::_dlog10:
269 case vmIntrinsics::_dpow:
270 case vmIntrinsics::_dcopySign:
271 case vmIntrinsics::_fcopySign:
272 case vmIntrinsics::_dsignum:
273 case vmIntrinsics::_roundF:
274 case vmIntrinsics::_roundD:
275 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
276
277 case vmIntrinsics::_notify:
278 case vmIntrinsics::_notifyAll:
279 return inline_notify(intrinsic_id());
280
281 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
282 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
283 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
284 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
285 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
286 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
287 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
288 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
289 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
290 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
291 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
292 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
293 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
294 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
295
296 case vmIntrinsics::_arraycopy: return inline_arraycopy();
297
298 case vmIntrinsics::_arraySort: return inline_array_sort();
299 case vmIntrinsics::_arrayPartition: return inline_array_partition();
300
301 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
302 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
303 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
304 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
305
306 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
307 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
308 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
309 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
310 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
311 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
312 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
313 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
314
315 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
316
317 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
318
319 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
320 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
321 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
322 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
323
324 case vmIntrinsics::_compressStringC:
325 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
326 case vmIntrinsics::_inflateStringC:
327 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
328
329 case vmIntrinsics::_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 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
492 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
493
494 #if INCLUDE_JVMTI
495 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
496 "notifyJvmtiStart", true, false);
497 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
498 "notifyJvmtiEnd", false, true);
499 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
500 "notifyJvmtiMount", false, false);
501 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
502 "notifyJvmtiUnmount", false, false);
503 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
504 #endif
505
506 #ifdef JFR_HAVE_INTRINSICS
507 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
508 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
509 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
510 #endif
511 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
512 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
513 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
514 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
515 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
516 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
517 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
518 case vmIntrinsics::_isFlatArray: return inline_unsafe_isFlatArray();
519 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
520 case vmIntrinsics::_getLength: return inline_native_getLength();
521 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
522 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
523 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
524 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
525 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
526 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
527 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
528
529 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
530 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
531 case vmIntrinsics::_newNullRestrictedArray: return inline_newNullRestrictedArray();
532
533 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
534
535 case vmIntrinsics::_isInstance:
536 case vmIntrinsics::_getModifiers:
537 case vmIntrinsics::_isInterface:
538 case vmIntrinsics::_isArray:
539 case vmIntrinsics::_isPrimitive:
540 case vmIntrinsics::_isHidden:
541 case vmIntrinsics::_getSuperclass:
542 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
543
544 case vmIntrinsics::_floatToRawIntBits:
545 case vmIntrinsics::_floatToIntBits:
546 case vmIntrinsics::_intBitsToFloat:
547 case vmIntrinsics::_doubleToRawLongBits:
548 case vmIntrinsics::_doubleToLongBits:
549 case vmIntrinsics::_longBitsToDouble:
550 case vmIntrinsics::_floatToFloat16:
551 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
552
553 case vmIntrinsics::_floatIsFinite:
554 case vmIntrinsics::_floatIsInfinite:
555 case vmIntrinsics::_doubleIsFinite:
556 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
557
558 case vmIntrinsics::_numberOfLeadingZeros_i:
559 case vmIntrinsics::_numberOfLeadingZeros_l:
560 case vmIntrinsics::_numberOfTrailingZeros_i:
561 case vmIntrinsics::_numberOfTrailingZeros_l:
562 case vmIntrinsics::_bitCount_i:
563 case vmIntrinsics::_bitCount_l:
564 case vmIntrinsics::_reverse_i:
565 case vmIntrinsics::_reverse_l:
566 case vmIntrinsics::_reverseBytes_i:
567 case vmIntrinsics::_reverseBytes_l:
568 case vmIntrinsics::_reverseBytes_s:
569 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
570
571 case vmIntrinsics::_compress_i:
572 case vmIntrinsics::_compress_l:
573 case vmIntrinsics::_expand_i:
574 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
575
576 case vmIntrinsics::_compareUnsigned_i:
577 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
578
579 case vmIntrinsics::_divideUnsigned_i:
580 case vmIntrinsics::_divideUnsigned_l:
581 case vmIntrinsics::_remainderUnsigned_i:
582 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
583
584 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
585
586 case vmIntrinsics::_Reference_get: return inline_reference_get();
587 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
588 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
589 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
590 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
591
592 case vmIntrinsics::_Class_cast: return inline_Class_cast();
593
594 case vmIntrinsics::_aescrypt_encryptBlock:
595 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
596
597 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
598 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
599 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
600
601 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
602 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
603 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
604
605 case vmIntrinsics::_counterMode_AESCrypt:
606 return inline_counterMode_AESCrypt(intrinsic_id());
607
608 case vmIntrinsics::_galoisCounterMode_AESCrypt:
609 return inline_galoisCounterMode_AESCrypt();
610
611 case vmIntrinsics::_md5_implCompress:
612 case vmIntrinsics::_sha_implCompress:
613 case vmIntrinsics::_sha2_implCompress:
614 case vmIntrinsics::_sha5_implCompress:
615 case vmIntrinsics::_sha3_implCompress:
616 return inline_digestBase_implCompress(intrinsic_id());
617
618 case vmIntrinsics::_digestBase_implCompressMB:
619 return inline_digestBase_implCompressMB(predicate);
620
621 case vmIntrinsics::_multiplyToLen:
622 return inline_multiplyToLen();
623
624 case vmIntrinsics::_squareToLen:
625 return inline_squareToLen();
626
627 case vmIntrinsics::_mulAdd:
628 return inline_mulAdd();
629
630 case vmIntrinsics::_montgomeryMultiply:
631 return inline_montgomeryMultiply();
632 case vmIntrinsics::_montgomerySquare:
633 return inline_montgomerySquare();
634
635 case vmIntrinsics::_bigIntegerRightShiftWorker:
636 return inline_bigIntegerShift(true);
637 case vmIntrinsics::_bigIntegerLeftShiftWorker:
638 return inline_bigIntegerShift(false);
639
640 case vmIntrinsics::_vectorizedMismatch:
641 return inline_vectorizedMismatch();
642
643 case vmIntrinsics::_ghash_processBlocks:
644 return inline_ghash_processBlocks();
645 case vmIntrinsics::_chacha20Block:
646 return inline_chacha20Block();
647 case vmIntrinsics::_base64_encodeBlock:
648 return inline_base64_encodeBlock();
649 case vmIntrinsics::_base64_decodeBlock:
650 return inline_base64_decodeBlock();
651 case vmIntrinsics::_poly1305_processBlocks:
652 return inline_poly1305_processBlocks();
653 case vmIntrinsics::_intpoly_montgomeryMult_P256:
654 return inline_intpoly_montgomeryMult_P256();
655 case vmIntrinsics::_intpoly_assign:
656 return inline_intpoly_assign();
657 case vmIntrinsics::_encodeISOArray:
658 case vmIntrinsics::_encodeByteISOArray:
659 return inline_encodeISOArray(false);
660 case vmIntrinsics::_encodeAsciiArray:
661 return inline_encodeISOArray(true);
662
663 case vmIntrinsics::_updateCRC32:
664 return inline_updateCRC32();
665 case vmIntrinsics::_updateBytesCRC32:
666 return inline_updateBytesCRC32();
667 case vmIntrinsics::_updateByteBufferCRC32:
668 return inline_updateByteBufferCRC32();
669
670 case vmIntrinsics::_updateBytesCRC32C:
671 return inline_updateBytesCRC32C();
672 case vmIntrinsics::_updateDirectByteBufferCRC32C:
673 return inline_updateDirectByteBufferCRC32C();
674
675 case vmIntrinsics::_updateBytesAdler32:
676 return inline_updateBytesAdler32();
677 case vmIntrinsics::_updateByteBufferAdler32:
678 return inline_updateByteBufferAdler32();
679
680 case vmIntrinsics::_profileBoolean:
681 return inline_profileBoolean();
682 case vmIntrinsics::_isCompileConstant:
683 return inline_isCompileConstant();
684
685 case vmIntrinsics::_countPositives:
686 return inline_countPositives();
687
688 case vmIntrinsics::_fmaD:
689 case vmIntrinsics::_fmaF:
690 return inline_fma(intrinsic_id());
691
692 case vmIntrinsics::_isDigit:
693 case vmIntrinsics::_isLowerCase:
694 case vmIntrinsics::_isUpperCase:
695 case vmIntrinsics::_isWhitespace:
696 return inline_character_compare(intrinsic_id());
697
698 case vmIntrinsics::_min:
699 case vmIntrinsics::_max:
700 case vmIntrinsics::_min_strict:
701 case vmIntrinsics::_max_strict:
702 return inline_min_max(intrinsic_id());
703
704 case vmIntrinsics::_maxF:
705 case vmIntrinsics::_minF:
706 case vmIntrinsics::_maxD:
707 case vmIntrinsics::_minD:
708 case vmIntrinsics::_maxF_strict:
709 case vmIntrinsics::_minF_strict:
710 case vmIntrinsics::_maxD_strict:
711 case vmIntrinsics::_minD_strict:
712 return inline_fp_min_max(intrinsic_id());
713
714 case vmIntrinsics::_VectorUnaryOp:
715 return inline_vector_nary_operation(1);
716 case vmIntrinsics::_VectorBinaryOp:
717 return inline_vector_nary_operation(2);
718 case vmIntrinsics::_VectorTernaryOp:
719 return inline_vector_nary_operation(3);
720 case vmIntrinsics::_VectorFromBitsCoerced:
721 return inline_vector_frombits_coerced();
722 case vmIntrinsics::_VectorShuffleIota:
723 return inline_vector_shuffle_iota();
724 case vmIntrinsics::_VectorMaskOp:
725 return inline_vector_mask_operation();
726 case vmIntrinsics::_VectorShuffleToVector:
727 return inline_vector_shuffle_to_vector();
728 case vmIntrinsics::_VectorWrapShuffleIndexes:
729 return inline_vector_wrap_shuffle_indexes();
730 case vmIntrinsics::_VectorLoadOp:
731 return inline_vector_mem_operation(/*is_store=*/false);
732 case vmIntrinsics::_VectorLoadMaskedOp:
733 return inline_vector_mem_masked_operation(/*is_store*/false);
734 case vmIntrinsics::_VectorStoreOp:
735 return inline_vector_mem_operation(/*is_store=*/true);
736 case vmIntrinsics::_VectorStoreMaskedOp:
737 return inline_vector_mem_masked_operation(/*is_store=*/true);
738 case vmIntrinsics::_VectorGatherOp:
739 return inline_vector_gather_scatter(/*is_scatter*/ false);
740 case vmIntrinsics::_VectorScatterOp:
741 return inline_vector_gather_scatter(/*is_scatter*/ true);
742 case vmIntrinsics::_VectorReductionCoerced:
743 return inline_vector_reduction();
744 case vmIntrinsics::_VectorTest:
745 return inline_vector_test();
746 case vmIntrinsics::_VectorBlend:
747 return inline_vector_blend();
748 case vmIntrinsics::_VectorRearrange:
749 return inline_vector_rearrange();
750 case vmIntrinsics::_VectorSelectFrom:
751 return inline_vector_select_from();
752 case vmIntrinsics::_VectorCompare:
753 return inline_vector_compare();
754 case vmIntrinsics::_VectorBroadcastInt:
755 return inline_vector_broadcast_int();
756 case vmIntrinsics::_VectorConvert:
757 return inline_vector_convert();
758 case vmIntrinsics::_VectorInsert:
759 return inline_vector_insert();
760 case vmIntrinsics::_VectorExtract:
761 return inline_vector_extract();
762 case vmIntrinsics::_VectorCompressExpand:
763 return inline_vector_compress_expand();
764 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
765 return inline_vector_select_from_two_vectors();
766 case vmIntrinsics::_IndexVector:
767 return inline_index_vector();
768 case vmIntrinsics::_IndexPartiallyInUpperRange:
769 return inline_index_partially_in_upper_range();
770
771 case vmIntrinsics::_getObjectSize:
772 return inline_getObjectSize();
773
774 case vmIntrinsics::_blackhole:
775 return inline_blackhole();
776
777 default:
778 // If you get here, it may be that someone has added a new intrinsic
779 // to the list in vmIntrinsics.hpp without implementing it here.
780 #ifndef PRODUCT
781 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
782 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
783 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
784 }
785 #endif
786 return false;
787 }
788 }
789
790 Node* LibraryCallKit::try_to_predicate(int predicate) {
791 if (!jvms()->has_method()) {
792 // Root JVMState has a null method.
793 assert(map()->memory()->Opcode() == Op_Parm, "");
794 // Insert the memory aliasing node
795 set_all_memory(reset_memory());
796 }
797 assert(merged_memory(), "");
798
799 switch (intrinsic_id()) {
800 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
801 return inline_cipherBlockChaining_AESCrypt_predicate(false);
802 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
803 return inline_cipherBlockChaining_AESCrypt_predicate(true);
804 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
805 return inline_electronicCodeBook_AESCrypt_predicate(false);
806 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
807 return inline_electronicCodeBook_AESCrypt_predicate(true);
808 case vmIntrinsics::_counterMode_AESCrypt:
809 return inline_counterMode_AESCrypt_predicate();
810 case vmIntrinsics::_digestBase_implCompressMB:
811 return inline_digestBase_implCompressMB_predicate(predicate);
812 case vmIntrinsics::_galoisCounterMode_AESCrypt:
813 return inline_galoisCounterMode_AESCrypt_predicate();
814
815 default:
816 // If you get here, it may be that someone has added a new intrinsic
817 // to the list in vmIntrinsics.hpp without implementing it here.
818 #ifndef PRODUCT
819 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
820 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
821 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
822 }
823 #endif
824 Node* slow_ctl = control();
825 set_control(top()); // No fast path intrinsic
826 return slow_ctl;
827 }
828 }
829
830 //------------------------------set_result-------------------------------
831 // Helper function for finishing intrinsics.
832 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
833 record_for_igvn(region);
834 set_control(_gvn.transform(region));
835 set_result( _gvn.transform(value));
836 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
837 }
838
839 //------------------------------generate_guard---------------------------
840 // Helper function for generating guarded fast-slow graph structures.
841 // The given 'test', if true, guards a slow path. If the test fails
842 // then a fast path can be taken. (We generally hope it fails.)
843 // In all cases, GraphKit::control() is updated to the fast path.
844 // The returned value represents the control for the slow path.
845 // The return value is never 'top'; it is either a valid control
846 // or null if it is obvious that the slow path can never be taken.
847 // Also, if region and the slow control are not null, the slow edge
848 // is appended to the region.
849 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
850 if (stopped()) {
851 // Already short circuited.
852 return nullptr;
853 }
854
855 // Build an if node and its projections.
856 // If test is true we take the slow path, which we assume is uncommon.
857 if (_gvn.type(test) == TypeInt::ZERO) {
858 // The slow branch is never taken. No need to build this guard.
859 return nullptr;
860 }
861
862 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
863
864 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
865 if (if_slow == top()) {
866 // The slow branch is never taken. No need to build this guard.
867 return nullptr;
868 }
869
870 if (region != nullptr)
871 region->add_req(if_slow);
872
873 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
874 set_control(if_fast);
875
876 return if_slow;
877 }
878
879 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
880 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
881 }
882 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
883 return generate_guard(test, region, PROB_FAIR);
884 }
885
886 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
887 Node* *pos_index) {
888 if (stopped())
889 return nullptr; // already stopped
890 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
891 return nullptr; // index is already adequately typed
892 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
893 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
894 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
895 if (is_neg != nullptr && pos_index != nullptr) {
896 // Emulate effect of Parse::adjust_map_after_if.
897 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
898 (*pos_index) = _gvn.transform(ccast);
899 }
900 return is_neg;
901 }
902
903 // Make sure that 'position' is a valid limit index, in [0..length].
904 // There are two equivalent plans for checking this:
905 // A. (offset + copyLength) unsigned<= arrayLength
906 // B. offset <= (arrayLength - copyLength)
907 // We require that all of the values above, except for the sum and
908 // difference, are already known to be non-negative.
909 // Plan A is robust in the face of overflow, if offset and copyLength
910 // are both hugely positive.
911 //
912 // Plan B is less direct and intuitive, but it does not overflow at
913 // all, since the difference of two non-negatives is always
914 // representable. Whenever Java methods must perform the equivalent
915 // check they generally use Plan B instead of Plan A.
916 // For the moment we use Plan A.
917 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
918 Node* subseq_length,
919 Node* array_length,
920 RegionNode* region) {
921 if (stopped())
922 return nullptr; // already stopped
923 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
924 if (zero_offset && subseq_length->eqv_uncast(array_length))
925 return nullptr; // common case of whole-array copy
926 Node* last = subseq_length;
927 if (!zero_offset) // last += offset
928 last = _gvn.transform(new AddINode(last, offset));
929 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
930 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
931 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
932 return is_over;
933 }
934
935 // Emit range checks for the given String.value byte array
936 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
937 if (stopped()) {
938 return; // already stopped
939 }
940 RegionNode* bailout = new RegionNode(1);
941 record_for_igvn(bailout);
942 if (char_count) {
943 // Convert char count to byte count
944 count = _gvn.transform(new LShiftINode(count, intcon(1)));
945 }
946
947 // Offset and count must not be negative
948 generate_negative_guard(offset, bailout);
949 generate_negative_guard(count, bailout);
950 // Offset + count must not exceed length of array
951 generate_limit_guard(offset, count, load_array_length(array), bailout);
952
953 if (bailout->req() > 1) {
954 PreserveJVMState pjvms(this);
955 set_control(_gvn.transform(bailout));
956 uncommon_trap(Deoptimization::Reason_intrinsic,
957 Deoptimization::Action_maybe_recompile);
958 }
959 }
960
961 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
962 bool is_immutable) {
963 ciKlass* thread_klass = env()->Thread_klass();
964 const Type* thread_type
965 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
966
967 Node* thread = _gvn.transform(new ThreadLocalNode());
968 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
969 tls_output = thread;
970
971 Node* thread_obj_handle
972 = (is_immutable
973 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
974 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
975 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
976 thread_obj_handle = _gvn.transform(thread_obj_handle);
977
978 DecoratorSet decorators = IN_NATIVE;
979 if (is_immutable) {
980 decorators |= C2_IMMUTABLE_MEMORY;
981 }
982 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
983 }
984
985 //--------------------------generate_current_thread--------------------
986 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
987 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
988 /*is_immutable*/false);
989 }
990
991 //--------------------------generate_virtual_thread--------------------
992 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
993 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
994 !C->method()->changes_current_thread());
995 }
996
997 //------------------------------make_string_method_node------------------------
998 // Helper method for String intrinsic functions. This version is called with
999 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1000 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1001 // containing the lengths of str1 and str2.
1002 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1003 Node* result = nullptr;
1004 switch (opcode) {
1005 case Op_StrIndexOf:
1006 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1007 str1_start, cnt1, str2_start, cnt2, ae);
1008 break;
1009 case Op_StrComp:
1010 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1011 str1_start, cnt1, str2_start, cnt2, ae);
1012 break;
1013 case Op_StrEquals:
1014 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1015 // Use the constant length if there is one because optimized match rule may exist.
1016 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1017 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1018 break;
1019 default:
1020 ShouldNotReachHere();
1021 return nullptr;
1022 }
1023
1024 // All these intrinsics have checks.
1025 C->set_has_split_ifs(true); // Has chance for split-if optimization
1026 clear_upper_avx();
1027
1028 return _gvn.transform(result);
1029 }
1030
1031 //------------------------------inline_string_compareTo------------------------
1032 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1033 Node* arg1 = argument(0);
1034 Node* arg2 = argument(1);
1035
1036 arg1 = must_be_not_null(arg1, true);
1037 arg2 = must_be_not_null(arg2, true);
1038
1039 // Get start addr and length of first argument
1040 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1041 Node* arg1_cnt = load_array_length(arg1);
1042
1043 // Get start addr and length of second argument
1044 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1045 Node* arg2_cnt = load_array_length(arg2);
1046
1047 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1048 set_result(result);
1049 return true;
1050 }
1051
1052 //------------------------------inline_string_equals------------------------
1053 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1054 Node* arg1 = argument(0);
1055 Node* arg2 = argument(1);
1056
1057 // paths (plus control) merge
1058 RegionNode* region = new RegionNode(3);
1059 Node* phi = new PhiNode(region, TypeInt::BOOL);
1060
1061 if (!stopped()) {
1062
1063 arg1 = must_be_not_null(arg1, true);
1064 arg2 = must_be_not_null(arg2, true);
1065
1066 // Get start addr and length of first argument
1067 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1068 Node* arg1_cnt = load_array_length(arg1);
1069
1070 // Get start addr and length of second argument
1071 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1072 Node* arg2_cnt = load_array_length(arg2);
1073
1074 // Check for arg1_cnt != arg2_cnt
1075 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1076 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1077 Node* if_ne = generate_slow_guard(bol, nullptr);
1078 if (if_ne != nullptr) {
1079 phi->init_req(2, intcon(0));
1080 region->init_req(2, if_ne);
1081 }
1082
1083 // Check for count == 0 is done by assembler code for StrEquals.
1084
1085 if (!stopped()) {
1086 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1087 phi->init_req(1, equals);
1088 region->init_req(1, control());
1089 }
1090 }
1091
1092 // post merge
1093 set_control(_gvn.transform(region));
1094 record_for_igvn(region);
1095
1096 set_result(_gvn.transform(phi));
1097 return true;
1098 }
1099
1100 //------------------------------inline_array_equals----------------------------
1101 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1102 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1103 Node* arg1 = argument(0);
1104 Node* arg2 = argument(1);
1105
1106 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1107 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1108 clear_upper_avx();
1109
1110 return true;
1111 }
1112
1113
1114 //------------------------------inline_countPositives------------------------------
1115 bool LibraryCallKit::inline_countPositives() {
1116 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1117 return false;
1118 }
1119
1120 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1121 // no receiver since it is static method
1122 Node* ba = argument(0);
1123 Node* offset = argument(1);
1124 Node* len = argument(2);
1125
1126 ba = must_be_not_null(ba, true);
1127
1128 // Range checks
1129 generate_string_range_check(ba, offset, len, false);
1130 if (stopped()) {
1131 return true;
1132 }
1133 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1134 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1135 set_result(_gvn.transform(result));
1136 clear_upper_avx();
1137 return true;
1138 }
1139
1140 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1141 Node* index = argument(0);
1142 Node* length = bt == T_INT ? argument(1) : argument(2);
1143 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1144 return false;
1145 }
1146
1147 // check that length is positive
1148 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1149 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1150
1151 {
1152 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1153 uncommon_trap(Deoptimization::Reason_intrinsic,
1154 Deoptimization::Action_make_not_entrant);
1155 }
1156
1157 if (stopped()) {
1158 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1159 return true;
1160 }
1161
1162 // length is now known positive, add a cast node to make this explicit
1163 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1164 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1165 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1166 ConstraintCastNode::RegularDependency, bt);
1167 casted_length = _gvn.transform(casted_length);
1168 replace_in_map(length, casted_length);
1169 length = casted_length;
1170
1171 // Use an unsigned comparison for the range check itself
1172 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1173 BoolTest::mask btest = BoolTest::lt;
1174 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1175 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1176 _gvn.set_type(rc, rc->Value(&_gvn));
1177 if (!rc_bool->is_Con()) {
1178 record_for_igvn(rc);
1179 }
1180 set_control(_gvn.transform(new IfTrueNode(rc)));
1181 {
1182 PreserveJVMState pjvms(this);
1183 set_control(_gvn.transform(new IfFalseNode(rc)));
1184 uncommon_trap(Deoptimization::Reason_range_check,
1185 Deoptimization::Action_make_not_entrant);
1186 }
1187
1188 if (stopped()) {
1189 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1190 return true;
1191 }
1192
1193 // index is now known to be >= 0 and < length, cast it
1194 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1195 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1196 ConstraintCastNode::RegularDependency, bt);
1197 result = _gvn.transform(result);
1198 set_result(result);
1199 replace_in_map(index, result);
1200 return true;
1201 }
1202
1203 //------------------------------inline_string_indexOf------------------------
1204 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1205 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1206 return false;
1207 }
1208 Node* src = argument(0);
1209 Node* tgt = argument(1);
1210
1211 // Make the merge point
1212 RegionNode* result_rgn = new RegionNode(4);
1213 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1214
1215 src = must_be_not_null(src, true);
1216 tgt = must_be_not_null(tgt, true);
1217
1218 // Get start addr and length of source string
1219 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1220 Node* src_count = load_array_length(src);
1221
1222 // Get start addr and length of substring
1223 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1224 Node* tgt_count = load_array_length(tgt);
1225
1226 Node* result = nullptr;
1227 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1228
1229 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1230 // Divide src size by 2 if String is UTF16 encoded
1231 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1232 }
1233 if (ae == StrIntrinsicNode::UU) {
1234 // Divide substring size by 2 if String is UTF16 encoded
1235 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1236 }
1237
1238 if (call_opt_stub) {
1239 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1240 StubRoutines::_string_indexof_array[ae],
1241 "stringIndexOf", TypePtr::BOTTOM, src_start,
1242 src_count, tgt_start, tgt_count);
1243 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1244 } else {
1245 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1246 result_rgn, result_phi, ae);
1247 }
1248 if (result != nullptr) {
1249 result_phi->init_req(3, result);
1250 result_rgn->init_req(3, control());
1251 }
1252 set_control(_gvn.transform(result_rgn));
1253 record_for_igvn(result_rgn);
1254 set_result(_gvn.transform(result_phi));
1255
1256 return true;
1257 }
1258
1259 //-----------------------------inline_string_indexOfI-----------------------
1260 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1261 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1262 return false;
1263 }
1264 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1265 return false;
1266 }
1267
1268 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1269 Node* src = argument(0); // byte[]
1270 Node* src_count = argument(1); // char count
1271 Node* tgt = argument(2); // byte[]
1272 Node* tgt_count = argument(3); // char count
1273 Node* from_index = argument(4); // char index
1274
1275 src = must_be_not_null(src, true);
1276 tgt = must_be_not_null(tgt, true);
1277
1278 // Multiply byte array index by 2 if String is UTF16 encoded
1279 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1280 src_count = _gvn.transform(new SubINode(src_count, from_index));
1281 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1282 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1283
1284 // Range checks
1285 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1286 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1287 if (stopped()) {
1288 return true;
1289 }
1290
1291 RegionNode* region = new RegionNode(5);
1292 Node* phi = new PhiNode(region, TypeInt::INT);
1293 Node* result = nullptr;
1294
1295 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1296
1297 if (call_opt_stub) {
1298 assert(arrayOopDesc::base_offset_in_bytes(T_BYTE) >= 16, "Needed for indexOf");
1299 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1300 StubRoutines::_string_indexof_array[ae],
1301 "stringIndexOf", TypePtr::BOTTOM, src_start,
1302 src_count, tgt_start, tgt_count);
1303 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1304 } else {
1305 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1306 region, phi, ae);
1307 }
1308 if (result != nullptr) {
1309 // The result is index relative to from_index if substring was found, -1 otherwise.
1310 // Generate code which will fold into cmove.
1311 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1312 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1313
1314 Node* if_lt = generate_slow_guard(bol, nullptr);
1315 if (if_lt != nullptr) {
1316 // result == -1
1317 phi->init_req(3, result);
1318 region->init_req(3, if_lt);
1319 }
1320 if (!stopped()) {
1321 result = _gvn.transform(new AddINode(result, from_index));
1322 phi->init_req(4, result);
1323 region->init_req(4, control());
1324 }
1325 }
1326
1327 set_control(_gvn.transform(region));
1328 record_for_igvn(region);
1329 set_result(_gvn.transform(phi));
1330 clear_upper_avx();
1331
1332 return true;
1333 }
1334
1335 // Create StrIndexOfNode with fast path checks
1336 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1337 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1338 // Check for substr count > string count
1339 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1340 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1341 Node* if_gt = generate_slow_guard(bol, nullptr);
1342 if (if_gt != nullptr) {
1343 phi->init_req(1, intcon(-1));
1344 region->init_req(1, if_gt);
1345 }
1346 if (!stopped()) {
1347 // Check for substr count == 0
1348 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1349 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1350 Node* if_zero = generate_slow_guard(bol, nullptr);
1351 if (if_zero != nullptr) {
1352 phi->init_req(2, intcon(0));
1353 region->init_req(2, if_zero);
1354 }
1355 }
1356 if (!stopped()) {
1357 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1358 }
1359 return nullptr;
1360 }
1361
1362 //-----------------------------inline_string_indexOfChar-----------------------
1363 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1364 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1365 return false;
1366 }
1367 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1368 return false;
1369 }
1370 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1371 Node* src = argument(0); // byte[]
1372 Node* int_ch = argument(1);
1373 Node* from_index = argument(2);
1374 Node* max = argument(3);
1375
1376 src = must_be_not_null(src, true);
1377
1378 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1379 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1380 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1381
1382 // Range checks
1383 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1384
1385 // Check for int_ch >= 0
1386 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1387 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1388 {
1389 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1390 uncommon_trap(Deoptimization::Reason_intrinsic,
1391 Deoptimization::Action_maybe_recompile);
1392 }
1393 if (stopped()) {
1394 return true;
1395 }
1396
1397 RegionNode* region = new RegionNode(3);
1398 Node* phi = new PhiNode(region, TypeInt::INT);
1399
1400 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1401 C->set_has_split_ifs(true); // Has chance for split-if optimization
1402 _gvn.transform(result);
1403
1404 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1405 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1406
1407 Node* if_lt = generate_slow_guard(bol, nullptr);
1408 if (if_lt != nullptr) {
1409 // result == -1
1410 phi->init_req(2, result);
1411 region->init_req(2, if_lt);
1412 }
1413 if (!stopped()) {
1414 result = _gvn.transform(new AddINode(result, from_index));
1415 phi->init_req(1, result);
1416 region->init_req(1, control());
1417 }
1418 set_control(_gvn.transform(region));
1419 record_for_igvn(region);
1420 set_result(_gvn.transform(phi));
1421 clear_upper_avx();
1422
1423 return true;
1424 }
1425 //---------------------------inline_string_copy---------------------
1426 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1427 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1428 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1429 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1430 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1431 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1432 bool LibraryCallKit::inline_string_copy(bool compress) {
1433 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1434 return false;
1435 }
1436 int nargs = 5; // 2 oops, 3 ints
1437 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1438
1439 Node* src = argument(0);
1440 Node* src_offset = argument(1);
1441 Node* dst = argument(2);
1442 Node* dst_offset = argument(3);
1443 Node* length = argument(4);
1444
1445 // Check for allocation before we add nodes that would confuse
1446 // tightly_coupled_allocation()
1447 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1448
1449 // Figure out the size and type of the elements we will be copying.
1450 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1451 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1452 if (src_type == nullptr || dst_type == nullptr) {
1453 return false;
1454 }
1455 BasicType src_elem = src_type->elem()->array_element_basic_type();
1456 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1457 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1458 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1459 "Unsupported array types for inline_string_copy");
1460
1461 src = must_be_not_null(src, true);
1462 dst = must_be_not_null(dst, true);
1463
1464 // Convert char[] offsets to byte[] offsets
1465 bool convert_src = (compress && src_elem == T_BYTE);
1466 bool convert_dst = (!compress && dst_elem == T_BYTE);
1467 if (convert_src) {
1468 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1469 } else if (convert_dst) {
1470 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1471 }
1472
1473 // Range checks
1474 generate_string_range_check(src, src_offset, length, convert_src);
1475 generate_string_range_check(dst, dst_offset, length, convert_dst);
1476 if (stopped()) {
1477 return true;
1478 }
1479
1480 Node* src_start = array_element_address(src, src_offset, src_elem);
1481 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1482 // 'src_start' points to src array + scaled offset
1483 // 'dst_start' points to dst array + scaled offset
1484 Node* count = nullptr;
1485 if (compress) {
1486 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1487 } else {
1488 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1489 }
1490
1491 if (alloc != nullptr) {
1492 if (alloc->maybe_set_complete(&_gvn)) {
1493 // "You break it, you buy it."
1494 InitializeNode* init = alloc->initialization();
1495 assert(init->is_complete(), "we just did this");
1496 init->set_complete_with_arraycopy();
1497 assert(dst->is_CheckCastPP(), "sanity");
1498 assert(dst->in(0)->in(0) == init, "dest pinned");
1499 }
1500 // Do not let stores that initialize this object be reordered with
1501 // a subsequent store that would make this object accessible by
1502 // other threads.
1503 // Record what AllocateNode this StoreStore protects so that
1504 // escape analysis can go from the MemBarStoreStoreNode to the
1505 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1506 // based on the escape status of the AllocateNode.
