1 /*
2 * Copyright (c) 1999, 2025, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "asm/macroAssembler.hpp"
26 #include "ci/ciUtilities.inline.hpp"
27 #include "ci/ciSymbols.hpp"
28 #include "classfile/vmIntrinsics.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "gc/shared/barrierSet.hpp"
32 #include "jfr/support/jfrIntrinsics.hpp"
33 #include "memory/resourceArea.hpp"
34 #include "oops/klass.inline.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "opto/addnode.hpp"
37 #include "opto/arraycopynode.hpp"
38 #include "opto/c2compiler.hpp"
39 #include "opto/castnode.hpp"
40 #include "opto/cfgnode.hpp"
41 #include "opto/convertnode.hpp"
42 #include "opto/countbitsnode.hpp"
43 #include "opto/idealKit.hpp"
44 #include "opto/library_call.hpp"
45 #include "opto/mathexactnode.hpp"
46 #include "opto/mulnode.hpp"
47 #include "opto/narrowptrnode.hpp"
48 #include "opto/opaquenode.hpp"
49 #include "opto/parse.hpp"
50 #include "opto/runtime.hpp"
51 #include "opto/rootnode.hpp"
52 #include "opto/subnode.hpp"
53 #include "opto/vectornode.hpp"
54 #include "prims/jvmtiExport.hpp"
55 #include "prims/jvmtiThreadState.hpp"
56 #include "prims/unsafe.hpp"
57 #include "runtime/jniHandles.inline.hpp"
58 #include "runtime/objectMonitor.hpp"
59 #include "runtime/sharedRuntime.hpp"
60 #include "runtime/stubRoutines.hpp"
61 #include "utilities/macros.hpp"
62 #include "utilities/powerOfTwo.hpp"
63
64 //---------------------------make_vm_intrinsic----------------------------
65 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
66 vmIntrinsicID id = m->intrinsic_id();
67 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
68
69 if (!m->is_loaded()) {
70 // Do not attempt to inline unloaded methods.
71 return nullptr;
72 }
73
74 C2Compiler* compiler = (C2Compiler*)CompilerThread::current()->compiler();
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 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
124 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
125 if (C->log()) {
126 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
127 vmIntrinsics::name_at(intrinsic_id()),
128 (is_virtual() ? " virtual='1'" : ""),
129 C->unique() - nodes);
130 }
131 // Push the result from the inlined method onto the stack.
132 kit.push_result();
133 return kit.transfer_exceptions_into_jvms();
134 }
135
136 // The intrinsic bailed out
137 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
138 if (jvms->has_method()) {
139 // Not a root compile.
140 const char* msg;
141 if (callee->intrinsic_candidate()) {
142 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
143 } else {
144 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
145 : "failed to inline (intrinsic), method not annotated";
146 }
147 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
148 C->inline_printer()->record(callee, jvms, InliningResult::FAILURE, msg);
149 } else {
150 // Root compile
151 ResourceMark rm;
152 stringStream msg_stream;
153 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
154 vmIntrinsics::name_at(intrinsic_id()),
155 is_virtual() ? " (virtual)" : "", bci);
156 const char *msg = msg_stream.freeze();
157 log_debug(jit, inlining)("%s", msg);
158 if (C->print_intrinsics() || C->print_inlining()) {
159 tty->print("%s", msg);
160 }
161 }
162 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
163
164 return nullptr;
165 }
166
167 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
168 LibraryCallKit kit(jvms, this);
169 Compile* C = kit.C;
170 int nodes = C->unique();
171 _last_predicate = predicate;
172 #ifndef PRODUCT
173 assert(is_predicated() && predicate < predicates_count(), "sanity");
174 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
175 char buf[1000];
176 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
177 tty->print_cr("Predicate for intrinsic %s", str);
178 }
179 #endif
180 ciMethod* callee = kit.callee();
181 const int bci = kit.bci();
182
183 Node* slow_ctl = kit.try_to_predicate(predicate);
184 if (!kit.failing()) {
185 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
186 : "(intrinsic, predicate)";
187 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
188 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
189
190 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
191 if (C->log()) {
192 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
193 vmIntrinsics::name_at(intrinsic_id()),
194 (is_virtual() ? " virtual='1'" : ""),
195 C->unique() - nodes);
196 }
197 return slow_ctl; // Could be null if the check folds.
198 }
199
200 // The intrinsic bailed out
201 if (jvms->has_method()) {
202 // Not a root compile.
203 const char* msg = "failed to generate predicate for intrinsic";
204 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
205 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
206 } else {
207 // Root compile
208 ResourceMark rm;
209 stringStream msg_stream;
210 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
211 vmIntrinsics::name_at(intrinsic_id()),
212 is_virtual() ? " (virtual)" : "", bci);
213 const char *msg = msg_stream.freeze();
214 log_debug(jit, inlining)("%s", msg);
215 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
216 }
217 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
218 return nullptr;
219 }
220
221 bool LibraryCallKit::try_to_inline(int predicate) {
222 // Handle symbolic names for otherwise undistinguished boolean switches:
223 const bool is_store = true;
224 const bool is_compress = true;
225 const bool is_static = true;
226 const bool is_volatile = true;
227
228 if (!jvms()->has_method()) {
229 // Root JVMState has a null method.
230 assert(map()->memory()->Opcode() == Op_Parm, "");
231 // Insert the memory aliasing node
232 set_all_memory(reset_memory());
233 }
234 assert(merged_memory(), "");
235
236 switch (intrinsic_id()) {
237 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
238 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
239 case vmIntrinsics::_getClass: return inline_native_getClass();
240
241 case vmIntrinsics::_ceil:
242 case vmIntrinsics::_floor:
243 case vmIntrinsics::_rint:
244 case vmIntrinsics::_dsin:
245 case vmIntrinsics::_dcos:
246 case vmIntrinsics::_dtan:
247 case vmIntrinsics::_dtanh:
248 case vmIntrinsics::_dabs:
249 case vmIntrinsics::_fabs:
250 case vmIntrinsics::_iabs:
251 case vmIntrinsics::_labs:
252 case vmIntrinsics::_datan2:
253 case vmIntrinsics::_dsqrt:
254 case vmIntrinsics::_dsqrt_strict:
255 case vmIntrinsics::_dexp:
256 case vmIntrinsics::_dlog:
257 case vmIntrinsics::_dlog10:
258 case vmIntrinsics::_dpow:
259 case vmIntrinsics::_dcopySign:
260 case vmIntrinsics::_fcopySign:
261 case vmIntrinsics::_dsignum:
262 case vmIntrinsics::_roundF:
263 case vmIntrinsics::_roundD:
264 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
265
266 case vmIntrinsics::_notify:
267 case vmIntrinsics::_notifyAll:
268 return inline_notify(intrinsic_id());
269
270 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
271 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
272 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
273 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
274 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
275 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
276 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
277 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
278 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
279 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
280 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
281 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
282 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
283 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
284
285 case vmIntrinsics::_arraycopy: return inline_arraycopy();
286
287 case vmIntrinsics::_arraySort: return inline_array_sort();
288 case vmIntrinsics::_arrayPartition: return inline_array_partition();
289
290 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
291 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
292 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
293 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
294
295 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
296 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
297 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
298 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
299 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
300 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
301 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
302 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
303
304 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
305
306 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
307
308 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
309 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
310 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
311 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
312
313 case vmIntrinsics::_compressStringC:
314 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
315 case vmIntrinsics::_inflateStringC:
316 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
317
318 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
319 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
320 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
321 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
322 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
323 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
324 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
325 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
326 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
327
328 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
329 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
330 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
331 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
332 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
333 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
334 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
335 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
336 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
337
338 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
339 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
340 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
341 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
342 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
343 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
344 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
345 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
346 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
347
348 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
349 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
350 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
351 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
352 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
353 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
354 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
355 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
356 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
357
358 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
359 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
360 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
361 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
362
363 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
364 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
365 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
366 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
367
368 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
369 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
370 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
371 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
372 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
373 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
374 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
375 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
376 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
377
378 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
379 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
380 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
381 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
382 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
383 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
384 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
385 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
386 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
387
388 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
389 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
390 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
391 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
392 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
393 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
394 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
395 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
396 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
397
398 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
399 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
400 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
401 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
402 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
403 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
404 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
405 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
406 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
407
408 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
409 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
410 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
411 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
412 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
413
414 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
415 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
416 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
417 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
418 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
419 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
420 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
421 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
422 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
423 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
424 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
425 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
426 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
427 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
428 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
429 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
430 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
431 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
432 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
433 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
434
435 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
436 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
437 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
438 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
439 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
440 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
441 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
442 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
443 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
444 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
445 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
446 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
447 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
448 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
449 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
450
451 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
452 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
453 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
454 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
455
456 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
457 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
458 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
459 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
460 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
461
462 case vmIntrinsics::_loadFence:
463 case vmIntrinsics::_storeFence:
464 case vmIntrinsics::_storeStoreFence:
465 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
466
467 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
468
469 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
470 case vmIntrinsics::_currentThread: return inline_native_currentThread();
471 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
472
473 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
474 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
475
476 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
477 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
478
479 #if INCLUDE_JVMTI
480 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
481 "notifyJvmtiStart", true, false);
482 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
483 "notifyJvmtiEnd", false, true);
484 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
485 "notifyJvmtiMount", false, false);
486 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
487 "notifyJvmtiUnmount", false, false);
488 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
489 #endif
490
491 #ifdef JFR_HAVE_INTRINSICS
492 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
493 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
494 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
495 #endif
496 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
497 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
498 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
499 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
500 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
501 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
502 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
503 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
504 case vmIntrinsics::_getLength: return inline_native_getLength();
505 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
506 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
507 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
508 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
509 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
510 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
511 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
512
513 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
514 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
515
516 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
517
518 case vmIntrinsics::_isInstance:
519 case vmIntrinsics::_isHidden:
520 case vmIntrinsics::_getSuperclass:
521 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
522
523 case vmIntrinsics::_floatToRawIntBits:
524 case vmIntrinsics::_floatToIntBits:
525 case vmIntrinsics::_intBitsToFloat:
526 case vmIntrinsics::_doubleToRawLongBits:
527 case vmIntrinsics::_doubleToLongBits:
528 case vmIntrinsics::_longBitsToDouble:
529 case vmIntrinsics::_floatToFloat16:
530 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
531 case vmIntrinsics::_sqrt_float16: return inline_fp16_operations(intrinsic_id(), 1);
532 case vmIntrinsics::_fma_float16: return inline_fp16_operations(intrinsic_id(), 3);
533 case vmIntrinsics::_floatIsFinite:
534 case vmIntrinsics::_floatIsInfinite:
535 case vmIntrinsics::_doubleIsFinite:
536 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
537
538 case vmIntrinsics::_numberOfLeadingZeros_i:
539 case vmIntrinsics::_numberOfLeadingZeros_l:
540 case vmIntrinsics::_numberOfTrailingZeros_i:
541 case vmIntrinsics::_numberOfTrailingZeros_l:
542 case vmIntrinsics::_bitCount_i:
543 case vmIntrinsics::_bitCount_l:
544 case vmIntrinsics::_reverse_i:
545 case vmIntrinsics::_reverse_l:
546 case vmIntrinsics::_reverseBytes_i:
547 case vmIntrinsics::_reverseBytes_l:
548 case vmIntrinsics::_reverseBytes_s:
549 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
550
551 case vmIntrinsics::_compress_i:
552 case vmIntrinsics::_compress_l:
553 case vmIntrinsics::_expand_i:
554 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
555
556 case vmIntrinsics::_compareUnsigned_i:
557 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
558
559 case vmIntrinsics::_divideUnsigned_i:
560 case vmIntrinsics::_divideUnsigned_l:
561 case vmIntrinsics::_remainderUnsigned_i:
562 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
563
564 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
565
566 case vmIntrinsics::_Reference_get: return inline_reference_get();
567 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
568 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
569 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
570 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
571
572 case vmIntrinsics::_Class_cast: return inline_Class_cast();
573
574 case vmIntrinsics::_aescrypt_encryptBlock:
575 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
576
577 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
578 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
579 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
580
581 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
582 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
583 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
584
585 case vmIntrinsics::_counterMode_AESCrypt:
586 return inline_counterMode_AESCrypt(intrinsic_id());
587
588 case vmIntrinsics::_galoisCounterMode_AESCrypt:
589 return inline_galoisCounterMode_AESCrypt();
590
591 case vmIntrinsics::_md5_implCompress:
592 case vmIntrinsics::_sha_implCompress:
593 case vmIntrinsics::_sha2_implCompress:
594 case vmIntrinsics::_sha5_implCompress:
595 case vmIntrinsics::_sha3_implCompress:
596 return inline_digestBase_implCompress(intrinsic_id());
597
598 case vmIntrinsics::_digestBase_implCompressMB:
599 return inline_digestBase_implCompressMB(predicate);
600
601 case vmIntrinsics::_multiplyToLen:
602 return inline_multiplyToLen();
603
604 case vmIntrinsics::_squareToLen:
605 return inline_squareToLen();
606
607 case vmIntrinsics::_mulAdd:
608 return inline_mulAdd();
609
610 case vmIntrinsics::_montgomeryMultiply:
611 return inline_montgomeryMultiply();
612 case vmIntrinsics::_montgomerySquare:
613 return inline_montgomerySquare();
614
615 case vmIntrinsics::_bigIntegerRightShiftWorker:
616 return inline_bigIntegerShift(true);
617 case vmIntrinsics::_bigIntegerLeftShiftWorker:
618 return inline_bigIntegerShift(false);
619
620 case vmIntrinsics::_vectorizedMismatch:
621 return inline_vectorizedMismatch();
622
623 case vmIntrinsics::_ghash_processBlocks:
624 return inline_ghash_processBlocks();
625 case vmIntrinsics::_chacha20Block:
626 return inline_chacha20Block();
627 case vmIntrinsics::_base64_encodeBlock:
628 return inline_base64_encodeBlock();
629 case vmIntrinsics::_base64_decodeBlock:
630 return inline_base64_decodeBlock();
631 case vmIntrinsics::_poly1305_processBlocks:
632 return inline_poly1305_processBlocks();
633 case vmIntrinsics::_intpoly_montgomeryMult_P256:
634 return inline_intpoly_montgomeryMult_P256();
635 case vmIntrinsics::_intpoly_assign:
636 return inline_intpoly_assign();
637 case vmIntrinsics::_encodeISOArray:
638 case vmIntrinsics::_encodeByteISOArray:
639 return inline_encodeISOArray(false);
640 case vmIntrinsics::_encodeAsciiArray:
641 return inline_encodeISOArray(true);
642
643 case vmIntrinsics::_updateCRC32:
644 return inline_updateCRC32();
645 case vmIntrinsics::_updateBytesCRC32:
646 return inline_updateBytesCRC32();
647 case vmIntrinsics::_updateByteBufferCRC32:
648 return inline_updateByteBufferCRC32();
649
650 case vmIntrinsics::_updateBytesCRC32C:
651 return inline_updateBytesCRC32C();
652 case vmIntrinsics::_updateDirectByteBufferCRC32C:
653 return inline_updateDirectByteBufferCRC32C();
654
655 case vmIntrinsics::_updateBytesAdler32:
656 return inline_updateBytesAdler32();
657 case vmIntrinsics::_updateByteBufferAdler32:
658 return inline_updateByteBufferAdler32();
659
660 case vmIntrinsics::_profileBoolean:
661 return inline_profileBoolean();
662 case vmIntrinsics::_isCompileConstant:
663 return inline_isCompileConstant();
664
665 case vmIntrinsics::_countPositives:
666 return inline_countPositives();
667
668 case vmIntrinsics::_fmaD:
669 case vmIntrinsics::_fmaF:
670 return inline_fma(intrinsic_id());
671
672 case vmIntrinsics::_isDigit:
673 case vmIntrinsics::_isLowerCase:
674 case vmIntrinsics::_isUpperCase:
675 case vmIntrinsics::_isWhitespace:
676 return inline_character_compare(intrinsic_id());
677
678 case vmIntrinsics::_min:
679 case vmIntrinsics::_max:
680 case vmIntrinsics::_min_strict:
681 case vmIntrinsics::_max_strict:
682 return inline_min_max(intrinsic_id());
683
684 case vmIntrinsics::_maxF:
685 case vmIntrinsics::_minF:
686 case vmIntrinsics::_maxD:
687 case vmIntrinsics::_minD:
688 case vmIntrinsics::_maxF_strict:
689 case vmIntrinsics::_minF_strict:
690 case vmIntrinsics::_maxD_strict:
691 case vmIntrinsics::_minD_strict:
692 return inline_fp_min_max(intrinsic_id());
693
694 case vmIntrinsics::_VectorUnaryOp:
695 return inline_vector_nary_operation(1);
696 case vmIntrinsics::_VectorBinaryOp:
697 return inline_vector_nary_operation(2);
698 case vmIntrinsics::_VectorTernaryOp:
699 return inline_vector_nary_operation(3);
700 case vmIntrinsics::_VectorFromBitsCoerced:
701 return inline_vector_frombits_coerced();
702 case vmIntrinsics::_VectorMaskOp:
703 return inline_vector_mask_operation();
704 case vmIntrinsics::_VectorLoadOp:
705 return inline_vector_mem_operation(/*is_store=*/false);
706 case vmIntrinsics::_VectorLoadMaskedOp:
707 return inline_vector_mem_masked_operation(/*is_store*/false);
708 case vmIntrinsics::_VectorStoreOp:
709 return inline_vector_mem_operation(/*is_store=*/true);
710 case vmIntrinsics::_VectorStoreMaskedOp:
711 return inline_vector_mem_masked_operation(/*is_store=*/true);
712 case vmIntrinsics::_VectorGatherOp:
713 return inline_vector_gather_scatter(/*is_scatter*/ false);
714 case vmIntrinsics::_VectorScatterOp:
715 return inline_vector_gather_scatter(/*is_scatter*/ true);
716 case vmIntrinsics::_VectorReductionCoerced:
717 return inline_vector_reduction();
718 case vmIntrinsics::_VectorTest:
719 return inline_vector_test();
720 case vmIntrinsics::_VectorBlend:
721 return inline_vector_blend();
722 case vmIntrinsics::_VectorRearrange:
723 return inline_vector_rearrange();
724 case vmIntrinsics::_VectorSelectFrom:
725 return inline_vector_select_from();
726 case vmIntrinsics::_VectorCompare:
727 return inline_vector_compare();
728 case vmIntrinsics::_VectorBroadcastInt:
729 return inline_vector_broadcast_int();
730 case vmIntrinsics::_VectorConvert:
731 return inline_vector_convert();
732 case vmIntrinsics::_VectorInsert:
733 return inline_vector_insert();
734 case vmIntrinsics::_VectorExtract:
735 return inline_vector_extract();
736 case vmIntrinsics::_VectorCompressExpand:
737 return inline_vector_compress_expand();
738 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
739 return inline_vector_select_from_two_vectors();
740 case vmIntrinsics::_IndexVector:
741 return inline_index_vector();
742 case vmIntrinsics::_IndexPartiallyInUpperRange:
743 return inline_index_partially_in_upper_range();
744
745 case vmIntrinsics::_getObjectSize:
746 return inline_getObjectSize();
747
748 case vmIntrinsics::_blackhole:
749 return inline_blackhole();
750
751 default:
752 // If you get here, it may be that someone has added a new intrinsic
753 // to the list in vmIntrinsics.hpp without implementing it here.
754 #ifndef PRODUCT
755 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
756 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
757 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
758 }
759 #endif
760 return false;
761 }
762 }
763
764 Node* LibraryCallKit::try_to_predicate(int predicate) {
765 if (!jvms()->has_method()) {
766 // Root JVMState has a null method.
767 assert(map()->memory()->Opcode() == Op_Parm, "");
768 // Insert the memory aliasing node
769 set_all_memory(reset_memory());
770 }
771 assert(merged_memory(), "");
772
773 switch (intrinsic_id()) {
774 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
775 return inline_cipherBlockChaining_AESCrypt_predicate(false);
776 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
777 return inline_cipherBlockChaining_AESCrypt_predicate(true);
778 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
779 return inline_electronicCodeBook_AESCrypt_predicate(false);
780 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
781 return inline_electronicCodeBook_AESCrypt_predicate(true);
782 case vmIntrinsics::_counterMode_AESCrypt:
783 return inline_counterMode_AESCrypt_predicate();
784 case vmIntrinsics::_digestBase_implCompressMB:
785 return inline_digestBase_implCompressMB_predicate(predicate);
786 case vmIntrinsics::_galoisCounterMode_AESCrypt:
787 return inline_galoisCounterMode_AESCrypt_predicate();
788
789 default:
790 // If you get here, it may be that someone has added a new intrinsic
791 // to the list in vmIntrinsics.hpp without implementing it here.