1507 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1508 }
1509 if (compress) {
1510 set_result(_gvn.transform(count));
1511 }
1512 clear_upper_avx();
1513
1514 return true;
1515 }
1516
1517 #ifdef _LP64
1518 #define XTOP ,top() /*additional argument*/
1519 #else //_LP64
1520 #define XTOP /*no additional argument*/
1521 #endif //_LP64
1522
1523 //------------------------inline_string_toBytesU--------------------------
1524 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1525 bool LibraryCallKit::inline_string_toBytesU() {
1526 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1527 return false;
1528 }
1529 // Get the arguments.
1530 Node* value = argument(0);
1531 Node* offset = argument(1);
1532 Node* length = argument(2);
1533
1534 Node* newcopy = nullptr;
1535
1536 // Set the original stack and the reexecute bit for the interpreter to reexecute
1537 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1538 { PreserveReexecuteState preexecs(this);
1539 jvms()->set_should_reexecute(true);
1540
1541 // Check if a null path was taken unconditionally.
1542 value = null_check(value);
1543
1544 RegionNode* bailout = new RegionNode(1);
1545 record_for_igvn(bailout);
1546
1547 // Range checks
1548 generate_negative_guard(offset, bailout);
1549 generate_negative_guard(length, bailout);
1550 generate_limit_guard(offset, length, load_array_length(value), bailout);
1551 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1552 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1553
1554 if (bailout->req() > 1) {
1555 PreserveJVMState pjvms(this);
1556 set_control(_gvn.transform(bailout));
1557 uncommon_trap(Deoptimization::Reason_intrinsic,
1558 Deoptimization::Action_maybe_recompile);
1559 }
1560 if (stopped()) {
1561 return true;
1562 }
1563
1564 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1565 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1566 newcopy = new_array(klass_node, size, 0); // no arguments to push
1567 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1568 guarantee(alloc != nullptr, "created above");
1569
1570 // Calculate starting addresses.
1571 Node* src_start = array_element_address(value, offset, T_CHAR);
1572 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1573
1574 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1575 const TypeInt* toffset = gvn().type(offset)->is_int();
1576 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1577
1578 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1579 const char* copyfunc_name = "arraycopy";
1580 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1581 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1582 OptoRuntime::fast_arraycopy_Type(),
1583 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1584 src_start, dst_start, ConvI2X(length) XTOP);
1585 // Do not let reads from the cloned object float above the arraycopy.
1586 if (alloc->maybe_set_complete(&_gvn)) {
1587 // "You break it, you buy it."
1588 InitializeNode* init = alloc->initialization();
1589 assert(init->is_complete(), "we just did this");
1590 init->set_complete_with_arraycopy();
1591 assert(newcopy->is_CheckCastPP(), "sanity");
1592 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1593 }
1594 // Do not let stores that initialize this object be reordered with
1595 // a subsequent store that would make this object accessible by
1596 // other threads.
1597 // Record what AllocateNode this StoreStore protects so that
1598 // escape analysis can go from the MemBarStoreStoreNode to the
1599 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1600 // based on the escape status of the AllocateNode.
1601 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1602 } // original reexecute is set back here
1603
1604 C->set_has_split_ifs(true); // Has chance for split-if optimization
1605 if (!stopped()) {
1606 set_result(newcopy);
1607 }
1608 clear_upper_avx();
1609
1610 return true;
1611 }
1612
1613 //------------------------inline_string_getCharsU--------------------------
1614 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1615 bool LibraryCallKit::inline_string_getCharsU() {
1616 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1617 return false;
1618 }
1619
1620 // Get the arguments.
1621 Node* src = argument(0);
1622 Node* src_begin = argument(1);
1623 Node* src_end = argument(2); // exclusive offset (i < src_end)
1624 Node* dst = argument(3);
1625 Node* dst_begin = argument(4);
1626
1627 // Check for allocation before we add nodes that would confuse
1628 // tightly_coupled_allocation()
1629 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1630
1631 // Check if a null path was taken unconditionally.
1632 src = null_check(src);
1633 dst = null_check(dst);
1634 if (stopped()) {
1635 return true;
1636 }
1637
1638 // Get length and convert char[] offset to byte[] offset
1639 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1640 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1641
1642 // Range checks
1643 generate_string_range_check(src, src_begin, length, true);
1644 generate_string_range_check(dst, dst_begin, length, false);
1645 if (stopped()) {
1646 return true;
1647 }
1648
1649 if (!stopped()) {
1650 // Calculate starting addresses.
1651 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1652 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1653
1654 // Check if array addresses are aligned to HeapWordSize
1655 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1656 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1657 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1658 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1659
1660 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1661 const char* copyfunc_name = "arraycopy";
1662 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1663 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1664 OptoRuntime::fast_arraycopy_Type(),
1665 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1666 src_start, dst_start, ConvI2X(length) XTOP);
1667 // Do not let reads from the cloned object float above the arraycopy.
1668 if (alloc != nullptr) {
1669 if (alloc->maybe_set_complete(&_gvn)) {
1670 // "You break it, you buy it."
1671 InitializeNode* init = alloc->initialization();
1672 assert(init->is_complete(), "we just did this");
1673 init->set_complete_with_arraycopy();
1674 assert(dst->is_CheckCastPP(), "sanity");
1675 assert(dst->in(0)->in(0) == init, "dest pinned");
1676 }
1677 // Do not let stores that initialize this object be reordered with
1678 // a subsequent store that would make this object accessible by
1679 // other threads.
1680 // Record what AllocateNode this StoreStore protects so that
1681 // escape analysis can go from the MemBarStoreStoreNode to the
1682 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1683 // based on the escape status of the AllocateNode.
1684 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1685 } else {
1686 insert_mem_bar(Op_MemBarCPUOrder);
1687 }
1688 }
1689
1690 C->set_has_split_ifs(true); // Has chance for split-if optimization
1691 return true;
1692 }
1693
1694 //----------------------inline_string_char_access----------------------------
1695 // Store/Load char to/from byte[] array.
1696 // static void StringUTF16.putChar(byte[] val, int index, int c)
1697 // static char StringUTF16.getChar(byte[] val, int index)
1698 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1699 Node* value = argument(0);
1700 Node* index = argument(1);
1701 Node* ch = is_store ? argument(2) : nullptr;
1702
1703 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1704 // correctly requires matched array shapes.
1705 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1706 "sanity: byte[] and char[] bases agree");
1707 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1708 "sanity: byte[] and char[] scales agree");
1709
1710 // Bail when getChar over constants is requested: constant folding would
1711 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1712 // Java method would constant fold nicely instead.
1713 if (!is_store && value->is_Con() && index->is_Con()) {
1714 return false;
1715 }
1716
1717 // Save state and restore on bailout
1718 uint old_sp = sp();
1719 SafePointNode* old_map = clone_map();
1720
1721 value = must_be_not_null(value, true);
1722
1723 Node* adr = array_element_address(value, index, T_CHAR);
1724 if (adr->is_top()) {
1725 set_map(old_map);
1726 set_sp(old_sp);
1727 return false;
1728 }
1729 destruct_map_clone(old_map);
1730 if (is_store) {
1731 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1732 } else {
1733 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);
1734 set_result(ch);
1735 }
1736 return true;
1737 }
1738
1739 //--------------------------round_double_node--------------------------------
1740 // Round a double node if necessary.
1741 Node* LibraryCallKit::round_double_node(Node* n) {
1742 if (Matcher::strict_fp_requires_explicit_rounding) {
1743 #ifdef IA32
1744 if (UseSSE < 2) {
1745 n = _gvn.transform(new RoundDoubleNode(nullptr, n));
1746 }
1747 #else
1748 Unimplemented();
1749 #endif // IA32
1750 }
1751 return n;
1752 }
1753
1754 //------------------------------inline_math-----------------------------------
1755 // public static double Math.abs(double)
1756 // public static double Math.sqrt(double)
1757 // public static double Math.log(double)
1758 // public static double Math.log10(double)
1759 // public static double Math.round(double)
1760 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1761 Node* arg = round_double_node(argument(0));
1762 Node* n = nullptr;
1763 switch (id) {
1764 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1765 case vmIntrinsics::_dsqrt:
1766 case vmIntrinsics::_dsqrt_strict:
1767 n = new SqrtDNode(C, control(), arg); break;
1768 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1769 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1770 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1771 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1772 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1773 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1774 default: fatal_unexpected_iid(id); break;
1775 }
1776 set_result(_gvn.transform(n));
1777 return true;
1778 }
1779
1780 //------------------------------inline_math-----------------------------------
1781 // public static float Math.abs(float)
1782 // public static int Math.abs(int)
1783 // public static long Math.abs(long)
1784 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1785 Node* arg = argument(0);
1786 Node* n = nullptr;
1787 switch (id) {
1788 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1789 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1790 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1791 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1792 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1793 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1794 default: fatal_unexpected_iid(id); break;
1795 }
1796 set_result(_gvn.transform(n));
1797 return true;
1798 }
1799
1800 //------------------------------runtime_math-----------------------------
1801 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1802 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1803 "must be (DD)D or (D)D type");
1804
1805 // Inputs
1806 Node* a = round_double_node(argument(0));
1807 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : nullptr;
1808
1809 const TypePtr* no_memory_effects = nullptr;
1810 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1811 no_memory_effects,
1812 a, top(), b, b ? top() : nullptr);
1813 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1814 #ifdef ASSERT
1815 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1816 assert(value_top == top(), "second value must be top");
1817 #endif
1818
1819 set_result(value);
1820 return true;
1821 }
1822
1823 //------------------------------inline_math_pow-----------------------------
1824 bool LibraryCallKit::inline_math_pow() {
1825 Node* exp = round_double_node(argument(2));
1826 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1827 if (d != nullptr) {
1828 if (d->getd() == 2.0) {
1829 // Special case: pow(x, 2.0) => x * x
1830 Node* base = round_double_node(argument(0));
1831 set_result(_gvn.transform(new MulDNode(base, base)));
1832 return true;
1833 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1834 // Special case: pow(x, 0.5) => sqrt(x)
1835 Node* base = round_double_node(argument(0));
1836 Node* zero = _gvn.zerocon(T_DOUBLE);
1837
1838 RegionNode* region = new RegionNode(3);
1839 Node* phi = new PhiNode(region, Type::DOUBLE);
1840
1841 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1842 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1843 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1844 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1845 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1846
1847 Node* if_pow = generate_slow_guard(test, nullptr);
1848 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1849 phi->init_req(1, value_sqrt);
1850 region->init_req(1, control());
1851
1852 if (if_pow != nullptr) {
1853 set_control(if_pow);
1854 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1855 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1856 const TypePtr* no_memory_effects = nullptr;
1857 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1858 no_memory_effects, base, top(), exp, top());
1859 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1860 #ifdef ASSERT
1861 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1862 assert(value_top == top(), "second value must be top");
1863 #endif
1864 phi->init_req(2, value_pow);
1865 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1866 }
1867
1868 C->set_has_split_ifs(true); // Has chance for split-if optimization
1869 set_control(_gvn.transform(region));
1870 record_for_igvn(region);
1871 set_result(_gvn.transform(phi));
1872
1873 return true;
1874 }
1875 }
1876
1877 return StubRoutines::dpow() != nullptr ?
1878 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1879 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1880 }
1881
1882 //------------------------------inline_math_native-----------------------------
1883 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1884 switch (id) {
1885 case vmIntrinsics::_dsin:
1886 return StubRoutines::dsin() != nullptr ?
1887 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1888 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1889 case vmIntrinsics::_dcos:
1890 return StubRoutines::dcos() != nullptr ?
1891 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1892 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1893 case vmIntrinsics::_dtan:
1894 return StubRoutines::dtan() != nullptr ?
1895 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1896 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1897 case vmIntrinsics::_dtanh:
1898 return StubRoutines::dtanh() != nullptr ?
1899 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1900 case vmIntrinsics::_dexp:
1901 return StubRoutines::dexp() != nullptr ?
1902 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1903 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1904 case vmIntrinsics::_dlog:
1905 return StubRoutines::dlog() != nullptr ?
1906 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1907 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1908 case vmIntrinsics::_dlog10:
1909 return StubRoutines::dlog10() != nullptr ?
1910 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1911 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1912
1913 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1914 case vmIntrinsics::_ceil:
1915 case vmIntrinsics::_floor:
1916 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1917
1918 case vmIntrinsics::_dsqrt:
1919 case vmIntrinsics::_dsqrt_strict:
1920 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1921 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1922 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1923 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1924 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1925
1926 case vmIntrinsics::_dpow: return inline_math_pow();
1927 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1928 case vmIntrinsics::_fcopySign: return inline_math(id);
1929 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1930 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1931 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1932
1933 // These intrinsics are not yet correctly implemented
1934 case vmIntrinsics::_datan2:
1935 return false;
1936
1937 default:
1938 fatal_unexpected_iid(id);
1939 return false;
1940 }
1941 }
1942
1943 //----------------------------inline_notify-----------------------------------*
1944 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1945 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1946 address func;
1947 if (id == vmIntrinsics::_notify) {
1948 func = OptoRuntime::monitor_notify_Java();
1949 } else {
1950 func = OptoRuntime::monitor_notifyAll_Java();
1951 }
1952 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1953 make_slow_call_ex(call, env()->Throwable_klass(), false);
1954 return true;
1955 }
1956
1957
1958 //----------------------------inline_min_max-----------------------------------
1959 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1960 set_result(generate_min_max(id, argument(0), argument(1)));
1961 return true;
1962 }
1963
1964 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1965 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1966 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1967 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1968 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1969
1970 {
1971 PreserveJVMState pjvms(this);
1972 PreserveReexecuteState preexecs(this);
1973 jvms()->set_should_reexecute(true);
1974
1975 set_control(slow_path);
1976 set_i_o(i_o());
1977
1978 uncommon_trap(Deoptimization::Reason_intrinsic,
1979 Deoptimization::Action_none);
1980 }
1981
1982 set_control(fast_path);
1983 set_result(math);
1984 }
1985
1986 template <typename OverflowOp>
1987 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1988 typedef typename OverflowOp::MathOp MathOp;
1989
1990 MathOp* mathOp = new MathOp(arg1, arg2);
1991 Node* operation = _gvn.transform( mathOp );
1992 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1993 inline_math_mathExact(operation, ofcheck);
1994 return true;
1995 }
1996
1997 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1998 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1999 }
2000
2001 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2002 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2003 }
2004
2005 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2006 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2007 }
2008
2009 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2010 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2011 }
2012
2013 bool LibraryCallKit::inline_math_negateExactI() {
2014 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2015 }
2016
2017 bool LibraryCallKit::inline_math_negateExactL() {
2018 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2019 }
2020
2021 bool LibraryCallKit::inline_math_multiplyExactI() {
2022 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2023 }
2024
2025 bool LibraryCallKit::inline_math_multiplyExactL() {
2026 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2027 }
2028
2029 bool LibraryCallKit::inline_math_multiplyHigh() {
2030 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2031 return true;
2032 }
2033
2034 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2035 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2036 return true;
2037 }
2038
2039 Node*
2040 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
2041 Node* result_val = nullptr;
2042 switch (id) {
2043 case vmIntrinsics::_min:
2044 case vmIntrinsics::_min_strict:
2045 result_val = _gvn.transform(new MinINode(x0, y0));
2046 break;
2047 case vmIntrinsics::_max:
2048 case vmIntrinsics::_max_strict:
2049 result_val = _gvn.transform(new MaxINode(x0, y0));
2050 break;
2051 default:
2052 fatal_unexpected_iid(id);
2053 break;
2054 }
2055 return result_val;
2056 }
2057
2058 inline int
2059 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2060 const TypePtr* base_type = TypePtr::NULL_PTR;
2061 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2062 if (base_type == nullptr) {
2063 // Unknown type.
2064 return Type::AnyPtr;
2065 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2066 // Since this is a null+long form, we have to switch to a rawptr.
2067 base = _gvn.transform(new CastX2PNode(offset));
2068 offset = MakeConX(0);
2069 return Type::RawPtr;
2070 } else if (base_type->base() == Type::RawPtr) {
2071 return Type::RawPtr;
2072 } else if (base_type->isa_oopptr()) {
2073 // Base is never null => always a heap address.
2074 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2075 return Type::OopPtr;
2076 }
2077 // Offset is small => always a heap address.
2078 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2079 if (offset_type != nullptr &&
2080 base_type->offset() == 0 && // (should always be?)
2081 offset_type->_lo >= 0 &&
2082 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2083 return Type::OopPtr;
2084 } else if (type == T_OBJECT) {
2085 // off heap access to an oop doesn't make any sense. Has to be on
2086 // heap.
2087 return Type::OopPtr;
2088 }
2089 // Otherwise, it might either be oop+off or null+addr.
2090 return Type::AnyPtr;
2091 } else {
2092 // No information:
2093 return Type::AnyPtr;
2094 }
2095 }
2096
2097 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2098 Node* uncasted_base = base;
2099 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2100 if (kind == Type::RawPtr) {
2101 return basic_plus_adr(top(), uncasted_base, offset);
2102 } else if (kind == Type::AnyPtr) {
2103 assert(base == uncasted_base, "unexpected base change");
2104 if (can_cast) {
2105 if (!_gvn.type(base)->speculative_maybe_null() &&
2106 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2107 // According to profiling, this access is always on
2108 // heap. Casting the base to not null and thus avoiding membars
2109 // around the access should allow better optimizations
2110 Node* null_ctl = top();
2111 base = null_check_oop(base, &null_ctl, true, true, true);
2112 assert(null_ctl->is_top(), "no null control here");
2113 return basic_plus_adr(base, offset);
2114 } else if (_gvn.type(base)->speculative_always_null() &&
2115 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2116 // According to profiling, this access is always off
2117 // heap.
2118 base = null_assert(base);
2119 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2120 offset = MakeConX(0);
2121 return basic_plus_adr(top(), raw_base, offset);
2122 }
2123 }
2124 // We don't know if it's an on heap or off heap access. Fall back
2125 // to raw memory access.
2126 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2127 return basic_plus_adr(top(), raw, offset);
2128 } else {
2129 assert(base == uncasted_base, "unexpected base change");
2130 // We know it's an on heap access so base can't be null
2131 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2132 base = must_be_not_null(base, true);
2133 }
2134 return basic_plus_adr(base, offset);
2135 }
2136 }
2137
2138 //--------------------------inline_number_methods-----------------------------
2139 // inline int Integer.numberOfLeadingZeros(int)
2140 // inline int Long.numberOfLeadingZeros(long)
2141 //
2142 // inline int Integer.numberOfTrailingZeros(int)
2143 // inline int Long.numberOfTrailingZeros(long)
2144 //
2145 // inline int Integer.bitCount(int)
2146 // inline int Long.bitCount(long)
2147 //
2148 // inline char Character.reverseBytes(char)
2149 // inline short Short.reverseBytes(short)
2150 // inline int Integer.reverseBytes(int)
2151 // inline long Long.reverseBytes(long)
2152 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2153 Node* arg = argument(0);
2154 Node* n = nullptr;
2155 switch (id) {
2156 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2157 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2158 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2159 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2160 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2161 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2162 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(nullptr, arg); break;
2163 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( nullptr, arg); break;
2164 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( nullptr, arg); break;
2165 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( nullptr, arg); break;
2166 case vmIntrinsics::_reverse_i: n = new ReverseINode(nullptr, arg); break;
2167 case vmIntrinsics::_reverse_l: n = new ReverseLNode(nullptr, arg); break;
2168 default: fatal_unexpected_iid(id); break;
2169 }
2170 set_result(_gvn.transform(n));
2171 return true;
2172 }
2173
2174 //--------------------------inline_bitshuffle_methods-----------------------------
2175 // inline int Integer.compress(int, int)
2176 // inline int Integer.expand(int, int)
2177 // inline long Long.compress(long, long)
2178 // inline long Long.expand(long, long)
2179 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2180 Node* n = nullptr;
2181 switch (id) {
2182 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2183 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2184 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2185 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2186 default: fatal_unexpected_iid(id); break;
2187 }
2188 set_result(_gvn.transform(n));
2189 return true;
2190 }
2191
2192 //--------------------------inline_number_methods-----------------------------
2193 // inline int Integer.compareUnsigned(int, int)
2194 // inline int Long.compareUnsigned(long, long)
2195 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2196 Node* arg1 = argument(0);
2197 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2198 Node* n = nullptr;
2199 switch (id) {
2200 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2201 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2202 default: fatal_unexpected_iid(id); break;
2203 }
2204 set_result(_gvn.transform(n));
2205 return true;
2206 }
2207
2208 //--------------------------inline_unsigned_divmod_methods-----------------------------
2209 // inline int Integer.divideUnsigned(int, int)
2210 // inline int Integer.remainderUnsigned(int, int)
2211 // inline long Long.divideUnsigned(long, long)
2212 // inline long Long.remainderUnsigned(long, long)
2213 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2214 Node* n = nullptr;
2215 switch (id) {
2216 case vmIntrinsics::_divideUnsigned_i: {
2217 zero_check_int(argument(1));
2218 // Compile-time detect of null-exception
2219 if (stopped()) {
2220 return true; // keep the graph constructed so far
2221 }
2222 n = new UDivINode(control(), argument(0), argument(1));
2223 break;
2224 }
2225 case vmIntrinsics::_divideUnsigned_l: {
2226 zero_check_long(argument(2));
2227 // Compile-time detect of null-exception
2228 if (stopped()) {
2229 return true; // keep the graph constructed so far
2230 }
2231 n = new UDivLNode(control(), argument(0), argument(2));
2232 break;
2233 }
2234 case vmIntrinsics::_remainderUnsigned_i: {
2235 zero_check_int(argument(1));
2236 // Compile-time detect of null-exception
2237 if (stopped()) {
2238 return true; // keep the graph constructed so far
2239 }
2240 n = new UModINode(control(), argument(0), argument(1));
2241 break;
2242 }
2243 case vmIntrinsics::_remainderUnsigned_l: {
2244 zero_check_long(argument(2));
2245 // Compile-time detect of null-exception
2246 if (stopped()) {
2247 return true; // keep the graph constructed so far
2248 }
2249 n = new UModLNode(control(), argument(0), argument(2));
2250 break;
2251 }
2252 default: fatal_unexpected_iid(id); break;
2253 }
2254 set_result(_gvn.transform(n));
2255 return true;
2256 }
2257
2258 //----------------------------inline_unsafe_access----------------------------
2259
2260 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2261 // Attempt to infer a sharper value type from the offset and base type.
2262 ciKlass* sharpened_klass = nullptr;
2263 bool null_free = false;
2264
2265 // See if it is an instance field, with an object type.
2266 if (alias_type->field() != nullptr) {
2267 if (alias_type->field()->type()->is_klass()) {
2268 sharpened_klass = alias_type->field()->type()->as_klass();
2269 null_free = alias_type->field()->is_null_free();
2270 }
2271 }
2272
2273 const TypeOopPtr* result = nullptr;
2274 // See if it is a narrow oop array.
2275 if (adr_type->isa_aryptr()) {
2276 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2277 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2278 null_free = adr_type->is_aryptr()->is_null_free();
2279 if (elem_type != nullptr && elem_type->is_loaded()) {
2280 // Sharpen the value type.
2281 result = elem_type;
2282 }
2283 }
2284 }
2285
2286 // The sharpened class might be unloaded if there is no class loader
2287 // contraint in place.
2288 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2289 // Sharpen the value type.
2290 result = TypeOopPtr::make_from_klass(sharpened_klass);
2291 if (null_free) {
2292 result = result->join_speculative(TypePtr::NOTNULL)->is_oopptr();
2293 }
2294 }
2295 if (result != nullptr) {
2296 #ifndef PRODUCT
2297 if (C->print_intrinsics() || C->print_inlining()) {
2298 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2299 tty->print(" sharpened value: "); result->dump(); tty->cr();
2300 }
2301 #endif
2302 }
2303 return result;
2304 }
2305
2306 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2307 switch (kind) {
2308 case Relaxed:
2309 return MO_UNORDERED;
2310 case Opaque:
2311 return MO_RELAXED;
2312 case Acquire:
2313 return MO_ACQUIRE;
2314 case Release:
2315 return MO_RELEASE;
2316 case Volatile:
2317 return MO_SEQ_CST;
2318 default:
2319 ShouldNotReachHere();
2320 return 0;
2321 }
2322 }
2323
2324 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned, const bool is_flat) {
2325 if (callee()->is_static()) return false; // caller must have the capability!
2326 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2327 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2328 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2329 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2330
2331 if (is_reference_type(type)) {
2332 decorators |= ON_UNKNOWN_OOP_REF;
2333 }
2334
2335 if (unaligned) {
2336 decorators |= C2_UNALIGNED;
2337 }
2338
2339 #ifndef PRODUCT
2340 {
2341 ResourceMark rm;
2342 // Check the signatures.
2343 ciSignature* sig = callee()->signature();
2344 #ifdef ASSERT
2345 if (!is_store) {
2346 // Object getReference(Object base, int/long offset), etc.
2347 BasicType rtype = sig->return_type()->basic_type();
2348 assert(rtype == type, "getter must return the expected value");
2349 assert(sig->count() == 2 || (is_flat && sig->count() == 3), "oop getter has 2 or 3 arguments");
2350 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2351 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2352 } else {
2353 // void putReference(Object base, int/long offset, Object x), etc.
2354 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2355 assert(sig->count() == 3 || (is_flat && sig->count() == 4), "oop putter has 3 arguments");
2356 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2357 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2358 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2359 assert(vtype == type, "putter must accept the expected value");
2360 }
2361 #endif // ASSERT
2362 }
2363 #endif //PRODUCT
2364
2365 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2366
2367 Node* receiver = argument(0); // type: oop
2368
2369 // Build address expression.
2370 Node* heap_base_oop = top();
2371
2372 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2373 Node* base = argument(1); // type: oop
2374 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2375 Node* offset = argument(2); // type: long
2376 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2377 // to be plain byte offsets, which are also the same as those accepted
2378 // by oopDesc::field_addr.
2379 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2380 "fieldOffset must be byte-scaled");
2381
2382 ciInlineKlass* inline_klass = nullptr;
2383 if (is_flat) {
2384 const TypeInstPtr* cls = _gvn.type(argument(4))->isa_instptr();
2385 if (cls == nullptr || cls->const_oop() == nullptr) {
2386 return false;
2387 }
2388 ciType* mirror_type = cls->const_oop()->as_instance()->java_mirror_type();
2389 if (!mirror_type->is_inlinetype()) {
2390 return false;
2391 }
2392 inline_klass = mirror_type->as_inline_klass();
2393 }
2394
2395 if (base->is_InlineType()) {
2396 InlineTypeNode* vt = base->as_InlineType();
2397 if (is_store) {
2398 if (!vt->is_allocated(&_gvn)) {
2399 return false;
2400 }
2401 base = vt->get_oop();
2402 } else {
2403 if (offset->is_Con()) {
2404 long off = find_long_con(offset, 0);
2405 ciInlineKlass* vk = vt->type()->inline_klass();
2406 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2407 return false;
2408 }
2409
2410 ciField* field = vk->get_non_flat_field_by_offset(off);
2411 if (field != nullptr) {
2412 BasicType bt = type2field[field->type()->basic_type()];
2413 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2414 bt = T_OBJECT;
2415 }
2416 if (bt == type && (!field->is_flat() || field->type() == inline_klass)) {
2417 Node* value = vt->field_value_by_offset(off, false);
2418 if (value->is_InlineType()) {
2419 value = value->as_InlineType()->adjust_scalarization_depth(this);
2420 }
2421 set_result(value);
2422 return true;
2423 }
2424 }
2425 }
2426 {
2427 // Re-execute the unsafe access if allocation triggers deoptimization.
2428 PreserveReexecuteState preexecs(this);
2429 jvms()->set_should_reexecute(true);
2430 vt = vt->buffer(this);
2431 }
2432 base = vt->get_oop();
2433 }
2434 }
2435
2436 // 32-bit machines ignore the high half!
2437 offset = ConvL2X(offset);
2438
2439 // Save state and restore on bailout
2440 uint old_sp = sp();
2441 SafePointNode* old_map = clone_map();
2442
2443 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2444 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2445
2446 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2447 if (type != T_OBJECT && (inline_klass == nullptr || !inline_klass->has_object_fields())) {
2448 decorators |= IN_NATIVE; // off-heap primitive access
2449 } else {
2450 set_map(old_map);
2451 set_sp(old_sp);
2452 return false; // off-heap oop accesses are not supported
2453 }
2454 } else {
2455 heap_base_oop = base; // on-heap or mixed access
2456 }
2457
2458 // Can base be null? Otherwise, always on-heap access.
2459 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2460
2461 if (!can_access_non_heap) {
2462 decorators |= IN_HEAP;
2463 }
2464
2465 Node* val = is_store ? argument(4 + (is_flat ? 1 : 0)) : nullptr;
2466
2467 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2468 if (adr_type == TypePtr::NULL_PTR) {
2469 set_map(old_map);
2470 set_sp(old_sp);
2471 return false; // off-heap access with zero address
2472 }
2473
2474 // Try to categorize the address.
2475 Compile::AliasType* alias_type = C->alias_type(adr_type);
2476 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2477
2478 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2479 alias_type->adr_type() == TypeAryPtr::RANGE) {
2480 set_map(old_map);
2481 set_sp(old_sp);
2482 return false; // not supported
2483 }
2484
2485 bool mismatched = false;
2486 BasicType bt = T_ILLEGAL;
2487 ciField* field = nullptr;
2488 if (adr_type->isa_instptr()) {
2489 const TypeInstPtr* instptr = adr_type->is_instptr();
2490 ciInstanceKlass* k = instptr->instance_klass();
2491 int off = instptr->offset();
2492 if (instptr->const_oop() != nullptr &&
2493 k == ciEnv::current()->Class_klass() &&
2494 instptr->offset() >= (k->size_helper() * wordSize)) {
2495 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2496 field = k->get_field_by_offset(off, true);
2497 } else {
2498 field = k->get_non_flat_field_by_offset(off);
2499 }
2500 if (field != nullptr) {
2501 bt = type2field[field->type()->basic_type()];
2502 }
2503 assert(bt == alias_type->basic_type() || is_flat, "should match");
2504 } else {
2505 bt = alias_type->basic_type();
2506 }
2507
2508 if (bt != T_ILLEGAL) {
2509 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2510 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2511 // Alias type doesn't differentiate between byte[] and boolean[]).
2512 // Use address type to get the element type.
2513 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2514 }
2515 if (is_reference_type(bt, true)) {
2516 // accessing an array field with getReference is not a mismatch
2517 bt = T_OBJECT;
2518 }
2519 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2520 // Don't intrinsify mismatched object accesses
2521 set_map(old_map);
2522 set_sp(old_sp);
2523 return false;
2524 }
2525 mismatched = (bt != type);
2526 } else if (alias_type->adr_type()->isa_oopptr()) {
2527 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2528 }
2529
2530 if (is_flat) {
2531 if (adr_type->isa_instptr()) {
2532 if (field == nullptr || field->type() != inline_klass) {
2533 mismatched = true;
2534 }
2535 } else if (adr_type->isa_aryptr()) {
2536 const Type* elem = adr_type->is_aryptr()->elem();
2537 if (!adr_type->is_flat() || elem->inline_klass() != inline_klass) {
2538 mismatched = true;
2539 }
2540 } else {
2541 mismatched = true;
2542 }
2543 if (is_store) {
2544 const Type* val_t = _gvn.type(val);
2545 if (!val_t->is_inlinetypeptr() || val_t->inline_klass() != inline_klass) {
2546 set_map(old_map);
2547 set_sp(old_sp);
2548 return false;
2549 }
2550 }
2551 }
2552
2553 destruct_map_clone(old_map);
2554 assert(!mismatched || is_flat || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2555
2556 if (mismatched) {
2557 decorators |= C2_MISMATCHED;
2558 }
2559
2560 // First guess at the value type.
2561 const Type *value_type = Type::get_const_basic_type(type);
2562
2563 // Figure out the memory ordering.
2564 decorators |= mo_decorator_for_access_kind(kind);
2565
2566 if (!is_store) {
2567 if (type == T_OBJECT && !is_flat) {
2568 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2569 if (tjp != nullptr) {
2570 value_type = tjp;
2571 }
2572 }
2573 }
2574
2575 receiver = null_check(receiver);
2576 if (stopped()) {
2577 return true;
2578 }
2579 // Heap pointers get a null-check from the interpreter,
2580 // as a courtesy. However, this is not guaranteed by Unsafe,
2581 // and it is not possible to fully distinguish unintended nulls
2582 // from intended ones in this API.
2583
2584 if (!is_store) {
2585 Node* p = nullptr;
2586 // Try to constant fold a load from a constant field
2587
2588 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !field->is_flat() && !mismatched) {
2589 // final or stable field
2590 p = make_constant_from_field(field, heap_base_oop);
2591 }
2592
2593 if (p == nullptr) { // Could not constant fold the load
2594 if (is_flat) {
2595 if (adr_type->isa_instptr() && !mismatched) {
2596 ciInstanceKlass* holder = adr_type->is_instptr()->instance_klass();
2597 int offset = adr_type->is_instptr()->offset();
2598 p = InlineTypeNode::make_from_flat(this, inline_klass, base, base, holder, offset, decorators);
2599 } else {
2600 p = InlineTypeNode::make_from_flat(this, inline_klass, base, adr, nullptr, 0, decorators);
2601 }
2602 } else {
2603 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2604 const TypeOopPtr* ptr = value_type->make_oopptr();
2605 if (ptr != nullptr && ptr->is_inlinetypeptr()) {
2606 // Load a non-flattened inline type from memory
2607 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass(), !ptr->maybe_null());
2608 }
2609 }
2610 // Normalize the value returned by getBoolean in the following cases
2611 if (type == T_BOOLEAN &&
2612 (mismatched ||
2613 heap_base_oop == top() || // - heap_base_oop is null or
2614 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2615 // and the unsafe access is made to large offset
2616 // (i.e., larger than the maximum offset necessary for any
2617 // field access)
2618 ) {
2619 IdealKit ideal = IdealKit(this);
2620 #define __ ideal.
2621 IdealVariable normalized_result(ideal);
2622 __ declarations_done();
2623 __ set(normalized_result, p);
2624 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2625 __ set(normalized_result, ideal.ConI(1));
2626 ideal.end_if();
2627 final_sync(ideal);
2628 p = __ value(normalized_result);
2629 #undef __
2630 }
2631 }
2632 if (type == T_ADDRESS) {
2633 p = gvn().transform(new CastP2XNode(nullptr, p));
2634 p = ConvX2UL(p);
2635 }
2636 // The load node has the control of the preceding MemBarCPUOrder. All
2637 // following nodes will have the control of the MemBarCPUOrder inserted at
2638 // the end of this method. So, pushing the load onto the stack at a later
2639 // point is fine.
2640 set_result(p);
2641 } else {
2642 if (bt == T_ADDRESS) {
2643 // Repackage the long as a pointer.
2644 val = ConvL2X(val);
2645 val = gvn().transform(new CastX2PNode(val));
2646 }
2647 if (is_flat) {
2648 if (adr_type->isa_instptr() && !mismatched) {
2649 ciInstanceKlass* holder = adr_type->is_instptr()->instance_klass();
2650 int offset = adr_type->is_instptr()->offset();
2651 val->as_InlineType()->store_flat(this, base, base, holder, offset, decorators);
2652 } else {
2653 val->as_InlineType()->store_flat(this, base, adr, nullptr, 0, decorators);
2654 }
2655 } else {
2656 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2657 }
2658 }
2659
2660 if (argument(1)->is_InlineType() && is_store) {
2661 InlineTypeNode* value = InlineTypeNode::make_from_oop(this, base, _gvn.type(argument(1))->inline_klass());
2662 value = value->make_larval(this, false);
2663 replace_in_map(argument(1), value);
2664 }
2665
2666 return true;
2667 }
2668
2669 bool LibraryCallKit::inline_unsafe_make_private_buffer() {
2670 Node* receiver = argument(0);
2671 Node* value = argument(1);
2672 if (!value->is_InlineType()) {
2673 return false;
2674 }
2675
2676 receiver = null_check(receiver);
2677 if (stopped()) {
2678 return true;
2679 }
2680
2681 set_result(value->as_InlineType()->make_larval(this, true));
2682 return true;
2683 }
2684
2685 bool LibraryCallKit::inline_unsafe_finish_private_buffer() {
2686 Node* receiver = argument(0);
2687 Node* buffer = argument(1);
2688 if (!buffer->is_InlineType()) {
2689 return false;
2690 }
2691 InlineTypeNode* vt = buffer->as_InlineType();
2692 if (!vt->is_allocated(&_gvn)) {
2693 return false;
2694 }
2695 // TODO 8239003 Why is this needed?
2696 if (AllocateNode::Ideal_allocation(vt->get_oop()) == nullptr) {
2697 return false;
2698 }
2699
2700 receiver = null_check(receiver);
2701 if (stopped()) {
2702 return true;
2703 }
2704
2705 set_result(vt->finish_larval(this));
2706 return true;
2707 }
2708
2709 //----------------------------inline_unsafe_load_store----------------------------
2710 // This method serves a couple of different customers (depending on LoadStoreKind):
2711 //
2712 // LS_cmp_swap:
2713 //
2714 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2715 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2716 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2717 //
2718 // LS_cmp_swap_weak:
2719 //
2720 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2721 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2722 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2723 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2724 //
2725 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2726 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2727 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2728 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2729 //
2730 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2731 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2732 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2733 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2734 //
2735 // LS_cmp_exchange:
2736 //
2737 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2738 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2739 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2740 //
2741 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2742 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2743 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2744 //
2745 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2746 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2747 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2748 //
2749 // LS_get_add:
2750 //
2751 // int getAndAddInt( Object o, long offset, int delta)
2752 // long getAndAddLong(Object o, long offset, long delta)
2753 //
2754 // LS_get_set:
2755 //
2756 // int getAndSet(Object o, long offset, int newValue)
2757 // long getAndSet(Object o, long offset, long newValue)
2758 // Object getAndSet(Object o, long offset, Object newValue)
2759 //
2760 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2761 // This basic scheme here is the same as inline_unsafe_access, but
2762 // differs in enough details that combining them would make the code
2763 // overly confusing. (This is a true fact! I originally combined
2764 // them, but even I was confused by it!) As much code/comments as
2765 // possible are retained from inline_unsafe_access though to make
2766 // the correspondences clearer. - dl
2767
2768 if (callee()->is_static()) return false; // caller must have the capability!
2769
2770 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2771 decorators |= mo_decorator_for_access_kind(access_kind);
2772
2773 #ifndef PRODUCT
2774 BasicType rtype;
2775 {
2776 ResourceMark rm;
2777 // Check the signatures.
2778 ciSignature* sig = callee()->signature();
2779 rtype = sig->return_type()->basic_type();
2780 switch(kind) {
2781 case LS_get_add:
2782 case LS_get_set: {
2783 // Check the signatures.
2784 #ifdef ASSERT
2785 assert(rtype == type, "get and set must return the expected type");
2786 assert(sig->count() == 3, "get and set has 3 arguments");
2787 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2788 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2789 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2790 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2791 #endif // ASSERT
2792 break;
2793 }
2794 case LS_cmp_swap:
2795 case LS_cmp_swap_weak: {
2796 // Check the signatures.
2797 #ifdef ASSERT
2798 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2799 assert(sig->count() == 4, "CAS has 4 arguments");
2800 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2801 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2802 #endif // ASSERT
2803 break;
2804 }
2805 case LS_cmp_exchange: {
2806 // Check the signatures.
2807 #ifdef ASSERT
2808 assert(rtype == type, "CAS must return the expected type");
2809 assert(sig->count() == 4, "CAS has 4 arguments");
2810 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2811 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2812 #endif // ASSERT
2813 break;
2814 }
2815 default:
2816 ShouldNotReachHere();
2817 }
2818 }
2819 #endif //PRODUCT
2820
2821 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2822
2823 // Get arguments:
2824 Node* receiver = nullptr;
2825 Node* base = nullptr;
2826 Node* offset = nullptr;
2827 Node* oldval = nullptr;
2828 Node* newval = nullptr;
2829 switch(kind) {
2830 case LS_cmp_swap:
2831 case LS_cmp_swap_weak:
2832 case LS_cmp_exchange: {
2833 const bool two_slot_type = type2size[type] == 2;
2834 receiver = argument(0); // type: oop
2835 base = argument(1); // type: oop
2836 offset = argument(2); // type: long
2837 oldval = argument(4); // type: oop, int, or long
2838 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2839 break;
2840 }
2841 case LS_get_add:
2842 case LS_get_set: {
2843 receiver = argument(0); // type: oop
2844 base = argument(1); // type: oop
2845 offset = argument(2); // type: long
2846 oldval = nullptr;
2847 newval = argument(4); // type: oop, int, or long
2848 break;
2849 }
2850 default:
2851 ShouldNotReachHere();
2852 }
2853
2854 // Build field offset expression.
2855 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2856 // to be plain byte offsets, which are also the same as those accepted
2857 // by oopDesc::field_addr.
2858 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2859 // 32-bit machines ignore the high half of long offsets
2860 offset = ConvL2X(offset);
2861 // Save state and restore on bailout
2862 uint old_sp = sp();
2863 SafePointNode* old_map = clone_map();
2864 Node* adr = make_unsafe_address(base, offset,type, false);
2865 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2866
2867 Compile::AliasType* alias_type = C->alias_type(adr_type);
2868 BasicType bt = alias_type->basic_type();
2869 if (bt != T_ILLEGAL &&
2870 (is_reference_type(bt) != (type == T_OBJECT))) {
2871 // Don't intrinsify mismatched object accesses.
2872 set_map(old_map);
2873 set_sp(old_sp);
2874 return false;
2875 }
2876
2877 destruct_map_clone(old_map);
2878
2879 // For CAS, unlike inline_unsafe_access, there seems no point in
2880 // trying to refine types. Just use the coarse types here.
2881 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2882 const Type *value_type = Type::get_const_basic_type(type);
2883
2884 switch (kind) {
2885 case LS_get_set:
2886 case LS_cmp_exchange: {
2887 if (type == T_OBJECT) {
2888 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2889 if (tjp != nullptr) {
2890 value_type = tjp;
2891 }
2892 }
2893 break;
2894 }
2895 case LS_cmp_swap:
2896 case LS_cmp_swap_weak:
2897 case LS_get_add:
2898 break;
2899 default:
2900 ShouldNotReachHere();
2901 }
2902
2903 // Null check receiver.
2904 receiver = null_check(receiver);
2905 if (stopped()) {
2906 return true;
2907 }
2908
2909 int alias_idx = C->get_alias_index(adr_type);
2910
2911 if (is_reference_type(type)) {
2912 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2913
2914 if (oldval != nullptr && oldval->is_InlineType()) {
2915 // Re-execute the unsafe access if allocation triggers deoptimization.
2916 PreserveReexecuteState preexecs(this);
2917 jvms()->set_should_reexecute(true);
2918 oldval = oldval->as_InlineType()->buffer(this)->get_oop();
2919 }
2920 if (newval != nullptr && newval->is_InlineType()) {
2921 // Re-execute the unsafe access if allocation triggers deoptimization.
2922 PreserveReexecuteState preexecs(this);
2923 jvms()->set_should_reexecute(true);
2924 newval = newval->as_InlineType()->buffer(this)->get_oop();
2925 }
2926
2927 // Transformation of a value which could be null pointer (CastPP #null)
2928 // could be delayed during Parse (for example, in adjust_map_after_if()).
2929 // Execute transformation here to avoid barrier generation in such case.
2930 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2931 newval = _gvn.makecon(TypePtr::NULL_PTR);
2932
2933 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2934 // Refine the value to a null constant, when it is known to be null
2935 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2936 }
2937 }
2938
2939 Node* result = nullptr;
2940 switch (kind) {
2941 case LS_cmp_exchange: {
2942 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2943 oldval, newval, value_type, type, decorators);
2944 break;
2945 }
2946 case LS_cmp_swap_weak:
2947 decorators |= C2_WEAK_CMPXCHG;
2948 case LS_cmp_swap: {
2949 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2950 oldval, newval, value_type, type, decorators);
2951 break;
2952 }
2953 case LS_get_set: {
2954 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2955 newval, value_type, type, decorators);
2956 break;
2957 }
2958 case LS_get_add: {
2959 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2960 newval, value_type, type, decorators);
2961 break;
2962 }
2963 default:
2964 ShouldNotReachHere();
2965 }
2966
2967 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2968 set_result(result);
2969 return true;
2970 }
2971
2972 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2973 // Regardless of form, don't allow previous ld/st to move down,
2974 // then issue acquire, release, or volatile mem_bar.
2975 insert_mem_bar(Op_MemBarCPUOrder);
2976 switch(id) {
2977 case vmIntrinsics::_loadFence:
2978 insert_mem_bar(Op_LoadFence);
2979 return true;
2980 case vmIntrinsics::_storeFence:
2981 insert_mem_bar(Op_StoreFence);
2982 return true;
2983 case vmIntrinsics::_storeStoreFence:
2984 insert_mem_bar(Op_StoreStoreFence);
2985 return true;
2986 case vmIntrinsics::_fullFence:
2987 insert_mem_bar(Op_MemBarVolatile);
2988 return true;
2989 default:
2990 fatal_unexpected_iid(id);
2991 return false;
2992 }
2993 }
2994
2995 bool LibraryCallKit::inline_onspinwait() {
2996 insert_mem_bar(Op_OnSpinWait);
2997 return true;
2998 }
2999
3000 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3001 if (!kls->is_Con()) {
3002 return true;
3003 }
3004 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
3005 if (klsptr == nullptr) {
3006 return true;
3007 }
3008 ciInstanceKlass* ik = klsptr->instance_klass();
3009 // don't need a guard for a klass that is already initialized
3010 return !ik->is_initialized();
3011 }
3012
3013 //----------------------------inline_unsafe_writeback0-------------------------
3014 // public native void Unsafe.writeback0(long address)
3015 bool LibraryCallKit::inline_unsafe_writeback0() {
3016 if (!Matcher::has_match_rule(Op_CacheWB)) {
3017 return false;
3018 }
3019 #ifndef PRODUCT
3020 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
3021 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
3022 ciSignature* sig = callee()->signature();
3023 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
3024 #endif
3025 null_check_receiver(); // null-check, then ignore
3026 Node *addr = argument(1);
3027 addr = new CastX2PNode(addr);
3028 addr = _gvn.transform(addr);
3029 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
3030 flush = _gvn.transform(flush);
3031 set_memory(flush, TypeRawPtr::BOTTOM);
3032 return true;
3033 }
3034
3035 //----------------------------inline_unsafe_writeback0-------------------------
3036 // public native void Unsafe.writeback0(long address)
3037 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
3038 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
3039 return false;
3040 }
3041 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
3042 return false;
3043 }
3044 #ifndef PRODUCT
3045 assert(Matcher::has_match_rule(Op_CacheWB),
3046 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
3047 : "found match rule for CacheWBPostSync but not CacheWB"));
3048
3049 #endif
3050 null_check_receiver(); // null-check, then ignore
3051 Node *sync;
3052 if (is_pre) {
3053 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3054 } else {
3055 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3056 }
3057 sync = _gvn.transform(sync);
3058 set_memory(sync, TypeRawPtr::BOTTOM);
3059 return true;
3060 }
3061
3062 //----------------------------inline_unsafe_allocate---------------------------
3063 // public native Object Unsafe.allocateInstance(Class<?> cls);
3064 bool LibraryCallKit::inline_unsafe_allocate() {
3065
3066 #if INCLUDE_JVMTI
3067 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3068 return false;
3069 }
3070 #endif //INCLUDE_JVMTI
3071
3072 if (callee()->is_static()) return false; // caller must have the capability!
3073
3074 null_check_receiver(); // null-check, then ignore
3075 Node* cls = null_check(argument(1));
3076 if (stopped()) return true;
3077
3078 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
3079 kls = null_check(kls);
3080 if (stopped()) return true; // argument was like int.class
3081
3082 #if INCLUDE_JVMTI
3083 // Don't try to access new allocated obj in the intrinsic.
3084 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
3085 // Deoptimize and allocate in interpreter instead.
3086 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
3087 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
3088 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
3089 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
3090 {
3091 BuildCutout unless(this, tst, PROB_MAX);
3092 uncommon_trap(Deoptimization::Reason_intrinsic,
3093 Deoptimization::Action_make_not_entrant);
3094 }
3095 if (stopped()) {
3096 return true;
3097 }
3098 #endif //INCLUDE_JVMTI
3099
3100 Node* test = nullptr;
3101 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3102 // Note: The argument might still be an illegal value like
3103 // Serializable.class or Object[].class. The runtime will handle it.
3104 // But we must make an explicit check for initialization.
3105 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3106 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3107 // can generate code to load it as unsigned byte.
3108 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
3109 Node* bits = intcon(InstanceKlass::fully_initialized);
3110 test = _gvn.transform(new SubINode(inst, bits));
3111 // The 'test' is non-zero if we need to take a slow path.
3112 }
3113 Node* obj = nullptr;
3114 const TypeInstKlassPtr* tkls = _gvn.type(kls)->isa_instklassptr();
3115 if (tkls != nullptr && tkls->instance_klass()->is_inlinetype()) {
3116 obj = InlineTypeNode::make_default(_gvn, tkls->instance_klass()->as_inline_klass())->buffer(this);
3117 } else {
3118 obj = new_instance(kls, test);
3119 }
3120 set_result(obj);
3121 return true;
3122 }
3123
3124 //------------------------inline_native_time_funcs--------------
3125 // inline code for System.currentTimeMillis() and System.nanoTime()
3126 // these have the same type and signature
3127 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3128 const TypeFunc* tf = OptoRuntime::void_long_Type();
3129 const TypePtr* no_memory_effects = nullptr;
3130 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3131 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3132 #ifdef ASSERT
3133 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3134 assert(value_top == top(), "second value must be top");
3135 #endif
3136 set_result(value);
3137 return true;
3138 }
3139
3140
3141 #if INCLUDE_JVMTI
3142
3143 // When notifications are disabled then just update the VTMS transition bit and return.
3144 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
3145 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
3146 if (!DoJVMTIVirtualThreadTransitions) {
3147 return true;
3148 }
3149 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3150 IdealKit ideal(this);
3151
3152 Node* ONE = ideal.ConI(1);
3153 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
3154 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
3155 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3156
3157 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
3158 sync_kit(ideal);
3159 // if notifyJvmti enabled then make a call to the given SharedRuntime function
3160 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
3161 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
3162 ideal.sync_kit(this);
3163 } ideal.else_(); {
3164 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
3165 Node* thread = ideal.thread();
3166 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
3167 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
3168
3169 sync_kit(ideal);
3170 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3171 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3172
3173 ideal.sync_kit(this);
3174 } ideal.end_if();
3175 final_sync(ideal);
3176
3177 return true;
3178 }
3179
3180 // Always update the is_disable_suspend bit.
3181 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3182 if (!DoJVMTIVirtualThreadTransitions) {
3183 return true;
3184 }
3185 IdealKit ideal(this);
3186
3187 {
3188 // unconditionally update the is_disable_suspend bit in current JavaThread
3189 Node* thread = ideal.thread();
3190 Node* arg = _gvn.transform(argument(0)); // argument for notification
3191 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3192 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3193
3194 sync_kit(ideal);
3195 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3196 ideal.sync_kit(this);
3197 }
3198 final_sync(ideal);
3199
3200 return true;
3201 }
3202
3203 #endif // INCLUDE_JVMTI
3204
3205 #ifdef JFR_HAVE_INTRINSICS
3206
3207 /**
3208 * if oop->klass != null
3209 * // normal class
3210 * epoch = _epoch_state ? 2 : 1
3211 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3212 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3213 * }
3214 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3215 * else
3216 * // primitive class
3217 * if oop->array_klass != null
3218 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3219 * else
3220 * id = LAST_TYPE_ID + 1 // void class path
3221 * if (!signaled)
3222 * signaled = true
3223 */
3224 bool LibraryCallKit::inline_native_classID() {
3225 Node* cls = argument(0);
3226
3227 IdealKit ideal(this);
3228 #define __ ideal.
3229 IdealVariable result(ideal); __ declarations_done();
3230 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3231 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3232 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3233
3234
3235 __ if_then(kls, BoolTest::ne, null()); {
3236 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3237 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3238
3239 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3240 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3241 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3242 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3243 mask = _gvn.transform(new OrLNode(mask, epoch));
3244 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3245
3246 float unlikely = PROB_UNLIKELY(0.999);
3247 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3248 sync_kit(ideal);
3249 make_runtime_call(RC_LEAF,
3250 OptoRuntime::class_id_load_barrier_Type(),
3251 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3252 "class id load barrier",
3253 TypePtr::BOTTOM,
3254 kls);
3255 ideal.sync_kit(this);
3256 } __ end_if();
3257
3258 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3259 } __ else_(); {
3260 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3261 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3262 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3263 __ if_then(array_kls, BoolTest::ne, null()); {
3264 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3265 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3266 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3267 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3268 } __ else_(); {
3269 // void class case
3270 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3271 } __ end_if();
3272
3273 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3274 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3275 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3276 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3277 } __ end_if();
3278 } __ end_if();
3279
3280 final_sync(ideal);
3281 set_result(ideal.value(result));
3282 #undef __
3283 return true;
3284 }
3285
3286 //------------------------inline_native_jvm_commit------------------
3287 bool LibraryCallKit::inline_native_jvm_commit() {
3288 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3289
3290 // Save input memory and i_o state.
3291 Node* input_memory_state = reset_memory();
3292 set_all_memory(input_memory_state);
3293 Node* input_io_state = i_o();
3294
3295 // TLS.
3296 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3297 // Jfr java buffer.
3298 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3299 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3300 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3301
3302 // Load the current value of the notified field in the JfrThreadLocal.
3303 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3304 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3305
3306 // Test for notification.
3307 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3308 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3309 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3310
3311 // True branch, is notified.
3312 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3313 set_control(is_notified);
3314
3315 // Reset notified state.
3316 Node* notified_reset_memory = store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered);
3317
3318 // 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.
3319 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3320 // Convert the machine-word to a long.
3321 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3322
3323 // False branch, not notified.
3324 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3325 set_control(not_notified);
3326 set_all_memory(input_memory_state);
3327
3328 // Arg is the next position as a long.
3329 Node* arg = argument(0);
3330 // Convert long to machine-word.
3331 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3332
3333 // Store the next_position to the underlying jfr java buffer.
3334 Node* commit_memory;
3335 #ifdef _LP64
3336 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_LONG, Compile::AliasIdxRaw, MemNode::release);
3337 #else
3338 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_INT, Compile::AliasIdxRaw, MemNode::release);
3339 #endif
3340
3341 // 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.
3342 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3343 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3344 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3345
3346 // And flags with lease constant.
3347 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3348
3349 // Branch on lease to conditionalize returning the leased java buffer.
3350 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3351 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3352 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3353
3354 // False branch, not a lease.
3355 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3356
3357 // True branch, is lease.
3358 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3359 set_control(is_lease);
3360
3361 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3362 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3363 OptoRuntime::void_void_Type(),
3364 SharedRuntime::jfr_return_lease(),
3365 "return_lease", TypePtr::BOTTOM);
3366 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3367
3368 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3369 record_for_igvn(lease_compare_rgn);
3370 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3371 record_for_igvn(lease_compare_mem);
3372 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3373 record_for_igvn(lease_compare_io);
3374 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3375 record_for_igvn(lease_result_value);
3376
3377 // Update control and phi nodes.
3378 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3379 lease_compare_rgn->init_req(_false_path, not_lease);
3380
3381 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3382 lease_compare_mem->init_req(_false_path, commit_memory);
3383
3384 lease_compare_io->init_req(_true_path, i_o());
3385 lease_compare_io->init_req(_false_path, input_io_state);
3386
3387 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3388 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3389
3390 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3391 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3392 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3393 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3394
3395 // Update control and phi nodes.
3396 result_rgn->init_req(_true_path, is_notified);
3397 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3398
3399 result_mem->init_req(_true_path, notified_reset_memory);
3400 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3401
3402 result_io->init_req(_true_path, input_io_state);
3403 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3404
3405 result_value->init_req(_true_path, current_pos);
3406 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3407
3408 // Set output state.
3409 set_control(_gvn.transform(result_rgn));
3410 set_all_memory(_gvn.transform(result_mem));
3411 set_i_o(_gvn.transform(result_io));
3412 set_result(result_rgn, result_value);
3413 return true;
3414 }
3415
3416 /*
3417 * The intrinsic is a model of this pseudo-code:
3418 *
3419 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3420 * jobject h_event_writer = tl->java_event_writer();
3421 * if (h_event_writer == nullptr) {
3422 * return nullptr;
3423 * }
3424 * oop threadObj = Thread::threadObj();
3425 * oop vthread = java_lang_Thread::vthread(threadObj);
3426 * traceid tid;
3427 * bool pinVirtualThread;
3428 * bool excluded;
3429 * if (vthread != threadObj) { // i.e. current thread is virtual
3430 * tid = java_lang_Thread::tid(vthread);
3431 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3432 * pinVirtualThread = VMContinuations;
3433 * excluded = vthread_epoch_raw & excluded_mask;
3434 * if (!excluded) {
3435 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3436 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3437 * if (vthread_epoch != current_epoch) {
3438 * write_checkpoint();
3439 * }
3440 * }
3441 * } else {
3442 * tid = java_lang_Thread::tid(threadObj);
3443 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3444 * pinVirtualThread = false;
3445 * excluded = thread_epoch_raw & excluded_mask;
3446 * }
3447 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3448 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3449 * if (tid_in_event_writer != tid) {
3450 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3451 * setField(event_writer, "excluded", excluded);
3452 * setField(event_writer, "threadID", tid);
3453 * }
3454 * return event_writer
3455 */
3456 bool LibraryCallKit::inline_native_getEventWriter() {
3457 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3458
3459 // Save input memory and i_o state.
3460 Node* input_memory_state = reset_memory();
3461 set_all_memory(input_memory_state);
3462 Node* input_io_state = i_o();
3463
3464 Node* excluded_mask = _gvn.intcon(32768);
3465 Node* epoch_mask = _gvn.intcon(32767);
3466
3467 // TLS
3468 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3469
3470 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3471 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3472
3473 // Load the eventwriter jobject handle.
3474 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3475
3476 // Null check the jobject handle.
3477 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3478 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3479 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3480
3481 // False path, jobj is null.
3482 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3483
3484 // True path, jobj is not null.
3485 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3486
3487 set_control(jobj_is_not_null);
3488
3489 // Load the threadObj for the CarrierThread.
3490 Node* threadObj = generate_current_thread(tls_ptr);
3491
3492 // Load the vthread.
3493 Node* vthread = generate_virtual_thread(tls_ptr);
3494
3495 // If vthread != threadObj, this is a virtual thread.
3496 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3497 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3498 IfNode* iff_vthread_not_equal_threadObj =
3499 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3500
3501 // False branch, fallback to threadObj.
3502 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3503 set_control(vthread_equal_threadObj);
3504
3505 // Load the tid field from the vthread object.
3506 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3507
3508 // Load the raw epoch value from the threadObj.
3509 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3510 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3511 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3512 TypeInt::CHAR, T_CHAR,
3513 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3514
3515 // Mask off the excluded information from the epoch.
3516 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3517
3518 // True branch, this is a virtual thread.
3519 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3520 set_control(vthread_not_equal_threadObj);
3521
3522 // Load the tid field from the vthread object.
3523 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3524
3525 // Continuation support determines if a virtual thread should be pinned.
3526 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3527 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3528
3529 // Load the raw epoch value from the vthread.
3530 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3531 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3532 TypeInt::CHAR, T_CHAR,
3533 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3534
3535 // Mask off the excluded information from the epoch.
3536 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3537
3538 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3539 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3540 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3541 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3542
3543 // False branch, vthread is excluded, no need to write epoch info.
3544 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3545
3546 // True branch, vthread is included, update epoch info.
3547 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3548 set_control(included);
3549
3550 // Get epoch value.
3551 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3552
3553 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3554 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3555 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3556
3557 // Compare the epoch in the vthread to the current epoch generation.
3558 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3559 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3560 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3561
3562 // False path, epoch is equal, checkpoint information is valid.
3563 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3564
3565 // True path, epoch is not equal, write a checkpoint for the vthread.
3566 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3567
3568 set_control(epoch_is_not_equal);
3569
3570 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3571 // The call also updates the native thread local thread id and the vthread with the current epoch.
3572 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3573 OptoRuntime::jfr_write_checkpoint_Type(),
3574 SharedRuntime::jfr_write_checkpoint(),
3575 "write_checkpoint", TypePtr::BOTTOM);
3576 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3577
3578 // vthread epoch != current epoch
3579 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3580 record_for_igvn(epoch_compare_rgn);
3581 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3582 record_for_igvn(epoch_compare_mem);
3583 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3584 record_for_igvn(epoch_compare_io);
3585
3586 // Update control and phi nodes.
3587 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3588 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3589 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3590 epoch_compare_mem->init_req(_false_path, input_memory_state);
3591 epoch_compare_io->init_req(_true_path, i_o());
3592 epoch_compare_io->init_req(_false_path, input_io_state);
3593
3594 // excluded != true
3595 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3596 record_for_igvn(exclude_compare_rgn);
3597 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3598 record_for_igvn(exclude_compare_mem);
3599 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3600 record_for_igvn(exclude_compare_io);
3601
3602 // Update control and phi nodes.
3603 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3604 exclude_compare_rgn->init_req(_false_path, excluded);
3605 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3606 exclude_compare_mem->init_req(_false_path, input_memory_state);
3607 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3608 exclude_compare_io->init_req(_false_path, input_io_state);
3609
3610 // vthread != threadObj
3611 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3612 record_for_igvn(vthread_compare_rgn);
3613 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3614 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3615 record_for_igvn(vthread_compare_io);
3616 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3617 record_for_igvn(tid);
3618 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3619 record_for_igvn(exclusion);
3620 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3621 record_for_igvn(pinVirtualThread);
3622
3623 // Update control and phi nodes.
3624 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3625 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3626 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3627 vthread_compare_mem->init_req(_false_path, input_memory_state);
3628 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3629 vthread_compare_io->init_req(_false_path, input_io_state);
3630 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3631 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3632 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3633 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3634 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3635 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3636
3637 // Update branch state.
3638 set_control(_gvn.transform(vthread_compare_rgn));
3639 set_all_memory(_gvn.transform(vthread_compare_mem));
3640 set_i_o(_gvn.transform(vthread_compare_io));
3641
3642 // Load the event writer oop by dereferencing the jobject handle.
3643 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3644 assert(klass_EventWriter->is_loaded(), "invariant");
3645 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3646 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3647 const TypeOopPtr* const xtype = aklass->as_instance_type();
3648 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3649 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3650
3651 // Load the current thread id from the event writer object.
3652 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3653 // Get the field offset to, conditionally, store an updated tid value later.
3654 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3655 const TypePtr* event_writer_tid_field_type = _gvn.type(event_writer_tid_field)->isa_ptr();
3656 // Get the field offset to, conditionally, store an updated exclusion value later.
3657 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3658 const TypePtr* event_writer_excluded_field_type = _gvn.type(event_writer_excluded_field)->isa_ptr();
3659 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3660 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3661 const TypePtr* event_writer_pin_field_type = _gvn.type(event_writer_pin_field)->isa_ptr();
3662
3663 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3664 record_for_igvn(event_writer_tid_compare_rgn);
3665 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3666 record_for_igvn(event_writer_tid_compare_mem);
3667 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3668 record_for_igvn(event_writer_tid_compare_io);
3669
3670 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3671 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3672 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3673 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3674
3675 // False path, tids are the same.
3676 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3677
3678 // True path, tid is not equal, need to update the tid in the event writer.
3679 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3680 record_for_igvn(tid_is_not_equal);
3681
3682 // Store the pin state to the event writer.
3683 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, event_writer_pin_field_type, MemNode::unordered);
3684
3685 // Store the exclusion state to the event writer.
3686 store_to_memory(tid_is_not_equal, event_writer_excluded_field, _gvn.transform(exclusion), T_BOOLEAN, event_writer_excluded_field_type, MemNode::unordered);
3687
3688 // Store the tid to the event writer.
3689 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, event_writer_tid_field_type, MemNode::unordered);
3690
3691 // Update control and phi nodes.
3692 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3693 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3694 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3695 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3696 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3697 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3698
3699 // Result of top level CFG, Memory, IO and Value.
3700 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3701 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3702 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3703 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3704
3705 // Result control.
3706 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3707 result_rgn->init_req(_false_path, jobj_is_null);
3708
3709 // Result memory.
3710 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3711 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3712
3713 // Result IO.
3714 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3715 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3716
3717 // Result value.
3718 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3719 result_value->init_req(_false_path, null()); // return null
3720
3721 // Set output state.
3722 set_control(_gvn.transform(result_rgn));
3723 set_all_memory(_gvn.transform(result_mem));
3724 set_i_o(_gvn.transform(result_io));
3725 set_result(result_rgn, result_value);
3726 return true;
3727 }
3728
3729 /*
3730 * The intrinsic is a model of this pseudo-code:
3731 *
3732 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3733 * if (carrierThread != thread) { // is virtual thread
3734 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3735 * bool excluded = vthread_epoch_raw & excluded_mask;
3736 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3737 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3738 * if (!excluded) {
3739 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3740 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3741 * }
3742 * Atomic::release_store(&tl->_vthread, true);
3743 * return;
3744 * }
3745 * Atomic::release_store(&tl->_vthread, false);
3746 */
3747 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3748 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3749
3750 Node* input_memory_state = reset_memory();
3751 set_all_memory(input_memory_state);
3752
3753 Node* excluded_mask = _gvn.intcon(32768);
3754 Node* epoch_mask = _gvn.intcon(32767);
3755
3756 Node* const carrierThread = generate_current_thread(jt);
3757 // If thread != carrierThread, this is a virtual thread.