792 #ifndef PRODUCT
793 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
794 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
795 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
796 }
797 #endif
798 Node* slow_ctl = control();
799 set_control(top()); // No fast path intrinsic
800 return slow_ctl;
801 }
802 }
803
804 //------------------------------set_result-------------------------------
805 // Helper function for finishing intrinsics.
806 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
807 record_for_igvn(region);
808 set_control(_gvn.transform(region));
809 set_result( _gvn.transform(value));
810 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
811 }
812
813 //------------------------------generate_guard---------------------------
814 // Helper function for generating guarded fast-slow graph structures.
815 // The given 'test', if true, guards a slow path. If the test fails
816 // then a fast path can be taken. (We generally hope it fails.)
817 // In all cases, GraphKit::control() is updated to the fast path.
818 // The returned value represents the control for the slow path.
819 // The return value is never 'top'; it is either a valid control
820 // or null if it is obvious that the slow path can never be taken.
821 // Also, if region and the slow control are not null, the slow edge
822 // is appended to the region.
823 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
824 if (stopped()) {
825 // Already short circuited.
826 return nullptr;
827 }
828
829 // Build an if node and its projections.
830 // If test is true we take the slow path, which we assume is uncommon.
831 if (_gvn.type(test) == TypeInt::ZERO) {
832 // The slow branch is never taken. No need to build this guard.
833 return nullptr;
834 }
835
836 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
837
838 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
839 if (if_slow == top()) {
840 // The slow branch is never taken. No need to build this guard.
841 return nullptr;
842 }
843
844 if (region != nullptr)
845 region->add_req(if_slow);
846
847 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
848 set_control(if_fast);
849
850 return if_slow;
851 }
852
853 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
854 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
855 }
856 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
857 return generate_guard(test, region, PROB_FAIR);
858 }
859
860 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
861 Node* *pos_index) {
862 if (stopped())
863 return nullptr; // already stopped
864 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
865 return nullptr; // index is already adequately typed
866 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
867 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
868 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
869 if (is_neg != nullptr && pos_index != nullptr) {
870 // Emulate effect of Parse::adjust_map_after_if.
871 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
872 (*pos_index) = _gvn.transform(ccast);
873 }
874 return is_neg;
875 }
876
877 // Make sure that 'position' is a valid limit index, in [0..length].
878 // There are two equivalent plans for checking this:
879 // A. (offset + copyLength) unsigned<= arrayLength
880 // B. offset <= (arrayLength - copyLength)
881 // We require that all of the values above, except for the sum and
882 // difference, are already known to be non-negative.
883 // Plan A is robust in the face of overflow, if offset and copyLength
884 // are both hugely positive.
885 //
886 // Plan B is less direct and intuitive, but it does not overflow at
887 // all, since the difference of two non-negatives is always
888 // representable. Whenever Java methods must perform the equivalent
889 // check they generally use Plan B instead of Plan A.
890 // For the moment we use Plan A.
891 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
892 Node* subseq_length,
893 Node* array_length,
894 RegionNode* region) {
895 if (stopped())
896 return nullptr; // already stopped
897 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
898 if (zero_offset && subseq_length->eqv_uncast(array_length))
899 return nullptr; // common case of whole-array copy
900 Node* last = subseq_length;
901 if (!zero_offset) // last += offset
902 last = _gvn.transform(new AddINode(last, offset));
903 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
904 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
905 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
906 return is_over;
907 }
908
909 // Emit range checks for the given String.value byte array
910 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
911 if (stopped()) {
912 return; // already stopped
913 }
914 RegionNode* bailout = new RegionNode(1);
915 record_for_igvn(bailout);
916 if (char_count) {
917 // Convert char count to byte count
918 count = _gvn.transform(new LShiftINode(count, intcon(1)));
919 }
920
921 // Offset and count must not be negative
922 generate_negative_guard(offset, bailout);
923 generate_negative_guard(count, bailout);
924 // Offset + count must not exceed length of array
925 generate_limit_guard(offset, count, load_array_length(array), bailout);
926
927 if (bailout->req() > 1) {
928 PreserveJVMState pjvms(this);
929 set_control(_gvn.transform(bailout));
930 uncommon_trap(Deoptimization::Reason_intrinsic,
931 Deoptimization::Action_maybe_recompile);
932 }
933 }
934
935 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
936 bool is_immutable) {
937 ciKlass* thread_klass = env()->Thread_klass();
938 const Type* thread_type
939 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
940
941 Node* thread = _gvn.transform(new ThreadLocalNode());
942 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
943 tls_output = thread;
944
945 Node* thread_obj_handle
946 = (is_immutable
947 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
948 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
949 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
950 thread_obj_handle = _gvn.transform(thread_obj_handle);
951
952 DecoratorSet decorators = IN_NATIVE;
953 if (is_immutable) {
954 decorators |= C2_IMMUTABLE_MEMORY;
955 }
956 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
957 }
958
959 //--------------------------generate_current_thread--------------------
960 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
961 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
962 /*is_immutable*/false);
963 }
964
965 //--------------------------generate_virtual_thread--------------------
966 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
967 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
968 !C->method()->changes_current_thread());
969 }
970
971 //------------------------------make_string_method_node------------------------
972 // Helper method for String intrinsic functions. This version is called with
973 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
974 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
975 // containing the lengths of str1 and str2.
976 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
977 Node* result = nullptr;
978 switch (opcode) {
979 case Op_StrIndexOf:
980 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
981 str1_start, cnt1, str2_start, cnt2, ae);
982 break;
983 case Op_StrComp:
984 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
985 str1_start, cnt1, str2_start, cnt2, ae);
986 break;
987 case Op_StrEquals:
988 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
989 // Use the constant length if there is one because optimized match rule may exist.
990 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
991 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
992 break;
993 default:
994 ShouldNotReachHere();
995 return nullptr;
996 }
997
998 // All these intrinsics have checks.
999 C->set_has_split_ifs(true); // Has chance for split-if optimization
1000 clear_upper_avx();
1001
1002 return _gvn.transform(result);
1003 }
1004
1005 //------------------------------inline_string_compareTo------------------------
1006 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1007 Node* arg1 = argument(0);
1008 Node* arg2 = argument(1);
1009
1010 arg1 = must_be_not_null(arg1, true);
1011 arg2 = must_be_not_null(arg2, true);
1012
1013 // Get start addr and length of first argument
1014 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1015 Node* arg1_cnt = load_array_length(arg1);
1016
1017 // Get start addr and length of second argument
1018 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1019 Node* arg2_cnt = load_array_length(arg2);
1020
1021 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1022 set_result(result);
1023 return true;
1024 }
1025
1026 //------------------------------inline_string_equals------------------------
1027 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1028 Node* arg1 = argument(0);
1029 Node* arg2 = argument(1);
1030
1031 // paths (plus control) merge
1032 RegionNode* region = new RegionNode(3);
1033 Node* phi = new PhiNode(region, TypeInt::BOOL);
1034
1035 if (!stopped()) {
1036
1037 arg1 = must_be_not_null(arg1, true);
1038 arg2 = must_be_not_null(arg2, true);
1039
1040 // Get start addr and length of first argument
1041 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1042 Node* arg1_cnt = load_array_length(arg1);
1043
1044 // Get start addr and length of second argument
1045 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1046 Node* arg2_cnt = load_array_length(arg2);
1047
1048 // Check for arg1_cnt != arg2_cnt
1049 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1050 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1051 Node* if_ne = generate_slow_guard(bol, nullptr);
1052 if (if_ne != nullptr) {
1053 phi->init_req(2, intcon(0));
1054 region->init_req(2, if_ne);
1055 }
1056
1057 // Check for count == 0 is done by assembler code for StrEquals.
1058
1059 if (!stopped()) {
1060 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1061 phi->init_req(1, equals);
1062 region->init_req(1, control());
1063 }
1064 }
1065
1066 // post merge
1067 set_control(_gvn.transform(region));
1068 record_for_igvn(region);
1069
1070 set_result(_gvn.transform(phi));
1071 return true;
1072 }
1073
1074 //------------------------------inline_array_equals----------------------------
1075 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1076 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1077 Node* arg1 = argument(0);
1078 Node* arg2 = argument(1);
1079
1080 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1081 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1082 clear_upper_avx();
1083
1084 return true;
1085 }
1086
1087
1088 //------------------------------inline_countPositives------------------------------
1089 bool LibraryCallKit::inline_countPositives() {
1090 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1091 return false;
1092 }
1093
1094 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1095 // no receiver since it is static method
1096 Node* ba = argument(0);
1097 Node* offset = argument(1);
1098 Node* len = argument(2);
1099
1100 ba = must_be_not_null(ba, true);
1101
1102 // Range checks
1103 generate_string_range_check(ba, offset, len, false);
1104 if (stopped()) {
1105 return true;
1106 }
1107 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1108 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1109 set_result(_gvn.transform(result));
1110 clear_upper_avx();
1111 return true;
1112 }
1113
1114 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1115 Node* index = argument(0);
1116 Node* length = bt == T_INT ? argument(1) : argument(2);
1117 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1118 return false;
1119 }
1120
1121 // check that length is positive
1122 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1123 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1124
1125 {
1126 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1127 uncommon_trap(Deoptimization::Reason_intrinsic,
1128 Deoptimization::Action_make_not_entrant);
1129 }
1130
1131 if (stopped()) {
1132 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1133 return true;
1134 }
1135
1136 // length is now known positive, add a cast node to make this explicit
1137 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1138 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1139 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1140 ConstraintCastNode::RegularDependency, bt);
1141 casted_length = _gvn.transform(casted_length);
1142 replace_in_map(length, casted_length);
1143 length = casted_length;
1144
1145 // Use an unsigned comparison for the range check itself
1146 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1147 BoolTest::mask btest = BoolTest::lt;
1148 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1149 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1150 _gvn.set_type(rc, rc->Value(&_gvn));
1151 if (!rc_bool->is_Con()) {
1152 record_for_igvn(rc);
1153 }
1154 set_control(_gvn.transform(new IfTrueNode(rc)));
1155 {
1156 PreserveJVMState pjvms(this);
1157 set_control(_gvn.transform(new IfFalseNode(rc)));
1158 uncommon_trap(Deoptimization::Reason_range_check,
1159 Deoptimization::Action_make_not_entrant);
1160 }
1161
1162 if (stopped()) {
1163 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1164 return true;
1165 }
1166
1167 // index is now known to be >= 0 and < length, cast it
1168 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1169 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1170 ConstraintCastNode::RegularDependency, bt);
1171 result = _gvn.transform(result);
1172 set_result(result);
1173 replace_in_map(index, result);
1174 return true;
1175 }
1176
1177 //------------------------------inline_string_indexOf------------------------
1178 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1179 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1180 return false;
1181 }
1182 Node* src = argument(0);
1183 Node* tgt = argument(1);
1184
1185 // Make the merge point
1186 RegionNode* result_rgn = new RegionNode(4);
1187 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1188
1189 src = must_be_not_null(src, true);
1190 tgt = must_be_not_null(tgt, true);
1191
1192 // Get start addr and length of source string
1193 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1194 Node* src_count = load_array_length(src);
1195
1196 // Get start addr and length of substring
1197 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1198 Node* tgt_count = load_array_length(tgt);
1199
1200 Node* result = nullptr;
1201 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1202
1203 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1204 // Divide src size by 2 if String is UTF16 encoded
1205 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1206 }
1207 if (ae == StrIntrinsicNode::UU) {
1208 // Divide substring size by 2 if String is UTF16 encoded
1209 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1210 }
1211
1212 if (call_opt_stub) {
1213 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1214 StubRoutines::_string_indexof_array[ae],
1215 "stringIndexOf", TypePtr::BOTTOM, src_start,
1216 src_count, tgt_start, tgt_count);
1217 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1218 } else {
1219 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1220 result_rgn, result_phi, ae);
1221 }
1222 if (result != nullptr) {
1223 result_phi->init_req(3, result);
1224 result_rgn->init_req(3, control());
1225 }
1226 set_control(_gvn.transform(result_rgn));
1227 record_for_igvn(result_rgn);
1228 set_result(_gvn.transform(result_phi));
1229
1230 return true;
1231 }
1232
1233 //-----------------------------inline_string_indexOfI-----------------------
1234 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1235 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1236 return false;
1237 }
1238 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1239 return false;
1240 }
1241
1242 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1243 Node* src = argument(0); // byte[]
1244 Node* src_count = argument(1); // char count
1245 Node* tgt = argument(2); // byte[]
1246 Node* tgt_count = argument(3); // char count
1247 Node* from_index = argument(4); // char index
1248
1249 src = must_be_not_null(src, true);
1250 tgt = must_be_not_null(tgt, true);
1251
1252 // Multiply byte array index by 2 if String is UTF16 encoded
1253 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1254 src_count = _gvn.transform(new SubINode(src_count, from_index));
1255 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1256 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1257
1258 // Range checks
1259 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1260 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1261 if (stopped()) {
1262 return true;
1263 }
1264
1265 RegionNode* region = new RegionNode(5);
1266 Node* phi = new PhiNode(region, TypeInt::INT);
1267 Node* result = nullptr;
1268
1269 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1270
1271 if (call_opt_stub) {
1272 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1273 StubRoutines::_string_indexof_array[ae],
1274 "stringIndexOf", TypePtr::BOTTOM, src_start,
1275 src_count, tgt_start, tgt_count);
1276 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1277 } else {
1278 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1279 region, phi, ae);
1280 }
1281 if (result != nullptr) {
1282 // The result is index relative to from_index if substring was found, -1 otherwise.
1283 // Generate code which will fold into cmove.
1284 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1285 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1286
1287 Node* if_lt = generate_slow_guard(bol, nullptr);
1288 if (if_lt != nullptr) {
1289 // result == -1
1290 phi->init_req(3, result);
1291 region->init_req(3, if_lt);
1292 }
1293 if (!stopped()) {
1294 result = _gvn.transform(new AddINode(result, from_index));
1295 phi->init_req(4, result);
1296 region->init_req(4, control());
1297 }
1298 }
1299
1300 set_control(_gvn.transform(region));
1301 record_for_igvn(region);
1302 set_result(_gvn.transform(phi));
1303 clear_upper_avx();
1304
1305 return true;
1306 }
1307
1308 // Create StrIndexOfNode with fast path checks
1309 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1310 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1311 // Check for substr count > string count
1312 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1313 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1314 Node* if_gt = generate_slow_guard(bol, nullptr);
1315 if (if_gt != nullptr) {
1316 phi->init_req(1, intcon(-1));
1317 region->init_req(1, if_gt);
1318 }
1319 if (!stopped()) {
1320 // Check for substr count == 0
1321 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1322 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1323 Node* if_zero = generate_slow_guard(bol, nullptr);
1324 if (if_zero != nullptr) {
1325 phi->init_req(2, intcon(0));
1326 region->init_req(2, if_zero);
1327 }
1328 }
1329 if (!stopped()) {
1330 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1331 }
1332 return nullptr;
1333 }
1334
1335 //-----------------------------inline_string_indexOfChar-----------------------
1336 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1337 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1338 return false;
1339 }
1340 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1341 return false;
1342 }
1343 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1344 Node* src = argument(0); // byte[]
1345 Node* int_ch = argument(1);
1346 Node* from_index = argument(2);
1347 Node* max = argument(3);
1348
1349 src = must_be_not_null(src, true);
1350
1351 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1352 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1353 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1354
1355 // Range checks
1356 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1357
1358 // Check for int_ch >= 0
1359 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1360 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1361 {
1362 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1363 uncommon_trap(Deoptimization::Reason_intrinsic,
1364 Deoptimization::Action_maybe_recompile);
1365 }
1366 if (stopped()) {
1367 return true;
1368 }
1369
1370 RegionNode* region = new RegionNode(3);
1371 Node* phi = new PhiNode(region, TypeInt::INT);
1372
1373 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1374 C->set_has_split_ifs(true); // Has chance for split-if optimization
1375 _gvn.transform(result);
1376
1377 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1378 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1379
1380 Node* if_lt = generate_slow_guard(bol, nullptr);
1381 if (if_lt != nullptr) {
1382 // result == -1
1383 phi->init_req(2, result);
1384 region->init_req(2, if_lt);
1385 }
1386 if (!stopped()) {
1387 result = _gvn.transform(new AddINode(result, from_index));
1388 phi->init_req(1, result);
1389 region->init_req(1, control());
1390 }
1391 set_control(_gvn.transform(region));
1392 record_for_igvn(region);
1393 set_result(_gvn.transform(phi));
1394 clear_upper_avx();
1395
1396 return true;
1397 }
1398 //---------------------------inline_string_copy---------------------
1399 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1400 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1401 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1402 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1403 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1404 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1405 bool LibraryCallKit::inline_string_copy(bool compress) {
1406 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1407 return false;
1408 }
1409 int nargs = 5; // 2 oops, 3 ints
1410 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1411
1412 Node* src = argument(0);
1413 Node* src_offset = argument(1);
1414 Node* dst = argument(2);
1415 Node* dst_offset = argument(3);
1416 Node* length = argument(4);
1417
1418 // Check for allocation before we add nodes that would confuse
1419 // tightly_coupled_allocation()
1420 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1421
1422 // Figure out the size and type of the elements we will be copying.
1423 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1424 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1425 if (src_type == nullptr || dst_type == nullptr) {
1426 return false;
1427 }
1428 BasicType src_elem = src_type->elem()->array_element_basic_type();
1429 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1430 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1431 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1432 "Unsupported array types for inline_string_copy");
1433
1434 src = must_be_not_null(src, true);
1435 dst = must_be_not_null(dst, true);
1436
1437 // Convert char[] offsets to byte[] offsets
1438 bool convert_src = (compress && src_elem == T_BYTE);
1439 bool convert_dst = (!compress && dst_elem == T_BYTE);
1440 if (convert_src) {
1441 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1442 } else if (convert_dst) {
1443 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1444 }
1445
1446 // Range checks
1447 generate_string_range_check(src, src_offset, length, convert_src);
1448 generate_string_range_check(dst, dst_offset, length, convert_dst);
1449 if (stopped()) {
1450 return true;
1451 }
1452
1453 Node* src_start = array_element_address(src, src_offset, src_elem);
1454 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1455 // 'src_start' points to src array + scaled offset
1456 // 'dst_start' points to dst array + scaled offset
1457 Node* count = nullptr;
1458 if (compress) {
1459 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1460 } else {
1461 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1462 }
1463
1464 if (alloc != nullptr) {
1465 if (alloc->maybe_set_complete(&_gvn)) {
1466 // "You break it, you buy it."
1467 InitializeNode* init = alloc->initialization();
1468 assert(init->is_complete(), "we just did this");
1469 init->set_complete_with_arraycopy();
1470 assert(dst->is_CheckCastPP(), "sanity");
1471 assert(dst->in(0)->in(0) == init, "dest pinned");
1472 }
1473 // Do not let stores that initialize this object be reordered with
1474 // a subsequent store that would make this object accessible by
1475 // other threads.
1476 // Record what AllocateNode this StoreStore protects so that
1477 // escape analysis can go from the MemBarStoreStoreNode to the
1478 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1479 // based on the escape status of the AllocateNode.
1480 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1481 }
1482 if (compress) {
1483 set_result(_gvn.transform(count));
1484 }
1485 clear_upper_avx();
1486
1487 return true;
1488 }
1489
1490 #ifdef _LP64
1491 #define XTOP ,top() /*additional argument*/
1492 #else //_LP64
1493 #define XTOP /*no additional argument*/
1494 #endif //_LP64
1495
1496 //------------------------inline_string_toBytesU--------------------------
1497 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1498 bool LibraryCallKit::inline_string_toBytesU() {
1499 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1500 return false;
1501 }
1502 // Get the arguments.
1503 Node* value = argument(0);
1504 Node* offset = argument(1);
1505 Node* length = argument(2);
1506
1507 Node* newcopy = nullptr;
1508
1509 // Set the original stack and the reexecute bit for the interpreter to reexecute
1510 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1511 { PreserveReexecuteState preexecs(this);
1512 jvms()->set_should_reexecute(true);
1513
1514 // Check if a null path was taken unconditionally.
1515 value = null_check(value);
1516
1517 RegionNode* bailout = new RegionNode(1);
1518 record_for_igvn(bailout);
1519
1520 // Range checks
1521 generate_negative_guard(offset, bailout);
1522 generate_negative_guard(length, bailout);
1523 generate_limit_guard(offset, length, load_array_length(value), bailout);
1524 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1525 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1526
1527 if (bailout->req() > 1) {
1528 PreserveJVMState pjvms(this);
1529 set_control(_gvn.transform(bailout));
1530 uncommon_trap(Deoptimization::Reason_intrinsic,
1531 Deoptimization::Action_maybe_recompile);
1532 }
1533 if (stopped()) {
1534 return true;
1535 }
1536
1537 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1538 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1539 newcopy = new_array(klass_node, size, 0); // no arguments to push
1540 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1541 guarantee(alloc != nullptr, "created above");
1542
1543 // Calculate starting addresses.
1544 Node* src_start = array_element_address(value, offset, T_CHAR);
1545 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1546
1547 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1548 const TypeInt* toffset = gvn().type(offset)->is_int();
1549 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1550
1551 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1552 const char* copyfunc_name = "arraycopy";
1553 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1554 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1555 OptoRuntime::fast_arraycopy_Type(),
1556 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1557 src_start, dst_start, ConvI2X(length) XTOP);
1558 // Do not let reads from the cloned object float above the arraycopy.
1559 if (alloc->maybe_set_complete(&_gvn)) {
1560 // "You break it, you buy it."
1561 InitializeNode* init = alloc->initialization();
1562 assert(init->is_complete(), "we just did this");
1563 init->set_complete_with_arraycopy();
1564 assert(newcopy->is_CheckCastPP(), "sanity");
1565 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1566 }
1567 // Do not let stores that initialize this object be reordered with
1568 // a subsequent store that would make this object accessible by
1569 // other threads.
1570 // Record what AllocateNode this StoreStore protects so that
1571 // escape analysis can go from the MemBarStoreStoreNode to the
1572 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1573 // based on the escape status of the AllocateNode.
1574 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1575 } // original reexecute is set back here
1576
1577 C->set_has_split_ifs(true); // Has chance for split-if optimization
1578 if (!stopped()) {
1579 set_result(newcopy);
1580 }
1581 clear_upper_avx();
1582
1583 return true;
1584 }
1585
1586 //------------------------inline_string_getCharsU--------------------------
1587 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1588 bool LibraryCallKit::inline_string_getCharsU() {
1589 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1590 return false;
1591 }
1592
1593 // Get the arguments.
1594 Node* src = argument(0);
1595 Node* src_begin = argument(1);
1596 Node* src_end = argument(2); // exclusive offset (i < src_end)
1597 Node* dst = argument(3);
1598 Node* dst_begin = argument(4);
1599
1600 // Check for allocation before we add nodes that would confuse
1601 // tightly_coupled_allocation()
1602 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1603
1604 // Check if a null path was taken unconditionally.
1605 src = null_check(src);
1606 dst = null_check(dst);
1607 if (stopped()) {
1608 return true;
1609 }
1610
1611 // Get length and convert char[] offset to byte[] offset
1612 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1613 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1614
1615 // Range checks
1616 generate_string_range_check(src, src_begin, length, true);
1617 generate_string_range_check(dst, dst_begin, length, false);
1618 if (stopped()) {
1619 return true;
1620 }
1621
1622 if (!stopped()) {
1623 // Calculate starting addresses.
1624 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1625 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1626
1627 // Check if array addresses are aligned to HeapWordSize
1628 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1629 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1630 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1631 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1632
1633 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1634 const char* copyfunc_name = "arraycopy";
1635 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1636 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1637 OptoRuntime::fast_arraycopy_Type(),
1638 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1639 src_start, dst_start, ConvI2X(length) XTOP);
1640 // Do not let reads from the cloned object float above the arraycopy.
1641 if (alloc != nullptr) {
1642 if (alloc->maybe_set_complete(&_gvn)) {
1643 // "You break it, you buy it."
1644 InitializeNode* init = alloc->initialization();
1645 assert(init->is_complete(), "we just did this");
1646 init->set_complete_with_arraycopy();
1647 assert(dst->is_CheckCastPP(), "sanity");
1648 assert(dst->in(0)->in(0) == init, "dest pinned");
1649 }
1650 // Do not let stores that initialize this object be reordered with
1651 // a subsequent store that would make this object accessible by
1652 // other threads.
1653 // Record what AllocateNode this StoreStore protects so that
1654 // escape analysis can go from the MemBarStoreStoreNode to the
1655 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1656 // based on the escape status of the AllocateNode.
1657 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1658 } else {
1659 insert_mem_bar(Op_MemBarCPUOrder);
1660 }
1661 }
1662
1663 C->set_has_split_ifs(true); // Has chance for split-if optimization
1664 return true;
1665 }
1666
1667 //----------------------inline_string_char_access----------------------------
1668 // Store/Load char to/from byte[] array.
1669 // static void StringUTF16.putChar(byte[] val, int index, int c)
1670 // static char StringUTF16.getChar(byte[] val, int index)
1671 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1672 Node* value = argument(0);
1673 Node* index = argument(1);
1674 Node* ch = is_store ? argument(2) : nullptr;
1675
1676 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1677 // correctly requires matched array shapes.
1678 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1679 "sanity: byte[] and char[] bases agree");
1680 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1681 "sanity: byte[] and char[] scales agree");
1682
1683 // Bail when getChar over constants is requested: constant folding would
1684 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1685 // Java method would constant fold nicely instead.
1686 if (!is_store && value->is_Con() && index->is_Con()) {
1687 return false;
1688 }
1689
1690 // Save state and restore on bailout
1691 uint old_sp = sp();
1692 SafePointNode* old_map = clone_map();
1693
1694 value = must_be_not_null(value, true);
1695
1696 Node* adr = array_element_address(value, index, T_CHAR);
1697 if (adr->is_top()) {
1698 set_map(old_map);
1699 set_sp(old_sp);
1700 return false;
1701 }
1702 destruct_map_clone(old_map);
1703 if (is_store) {
1704 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1705 } else {
1706 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);
1707 set_result(ch);
1708 }
1709 return true;
1710 }
1711
1712 //--------------------------round_double_node--------------------------------
1713 // Round a double node if necessary.
1714 Node* LibraryCallKit::round_double_node(Node* n) {
1715 if (Matcher::strict_fp_requires_explicit_rounding) {
1716 #ifdef IA32
1717 if (UseSSE < 2) {
1718 n = _gvn.transform(new RoundDoubleNode(nullptr, n));
1719 }
1720 #else
1721 Unimplemented();
1722 #endif // IA32
1723 }
1724 return n;
1725 }
1726
1727 //------------------------------inline_math-----------------------------------
1728 // public static double Math.abs(double)
1729 // public static double Math.sqrt(double)
1730 // public static double Math.log(double)
1731 // public static double Math.log10(double)
1732 // public static double Math.round(double)
1733 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1734 Node* arg = round_double_node(argument(0));
1735 Node* n = nullptr;
1736 switch (id) {
1737 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1738 case vmIntrinsics::_dsqrt:
1739 case vmIntrinsics::_dsqrt_strict:
1740 n = new SqrtDNode(C, control(), arg); break;
1741 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1742 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1743 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1744 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1745 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1746 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1747 default: fatal_unexpected_iid(id); break;
1748 }
1749 set_result(_gvn.transform(n));
1750 return true;
1751 }
1752
1753 //------------------------------inline_math-----------------------------------
1754 // public static float Math.abs(float)
1755 // public static int Math.abs(int)
1756 // public static long Math.abs(long)
1757 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1758 Node* arg = argument(0);
1759 Node* n = nullptr;
1760 switch (id) {
1761 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1762 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1763 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1764 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1765 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1766 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1767 default: fatal_unexpected_iid(id); break;
1768 }
1769 set_result(_gvn.transform(n));
1770 return true;
1771 }
1772
1773 //------------------------------runtime_math-----------------------------
1774 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1775 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1776 "must be (DD)D or (D)D type");
1777
1778 // Inputs
1779 Node* a = round_double_node(argument(0));
1780 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : nullptr;
1781
1782 const TypePtr* no_memory_effects = nullptr;
1783 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1784 no_memory_effects,
1785 a, top(), b, b ? top() : nullptr);
1786 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1787 #ifdef ASSERT
1788 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1789 assert(value_top == top(), "second value must be top");
1790 #endif
1791
1792 set_result(value);
1793 return true;
1794 }
1795
1796 //------------------------------inline_math_pow-----------------------------
1797 bool LibraryCallKit::inline_math_pow() {
1798 Node* exp = round_double_node(argument(2));
1799 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1800 if (d != nullptr) {
1801 if (d->getd() == 2.0) {
1802 // Special case: pow(x, 2.0) => x * x
1803 Node* base = round_double_node(argument(0));
1804 set_result(_gvn.transform(new MulDNode(base, base)));
1805 return true;
1806 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1807 // Special case: pow(x, 0.5) => sqrt(x)
1808 Node* base = round_double_node(argument(0));
1809 Node* zero = _gvn.zerocon(T_DOUBLE);
1810
1811 RegionNode* region = new RegionNode(3);
1812 Node* phi = new PhiNode(region, Type::DOUBLE);
1813
1814 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1815 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1816 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1817 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1818 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1819
1820 Node* if_pow = generate_slow_guard(test, nullptr);
1821 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1822 phi->init_req(1, value_sqrt);
1823 region->init_req(1, control());
1824
1825 if (if_pow != nullptr) {
1826 set_control(if_pow);
1827 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1828 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1829 const TypePtr* no_memory_effects = nullptr;
1830 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1831 no_memory_effects, base, top(), exp, top());
1832 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1833 #ifdef ASSERT
1834 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1835 assert(value_top == top(), "second value must be top");
1836 #endif
1837 phi->init_req(2, value_pow);
1838 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1839 }
1840
1841 C->set_has_split_ifs(true); // Has chance for split-if optimization
1842 set_control(_gvn.transform(region));
1843 record_for_igvn(region);
1844 set_result(_gvn.transform(phi));
1845
1846 return true;
1847 }
1848 }
1849
1850 return StubRoutines::dpow() != nullptr ?
1851 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1852 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1853 }
1854
1855 //------------------------------inline_math_native-----------------------------
1856 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1857 switch (id) {
1858 case vmIntrinsics::_dsin:
1859 return StubRoutines::dsin() != nullptr ?
1860 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1861 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1862 case vmIntrinsics::_dcos:
1863 return StubRoutines::dcos() != nullptr ?
1864 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1865 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1866 case vmIntrinsics::_dtan:
1867 return StubRoutines::dtan() != nullptr ?
1868 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1869 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1870 case vmIntrinsics::_dtanh:
1871 return StubRoutines::dtanh() != nullptr ?
1872 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1873 case vmIntrinsics::_dexp:
1874 return StubRoutines::dexp() != nullptr ?
1875 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1876 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1877 case vmIntrinsics::_dlog:
1878 return StubRoutines::dlog() != nullptr ?
1879 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1880 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1881 case vmIntrinsics::_dlog10:
1882 return StubRoutines::dlog10() != nullptr ?
1883 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1884 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1885
1886 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1887 case vmIntrinsics::_ceil:
1888 case vmIntrinsics::_floor:
1889 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1890
1891 case vmIntrinsics::_dsqrt:
1892 case vmIntrinsics::_dsqrt_strict:
1893 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1894 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1895 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1896 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1897 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1898
1899 case vmIntrinsics::_dpow: return inline_math_pow();
1900 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1901 case vmIntrinsics::_fcopySign: return inline_math(id);
1902 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1903 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1904 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1905
1906 // These intrinsics are not yet correctly implemented
1907 case vmIntrinsics::_datan2:
1908 return false;
1909
1910 default:
1911 fatal_unexpected_iid(id);
1912 return false;
1913 }
1914 }
1915
1916 //----------------------------inline_notify-----------------------------------*
1917 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1918 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1919 address func;
1920 if (id == vmIntrinsics::_notify) {
1921 func = OptoRuntime::monitor_notify_Java();
1922 } else {
1923 func = OptoRuntime::monitor_notifyAll_Java();
1924 }
1925 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1926 make_slow_call_ex(call, env()->Throwable_klass(), false);
1927 return true;
1928 }
1929
1930
1931 //----------------------------inline_min_max-----------------------------------
1932 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1933 set_result(generate_min_max(id, argument(0), argument(1)));
1934 return true;
1935 }
1936
1937 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1938 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1939 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1940 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1941 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1942
1943 {
1944 PreserveJVMState pjvms(this);
1945 PreserveReexecuteState preexecs(this);
1946 jvms()->set_should_reexecute(true);
1947
1948 set_control(slow_path);
1949 set_i_o(i_o());
1950
1951 uncommon_trap(Deoptimization::Reason_intrinsic,
1952 Deoptimization::Action_none);
1953 }
1954
1955 set_control(fast_path);
1956 set_result(math);
1957 }
1958
1959 template <typename OverflowOp>
1960 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1961 typedef typename OverflowOp::MathOp MathOp;
1962
1963 MathOp* mathOp = new MathOp(arg1, arg2);
1964 Node* operation = _gvn.transform( mathOp );
1965 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1966 inline_math_mathExact(operation, ofcheck);
1967 return true;
1968 }
1969
1970 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1971 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1972 }
1973
1974 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1975 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1976 }
1977
1978 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1979 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1980 }
1981
1982 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1983 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1984 }
1985
1986 bool LibraryCallKit::inline_math_negateExactI() {
1987 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1988 }
1989
1990 bool LibraryCallKit::inline_math_negateExactL() {
1991 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1992 }
1993
1994 bool LibraryCallKit::inline_math_multiplyExactI() {
1995 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1996 }
1997
1998 bool LibraryCallKit::inline_math_multiplyExactL() {
1999 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2000 }
2001
2002 bool LibraryCallKit::inline_math_multiplyHigh() {
2003 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2004 return true;
2005 }
2006
2007 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2008 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2009 return true;
2010 }
2011
2012 Node*
2013 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
2014 Node* result_val = nullptr;
2015 switch (id) {
2016 case vmIntrinsics::_min:
2017 case vmIntrinsics::_min_strict:
2018 result_val = _gvn.transform(new MinINode(x0, y0));
2019 break;
2020 case vmIntrinsics::_max:
2021 case vmIntrinsics::_max_strict:
2022 result_val = _gvn.transform(new MaxINode(x0, y0));
2023 break;
2024 default:
2025 fatal_unexpected_iid(id);
2026 break;
2027 }
2028 return result_val;
2029 }
2030
2031 inline int
2032 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2033 const TypePtr* base_type = TypePtr::NULL_PTR;
2034 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2035 if (base_type == nullptr) {
2036 // Unknown type.
2037 return Type::AnyPtr;
2038 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2039 // Since this is a null+long form, we have to switch to a rawptr.
2040 base = _gvn.transform(new CastX2PNode(offset));
2041 offset = MakeConX(0);
2042 return Type::RawPtr;
2043 } else if (base_type->base() == Type::RawPtr) {
2044 return Type::RawPtr;
2045 } else if (base_type->isa_oopptr()) {
2046 // Base is never null => always a heap address.
2047 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2048 return Type::OopPtr;
2049 }
2050 // Offset is small => always a heap address.
2051 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2052 if (offset_type != nullptr &&
2053 base_type->offset() == 0 && // (should always be?)
2054 offset_type->_lo >= 0 &&
2055 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2056 return Type::OopPtr;
2057 } else if (type == T_OBJECT) {
2058 // off heap access to an oop doesn't make any sense. Has to be on
2059 // heap.
2060 return Type::OopPtr;
2061 }
2062 // Otherwise, it might either be oop+off or null+addr.
2063 return Type::AnyPtr;
2064 } else {
2065 // No information:
2066 return Type::AnyPtr;
2067 }
2068 }
2069
2070 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2071 Node* uncasted_base = base;
2072 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2073 if (kind == Type::RawPtr) {
2074 return basic_plus_adr(top(), uncasted_base, offset);
2075 } else if (kind == Type::AnyPtr) {
2076 assert(base == uncasted_base, "unexpected base change");
2077 if (can_cast) {
2078 if (!_gvn.type(base)->speculative_maybe_null() &&
2079 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2080 // According to profiling, this access is always on
2081 // heap. Casting the base to not null and thus avoiding membars
2082 // around the access should allow better optimizations
2083 Node* null_ctl = top();
2084 base = null_check_oop(base, &null_ctl, true, true, true);
2085 assert(null_ctl->is_top(), "no null control here");
2086 return basic_plus_adr(base, offset);
2087 } else if (_gvn.type(base)->speculative_always_null() &&
2088 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2089 // According to profiling, this access is always off
2090 // heap.
2091 base = null_assert(base);
2092 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2093 offset = MakeConX(0);
2094 return basic_plus_adr(top(), raw_base, offset);
2095 }
2096 }
2097 // We don't know if it's an on heap or off heap access. Fall back
2098 // to raw memory access.
2099 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2100 return basic_plus_adr(top(), raw, offset);
2101 } else {
2102 assert(base == uncasted_base, "unexpected base change");
2103 // We know it's an on heap access so base can't be null
2104 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2105 base = must_be_not_null(base, true);
2106 }
2107 return basic_plus_adr(base, offset);
2108 }
2109 }
2110
2111 //--------------------------inline_number_methods-----------------------------
2112 // inline int Integer.numberOfLeadingZeros(int)
2113 // inline int Long.numberOfLeadingZeros(long)
2114 //
2115 // inline int Integer.numberOfTrailingZeros(int)
2116 // inline int Long.numberOfTrailingZeros(long)
2117 //
2118 // inline int Integer.bitCount(int)
2119 // inline int Long.bitCount(long)
2120 //
2121 // inline char Character.reverseBytes(char)
2122 // inline short Short.reverseBytes(short)
2123 // inline int Integer.reverseBytes(int)
2124 // inline long Long.reverseBytes(long)
2125 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2126 Node* arg = argument(0);
2127 Node* n = nullptr;
2128 switch (id) {
2129 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2130 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2131 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2132 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2133 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2134 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2135 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2136 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2137 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2138 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2139 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2140 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2141 default: fatal_unexpected_iid(id); break;
2142 }
2143 set_result(_gvn.transform(n));
2144 return true;
2145 }
2146
2147 //--------------------------inline_bitshuffle_methods-----------------------------
2148 // inline int Integer.compress(int, int)
2149 // inline int Integer.expand(int, int)
2150 // inline long Long.compress(long, long)
2151 // inline long Long.expand(long, long)
2152 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2153 Node* n = nullptr;
2154 switch (id) {
2155 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2156 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2157 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2158 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2159 default: fatal_unexpected_iid(id); break;
2160 }
2161 set_result(_gvn.transform(n));
2162 return true;
2163 }
2164
2165 //--------------------------inline_number_methods-----------------------------
2166 // inline int Integer.compareUnsigned(int, int)
2167 // inline int Long.compareUnsigned(long, long)
2168 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2169 Node* arg1 = argument(0);
2170 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2171 Node* n = nullptr;
2172 switch (id) {
2173 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2174 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2175 default: fatal_unexpected_iid(id); break;
2176 }
2177 set_result(_gvn.transform(n));
2178 return true;
2179 }
2180
2181 //--------------------------inline_unsigned_divmod_methods-----------------------------
2182 // inline int Integer.divideUnsigned(int, int)
2183 // inline int Integer.remainderUnsigned(int, int)
2184 // inline long Long.divideUnsigned(long, long)
2185 // inline long Long.remainderUnsigned(long, long)
2186 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2187 Node* n = nullptr;
2188 switch (id) {
2189 case vmIntrinsics::_divideUnsigned_i: {
2190 zero_check_int(argument(1));
2191 // Compile-time detect of null-exception
2192 if (stopped()) {
2193 return true; // keep the graph constructed so far
2194 }
2195 n = new UDivINode(control(), argument(0), argument(1));
2196 break;
2197 }
2198 case vmIntrinsics::_divideUnsigned_l: {
2199 zero_check_long(argument(2));
2200 // Compile-time detect of null-exception
2201 if (stopped()) {
2202 return true; // keep the graph constructed so far
2203 }
2204 n = new UDivLNode(control(), argument(0), argument(2));
2205 break;
2206 }
2207 case vmIntrinsics::_remainderUnsigned_i: {
2208 zero_check_int(argument(1));
2209 // Compile-time detect of null-exception
2210 if (stopped()) {
2211 return true; // keep the graph constructed so far
2212 }
2213 n = new UModINode(control(), argument(0), argument(1));
2214 break;
2215 }
2216 case vmIntrinsics::_remainderUnsigned_l: {
2217 zero_check_long(argument(2));
2218 // Compile-time detect of null-exception
2219 if (stopped()) {
2220 return true; // keep the graph constructed so far
2221 }
2222 n = new UModLNode(control(), argument(0), argument(2));
2223 break;
2224 }
2225 default: fatal_unexpected_iid(id); break;
2226 }
2227 set_result(_gvn.transform(n));
2228 return true;
2229 }
2230
2231 //----------------------------inline_unsafe_access----------------------------
2232
2233 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2234 // Attempt to infer a sharper value type from the offset and base type.