3758 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3759 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3760 IfNode* iff_thread_not_equal_carrierThread =
3761 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3762
3763 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3764
3765 // False branch, is carrierThread.
3766 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3767 // Store release
3768 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3769
3770 set_all_memory(input_memory_state);
3771
3772 // True branch, is virtual thread.
3773 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3774 set_control(thread_not_equal_carrierThread);
3775
3776 // Load the raw epoch value from the vthread.
3777 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3778 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
3779 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3780
3781 // Mask off the excluded information from the epoch.
3782 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3783
3784 // Load the tid field from the thread.
3785 Node* tid = load_field_from_object(thread, "tid", "J");
3786
3787 // Store the vthread tid to the jfr thread local.
3788 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3789 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, Compile::AliasIdxRaw, MemNode::unordered, true);
3790
3791 // Branch is_excluded to conditionalize updating the epoch .
3792 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3793 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3794 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3795
3796 // True branch, vthread is excluded, no need to write epoch info.
3797 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3798 set_control(excluded);
3799 Node* vthread_is_excluded = _gvn.intcon(1);
3800
3801 // False branch, vthread is included, update epoch info.
3802 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3803 set_control(included);
3804 Node* vthread_is_included = _gvn.intcon(0);
3805
3806 // Get epoch value.
3807 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3808
3809 // Store the vthread epoch to the jfr thread local.
3810 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3811 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, Compile::AliasIdxRaw, MemNode::unordered, true);
3812
3813 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3814 record_for_igvn(excluded_rgn);
3815 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3816 record_for_igvn(excluded_mem);
3817 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3818 record_for_igvn(exclusion);
3819
3820 // Merge the excluded control and memory.
3821 excluded_rgn->init_req(_true_path, excluded);
3822 excluded_rgn->init_req(_false_path, included);
3823 excluded_mem->init_req(_true_path, tid_memory);
3824 excluded_mem->init_req(_false_path, included_memory);
3825 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3826 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3827
3828 // Set intermediate state.
3829 set_control(_gvn.transform(excluded_rgn));
3830 set_all_memory(excluded_mem);
3831
3832 // Store the vthread exclusion state to the jfr thread local.
3833 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3834 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered, true);
3835
3836 // Store release
3837 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3838
3839 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3840 record_for_igvn(thread_compare_rgn);
3841 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3842 record_for_igvn(thread_compare_mem);
3843 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3844 record_for_igvn(vthread);
3845
3846 // Merge the thread_compare control and memory.
3847 thread_compare_rgn->init_req(_true_path, control());
3848 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3849 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3850 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3851
3852 // Set output state.
3853 set_control(_gvn.transform(thread_compare_rgn));
3854 set_all_memory(_gvn.transform(thread_compare_mem));
3855 }
3856
3857 #endif // JFR_HAVE_INTRINSICS
3858
3859 //------------------------inline_native_currentCarrierThread------------------
3860 bool LibraryCallKit::inline_native_currentCarrierThread() {
3861 Node* junk = nullptr;
3862 set_result(generate_current_thread(junk));
3863 return true;
3864 }
3865
3866 //------------------------inline_native_currentThread------------------
3867 bool LibraryCallKit::inline_native_currentThread() {
3868 Node* junk = nullptr;
3869 set_result(generate_virtual_thread(junk));
3870 return true;
3871 }
3872
3873 //------------------------inline_native_setVthread------------------
3874 bool LibraryCallKit::inline_native_setCurrentThread() {
3875 assert(C->method()->changes_current_thread(),
3876 "method changes current Thread but is not annotated ChangesCurrentThread");
3877 Node* arr = argument(1);
3878 Node* thread = _gvn.transform(new ThreadLocalNode());
3879 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3880 Node* thread_obj_handle
3881 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3882 thread_obj_handle = _gvn.transform(thread_obj_handle);
3883 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3884 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3885 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3886 return true;
3887 }
3888
3889 const Type* LibraryCallKit::scopedValueCache_type() {
3890 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3891 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3892 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS, /* stable= */ false, /* flat= */ false, /* not_flat= */ true, /* not_null_free= */ true);
3893
3894 // Because we create the scopedValue cache lazily we have to make the
3895 // type of the result BotPTR.
3896 bool xk = etype->klass_is_exact();
3897 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, TypeAryPtr::Offset(0));
3898 return objects_type;
3899 }
3900
3901 Node* LibraryCallKit::scopedValueCache_helper() {
3902 Node* thread = _gvn.transform(new ThreadLocalNode());
3903 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3904 // We cannot use immutable_memory() because we might flip onto a
3905 // different carrier thread, at which point we'll need to use that
3906 // carrier thread's cache.
3907 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3908 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3909 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3910 }
3911
3912 //------------------------inline_native_scopedValueCache------------------
3913 bool LibraryCallKit::inline_native_scopedValueCache() {
3914 Node* cache_obj_handle = scopedValueCache_helper();
3915 const Type* objects_type = scopedValueCache_type();
3916 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3917
3918 return true;
3919 }
3920
3921 //------------------------inline_native_setScopedValueCache------------------
3922 bool LibraryCallKit::inline_native_setScopedValueCache() {
3923 Node* arr = argument(0);
3924 Node* cache_obj_handle = scopedValueCache_helper();
3925 const Type* objects_type = scopedValueCache_type();
3926
3927 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3928 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3929
3930 return true;
3931 }
3932
3933 //------------------------inline_native_Continuation_pin and unpin-----------
3934
3935 // Shared implementation routine for both pin and unpin.
3936 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
3937 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3938
3939 // Save input memory.
3940 Node* input_memory_state = reset_memory();
3941 set_all_memory(input_memory_state);
3942
3943 // TLS
3944 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3945 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
3946 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
3947
3948 // Null check the last continuation object.
3949 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
3950 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
3951 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3952
3953 // False path, last continuation is null.
3954 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
3955
3956 // True path, last continuation is not null.
3957 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
3958
3959 set_control(continuation_is_not_null);
3960
3961 // Load the pin count from the last continuation.
3962 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
3963 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
3964
3965 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
3966 Node* pin_count_rhs;
3967 if (unpin) {
3968 pin_count_rhs = _gvn.intcon(0);
3969 } else {
3970 pin_count_rhs = _gvn.intcon(UINT32_MAX);
3971 }
3972 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
3973 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
3974 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
3975
3976 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
3977 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
3978 set_control(valid_pin_count);
3979
3980 Node* next_pin_count;
3981 if (unpin) {
3982 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
3983 } else {
3984 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
3985 }
3986
3987 Node* updated_pin_count_memory = store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, Compile::AliasIdxRaw, MemNode::unordered);
3988
3989 // True branch, pin count over/underflow.
3990 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
3991 {
3992 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
3993 // which will throw IllegalStateException for pin count over/underflow.
3994 PreserveJVMState pjvms(this);
3995 set_control(pin_count_over_underflow);
3996 set_all_memory(input_memory_state);
3997 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
3998 Deoptimization::Action_none);
3999 assert(stopped(), "invariant");
4000 }
4001
4002 // Result of top level CFG and Memory.
4003 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
4004 record_for_igvn(result_rgn);
4005 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
4006 record_for_igvn(result_mem);
4007
4008 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
4009 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
4010 result_mem->init_req(_true_path, _gvn.transform(updated_pin_count_memory));
4011 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
4012
4013 // Set output state.
4014 set_control(_gvn.transform(result_rgn));
4015 set_all_memory(_gvn.transform(result_mem));
4016
4017 return true;
4018 }
4019
4020 //-----------------------load_klass_from_mirror_common-------------------------
4021 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
4022 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
4023 // and branch to the given path on the region.
4024 // If never_see_null, take an uncommon trap on null, so we can optimistically
4025 // compile for the non-null case.
4026 // If the region is null, force never_see_null = true.
4027 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
4028 bool never_see_null,
4029 RegionNode* region,
4030 int null_path,
4031 int offset) {
4032 if (region == nullptr) never_see_null = true;
4033 Node* p = basic_plus_adr(mirror, offset);
4034 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4035 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
4036 Node* null_ctl = top();
4037 kls = null_check_oop(kls, &null_ctl, never_see_null);
4038 if (region != nullptr) {
4039 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
4040 region->init_req(null_path, null_ctl);
4041 } else {
4042 assert(null_ctl == top(), "no loose ends");
4043 }
4044 return kls;
4045 }
4046
4047 //--------------------(inline_native_Class_query helpers)---------------------
4048 // Use this for JVM_ACC_INTERFACE.
4049 // Fall through if (mods & mask) == bits, take the guard otherwise.
4050 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
4051 ByteSize offset, const Type* type, BasicType bt) {
4052 // Branch around if the given klass has the given modifier bit set.
4053 // Like generate_guard, adds a new path onto the region.
4054 Node* modp = basic_plus_adr(kls, in_bytes(offset));
4055 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
4056 Node* mask = intcon(modifier_mask);
4057 Node* bits = intcon(modifier_bits);
4058 Node* mbit = _gvn.transform(new AndINode(mods, mask));
4059 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
4060 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
4061 return generate_fair_guard(bol, region);
4062 }
4063
4064 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
4065 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
4066 Klass::access_flags_offset(), TypeInt::INT, T_INT);
4067 }
4068
4069 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
4070 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
4071 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
4072 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
4073 }
4074
4075 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
4076 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
4077 }
4078
4079 //-------------------------inline_native_Class_query-------------------
4080 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
4081 const Type* return_type = TypeInt::BOOL;
4082 Node* prim_return_value = top(); // what happens if it's a primitive class?
4083 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4084 bool expect_prim = false; // most of these guys expect to work on refs
4085
4086 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
4087
4088 Node* mirror = argument(0);
4089 Node* obj = top();
4090
4091 switch (id) {
4092 case vmIntrinsics::_isInstance:
4093 // nothing is an instance of a primitive type
4094 prim_return_value = intcon(0);
4095 obj = argument(1);
4096 break;
4097 case vmIntrinsics::_getModifiers:
4098 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
4099 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
4100 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
4101 break;
4102 case vmIntrinsics::_isInterface:
4103 prim_return_value = intcon(0);
4104 break;
4105 case vmIntrinsics::_isArray:
4106 prim_return_value = intcon(0);
4107 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
4108 break;
4109 case vmIntrinsics::_isPrimitive:
4110 prim_return_value = intcon(1);
4111 expect_prim = true; // obviously
4112 break;
4113 case vmIntrinsics::_isHidden:
4114 prim_return_value = intcon(0);
4115 break;
4116 case vmIntrinsics::_getSuperclass:
4117 prim_return_value = null();
4118 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
4119 break;
4120 case vmIntrinsics::_getClassAccessFlags:
4121 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
4122 return_type = TypeInt::INT; // not bool! 6297094
4123 break;
4124 default:
4125 fatal_unexpected_iid(id);
4126 break;
4127 }
4128
4129 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4130 if (mirror_con == nullptr) return false; // cannot happen?
4131
4132 #ifndef PRODUCT
4133 if (C->print_intrinsics() || C->print_inlining()) {
4134 ciType* k = mirror_con->java_mirror_type();
4135 if (k) {
4136 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4137 k->print_name();
4138 tty->cr();
4139 }
4140 }
4141 #endif
4142
4143 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4144 RegionNode* region = new RegionNode(PATH_LIMIT);
4145 record_for_igvn(region);
4146 PhiNode* phi = new PhiNode(region, return_type);
4147
4148 // The mirror will never be null of Reflection.getClassAccessFlags, however
4149 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4150 // if it is. See bug 4774291.
4151
4152 // For Reflection.getClassAccessFlags(), the null check occurs in
4153 // the wrong place; see inline_unsafe_access(), above, for a similar
4154 // situation.
4155 mirror = null_check(mirror);
4156 // If mirror or obj is dead, only null-path is taken.
4157 if (stopped()) return true;
4158
4159 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4160
4161 // Now load the mirror's klass metaobject, and null-check it.
4162 // Side-effects region with the control path if the klass is null.
4163 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4164 // If kls is null, we have a primitive mirror.
4165 phi->init_req(_prim_path, prim_return_value);
4166 if (stopped()) { set_result(region, phi); return true; }
4167 bool safe_for_replace = (region->in(_prim_path) == top());
4168
4169 Node* p; // handy temp
4170 Node* null_ctl;
4171
4172 // Now that we have the non-null klass, we can perform the real query.
4173 // For constant classes, the query will constant-fold in LoadNode::Value.
4174 Node* query_value = top();
4175 switch (id) {
4176 case vmIntrinsics::_isInstance:
4177 // nothing is an instance of a primitive type
4178 query_value = gen_instanceof(obj, kls, safe_for_replace);
4179 break;
4180
4181 case vmIntrinsics::_getModifiers:
4182 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
4183 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
4184 break;
4185
4186 case vmIntrinsics::_isInterface:
4187 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4188 if (generate_interface_guard(kls, region) != nullptr)
4189 // A guard was added. If the guard is taken, it was an interface.
4190 phi->add_req(intcon(1));
4191 // If we fall through, it's a plain class.
4192 query_value = intcon(0);
4193 break;
4194
4195 case vmIntrinsics::_isArray:
4196 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
4197 if (generate_array_guard(kls, region) != nullptr)
4198 // A guard was added. If the guard is taken, it was an array.
4199 phi->add_req(intcon(1));
4200 // If we fall through, it's a plain class.
4201 query_value = intcon(0);
4202 break;
4203
4204 case vmIntrinsics::_isPrimitive:
4205 query_value = intcon(0); // "normal" path produces false
4206 break;
4207
4208 case vmIntrinsics::_isHidden:
4209 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4210 if (generate_hidden_class_guard(kls, region) != nullptr)
4211 // A guard was added. If the guard is taken, it was an hidden class.
4212 phi->add_req(intcon(1));
4213 // If we fall through, it's a plain class.
4214 query_value = intcon(0);
4215 break;
4216
4217
4218 case vmIntrinsics::_getSuperclass:
4219 // The rules here are somewhat unfortunate, but we can still do better
4220 // with random logic than with a JNI call.
4221 // Interfaces store null or Object as _super, but must report null.
4222 // Arrays store an intermediate super as _super, but must report Object.
4223 // Other types can report the actual _super.
4224 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4225 if (generate_interface_guard(kls, region) != nullptr)
4226 // A guard was added. If the guard is taken, it was an interface.
4227 phi->add_req(null());
4228 if (generate_array_guard(kls, region) != nullptr)
4229 // A guard was added. If the guard is taken, it was an array.
4230 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4231 // If we fall through, it's a plain class. Get its _super.
4232 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4233 kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4234 null_ctl = top();
4235 kls = null_check_oop(kls, &null_ctl);
4236 if (null_ctl != top()) {
4237 // If the guard is taken, Object.superClass is null (both klass and mirror).
4238 region->add_req(null_ctl);
4239 phi ->add_req(null());
4240 }
4241 if (!stopped()) {
4242 query_value = load_mirror_from_klass(kls);
4243 }
4244 break;
4245
4246 case vmIntrinsics::_getClassAccessFlags:
4247 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
4248 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
4249 break;
4250
4251 default:
4252 fatal_unexpected_iid(id);
4253 break;
4254 }
4255
4256 // Fall-through is the normal case of a query to a real class.
4257 phi->init_req(1, query_value);
4258 region->init_req(1, control());
4259
4260 C->set_has_split_ifs(true); // Has chance for split-if optimization
4261 set_result(region, phi);
4262 return true;
4263 }
4264
4265
4266 //-------------------------inline_Class_cast-------------------
4267 bool LibraryCallKit::inline_Class_cast() {
4268 Node* mirror = argument(0); // Class
4269 Node* obj = argument(1);
4270 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4271 if (mirror_con == nullptr) {
4272 return false; // dead path (mirror->is_top()).
4273 }
4274 if (obj == nullptr || obj->is_top()) {
4275 return false; // dead path
4276 }
4277 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4278
4279 // First, see if Class.cast() can be folded statically.
4280 // java_mirror_type() returns non-null for compile-time Class constants.
4281 bool is_null_free_array = false;
4282 ciType* tm = mirror_con->java_mirror_type(&is_null_free_array);
4283 if (tm != nullptr && tm->is_klass() &&
4284 tp != nullptr) {
4285 if (!tp->is_loaded()) {
4286 // Don't use intrinsic when class is not loaded.
4287 return false;
4288 } else {
4289 const TypeKlassPtr* tklass = TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces);
4290 if (is_null_free_array) {
4291 tklass = tklass->is_aryklassptr()->cast_to_null_free();
4292 }
4293 int static_res = C->static_subtype_check(tklass, tp->as_klass_type());
4294 if (static_res == Compile::SSC_always_true) {
4295 // isInstance() is true - fold the code.
4296 set_result(obj);
4297 return true;
4298 } else if (static_res == Compile::SSC_always_false) {
4299 // Don't use intrinsic, have to throw ClassCastException.
4300 // If the reference is null, the non-intrinsic bytecode will
4301 // be optimized appropriately.
4302 return false;
4303 }
4304 }
4305 }
4306
4307 // Bailout intrinsic and do normal inlining if exception path is frequent.
4308 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4309 return false;
4310 }
4311
4312 // Generate dynamic checks.
4313 // Class.cast() is java implementation of _checkcast bytecode.
4314 // Do checkcast (Parse::do_checkcast()) optimizations here.
4315
4316 mirror = null_check(mirror);
4317 // If mirror is dead, only null-path is taken.
4318 if (stopped()) {
4319 return true;
4320 }
4321
4322 // Not-subtype or the mirror's klass ptr is nullptr (in case it is a primitive).
4323 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
4324 RegionNode* region = new RegionNode(PATH_LIMIT);
4325 record_for_igvn(region);
4326
4327 // Now load the mirror's klass metaobject, and null-check it.
4328 // If kls is null, we have a primitive mirror and
4329 // nothing is an instance of a primitive type.
4330 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4331
4332 Node* res = top();
4333 Node* io = i_o();
4334 Node* mem = merged_memory();
4335 if (!stopped()) {
4336
4337 Node* bad_type_ctrl = top();
4338 // Do checkcast optimizations.
4339 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4340 region->init_req(_bad_type_path, bad_type_ctrl);
4341 }
4342 if (region->in(_prim_path) != top() ||
4343 region->in(_bad_type_path) != top() ||
4344 region->in(_npe_path) != top()) {
4345 // Let Interpreter throw ClassCastException.
4346 PreserveJVMState pjvms(this);
4347 set_control(_gvn.transform(region));
4348 // Set IO and memory because gen_checkcast may override them when buffering inline types
4349 set_i_o(io);
4350 set_all_memory(mem);
4351 uncommon_trap(Deoptimization::Reason_intrinsic,
4352 Deoptimization::Action_maybe_recompile);
4353 }
4354 if (!stopped()) {
4355 set_result(res);
4356 }
4357 return true;
4358 }
4359
4360
4361 //--------------------------inline_native_subtype_check------------------------
4362 // This intrinsic takes the JNI calls out of the heart of
4363 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4364 bool LibraryCallKit::inline_native_subtype_check() {
4365 // Pull both arguments off the stack.
4366 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4367 args[0] = argument(0);
4368 args[1] = argument(1);
4369 Node* klasses[2]; // corresponding Klasses: superk, subk
4370 klasses[0] = klasses[1] = top();
4371
4372 enum {
4373 // A full decision tree on {superc is prim, subc is prim}:
4374 _prim_0_path = 1, // {P,N} => false
4375 // {P,P} & superc!=subc => false
4376 _prim_same_path, // {P,P} & superc==subc => true
4377 _prim_1_path, // {N,P} => false
4378 _ref_subtype_path, // {N,N} & subtype check wins => true
4379 _both_ref_path, // {N,N} & subtype check loses => false
4380 PATH_LIMIT
4381 };
4382
4383 RegionNode* region = new RegionNode(PATH_LIMIT);
4384 RegionNode* prim_region = new RegionNode(2);
4385 Node* phi = new PhiNode(region, TypeInt::BOOL);
4386 record_for_igvn(region);
4387 record_for_igvn(prim_region);
4388
4389 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4390 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4391 int class_klass_offset = java_lang_Class::klass_offset();
4392
4393 // First null-check both mirrors and load each mirror's klass metaobject.
4394 int which_arg;
4395 for (which_arg = 0; which_arg <= 1; which_arg++) {
4396 Node* arg = args[which_arg];
4397 arg = null_check(arg);
4398 if (stopped()) break;
4399 args[which_arg] = arg;
4400
4401 Node* p = basic_plus_adr(arg, class_klass_offset);
4402 Node* kls = LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, adr_type, kls_type);
4403 klasses[which_arg] = _gvn.transform(kls);
4404 }
4405
4406 // Having loaded both klasses, test each for null.
4407 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4408 for (which_arg = 0; which_arg <= 1; which_arg++) {
4409 Node* kls = klasses[which_arg];
4410 Node* null_ctl = top();
4411 kls = null_check_oop(kls, &null_ctl, never_see_null);
4412 if (which_arg == 0) {
4413 prim_region->init_req(1, null_ctl);
4414 } else {
4415 region->init_req(_prim_1_path, null_ctl);
4416 }
4417 if (stopped()) break;
4418 klasses[which_arg] = kls;
4419 }
4420
4421 if (!stopped()) {
4422 // now we have two reference types, in klasses[0..1]
4423 Node* subk = klasses[1]; // the argument to isAssignableFrom
4424 Node* superk = klasses[0]; // the receiver
4425 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4426 region->set_req(_ref_subtype_path, control());
4427 }
4428
4429 // If both operands are primitive (both klasses null), then
4430 // we must return true when they are identical primitives.
4431 // It is convenient to test this after the first null klass check.
4432 // This path is also used if superc is a value mirror.
4433 set_control(_gvn.transform(prim_region));
4434 if (!stopped()) {
4435 // Since superc is primitive, make a guard for the superc==subc case.
4436 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4437 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4438 generate_fair_guard(bol_eq, region);
4439 if (region->req() == PATH_LIMIT+1) {
4440 // A guard was added. If the added guard is taken, superc==subc.
4441 region->swap_edges(PATH_LIMIT, _prim_same_path);
4442 region->del_req(PATH_LIMIT);
4443 }
4444 region->set_req(_prim_0_path, control()); // Not equal after all.
4445 }
4446
4447 // these are the only paths that produce 'true':
4448 phi->set_req(_prim_same_path, intcon(1));
4449 phi->set_req(_ref_subtype_path, intcon(1));
4450
4451 // pull together the cases:
4452 assert(region->req() == PATH_LIMIT, "sane region");
4453 for (uint i = 1; i < region->req(); i++) {
4454 Node* ctl = region->in(i);
4455 if (ctl == nullptr || ctl == top()) {
4456 region->set_req(i, top());
4457 phi ->set_req(i, top());
4458 } else if (phi->in(i) == nullptr) {
4459 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4460 }
4461 }
4462
4463 set_control(_gvn.transform(region));
4464 set_result(_gvn.transform(phi));
4465 return true;
4466 }
4467
4468 //---------------------generate_array_guard_common------------------------
4469 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind) {
4470
4471 if (stopped()) {
4472 return nullptr;
4473 }
4474
4475 // Like generate_guard, adds a new path onto the region.
4476 jint layout_con = 0;
4477 Node* layout_val = get_layout_helper(kls, layout_con);
4478 if (layout_val == nullptr) {
4479 bool query = 0;
4480 switch(kind) {
4481 case ObjectArray: query = Klass::layout_helper_is_objArray(layout_con); break;
4482 case NonObjectArray: query = !Klass::layout_helper_is_objArray(layout_con); break;
4483 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
4484 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
4485 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
4486 default:
4487 ShouldNotReachHere();
4488 }
4489 if (!query) {
4490 return nullptr; // never a branch
4491 } else { // always a branch
4492 Node* always_branch = control();
4493 if (region != nullptr)
4494 region->add_req(always_branch);
4495 set_control(top());
4496 return always_branch;
4497 }
4498 }
4499 unsigned int value = 0;
4500 BoolTest::mask btest = BoolTest::illegal;
4501 switch(kind) {
4502 case ObjectArray:
4503 case NonObjectArray: {
4504 value = Klass::_lh_array_tag_obj_value;
4505 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4506 btest = (kind == ObjectArray) ? BoolTest::eq : BoolTest::ne;
4507 break;
4508 }
4509 case TypeArray: {
4510 value = Klass::_lh_array_tag_type_value;
4511 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4512 btest = BoolTest::eq;
4513 break;
4514 }
4515 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
4516 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
4517 default:
4518 ShouldNotReachHere();
4519 }
4520 // Now test the correct condition.
4521 jint nval = (jint)value;
4522 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4523 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4524 return generate_fair_guard(bol, region);
4525 }
4526
4527 //-----------------------inline_newNullRestrictedArray--------------------------
4528 // public static native Object[] newNullRestrictedArray(Class<?> componentType, int length);
4529 bool LibraryCallKit::inline_newNullRestrictedArray() {
4530 Node* componentType = argument(0);
4531 Node* length = argument(1);
4532
4533 const TypeInstPtr* tp = _gvn.type(componentType)->isa_instptr();
4534 if (tp != nullptr) {
4535 ciInstanceKlass* ik = tp->instance_klass();
4536 if (ik == C->env()->Class_klass()) {
4537 ciType* t = tp->java_mirror_type();
4538 if (t != nullptr && t->is_inlinetype()) {
4539 ciArrayKlass* array_klass = ciArrayKlass::make(t, true);
4540 if (array_klass->is_loaded() && array_klass->element_klass()->as_inline_klass()->is_initialized()) {
4541 const TypeAryKlassPtr* array_klass_type = TypeKlassPtr::make(array_klass, Type::trust_interfaces)->is_aryklassptr();
4542 array_klass_type = array_klass_type->cast_to_null_free();
4543 Node* obj = new_array(makecon(array_klass_type), length, 0, nullptr, false); // no arguments to push
4544 set_result(obj);
4545 assert(gvn().type(obj)->is_aryptr()->is_null_free(), "must be null-free");
4546 return true;
4547 }
4548 }
4549 }
4550 }
4551 return false;
4552 }
4553
4554 //-----------------------inline_native_newArray--------------------------
4555 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
4556 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4557 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4558 Node* mirror;
4559 Node* count_val;
4560 if (uninitialized) {
4561 null_check_receiver();
4562 mirror = argument(1);
4563 count_val = argument(2);
4564 } else {
4565 mirror = argument(0);
4566 count_val = argument(1);
4567 }
4568
4569 mirror = null_check(mirror);
4570 // If mirror or obj is dead, only null-path is taken.
4571 if (stopped()) return true;
4572
4573 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4574 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4575 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4576 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4577 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4578
4579 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4580 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4581 result_reg, _slow_path);
4582 Node* normal_ctl = control();
4583 Node* no_array_ctl = result_reg->in(_slow_path);
4584
4585 // Generate code for the slow case. We make a call to newArray().
4586 set_control(no_array_ctl);
4587 if (!stopped()) {
4588 // Either the input type is void.class, or else the
4589 // array klass has not yet been cached. Either the
4590 // ensuing call will throw an exception, or else it
4591 // will cache the array klass for next time.
4592 PreserveJVMState pjvms(this);
4593 CallJavaNode* slow_call = nullptr;
4594 if (uninitialized) {
4595 // Generate optimized virtual call (holder class 'Unsafe' is final)
4596 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4597 } else {
4598 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4599 }
4600 Node* slow_result = set_results_for_java_call(slow_call);
4601 // this->control() comes from set_results_for_java_call
4602 result_reg->set_req(_slow_path, control());
4603 result_val->set_req(_slow_path, slow_result);
4604 result_io ->set_req(_slow_path, i_o());
4605 result_mem->set_req(_slow_path, reset_memory());
4606 }
4607
4608 set_control(normal_ctl);
4609 if (!stopped()) {
4610 // Normal case: The array type has been cached in the java.lang.Class.
4611 // The following call works fine even if the array type is polymorphic.
4612 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4613 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4614 result_reg->init_req(_normal_path, control());
4615 result_val->init_req(_normal_path, obj);
4616 result_io ->init_req(_normal_path, i_o());
4617 result_mem->init_req(_normal_path, reset_memory());
4618
4619 if (uninitialized) {
4620 // Mark the allocation so that zeroing is skipped
4621 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4622 alloc->maybe_set_complete(&_gvn);
4623 }
4624 }
4625
4626 // Return the combined state.
4627 set_i_o( _gvn.transform(result_io) );
4628 set_all_memory( _gvn.transform(result_mem));
4629
4630 C->set_has_split_ifs(true); // Has chance for split-if optimization
4631 set_result(result_reg, result_val);
4632 return true;
4633 }
4634
4635 //----------------------inline_native_getLength--------------------------
4636 // public static native int java.lang.reflect.Array.getLength(Object array);
4637 bool LibraryCallKit::inline_native_getLength() {
4638 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4639
4640 Node* array = null_check(argument(0));
4641 // If array is dead, only null-path is taken.
4642 if (stopped()) return true;
4643
4644 // Deoptimize if it is a non-array.
4645 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr);
4646
4647 if (non_array != nullptr) {
4648 PreserveJVMState pjvms(this);
4649 set_control(non_array);
4650 uncommon_trap(Deoptimization::Reason_intrinsic,
4651 Deoptimization::Action_maybe_recompile);
4652 }
4653
4654 // If control is dead, only non-array-path is taken.
4655 if (stopped()) return true;
4656
4657 // The works fine even if the array type is polymorphic.
4658 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4659 Node* result = load_array_length(array);
4660
4661 C->set_has_split_ifs(true); // Has chance for split-if optimization
4662 set_result(result);
4663 return true;
4664 }
4665
4666 //------------------------inline_array_copyOf----------------------------
4667 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4668 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4669 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4670 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4671
4672 // Get the arguments.
4673 Node* original = argument(0);
4674 Node* start = is_copyOfRange? argument(1): intcon(0);
4675 Node* end = is_copyOfRange? argument(2): argument(1);
4676 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4677
4678 Node* newcopy = nullptr;
4679
4680 // Set the original stack and the reexecute bit for the interpreter to reexecute
4681 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4682 { PreserveReexecuteState preexecs(this);
4683 jvms()->set_should_reexecute(true);
4684
4685 array_type_mirror = null_check(array_type_mirror);
4686 original = null_check(original);
4687
4688 // Check if a null path was taken unconditionally.
4689 if (stopped()) return true;
4690
4691 Node* orig_length = load_array_length(original);
4692
4693 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4694 klass_node = null_check(klass_node);
4695
4696 RegionNode* bailout = new RegionNode(1);
4697 record_for_igvn(bailout);
4698
4699 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4700 // Bail out if that is so.
4701 // Inline type array may have object field that would require a
4702 // write barrier. Conservatively, go to slow path.
4703 // TODO 8251971: Optimize for the case when flat src/dst are later found
4704 // to not contain oops (i.e., move this check to the macro expansion phase).
4705 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4706 const TypeAryPtr* orig_t = _gvn.type(original)->isa_aryptr();
4707 const TypeKlassPtr* tklass = _gvn.type(klass_node)->is_klassptr();
4708 bool exclude_flat = UseFlatArray && bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing) &&
4709 // Can src array be flat and contain oops?
4710 (orig_t == nullptr || (!orig_t->is_not_flat() && (!orig_t->is_flat() || orig_t->elem()->inline_klass()->contains_oops()))) &&
4711 // Can dest array be flat and contain oops?
4712 tklass->can_be_inline_array() && (!tklass->is_flat() || tklass->is_aryklassptr()->elem()->is_instklassptr()->instance_klass()->as_inline_klass()->contains_oops());
4713 Node* not_objArray = exclude_flat ? generate_non_objArray_guard(klass_node, bailout) : generate_typeArray_guard(klass_node, bailout);
4714 if (not_objArray != nullptr) {
4715 // Improve the klass node's type from the new optimistic assumption:
4716 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4717 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0));
4718 Node* cast = new CastPPNode(control(), klass_node, akls);
4719 klass_node = _gvn.transform(cast);
4720 }
4721
4722 // Bail out if either start or end is negative.
4723 generate_negative_guard(start, bailout, &start);
4724 generate_negative_guard(end, bailout, &end);
4725
4726 Node* length = end;
4727 if (_gvn.type(start) != TypeInt::ZERO) {
4728 length = _gvn.transform(new SubINode(end, start));
4729 }
4730
4731 // Bail out if length is negative (i.e., if start > end).