2235 ciKlass* sharpened_klass = nullptr;
2236
2237 // See if it is an instance field, with an object type.
2238 if (alias_type->field() != nullptr) {
2239 if (alias_type->field()->type()->is_klass()) {
2240 sharpened_klass = alias_type->field()->type()->as_klass();
2241 }
2242 }
2243
2244 const TypeOopPtr* result = nullptr;
2245 // See if it is a narrow oop array.
2246 if (adr_type->isa_aryptr()) {
2247 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2248 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2249 if (elem_type != nullptr && elem_type->is_loaded()) {
2250 // Sharpen the value type.
2251 result = elem_type;
2252 }
2253 }
2254 }
2255
2256 // The sharpened class might be unloaded if there is no class loader
2257 // contraint in place.
2258 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2259 // Sharpen the value type.
2260 result = TypeOopPtr::make_from_klass(sharpened_klass);
2261 }
2262 if (result != nullptr) {
2263 #ifndef PRODUCT
2264 if (C->print_intrinsics() || C->print_inlining()) {
2265 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2266 tty->print(" sharpened value: "); result->dump(); tty->cr();
2267 }
2268 #endif
2269 }
2270 return result;
2271 }
2272
2273 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2274 switch (kind) {
2275 case Relaxed:
2276 return MO_UNORDERED;
2277 case Opaque:
2278 return MO_RELAXED;
2279 case Acquire:
2280 return MO_ACQUIRE;
2281 case Release:
2282 return MO_RELEASE;
2283 case Volatile:
2284 return MO_SEQ_CST;
2285 default:
2286 ShouldNotReachHere();
2287 return 0;
2288 }
2289 }
2290
2291 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2292 if (callee()->is_static()) return false; // caller must have the capability!
2293 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2294 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2295 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2296 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2297
2298 if (is_reference_type(type)) {
2299 decorators |= ON_UNKNOWN_OOP_REF;
2300 }
2301
2302 if (unaligned) {
2303 decorators |= C2_UNALIGNED;
2304 }
2305
2306 #ifndef PRODUCT
2307 {
2308 ResourceMark rm;
2309 // Check the signatures.
2310 ciSignature* sig = callee()->signature();
2311 #ifdef ASSERT
2312 if (!is_store) {
2313 // Object getReference(Object base, int/long offset), etc.
2314 BasicType rtype = sig->return_type()->basic_type();
2315 assert(rtype == type, "getter must return the expected value");
2316 assert(sig->count() == 2, "oop getter has 2 arguments");
2317 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2318 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2319 } else {
2320 // void putReference(Object base, int/long offset, Object x), etc.
2321 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2322 assert(sig->count() == 3, "oop putter has 3 arguments");
2323 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2324 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2325 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2326 assert(vtype == type, "putter must accept the expected value");
2327 }
2328 #endif // ASSERT
2329 }
2330 #endif //PRODUCT
2331
2332 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2333
2334 Node* receiver = argument(0); // type: oop
2335
2336 // Build address expression.
2337 Node* heap_base_oop = top();
2338
2339 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2340 Node* base = argument(1); // type: oop
2341 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2342 Node* offset = argument(2); // type: long
2343 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2344 // to be plain byte offsets, which are also the same as those accepted
2345 // by oopDesc::field_addr.
2346 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2347 "fieldOffset must be byte-scaled");
2348 // 32-bit machines ignore the high half!
2349 offset = ConvL2X(offset);
2350
2351 // Save state and restore on bailout
2352 uint old_sp = sp();
2353 SafePointNode* old_map = clone_map();
2354
2355 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2356 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2357
2358 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2359 if (type != T_OBJECT) {
2360 decorators |= IN_NATIVE; // off-heap primitive access
2361 } else {
2362 set_map(old_map);
2363 set_sp(old_sp);
2364 return false; // off-heap oop accesses are not supported
2365 }
2366 } else {
2367 heap_base_oop = base; // on-heap or mixed access
2368 }
2369
2370 // Can base be null? Otherwise, always on-heap access.
2371 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2372
2373 if (!can_access_non_heap) {
2374 decorators |= IN_HEAP;
2375 }
2376
2377 Node* val = is_store ? argument(4) : nullptr;
2378
2379 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2380 if (adr_type == TypePtr::NULL_PTR) {
2381 set_map(old_map);
2382 set_sp(old_sp);
2383 return false; // off-heap access with zero address
2384 }
2385
2386 // Try to categorize the address.
2387 Compile::AliasType* alias_type = C->alias_type(adr_type);
2388 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2389
2390 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2391 alias_type->adr_type() == TypeAryPtr::RANGE) {
2392 set_map(old_map);
2393 set_sp(old_sp);
2394 return false; // not supported
2395 }
2396
2397 bool mismatched = false;
2398 BasicType bt = alias_type->basic_type();
2399 if (bt != T_ILLEGAL) {
2400 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2401 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2402 // Alias type doesn't differentiate between byte[] and boolean[]).
2403 // Use address type to get the element type.
2404 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2405 }
2406 if (is_reference_type(bt, true)) {
2407 // accessing an array field with getReference is not a mismatch
2408 bt = T_OBJECT;
2409 }
2410 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2411 // Don't intrinsify mismatched object accesses
2412 set_map(old_map);
2413 set_sp(old_sp);
2414 return false;
2415 }
2416 mismatched = (bt != type);
2417 } else if (alias_type->adr_type()->isa_oopptr()) {
2418 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2419 }
2420
2421 destruct_map_clone(old_map);
2422 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2423
2424 if (mismatched) {
2425 decorators |= C2_MISMATCHED;
2426 }
2427
2428 // First guess at the value type.
2429 const Type *value_type = Type::get_const_basic_type(type);
2430
2431 // Figure out the memory ordering.
2432 decorators |= mo_decorator_for_access_kind(kind);
2433
2434 if (!is_store && type == T_OBJECT) {
2435 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2436 if (tjp != nullptr) {
2437 value_type = tjp;
2438 }
2439 }
2440
2441 receiver = null_check(receiver);
2442 if (stopped()) {
2443 return true;
2444 }
2445 // Heap pointers get a null-check from the interpreter,
2446 // as a courtesy. However, this is not guaranteed by Unsafe,
2447 // and it is not possible to fully distinguish unintended nulls
2448 // from intended ones in this API.
2449
2450 if (!is_store) {
2451 Node* p = nullptr;
2452 // Try to constant fold a load from a constant field
2453 ciField* field = alias_type->field();
2454 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) {
2455 // final or stable field
2456 p = make_constant_from_field(field, heap_base_oop);
2457 }
2458
2459 if (p == nullptr) { // Could not constant fold the load
2460 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2461 // Normalize the value returned by getBoolean in the following cases
2462 if (type == T_BOOLEAN &&
2463 (mismatched ||
2464 heap_base_oop == top() || // - heap_base_oop is null or
2465 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2466 // and the unsafe access is made to large offset
2467 // (i.e., larger than the maximum offset necessary for any
2468 // field access)
2469 ) {
2470 IdealKit ideal = IdealKit(this);
2471 #define __ ideal.
2472 IdealVariable normalized_result(ideal);
2473 __ declarations_done();
2474 __ set(normalized_result, p);
2475 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2476 __ set(normalized_result, ideal.ConI(1));
2477 ideal.end_if();
2478 final_sync(ideal);
2479 p = __ value(normalized_result);
2480 #undef __
2481 }
2482 }
2483 if (type == T_ADDRESS) {
2484 p = gvn().transform(new CastP2XNode(nullptr, p));
2485 p = ConvX2UL(p);
2486 }
2487 // The load node has the control of the preceding MemBarCPUOrder. All
2488 // following nodes will have the control of the MemBarCPUOrder inserted at
2489 // the end of this method. So, pushing the load onto the stack at a later
2490 // point is fine.
2491 set_result(p);
2492 } else {
2493 if (bt == T_ADDRESS) {
2494 // Repackage the long as a pointer.
2495 val = ConvL2X(val);
2496 val = gvn().transform(new CastX2PNode(val));
2497 }
2498 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2499 }
2500
2501 return true;
2502 }
2503
2504 //----------------------------inline_unsafe_load_store----------------------------
2505 // This method serves a couple of different customers (depending on LoadStoreKind):
2506 //
2507 // LS_cmp_swap:
2508 //
2509 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2510 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2511 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2512 //
2513 // LS_cmp_swap_weak:
2514 //
2515 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2516 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2517 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2518 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2519 //
2520 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2521 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2522 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2523 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2524 //
2525 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2526 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2527 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2528 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2529 //
2530 // LS_cmp_exchange:
2531 //
2532 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2533 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2534 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2535 //
2536 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2537 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2538 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2539 //
2540 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2541 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2542 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2543 //
2544 // LS_get_add:
2545 //
2546 // int getAndAddInt( Object o, long offset, int delta)
2547 // long getAndAddLong(Object o, long offset, long delta)
2548 //
2549 // LS_get_set:
2550 //
2551 // int getAndSet(Object o, long offset, int newValue)
2552 // long getAndSet(Object o, long offset, long newValue)
2553 // Object getAndSet(Object o, long offset, Object newValue)
2554 //
2555 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2556 // This basic scheme here is the same as inline_unsafe_access, but
2557 // differs in enough details that combining them would make the code
2558 // overly confusing. (This is a true fact! I originally combined
2559 // them, but even I was confused by it!) As much code/comments as
2560 // possible are retained from inline_unsafe_access though to make
2561 // the correspondences clearer. - dl
2562
2563 if (callee()->is_static()) return false; // caller must have the capability!
2564
2565 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2566 decorators |= mo_decorator_for_access_kind(access_kind);
2567
2568 #ifndef PRODUCT
2569 BasicType rtype;
2570 {
2571 ResourceMark rm;
2572 // Check the signatures.
2573 ciSignature* sig = callee()->signature();
2574 rtype = sig->return_type()->basic_type();
2575 switch(kind) {
2576 case LS_get_add:
2577 case LS_get_set: {
2578 // Check the signatures.
2579 #ifdef ASSERT
2580 assert(rtype == type, "get and set must return the expected type");
2581 assert(sig->count() == 3, "get and set has 3 arguments");
2582 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2583 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2584 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2585 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2586 #endif // ASSERT
2587 break;
2588 }
2589 case LS_cmp_swap:
2590 case LS_cmp_swap_weak: {
2591 // Check the signatures.
2592 #ifdef ASSERT
2593 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2594 assert(sig->count() == 4, "CAS has 4 arguments");
2595 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2596 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2597 #endif // ASSERT
2598 break;
2599 }
2600 case LS_cmp_exchange: {
2601 // Check the signatures.
2602 #ifdef ASSERT
2603 assert(rtype == type, "CAS must return the expected type");
2604 assert(sig->count() == 4, "CAS has 4 arguments");
2605 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2606 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2607 #endif // ASSERT
2608 break;
2609 }
2610 default:
2611 ShouldNotReachHere();
2612 }
2613 }
2614 #endif //PRODUCT
2615
2616 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2617
2618 // Get arguments:
2619 Node* receiver = nullptr;
2620 Node* base = nullptr;
2621 Node* offset = nullptr;
2622 Node* oldval = nullptr;
2623 Node* newval = nullptr;
2624 switch(kind) {
2625 case LS_cmp_swap:
2626 case LS_cmp_swap_weak:
2627 case LS_cmp_exchange: {
2628 const bool two_slot_type = type2size[type] == 2;
2629 receiver = argument(0); // type: oop
2630 base = argument(1); // type: oop
2631 offset = argument(2); // type: long
2632 oldval = argument(4); // type: oop, int, or long
2633 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2634 break;
2635 }
2636 case LS_get_add:
2637 case LS_get_set: {
2638 receiver = argument(0); // type: oop
2639 base = argument(1); // type: oop
2640 offset = argument(2); // type: long
2641 oldval = nullptr;
2642 newval = argument(4); // type: oop, int, or long
2643 break;
2644 }
2645 default:
2646 ShouldNotReachHere();
2647 }
2648
2649 // Build field offset expression.
2650 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2651 // to be plain byte offsets, which are also the same as those accepted
2652 // by oopDesc::field_addr.
2653 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2654 // 32-bit machines ignore the high half of long offsets
2655 offset = ConvL2X(offset);
2656 // Save state and restore on bailout
2657 uint old_sp = sp();
2658 SafePointNode* old_map = clone_map();
2659 Node* adr = make_unsafe_address(base, offset,type, false);
2660 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2661
2662 Compile::AliasType* alias_type = C->alias_type(adr_type);
2663 BasicType bt = alias_type->basic_type();
2664 if (bt != T_ILLEGAL &&
2665 (is_reference_type(bt) != (type == T_OBJECT))) {
2666 // Don't intrinsify mismatched object accesses.
2667 set_map(old_map);
2668 set_sp(old_sp);
2669 return false;
2670 }
2671
2672 destruct_map_clone(old_map);
2673
2674 // For CAS, unlike inline_unsafe_access, there seems no point in
2675 // trying to refine types. Just use the coarse types here.
2676 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2677 const Type *value_type = Type::get_const_basic_type(type);
2678
2679 switch (kind) {
2680 case LS_get_set:
2681 case LS_cmp_exchange: {
2682 if (type == T_OBJECT) {
2683 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2684 if (tjp != nullptr) {
2685 value_type = tjp;
2686 }
2687 }
2688 break;
2689 }
2690 case LS_cmp_swap:
2691 case LS_cmp_swap_weak:
2692 case LS_get_add:
2693 break;
2694 default:
2695 ShouldNotReachHere();
2696 }
2697
2698 // Null check receiver.
2699 receiver = null_check(receiver);
2700 if (stopped()) {
2701 return true;
2702 }
2703
2704 int alias_idx = C->get_alias_index(adr_type);
2705
2706 if (is_reference_type(type)) {
2707 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2708
2709 // Transformation of a value which could be null pointer (CastPP #null)
2710 // could be delayed during Parse (for example, in adjust_map_after_if()).
2711 // Execute transformation here to avoid barrier generation in such case.
2712 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2713 newval = _gvn.makecon(TypePtr::NULL_PTR);
2714
2715 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2716 // Refine the value to a null constant, when it is known to be null
2717 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2718 }
2719 }
2720
2721 Node* result = nullptr;
2722 switch (kind) {
2723 case LS_cmp_exchange: {
2724 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2725 oldval, newval, value_type, type, decorators);
2726 break;
2727 }
2728 case LS_cmp_swap_weak:
2729 decorators |= C2_WEAK_CMPXCHG;
2730 case LS_cmp_swap: {
2731 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2732 oldval, newval, value_type, type, decorators);
2733 break;
2734 }
2735 case LS_get_set: {
2736 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2737 newval, value_type, type, decorators);
2738 break;
2739 }
2740 case LS_get_add: {
2741 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2742 newval, value_type, type, decorators);
2743 break;
2744 }
2745 default:
2746 ShouldNotReachHere();
2747 }
2748
2749 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2750 set_result(result);
2751 return true;
2752 }
2753
2754 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2755 // Regardless of form, don't allow previous ld/st to move down,
2756 // then issue acquire, release, or volatile mem_bar.
2757 insert_mem_bar(Op_MemBarCPUOrder);
2758 switch(id) {
2759 case vmIntrinsics::_loadFence:
2760 insert_mem_bar(Op_LoadFence);
2761 return true;
2762 case vmIntrinsics::_storeFence:
2763 insert_mem_bar(Op_StoreFence);
2764 return true;
2765 case vmIntrinsics::_storeStoreFence:
2766 insert_mem_bar(Op_StoreStoreFence);
2767 return true;
2768 case vmIntrinsics::_fullFence:
2769 insert_mem_bar(Op_MemBarVolatile);
2770 return true;
2771 default:
2772 fatal_unexpected_iid(id);
2773 return false;
2774 }
2775 }
2776
2777 bool LibraryCallKit::inline_onspinwait() {
2778 insert_mem_bar(Op_OnSpinWait);
2779 return true;
2780 }
2781
2782 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2783 if (!kls->is_Con()) {
2784 return true;
2785 }
2786 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2787 if (klsptr == nullptr) {
2788 return true;
2789 }
2790 ciInstanceKlass* ik = klsptr->instance_klass();
2791 // don't need a guard for a klass that is already initialized
2792 return !ik->is_initialized();
2793 }
2794
2795 //----------------------------inline_unsafe_writeback0-------------------------
2796 // public native void Unsafe.writeback0(long address)
2797 bool LibraryCallKit::inline_unsafe_writeback0() {
2798 if (!Matcher::has_match_rule(Op_CacheWB)) {
2799 return false;
2800 }
2801 #ifndef PRODUCT
2802 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2803 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2804 ciSignature* sig = callee()->signature();
2805 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2806 #endif
2807 null_check_receiver(); // null-check, then ignore
2808 Node *addr = argument(1);
2809 addr = new CastX2PNode(addr);
2810 addr = _gvn.transform(addr);
2811 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2812 flush = _gvn.transform(flush);
2813 set_memory(flush, TypeRawPtr::BOTTOM);
2814 return true;
2815 }
2816
2817 //----------------------------inline_unsafe_writeback0-------------------------
2818 // public native void Unsafe.writeback0(long address)
2819 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2820 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2821 return false;
2822 }
2823 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2824 return false;
2825 }
2826 #ifndef PRODUCT
2827 assert(Matcher::has_match_rule(Op_CacheWB),
2828 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2829 : "found match rule for CacheWBPostSync but not CacheWB"));
2830
2831 #endif
2832 null_check_receiver(); // null-check, then ignore
2833 Node *sync;
2834 if (is_pre) {
2835 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2836 } else {
2837 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2838 }
2839 sync = _gvn.transform(sync);
2840 set_memory(sync, TypeRawPtr::BOTTOM);
2841 return true;
2842 }
2843
2844 //----------------------------inline_unsafe_allocate---------------------------
2845 // public native Object Unsafe.allocateInstance(Class<?> cls);
2846 bool LibraryCallKit::inline_unsafe_allocate() {
2847
2848 #if INCLUDE_JVMTI
2849 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
2850 return false;
2851 }
2852 #endif //INCLUDE_JVMTI
2853
2854 if (callee()->is_static()) return false; // caller must have the capability!
2855
2856 null_check_receiver(); // null-check, then ignore
2857 Node* cls = null_check(argument(1));
2858 if (stopped()) return true;
2859
2860 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
2861 kls = null_check(kls);
2862 if (stopped()) return true; // argument was like int.class
2863
2864 #if INCLUDE_JVMTI
2865 // Don't try to access new allocated obj in the intrinsic.
2866 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
2867 // Deoptimize and allocate in interpreter instead.
2868 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
2869 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
2870 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
2871 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
2872 {
2873 BuildCutout unless(this, tst, PROB_MAX);
2874 uncommon_trap(Deoptimization::Reason_intrinsic,
2875 Deoptimization::Action_make_not_entrant);
2876 }
2877 if (stopped()) {
2878 return true;
2879 }
2880 #endif //INCLUDE_JVMTI
2881
2882 Node* test = nullptr;
2883 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2884 // Note: The argument might still be an illegal value like
2885 // Serializable.class or Object[].class. The runtime will handle it.
2886 // But we must make an explicit check for initialization.
2887 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2888 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2889 // can generate code to load it as unsigned byte.
2890 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
2891 Node* bits = intcon(InstanceKlass::fully_initialized);
2892 test = _gvn.transform(new SubINode(inst, bits));
2893 // The 'test' is non-zero if we need to take a slow path.
2894 }
2895
2896 Node* obj = new_instance(kls, test);
2897 set_result(obj);
2898 return true;
2899 }
2900
2901 //------------------------inline_native_time_funcs--------------
2902 // inline code for System.currentTimeMillis() and System.nanoTime()
2903 // these have the same type and signature
2904 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2905 const TypeFunc* tf = OptoRuntime::void_long_Type();
2906 const TypePtr* no_memory_effects = nullptr;
2907 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2908 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2909 #ifdef ASSERT
2910 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2911 assert(value_top == top(), "second value must be top");
2912 #endif
2913 set_result(value);
2914 return true;
2915 }
2916
2917
2918 #if INCLUDE_JVMTI
2919
2920 // When notifications are disabled then just update the VTMS transition bit and return.
2921 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
2922 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
2923 if (!DoJVMTIVirtualThreadTransitions) {
2924 return true;
2925 }
2926 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
2927 IdealKit ideal(this);
2928
2929 Node* ONE = ideal.ConI(1);
2930 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
2931 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
2932 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
2933
2934 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
2935 sync_kit(ideal);
2936 // if notifyJvmti enabled then make a call to the given SharedRuntime function
2937 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
2938 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
2939 ideal.sync_kit(this);
2940 } ideal.else_(); {
2941 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
2942 Node* thread = ideal.thread();
2943 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
2944 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
2945
2946 sync_kit(ideal);
2947 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2948 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2949
2950 ideal.sync_kit(this);
2951 } ideal.end_if();
2952 final_sync(ideal);
2953
2954 return true;
2955 }
2956
2957 // Always update the is_disable_suspend bit.
2958 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
2959 if (!DoJVMTIVirtualThreadTransitions) {
2960 return true;
2961 }
2962 IdealKit ideal(this);
2963
2964 {
2965 // unconditionally update the is_disable_suspend bit in current JavaThread
2966 Node* thread = ideal.thread();
2967 Node* arg = _gvn.transform(argument(0)); // argument for notification
2968 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
2969 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
2970
2971 sync_kit(ideal);
2972 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2973 ideal.sync_kit(this);
2974 }
2975 final_sync(ideal);
2976
2977 return true;
2978 }
2979
2980 #endif // INCLUDE_JVMTI
2981
2982 #ifdef JFR_HAVE_INTRINSICS
2983
2984 /**
2985 * if oop->klass != null
2986 * // normal class
2987 * epoch = _epoch_state ? 2 : 1
2988 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
2989 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
2990 * }
2991 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
2992 * else
2993 * // primitive class
2994 * if oop->array_klass != null
2995 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
2996 * else
2997 * id = LAST_TYPE_ID + 1 // void class path
2998 * if (!signaled)
2999 * signaled = true
3000 */
3001 bool LibraryCallKit::inline_native_classID() {
3002 Node* cls = argument(0);
3003
3004 IdealKit ideal(this);
3005 #define __ ideal.
3006 IdealVariable result(ideal); __ declarations_done();
3007 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3008 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3009 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3010
3011
3012 __ if_then(kls, BoolTest::ne, null()); {
3013 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3014 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3015
3016 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3017 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3018 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3019 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3020 mask = _gvn.transform(new OrLNode(mask, epoch));
3021 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3022
3023 float unlikely = PROB_UNLIKELY(0.999);
3024 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3025 sync_kit(ideal);
3026 make_runtime_call(RC_LEAF,
3027 OptoRuntime::class_id_load_barrier_Type(),
3028 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3029 "class id load barrier",
3030 TypePtr::BOTTOM,
3031 kls);
3032 ideal.sync_kit(this);
3033 } __ end_if();
3034
3035 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3036 } __ else_(); {
3037 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3038 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3039 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3040 __ if_then(array_kls, BoolTest::ne, null()); {
3041 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3042 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3043 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3044 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3045 } __ else_(); {
3046 // void class case
3047 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3048 } __ end_if();
3049
3050 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3051 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3052 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3053 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3054 } __ end_if();
3055 } __ end_if();
3056
3057 final_sync(ideal);
3058 set_result(ideal.value(result));
3059 #undef __
3060 return true;
3061 }
3062
3063 //------------------------inline_native_jvm_commit------------------
3064 bool LibraryCallKit::inline_native_jvm_commit() {
3065 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3066
3067 // Save input memory and i_o state.
3068 Node* input_memory_state = reset_memory();
3069 set_all_memory(input_memory_state);
3070 Node* input_io_state = i_o();
3071
3072 // TLS.
3073 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3074 // Jfr java buffer.
3075 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3076 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3077 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3078
3079 // Load the current value of the notified field in the JfrThreadLocal.
3080 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3081 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3082
3083 // Test for notification.
3084 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3085 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3086 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3087
3088 // True branch, is notified.
3089 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3090 set_control(is_notified);
3091
3092 // Reset notified state.
3093 Node* notified_reset_memory = store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3094
3095 // 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.
3096 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3097 // Convert the machine-word to a long.
3098 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3099
3100 // False branch, not notified.
3101 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3102 set_control(not_notified);
3103 set_all_memory(input_memory_state);
3104
3105 // Arg is the next position as a long.
3106 Node* arg = argument(0);
3107 // Convert long to machine-word.
3108 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3109
3110 // Store the next_position to the underlying jfr java buffer.
3111 Node* commit_memory;
3112 #ifdef _LP64
3113 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_LONG, MemNode::release);
3114 #else
3115 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_INT, MemNode::release);
3116 #endif
3117
3118 // 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.
3119 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3120 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3121 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3122
3123 // And flags with lease constant.
3124 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3125
3126 // Branch on lease to conditionalize returning the leased java buffer.
3127 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3128 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3129 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3130
3131 // False branch, not a lease.
3132 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3133
3134 // True branch, is lease.
3135 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3136 set_control(is_lease);
3137
3138 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3139 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3140 OptoRuntime::void_void_Type(),
3141 SharedRuntime::jfr_return_lease(),
3142 "return_lease", TypePtr::BOTTOM);
3143 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3144
3145 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3146 record_for_igvn(lease_compare_rgn);
3147 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3148 record_for_igvn(lease_compare_mem);
3149 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3150 record_for_igvn(lease_compare_io);
3151 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3152 record_for_igvn(lease_result_value);
3153
3154 // Update control and phi nodes.
3155 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3156 lease_compare_rgn->init_req(_false_path, not_lease);
3157
3158 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3159 lease_compare_mem->init_req(_false_path, commit_memory);
3160
3161 lease_compare_io->init_req(_true_path, i_o());
3162 lease_compare_io->init_req(_false_path, input_io_state);
3163
3164 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3165 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3166
3167 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3168 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3169 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3170 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3171
3172 // Update control and phi nodes.
3173 result_rgn->init_req(_true_path, is_notified);
3174 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3175
3176 result_mem->init_req(_true_path, notified_reset_memory);
3177 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3178
3179 result_io->init_req(_true_path, input_io_state);
3180 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3181
3182 result_value->init_req(_true_path, current_pos);
3183 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3184
3185 // Set output state.
3186 set_control(_gvn.transform(result_rgn));
3187 set_all_memory(_gvn.transform(result_mem));
3188 set_i_o(_gvn.transform(result_io));
3189 set_result(result_rgn, result_value);
3190 return true;
3191 }
3192
3193 /*
3194 * The intrinsic is a model of this pseudo-code:
3195 *
3196 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3197 * jobject h_event_writer = tl->java_event_writer();
3198 * if (h_event_writer == nullptr) {
3199 * return nullptr;
3200 * }
3201 * oop threadObj = Thread::threadObj();
3202 * oop vthread = java_lang_Thread::vthread(threadObj);
3203 * traceid tid;
3204 * bool pinVirtualThread;
3205 * bool excluded;
3206 * if (vthread != threadObj) { // i.e. current thread is virtual
3207 * tid = java_lang_Thread::tid(vthread);
3208 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3209 * pinVirtualThread = VMContinuations;
3210 * excluded = vthread_epoch_raw & excluded_mask;
3211 * if (!excluded) {
3212 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3213 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3214 * if (vthread_epoch != current_epoch) {
3215 * write_checkpoint();
3216 * }
3217 * }
3218 * } else {
3219 * tid = java_lang_Thread::tid(threadObj);
3220 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3221 * pinVirtualThread = false;
3222 * excluded = thread_epoch_raw & excluded_mask;
3223 * }
3224 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3225 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3226 * if (tid_in_event_writer != tid) {
3227 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3228 * setField(event_writer, "excluded", excluded);
3229 * setField(event_writer, "threadID", tid);
3230 * }
3231 * return event_writer
3232 */
3233 bool LibraryCallKit::inline_native_getEventWriter() {
3234 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3235
3236 // Save input memory and i_o state.
3237 Node* input_memory_state = reset_memory();
3238 set_all_memory(input_memory_state);
3239 Node* input_io_state = i_o();
3240
3241 // The most significant bit of the u2 is used to denote thread exclusion
3242 Node* excluded_shift = _gvn.intcon(15);
3243 Node* excluded_mask = _gvn.intcon(1 << 15);
3244 // The epoch generation is the range [1-32767]
3245 Node* epoch_mask = _gvn.intcon(32767);
3246
3247 // TLS
3248 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3249
3250 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3251 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3252
3253 // Load the eventwriter jobject handle.
3254 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3255
3256 // Null check the jobject handle.
3257 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3258 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3259 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3260
3261 // False path, jobj is null.
3262 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3263
3264 // True path, jobj is not null.
3265 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3266
3267 set_control(jobj_is_not_null);
3268
3269 // Load the threadObj for the CarrierThread.
3270 Node* threadObj = generate_current_thread(tls_ptr);
3271
3272 // Load the vthread.
3273 Node* vthread = generate_virtual_thread(tls_ptr);
3274
3275 // If vthread != threadObj, this is a virtual thread.
3276 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3277 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3278 IfNode* iff_vthread_not_equal_threadObj =
3279 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3280
3281 // False branch, fallback to threadObj.
3282 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3283 set_control(vthread_equal_threadObj);
3284
3285 // Load the tid field from the vthread object.
3286 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3287
3288 // Load the raw epoch value from the threadObj.
3289 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3290 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3291 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3292 TypeInt::CHAR, T_CHAR,
3293 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3294
3295 // Mask off the excluded information from the epoch.
3296 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3297
3298 // True branch, this is a virtual thread.
3299 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3300 set_control(vthread_not_equal_threadObj);
3301
3302 // Load the tid field from the vthread object.
3303 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3304
3305 // Continuation support determines if a virtual thread should be pinned.
3306 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3307 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3308
3309 // Load the raw epoch value from the vthread.
3310 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3311 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3312 TypeInt::CHAR, T_CHAR,
3313 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3314
3315 // Mask off the excluded information from the epoch.
3316 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3317
3318 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3319 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3320 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3321 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3322
3323 // False branch, vthread is excluded, no need to write epoch info.
3324 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3325
3326 // True branch, vthread is included, update epoch info.
3327 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3328 set_control(included);
3329
3330 // Get epoch value.
3331 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3332
3333 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3334 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3335 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3336
3337 // Compare the epoch in the vthread to the current epoch generation.
3338 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3339 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3340 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3341
3342 // False path, epoch is equal, checkpoint information is valid.
3343 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3344
3345 // True path, epoch is not equal, write a checkpoint for the vthread.
3346 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3347
3348 set_control(epoch_is_not_equal);
3349
3350 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3351 // The call also updates the native thread local thread id and the vthread with the current epoch.
3352 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3353 OptoRuntime::jfr_write_checkpoint_Type(),
3354 SharedRuntime::jfr_write_checkpoint(),
3355 "write_checkpoint", TypePtr::BOTTOM);
3356 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3357
3358 // vthread epoch != current epoch
3359 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3360 record_for_igvn(epoch_compare_rgn);
3361 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3362 record_for_igvn(epoch_compare_mem);
3363 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3364 record_for_igvn(epoch_compare_io);
3365
3366 // Update control and phi nodes.
3367 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3368 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3369 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3370 epoch_compare_mem->init_req(_false_path, input_memory_state);
3371 epoch_compare_io->init_req(_true_path, i_o());
3372 epoch_compare_io->init_req(_false_path, input_io_state);
3373
3374 // excluded != true
3375 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3376 record_for_igvn(exclude_compare_rgn);
3377 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3378 record_for_igvn(exclude_compare_mem);
3379 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3380 record_for_igvn(exclude_compare_io);
3381
3382 // Update control and phi nodes.
3383 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3384 exclude_compare_rgn->init_req(_false_path, excluded);
3385 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3386 exclude_compare_mem->init_req(_false_path, input_memory_state);
3387 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3388 exclude_compare_io->init_req(_false_path, input_io_state);
3389
3390 // vthread != threadObj
3391 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3392 record_for_igvn(vthread_compare_rgn);
3393 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3394 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3395 record_for_igvn(vthread_compare_io);
3396 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3397 record_for_igvn(tid);
3398 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3399 record_for_igvn(exclusion);
3400 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3401 record_for_igvn(pinVirtualThread);
3402
3403 // Update control and phi nodes.
3404 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3405 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3406 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3407 vthread_compare_mem->init_req(_false_path, input_memory_state);
3408 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3409 vthread_compare_io->init_req(_false_path, input_io_state);
3410 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3411 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3412 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3413 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3414 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3415 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3416
3417 // Update branch state.
3418 set_control(_gvn.transform(vthread_compare_rgn));
3419 set_all_memory(_gvn.transform(vthread_compare_mem));
3420 set_i_o(_gvn.transform(vthread_compare_io));
3421
3422 // Load the event writer oop by dereferencing the jobject handle.
3423 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3424 assert(klass_EventWriter->is_loaded(), "invariant");
3425 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3426 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3427 const TypeOopPtr* const xtype = aklass->as_instance_type();
3428 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3429 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3430
3431 // Load the current thread id from the event writer object.
3432 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3433 // Get the field offset to, conditionally, store an updated tid value later.
3434 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3435 // Get the field offset to, conditionally, store an updated exclusion value later.
3436 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3437 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3438 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3439
3440 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3441 record_for_igvn(event_writer_tid_compare_rgn);
3442 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3443 record_for_igvn(event_writer_tid_compare_mem);
3444 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3445 record_for_igvn(event_writer_tid_compare_io);
3446
3447 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3448 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3449 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3450 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3451
3452 // False path, tids are the same.
3453 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3454
3455 // True path, tid is not equal, need to update the tid in the event writer.
3456 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3457 record_for_igvn(tid_is_not_equal);
3458
3459 // Store the pin state to the event writer.
3460 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3461
3462 // Store the exclusion state to the event writer.
3463 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3464 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3465
3466 // Store the tid to the event writer.
3467 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3468
3469 // Update control and phi nodes.
3470 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3471 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3472 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3473 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3474 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3475 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3476
3477 // Result of top level CFG, Memory, IO and Value.
3478 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3479 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3480 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3481 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3482
3483 // Result control.
3484 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3485 result_rgn->init_req(_false_path, jobj_is_null);
3486
3487 // Result memory.
3488 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3489 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3490
3491 // Result IO.
3492 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3493 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3494
3495 // Result value.
3496 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3497 result_value->init_req(_false_path, null()); // return null
3498
3499 // Set output state.
3500 set_control(_gvn.transform(result_rgn));
3501 set_all_memory(_gvn.transform(result_mem));
3502 set_i_o(_gvn.transform(result_io));
3503 set_result(result_rgn, result_value);
3504 return true;
3505 }
3506
3507 /*
3508 * The intrinsic is a model of this pseudo-code:
3509 *
3510 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3511 * if (carrierThread != thread) { // is virtual thread
3512 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3513 * bool excluded = vthread_epoch_raw & excluded_mask;
3514 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3515 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3516 * if (!excluded) {
3517 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3518 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3519 * }
3520 * Atomic::release_store(&tl->_vthread, true);
3521 * return;
3522 * }
3523 * Atomic::release_store(&tl->_vthread, false);
3524 */
3525 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3526 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3527
3528 Node* input_memory_state = reset_memory();
3529 set_all_memory(input_memory_state);
3530
3531 // The most significant bit of the u2 is used to denote thread exclusion
3532 Node* excluded_mask = _gvn.intcon(1 << 15);
3533 // The epoch generation is the range [1-32767]
3534 Node* epoch_mask = _gvn.intcon(32767);
3535
3536 Node* const carrierThread = generate_current_thread(jt);
3537 // If thread != carrierThread, this is a virtual thread.
3538 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3539 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3540 IfNode* iff_thread_not_equal_carrierThread =
3541 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3542
3543 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3544
3545 // False branch, is carrierThread.
3546 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3547 // Store release
3548 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3549
3550 set_all_memory(input_memory_state);
3551
3552 // True branch, is virtual thread.
3553 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3554 set_control(thread_not_equal_carrierThread);
3555
3556 // Load the raw epoch value from the vthread.
3557 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3558 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
3559 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3560
3561 // Mask off the excluded information from the epoch.
3562 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3563
3564 // Load the tid field from the thread.
3565 Node* tid = load_field_from_object(thread, "tid", "J");
3566
3567 // Store the vthread tid to the jfr thread local.
3568 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3569 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
3570
3571 // Branch is_excluded to conditionalize updating the epoch .
3572 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3573 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3574 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3575
3576 // True branch, vthread is excluded, no need to write epoch info.
3577 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3578 set_control(excluded);
3579 Node* vthread_is_excluded = _gvn.intcon(1);
3580
3581 // False branch, vthread is included, update epoch info.
3582 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3583 set_control(included);
3584 Node* vthread_is_included = _gvn.intcon(0);
3585
3586 // Get epoch value.
3587 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3588
3589 // Store the vthread epoch to the jfr thread local.
3590 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3591 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
3592
3593 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3594 record_for_igvn(excluded_rgn);
3595 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3596 record_for_igvn(excluded_mem);
3597 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3598 record_for_igvn(exclusion);
3599
3600 // Merge the excluded control and memory.
3601 excluded_rgn->init_req(_true_path, excluded);
3602 excluded_rgn->init_req(_false_path, included);
3603 excluded_mem->init_req(_true_path, tid_memory);
3604 excluded_mem->init_req(_false_path, included_memory);
3605 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3606 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3607
3608 // Set intermediate state.
3609 set_control(_gvn.transform(excluded_rgn));
3610 set_all_memory(excluded_mem);
3611
3612 // Store the vthread exclusion state to the jfr thread local.
3613 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3614 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
3615
3616 // Store release
3617 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
3618
3619 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3620 record_for_igvn(thread_compare_rgn);
3621 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3622 record_for_igvn(thread_compare_mem);
3623 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3624 record_for_igvn(vthread);
3625
3626 // Merge the thread_compare control and memory.
3627 thread_compare_rgn->init_req(_true_path, control());
3628 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3629 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3630 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3631
3632 // Set output state.
3633 set_control(_gvn.transform(thread_compare_rgn));
3634 set_all_memory(_gvn.transform(thread_compare_mem));
3635 }
3636
3637 #endif // JFR_HAVE_INTRINSICS
3638
3639 //------------------------inline_native_currentCarrierThread------------------
3640 bool LibraryCallKit::inline_native_currentCarrierThread() {
3641 Node* junk = nullptr;
3642 set_result(generate_current_thread(junk));
3643 return true;
3644 }
3645
3646 //------------------------inline_native_currentThread------------------
3647 bool LibraryCallKit::inline_native_currentThread() {
3648 Node* junk = nullptr;
3649 set_result(generate_virtual_thread(junk));
3650 return true;
3651 }
3652
3653 //------------------------inline_native_setVthread------------------
3654 bool LibraryCallKit::inline_native_setCurrentThread() {
3655 assert(C->method()->changes_current_thread(),
3656 "method changes current Thread but is not annotated ChangesCurrentThread");
3657 Node* arr = argument(1);
3658 Node* thread = _gvn.transform(new ThreadLocalNode());
3659 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3660 Node* thread_obj_handle
3661 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3662 thread_obj_handle = _gvn.transform(thread_obj_handle);
3663 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3664 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3665
3666 // Change the _monitor_owner_id of the JavaThread
3667 Node* tid = load_field_from_object(arr, "tid", "J");
3668 Node* monitor_owner_id_offset = basic_plus_adr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
3669 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
3670
3671 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3672 return true;
3673 }
3674
3675 const Type* LibraryCallKit::scopedValueCache_type() {
3676 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3677 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3678 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3679
3680 // Because we create the scopedValue cache lazily we have to make the
3681 // type of the result BotPTR.
3682 bool xk = etype->klass_is_exact();
3683 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3684 return objects_type;
3685 }
3686
3687 Node* LibraryCallKit::scopedValueCache_helper() {
3688 Node* thread = _gvn.transform(new ThreadLocalNode());
3689 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3690 // We cannot use immutable_memory() because we might flip onto a
3691 // different carrier thread, at which point we'll need to use that
3692 // carrier thread's cache.
3693 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3694 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3695 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3696 }
3697
3698 //------------------------inline_native_scopedValueCache------------------
3699 bool LibraryCallKit::inline_native_scopedValueCache() {
3700 Node* cache_obj_handle = scopedValueCache_helper();
3701 const Type* objects_type = scopedValueCache_type();
3702 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3703
3704 return true;
3705 }
3706
3707 //------------------------inline_native_setScopedValueCache------------------
3708 bool LibraryCallKit::inline_native_setScopedValueCache() {
3709 Node* arr = argument(0);
3710 Node* cache_obj_handle = scopedValueCache_helper();
3711 const Type* objects_type = scopedValueCache_type();
3712
3713 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3714 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3715
3716 return true;
3717 }
3718
3719 //------------------------inline_native_Continuation_pin and unpin-----------
3720
3721 // Shared implementation routine for both pin and unpin.