4732 // Without this the new_array would throw
4733 // NegativeArraySizeException but IllegalArgumentException is what
4734 // should be thrown
4735 generate_negative_guard(length, bailout, &length);
4736
4737 // Handle inline type arrays
4738 bool can_validate = !too_many_traps(Deoptimization::Reason_class_check);
4739 if (!stopped()) {
4740 // TODO JDK-8329224
4741 if (!orig_t->is_null_free()) {
4742 // Not statically known to be null free, add a check
4743 generate_fair_guard(null_free_array_test(original), bailout);
4744 }
4745 orig_t = _gvn.type(original)->isa_aryptr();
4746 if (orig_t != nullptr && orig_t->is_flat()) {
4747 // Src is flat, check that dest is flat as well
4748 if (exclude_flat) {
4749 // Dest can't be flat, bail out
4750 bailout->add_req(control());
4751 set_control(top());
4752 } else {
4753 generate_fair_guard(flat_array_test(klass_node, /* flat = */ false), bailout);
4754 }
4755 } else if (UseFlatArray && (orig_t == nullptr || !orig_t->is_not_flat()) &&
4756 // If dest is flat, src must be flat as well (guaranteed by src <: dest check if validated).
4757 ((!tklass->is_flat() && tklass->can_be_inline_array()) || !can_validate)) {
4758 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
4759 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
4760 generate_fair_guard(flat_array_test(load_object_klass(original)), bailout);
4761 if (orig_t != nullptr) {
4762 orig_t = orig_t->cast_to_not_flat();
4763 original = _gvn.transform(new CheckCastPPNode(control(), original, orig_t));
4764 }
4765 }
4766 if (!can_validate) {
4767 // No validation. The subtype check emitted at macro expansion time will not go to the slow
4768 // path but call checkcast_arraycopy which can not handle flat/null-free inline type arrays.
4769 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat/null-free.
4770 generate_fair_guard(flat_array_test(klass_node), bailout);
4771 generate_fair_guard(null_free_array_test(original), bailout);
4772 }
4773 }
4774
4775 // Bail out if start is larger than the original length
4776 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4777 generate_negative_guard(orig_tail, bailout, &orig_tail);
4778
4779 if (bailout->req() > 1) {
4780 PreserveJVMState pjvms(this);
4781 set_control(_gvn.transform(bailout));
4782 uncommon_trap(Deoptimization::Reason_intrinsic,
4783 Deoptimization::Action_maybe_recompile);
4784 }
4785
4786 if (!stopped()) {
4787 // How many elements will we copy from the original?
4788 // The answer is MinI(orig_tail, length).
4789 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4790
4791 // Generate a direct call to the right arraycopy function(s).
4792 // We know the copy is disjoint but we might not know if the
4793 // oop stores need checking.
4794 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4795 // This will fail a store-check if x contains any non-nulls.
4796
4797 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4798 // loads/stores but it is legal only if we're sure the
4799 // Arrays.copyOf would succeed. So we need all input arguments
4800 // to the copyOf to be validated, including that the copy to the
4801 // new array won't trigger an ArrayStoreException. That subtype
4802 // check can be optimized if we know something on the type of
4803 // the input array from type speculation.
4804 if (_gvn.type(klass_node)->singleton()) {
4805 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4806 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4807
4808 int test = C->static_subtype_check(superk, subk);
4809 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4810 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4811 if (t_original->speculative_type() != nullptr) {
4812 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4813 }
4814 }
4815 }
4816
4817 bool validated = false;
4818 // Reason_class_check rather than Reason_intrinsic because we
4819 // want to intrinsify even if this traps.
4820 if (can_validate) {
4821 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4822
4823 if (not_subtype_ctrl != top()) {
4824 PreserveJVMState pjvms(this);
4825 set_control(not_subtype_ctrl);
4826 uncommon_trap(Deoptimization::Reason_class_check,
4827 Deoptimization::Action_make_not_entrant);
4828 assert(stopped(), "Should be stopped");
4829 }
4830 validated = true;
4831 }
4832
4833 if (!stopped()) {
4834 newcopy = new_array(klass_node, length, 0); // no arguments to push
4835
4836 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4837 load_object_klass(original), klass_node);
4838 if (!is_copyOfRange) {
4839 ac->set_copyof(validated);
4840 } else {
4841 ac->set_copyofrange(validated);
4842 }
4843 Node* n = _gvn.transform(ac);
4844 if (n == ac) {
4845 ac->connect_outputs(this);
4846 } else {
4847 assert(validated, "shouldn't transform if all arguments not validated");
4848 set_all_memory(n);
4849 }
4850 }
4851 }
4852 } // original reexecute is set back here
4853
4854 C->set_has_split_ifs(true); // Has chance for split-if optimization
4855 if (!stopped()) {
4856 set_result(newcopy);
4857 }
4858 return true;
4859 }
4860
4861
4862 //----------------------generate_virtual_guard---------------------------
4863 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4864 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4865 RegionNode* slow_region) {
4866 ciMethod* method = callee();
4867 int vtable_index = method->vtable_index();
4868 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4869 "bad index %d", vtable_index);
4870 // Get the Method* out of the appropriate vtable entry.
4871 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4872 vtable_index*vtableEntry::size_in_bytes() +
4873 in_bytes(vtableEntry::method_offset());
4874 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4875 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4876
4877 // Compare the target method with the expected method (e.g., Object.hashCode).
4878 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4879
4880 Node* native_call = makecon(native_call_addr);
4881 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4882 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4883
4884 return generate_slow_guard(test_native, slow_region);
4885 }
4886
4887 //-----------------------generate_method_call----------------------------
4888 // Use generate_method_call to make a slow-call to the real
4889 // method if the fast path fails. An alternative would be to
4890 // use a stub like OptoRuntime::slow_arraycopy_Java.
4891 // This only works for expanding the current library call,
4892 // not another intrinsic. (E.g., don't use this for making an
4893 // arraycopy call inside of the copyOf intrinsic.)
4894 CallJavaNode*
4895 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4896 // When compiling the intrinsic method itself, do not use this technique.
4897 guarantee(callee() != C->method(), "cannot make slow-call to self");
4898
4899 ciMethod* method = callee();
4900 // ensure the JVMS we have will be correct for this call
4901 guarantee(method_id == method->intrinsic_id(), "must match");
4902
4903 const TypeFunc* tf = TypeFunc::make(method);
4904 if (res_not_null) {
4905 assert(tf->return_type() == T_OBJECT, "");
4906 const TypeTuple* range = tf->range_cc();
4907 const Type** fields = TypeTuple::fields(range->cnt());
4908 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4909 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4910 tf = TypeFunc::make(tf->domain_cc(), new_range);
4911 }
4912 CallJavaNode* slow_call;
4913 if (is_static) {
4914 assert(!is_virtual, "");
4915 slow_call = new CallStaticJavaNode(C, tf,
4916 SharedRuntime::get_resolve_static_call_stub(), method);
4917 } else if (is_virtual) {
4918 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4919 int vtable_index = Method::invalid_vtable_index;
4920 if (UseInlineCaches) {
4921 // Suppress the vtable call
4922 } else {
4923 // hashCode and clone are not a miranda methods,
4924 // so the vtable index is fixed.
4925 // No need to use the linkResolver to get it.
4926 vtable_index = method->vtable_index();
4927 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4928 "bad index %d", vtable_index);
4929 }
4930 slow_call = new CallDynamicJavaNode(tf,
4931 SharedRuntime::get_resolve_virtual_call_stub(),
4932 method, vtable_index);
4933 } else { // neither virtual nor static: opt_virtual
4934 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4935 slow_call = new CallStaticJavaNode(C, tf,
4936 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4937 slow_call->set_optimized_virtual(true);
4938 }
4939 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4940 // To be able to issue a direct call (optimized virtual or virtual)
4941 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4942 // about the method being invoked should be attached to the call site to
4943 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4944 slow_call->set_override_symbolic_info(true);
4945 }
4946 set_arguments_for_java_call(slow_call);
4947 set_edges_for_java_call(slow_call);
4948 return slow_call;
4949 }
4950
4951
4952 /**
4953 * Build special case code for calls to hashCode on an object. This call may
4954 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4955 * slightly different code.
4956 */
4957 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4958 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4959 assert(!(is_virtual && is_static), "either virtual, special, or static");
4960
4961 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4962
4963 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4964 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4965 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4966 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4967 Node* obj = argument(0);
4968
4969 // Don't intrinsify hashcode on inline types for now.
4970 // The "is locked" runtime check below also serves as inline type check and goes to the slow path.
4971 if (gvn().type(obj)->is_inlinetypeptr()) {
4972 return false;
4973 }
4974
4975 if (!is_static) {
4976 // Check for hashing null object
4977 obj = null_check_receiver();
4978 if (stopped()) return true; // unconditionally null
4979 result_reg->init_req(_null_path, top());
4980 result_val->init_req(_null_path, top());
4981 } else {
4982 // Do a null check, and return zero if null.
4983 // System.identityHashCode(null) == 0
4984 Node* null_ctl = top();
4985 obj = null_check_oop(obj, &null_ctl);
4986 result_reg->init_req(_null_path, null_ctl);
4987 result_val->init_req(_null_path, _gvn.intcon(0));
4988 }
4989
4990 // Unconditionally null? Then return right away.
4991 if (stopped()) {
4992 set_control( result_reg->in(_null_path));
4993 if (!stopped())
4994 set_result(result_val->in(_null_path));
4995 return true;
4996 }
4997
4998 // We only go to the fast case code if we pass a number of guards. The
4999 // paths which do not pass are accumulated in the slow_region.
5000 RegionNode* slow_region = new RegionNode(1);
5001 record_for_igvn(slow_region);
5002
5003 // If this is a virtual call, we generate a funny guard. We pull out
5004 // the vtable entry corresponding to hashCode() from the target object.
5005 // If the target method which we are calling happens to be the native
5006 // Object hashCode() method, we pass the guard. We do not need this
5007 // guard for non-virtual calls -- the caller is known to be the native
5008 // Object hashCode().
5009 if (is_virtual) {
5010 // After null check, get the object's klass.
5011 Node* obj_klass = load_object_klass(obj);
5012 generate_virtual_guard(obj_klass, slow_region);
5013 }
5014
5015 // Get the header out of the object, use LoadMarkNode when available
5016 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
5017 // The control of the load must be null. Otherwise, the load can move before
5018 // the null check after castPP removal.
5019 Node* no_ctrl = nullptr;
5020 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
5021
5022 if (!UseObjectMonitorTable) {
5023 // Test the header to see if it is safe to read w.r.t. locking.
5024 // This also serves as guard against inline types
5025 Node *lock_mask = _gvn.MakeConX(markWord::inline_type_mask_in_place);
5026 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
5027 if (LockingMode == LM_LIGHTWEIGHT) {
5028 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
5029 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
5030 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
5031
5032 generate_slow_guard(test_monitor, slow_region);
5033 } else {
5034 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
5035 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val));
5036 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne));
5037
5038 generate_slow_guard(test_not_unlocked, slow_region);
5039 }
5040 }
5041
5042 // Get the hash value and check to see that it has been properly assigned.
5043 // We depend on hash_mask being at most 32 bits and avoid the use of
5044 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
5045 // vm: see markWord.hpp.
5046 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
5047 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
5048 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
5049 // This hack lets the hash bits live anywhere in the mark object now, as long
5050 // as the shift drops the relevant bits into the low 32 bits. Note that
5051 // Java spec says that HashCode is an int so there's no point in capturing
5052 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
5053 hshifted_header = ConvX2I(hshifted_header);
5054 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
5055
5056 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
5057 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
5058 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
5059
5060 generate_slow_guard(test_assigned, slow_region);
5061
5062 Node* init_mem = reset_memory();
5063 // fill in the rest of the null path:
5064 result_io ->init_req(_null_path, i_o());
5065 result_mem->init_req(_null_path, init_mem);
5066
5067 result_val->init_req(_fast_path, hash_val);
5068 result_reg->init_req(_fast_path, control());
5069 result_io ->init_req(_fast_path, i_o());
5070 result_mem->init_req(_fast_path, init_mem);
5071
5072 // Generate code for the slow case. We make a call to hashCode().
5073 set_control(_gvn.transform(slow_region));
5074 if (!stopped()) {
5075 // No need for PreserveJVMState, because we're using up the present state.
5076 set_all_memory(init_mem);
5077 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
5078 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
5079 Node* slow_result = set_results_for_java_call(slow_call);
5080 // this->control() comes from set_results_for_java_call
5081 result_reg->init_req(_slow_path, control());
5082 result_val->init_req(_slow_path, slow_result);
5083 result_io ->set_req(_slow_path, i_o());
5084 result_mem ->set_req(_slow_path, reset_memory());
5085 }
5086
5087 // Return the combined state.
5088 set_i_o( _gvn.transform(result_io) );
5089 set_all_memory( _gvn.transform(result_mem));
5090
5091 set_result(result_reg, result_val);
5092 return true;
5093 }
5094
5095 //---------------------------inline_native_getClass----------------------------
5096 // public final native Class<?> java.lang.Object.getClass();
5097 //
5098 // Build special case code for calls to getClass on an object.
5099 bool LibraryCallKit::inline_native_getClass() {
5100 Node* obj = argument(0);
5101 if (obj->is_InlineType()) {
5102 const Type* t = _gvn.type(obj);
5103 if (t->maybe_null()) {
5104 null_check(obj);
5105 }
5106 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror())));
5107 return true;
5108 }
5109 obj = null_check_receiver();
5110 if (stopped()) return true;
5111 set_result(load_mirror_from_klass(load_object_klass(obj)));
5112 return true;
5113 }
5114
5115 //-----------------inline_native_Reflection_getCallerClass---------------------
5116 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
5117 //
5118 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
5119 //
5120 // NOTE: This code must perform the same logic as JVM_GetCallerClass
5121 // in that it must skip particular security frames and checks for
5122 // caller sensitive methods.
5123 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
5124 #ifndef PRODUCT
5125 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5126 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
5127 }
5128 #endif
5129
5130 if (!jvms()->has_method()) {
5131 #ifndef PRODUCT
5132 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5133 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
5134 }
5135 #endif
5136 return false;
5137 }
5138
5139 // Walk back up the JVM state to find the caller at the required
5140 // depth.
5141 JVMState* caller_jvms = jvms();
5142
5143 // Cf. JVM_GetCallerClass
5144 // NOTE: Start the loop at depth 1 because the current JVM state does
5145 // not include the Reflection.getCallerClass() frame.
5146 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
5147 ciMethod* m = caller_jvms->method();
5148 switch (n) {
5149 case 0:
5150 fatal("current JVM state does not include the Reflection.getCallerClass frame");
5151 break;
5152 case 1:
5153 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
5154 if (!m->caller_sensitive()) {
5155 #ifndef PRODUCT
5156 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5157 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
5158 }
5159 #endif
5160 return false; // bail-out; let JVM_GetCallerClass do the work
5161 }
5162 break;
5163 default:
5164 if (!m->is_ignored_by_security_stack_walk()) {
5165 // We have reached the desired frame; return the holder class.
5166 // Acquire method holder as java.lang.Class and push as constant.
5167 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
5168 ciInstance* caller_mirror = caller_klass->java_mirror();
5169 set_result(makecon(TypeInstPtr::make(caller_mirror)));
5170
5171 #ifndef PRODUCT
5172 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5173 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());
5174 tty->print_cr(" JVM state at this point:");
5175 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5176 ciMethod* m = jvms()->of_depth(i)->method();
5177 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5178 }
5179 }
5180 #endif
5181 return true;
5182 }
5183 break;
5184 }
5185 }
5186
5187 #ifndef PRODUCT
5188 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5189 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
5190 tty->print_cr(" JVM state at this point:");
5191 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5192 ciMethod* m = jvms()->of_depth(i)->method();
5193 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5194 }
5195 }
5196 #endif
5197
5198 return false; // bail-out; let JVM_GetCallerClass do the work
5199 }
5200
5201 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
5202 Node* arg = argument(0);
5203 Node* result = nullptr;
5204
5205 switch (id) {
5206 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
5207 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
5208 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5209 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5210 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5211 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5212
5213 case vmIntrinsics::_doubleToLongBits: {
5214 // two paths (plus control) merge in a wood
5215 RegionNode *r = new RegionNode(3);
5216 Node *phi = new PhiNode(r, TypeLong::LONG);
5217
5218 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5219 // Build the boolean node
5220 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5221
5222 // Branch either way.
5223 // NaN case is less traveled, which makes all the difference.
5224 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5225 Node *opt_isnan = _gvn.transform(ifisnan);
5226 assert( opt_isnan->is_If(), "Expect an IfNode");
5227 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5228 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5229
5230 set_control(iftrue);
5231
5232 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5233 Node *slow_result = longcon(nan_bits); // return NaN
5234 phi->init_req(1, _gvn.transform( slow_result ));
5235 r->init_req(1, iftrue);
5236
5237 // Else fall through
5238 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5239 set_control(iffalse);
5240
5241 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5242 r->init_req(2, iffalse);
5243
5244 // Post merge
5245 set_control(_gvn.transform(r));
5246 record_for_igvn(r);
5247
5248 C->set_has_split_ifs(true); // Has chance for split-if optimization
5249 result = phi;
5250 assert(result->bottom_type()->isa_long(), "must be");
5251 break;
5252 }
5253
5254 case vmIntrinsics::_floatToIntBits: {
5255 // two paths (plus control) merge in a wood
5256 RegionNode *r = new RegionNode(3);
5257 Node *phi = new PhiNode(r, TypeInt::INT);
5258
5259 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5260 // Build the boolean node
5261 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5262
5263 // Branch either way.
5264 // NaN case is less traveled, which makes all the difference.
5265 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5266 Node *opt_isnan = _gvn.transform(ifisnan);
5267 assert( opt_isnan->is_If(), "Expect an IfNode");
5268 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5269 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5270
5271 set_control(iftrue);
5272
5273 static const jint nan_bits = 0x7fc00000;
5274 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5275 phi->init_req(1, _gvn.transform( slow_result ));
5276 r->init_req(1, iftrue);
5277
5278 // Else fall through
5279 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5280 set_control(iffalse);
5281
5282 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5283 r->init_req(2, iffalse);
5284
5285 // Post merge
5286 set_control(_gvn.transform(r));
5287 record_for_igvn(r);
5288
5289 C->set_has_split_ifs(true); // Has chance for split-if optimization
5290 result = phi;
5291 assert(result->bottom_type()->isa_int(), "must be");
5292 break;
5293 }
5294
5295 default:
5296 fatal_unexpected_iid(id);
5297 break;
5298 }
5299 set_result(_gvn.transform(result));
5300 return true;
5301 }
5302
5303 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5304 Node* arg = argument(0);
5305 Node* result = nullptr;
5306
5307 switch (id) {
5308 case vmIntrinsics::_floatIsInfinite:
5309 result = new IsInfiniteFNode(arg);
5310 break;
5311 case vmIntrinsics::_floatIsFinite:
5312 result = new IsFiniteFNode(arg);
5313 break;
5314 case vmIntrinsics::_doubleIsInfinite:
5315 result = new IsInfiniteDNode(arg);
5316 break;
5317 case vmIntrinsics::_doubleIsFinite:
5318 result = new IsFiniteDNode(arg);
5319 break;
5320 default:
5321 fatal_unexpected_iid(id);
5322 break;
5323 }
5324 set_result(_gvn.transform(result));
5325 return true;
5326 }
5327
5328 //----------------------inline_unsafe_copyMemory-------------------------
5329 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5330
5331 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5332 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5333 const Type* base_t = gvn.type(base);
5334
5335 bool in_native = (base_t == TypePtr::NULL_PTR);
5336 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5337 bool is_mixed = !in_heap && !in_native;
5338
5339 if (is_mixed) {
5340 return true; // mixed accesses can touch both on-heap and off-heap memory
5341 }
5342 if (in_heap) {
5343 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5344 if (!is_prim_array) {
5345 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5346 // there's not enough type information available to determine proper memory slice for it.
5347 return true;
5348 }
5349 }
5350 return false;
5351 }
5352
5353 bool LibraryCallKit::inline_unsafe_copyMemory() {
5354 if (callee()->is_static()) return false; // caller must have the capability!
5355 null_check_receiver(); // null-check receiver
5356 if (stopped()) return true;
5357
5358 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5359
5360 Node* src_base = argument(1); // type: oop
5361 Node* src_off = ConvL2X(argument(2)); // type: long
5362 Node* dst_base = argument(4); // type: oop
5363 Node* dst_off = ConvL2X(argument(5)); // type: long
5364 Node* size = ConvL2X(argument(7)); // type: long
5365
5366 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5367 "fieldOffset must be byte-scaled");
5368
5369 Node* src_addr = make_unsafe_address(src_base, src_off);
5370 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5371
5372 Node* thread = _gvn.transform(new ThreadLocalNode());
5373 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5374 BasicType doing_unsafe_access_bt = T_BYTE;
5375 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5376
5377 // update volatile field
5378 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5379
5380 int flags = RC_LEAF | RC_NO_FP;
5381
5382 const TypePtr* dst_type = TypePtr::BOTTOM;
5383
5384 // Adjust memory effects of the runtime call based on input values.
5385 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5386 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5387 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5388
5389 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5390 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5391 flags |= RC_NARROW_MEM; // narrow in memory
5392 }
5393 }
5394
5395 // Call it. Note that the length argument is not scaled.
5396 make_runtime_call(flags,
5397 OptoRuntime::fast_arraycopy_Type(),
5398 StubRoutines::unsafe_arraycopy(),
5399 "unsafe_arraycopy",
5400 dst_type,
5401 src_addr, dst_addr, size XTOP);
5402
5403 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5404
5405 return true;
5406 }
5407
5408 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5409 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5410 bool LibraryCallKit::inline_unsafe_setMemory() {
5411 if (callee()->is_static()) return false; // caller must have the capability!
5412 null_check_receiver(); // null-check receiver
5413 if (stopped()) return true;
5414
5415 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5416
5417 Node* dst_base = argument(1); // type: oop
5418 Node* dst_off = ConvL2X(argument(2)); // type: long
5419 Node* size = ConvL2X(argument(4)); // type: long
5420 Node* byte = argument(6); // type: byte
5421
5422 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5423 "fieldOffset must be byte-scaled");
5424
5425 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5426
5427 Node* thread = _gvn.transform(new ThreadLocalNode());
5428 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5429 BasicType doing_unsafe_access_bt = T_BYTE;
5430 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5431
5432 // update volatile field
5433 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5434
5435 int flags = RC_LEAF | RC_NO_FP;
5436
5437 const TypePtr* dst_type = TypePtr::BOTTOM;
5438
5439 // Adjust memory effects of the runtime call based on input values.
5440 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5441 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5442
5443 flags |= RC_NARROW_MEM; // narrow in memory
5444 }
5445
5446 // Call it. Note that the length argument is not scaled.
5447 make_runtime_call(flags,
5448 OptoRuntime::make_setmemory_Type(),
5449 StubRoutines::unsafe_setmemory(),
5450 "unsafe_setmemory",
5451 dst_type,
5452 dst_addr, size XTOP, byte);
5453
5454 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5455
5456 return true;
5457 }
5458
5459 #undef XTOP
5460
5461 //----------------------inline_unsafe_isFlatArray------------------------
5462 // public native boolean Unsafe.isFlatArray(Class<?> arrayClass);
5463 // This intrinsic exploits assumptions made by the native implementation
5464 // (arrayClass is neither null nor primitive) to avoid unnecessary null checks.
5465 bool LibraryCallKit::inline_unsafe_isFlatArray() {
5466 Node* cls = argument(1);
5467 Node* p = basic_plus_adr(cls, java_lang_Class::klass_offset());
5468 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p,
5469 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT));
5470 Node* result = flat_array_test(kls);
5471 set_result(result);
5472 return true;
5473 }
5474
5475 //------------------------clone_coping-----------------------------------
5476 // Helper function for inline_native_clone.
5477 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5478 assert(obj_size != nullptr, "");
5479 Node* raw_obj = alloc_obj->in(1);
5480 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5481
5482 AllocateNode* alloc = nullptr;
5483 if (ReduceBulkZeroing &&
5484 // If we are implementing an array clone without knowing its source type
5485 // (can happen when compiling the array-guarded branch of a reflective
5486 // Object.clone() invocation), initialize the array within the allocation.
5487 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5488 // to a runtime clone call that assumes fully initialized source arrays.
5489 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5490 // We will be completely responsible for initializing this object -
5491 // mark Initialize node as complete.
5492 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5493 // The object was just allocated - there should be no any stores!
5494 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5495 // Mark as complete_with_arraycopy so that on AllocateNode
5496 // expansion, we know this AllocateNode is initialized by an array
5497 // copy and a StoreStore barrier exists after the array copy.
5498 alloc->initialization()->set_complete_with_arraycopy();
5499 }
5500
5501 Node* size = _gvn.transform(obj_size);
5502 access_clone(obj, alloc_obj, size, is_array);
5503
5504 // Do not let reads from the cloned object float above the arraycopy.
5505 if (alloc != nullptr) {
5506 // Do not let stores that initialize this object be reordered with
5507 // a subsequent store that would make this object accessible by
5508 // other threads.
5509 // Record what AllocateNode this StoreStore protects so that
5510 // escape analysis can go from the MemBarStoreStoreNode to the
5511 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5512 // based on the escape status of the AllocateNode.
5513 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5514 } else {
5515 insert_mem_bar(Op_MemBarCPUOrder);
5516 }
5517 }
5518
5519 //------------------------inline_native_clone----------------------------
5520 // protected native Object java.lang.Object.clone();
5521 //
5522 // Here are the simple edge cases:
5523 // null receiver => normal trap
5524 // virtual and clone was overridden => slow path to out-of-line clone
5525 // not cloneable or finalizer => slow path to out-of-line Object.clone
5526 //
5527 // The general case has two steps, allocation and copying.
5528 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5529 //
5530 // Copying also has two cases, oop arrays and everything else.
5531 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5532 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5533 //
5534 // These steps fold up nicely if and when the cloned object's klass
5535 // can be sharply typed as an object array, a type array, or an instance.
5536 //
5537 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5538 PhiNode* result_val;
5539
5540 // Set the reexecute bit for the interpreter to reexecute
5541 // the bytecode that invokes Object.clone if deoptimization happens.
5542 { PreserveReexecuteState preexecs(this);
5543 jvms()->set_should_reexecute(true);
5544
5545 Node* obj = argument(0);
5546 obj = null_check_receiver();
5547 if (stopped()) return true;
5548
5549 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5550 if (obj_type->is_inlinetypeptr()) {
5551 // If the object to clone is an inline type, we can simply return it (i.e. a nop) since inline types have
5552 // no identity.
5553 set_result(obj);
5554 return true;
5555 }
5556
5557 // If we are going to clone an instance, we need its exact type to
5558 // know the number and types of fields to convert the clone to
5559 // loads/stores. Maybe a speculative type can help us.
5560 if (!obj_type->klass_is_exact() &&
5561 obj_type->speculative_type() != nullptr &&
5562 obj_type->speculative_type()->is_instance_klass() &&
5563 !obj_type->speculative_type()->is_inlinetype()) {
5564 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5565 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5566 !spec_ik->has_injected_fields()) {
5567 if (!obj_type->isa_instptr() ||
5568 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5569 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5570 }
5571 }
5572 }
5573
5574 // Conservatively insert a memory barrier on all memory slices.
5575 // Do not let writes into the original float below the clone.
5576 insert_mem_bar(Op_MemBarCPUOrder);
5577
5578 // paths into result_reg:
5579 enum {
5580 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5581 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5582 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5583 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5584 PATH_LIMIT
5585 };
5586 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5587 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5588 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5589 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5590 record_for_igvn(result_reg);
5591
5592 Node* obj_klass = load_object_klass(obj);
5593 // We only go to the fast case code if we pass a number of guards.
5594 // The paths which do not pass are accumulated in the slow_region.
5595 RegionNode* slow_region = new RegionNode(1);
5596 record_for_igvn(slow_region);
5597
5598 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr);
5599 if (array_ctl != nullptr) {
5600 // It's an array.
5601 PreserveJVMState pjvms(this);
5602 set_control(array_ctl);
5603
5604 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5605 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr();
5606 if (UseFlatArray && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) &&
5607 obj_type->can_be_inline_array() &&
5608 (ary_ptr == nullptr || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) {
5609 // Flat inline type array may have object field that would require a
5610 // write barrier. Conservatively, go to slow path.
5611 generate_fair_guard(flat_array_test(obj_klass), slow_region);
5612 }
5613
5614 if (!stopped()) {
5615 Node* obj_length = load_array_length(obj);
5616 Node* array_size = nullptr; // Size of the array without object alignment padding.
5617 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5618
5619 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5620 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5621 // If it is an oop array, it requires very special treatment,
5622 // because gc barriers are required when accessing the array.
5623 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5624 if (is_obja != nullptr) {
5625 PreserveJVMState pjvms2(this);
5626 set_control(is_obja);
5627 // Generate a direct call to the right arraycopy function(s).
5628 // Clones are always tightly coupled.
5629 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5630 ac->set_clone_oop_array();
5631 Node* n = _gvn.transform(ac);
5632 assert(n == ac, "cannot disappear");
5633 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5634
5635 result_reg->init_req(_objArray_path, control());
5636 result_val->init_req(_objArray_path, alloc_obj);
5637 result_i_o ->set_req(_objArray_path, i_o());
5638 result_mem ->set_req(_objArray_path, reset_memory());
5639 }
5640 }
5641 // Otherwise, there are no barriers to worry about.
5642 // (We can dispense with card marks if we know the allocation
5643 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5644 // causes the non-eden paths to take compensating steps to
5645 // simulate a fresh allocation, so that no further
5646 // card marks are required in compiled code to initialize
5647 // the object.)
5648
5649 if (!stopped()) {
5650 copy_to_clone(obj, alloc_obj, array_size, true);
5651
5652 // Present the results of the copy.
5653 result_reg->init_req(_array_path, control());
5654 result_val->init_req(_array_path, alloc_obj);
5655 result_i_o ->set_req(_array_path, i_o());
5656 result_mem ->set_req(_array_path, reset_memory());
5657 }
5658 }
5659 }
5660
5661 if (!stopped()) {
5662 // It's an instance (we did array above). Make the slow-path tests.
5663 // If this is a virtual call, we generate a funny guard. We grab
5664 // the vtable entry corresponding to clone() from the target object.
5665 // If the target method which we are calling happens to be the
5666 // Object clone() method, we pass the guard. We do not need this
5667 // guard for non-virtual calls; the caller is known to be the native
5668 // Object clone().
5669 if (is_virtual) {
5670 generate_virtual_guard(obj_klass, slow_region);
5671 }
5672
5673 // The object must be easily cloneable and must not have a finalizer.
5674 // Both of these conditions may be checked in a single test.
5675 // We could optimize the test further, but we don't care.
5676 generate_misc_flags_guard(obj_klass,
5677 // Test both conditions:
5678 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5679 // Must be cloneable but not finalizer:
5680 KlassFlags::_misc_is_cloneable_fast,
5681 slow_region);
5682 }
5683
5684 if (!stopped()) {
5685 // It's an instance, and it passed the slow-path tests.
5686 PreserveJVMState pjvms(this);
5687 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5688 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5689 // is reexecuted if deoptimization occurs and there could be problems when merging
5690 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5691 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5692
5693 copy_to_clone(obj, alloc_obj, obj_size, false);
5694
5695 // Present the results of the slow call.
5696 result_reg->init_req(_instance_path, control());
5697 result_val->init_req(_instance_path, alloc_obj);
5698 result_i_o ->set_req(_instance_path, i_o());
5699 result_mem ->set_req(_instance_path, reset_memory());
5700 }
5701
5702 // Generate code for the slow case. We make a call to clone().
5703 set_control(_gvn.transform(slow_region));
5704 if (!stopped()) {
5705 PreserveJVMState pjvms(this);
5706 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5707 // We need to deoptimize on exception (see comment above)
5708 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5709 // this->control() comes from set_results_for_java_call
5710 result_reg->init_req(_slow_path, control());
5711 result_val->init_req(_slow_path, slow_result);
5712 result_i_o ->set_req(_slow_path, i_o());
5713 result_mem ->set_req(_slow_path, reset_memory());
5714 }
5715
5716 // Return the combined state.
5717 set_control( _gvn.transform(result_reg));
5718 set_i_o( _gvn.transform(result_i_o));
5719 set_all_memory( _gvn.transform(result_mem));
5720 } // original reexecute is set back here
5721
5722 set_result(_gvn.transform(result_val));
5723 return true;
5724 }
5725
5726 // If we have a tightly coupled allocation, the arraycopy may take care
5727 // of the array initialization. If one of the guards we insert between
5728 // the allocation and the arraycopy causes a deoptimization, an
5729 // uninitialized array will escape the compiled method. To prevent that
5730 // we set the JVM state for uncommon traps between the allocation and
5731 // the arraycopy to the state before the allocation so, in case of
5732 // deoptimization, we'll reexecute the allocation and the
5733 // initialization.