3722 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
3723 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3724
3725 // Save input memory.
3726 Node* input_memory_state = reset_memory();
3727 set_all_memory(input_memory_state);
3728
3729 // TLS
3730 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3731 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
3732 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
3733
3734 // Null check the last continuation object.
3735 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
3736 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
3737 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3738
3739 // False path, last continuation is null.
3740 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
3741
3742 // True path, last continuation is not null.
3743 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
3744
3745 set_control(continuation_is_not_null);
3746
3747 // Load the pin count from the last continuation.
3748 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
3749 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
3750
3751 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
3752 Node* pin_count_rhs;
3753 if (unpin) {
3754 pin_count_rhs = _gvn.intcon(0);
3755 } else {
3756 pin_count_rhs = _gvn.intcon(UINT32_MAX);
3757 }
3758 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
3759 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
3760 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
3761
3762 // True branch, pin count over/underflow.
3763 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
3764 {
3765 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
3766 // which will throw IllegalStateException for pin count over/underflow.
3767 // No memory changed so far - we can use memory create by reset_memory()
3768 // at the beginning of this intrinsic. No need to call reset_memory() again.
3769 PreserveJVMState pjvms(this);
3770 set_control(pin_count_over_underflow);
3771 uncommon_trap(Deoptimization::Reason_intrinsic,
3772 Deoptimization::Action_none);
3773 assert(stopped(), "invariant");
3774 }
3775
3776 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
3777 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
3778 set_control(valid_pin_count);
3779
3780 Node* next_pin_count;
3781 if (unpin) {
3782 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
3783 } else {
3784 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
3785 }
3786
3787 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
3788
3789 // Result of top level CFG and Memory.
3790 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3791 record_for_igvn(result_rgn);
3792 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3793 record_for_igvn(result_mem);
3794
3795 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
3796 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
3797 result_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3798 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3799
3800 // Set output state.
3801 set_control(_gvn.transform(result_rgn));
3802 set_all_memory(_gvn.transform(result_mem));
3803
3804 return true;
3805 }
3806
3807 //---------------------------load_mirror_from_klass----------------------------
3808 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3809 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3810 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3811 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3812 // mirror = ((OopHandle)mirror)->resolve();
3813 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3814 }
3815
3816 //-----------------------load_klass_from_mirror_common-------------------------
3817 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3818 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3819 // and branch to the given path on the region.
3820 // If never_see_null, take an uncommon trap on null, so we can optimistically
3821 // compile for the non-null case.
3822 // If the region is null, force never_see_null = true.
3823 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3824 bool never_see_null,
3825 RegionNode* region,
3826 int null_path,
3827 int offset) {
3828 if (region == nullptr) never_see_null = true;
3829 Node* p = basic_plus_adr(mirror, offset);
3830 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3831 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3832 Node* null_ctl = top();
3833 kls = null_check_oop(kls, &null_ctl, never_see_null);
3834 if (region != nullptr) {
3835 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3836 region->init_req(null_path, null_ctl);
3837 } else {
3838 assert(null_ctl == top(), "no loose ends");
3839 }
3840 return kls;
3841 }
3842
3843 //--------------------(inline_native_Class_query helpers)---------------------
3844 // Use this for JVM_ACC_INTERFACE.
3845 // Fall through if (mods & mask) == bits, take the guard otherwise.
3846 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
3847 ByteSize offset, const Type* type, BasicType bt) {
3848 // Branch around if the given klass has the given modifier bit set.
3849 // Like generate_guard, adds a new path onto the region.
3850 Node* modp = basic_plus_adr(kls, in_bytes(offset));
3851 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
3852 Node* mask = intcon(modifier_mask);
3853 Node* bits = intcon(modifier_bits);
3854 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3855 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3856 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3857 return generate_fair_guard(bol, region);
3858 }
3859 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3860 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
3861 Klass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
3862 }
3863
3864 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
3865 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3866 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
3867 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
3868 }
3869
3870 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3871 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
3872 }
3873
3874 //-------------------------inline_native_Class_query-------------------
3875 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3876 const Type* return_type = TypeInt::BOOL;
3877 Node* prim_return_value = top(); // what happens if it's a primitive class?
3878 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3879 bool expect_prim = false; // most of these guys expect to work on refs
3880
3881 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3882
3883 Node* mirror = argument(0);
3884 Node* obj = top();
3885
3886 switch (id) {
3887 case vmIntrinsics::_isInstance:
3888 // nothing is an instance of a primitive type
3889 prim_return_value = intcon(0);
3890 obj = argument(1);
3891 break;
3892 case vmIntrinsics::_isHidden:
3893 prim_return_value = intcon(0);
3894 break;
3895 case vmIntrinsics::_getSuperclass:
3896 prim_return_value = null();
3897 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3898 break;
3899 case vmIntrinsics::_getClassAccessFlags:
3900 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3901 return_type = TypeInt::CHAR;
3902 break;
3903 default:
3904 fatal_unexpected_iid(id);
3905 break;
3906 }
3907
3908 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3909 if (mirror_con == nullptr) return false; // cannot happen?
3910
3911 #ifndef PRODUCT
3912 if (C->print_intrinsics() || C->print_inlining()) {
3913 ciType* k = mirror_con->java_mirror_type();
3914 if (k) {
3915 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3916 k->print_name();
3917 tty->cr();
3918 }
3919 }
3920 #endif
3921
3922 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3923 RegionNode* region = new RegionNode(PATH_LIMIT);
3924 record_for_igvn(region);
3925 PhiNode* phi = new PhiNode(region, return_type);
3926
3927 // The mirror will never be null of Reflection.getClassAccessFlags, however
3928 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3929 // if it is. See bug 4774291.
3930
3931 // For Reflection.getClassAccessFlags(), the null check occurs in
3932 // the wrong place; see inline_unsafe_access(), above, for a similar
3933 // situation.
3934 mirror = null_check(mirror);
3935 // If mirror or obj is dead, only null-path is taken.
3936 if (stopped()) return true;
3937
3938 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3939
3940 // Now load the mirror's klass metaobject, and null-check it.
3941 // Side-effects region with the control path if the klass is null.
3942 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3943 // If kls is null, we have a primitive mirror.
3944 phi->init_req(_prim_path, prim_return_value);
3945 if (stopped()) { set_result(region, phi); return true; }
3946 bool safe_for_replace = (region->in(_prim_path) == top());
3947
3948 Node* p; // handy temp
3949 Node* null_ctl;
3950
3951 // Now that we have the non-null klass, we can perform the real query.
3952 // For constant classes, the query will constant-fold in LoadNode::Value.
3953 Node* query_value = top();
3954 switch (id) {
3955 case vmIntrinsics::_isInstance:
3956 // nothing is an instance of a primitive type
3957 query_value = gen_instanceof(obj, kls, safe_for_replace);
3958 break;
3959
3960 case vmIntrinsics::_isHidden:
3961 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
3962 if (generate_hidden_class_guard(kls, region) != nullptr)
3963 // A guard was added. If the guard is taken, it was an hidden class.
3964 phi->add_req(intcon(1));
3965 // If we fall through, it's a plain class.
3966 query_value = intcon(0);
3967 break;
3968
3969
3970 case vmIntrinsics::_getSuperclass:
3971 // The rules here are somewhat unfortunate, but we can still do better
3972 // with random logic than with a JNI call.
3973 // Interfaces store null or Object as _super, but must report null.
3974 // Arrays store an intermediate super as _super, but must report Object.
3975 // Other types can report the actual _super.
3976 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3977 if (generate_interface_guard(kls, region) != nullptr)
3978 // A guard was added. If the guard is taken, it was an interface.
3979 phi->add_req(null());
3980 if (generate_array_guard(kls, region) != nullptr)
3981 // A guard was added. If the guard is taken, it was an array.
3982 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3983 // If we fall through, it's a plain class. Get its _super.
3984 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3985 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3986 null_ctl = top();
3987 kls = null_check_oop(kls, &null_ctl);
3988 if (null_ctl != top()) {
3989 // If the guard is taken, Object.superClass is null (both klass and mirror).
3990 region->add_req(null_ctl);
3991 phi ->add_req(null());
3992 }
3993 if (!stopped()) {
3994 query_value = load_mirror_from_klass(kls);
3995 }
3996 break;
3997
3998 case vmIntrinsics::_getClassAccessFlags:
3999 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
4000 query_value = make_load(nullptr, p, TypeInt::CHAR, T_CHAR, MemNode::unordered);
4001 break;
4002
4003 default:
4004 fatal_unexpected_iid(id);
4005 break;
4006 }
4007
4008 // Fall-through is the normal case of a query to a real class.
4009 phi->init_req(1, query_value);
4010 region->init_req(1, control());
4011
4012 C->set_has_split_ifs(true); // Has chance for split-if optimization
4013 set_result(region, phi);
4014 return true;
4015 }
4016
4017 //-------------------------inline_Class_cast-------------------
4018 bool LibraryCallKit::inline_Class_cast() {
4019 Node* mirror = argument(0); // Class
4020 Node* obj = argument(1);
4021 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4022 if (mirror_con == nullptr) {
4023 return false; // dead path (mirror->is_top()).
4024 }
4025 if (obj == nullptr || obj->is_top()) {
4026 return false; // dead path
4027 }
4028 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4029
4030 // First, see if Class.cast() can be folded statically.
4031 // java_mirror_type() returns non-null for compile-time Class constants.
4032 ciType* tm = mirror_con->java_mirror_type();
4033 if (tm != nullptr && tm->is_klass() &&
4034 tp != nullptr) {
4035 if (!tp->is_loaded()) {
4036 // Don't use intrinsic when class is not loaded.
4037 return false;
4038 } else {
4039 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
4040 if (static_res == Compile::SSC_always_true) {
4041 // isInstance() is true - fold the code.
4042 set_result(obj);
4043 return true;
4044 } else if (static_res == Compile::SSC_always_false) {
4045 // Don't use intrinsic, have to throw ClassCastException.
4046 // If the reference is null, the non-intrinsic bytecode will
4047 // be optimized appropriately.
4048 return false;
4049 }
4050 }
4051 }
4052
4053 // Bailout intrinsic and do normal inlining if exception path is frequent.
4054 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4055 return false;
4056 }
4057
4058 // Generate dynamic checks.
4059 // Class.cast() is java implementation of _checkcast bytecode.
4060 // Do checkcast (Parse::do_checkcast()) optimizations here.
4061
4062 mirror = null_check(mirror);
4063 // If mirror is dead, only null-path is taken.
4064 if (stopped()) {
4065 return true;
4066 }
4067
4068 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive).
4069 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
4070 RegionNode* region = new RegionNode(PATH_LIMIT);
4071 record_for_igvn(region);
4072
4073 // Now load the mirror's klass metaobject, and null-check it.
4074 // If kls is null, we have a primitive mirror and
4075 // nothing is an instance of a primitive type.
4076 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4077
4078 Node* res = top();
4079 if (!stopped()) {
4080 Node* bad_type_ctrl = top();
4081 // Do checkcast optimizations.
4082 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4083 region->init_req(_bad_type_path, bad_type_ctrl);
4084 }
4085 if (region->in(_prim_path) != top() ||
4086 region->in(_bad_type_path) != top()) {
4087 // Let Interpreter throw ClassCastException.
4088 PreserveJVMState pjvms(this);
4089 set_control(_gvn.transform(region));
4090 uncommon_trap(Deoptimization::Reason_intrinsic,
4091 Deoptimization::Action_maybe_recompile);
4092 }
4093 if (!stopped()) {
4094 set_result(res);
4095 }
4096 return true;
4097 }
4098
4099
4100 //--------------------------inline_native_subtype_check------------------------
4101 // This intrinsic takes the JNI calls out of the heart of
4102 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4103 bool LibraryCallKit::inline_native_subtype_check() {
4104 // Pull both arguments off the stack.
4105 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4106 args[0] = argument(0);
4107 args[1] = argument(1);
4108 Node* klasses[2]; // corresponding Klasses: superk, subk
4109 klasses[0] = klasses[1] = top();
4110
4111 enum {
4112 // A full decision tree on {superc is prim, subc is prim}:
4113 _prim_0_path = 1, // {P,N} => false
4114 // {P,P} & superc!=subc => false
4115 _prim_same_path, // {P,P} & superc==subc => true
4116 _prim_1_path, // {N,P} => false
4117 _ref_subtype_path, // {N,N} & subtype check wins => true
4118 _both_ref_path, // {N,N} & subtype check loses => false
4119 PATH_LIMIT
4120 };
4121
4122 RegionNode* region = new RegionNode(PATH_LIMIT);
4123 Node* phi = new PhiNode(region, TypeInt::BOOL);
4124 record_for_igvn(region);
4125
4126 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4127 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4128 int class_klass_offset = java_lang_Class::klass_offset();
4129
4130 // First null-check both mirrors and load each mirror's klass metaobject.
4131 int which_arg;
4132 for (which_arg = 0; which_arg <= 1; which_arg++) {
4133 Node* arg = args[which_arg];
4134 arg = null_check(arg);
4135 if (stopped()) break;
4136 args[which_arg] = arg;
4137
4138 Node* p = basic_plus_adr(arg, class_klass_offset);
4139 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4140 klasses[which_arg] = _gvn.transform(kls);
4141 }
4142
4143 // Having loaded both klasses, test each for null.
4144 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4145 for (which_arg = 0; which_arg <= 1; which_arg++) {
4146 Node* kls = klasses[which_arg];
4147 Node* null_ctl = top();
4148 kls = null_check_oop(kls, &null_ctl, never_see_null);
4149 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
4150 region->init_req(prim_path, null_ctl);
4151 if (stopped()) break;
4152 klasses[which_arg] = kls;
4153 }
4154
4155 if (!stopped()) {
4156 // now we have two reference types, in klasses[0..1]
4157 Node* subk = klasses[1]; // the argument to isAssignableFrom
4158 Node* superk = klasses[0]; // the receiver
4159 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4160 // now we have a successful reference subtype check
4161 region->set_req(_ref_subtype_path, control());
4162 }
4163
4164 // If both operands are primitive (both klasses null), then
4165 // we must return true when they are identical primitives.
4166 // It is convenient to test this after the first null klass check.
4167 set_control(region->in(_prim_0_path)); // go back to first null check
4168 if (!stopped()) {
4169 // Since superc is primitive, make a guard for the superc==subc case.
4170 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4171 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4172 generate_guard(bol_eq, region, PROB_FAIR);
4173 if (region->req() == PATH_LIMIT+1) {
4174 // A guard was added. If the added guard is taken, superc==subc.
4175 region->swap_edges(PATH_LIMIT, _prim_same_path);
4176 region->del_req(PATH_LIMIT);
4177 }
4178 region->set_req(_prim_0_path, control()); // Not equal after all.
4179 }
4180
4181 // these are the only paths that produce 'true':
4182 phi->set_req(_prim_same_path, intcon(1));
4183 phi->set_req(_ref_subtype_path, intcon(1));
4184
4185 // pull together the cases:
4186 assert(region->req() == PATH_LIMIT, "sane region");
4187 for (uint i = 1; i < region->req(); i++) {
4188 Node* ctl = region->in(i);
4189 if (ctl == nullptr || ctl == top()) {
4190 region->set_req(i, top());
4191 phi ->set_req(i, top());
4192 } else if (phi->in(i) == nullptr) {
4193 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4194 }
4195 }
4196
4197 set_control(_gvn.transform(region));
4198 set_result(_gvn.transform(phi));
4199 return true;
4200 }
4201
4202 //---------------------generate_array_guard_common------------------------
4203 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
4204 bool obj_array, bool not_array, Node** obj) {
4205
4206 if (stopped()) {
4207 return nullptr;
4208 }
4209
4210 // If obj_array/non_array==false/false:
4211 // Branch around if the given klass is in fact an array (either obj or prim).
4212 // If obj_array/non_array==false/true:
4213 // Branch around if the given klass is not an array klass of any kind.
4214 // If obj_array/non_array==true/true:
4215 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
4216 // If obj_array/non_array==true/false:
4217 // Branch around if the kls is an oop array (Object[] or subtype)
4218 //
4219 // Like generate_guard, adds a new path onto the region.
4220 jint layout_con = 0;
4221 Node* layout_val = get_layout_helper(kls, layout_con);
4222 if (layout_val == nullptr) {
4223 bool query = (obj_array
4224 ? Klass::layout_helper_is_objArray(layout_con)
4225 : Klass::layout_helper_is_array(layout_con));
4226 if (query == not_array) {
4227 return nullptr; // never a branch
4228 } else { // always a branch
4229 Node* always_branch = control();
4230 if (region != nullptr)
4231 region->add_req(always_branch);
4232 set_control(top());
4233 return always_branch;
4234 }
4235 }
4236 // Now test the correct condition.
4237 jint nval = (obj_array
4238 ? (jint)(Klass::_lh_array_tag_type_value
4239 << Klass::_lh_array_tag_shift)
4240 : Klass::_lh_neutral_value);
4241 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4242 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
4243 // invert the test if we are looking for a non-array
4244 if (not_array) btest = BoolTest(btest).negate();
4245 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4246 Node* ctrl = generate_fair_guard(bol, region);
4247 Node* is_array_ctrl = not_array ? control() : ctrl;
4248 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4249 // Keep track of the fact that 'obj' is an array to prevent
4250 // array specific accesses from floating above the guard.
4251 Node* cast = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4252 // Check for top because in rare cases, the type system can determine that
4253 // the object can't be an array but the layout helper check is not folded.
4254 if (!cast->is_top()) {
4255 *obj = cast;
4256 }
4257 }
4258 return ctrl;
4259 }
4260
4261
4262 //-----------------------inline_native_newArray--------------------------
4263 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4264 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4265 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4266 Node* mirror;
4267 Node* count_val;
4268 if (uninitialized) {
4269 null_check_receiver();
4270 mirror = argument(1);
4271 count_val = argument(2);
4272 } else {
4273 mirror = argument(0);
4274 count_val = argument(1);
4275 }
4276
4277 mirror = null_check(mirror);
4278 // If mirror or obj is dead, only null-path is taken.
4279 if (stopped()) return true;
4280
4281 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4282 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4283 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4284 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4285 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4286
4287 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4288 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4289 result_reg, _slow_path);
4290 Node* normal_ctl = control();
4291 Node* no_array_ctl = result_reg->in(_slow_path);
4292
4293 // Generate code for the slow case. We make a call to newArray().
4294 set_control(no_array_ctl);
4295 if (!stopped()) {
4296 // Either the input type is void.class, or else the
4297 // array klass has not yet been cached. Either the
4298 // ensuing call will throw an exception, or else it
4299 // will cache the array klass for next time.
4300 PreserveJVMState pjvms(this);
4301 CallJavaNode* slow_call = nullptr;
4302 if (uninitialized) {
4303 // Generate optimized virtual call (holder class 'Unsafe' is final)
4304 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4305 } else {
4306 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4307 }
4308 Node* slow_result = set_results_for_java_call(slow_call);
4309 // this->control() comes from set_results_for_java_call
4310 result_reg->set_req(_slow_path, control());
4311 result_val->set_req(_slow_path, slow_result);
4312 result_io ->set_req(_slow_path, i_o());
4313 result_mem->set_req(_slow_path, reset_memory());
4314 }
4315
4316 set_control(normal_ctl);
4317 if (!stopped()) {
4318 // Normal case: The array type has been cached in the java.lang.Class.
4319 // The following call works fine even if the array type is polymorphic.
4320 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4321 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4322 result_reg->init_req(_normal_path, control());
4323 result_val->init_req(_normal_path, obj);
4324 result_io ->init_req(_normal_path, i_o());
4325 result_mem->init_req(_normal_path, reset_memory());
4326
4327 if (uninitialized) {
4328 // Mark the allocation so that zeroing is skipped
4329 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4330 alloc->maybe_set_complete(&_gvn);
4331 }
4332 }
4333
4334 // Return the combined state.
4335 set_i_o( _gvn.transform(result_io) );
4336 set_all_memory( _gvn.transform(result_mem));
4337
4338 C->set_has_split_ifs(true); // Has chance for split-if optimization
4339 set_result(result_reg, result_val);
4340 return true;
4341 }
4342
4343 //----------------------inline_native_getLength--------------------------
4344 // public static native int java.lang.reflect.Array.getLength(Object array);
4345 bool LibraryCallKit::inline_native_getLength() {
4346 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4347
4348 Node* array = null_check(argument(0));
4349 // If array is dead, only null-path is taken.
4350 if (stopped()) return true;
4351
4352 // Deoptimize if it is a non-array.