5734 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5735 if (alloc != nullptr) {
5736 ciMethod* trap_method = alloc->jvms()->method();
5737 int trap_bci = alloc->jvms()->bci();
5738
5739 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5740 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5741 // Make sure there's no store between the allocation and the
5742 // arraycopy otherwise visible side effects could be rexecuted
5743 // in case of deoptimization and cause incorrect execution.
5744 bool no_interfering_store = true;
5745 Node* mem = alloc->in(TypeFunc::Memory);
5746 if (mem->is_MergeMem()) {
5747 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5748 Node* n = mms.memory();
5749 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5750 assert(n->is_Store(), "what else?");
5751 no_interfering_store = false;
5752 break;
5753 }
5754 }
5755 } else {
5756 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5757 Node* n = mms.memory();
5758 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5759 assert(n->is_Store(), "what else?");
5760 no_interfering_store = false;
5761 break;
5762 }
5763 }
5764 }
5765
5766 if (no_interfering_store) {
5767 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5768
5769 JVMState* saved_jvms = jvms();
5770 saved_reexecute_sp = _reexecute_sp;
5771
5772 set_jvms(sfpt->jvms());
5773 _reexecute_sp = jvms()->sp();
5774
5775 return saved_jvms;
5776 }
5777 }
5778 }
5779 return nullptr;
5780 }
5781
5782 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5783 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5784 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5785 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5786 uint size = alloc->req();
5787 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5788 old_jvms->set_map(sfpt);
5789 for (uint i = 0; i < size; i++) {
5790 sfpt->init_req(i, alloc->in(i));
5791 }
5792 int adjustment = 1;
5793 const TypeAryKlassPtr* ary_klass_ptr = alloc->in(AllocateNode::KlassNode)->bottom_type()->is_aryklassptr();
5794 if (ary_klass_ptr->is_null_free()) {
5795 // A null-free, tightly coupled array allocation can only come from LibraryCallKit::inline_newNullRestrictedArray
5796 // which requires both the component type and the array length on stack for re-execution. Re-create and push
5797 // the component type.
5798 ciArrayKlass* klass = ary_klass_ptr->exact_klass()->as_array_klass();
5799 ciInstance* instance = klass->component_mirror_instance();
5800 const TypeInstPtr* t_instance = TypeInstPtr::make(instance);
5801 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), makecon(t_instance));
5802 adjustment++;
5803 }
5804 // re-push array length for deoptimization
5805 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment - 1, alloc->in(AllocateNode::ALength));
5806 old_jvms->set_sp(old_jvms->sp() + adjustment);
5807 old_jvms->set_monoff(old_jvms->monoff() + adjustment);
5808 old_jvms->set_scloff(old_jvms->scloff() + adjustment);
5809 old_jvms->set_endoff(old_jvms->endoff() + adjustment);
5810 old_jvms->set_should_reexecute(true);
5811
5812 sfpt->set_i_o(map()->i_o());
5813 sfpt->set_memory(map()->memory());
5814 sfpt->set_control(map()->control());
5815 return sfpt;
5816 }
5817
5818 // In case of a deoptimization, we restart execution at the
5819 // allocation, allocating a new array. We would leave an uninitialized
5820 // array in the heap that GCs wouldn't expect. Move the allocation
5821 // after the traps so we don't allocate the array if we
5822 // deoptimize. This is possible because tightly_coupled_allocation()
5823 // guarantees there's no observer of the allocated array at this point
5824 // and the control flow is simple enough.
5825 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5826 int saved_reexecute_sp, uint new_idx) {
5827 if (saved_jvms_before_guards != nullptr && !stopped()) {
5828 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5829
5830 assert(alloc != nullptr, "only with a tightly coupled allocation");
5831 // restore JVM state to the state at the arraycopy
5832 saved_jvms_before_guards->map()->set_control(map()->control());
5833 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5834 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5835 // If we've improved the types of some nodes (null check) while
5836 // emitting the guards, propagate them to the current state
5837 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5838 set_jvms(saved_jvms_before_guards);
5839 _reexecute_sp = saved_reexecute_sp;
5840
5841 // Remove the allocation from above the guards
5842 CallProjections* callprojs = alloc->extract_projections(true);
5843 InitializeNode* init = alloc->initialization();
5844 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5845 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5846 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5847
5848 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5849 // the allocation (i.e. is only valid if the allocation succeeds):
5850 // 1) replace CastIINode with AllocateArrayNode's length here
5851 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5852 //
5853 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5854 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5855 Node* init_control = init->proj_out(TypeFunc::Control);
5856 Node* alloc_length = alloc->Ideal_length();
5857 #ifdef ASSERT
5858 Node* prev_cast = nullptr;
5859 #endif
5860 for (uint i = 0; i < init_control->outcnt(); i++) {
5861 Node* init_out = init_control->raw_out(i);
5862 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5863 #ifdef ASSERT
5864 if (prev_cast == nullptr) {
5865 prev_cast = init_out;
5866 } else {
5867 if (prev_cast->cmp(*init_out) == false) {
5868 prev_cast->dump();
5869 init_out->dump();
5870 assert(false, "not equal CastIINode");
5871 }
5872 }
5873 #endif
5874 C->gvn_replace_by(init_out, alloc_length);
5875 }
5876 }
5877 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5878
5879 // move the allocation here (after the guards)
5880 _gvn.hash_delete(alloc);
5881 alloc->set_req(TypeFunc::Control, control());
5882 alloc->set_req(TypeFunc::I_O, i_o());
5883 Node *mem = reset_memory();
5884 set_all_memory(mem);
5885 alloc->set_req(TypeFunc::Memory, mem);
5886 set_control(init->proj_out_or_null(TypeFunc::Control));
5887 set_i_o(callprojs->fallthrough_ioproj);
5888
5889 // Update memory as done in GraphKit::set_output_for_allocation()
5890 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5891 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5892 if (ary_type->isa_aryptr() && length_type != nullptr) {
5893 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5894 }
5895 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5896 int elemidx = C->get_alias_index(telemref);
5897 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5898 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5899
5900 Node* allocx = _gvn.transform(alloc);
5901 assert(allocx == alloc, "where has the allocation gone?");
5902 assert(dest->is_CheckCastPP(), "not an allocation result?");
5903
5904 _gvn.hash_delete(dest);
5905 dest->set_req(0, control());
5906 Node* destx = _gvn.transform(dest);
5907 assert(destx == dest, "where has the allocation result gone?");
5908
5909 array_ideal_length(alloc, ary_type, true);
5910 }
5911 }
5912
5913 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5914 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5915 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5916 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5917 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5918 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5919 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5920 JVMState* saved_jvms_before_guards) {
5921 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5922 // There is at least one unrelated uncommon trap which needs to be replaced.
5923 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5924
5925 JVMState* saved_jvms = jvms();
5926 const int saved_reexecute_sp = _reexecute_sp;
5927 set_jvms(sfpt->jvms());
5928 _reexecute_sp = jvms()->sp();
5929
5930 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5931
5932 // Restore state
5933 set_jvms(saved_jvms);
5934 _reexecute_sp = saved_reexecute_sp;
5935 }
5936 }
5937
5938 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5939 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5940 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5941 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5942 while (if_proj->is_IfProj()) {
5943 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5944 if (uncommon_trap != nullptr) {
5945 create_new_uncommon_trap(uncommon_trap);
5946 }
5947 assert(if_proj->in(0)->is_If(), "must be If");
5948 if_proj = if_proj->in(0)->in(0);
5949 }
5950 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5951 "must have reached control projection of init node");
5952 }
5953
5954 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5955 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5956 assert(trap_request != 0, "no valid UCT trap request");
5957 PreserveJVMState pjvms(this);
5958 set_control(uncommon_trap_call->in(0));
5959 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5960 Deoptimization::trap_request_action(trap_request));
5961 assert(stopped(), "Should be stopped");
5962 _gvn.hash_delete(uncommon_trap_call);
5963 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5964 }
5965
5966 // Common checks for array sorting intrinsics arguments.
5967 // Returns `true` if checks passed.
5968 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
5969 // check address of the class
5970 if (elementType == nullptr || elementType->is_top()) {
5971 return false; // dead path
5972 }
5973 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5974 if (elem_klass == nullptr) {
5975 return false; // dead path
5976 }
5977 // java_mirror_type() returns non-null for compile-time Class constants only
5978 ciType* elem_type = elem_klass->java_mirror_type();
5979 if (elem_type == nullptr) {
5980 return false;
5981 }
5982 bt = elem_type->basic_type();
5983 // Disable the intrinsic if the CPU does not support SIMD sort
5984 if (!Matcher::supports_simd_sort(bt)) {
5985 return false;
5986 }
5987 // check address of the array
5988 if (obj == nullptr || obj->is_top()) {
5989 return false; // dead path
5990 }
5991 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5992 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
5993 return false; // failed input validation
5994 }
5995 return true;
5996 }
5997
5998 //------------------------------inline_array_partition-----------------------
5999 bool LibraryCallKit::inline_array_partition() {
6000 address stubAddr = StubRoutines::select_array_partition_function();
6001 if (stubAddr == nullptr) {
6002 return false; // Intrinsic's stub is not implemented on this platform
6003 }
6004 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
6005
6006 // no receiver because it is a static method
6007 Node* elementType = argument(0);
6008 Node* obj = argument(1);
6009 Node* offset = argument(2); // long
6010 Node* fromIndex = argument(4);
6011 Node* toIndex = argument(5);
6012 Node* indexPivot1 = argument(6);
6013 Node* indexPivot2 = argument(7);
6014 // PartitionOperation: argument(8) is ignored
6015
6016 Node* pivotIndices = nullptr;
6017 BasicType bt = T_ILLEGAL;
6018
6019 if (!check_array_sort_arguments(elementType, obj, bt)) {
6020 return false;
6021 }
6022 null_check(obj);
6023 // If obj is dead, only null-path is taken.
6024 if (stopped()) {
6025 return true;
6026 }
6027 // Set the original stack and the reexecute bit for the interpreter to reexecute
6028 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
6029 { PreserveReexecuteState preexecs(this);
6030 jvms()->set_should_reexecute(true);
6031
6032 Node* obj_adr = make_unsafe_address(obj, offset);
6033
6034 // create the pivotIndices array of type int and size = 2
6035 Node* size = intcon(2);
6036 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
6037 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
6038 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
6039 guarantee(alloc != nullptr, "created above");
6040 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
6041
6042 // pass the basic type enum to the stub
6043 Node* elemType = intcon(bt);
6044
6045 // Call the stub
6046 const char *stubName = "array_partition_stub";
6047 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
6048 stubAddr, stubName, TypePtr::BOTTOM,
6049 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
6050 indexPivot1, indexPivot2);
6051
6052 } // original reexecute is set back here
6053
6054 if (!stopped()) {
6055 set_result(pivotIndices);
6056 }
6057
6058 return true;
6059 }
6060
6061
6062 //------------------------------inline_array_sort-----------------------
6063 bool LibraryCallKit::inline_array_sort() {
6064 address stubAddr = StubRoutines::select_arraysort_function();
6065 if (stubAddr == nullptr) {
6066 return false; // Intrinsic's stub is not implemented on this platform
6067 }
6068 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
6069
6070 // no receiver because it is a static method
6071 Node* elementType = argument(0);
6072 Node* obj = argument(1);
6073 Node* offset = argument(2); // long
6074 Node* fromIndex = argument(4);
6075 Node* toIndex = argument(5);
6076 // SortOperation: argument(6) is ignored
6077
6078 BasicType bt = T_ILLEGAL;
6079
6080 if (!check_array_sort_arguments(elementType, obj, bt)) {
6081 return false;
6082 }
6083 null_check(obj);
6084 // If obj is dead, only null-path is taken.
6085 if (stopped()) {
6086 return true;
6087 }
6088 Node* obj_adr = make_unsafe_address(obj, offset);
6089
6090 // pass the basic type enum to the stub
6091 Node* elemType = intcon(bt);
6092
6093 // Call the stub.
6094 const char *stubName = "arraysort_stub";
6095 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
6096 stubAddr, stubName, TypePtr::BOTTOM,
6097 obj_adr, elemType, fromIndex, toIndex);
6098
6099 return true;
6100 }
6101
6102
6103 //------------------------------inline_arraycopy-----------------------
6104 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
6105 // Object dest, int destPos,
6106 // int length);
6107 bool LibraryCallKit::inline_arraycopy() {
6108 // Get the arguments.
6109 Node* src = argument(0); // type: oop
6110 Node* src_offset = argument(1); // type: int
6111 Node* dest = argument(2); // type: oop
6112 Node* dest_offset = argument(3); // type: int
6113 Node* length = argument(4); // type: int
6114
6115 uint new_idx = C->unique();
6116
6117 // Check for allocation before we add nodes that would confuse
6118 // tightly_coupled_allocation()
6119 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
6120
6121 int saved_reexecute_sp = -1;
6122 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
6123 // See arraycopy_restore_alloc_state() comment
6124 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
6125 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
6126 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
6127 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
6128
6129 // The following tests must be performed
6130 // (1) src and dest are arrays.
6131 // (2) src and dest arrays must have elements of the same BasicType
6132 // (3) src and dest must not be null.
6133 // (4) src_offset must not be negative.
6134 // (5) dest_offset must not be negative.
6135 // (6) length must not be negative.
6136 // (7) src_offset + length must not exceed length of src.
6137 // (8) dest_offset + length must not exceed length of dest.
6138 // (9) each element of an oop array must be assignable
6139
6140 // (3) src and dest must not be null.
6141 // always do this here because we need the JVM state for uncommon traps
6142 Node* null_ctl = top();
6143 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
6144 assert(null_ctl->is_top(), "no null control here");
6145 dest = null_check(dest, T_ARRAY);
6146
6147 if (!can_emit_guards) {
6148 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
6149 // guards but the arraycopy node could still take advantage of a
6150 // tightly allocated allocation. tightly_coupled_allocation() is
6151 // called again to make sure it takes the null check above into
6152 // account: the null check is mandatory and if it caused an
6153 // uncommon trap to be emitted then the allocation can't be
6154 // considered tightly coupled in this context.
6155 alloc = tightly_coupled_allocation(dest);
6156 }
6157
6158 bool validated = false;
6159
6160 const Type* src_type = _gvn.type(src);
6161 const Type* dest_type = _gvn.type(dest);
6162 const TypeAryPtr* top_src = src_type->isa_aryptr();
6163 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6164
6165 // Do we have the type of src?
6166 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6167 // Do we have the type of dest?
6168 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6169 // Is the type for src from speculation?
6170 bool src_spec = false;
6171 // Is the type for dest from speculation?
6172 bool dest_spec = false;
6173
6174 if ((!has_src || !has_dest) && can_emit_guards) {
6175 // We don't have sufficient type information, let's see if
6176 // speculative types can help. We need to have types for both src
6177 // and dest so that it pays off.
6178
6179 // Do we already have or could we have type information for src
6180 bool could_have_src = has_src;
6181 // Do we already have or could we have type information for dest
6182 bool could_have_dest = has_dest;
6183
6184 ciKlass* src_k = nullptr;
6185 if (!has_src) {
6186 src_k = src_type->speculative_type_not_null();
6187 if (src_k != nullptr && src_k->is_array_klass()) {
6188 could_have_src = true;
6189 }
6190 }
6191
6192 ciKlass* dest_k = nullptr;
6193 if (!has_dest) {
6194 dest_k = dest_type->speculative_type_not_null();
6195 if (dest_k != nullptr && dest_k->is_array_klass()) {
6196 could_have_dest = true;
6197 }
6198 }
6199
6200 if (could_have_src && could_have_dest) {
6201 // This is going to pay off so emit the required guards
6202 if (!has_src) {
6203 src = maybe_cast_profiled_obj(src, src_k, true);
6204 src_type = _gvn.type(src);
6205 top_src = src_type->isa_aryptr();
6206 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6207 src_spec = true;
6208 }
6209 if (!has_dest) {
6210 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6211 dest_type = _gvn.type(dest);
6212 top_dest = dest_type->isa_aryptr();
6213 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6214 dest_spec = true;
6215 }
6216 }
6217 }
6218
6219 if (has_src && has_dest && can_emit_guards) {
6220 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
6221 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
6222 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
6223 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
6224
6225 if (src_elem == dest_elem && top_src->is_flat() == top_dest->is_flat() && src_elem == T_OBJECT) {
6226 // If both arrays are object arrays then having the exact types
6227 // for both will remove the need for a subtype check at runtime
6228 // before the call and may make it possible to pick a faster copy
6229 // routine (without a subtype check on every element)
6230 // Do we have the exact type of src?
6231 bool could_have_src = src_spec;
6232 // Do we have the exact type of dest?
6233 bool could_have_dest = dest_spec;
6234 ciKlass* src_k = nullptr;
6235 ciKlass* dest_k = nullptr;
6236 if (!src_spec) {
6237 src_k = src_type->speculative_type_not_null();
6238 if (src_k != nullptr && src_k->is_array_klass()) {
6239 could_have_src = true;
6240 }
6241 }
6242 if (!dest_spec) {
6243 dest_k = dest_type->speculative_type_not_null();
6244 if (dest_k != nullptr && dest_k->is_array_klass()) {
6245 could_have_dest = true;
6246 }
6247 }
6248 if (could_have_src && could_have_dest) {
6249 // If we can have both exact types, emit the missing guards
6250 if (could_have_src && !src_spec) {
6251 src = maybe_cast_profiled_obj(src, src_k, true);
6252 src_type = _gvn.type(src);
6253 top_src = src_type->isa_aryptr();
6254 }
6255 if (could_have_dest && !dest_spec) {
6256 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6257 dest_type = _gvn.type(dest);
6258 top_dest = dest_type->isa_aryptr();
6259 }
6260 }
6261 }
6262 }
6263
6264 ciMethod* trap_method = method();
6265 int trap_bci = bci();
6266 if (saved_jvms_before_guards != nullptr) {
6267 trap_method = alloc->jvms()->method();
6268 trap_bci = alloc->jvms()->bci();
6269 }
6270
6271 bool negative_length_guard_generated = false;
6272
6273 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6274 can_emit_guards && !src->is_top() && !dest->is_top()) {
6275 // validate arguments: enables transformation the ArrayCopyNode
6276 validated = true;
6277
6278 RegionNode* slow_region = new RegionNode(1);
6279 record_for_igvn(slow_region);
6280
6281 // (1) src and dest are arrays.
6282 generate_non_array_guard(load_object_klass(src), slow_region);
6283 generate_non_array_guard(load_object_klass(dest), slow_region);
6284
6285 // (2) src and dest arrays must have elements of the same BasicType
6286 // done at macro expansion or at Ideal transformation time
6287
6288 // (4) src_offset must not be negative.
6289 generate_negative_guard(src_offset, slow_region);
6290
6291 // (5) dest_offset must not be negative.
6292 generate_negative_guard(dest_offset, slow_region);
6293
6294 // (7) src_offset + length must not exceed length of src.
6295 generate_limit_guard(src_offset, length,
6296 load_array_length(src),
6297 slow_region);
6298
6299 // (8) dest_offset + length must not exceed length of dest.
6300 generate_limit_guard(dest_offset, length,
6301 load_array_length(dest),
6302 slow_region);
6303
6304 // (6) length must not be negative.
6305 // This is also checked in generate_arraycopy() during macro expansion, but
6306 // we also have to check it here for the case where the ArrayCopyNode will
6307 // be eliminated by Escape Analysis.
6308 if (EliminateAllocations) {
6309 generate_negative_guard(length, slow_region);
6310 negative_length_guard_generated = true;
6311 }
6312
6313 // (9) each element of an oop array must be assignable
6314 Node* dest_klass = load_object_klass(dest);
6315 if (src != dest) {
6316 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6317 slow_region->add_req(not_subtype_ctrl);
6318 }
6319
6320 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
6321 const Type* toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6322 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6323 src_type = _gvn.type(src);
6324 top_src = src_type->isa_aryptr();
6325
6326 // Handle flat inline type arrays (null-free arrays are handled by the subtype check above)
6327 if (!stopped() && UseFlatArray) {
6328 // If dest is flat, src must be flat as well (guaranteed by src <: dest check). Handle flat src here.
6329 assert(top_dest == nullptr || !top_dest->is_flat() || top_src->is_flat(), "src array must be flat");
6330 if (top_src != nullptr && top_src->is_flat()) {
6331 // Src is flat, check that dest is flat as well
6332 if (top_dest != nullptr && !top_dest->is_flat()) {
6333 generate_fair_guard(flat_array_test(dest_klass, /* flat = */ false), slow_region);
6334 // Since dest is flat and src <: dest, dest must have the same type as src.
6335 top_dest = top_src->cast_to_exactness(false);
6336 assert(top_dest->is_flat(), "dest must be flat");
6337 dest = _gvn.transform(new CheckCastPPNode(control(), dest, top_dest));
6338 }
6339 } else if (top_src == nullptr || !top_src->is_not_flat()) {
6340 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
6341 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
6342 assert(top_dest == nullptr || !top_dest->is_flat(), "dest array must not be flat");
6343 generate_fair_guard(flat_array_test(src), slow_region);
6344 if (top_src != nullptr) {
6345 top_src = top_src->cast_to_not_flat();
6346 src = _gvn.transform(new CheckCastPPNode(control(), src, top_src));
6347 }
6348 }
6349 }
6350
6351 {
6352 PreserveJVMState pjvms(this);
6353 set_control(_gvn.transform(slow_region));
6354 uncommon_trap(Deoptimization::Reason_intrinsic,
6355 Deoptimization::Action_make_not_entrant);
6356 assert(stopped(), "Should be stopped");
6357 }
6358 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6359 }
6360
6361 if (stopped()) {
6362 return true;
6363 }
6364
6365 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6366 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6367 // so the compiler has a chance to eliminate them: during macro expansion,
6368 // we have to set their control (CastPP nodes are eliminated).
6369 load_object_klass(src), load_object_klass(dest),
6370 load_array_length(src), load_array_length(dest));
6371
6372 ac->set_arraycopy(validated);
6373
6374 Node* n = _gvn.transform(ac);
6375 if (n == ac) {
6376 ac->connect_outputs(this);
6377 } else {
6378 assert(validated, "shouldn't transform if all arguments not validated");
6379 set_all_memory(n);
6380 }
6381 clear_upper_avx();
6382
6383
6384 return true;
6385 }
6386
6387
6388 // Helper function which determines if an arraycopy immediately follows
6389 // an allocation, with no intervening tests or other escapes for the object.
6390 AllocateArrayNode*
6391 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6392 if (stopped()) return nullptr; // no fast path
6393 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6394
6395 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6396 if (alloc == nullptr) return nullptr;
6397
6398 Node* rawmem = memory(Compile::AliasIdxRaw);
6399 // Is the allocation's memory state untouched?
6400 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6401 // Bail out if there have been raw-memory effects since the allocation.
6402 // (Example: There might have been a call or safepoint.)
6403 return nullptr;
6404 }
6405 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6406 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6407 return nullptr;
6408 }
6409
6410 // There must be no unexpected observers of this allocation.
6411 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6412 Node* obs = ptr->fast_out(i);
6413 if (obs != this->map()) {
6414 return nullptr;
6415 }
6416 }
6417
6418 // This arraycopy must unconditionally follow the allocation of the ptr.
6419 Node* alloc_ctl = ptr->in(0);
6420 Node* ctl = control();
6421 while (ctl != alloc_ctl) {
6422 // There may be guards which feed into the slow_region.
6423 // Any other control flow means that we might not get a chance
6424 // to finish initializing the allocated object.
6425 // Various low-level checks bottom out in uncommon traps. These
6426 // are considered safe since we've already checked above that
6427 // there is no unexpected observer of this allocation.
6428 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6429 assert(ctl->in(0)->is_If(), "must be If");
6430 ctl = ctl->in(0)->in(0);
6431 } else {
6432 return nullptr;
6433 }
6434 }
6435
6436 // If we get this far, we have an allocation which immediately
6437 // precedes the arraycopy, and we can take over zeroing the new object.
6438 // The arraycopy will finish the initialization, and provide
6439 // a new control state to which we will anchor the destination pointer.
6440
6441 return alloc;
6442 }
6443
6444 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6445 if (node->is_IfProj()) {
6446 Node* other_proj = node->as_IfProj()->other_if_proj();
6447 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6448 Node* obs = other_proj->fast_out(j);
6449 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6450 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6451 return obs->as_CallStaticJava();
6452 }
6453 }
6454 }
6455 return nullptr;
6456 }
6457
6458 //-------------inline_encodeISOArray-----------------------------------
6459 // encode char[] to byte[] in ISO_8859_1 or ASCII
6460 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6461 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6462 // no receiver since it is static method
6463 Node *src = argument(0);
6464 Node *src_offset = argument(1);
6465 Node *dst = argument(2);
6466 Node *dst_offset = argument(3);
6467 Node *length = argument(4);
6468
6469 src = must_be_not_null(src, true);
6470 dst = must_be_not_null(dst, true);
6471
6472 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6473 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6474 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6475 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6476 // failed array check
6477 return false;
6478 }
6479
6480 // Figure out the size and type of the elements we will be copying.
6481 BasicType src_elem = src_type->elem()->array_element_basic_type();
6482 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6483 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6484 return false;
6485 }
6486
6487 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6488 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6489 // 'src_start' points to src array + scaled offset
6490 // 'dst_start' points to dst array + scaled offset
6491
6492 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6493 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6494 enc = _gvn.transform(enc);
6495 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6496 set_memory(res_mem, mtype);
6497 set_result(enc);
6498 clear_upper_avx();
6499
6500 return true;
6501 }
6502
6503 //-------------inline_multiplyToLen-----------------------------------
6504 bool LibraryCallKit::inline_multiplyToLen() {
6505 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6506
6507 address stubAddr = StubRoutines::multiplyToLen();
6508 if (stubAddr == nullptr) {
6509 return false; // Intrinsic's stub is not implemented on this platform
6510 }
6511 const char* stubName = "multiplyToLen";
6512
6513 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6514
6515 // no receiver because it is a static method
6516 Node* x = argument(0);
6517 Node* xlen = argument(1);
6518 Node* y = argument(2);
6519 Node* ylen = argument(3);
6520 Node* z = argument(4);
6521
6522 x = must_be_not_null(x, true);
6523 y = must_be_not_null(y, true);
6524
6525 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6526 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6527 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6528 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6529 // failed array check
6530 return false;
6531 }
6532
6533 BasicType x_elem = x_type->elem()->array_element_basic_type();
6534 BasicType y_elem = y_type->elem()->array_element_basic_type();
6535 if (x_elem != T_INT || y_elem != T_INT) {
6536 return false;
6537 }
6538
6539 Node* x_start = array_element_address(x, intcon(0), x_elem);
6540 Node* y_start = array_element_address(y, intcon(0), y_elem);
6541 // 'x_start' points to x array + scaled xlen
6542 // 'y_start' points to y array + scaled ylen
6543
6544 Node* z_start = array_element_address(z, intcon(0), T_INT);
6545
6546 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6547 OptoRuntime::multiplyToLen_Type(),
6548 stubAddr, stubName, TypePtr::BOTTOM,
6549 x_start, xlen, y_start, ylen, z_start);
6550
6551 C->set_has_split_ifs(true); // Has chance for split-if optimization
6552 set_result(z);
6553 return true;
6554 }
6555
6556 //-------------inline_squareToLen------------------------------------
6557 bool LibraryCallKit::inline_squareToLen() {
6558 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6559
6560 address stubAddr = StubRoutines::squareToLen();
6561 if (stubAddr == nullptr) {
6562 return false; // Intrinsic's stub is not implemented on this platform
6563 }
6564 const char* stubName = "squareToLen";
6565
6566 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6567
6568 Node* x = argument(0);
6569 Node* len = argument(1);
6570 Node* z = argument(2);
6571 Node* zlen = argument(3);
6572
6573 x = must_be_not_null(x, true);
6574 z = must_be_not_null(z, true);
6575
6576 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6577 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6578 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6579 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6580 // failed array check
6581 return false;
6582 }
6583
6584 BasicType x_elem = x_type->elem()->array_element_basic_type();
6585 BasicType z_elem = z_type->elem()->array_element_basic_type();
6586 if (x_elem != T_INT || z_elem != T_INT) {
6587 return false;
6588 }
6589
6590
6591 Node* x_start = array_element_address(x, intcon(0), x_elem);
6592 Node* z_start = array_element_address(z, intcon(0), z_elem);
6593
6594 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6595 OptoRuntime::squareToLen_Type(),
6596 stubAddr, stubName, TypePtr::BOTTOM,
6597 x_start, len, z_start, zlen);
6598
6599 set_result(z);
6600 return true;
6601 }
6602
6603 //-------------inline_mulAdd------------------------------------------
6604 bool LibraryCallKit::inline_mulAdd() {
6605 assert(UseMulAddIntrinsic, "not implemented on this platform");
6606
6607 address stubAddr = StubRoutines::mulAdd();
6608 if (stubAddr == nullptr) {
6609 return false; // Intrinsic's stub is not implemented on this platform
6610 }
6611 const char* stubName = "mulAdd";
6612
6613 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6614
6615 Node* out = argument(0);
6616 Node* in = argument(1);
6617 Node* offset = argument(2);
6618 Node* len = argument(3);
6619 Node* k = argument(4);
6620
6621 in = must_be_not_null(in, true);
6622 out = must_be_not_null(out, true);
6623
6624 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6625 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6626 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6627 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6628 // failed array check
6629 return false;
6630 }
6631
6632 BasicType out_elem = out_type->elem()->array_element_basic_type();
6633 BasicType in_elem = in_type->elem()->array_element_basic_type();
6634 if (out_elem != T_INT || in_elem != T_INT) {
6635 return false;
6636 }
6637
6638 Node* outlen = load_array_length(out);
6639 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6640 Node* out_start = array_element_address(out, intcon(0), out_elem);
6641 Node* in_start = array_element_address(in, intcon(0), in_elem);
6642
6643 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6644 OptoRuntime::mulAdd_Type(),
6645 stubAddr, stubName, TypePtr::BOTTOM,
6646 out_start,in_start, new_offset, len, k);
6647 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6648 set_result(result);
6649 return true;
6650 }
6651
6652 //-------------inline_montgomeryMultiply-----------------------------------
6653 bool LibraryCallKit::inline_montgomeryMultiply() {
6654 address stubAddr = StubRoutines::montgomeryMultiply();
6655 if (stubAddr == nullptr) {
6656 return false; // Intrinsic's stub is not implemented on this platform
6657 }
6658
6659 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6660 const char* stubName = "montgomery_multiply";
6661
6662 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6663
6664 Node* a = argument(0);
6665 Node* b = argument(1);
6666 Node* n = argument(2);
6667 Node* len = argument(3);
6668 Node* inv = argument(4);
6669 Node* m = argument(6);
6670
6671 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6672 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6673 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6674 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6675 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6676 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6677 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6678 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6679 // failed array check
6680 return false;
6681 }
6682
6683 BasicType a_elem = a_type->elem()->array_element_basic_type();
6684 BasicType b_elem = b_type->elem()->array_element_basic_type();
6685 BasicType n_elem = n_type->elem()->array_element_basic_type();
6686 BasicType m_elem = m_type->elem()->array_element_basic_type();
6687 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6688 return false;
6689 }
6690
6691 // Make the call
6692 {
6693 Node* a_start = array_element_address(a, intcon(0), a_elem);
6694 Node* b_start = array_element_address(b, intcon(0), b_elem);
6695 Node* n_start = array_element_address(n, intcon(0), n_elem);
6696 Node* m_start = array_element_address(m, intcon(0), m_elem);
6697
6698 Node* call = make_runtime_call(RC_LEAF,
6699 OptoRuntime::montgomeryMultiply_Type(),
6700 stubAddr, stubName, TypePtr::BOTTOM,
6701 a_start, b_start, n_start, len, inv, top(),
6702 m_start);
6703 set_result(m);
6704 }
6705
6706 return true;
6707 }
6708
6709 bool LibraryCallKit::inline_montgomerySquare() {
6710 address stubAddr = StubRoutines::montgomerySquare();
6711 if (stubAddr == nullptr) {
6712 return false; // Intrinsic's stub is not implemented on this platform
6713 }
6714
6715 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6716 const char* stubName = "montgomery_square";
6717
6718 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6719
6720 Node* a = argument(0);
6721 Node* n = argument(1);
6722 Node* len = argument(2);
6723 Node* inv = argument(3);
6724 Node* m = argument(5);
6725
6726 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6727 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6728 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6729 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6730 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6731 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6732 // failed array check
6733 return false;
6734 }
6735
6736 BasicType a_elem = a_type->elem()->array_element_basic_type();
6737 BasicType n_elem = n_type->elem()->array_element_basic_type();
6738 BasicType m_elem = m_type->elem()->array_element_basic_type();
6739 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6740 return false;
6741 }
6742
6743 // Make the call
6744 {
6745 Node* a_start = array_element_address(a, intcon(0), a_elem);
6746 Node* n_start = array_element_address(n, intcon(0), n_elem);
6747 Node* m_start = array_element_address(m, intcon(0), m_elem);
6748
6749 Node* call = make_runtime_call(RC_LEAF,
6750 OptoRuntime::montgomerySquare_Type(),
6751 stubAddr, stubName, TypePtr::BOTTOM,
6752 a_start, n_start, len, inv, top(),
6753 m_start);
6754 set_result(m);
6755 }
6756
6757 return true;
6758 }
6759
6760 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6761 address stubAddr = nullptr;
6762 const char* stubName = nullptr;
6763
6764 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6765 if (stubAddr == nullptr) {
6766 return false; // Intrinsic's stub is not implemented on this platform
6767 }
6768
6769 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6770
6771 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6772
6773 Node* newArr = argument(0);
6774 Node* oldArr = argument(1);
6775 Node* newIdx = argument(2);
6776 Node* shiftCount = argument(3);
6777 Node* numIter = argument(4);
6778
6779 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6780 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6781 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6782 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6783 return false;
6784 }
6785
6786 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6787 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6788 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6789 return false;
6790 }
6791
6792 // Make the call
6793 {
6794 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6795 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6796
6797 Node* call = make_runtime_call(RC_LEAF,
6798 OptoRuntime::bigIntegerShift_Type(),
6799 stubAddr,
6800 stubName,
6801 TypePtr::BOTTOM,
6802 newArr_start,
6803 oldArr_start,
6804 newIdx,
6805 shiftCount,
6806 numIter);
6807 }
6808
6809 return true;
6810 }
6811
6812 //-------------inline_vectorizedMismatch------------------------------
6813 bool LibraryCallKit::inline_vectorizedMismatch() {
6814 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6815
6816 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6817 Node* obja = argument(0); // Object
6818 Node* aoffset = argument(1); // long
6819 Node* objb = argument(3); // Object
6820 Node* boffset = argument(4); // long
6821 Node* length = argument(6); // int
6822 Node* scale = argument(7); // int
6823
6824 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6825 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6826 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6827 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6828 scale == top()) {
6829 return false; // failed input validation
6830 }
6831
6832 Node* obja_adr = make_unsafe_address(obja, aoffset);
6833 Node* objb_adr = make_unsafe_address(objb, boffset);
6834
6835 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6836 //
6837 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6838 // if (length <= inline_limit) {
6839 // inline_path:
6840 // vmask = VectorMaskGen length
6841 // vload1 = LoadVectorMasked obja, vmask
6842 // vload2 = LoadVectorMasked objb, vmask
6843 // result1 = VectorCmpMasked vload1, vload2, vmask
6844 // } else {
6845 // call_stub_path:
6846 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6847 // }
6848 // exit_block:
6849 // return Phi(result1, result2);
6850 //
6851 enum { inline_path = 1, // input is small enough to process it all at once
6852 stub_path = 2, // input is too large; call into the VM
6853 PATH_LIMIT = 3
6854 };
6855
6856 Node* exit_block = new RegionNode(PATH_LIMIT);
6857 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6858 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6859
6860 Node* call_stub_path = control();
6861
6862 BasicType elem_bt = T_ILLEGAL;
6863
6864 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6865 if (scale_t->is_con()) {
6866 switch (scale_t->get_con()) {
6867 case 0: elem_bt = T_BYTE; break;
6868 case 1: elem_bt = T_SHORT; break;
6869 case 2: elem_bt = T_INT; break;
6870 case 3: elem_bt = T_LONG; break;
6871
6872 default: elem_bt = T_ILLEGAL; break; // not supported
6873 }
6874 }
6875
6876 int inline_limit = 0;
6877 bool do_partial_inline = false;
6878
6879 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6880 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6881 do_partial_inline = inline_limit >= 16;
6882 }
6883
6884 if (do_partial_inline) {
6885 assert(elem_bt != T_ILLEGAL, "sanity");
6886
6887 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6888 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6889 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6890
6891 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6892 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6893 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6894
6895 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6896
6897 if (!stopped()) {
6898 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6899
6900 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6901 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6902 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6903 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6904
6905 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6906 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6907 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6908 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6909
6910 exit_block->init_req(inline_path, control());
6911 memory_phi->init_req(inline_path, map()->memory());
6912 result_phi->init_req(inline_path, result);
6913
6914 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6915 clear_upper_avx();
6916 }
6917 }
6918 }
6919
6920 if (call_stub_path != nullptr) {
6921 set_control(call_stub_path);
6922
6923 Node* call = make_runtime_call(RC_LEAF,
6924 OptoRuntime::vectorizedMismatch_Type(),
6925 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6926 obja_adr, objb_adr, length, scale);
6927
6928 exit_block->init_req(stub_path, control());
6929 memory_phi->init_req(stub_path, map()->memory());
6930 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6931 }
6932
6933 exit_block = _gvn.transform(exit_block);
6934 memory_phi = _gvn.transform(memory_phi);
6935 result_phi = _gvn.transform(result_phi);
6936
6937 set_control(exit_block);
6938 set_all_memory(memory_phi);
6939 set_result(result_phi);
6940
6941 return true;
6942 }
6943
6944 //------------------------------inline_vectorizedHashcode----------------------------
6945 bool LibraryCallKit::inline_vectorizedHashCode() {
6946 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6947
6948 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6949 Node* array = argument(0);
6950 Node* offset = argument(1);
6951 Node* length = argument(2);
6952 Node* initialValue = argument(3);
6953 Node* basic_type = argument(4);
6954
6955 if (basic_type == top()) {
6956 return false; // failed input validation
6957 }
6958
6959 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6960 if (!basic_type_t->is_con()) {
6961 return false; // Only intrinsify if mode argument is constant
6962 }
6963
6964 array = must_be_not_null(array, true);
6965
6966 BasicType bt = (BasicType)basic_type_t->get_con();
6967
6968 // Resolve address of first element
6969 Node* array_start = array_element_address(array, offset, bt);
6970
6971 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6972 array_start, length, initialValue, basic_type)));
6973 clear_upper_avx();
6974
6975 return true;
6976 }
6977
6978 /**
6979 * Calculate CRC32 for byte.