4353 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
4354
4355 if (non_array != nullptr) {
4356 PreserveJVMState pjvms(this);
4357 set_control(non_array);
4358 uncommon_trap(Deoptimization::Reason_intrinsic,
4359 Deoptimization::Action_maybe_recompile);
4360 }
4361
4362 // If control is dead, only non-array-path is taken.
4363 if (stopped()) return true;
4364
4365 // The works fine even if the array type is polymorphic.
4366 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4367 Node* result = load_array_length(array);
4368
4369 C->set_has_split_ifs(true); // Has chance for split-if optimization
4370 set_result(result);
4371 return true;
4372 }
4373
4374 //------------------------inline_array_copyOf----------------------------
4375 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4376 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4377 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4378 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4379
4380 // Get the arguments.
4381 Node* original = argument(0);
4382 Node* start = is_copyOfRange? argument(1): intcon(0);
4383 Node* end = is_copyOfRange? argument(2): argument(1);
4384 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4385
4386 Node* newcopy = nullptr;
4387
4388 // Set the original stack and the reexecute bit for the interpreter to reexecute
4389 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4390 { PreserveReexecuteState preexecs(this);
4391 jvms()->set_should_reexecute(true);
4392
4393 array_type_mirror = null_check(array_type_mirror);
4394 original = null_check(original);
4395
4396 // Check if a null path was taken unconditionally.
4397 if (stopped()) return true;
4398
4399 Node* orig_length = load_array_length(original);
4400
4401 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4402 klass_node = null_check(klass_node);
4403
4404 RegionNode* bailout = new RegionNode(1);
4405 record_for_igvn(bailout);
4406
4407 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4408 // Bail out if that is so.
4409 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4410 if (not_objArray != nullptr) {
4411 // Improve the klass node's type from the new optimistic assumption:
4412 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4413 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4414 Node* cast = new CastPPNode(control(), klass_node, akls);
4415 klass_node = _gvn.transform(cast);
4416 }
4417
4418 // Bail out if either start or end is negative.
4419 generate_negative_guard(start, bailout, &start);
4420 generate_negative_guard(end, bailout, &end);
4421
4422 Node* length = end;
4423 if (_gvn.type(start) != TypeInt::ZERO) {
4424 length = _gvn.transform(new SubINode(end, start));
4425 }
4426
4427 // Bail out if length is negative (i.e., if start > end).
4428 // Without this the new_array would throw
4429 // NegativeArraySizeException but IllegalArgumentException is what
4430 // should be thrown
4431 generate_negative_guard(length, bailout, &length);
4432
4433 // Bail out if start is larger than the original length
4434 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4435 generate_negative_guard(orig_tail, bailout, &orig_tail);
4436
4437 if (bailout->req() > 1) {
4438 PreserveJVMState pjvms(this);
4439 set_control(_gvn.transform(bailout));
4440 uncommon_trap(Deoptimization::Reason_intrinsic,
4441 Deoptimization::Action_maybe_recompile);
4442 }
4443
4444 if (!stopped()) {
4445 // How many elements will we copy from the original?
4446 // The answer is MinI(orig_tail, length).
4447 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4448
4449 // Generate a direct call to the right arraycopy function(s).
4450 // We know the copy is disjoint but we might not know if the
4451 // oop stores need checking.
4452 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4453 // This will fail a store-check if x contains any non-nulls.
4454
4455 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4456 // loads/stores but it is legal only if we're sure the
4457 // Arrays.copyOf would succeed. So we need all input arguments
4458 // to the copyOf to be validated, including that the copy to the
4459 // new array won't trigger an ArrayStoreException. That subtype
4460 // check can be optimized if we know something on the type of
4461 // the input array from type speculation.
4462 if (_gvn.type(klass_node)->singleton()) {
4463 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4464 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4465
4466 int test = C->static_subtype_check(superk, subk);
4467 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4468 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4469 if (t_original->speculative_type() != nullptr) {
4470 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4471 }
4472 }
4473 }
4474
4475 bool validated = false;
4476 // Reason_class_check rather than Reason_intrinsic because we
4477 // want to intrinsify even if this traps.
4478 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4479 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4480
4481 if (not_subtype_ctrl != top()) {
4482 PreserveJVMState pjvms(this);
4483 set_control(not_subtype_ctrl);
4484 uncommon_trap(Deoptimization::Reason_class_check,
4485 Deoptimization::Action_make_not_entrant);
4486 assert(stopped(), "Should be stopped");
4487 }
4488 validated = true;
4489 }
4490
4491 if (!stopped()) {
4492 newcopy = new_array(klass_node, length, 0); // no arguments to push
4493
4494 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4495 load_object_klass(original), klass_node);
4496 if (!is_copyOfRange) {
4497 ac->set_copyof(validated);
4498 } else {
4499 ac->set_copyofrange(validated);
4500 }
4501 Node* n = _gvn.transform(ac);
4502 if (n == ac) {
4503 ac->connect_outputs(this);
4504 } else {
4505 assert(validated, "shouldn't transform if all arguments not validated");
4506 set_all_memory(n);
4507 }
4508 }
4509 }
4510 } // original reexecute is set back here
4511
4512 C->set_has_split_ifs(true); // Has chance for split-if optimization
4513 if (!stopped()) {
4514 set_result(newcopy);
4515 }
4516 return true;
4517 }
4518
4519
4520 //----------------------generate_virtual_guard---------------------------
4521 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4522 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4523 RegionNode* slow_region) {
4524 ciMethod* method = callee();
4525 int vtable_index = method->vtable_index();
4526 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4527 "bad index %d", vtable_index);
4528 // Get the Method* out of the appropriate vtable entry.
4529 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4530 vtable_index*vtableEntry::size_in_bytes() +
4531 in_bytes(vtableEntry::method_offset());
4532 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4533 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4534
4535 // Compare the target method with the expected method (e.g., Object.hashCode).
4536 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4537
4538 Node* native_call = makecon(native_call_addr);
4539 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4540 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4541
4542 return generate_slow_guard(test_native, slow_region);
4543 }
4544
4545 //-----------------------generate_method_call----------------------------
4546 // Use generate_method_call to make a slow-call to the real
4547 // method if the fast path fails. An alternative would be to
4548 // use a stub like OptoRuntime::slow_arraycopy_Java.
4549 // This only works for expanding the current library call,
4550 // not another intrinsic. (E.g., don't use this for making an
4551 // arraycopy call inside of the copyOf intrinsic.)
4552 CallJavaNode*
4553 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4554 // When compiling the intrinsic method itself, do not use this technique.
4555 guarantee(callee() != C->method(), "cannot make slow-call to self");
4556
4557 ciMethod* method = callee();
4558 // ensure the JVMS we have will be correct for this call
4559 guarantee(method_id == method->intrinsic_id(), "must match");
4560
4561 const TypeFunc* tf = TypeFunc::make(method);
4562 if (res_not_null) {
4563 assert(tf->return_type() == T_OBJECT, "");
4564 const TypeTuple* range = tf->range();
4565 const Type** fields = TypeTuple::fields(range->cnt());
4566 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4567 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4568 tf = TypeFunc::make(tf->domain(), new_range);
4569 }
4570 CallJavaNode* slow_call;
4571 if (is_static) {
4572 assert(!is_virtual, "");
4573 slow_call = new CallStaticJavaNode(C, tf,
4574 SharedRuntime::get_resolve_static_call_stub(), method);
4575 } else if (is_virtual) {
4576 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4577 int vtable_index = Method::invalid_vtable_index;
4578 if (UseInlineCaches) {
4579 // Suppress the vtable call
4580 } else {
4581 // hashCode and clone are not a miranda methods,
4582 // so the vtable index is fixed.
4583 // No need to use the linkResolver to get it.
4584 vtable_index = method->vtable_index();
4585 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4586 "bad index %d", vtable_index);
4587 }
4588 slow_call = new CallDynamicJavaNode(tf,
4589 SharedRuntime::get_resolve_virtual_call_stub(),
4590 method, vtable_index);
4591 } else { // neither virtual nor static: opt_virtual
4592 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4593 slow_call = new CallStaticJavaNode(C, tf,
4594 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4595 slow_call->set_optimized_virtual(true);
4596 }
4597 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4598 // To be able to issue a direct call (optimized virtual or virtual)
4599 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4600 // about the method being invoked should be attached to the call site to
4601 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4602 slow_call->set_override_symbolic_info(true);
4603 }
4604 set_arguments_for_java_call(slow_call);
4605 set_edges_for_java_call(slow_call);
4606 return slow_call;
4607 }
4608
4609
4610 /**
4611 * Build special case code for calls to hashCode on an object. This call may
4612 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4613 * slightly different code.
4614 */
4615 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4616 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4617 assert(!(is_virtual && is_static), "either virtual, special, or static");
4618
4619 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4620
4621 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4622 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4623 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4624 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4625 Node* obj = nullptr;
4626 if (!is_static) {
4627 // Check for hashing null object
4628 obj = null_check_receiver();
4629 if (stopped()) return true; // unconditionally null
4630 result_reg->init_req(_null_path, top());
4631 result_val->init_req(_null_path, top());
4632 } else {
4633 // Do a null check, and return zero if null.
4634 // System.identityHashCode(null) == 0
4635 obj = argument(0);
4636 Node* null_ctl = top();
4637 obj = null_check_oop(obj, &null_ctl);
4638 result_reg->init_req(_null_path, null_ctl);
4639 result_val->init_req(_null_path, _gvn.intcon(0));
4640 }
4641
4642 // Unconditionally null? Then return right away.
4643 if (stopped()) {
4644 set_control( result_reg->in(_null_path));
4645 if (!stopped())
4646 set_result(result_val->in(_null_path));
4647 return true;
4648 }
4649
4650 // We only go to the fast case code if we pass a number of guards. The
4651 // paths which do not pass are accumulated in the slow_region.
4652 RegionNode* slow_region = new RegionNode(1);
4653 record_for_igvn(slow_region);
4654
4655 // If this is a virtual call, we generate a funny guard. We pull out
4656 // the vtable entry corresponding to hashCode() from the target object.
4657 // If the target method which we are calling happens to be the native
4658 // Object hashCode() method, we pass the guard. We do not need this
4659 // guard for non-virtual calls -- the caller is known to be the native
4660 // Object hashCode().
4661 if (is_virtual) {
4662 // After null check, get the object's klass.
4663 Node* obj_klass = load_object_klass(obj);
4664 generate_virtual_guard(obj_klass, slow_region);
4665 }
4666
4667 // Get the header out of the object, use LoadMarkNode when available
4668 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4669 // The control of the load must be null. Otherwise, the load can move before
4670 // the null check after castPP removal.
4671 Node* no_ctrl = nullptr;
4672 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4673
4674 if (!UseObjectMonitorTable) {
4675 // Test the header to see if it is safe to read w.r.t. locking.
4676 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4677 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4678 if (LockingMode == LM_LIGHTWEIGHT) {
4679 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
4680 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
4681 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
4682
4683 generate_slow_guard(test_monitor, slow_region);
4684 } else {
4685 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4686 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val));
4687 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne));
4688
4689 generate_slow_guard(test_not_unlocked, slow_region);
4690 }
4691 }
4692
4693 // Get the hash value and check to see that it has been properly assigned.
4694 // We depend on hash_mask being at most 32 bits and avoid the use of
4695 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4696 // vm: see markWord.hpp.
4697 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4698 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4699 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4700 // This hack lets the hash bits live anywhere in the mark object now, as long
4701 // as the shift drops the relevant bits into the low 32 bits. Note that
4702 // Java spec says that HashCode is an int so there's no point in capturing
4703 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4704 hshifted_header = ConvX2I(hshifted_header);
4705 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4706
4707 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4708 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4709 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4710
4711 generate_slow_guard(test_assigned, slow_region);
4712
4713 Node* init_mem = reset_memory();
4714 // fill in the rest of the null path:
4715 result_io ->init_req(_null_path, i_o());
4716 result_mem->init_req(_null_path, init_mem);
4717
4718 result_val->init_req(_fast_path, hash_val);
4719 result_reg->init_req(_fast_path, control());
4720 result_io ->init_req(_fast_path, i_o());
4721 result_mem->init_req(_fast_path, init_mem);
4722
4723 // Generate code for the slow case. We make a call to hashCode().
4724 set_control(_gvn.transform(slow_region));
4725 if (!stopped()) {
4726 // No need for PreserveJVMState, because we're using up the present state.
4727 set_all_memory(init_mem);
4728 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4729 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4730 Node* slow_result = set_results_for_java_call(slow_call);
4731 // this->control() comes from set_results_for_java_call
4732 result_reg->init_req(_slow_path, control());
4733 result_val->init_req(_slow_path, slow_result);
4734 result_io ->set_req(_slow_path, i_o());
4735 result_mem ->set_req(_slow_path, reset_memory());
4736 }
4737
4738 // Return the combined state.
4739 set_i_o( _gvn.transform(result_io) );
4740 set_all_memory( _gvn.transform(result_mem));
4741
4742 set_result(result_reg, result_val);
4743 return true;
4744 }
4745
4746 //---------------------------inline_native_getClass----------------------------
4747 // public final native Class<?> java.lang.Object.getClass();
4748 //
4749 // Build special case code for calls to getClass on an object.
4750 bool LibraryCallKit::inline_native_getClass() {
4751 Node* obj = null_check_receiver();
4752 if (stopped()) return true;
4753 set_result(load_mirror_from_klass(load_object_klass(obj)));
4754 return true;
4755 }
4756
4757 //-----------------inline_native_Reflection_getCallerClass---------------------
4758 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4759 //
4760 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4761 //
4762 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4763 // in that it must skip particular security frames and checks for
4764 // caller sensitive methods.
4765 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4766 #ifndef PRODUCT
4767 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4768 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4769 }
4770 #endif
4771
4772 if (!jvms()->has_method()) {
4773 #ifndef PRODUCT
4774 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4775 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4776 }
4777 #endif
4778 return false;
4779 }
4780
4781 // Walk back up the JVM state to find the caller at the required
4782 // depth.
4783 JVMState* caller_jvms = jvms();
4784
4785 // Cf. JVM_GetCallerClass
4786 // NOTE: Start the loop at depth 1 because the current JVM state does
4787 // not include the Reflection.getCallerClass() frame.
4788 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4789 ciMethod* m = caller_jvms->method();
4790 switch (n) {
4791 case 0:
4792 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4793 break;
4794 case 1:
4795 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4796 if (!m->caller_sensitive()) {
4797 #ifndef PRODUCT
4798 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4799 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4800 }
4801 #endif
4802 return false; // bail-out; let JVM_GetCallerClass do the work
4803 }
4804 break;
4805 default:
4806 if (!m->is_ignored_by_security_stack_walk()) {
4807 // We have reached the desired frame; return the holder class.
4808 // Acquire method holder as java.lang.Class and push as constant.
4809 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4810 ciInstance* caller_mirror = caller_klass->java_mirror();
4811 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4812
4813 #ifndef PRODUCT
4814 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4815 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());
4816 tty->print_cr(" JVM state at this point:");
4817 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4818 ciMethod* m = jvms()->of_depth(i)->method();
4819 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4820 }
4821 }
4822 #endif
4823 return true;
4824 }
4825 break;
4826 }
4827 }
4828
4829 #ifndef PRODUCT
4830 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4831 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4832 tty->print_cr(" JVM state at this point:");
4833 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4834 ciMethod* m = jvms()->of_depth(i)->method();
4835 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4836 }
4837 }
4838 #endif
4839
4840 return false; // bail-out; let JVM_GetCallerClass do the work
4841 }
4842
4843 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4844 Node* arg = argument(0);
4845 Node* result = nullptr;
4846
4847 switch (id) {
4848 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4849 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4850 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4851 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4852 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
4853 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
4854
4855 case vmIntrinsics::_doubleToLongBits: {
4856 // two paths (plus control) merge in a wood
4857 RegionNode *r = new RegionNode(3);
4858 Node *phi = new PhiNode(r, TypeLong::LONG);
4859
4860 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4861 // Build the boolean node
4862 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4863
4864 // Branch either way.
4865 // NaN case is less traveled, which makes all the difference.
4866 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4867 Node *opt_isnan = _gvn.transform(ifisnan);
4868 assert( opt_isnan->is_If(), "Expect an IfNode");
4869 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4870 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4871
4872 set_control(iftrue);
4873
4874 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4875 Node *slow_result = longcon(nan_bits); // return NaN
4876 phi->init_req(1, _gvn.transform( slow_result ));
4877 r->init_req(1, iftrue);
4878
4879 // Else fall through
4880 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4881 set_control(iffalse);
4882
4883 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4884 r->init_req(2, iffalse);
4885
4886 // Post merge
4887 set_control(_gvn.transform(r));
4888 record_for_igvn(r);
4889
4890 C->set_has_split_ifs(true); // Has chance for split-if optimization
4891 result = phi;
4892 assert(result->bottom_type()->isa_long(), "must be");
4893 break;
4894 }
4895
4896 case vmIntrinsics::_floatToIntBits: {
4897 // two paths (plus control) merge in a wood
4898 RegionNode *r = new RegionNode(3);
4899 Node *phi = new PhiNode(r, TypeInt::INT);
4900
4901 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4902 // Build the boolean node
4903 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4904
4905 // Branch either way.
4906 // NaN case is less traveled, which makes all the difference.
4907 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4908 Node *opt_isnan = _gvn.transform(ifisnan);
4909 assert( opt_isnan->is_If(), "Expect an IfNode");
4910 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4911 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4912
4913 set_control(iftrue);
4914
4915 static const jint nan_bits = 0x7fc00000;
4916 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4917 phi->init_req(1, _gvn.transform( slow_result ));
4918 r->init_req(1, iftrue);
4919
4920 // Else fall through
4921 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4922 set_control(iffalse);
4923
4924 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4925 r->init_req(2, iffalse);
4926
4927 // Post merge
4928 set_control(_gvn.transform(r));
4929 record_for_igvn(r);
4930
4931 C->set_has_split_ifs(true); // Has chance for split-if optimization
4932 result = phi;
4933 assert(result->bottom_type()->isa_int(), "must be");
4934 break;
4935 }
4936
4937 default:
4938 fatal_unexpected_iid(id);
4939 break;
4940 }
4941 set_result(_gvn.transform(result));
4942 return true;
4943 }
4944
4945 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
4946 Node* arg = argument(0);
4947 Node* result = nullptr;
4948
4949 switch (id) {
4950 case vmIntrinsics::_floatIsInfinite:
4951 result = new IsInfiniteFNode(arg);
4952 break;
4953 case vmIntrinsics::_floatIsFinite:
4954 result = new IsFiniteFNode(arg);
4955 break;
4956 case vmIntrinsics::_doubleIsInfinite:
4957 result = new IsInfiniteDNode(arg);
4958 break;
4959 case vmIntrinsics::_doubleIsFinite:
4960 result = new IsFiniteDNode(arg);
4961 break;
4962 default:
4963 fatal_unexpected_iid(id);
4964 break;
4965 }
4966 set_result(_gvn.transform(result));
4967 return true;
4968 }
4969
4970 //----------------------inline_unsafe_copyMemory-------------------------
4971 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4972
4973 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
4974 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
4975 const Type* base_t = gvn.type(base);
4976
4977 bool in_native = (base_t == TypePtr::NULL_PTR);
4978 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
4979 bool is_mixed = !in_heap && !in_native;
4980
4981 if (is_mixed) {
4982 return true; // mixed accesses can touch both on-heap and off-heap memory
4983 }
4984 if (in_heap) {
4985 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
4986 if (!is_prim_array) {
4987 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
4988 // there's not enough type information available to determine proper memory slice for it.
4989 return true;
4990 }
4991 }
4992 return false;
4993 }
4994
4995 bool LibraryCallKit::inline_unsafe_copyMemory() {
4996 if (callee()->is_static()) return false; // caller must have the capability!
4997 null_check_receiver(); // null-check receiver
4998 if (stopped()) return true;
4999
5000 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5001
5002 Node* src_base = argument(1); // type: oop
5003 Node* src_off = ConvL2X(argument(2)); // type: long
5004 Node* dst_base = argument(4); // type: oop
5005 Node* dst_off = ConvL2X(argument(5)); // type: long
5006 Node* size = ConvL2X(argument(7)); // type: long
5007
5008 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5009 "fieldOffset must be byte-scaled");
5010
5011 Node* src_addr = make_unsafe_address(src_base, src_off);
5012 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5013
5014 Node* thread = _gvn.transform(new ThreadLocalNode());
5015 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5016 BasicType doing_unsafe_access_bt = T_BYTE;
5017 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5018
5019 // update volatile field
5020 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5021
5022 int flags = RC_LEAF | RC_NO_FP;
5023
5024 const TypePtr* dst_type = TypePtr::BOTTOM;
5025
5026 // Adjust memory effects of the runtime call based on input values.