6980 * int java.util.zip.CRC32.update(int crc, int b)
6981 */
6982 bool LibraryCallKit::inline_updateCRC32() {
6983 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6984 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6985 // no receiver since it is static method
6986 Node* crc = argument(0); // type: int
6987 Node* b = argument(1); // type: int
6988
6989 /*
6990 * int c = ~ crc;
6991 * b = timesXtoThe32[(b ^ c) & 0xFF];
6992 * b = b ^ (c >>> 8);
6993 * crc = ~b;
6994 */
6995
6996 Node* M1 = intcon(-1);
6997 crc = _gvn.transform(new XorINode(crc, M1));
6998 Node* result = _gvn.transform(new XorINode(crc, b));
6999 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
7000
7001 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
7002 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
7003 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
7004 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
7005
7006 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
7007 result = _gvn.transform(new XorINode(crc, result));
7008 result = _gvn.transform(new XorINode(result, M1));
7009 set_result(result);
7010 return true;
7011 }
7012
7013 /**
7014 * Calculate CRC32 for byte[] array.
7015 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
7016 */
7017 bool LibraryCallKit::inline_updateBytesCRC32() {
7018 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
7019 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7020 // no receiver since it is static method
7021 Node* crc = argument(0); // type: int
7022 Node* src = argument(1); // type: oop
7023 Node* offset = argument(2); // type: int
7024 Node* length = argument(3); // type: int
7025
7026 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7027 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7028 // failed array check
7029 return false;
7030 }
7031
7032 // Figure out the size and type of the elements we will be copying.
7033 BasicType src_elem = src_type->elem()->array_element_basic_type();
7034 if (src_elem != T_BYTE) {
7035 return false;
7036 }
7037
7038 // 'src_start' points to src array + scaled offset
7039 src = must_be_not_null(src, true);
7040 Node* src_start = array_element_address(src, offset, src_elem);
7041
7042 // We assume that range check is done by caller.
7043 // TODO: generate range check (offset+length < src.length) in debug VM.
7044
7045 // Call the stub.
7046 address stubAddr = StubRoutines::updateBytesCRC32();
7047 const char *stubName = "updateBytesCRC32";
7048
7049 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7050 stubAddr, stubName, TypePtr::BOTTOM,
7051 crc, src_start, length);
7052 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7053 set_result(result);
7054 return true;
7055 }
7056
7057 /**
7058 * Calculate CRC32 for ByteBuffer.
7059 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
7060 */
7061 bool LibraryCallKit::inline_updateByteBufferCRC32() {
7062 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
7063 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7064 // no receiver since it is static method
7065 Node* crc = argument(0); // type: int
7066 Node* src = argument(1); // type: long
7067 Node* offset = argument(3); // type: int
7068 Node* length = argument(4); // type: int
7069
7070 src = ConvL2X(src); // adjust Java long to machine word
7071 Node* base = _gvn.transform(new CastX2PNode(src));
7072 offset = ConvI2X(offset);
7073
7074 // 'src_start' points to src array + scaled offset
7075 Node* src_start = basic_plus_adr(top(), base, offset);
7076
7077 // Call the stub.
7078 address stubAddr = StubRoutines::updateBytesCRC32();
7079 const char *stubName = "updateBytesCRC32";
7080
7081 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7082 stubAddr, stubName, TypePtr::BOTTOM,
7083 crc, src_start, length);
7084 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7085 set_result(result);
7086 return true;
7087 }
7088
7089 //------------------------------get_table_from_crc32c_class-----------------------
7090 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
7091 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
7092 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
7093
7094 return table;
7095 }
7096
7097 //------------------------------inline_updateBytesCRC32C-----------------------
7098 //
7099 // Calculate CRC32C for byte[] array.
7100 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
7101 //
7102 bool LibraryCallKit::inline_updateBytesCRC32C() {
7103 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7104 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7105 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7106 // no receiver since it is a static method
7107 Node* crc = argument(0); // type: int
7108 Node* src = argument(1); // type: oop
7109 Node* offset = argument(2); // type: int
7110 Node* end = argument(3); // type: int
7111
7112 Node* length = _gvn.transform(new SubINode(end, offset));
7113
7114 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7115 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7116 // failed array check
7117 return false;
7118 }
7119
7120 // Figure out the size and type of the elements we will be copying.
7121 BasicType src_elem = src_type->elem()->array_element_basic_type();
7122 if (src_elem != T_BYTE) {
7123 return false;
7124 }
7125
7126 // 'src_start' points to src array + scaled offset
7127 src = must_be_not_null(src, true);
7128 Node* src_start = array_element_address(src, offset, src_elem);
7129
7130 // static final int[] byteTable in class CRC32C
7131 Node* table = get_table_from_crc32c_class(callee()->holder());
7132 table = must_be_not_null(table, true);
7133 Node* table_start = array_element_address(table, intcon(0), T_INT);
7134
7135 // We assume that range check is done by caller.
7136 // TODO: generate range check (offset+length < src.length) in debug VM.
7137
7138 // Call the stub.
7139 address stubAddr = StubRoutines::updateBytesCRC32C();
7140 const char *stubName = "updateBytesCRC32C";
7141
7142 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7143 stubAddr, stubName, TypePtr::BOTTOM,
7144 crc, src_start, length, table_start);
7145 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7146 set_result(result);
7147 return true;
7148 }
7149
7150 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
7151 //
7152 // Calculate CRC32C for DirectByteBuffer.
7153 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
7154 //
7155 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
7156 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7157 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
7158 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7159 // no receiver since it is a static method
7160 Node* crc = argument(0); // type: int
7161 Node* src = argument(1); // type: long
7162 Node* offset = argument(3); // type: int
7163 Node* end = argument(4); // type: int
7164
7165 Node* length = _gvn.transform(new SubINode(end, offset));
7166
7167 src = ConvL2X(src); // adjust Java long to machine word
7168 Node* base = _gvn.transform(new CastX2PNode(src));
7169 offset = ConvI2X(offset);
7170
7171 // 'src_start' points to src array + scaled offset
7172 Node* src_start = basic_plus_adr(top(), base, offset);
7173
7174 // static final int[] byteTable in class CRC32C
7175 Node* table = get_table_from_crc32c_class(callee()->holder());
7176 table = must_be_not_null(table, true);
7177 Node* table_start = array_element_address(table, intcon(0), T_INT);
7178
7179 // Call the stub.
7180 address stubAddr = StubRoutines::updateBytesCRC32C();
7181 const char *stubName = "updateBytesCRC32C";
7182
7183 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7184 stubAddr, stubName, TypePtr::BOTTOM,
7185 crc, src_start, length, table_start);
7186 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7187 set_result(result);
7188 return true;
7189 }
7190
7191 //------------------------------inline_updateBytesAdler32----------------------
7192 //
7193 // Calculate Adler32 checksum for byte[] array.
7194 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
7195 //
7196 bool LibraryCallKit::inline_updateBytesAdler32() {
7197 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7198 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7199 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7200 // no receiver since it is static method
7201 Node* crc = argument(0); // type: int
7202 Node* src = argument(1); // type: oop
7203 Node* offset = argument(2); // type: int
7204 Node* length = argument(3); // type: int
7205
7206 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7207 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7208 // failed array check
7209 return false;
7210 }
7211
7212 // Figure out the size and type of the elements we will be copying.
7213 BasicType src_elem = src_type->elem()->array_element_basic_type();
7214 if (src_elem != T_BYTE) {
7215 return false;
7216 }
7217
7218 // 'src_start' points to src array + scaled offset
7219 Node* src_start = array_element_address(src, offset, src_elem);
7220
7221 // We assume that range check is done by caller.
7222 // TODO: generate range check (offset+length < src.length) in debug VM.
7223
7224 // Call the stub.
7225 address stubAddr = StubRoutines::updateBytesAdler32();
7226 const char *stubName = "updateBytesAdler32";
7227
7228 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7229 stubAddr, stubName, TypePtr::BOTTOM,
7230 crc, src_start, length);
7231 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7232 set_result(result);
7233 return true;
7234 }
7235
7236 //------------------------------inline_updateByteBufferAdler32---------------
7237 //
7238 // Calculate Adler32 checksum for DirectByteBuffer.
7239 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
7240 //
7241 bool LibraryCallKit::inline_updateByteBufferAdler32() {
7242 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7243 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7244 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7245 // no receiver since it is static method
7246 Node* crc = argument(0); // type: int
7247 Node* src = argument(1); // type: long
7248 Node* offset = argument(3); // type: int
7249 Node* length = argument(4); // type: int
7250
7251 src = ConvL2X(src); // adjust Java long to machine word
7252 Node* base = _gvn.transform(new CastX2PNode(src));
7253 offset = ConvI2X(offset);
7254
7255 // 'src_start' points to src array + scaled offset
7256 Node* src_start = basic_plus_adr(top(), base, offset);
7257
7258 // Call the stub.
7259 address stubAddr = StubRoutines::updateBytesAdler32();
7260 const char *stubName = "updateBytesAdler32";
7261
7262 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7263 stubAddr, stubName, TypePtr::BOTTOM,
7264 crc, src_start, length);
7265
7266 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7267 set_result(result);
7268 return true;
7269 }
7270
7271 //----------------------------inline_reference_get----------------------------
7272 // public T java.lang.ref.Reference.get();
7273 bool LibraryCallKit::inline_reference_get() {
7274 const int referent_offset = java_lang_ref_Reference::referent_offset();
7275
7276 // Get the argument:
7277 Node* reference_obj = null_check_receiver();
7278 if (stopped()) return true;
7279
7280 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7281 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7282 decorators, /*is_static*/ false, nullptr);
7283 if (result == nullptr) return false;
7284
7285 // Add memory barrier to prevent commoning reads from this field
7286 // across safepoint since GC can change its value.
7287 insert_mem_bar(Op_MemBarCPUOrder);
7288
7289 set_result(result);
7290 return true;
7291 }
7292
7293 //----------------------------inline_reference_refersTo0----------------------------
7294 // bool java.lang.ref.Reference.refersTo0();
7295 // bool java.lang.ref.PhantomReference.refersTo0();
7296 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7297 // Get arguments:
7298 Node* reference_obj = null_check_receiver();
7299 Node* other_obj = argument(1);
7300 if (stopped()) return true;
7301
7302 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7303 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7304 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7305 decorators, /*is_static*/ false, nullptr);
7306 if (referent == nullptr) return false;
7307
7308 // Add memory barrier to prevent commoning reads from this field
7309 // across safepoint since GC can change its value.
7310 insert_mem_bar(Op_MemBarCPUOrder);
7311
7312 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7313 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7314 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7315
7316 RegionNode* region = new RegionNode(3);
7317 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7318
7319 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7320 region->init_req(1, if_true);
7321 phi->init_req(1, intcon(1));
7322
7323 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7324 region->init_req(2, if_false);
7325 phi->init_req(2, intcon(0));
7326
7327 set_control(_gvn.transform(region));
7328 record_for_igvn(region);
7329 set_result(_gvn.transform(phi));
7330 return true;
7331 }
7332
7333 //----------------------------inline_reference_clear0----------------------------
7334 // void java.lang.ref.Reference.clear0();
7335 // void java.lang.ref.PhantomReference.clear0();
7336 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7337 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7338
7339 // Get arguments
7340 Node* reference_obj = null_check_receiver();
7341 if (stopped()) return true;
7342
7343 // Common access parameters
7344 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7345 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7346 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7347 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7348 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7349
7350 Node* referent = access_load_at(reference_obj,
7351 referent_field_addr,
7352 referent_field_addr_type,
7353 val_type,
7354 T_OBJECT,
7355 decorators);
7356
7357 IdealKit ideal(this);
7358 #define __ ideal.
7359 __ if_then(referent, BoolTest::ne, null());
7360 sync_kit(ideal);
7361 access_store_at(reference_obj,
7362 referent_field_addr,
7363 referent_field_addr_type,
7364 null(),
7365 val_type,
7366 T_OBJECT,
7367 decorators);
7368 __ sync_kit(this);
7369 __ end_if();
7370 final_sync(ideal);
7371 #undef __
7372
7373 return true;
7374 }
7375
7376 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7377 DecoratorSet decorators, bool is_static,
7378 ciInstanceKlass* fromKls) {
7379 if (fromKls == nullptr) {
7380 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7381 assert(tinst != nullptr, "obj is null");
7382 assert(tinst->is_loaded(), "obj is not loaded");
7383 fromKls = tinst->instance_klass();
7384 } else {
7385 assert(is_static, "only for static field access");
7386 }
7387 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7388 ciSymbol::make(fieldTypeString),
7389 is_static);
7390
7391 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7392 if (field == nullptr) return (Node *) nullptr;
7393
7394 if (is_static) {
7395 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7396 fromObj = makecon(tip);
7397 }
7398
7399 // Next code copied from Parse::do_get_xxx():
7400
7401 // Compute address and memory type.
7402 int offset = field->offset_in_bytes();
7403 bool is_vol = field->is_volatile();
7404 ciType* field_klass = field->type();
7405 assert(field_klass->is_loaded(), "should be loaded");
7406 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7407 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7408 BasicType bt = field->layout_type();
7409
7410 // Build the resultant type of the load
7411 const Type *type;
7412 if (bt == T_OBJECT) {
7413 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7414 } else {
7415 type = Type::get_const_basic_type(bt);
7416 }
7417
7418 if (is_vol) {
7419 decorators |= MO_SEQ_CST;
7420 }
7421
7422 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7423 }
7424
7425 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7426 bool is_exact /* true */, bool is_static /* false */,
7427 ciInstanceKlass * fromKls /* nullptr */) {
7428 if (fromKls == nullptr) {
7429 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7430 assert(tinst != nullptr, "obj is null");
7431 assert(tinst->is_loaded(), "obj is not loaded");
7432 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7433 fromKls = tinst->instance_klass();
7434 }
7435 else {
7436 assert(is_static, "only for static field access");
7437 }
7438 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7439 ciSymbol::make(fieldTypeString),
7440 is_static);
7441
7442 assert(field != nullptr, "undefined field");
7443 assert(!field->is_volatile(), "not defined for volatile fields");
7444
7445 if (is_static) {
7446 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7447 fromObj = makecon(tip);
7448 }
7449
7450 // Next code copied from Parse::do_get_xxx():
7451
7452 // Compute address and memory type.
7453 int offset = field->offset_in_bytes();
7454 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7455
7456 return adr;
7457 }
7458
7459 //------------------------------inline_aescrypt_Block-----------------------
7460 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7461 address stubAddr = nullptr;
7462 const char *stubName;
7463 assert(UseAES, "need AES instruction support");
7464
7465 switch(id) {
7466 case vmIntrinsics::_aescrypt_encryptBlock:
7467 stubAddr = StubRoutines::aescrypt_encryptBlock();
7468 stubName = "aescrypt_encryptBlock";
7469 break;
7470 case vmIntrinsics::_aescrypt_decryptBlock:
7471 stubAddr = StubRoutines::aescrypt_decryptBlock();
7472 stubName = "aescrypt_decryptBlock";
7473 break;
7474 default:
7475 break;
7476 }
7477 if (stubAddr == nullptr) return false;
7478
7479 Node* aescrypt_object = argument(0);
7480 Node* src = argument(1);
7481 Node* src_offset = argument(2);
7482 Node* dest = argument(3);
7483 Node* dest_offset = argument(4);
7484
7485 src = must_be_not_null(src, true);
7486 dest = must_be_not_null(dest, true);
7487
7488 // (1) src and dest are arrays.
7489 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7490 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7491 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7492 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7493
7494 // for the quick and dirty code we will skip all the checks.
7495 // we are just trying to get the call to be generated.
7496 Node* src_start = src;
7497 Node* dest_start = dest;
7498 if (src_offset != nullptr || dest_offset != nullptr) {
7499 assert(src_offset != nullptr && dest_offset != nullptr, "");
7500 src_start = array_element_address(src, src_offset, T_BYTE);
7501 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7502 }
7503
7504 // now need to get the start of its expanded key array
7505 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7506 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7507 if (k_start == nullptr) return false;
7508
7509 // Call the stub.
7510 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7511 stubAddr, stubName, TypePtr::BOTTOM,
7512 src_start, dest_start, k_start);
7513
7514 return true;
7515 }
7516
7517 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7518 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7519 address stubAddr = nullptr;
7520 const char *stubName = nullptr;
7521
7522 assert(UseAES, "need AES instruction support");
7523
7524 switch(id) {
7525 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7526 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7527 stubName = "cipherBlockChaining_encryptAESCrypt";
7528 break;
7529 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7530 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7531 stubName = "cipherBlockChaining_decryptAESCrypt";
7532 break;
7533 default:
7534 break;
7535 }
7536 if (stubAddr == nullptr) return false;
7537
7538 Node* cipherBlockChaining_object = argument(0);
7539 Node* src = argument(1);
7540 Node* src_offset = argument(2);
7541 Node* len = argument(3);
7542 Node* dest = argument(4);
7543 Node* dest_offset = argument(5);
7544
7545 src = must_be_not_null(src, false);
7546 dest = must_be_not_null(dest, false);
7547
7548 // (1) src and dest are arrays.
7549 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7550 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7551 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7552 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7553
7554 // checks are the responsibility of the caller
7555 Node* src_start = src;
7556 Node* dest_start = dest;
7557 if (src_offset != nullptr || dest_offset != nullptr) {
7558 assert(src_offset != nullptr && dest_offset != nullptr, "");
7559 src_start = array_element_address(src, src_offset, T_BYTE);
7560 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7561 }
7562
7563 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7564 // (because of the predicated logic executed earlier).
7565 // so we cast it here safely.
7566 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7567
7568 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7569 if (embeddedCipherObj == nullptr) return false;
7570
7571 // cast it to what we know it will be at runtime
7572 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7573 assert(tinst != nullptr, "CBC obj is null");
7574 assert(tinst->is_loaded(), "CBC obj is not loaded");
7575 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7576 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7577
7578 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7579 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7580 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7581 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7582 aescrypt_object = _gvn.transform(aescrypt_object);
7583
7584 // we need to get the start of the aescrypt_object's expanded key array
7585 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7586 if (k_start == nullptr) return false;
7587
7588 // similarly, get the start address of the r vector
7589 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7590 if (objRvec == nullptr) return false;
7591 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7592
7593 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7594 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7595 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7596 stubAddr, stubName, TypePtr::BOTTOM,
7597 src_start, dest_start, k_start, r_start, len);
7598
7599 // return cipher length (int)
7600 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7601 set_result(retvalue);
7602 return true;
7603 }
7604
7605 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7606 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7607 address stubAddr = nullptr;
7608 const char *stubName = nullptr;
7609
7610 assert(UseAES, "need AES instruction support");
7611
7612 switch (id) {
7613 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7614 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7615 stubName = "electronicCodeBook_encryptAESCrypt";
7616 break;
7617 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7618 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7619 stubName = "electronicCodeBook_decryptAESCrypt";
7620 break;
7621 default:
7622 break;
7623 }
7624
7625 if (stubAddr == nullptr) return false;
7626
7627 Node* electronicCodeBook_object = argument(0);
7628 Node* src = argument(1);
7629 Node* src_offset = argument(2);
7630 Node* len = argument(3);
7631 Node* dest = argument(4);
7632 Node* dest_offset = argument(5);
7633
7634 // (1) src and dest are arrays.
7635 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7636 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7637 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7638 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7639
7640 // checks are the responsibility of the caller
7641 Node* src_start = src;
7642 Node* dest_start = dest;
7643 if (src_offset != nullptr || dest_offset != nullptr) {
7644 assert(src_offset != nullptr && dest_offset != nullptr, "");
7645 src_start = array_element_address(src, src_offset, T_BYTE);
7646 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7647 }
7648
7649 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7650 // (because of the predicated logic executed earlier).
7651 // so we cast it here safely.
7652 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7653
7654 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7655 if (embeddedCipherObj == nullptr) return false;
7656
7657 // cast it to what we know it will be at runtime
7658 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7659 assert(tinst != nullptr, "ECB obj is null");
7660 assert(tinst->is_loaded(), "ECB obj is not loaded");
7661 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7662 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7663
7664 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7665 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7666 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7667 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7668 aescrypt_object = _gvn.transform(aescrypt_object);
7669
7670 // we need to get the start of the aescrypt_object's expanded key array
7671 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7672 if (k_start == nullptr) return false;
7673
7674 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7675 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7676 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7677 stubAddr, stubName, TypePtr::BOTTOM,
7678 src_start, dest_start, k_start, len);
7679
7680 // return cipher length (int)
7681 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7682 set_result(retvalue);
7683 return true;
7684 }
7685
7686 //------------------------------inline_counterMode_AESCrypt-----------------------
7687 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7688 assert(UseAES, "need AES instruction support");
7689 if (!UseAESCTRIntrinsics) return false;
7690
7691 address stubAddr = nullptr;
7692 const char *stubName = nullptr;
7693 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7694 stubAddr = StubRoutines::counterMode_AESCrypt();
7695 stubName = "counterMode_AESCrypt";
7696 }
7697 if (stubAddr == nullptr) return false;
7698
7699 Node* counterMode_object = argument(0);
7700 Node* src = argument(1);
7701 Node* src_offset = argument(2);
7702 Node* len = argument(3);
7703 Node* dest = argument(4);
7704 Node* dest_offset = argument(5);
7705
7706 // (1) src and dest are arrays.
7707 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7708 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7709 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7710 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7711
7712 // checks are the responsibility of the caller
7713 Node* src_start = src;
7714 Node* dest_start = dest;
7715 if (src_offset != nullptr || dest_offset != nullptr) {
7716 assert(src_offset != nullptr && dest_offset != nullptr, "");
7717 src_start = array_element_address(src, src_offset, T_BYTE);
7718 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7719 }
7720
7721 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7722 // (because of the predicated logic executed earlier).
7723 // so we cast it here safely.
7724 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7725 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7726 if (embeddedCipherObj == nullptr) return false;
7727 // cast it to what we know it will be at runtime
7728 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7729 assert(tinst != nullptr, "CTR obj is null");
7730 assert(tinst->is_loaded(), "CTR obj is not loaded");
7731 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7732 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7733 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7734 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7735 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7736 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7737 aescrypt_object = _gvn.transform(aescrypt_object);
7738 // we need to get the start of the aescrypt_object's expanded key array
7739 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7740 if (k_start == nullptr) return false;
7741 // similarly, get the start address of the r vector
7742 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7743 if (obj_counter == nullptr) return false;
7744 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7745
7746 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7747 if (saved_encCounter == nullptr) return false;
7748 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7749 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7750
7751 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7752 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7753 OptoRuntime::counterMode_aescrypt_Type(),
7754 stubAddr, stubName, TypePtr::BOTTOM,
7755 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7756
7757 // return cipher length (int)
7758 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7759 set_result(retvalue);
7760 return true;
7761 }
7762
7763 //------------------------------get_key_start_from_aescrypt_object-----------------------
7764 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7765 #if defined(PPC64) || defined(S390) || defined(RISCV64)
7766 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7767 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7768 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7769 // The ppc64 and riscv64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7770 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7771 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7772 if (objSessionK == nullptr) {
7773 return (Node *) nullptr;
7774 }
7775 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7776 #else
7777 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7778 #endif // PPC64
7779 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7780 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7781
7782 // now have the array, need to get the start address of the K array
7783 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7784 return k_start;
7785 }
7786
7787 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7788 // Return node representing slow path of predicate check.
7789 // the pseudo code we want to emulate with this predicate is:
7790 // for encryption:
7791 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7792 // for decryption:
7793 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7794 // note cipher==plain is more conservative than the original java code but that's OK
7795 //
7796 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7797 // The receiver was checked for null already.
7798 Node* objCBC = argument(0);
7799
7800 Node* src = argument(1);
7801 Node* dest = argument(4);
7802
7803 // Load embeddedCipher field of CipherBlockChaining object.
7804 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7805
7806 // get AESCrypt klass for instanceOf check
7807 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7808 // will have same classloader as CipherBlockChaining object
7809 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7810 assert(tinst != nullptr, "CBCobj is null");
7811 assert(tinst->is_loaded(), "CBCobj is not loaded");
7812
7813 // we want to do an instanceof comparison against the AESCrypt class
7814 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7815 if (!klass_AESCrypt->is_loaded()) {
7816 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7817 Node* ctrl = control();
7818 set_control(top()); // no regular fast path
7819 return ctrl;
7820 }
7821
7822 src = must_be_not_null(src, true);
7823 dest = must_be_not_null(dest, true);
7824
7825 // Resolve oops to stable for CmpP below.
7826 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7827
7828 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7829 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7830 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7831
7832 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7833
7834 // for encryption, we are done
7835 if (!decrypting)
7836 return instof_false; // even if it is null
7837
7838 // for decryption, we need to add a further check to avoid
7839 // taking the intrinsic path when cipher and plain are the same
7840 // see the original java code for why.
7841 RegionNode* region = new RegionNode(3);
7842 region->init_req(1, instof_false);
7843
7844 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7845 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7846 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7847 region->init_req(2, src_dest_conjoint);
7848
7849 record_for_igvn(region);
7850 return _gvn.transform(region);
7851 }
7852
7853 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7854 // Return node representing slow path of predicate check.
7855 // the pseudo code we want to emulate with this predicate is:
7856 // for encryption:
7857 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7858 // for decryption:
7859 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7860 // note cipher==plain is more conservative than the original java code but that's OK
7861 //
7862 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7863 // The receiver was checked for null already.
7864 Node* objECB = argument(0);
7865
7866 // Load embeddedCipher field of ElectronicCodeBook object.
7867 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7868
7869 // get AESCrypt klass for instanceOf check
7870 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7871 // will have same classloader as ElectronicCodeBook object
7872 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7873 assert(tinst != nullptr, "ECBobj is null");
7874 assert(tinst->is_loaded(), "ECBobj is not loaded");
7875
7876 // we want to do an instanceof comparison against the AESCrypt class
7877 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7878 if (!klass_AESCrypt->is_loaded()) {
7879 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7880 Node* ctrl = control();
7881 set_control(top()); // no regular fast path
7882 return ctrl;
7883 }
7884 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7885
7886 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7887 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7888 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7889
7890 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7891
7892 // for encryption, we are done
7893 if (!decrypting)
7894 return instof_false; // even if it is null
7895
7896 // for decryption, we need to add a further check to avoid
7897 // taking the intrinsic path when cipher and plain are the same
7898 // see the original java code for why.
7899 RegionNode* region = new RegionNode(3);
7900 region->init_req(1, instof_false);
7901 Node* src = argument(1);
7902 Node* dest = argument(4);
7903 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7904 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7905 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7906 region->init_req(2, src_dest_conjoint);
7907
7908 record_for_igvn(region);
7909 return _gvn.transform(region);
7910 }
7911
7912 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7913 // Return node representing slow path of predicate check.
7914 // the pseudo code we want to emulate with this predicate is:
7915 // for encryption:
7916 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7917 // for decryption:
7918 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7919 // note cipher==plain is more conservative than the original java code but that's OK
7920 //
7921
7922 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7923 // The receiver was checked for null already.
7924 Node* objCTR = argument(0);
7925
7926 // Load embeddedCipher field of CipherBlockChaining object.
7927 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7928
7929 // get AESCrypt klass for instanceOf check
7930 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7931 // will have same classloader as CipherBlockChaining object
7932 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7933 assert(tinst != nullptr, "CTRobj is null");
7934 assert(tinst->is_loaded(), "CTRobj is not loaded");
7935
7936 // we want to do an instanceof comparison against the AESCrypt class
7937 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7938 if (!klass_AESCrypt->is_loaded()) {
7939 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7940 Node* ctrl = control();
7941 set_control(top()); // no regular fast path
7942 return ctrl;
7943 }
7944
7945 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7946 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7947 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7948 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7949 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7950
7951 return instof_false; // even if it is null
7952 }
7953
7954 //------------------------------inline_ghash_processBlocks
7955 bool LibraryCallKit::inline_ghash_processBlocks() {
7956 address stubAddr;
7957 const char *stubName;
7958 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7959
7960 stubAddr = StubRoutines::ghash_processBlocks();
7961 stubName = "ghash_processBlocks";
7962
7963 Node* data = argument(0);
7964 Node* offset = argument(1);
7965 Node* len = argument(2);
7966 Node* state = argument(3);
7967 Node* subkeyH = argument(4);
7968
7969 state = must_be_not_null(state, true);
7970 subkeyH = must_be_not_null(subkeyH, true);
7971 data = must_be_not_null(data, true);
7972
7973 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7974 assert(state_start, "state is null");
7975 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7976 assert(subkeyH_start, "subkeyH is null");
7977 Node* data_start = array_element_address(data, offset, T_BYTE);
7978 assert(data_start, "data is null");
7979
7980 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7981 OptoRuntime::ghash_processBlocks_Type(),
7982 stubAddr, stubName, TypePtr::BOTTOM,
7983 state_start, subkeyH_start, data_start, len);
7984 return true;
7985 }
7986
7987 //------------------------------inline_chacha20Block
7988 bool LibraryCallKit::inline_chacha20Block() {
7989 address stubAddr;
7990 const char *stubName;
7991 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7992
7993 stubAddr = StubRoutines::chacha20Block();
7994 stubName = "chacha20Block";
7995
7996 Node* state = argument(0);
7997 Node* result = argument(1);
7998
7999 state = must_be_not_null(state, true);
8000 result = must_be_not_null(result, true);
8001
8002 Node* state_start = array_element_address(state, intcon(0), T_INT);
8003 assert(state_start, "state is null");
8004 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
8005 assert(result_start, "result is null");
8006
8007 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
8008 OptoRuntime::chacha20Block_Type(),
8009 stubAddr, stubName, TypePtr::BOTTOM,
8010 state_start, result_start);
8011 // return key stream length (int)
8012 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
8013 set_result(retvalue);
8014 return true;
8015 }
8016
8017 bool LibraryCallKit::inline_base64_encodeBlock() {
8018 address stubAddr;
8019 const char *stubName;
8020 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8021 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8022 stubAddr = StubRoutines::base64_encodeBlock();
8023 stubName = "encodeBlock";
8024
8025 if (!stubAddr) return false;
8026 Node* base64obj = argument(0);
8027 Node* src = argument(1);
8028 Node* offset = argument(2);
8029 Node* len = argument(3);
8030 Node* dest = argument(4);
8031 Node* dp = argument(5);
8032 Node* isURL = argument(6);
8033
8034 src = must_be_not_null(src, true);
8035 dest = must_be_not_null(dest, true);
8036
8037 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8038 assert(src_start, "source array is null");
8039 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8040 assert(dest_start, "destination array is null");
8041
8042 Node* base64 = make_runtime_call(RC_LEAF,
8043 OptoRuntime::base64_encodeBlock_Type(),
8044 stubAddr, stubName, TypePtr::BOTTOM,
8045 src_start, offset, len, dest_start, dp, isURL);
8046 return true;
8047 }
8048
8049 bool LibraryCallKit::inline_base64_decodeBlock() {
8050 address stubAddr;
8051 const char *stubName;
8052 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8053 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8054 stubAddr = StubRoutines::base64_decodeBlock();
8055 stubName = "decodeBlock";
8056
8057 if (!stubAddr) return false;
8058 Node* base64obj = argument(0);
8059 Node* src = argument(1);
8060 Node* src_offset = argument(2);
8061 Node* len = argument(3);
8062 Node* dest = argument(4);
8063 Node* dest_offset = argument(5);
8064 Node* isURL = argument(6);
8065 Node* isMIME = argument(7);
8066
8067 src = must_be_not_null(src, true);
8068 dest = must_be_not_null(dest, true);
8069
8070 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8071 assert(src_start, "source array is null");
8072 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8073 assert(dest_start, "destination array is null");
8074
8075 Node* call = make_runtime_call(RC_LEAF,
8076 OptoRuntime::base64_decodeBlock_Type(),
8077 stubAddr, stubName, TypePtr::BOTTOM,
8078 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8079 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8080 set_result(result);
8081 return true;
8082 }
8083
8084 bool LibraryCallKit::inline_poly1305_processBlocks() {
8085 address stubAddr;
8086 const char *stubName;
8087 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8088 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8089 stubAddr = StubRoutines::poly1305_processBlocks();
8090 stubName = "poly1305_processBlocks";
8091
8092 if (!stubAddr) return false;
8093 null_check_receiver(); // null-check receiver
8094 if (stopped()) return true;
8095
8096 Node* input = argument(1);
8097 Node* input_offset = argument(2);
8098 Node* len = argument(3);
8099 Node* alimbs = argument(4);
8100 Node* rlimbs = argument(5);
8101
8102 input = must_be_not_null(input, true);
8103 alimbs = must_be_not_null(alimbs, true);
8104 rlimbs = must_be_not_null(rlimbs, true);
8105
8106 Node* input_start = array_element_address(input, input_offset, T_BYTE);
8107 assert(input_start, "input array is null");
8108 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
8109 assert(acc_start, "acc array is null");
8110 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
8111 assert(r_start, "r array is null");
8112
8113 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8114 OptoRuntime::poly1305_processBlocks_Type(),
8115 stubAddr, stubName, TypePtr::BOTTOM,
8116 input_start, len, acc_start, r_start);
8117 return true;
8118 }
8119
8120 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
8121 address stubAddr;
8122 const char *stubName;
8123 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8124 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
8125 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
8126 stubName = "intpoly_montgomeryMult_P256";
8127
8128 if (!stubAddr) return false;
8129 null_check_receiver(); // null-check receiver
8130 if (stopped()) return true;
8131
8132 Node* a = argument(1);
8133 Node* b = argument(2);
8134 Node* r = argument(3);
8135
8136 a = must_be_not_null(a, true);
8137 b = must_be_not_null(b, true);
8138 r = must_be_not_null(r, true);
8139
8140 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8141 assert(a_start, "a array is NULL");
8142 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8143 assert(b_start, "b array is NULL");
8144 Node* r_start = array_element_address(r, intcon(0), T_LONG);
8145 assert(r_start, "r array is NULL");
8146
8147 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8148 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
8149 stubAddr, stubName, TypePtr::BOTTOM,
8150 a_start, b_start, r_start);
8151 return true;
8152 }
8153
8154 bool LibraryCallKit::inline_intpoly_assign() {
8155 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8156 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
8157 const char *stubName = "intpoly_assign";
8158 address stubAddr = StubRoutines::intpoly_assign();
8159 if (!stubAddr) return false;
8160
8161 Node* set = argument(0);
8162 Node* a = argument(1);
8163 Node* b = argument(2);
8164 Node* arr_length = load_array_length(a);
8165
8166 a = must_be_not_null(a, true);
8167 b = must_be_not_null(b, true);
8168
8169 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8170 assert(a_start, "a array is NULL");
8171 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8172 assert(b_start, "b array is NULL");
8173
8174 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8175 OptoRuntime::intpoly_assign_Type(),
8176 stubAddr, stubName, TypePtr::BOTTOM,
8177 set, a_start, b_start, arr_length);
8178 return true;
8179 }
8180
8181 //------------------------------inline_digestBase_implCompress-----------------------
8182 //
8183 // Calculate MD5 for single-block byte[] array.
8184 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
8185 //
8186 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
8187 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
8188 //
8189 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
8190 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
8191 //
8192 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
8193 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
8194 //
8195 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
8196 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
8197 //
8198 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
8199 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
8200
8201 Node* digestBase_obj = argument(0);
8202 Node* src = argument(1); // type oop
8203 Node* ofs = argument(2); // type int
8204
8205 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8206 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8207 // failed array check
8208 return false;
8209 }
8210 // Figure out the size and type of the elements we will be copying.
8211 BasicType src_elem = src_type->elem()->array_element_basic_type();
8212 if (src_elem != T_BYTE) {
8213 return false;
8214 }
8215 // 'src_start' points to src array + offset
8216 src = must_be_not_null(src, true);
8217 Node* src_start = array_element_address(src, ofs, src_elem);
8218 Node* state = nullptr;
8219 Node* block_size = nullptr;
8220 address stubAddr;
8221 const char *stubName;
8222
8223 switch(id) {
8224 case vmIntrinsics::_md5_implCompress:
8225 assert(UseMD5Intrinsics, "need MD5 instruction support");
8226 state = get_state_from_digest_object(digestBase_obj, T_INT);
8227 stubAddr = StubRoutines::md5_implCompress();
8228 stubName = "md5_implCompress";
8229 break;
8230 case vmIntrinsics::_sha_implCompress:
8231 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
8232 state = get_state_from_digest_object(digestBase_obj, T_INT);
8233 stubAddr = StubRoutines::sha1_implCompress();
8234 stubName = "sha1_implCompress";
8235 break;
8236 case vmIntrinsics::_sha2_implCompress:
8237 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
8238 state = get_state_from_digest_object(digestBase_obj, T_INT);
8239 stubAddr = StubRoutines::sha256_implCompress();
8240 stubName = "sha256_implCompress";
8241 break;
8242 case vmIntrinsics::_sha5_implCompress:
8243 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
8244 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8245 stubAddr = StubRoutines::sha512_implCompress();
8246 stubName = "sha512_implCompress";
8247 break;
8248 case vmIntrinsics::_sha3_implCompress:
8249 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
8250 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8251 stubAddr = StubRoutines::sha3_implCompress();
8252 stubName = "sha3_implCompress";
8253 block_size = get_block_size_from_digest_object(digestBase_obj);
8254 if (block_size == nullptr) return false;
8255 break;
8256 default:
8257 fatal_unexpected_iid(id);
8258 return false;
8259 }
8260 if (state == nullptr) return false;
8261
8262 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
8263 if (stubAddr == nullptr) return false;
8264
8265 // Call the stub.
8266 Node* call;
8267 if (block_size == nullptr) {
8268 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
8269 stubAddr, stubName, TypePtr::BOTTOM,
8270 src_start, state);
8271 } else {
8272 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
8273 stubAddr, stubName, TypePtr::BOTTOM,
8274 src_start, state, block_size);
8275 }
8276
8277 return true;
8278 }
8279
8280 //------------------------------inline_digestBase_implCompressMB-----------------------
8281 //
8282 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
8283 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
8284 //
8285 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
8286 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8287 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8288 assert((uint)predicate < 5, "sanity");
8289 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
8290
8291 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
8292 Node* src = argument(1); // byte[] array
8293 Node* ofs = argument(2); // type int
8294 Node* limit = argument(3); // type int
8295
8296 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8297 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8298 // failed array check
8299 return false;
8300 }
8301 // Figure out the size and type of the elements we will be copying.
8302 BasicType src_elem = src_type->elem()->array_element_basic_type();
8303 if (src_elem != T_BYTE) {
8304 return false;
8305 }
8306 // 'src_start' points to src array + offset
8307 src = must_be_not_null(src, false);
8308 Node* src_start = array_element_address(src, ofs, src_elem);
8309
8310 const char* klass_digestBase_name = nullptr;
8311 const char* stub_name = nullptr;
8312 address stub_addr = nullptr;
8313 BasicType elem_type = T_INT;
8314
8315 switch (predicate) {
8316 case 0:
8317 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
8318 klass_digestBase_name = "sun/security/provider/MD5";
8319 stub_name = "md5_implCompressMB";
8320 stub_addr = StubRoutines::md5_implCompressMB();
8321 }
8322 break;
8323 case 1:
8324 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
8325 klass_digestBase_name = "sun/security/provider/SHA";
8326 stub_name = "sha1_implCompressMB";
8327 stub_addr = StubRoutines::sha1_implCompressMB();
8328 }
8329 break;
8330 case 2:
8331 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
8332 klass_digestBase_name = "sun/security/provider/SHA2";
8333 stub_name = "sha256_implCompressMB";
8334 stub_addr = StubRoutines::sha256_implCompressMB();
8335 }
8336 break;
8337 case 3:
8338 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
8339 klass_digestBase_name = "sun/security/provider/SHA5";
8340 stub_name = "sha512_implCompressMB";
8341 stub_addr = StubRoutines::sha512_implCompressMB();
8342 elem_type = T_LONG;
8343 }
8344 break;
8345 case 4:
8346 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
8347 klass_digestBase_name = "sun/security/provider/SHA3";
8348 stub_name = "sha3_implCompressMB";
8349 stub_addr = StubRoutines::sha3_implCompressMB();
8350 elem_type = T_LONG;
8351 }
8352 break;
8353 default:
8354 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
8355 }
8356 if (klass_digestBase_name != nullptr) {
8357 assert(stub_addr != nullptr, "Stub is generated");
8358 if (stub_addr == nullptr) return false;
8359
8360 // get DigestBase klass to lookup for SHA klass
8361 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
8362 assert(tinst != nullptr, "digestBase_obj is not instance???");
8363 assert(tinst->is_loaded(), "DigestBase is not loaded");
8364
8365 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
8366 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
8367 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
8368 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
8369 }
8370 return false;
8371 }
8372
8373 //------------------------------inline_digestBase_implCompressMB-----------------------
8374 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
8375 BasicType elem_type, address stubAddr, const char *stubName,
8376 Node* src_start, Node* ofs, Node* limit) {
8377 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
8378 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8379 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
8380 digest_obj = _gvn.transform(digest_obj);
8381
8382 Node* state = get_state_from_digest_object(digest_obj, elem_type);
8383 if (state == nullptr) return false;
8384
8385 Node* block_size = nullptr;
8386 if (strcmp("sha3_implCompressMB", stubName) == 0) {
8387 block_size = get_block_size_from_digest_object(digest_obj);
8388 if (block_size == nullptr) return false;
8389 }
8390
8391 // Call the stub.
8392 Node* call;
8393 if (block_size == nullptr) {
8394 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8395 OptoRuntime::digestBase_implCompressMB_Type(false),
8396 stubAddr, stubName, TypePtr::BOTTOM,
8397 src_start, state, ofs, limit);
8398 } else {
8399 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8400 OptoRuntime::digestBase_implCompressMB_Type(true),
8401 stubAddr, stubName, TypePtr::BOTTOM,
8402 src_start, state, block_size, ofs, limit);
8403 }
8404
8405 // return ofs (int)
8406 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8407 set_result(result);
8408
8409 return true;
8410 }
8411
8412 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8413 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8414 assert(UseAES, "need AES instruction support");
8415 address stubAddr = nullptr;
8416 const char *stubName = nullptr;
8417 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8418 stubName = "galoisCounterMode_AESCrypt";
8419
8420 if (stubAddr == nullptr) return false;
8421
8422 Node* in = argument(0);
8423 Node* inOfs = argument(1);
8424 Node* len = argument(2);
8425 Node* ct = argument(3);
8426 Node* ctOfs = argument(4);
8427 Node* out = argument(5);
8428 Node* outOfs = argument(6);
8429 Node* gctr_object = argument(7);
8430 Node* ghash_object = argument(8);
8431
8432 // (1) in, ct and out are arrays.
8433 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8434 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8435 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8436 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8437 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8438 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8439
8440 // checks are the responsibility of the caller
8441 Node* in_start = in;
8442 Node* ct_start = ct;
8443 Node* out_start = out;
8444 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8445 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8446 in_start = array_element_address(in, inOfs, T_BYTE);
8447 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8448 out_start = array_element_address(out, outOfs, T_BYTE);
8449 }
8450
8451 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8452 // (because of the predicated logic executed earlier).
8453 // so we cast it here safely.
8454 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8455 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8456 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8457 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8458 Node* state = load_field_from_object(ghash_object, "state", "[J");
8459
8460 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8461 return false;
8462 }
8463 // cast it to what we know it will be at runtime
8464 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8465 assert(tinst != nullptr, "GCTR obj is null");
8466 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8467 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8468 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8469 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8470 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8471 const TypeOopPtr* xtype = aklass->as_instance_type();
8472 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8473 aescrypt_object = _gvn.transform(aescrypt_object);
8474 // we need to get the start of the aescrypt_object's expanded key array
8475 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8476 if (k_start == nullptr) return false;
8477 // similarly, get the start address of the r vector
8478 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8479 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8480 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8481
8482
8483 // Call the stub, passing params
8484 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8485 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8486 stubAddr, stubName, TypePtr::BOTTOM,
8487 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8488
8489 // return cipher length (int)
8490 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8491 set_result(retvalue);
8492
8493 return true;
8494 }
8495
8496 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8497 // Return node representing slow path of predicate check.
8498 // the pseudo code we want to emulate with this predicate is:
8499 // for encryption:
8500 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8501 // for decryption:
8502 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8503 // note cipher==plain is more conservative than the original java code but that's OK
8504 //
8505
8506 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8507 // The receiver was checked for null already.
8508 Node* objGCTR = argument(7);
8509 // Load embeddedCipher field of GCTR object.
8510 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8511 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8512
8513 // get AESCrypt klass for instanceOf check
8514 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8515 // will have same classloader as CipherBlockChaining object
8516 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8517 assert(tinst != nullptr, "GCTR obj is null");
8518 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8519
8520 // we want to do an instanceof comparison against the AESCrypt class
8521 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8522 if (!klass_AESCrypt->is_loaded()) {
8523 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8524 Node* ctrl = control();
8525 set_control(top()); // no regular fast path
8526 return ctrl;
8527 }
8528
8529 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8530 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8531 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8532 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8533 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8534
8535 return instof_false; // even if it is null
8536 }
8537
8538 //------------------------------get_state_from_digest_object-----------------------
8539 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8540 const char* state_type;
8541 switch (elem_type) {
8542 case T_BYTE: state_type = "[B"; break;
8543 case T_INT: state_type = "[I"; break;
8544 case T_LONG: state_type = "[J"; break;
8545 default: ShouldNotReachHere();
8546 }
8547 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8548 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8549 if (digest_state == nullptr) return (Node *) nullptr;
8550
8551 // now have the array, need to get the start address of the state array
8552 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8553 return state;
8554 }
8555
8556 //------------------------------get_block_size_from_sha3_object----------------------------------
8557 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8558 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8559 assert (block_size != nullptr, "sanity");
8560 return block_size;
8561 }
8562
8563 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8564 // Return node representing slow path of predicate check.
8565 // the pseudo code we want to emulate with this predicate is:
8566 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8567 //
8568 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8569 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8570 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8571 assert((uint)predicate < 5, "sanity");
8572
8573 // The receiver was checked for null already.
8574 Node* digestBaseObj = argument(0);
8575
8576 // get DigestBase klass for instanceOf check
8577 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8578 assert(tinst != nullptr, "digestBaseObj is null");
8579 assert(tinst->is_loaded(), "DigestBase is not loaded");
8580
8581 const char* klass_name = nullptr;
8582 switch (predicate) {
8583 case 0:
8584 if (UseMD5Intrinsics) {
8585 // we want to do an instanceof comparison against the MD5 class
8586 klass_name = "sun/security/provider/MD5";
8587 }
8588 break;
8589 case 1:
8590 if (UseSHA1Intrinsics) {
8591 // we want to do an instanceof comparison against the SHA class
8592 klass_name = "sun/security/provider/SHA";
8593 }
8594 break;
8595 case 2:
8596 if (UseSHA256Intrinsics) {
8597 // we want to do an instanceof comparison against the SHA2 class
8598 klass_name = "sun/security/provider/SHA2";
8599 }
8600 break;
8601 case 3:
8602 if (UseSHA512Intrinsics) {
8603 // we want to do an instanceof comparison against the SHA5 class
8604 klass_name = "sun/security/provider/SHA5";
8605 }
8606 break;
8607 case 4:
8608 if (UseSHA3Intrinsics) {
8609 // we want to do an instanceof comparison against the SHA3 class
8610 klass_name = "sun/security/provider/SHA3";
8611 }
8612 break;
8613 default:
8614 fatal("unknown SHA intrinsic predicate: %d", predicate);
8615 }
8616
8617 ciKlass* klass = nullptr;
8618 if (klass_name != nullptr) {
8619 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8620 }
8621 if ((klass == nullptr) || !klass->is_loaded()) {
8622 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8623 Node* ctrl = control();
8624 set_control(top()); // no intrinsic path
8625 return ctrl;
8626 }
8627 ciInstanceKlass* instklass = klass->as_instance_klass();
8628
8629 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8630 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8631 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8632 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8633
8634 return instof_false; // even if it is null
8635 }
8636
8637 //-------------inline_fma-----------------------------------
8638 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8639 Node *a = nullptr;
8640 Node *b = nullptr;
8641 Node *c = nullptr;
8642 Node* result = nullptr;
8643 switch (id) {
8644 case vmIntrinsics::_fmaD:
8645 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8646 // no receiver since it is static method
8647 a = round_double_node(argument(0));
8648 b = round_double_node(argument(2));
8649 c = round_double_node(argument(4));
8650 result = _gvn.transform(new FmaDNode(control(), a, b, c));
8651 break;
8652 case vmIntrinsics::_fmaF:
8653 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8654 a = argument(0);
8655 b = argument(1);
8656 c = argument(2);
8657 result = _gvn.transform(new FmaFNode(control(), a, b, c));
8658 break;
8659 default:
8660 fatal_unexpected_iid(id); break;
8661 }
8662 set_result(result);
8663 return true;
8664 }
8665
8666 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8667 // argument(0) is receiver
8668 Node* codePoint = argument(1);
8669 Node* n = nullptr;
8670
8671 switch (id) {
8672 case vmIntrinsics::_isDigit :
8673 n = new DigitNode(control(), codePoint);
8674 break;
8675 case vmIntrinsics::_isLowerCase :
8676 n = new LowerCaseNode(control(), codePoint);
8677 break;
8678 case vmIntrinsics::_isUpperCase :
8679 n = new UpperCaseNode(control(), codePoint);
8680 break;
8681 case vmIntrinsics::_isWhitespace :
8682 n = new WhitespaceNode(control(), codePoint);
8683 break;
8684 default:
8685 fatal_unexpected_iid(id);
8686 }
8687
8688 set_result(_gvn.transform(n));
8689 return true;
8690 }
8691
8692 //------------------------------inline_fp_min_max------------------------------
8693 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
8694 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
8695
8696 // The intrinsic should be used only when the API branches aren't predictable,
8697 // the last one performing the most important comparison. The following heuristic
8698 // uses the branch statistics to eventually bail out if necessary.
8699
8700 ciMethodData *md = callee()->method_data();
8701
8702 if ( md != nullptr && md->is_mature() && md->invocation_count() > 0 ) {
8703 ciCallProfile cp = caller()->call_profile_at_bci(bci());
8704
8705 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
8706 // Bail out if the call-site didn't contribute enough to the statistics.
8707 return false;
8708 }
8709
8710 uint taken = 0, not_taken = 0;
8711
8712 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
8713 if (p->is_BranchData()) {
8714 taken = ((ciBranchData*)p)->taken();
8715 not_taken = ((ciBranchData*)p)->not_taken();
8716 }
8717 }
8718
8719 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
8720 balance = balance < 0 ? -balance : balance;
8721 if ( balance > 0.2 ) {
8722 // Bail out if the most important branch is predictable enough.
8723 return false;
8724 }
8725 }
8726 */
8727
8728 Node *a = nullptr;
8729 Node *b = nullptr;
8730 Node *n = nullptr;
8731 switch (id) {
8732 case vmIntrinsics::_maxF:
8733 case vmIntrinsics::_minF:
8734 case vmIntrinsics::_maxF_strict:
8735 case vmIntrinsics::_minF_strict:
8736 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
8737 a = argument(0);
8738 b = argument(1);
8739 break;
8740 case vmIntrinsics::_maxD:
8741 case vmIntrinsics::_minD:
8742 case vmIntrinsics::_maxD_strict:
8743 case vmIntrinsics::_minD_strict:
8744 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
8745 a = round_double_node(argument(0));
8746 b = round_double_node(argument(2));
8747 break;
8748 default:
8749 fatal_unexpected_iid(id);
8750 break;
8751 }
8752 switch (id) {
8753 case vmIntrinsics::_maxF:
8754 case vmIntrinsics::_maxF_strict:
8755 n = new MaxFNode(a, b);
8756 break;
8757 case vmIntrinsics::_minF:
8758 case vmIntrinsics::_minF_strict:
8759 n = new MinFNode(a, b);
8760 break;
8761 case vmIntrinsics::_maxD:
8762 case vmIntrinsics::_maxD_strict:
8763 n = new MaxDNode(a, b);
8764 break;
8765 case vmIntrinsics::_minD:
8766 case vmIntrinsics::_minD_strict:
8767 n = new MinDNode(a, b);
8768 break;
8769 default:
8770 fatal_unexpected_iid(id);
8771 break;
8772 }
8773 set_result(_gvn.transform(n));
8774 return true;
8775 }
8776
8777 bool LibraryCallKit::inline_profileBoolean() {
8778 Node* counts = argument(1);
8779 const TypeAryPtr* ary = nullptr;
8780 ciArray* aobj = nullptr;
8781 if (counts->is_Con()
8782 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8783 && (aobj = ary->const_oop()->as_array()) != nullptr
8784 && (aobj->length() == 2)) {
8785 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8786 jint false_cnt = aobj->element_value(0).as_int();
8787 jint true_cnt = aobj->element_value(1).as_int();
8788
8789 if (C->log() != nullptr) {
8790 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8791 false_cnt, true_cnt);
8792 }
8793
8794 if (false_cnt + true_cnt == 0) {
8795 // According to profile, never executed.
8796 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8797 Deoptimization::Action_reinterpret);
8798 return true;
8799 }
8800
8801 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8802 // is a number of each value occurrences.
8803 Node* result = argument(0);
8804 if (false_cnt == 0 || true_cnt == 0) {
8805 // According to profile, one value has been never seen.
8806 int expected_val = (false_cnt == 0) ? 1 : 0;
8807
8808 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8809 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8810
8811 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8812 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8813 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8814
8815 { // Slow path: uncommon trap for never seen value and then reexecute
8816 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8817 // the value has been seen at least once.
8818 PreserveJVMState pjvms(this);
8819 PreserveReexecuteState preexecs(this);
8820 jvms()->set_should_reexecute(true);
8821
8822 set_control(slow_path);
8823 set_i_o(i_o());
8824
8825 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8826 Deoptimization::Action_reinterpret);
8827 }
8828 // The guard for never seen value enables sharpening of the result and
8829 // returning a constant. It allows to eliminate branches on the same value
8830 // later on.
8831 set_control(fast_path);
8832 result = intcon(expected_val);
8833 }
8834 // Stop profiling.
8835 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8836 // By replacing method body with profile data (represented as ProfileBooleanNode
8837 // on IR level) we effectively disable profiling.
8838 // It enables full speed execution once optimized code is generated.
8839 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8840 C->record_for_igvn(profile);
8841 set_result(profile);
8842 return true;
8843 } else {
8844 // Continue profiling.
8845 // Profile data isn't available at the moment. So, execute method's bytecode version.
8846 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8847 // is compiled and counters aren't available since corresponding MethodHandle
8848 // isn't a compile-time constant.
8849 return false;
8850 }
8851 }
8852
8853 bool LibraryCallKit::inline_isCompileConstant() {
8854 Node* n = argument(0);
8855 set_result(n->is_Con() ? intcon(1) : intcon(0));
8856 return true;
8857 }
8858
8859 //------------------------------- inline_getObjectSize --------------------------------------
8860 //
8861 // Calculate the runtime size of the object/array.
8862 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8863 //
8864 bool LibraryCallKit::inline_getObjectSize() {
8865 Node* obj = argument(3);
8866 Node* klass_node = load_object_klass(obj);
8867
8868 jint layout_con = Klass::_lh_neutral_value;
8869 Node* layout_val = get_layout_helper(klass_node, layout_con);
8870 int layout_is_con = (layout_val == nullptr);
8871
8872 if (layout_is_con) {
8873 // Layout helper is constant, can figure out things at compile time.
8874
8875 if (Klass::layout_helper_is_instance(layout_con)) {
8876 // Instance case: layout_con contains the size itself.
8877 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8878 set_result(size);
8879 } else {
8880 // Array case: size is round(header + element_size*arraylength).
8881 // Since arraylength is different for every array instance, we have to
8882 // compute the whole thing at runtime.
8883
8884 Node* arr_length = load_array_length(obj);
8885
8886 int round_mask = MinObjAlignmentInBytes - 1;
8887 int hsize = Klass::layout_helper_header_size(layout_con);
8888 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8889
8890 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8891 round_mask = 0; // strength-reduce it if it goes away completely
8892 }
8893 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8894 Node* header_size = intcon(hsize + round_mask);
8895
8896 Node* lengthx = ConvI2X(arr_length);
8897 Node* headerx = ConvI2X(header_size);
8898
8899 Node* abody = lengthx;
8900 if (eshift != 0) {
8901 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8902 }
8903 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8904 if (round_mask != 0) {
8905 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8906 }
8907 size = ConvX2L(size);
8908 set_result(size);
8909 }
8910 } else {
8911 // Layout helper is not constant, need to test for array-ness at runtime.
8912
8913 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8914 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8915 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8916 record_for_igvn(result_reg);
8917
8918 Node* array_ctl = generate_array_guard(klass_node, nullptr);
8919 if (array_ctl != nullptr) {
8920 // Array case: size is round(header + element_size*arraylength).
8921 // Since arraylength is different for every array instance, we have to
8922 // compute the whole thing at runtime.
8923
8924 PreserveJVMState pjvms(this);
8925 set_control(array_ctl);
8926 Node* arr_length = load_array_length(obj);
8927
8928 int round_mask = MinObjAlignmentInBytes - 1;
8929 Node* mask = intcon(round_mask);
8930
8931 Node* hss = intcon(Klass::_lh_header_size_shift);
8932 Node* hsm = intcon(Klass::_lh_header_size_mask);
8933 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8934 header_size = _gvn.transform(new AndINode(header_size, hsm));
8935 header_size = _gvn.transform(new AddINode(header_size, mask));
8936
8937 // There is no need to mask or shift this value.
8938 // The semantics of LShiftINode include an implicit mask to 0x1F.
8939 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8940 Node* elem_shift = layout_val;
8941
8942 Node* lengthx = ConvI2X(arr_length);
8943 Node* headerx = ConvI2X(header_size);
8944
8945 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
8946 Node* size = _gvn.transform(new AddXNode(headerx, abody));
8947 if (round_mask != 0) {
8948 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
8949 }
8950 size = ConvX2L(size);
8951
8952 result_reg->init_req(_array_path, control());
8953 result_val->init_req(_array_path, size);
8954 }
8955
8956 if (!stopped()) {
8957 // Instance case: the layout helper gives us instance size almost directly,
8958 // but we need to mask out the _lh_instance_slow_path_bit.
8959 Node* size = ConvI2X(layout_val);
8960 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
8961 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
8962 size = _gvn.transform(new AndXNode(size, mask));
8963 size = ConvX2L(size);
8964
8965 result_reg->init_req(_instance_path, control());
8966 result_val->init_req(_instance_path, size);
8967 }
8968
8969 set_result(result_reg, result_val);
8970 }
8971
8972 return true;
8973 }
8974
8975 //------------------------------- inline_blackhole --------------------------------------
8976 //
8977 // Make sure all arguments to this node are alive.
8978 // This matches methods that were requested to be blackholed through compile commands.
8979 //
8980 bool LibraryCallKit::inline_blackhole() {
8981 assert(callee()->is_static(), "Should have been checked before: only static methods here");
8982 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
8983 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
8984
8985 // Blackhole node pinches only the control, not memory. This allows
8986 // the blackhole to be pinned in the loop that computes blackholed
8987 // values, but have no other side effects, like breaking the optimizations
8988 // across the blackhole.
8989
8990 Node* bh = _gvn.transform(new BlackholeNode(control()));
8991 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
8992
8993 // Bind call arguments as blackhole arguments to keep them alive
8994 uint nargs = callee()->arg_size();
8995 for (uint i = 0; i < nargs; i++) {
8996 bh->add_req(argument(i));
8997 }
8998
8999 return true;
9000 }