5027 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5028 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5029 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5030
5031 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5032 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5033 flags |= RC_NARROW_MEM; // narrow in memory
5034 }
5035 }
5036
5037 // Call it. Note that the length argument is not scaled.
5038 make_runtime_call(flags,
5039 OptoRuntime::fast_arraycopy_Type(),
5040 StubRoutines::unsafe_arraycopy(),
5041 "unsafe_arraycopy",
5042 dst_type,
5043 src_addr, dst_addr, size XTOP);
5044
5045 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5046
5047 return true;
5048 }
5049
5050 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5051 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5052 bool LibraryCallKit::inline_unsafe_setMemory() {
5053 if (callee()->is_static()) return false; // caller must have the capability!
5054 null_check_receiver(); // null-check receiver
5055 if (stopped()) return true;
5056
5057 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5058
5059 Node* dst_base = argument(1); // type: oop
5060 Node* dst_off = ConvL2X(argument(2)); // type: long
5061 Node* size = ConvL2X(argument(4)); // type: long
5062 Node* byte = argument(6); // type: byte
5063
5064 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5065 "fieldOffset must be byte-scaled");
5066
5067 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5068
5069 Node* thread = _gvn.transform(new ThreadLocalNode());
5070 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5071 BasicType doing_unsafe_access_bt = T_BYTE;
5072 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5073
5074 // update volatile field
5075 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5076
5077 int flags = RC_LEAF | RC_NO_FP;
5078
5079 const TypePtr* dst_type = TypePtr::BOTTOM;
5080
5081 // Adjust memory effects of the runtime call based on input values.
5082 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5083 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5084
5085 flags |= RC_NARROW_MEM; // narrow in memory
5086 }
5087
5088 // Call it. Note that the length argument is not scaled.
5089 make_runtime_call(flags,
5090 OptoRuntime::unsafe_setmemory_Type(),
5091 StubRoutines::unsafe_setmemory(),
5092 "unsafe_setmemory",
5093 dst_type,
5094 dst_addr, size XTOP, byte);
5095
5096 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5097
5098 return true;
5099 }
5100
5101 #undef XTOP
5102
5103 //------------------------clone_coping-----------------------------------
5104 // Helper function for inline_native_clone.
5105 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5106 assert(obj_size != nullptr, "");
5107 Node* raw_obj = alloc_obj->in(1);
5108 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5109
5110 AllocateNode* alloc = nullptr;
5111 if (ReduceBulkZeroing &&
5112 // If we are implementing an array clone without knowing its source type
5113 // (can happen when compiling the array-guarded branch of a reflective
5114 // Object.clone() invocation), initialize the array within the allocation.
5115 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5116 // to a runtime clone call that assumes fully initialized source arrays.
5117 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5118 // We will be completely responsible for initializing this object -
5119 // mark Initialize node as complete.
5120 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5121 // The object was just allocated - there should be no any stores!
5122 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5123 // Mark as complete_with_arraycopy so that on AllocateNode
5124 // expansion, we know this AllocateNode is initialized by an array
5125 // copy and a StoreStore barrier exists after the array copy.
5126 alloc->initialization()->set_complete_with_arraycopy();
5127 }
5128
5129 Node* size = _gvn.transform(obj_size);
5130 access_clone(obj, alloc_obj, size, is_array);
5131
5132 // Do not let reads from the cloned object float above the arraycopy.
5133 if (alloc != nullptr) {
5134 // Do not let stores that initialize this object be reordered with
5135 // a subsequent store that would make this object accessible by
5136 // other threads.
5137 // Record what AllocateNode this StoreStore protects so that
5138 // escape analysis can go from the MemBarStoreStoreNode to the
5139 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5140 // based on the escape status of the AllocateNode.
5141 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5142 } else {
5143 insert_mem_bar(Op_MemBarCPUOrder);
5144 }
5145 }
5146
5147 //------------------------inline_native_clone----------------------------
5148 // protected native Object java.lang.Object.clone();
5149 //
5150 // Here are the simple edge cases:
5151 // null receiver => normal trap
5152 // virtual and clone was overridden => slow path to out-of-line clone
5153 // not cloneable or finalizer => slow path to out-of-line Object.clone
5154 //
5155 // The general case has two steps, allocation and copying.
5156 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5157 //
5158 // Copying also has two cases, oop arrays and everything else.
5159 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5160 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5161 //
5162 // These steps fold up nicely if and when the cloned object's klass
5163 // can be sharply typed as an object array, a type array, or an instance.
5164 //
5165 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5166 PhiNode* result_val;
5167
5168 // Set the reexecute bit for the interpreter to reexecute
5169 // the bytecode that invokes Object.clone if deoptimization happens.
5170 { PreserveReexecuteState preexecs(this);
5171 jvms()->set_should_reexecute(true);
5172
5173 Node* obj = null_check_receiver();
5174 if (stopped()) return true;
5175
5176 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5177
5178 // If we are going to clone an instance, we need its exact type to
5179 // know the number and types of fields to convert the clone to
5180 // loads/stores. Maybe a speculative type can help us.
5181 if (!obj_type->klass_is_exact() &&
5182 obj_type->speculative_type() != nullptr &&
5183 obj_type->speculative_type()->is_instance_klass()) {
5184 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5185 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5186 !spec_ik->has_injected_fields()) {
5187 if (!obj_type->isa_instptr() ||
5188 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5189 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5190 }
5191 }
5192 }
5193
5194 // Conservatively insert a memory barrier on all memory slices.
5195 // Do not let writes into the original float below the clone.
5196 insert_mem_bar(Op_MemBarCPUOrder);
5197
5198 // paths into result_reg:
5199 enum {
5200 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5201 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5202 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5203 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5204 PATH_LIMIT
5205 };
5206 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5207 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5208 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5209 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5210 record_for_igvn(result_reg);
5211
5212 Node* obj_klass = load_object_klass(obj);
5213 Node* array_obj = obj;
5214 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5215 if (array_ctl != nullptr) {
5216 // It's an array.
5217 PreserveJVMState pjvms(this);
5218 set_control(array_ctl);
5219 Node* obj_length = load_array_length(array_obj);
5220 Node* array_size = nullptr; // Size of the array without object alignment padding.
5221 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5222
5223 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5224 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5225 // If it is an oop array, it requires very special treatment,
5226 // because gc barriers are required when accessing the array.
5227 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5228 if (is_obja != nullptr) {
5229 PreserveJVMState pjvms2(this);
5230 set_control(is_obja);
5231 // Generate a direct call to the right arraycopy function(s).
5232 // Clones are always tightly coupled.
5233 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5234 ac->set_clone_oop_array();
5235 Node* n = _gvn.transform(ac);
5236 assert(n == ac, "cannot disappear");
5237 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5238
5239 result_reg->init_req(_objArray_path, control());
5240 result_val->init_req(_objArray_path, alloc_obj);
5241 result_i_o ->set_req(_objArray_path, i_o());
5242 result_mem ->set_req(_objArray_path, reset_memory());
5243 }
5244 }
5245 // Otherwise, there are no barriers to worry about.
5246 // (We can dispense with card marks if we know the allocation
5247 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5248 // causes the non-eden paths to take compensating steps to
5249 // simulate a fresh allocation, so that no further
5250 // card marks are required in compiled code to initialize
5251 // the object.)
5252
5253 if (!stopped()) {
5254 copy_to_clone(array_obj, alloc_obj, array_size, true);
5255
5256 // Present the results of the copy.
5257 result_reg->init_req(_array_path, control());
5258 result_val->init_req(_array_path, alloc_obj);
5259 result_i_o ->set_req(_array_path, i_o());
5260 result_mem ->set_req(_array_path, reset_memory());
5261 }
5262 }
5263
5264 // We only go to the instance fast case code if we pass a number of guards.
5265 // The paths which do not pass are accumulated in the slow_region.
5266 RegionNode* slow_region = new RegionNode(1);
5267 record_for_igvn(slow_region);
5268 if (!stopped()) {
5269 // It's an instance (we did array above). Make the slow-path tests.
5270 // If this is a virtual call, we generate a funny guard. We grab
5271 // the vtable entry corresponding to clone() from the target object.
5272 // If the target method which we are calling happens to be the
5273 // Object clone() method, we pass the guard. We do not need this
5274 // guard for non-virtual calls; the caller is known to be the native
5275 // Object clone().
5276 if (is_virtual) {
5277 generate_virtual_guard(obj_klass, slow_region);
5278 }
5279
5280 // The object must be easily cloneable and must not have a finalizer.
5281 // Both of these conditions may be checked in a single test.
5282 // We could optimize the test further, but we don't care.
5283 generate_misc_flags_guard(obj_klass,
5284 // Test both conditions:
5285 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5286 // Must be cloneable but not finalizer:
5287 KlassFlags::_misc_is_cloneable_fast,
5288 slow_region);
5289 }
5290
5291 if (!stopped()) {
5292 // It's an instance, and it passed the slow-path tests.
5293 PreserveJVMState pjvms(this);
5294 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5295 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5296 // is reexecuted if deoptimization occurs and there could be problems when merging
5297 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5298 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5299
5300 copy_to_clone(obj, alloc_obj, obj_size, false);
5301
5302 // Present the results of the slow call.
5303 result_reg->init_req(_instance_path, control());
5304 result_val->init_req(_instance_path, alloc_obj);
5305 result_i_o ->set_req(_instance_path, i_o());
5306 result_mem ->set_req(_instance_path, reset_memory());
5307 }
5308
5309 // Generate code for the slow case. We make a call to clone().
5310 set_control(_gvn.transform(slow_region));
5311 if (!stopped()) {
5312 PreserveJVMState pjvms(this);
5313 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5314 // We need to deoptimize on exception (see comment above)
5315 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5316 // this->control() comes from set_results_for_java_call
5317 result_reg->init_req(_slow_path, control());
5318 result_val->init_req(_slow_path, slow_result);
5319 result_i_o ->set_req(_slow_path, i_o());
5320 result_mem ->set_req(_slow_path, reset_memory());
5321 }
5322
5323 // Return the combined state.
5324 set_control( _gvn.transform(result_reg));
5325 set_i_o( _gvn.transform(result_i_o));
5326 set_all_memory( _gvn.transform(result_mem));
5327 } // original reexecute is set back here
5328
5329 set_result(_gvn.transform(result_val));
5330 return true;
5331 }
5332
5333 // If we have a tightly coupled allocation, the arraycopy may take care
5334 // of the array initialization. If one of the guards we insert between
5335 // the allocation and the arraycopy causes a deoptimization, an
5336 // uninitialized array will escape the compiled method. To prevent that
5337 // we set the JVM state for uncommon traps between the allocation and
5338 // the arraycopy to the state before the allocation so, in case of
5339 // deoptimization, we'll reexecute the allocation and the
5340 // initialization.
5341 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5342 if (alloc != nullptr) {
5343 ciMethod* trap_method = alloc->jvms()->method();
5344 int trap_bci = alloc->jvms()->bci();
5345
5346 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5347 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5348 // Make sure there's no store between the allocation and the
5349 // arraycopy otherwise visible side effects could be rexecuted
5350 // in case of deoptimization and cause incorrect execution.
5351 bool no_interfering_store = true;
5352 Node* mem = alloc->in(TypeFunc::Memory);
5353 if (mem->is_MergeMem()) {
5354 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5355 Node* n = mms.memory();
5356 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5357 assert(n->is_Store(), "what else?");
5358 no_interfering_store = false;
5359 break;
5360 }
5361 }
5362 } else {
5363 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5364 Node* n = mms.memory();
5365 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5366 assert(n->is_Store(), "what else?");
5367 no_interfering_store = false;
5368 break;
5369 }
5370 }
5371 }
5372
5373 if (no_interfering_store) {
5374 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5375
5376 JVMState* saved_jvms = jvms();
5377 saved_reexecute_sp = _reexecute_sp;
5378
5379 set_jvms(sfpt->jvms());
5380 _reexecute_sp = jvms()->sp();
5381
5382 return saved_jvms;
5383 }
5384 }
5385 }
5386 return nullptr;
5387 }
5388
5389 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5390 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5391 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5392 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5393 uint size = alloc->req();
5394 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5395 old_jvms->set_map(sfpt);
5396 for (uint i = 0; i < size; i++) {
5397 sfpt->init_req(i, alloc->in(i));
5398 }
5399 // re-push array length for deoptimization
5400 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5401 old_jvms->set_sp(old_jvms->sp()+1);
5402 old_jvms->set_monoff(old_jvms->monoff()+1);
5403 old_jvms->set_scloff(old_jvms->scloff()+1);
5404 old_jvms->set_endoff(old_jvms->endoff()+1);
5405 old_jvms->set_should_reexecute(true);
5406
5407 sfpt->set_i_o(map()->i_o());
5408 sfpt->set_memory(map()->memory());
5409 sfpt->set_control(map()->control());
5410 return sfpt;
5411 }
5412
5413 // In case of a deoptimization, we restart execution at the
5414 // allocation, allocating a new array. We would leave an uninitialized
5415 // array in the heap that GCs wouldn't expect. Move the allocation
5416 // after the traps so we don't allocate the array if we
5417 // deoptimize. This is possible because tightly_coupled_allocation()
5418 // guarantees there's no observer of the allocated array at this point
5419 // and the control flow is simple enough.
5420 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5421 int saved_reexecute_sp, uint new_idx) {
5422 if (saved_jvms_before_guards != nullptr && !stopped()) {
5423 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5424
5425 assert(alloc != nullptr, "only with a tightly coupled allocation");
5426 // restore JVM state to the state at the arraycopy
5427 saved_jvms_before_guards->map()->set_control(map()->control());
5428 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5429 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5430 // If we've improved the types of some nodes (null check) while
5431 // emitting the guards, propagate them to the current state
5432 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5433 set_jvms(saved_jvms_before_guards);
5434 _reexecute_sp = saved_reexecute_sp;
5435
5436 // Remove the allocation from above the guards
5437 CallProjections callprojs;
5438 alloc->extract_projections(&callprojs, true);
5439 InitializeNode* init = alloc->initialization();
5440 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5441 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5442 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5443
5444 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5445 // the allocation (i.e. is only valid if the allocation succeeds):
5446 // 1) replace CastIINode with AllocateArrayNode's length here
5447 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5448 //
5449 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5450 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5451 Node* init_control = init->proj_out(TypeFunc::Control);
5452 Node* alloc_length = alloc->Ideal_length();
5453 #ifdef ASSERT
5454 Node* prev_cast = nullptr;
5455 #endif
5456 for (uint i = 0; i < init_control->outcnt(); i++) {
5457 Node* init_out = init_control->raw_out(i);
5458 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5459 #ifdef ASSERT
5460 if (prev_cast == nullptr) {
5461 prev_cast = init_out;
5462 } else {
5463 if (prev_cast->cmp(*init_out) == false) {
5464 prev_cast->dump();
5465 init_out->dump();
5466 assert(false, "not equal CastIINode");
5467 }
5468 }
5469 #endif
5470 C->gvn_replace_by(init_out, alloc_length);
5471 }
5472 }
5473 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5474
5475 // move the allocation here (after the guards)
5476 _gvn.hash_delete(alloc);
5477 alloc->set_req(TypeFunc::Control, control());
5478 alloc->set_req(TypeFunc::I_O, i_o());
5479 Node *mem = reset_memory();
5480 set_all_memory(mem);
5481 alloc->set_req(TypeFunc::Memory, mem);
5482 set_control(init->proj_out_or_null(TypeFunc::Control));
5483 set_i_o(callprojs.fallthrough_ioproj);
5484
5485 // Update memory as done in GraphKit::set_output_for_allocation()
5486 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5487 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5488 if (ary_type->isa_aryptr() && length_type != nullptr) {
5489 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5490 }
5491 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5492 int elemidx = C->get_alias_index(telemref);
5493 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5494 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5495
5496 Node* allocx = _gvn.transform(alloc);
5497 assert(allocx == alloc, "where has the allocation gone?");
5498 assert(dest->is_CheckCastPP(), "not an allocation result?");
5499
5500 _gvn.hash_delete(dest);
5501 dest->set_req(0, control());
5502 Node* destx = _gvn.transform(dest);
5503 assert(destx == dest, "where has the allocation result gone?");
5504
5505 array_ideal_length(alloc, ary_type, true);
5506 }
5507 }
5508
5509 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5510 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5511 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5512 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5513 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5514 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5515 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5516 JVMState* saved_jvms_before_guards) {
5517 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5518 // There is at least one unrelated uncommon trap which needs to be replaced.
5519 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5520
5521 JVMState* saved_jvms = jvms();
5522 const int saved_reexecute_sp = _reexecute_sp;
5523 set_jvms(sfpt->jvms());
5524 _reexecute_sp = jvms()->sp();
5525
5526 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5527
5528 // Restore state
5529 set_jvms(saved_jvms);
5530 _reexecute_sp = saved_reexecute_sp;
5531 }
5532 }
5533
5534 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5535 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5536 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5537 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5538 while (if_proj->is_IfProj()) {
5539 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5540 if (uncommon_trap != nullptr) {
5541 create_new_uncommon_trap(uncommon_trap);
5542 }
5543 assert(if_proj->in(0)->is_If(), "must be If");
5544 if_proj = if_proj->in(0)->in(0);
5545 }
5546 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5547 "must have reached control projection of init node");
5548 }
5549
5550 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5551 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5552 assert(trap_request != 0, "no valid UCT trap request");
5553 PreserveJVMState pjvms(this);
5554 set_control(uncommon_trap_call->in(0));
5555 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5556 Deoptimization::trap_request_action(trap_request));
5557 assert(stopped(), "Should be stopped");
5558 _gvn.hash_delete(uncommon_trap_call);
5559 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5560 }
5561
5562 // Common checks for array sorting intrinsics arguments.
5563 // Returns `true` if checks passed.
5564 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
5565 // check address of the class
5566 if (elementType == nullptr || elementType->is_top()) {
5567 return false; // dead path
5568 }
5569 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5570 if (elem_klass == nullptr) {
5571 return false; // dead path
5572 }
5573 // java_mirror_type() returns non-null for compile-time Class constants only
5574 ciType* elem_type = elem_klass->java_mirror_type();
5575 if (elem_type == nullptr) {
5576 return false;
5577 }
5578 bt = elem_type->basic_type();
5579 // Disable the intrinsic if the CPU does not support SIMD sort
5580 if (!Matcher::supports_simd_sort(bt)) {
5581 return false;
5582 }
5583 // check address of the array
5584 if (obj == nullptr || obj->is_top()) {
5585 return false; // dead path
5586 }
5587 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5588 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
5589 return false; // failed input validation
5590 }
5591 return true;
5592 }
5593
5594 //------------------------------inline_array_partition-----------------------
5595 bool LibraryCallKit::inline_array_partition() {
5596 address stubAddr = StubRoutines::select_array_partition_function();
5597 if (stubAddr == nullptr) {
5598 return false; // Intrinsic's stub is not implemented on this platform
5599 }
5600 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
5601
5602 // no receiver because it is a static method
5603 Node* elementType = argument(0);
5604 Node* obj = argument(1);
5605 Node* offset = argument(2); // long
5606 Node* fromIndex = argument(4);
5607 Node* toIndex = argument(5);
5608 Node* indexPivot1 = argument(6);
5609 Node* indexPivot2 = argument(7);
5610 // PartitionOperation: argument(8) is ignored
5611
5612 Node* pivotIndices = nullptr;
5613 BasicType bt = T_ILLEGAL;
5614
5615 if (!check_array_sort_arguments(elementType, obj, bt)) {
5616 return false;
5617 }
5618 null_check(obj);
5619 // If obj is dead, only null-path is taken.
5620 if (stopped()) {
5621 return true;
5622 }
5623 // Set the original stack and the reexecute bit for the interpreter to reexecute
5624 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
5625 { PreserveReexecuteState preexecs(this);
5626 jvms()->set_should_reexecute(true);
5627
5628 Node* obj_adr = make_unsafe_address(obj, offset);
5629
5630 // create the pivotIndices array of type int and size = 2
5631 Node* size = intcon(2);
5632 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
5633 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
5634 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
5635 guarantee(alloc != nullptr, "created above");
5636 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
5637
5638 // pass the basic type enum to the stub
5639 Node* elemType = intcon(bt);
5640
5641 // Call the stub
5642 const char *stubName = "array_partition_stub";
5643 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
5644 stubAddr, stubName, TypePtr::BOTTOM,
5645 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
5646 indexPivot1, indexPivot2);
5647
5648 } // original reexecute is set back here
5649
5650 if (!stopped()) {
5651 set_result(pivotIndices);
5652 }
5653
5654 return true;
5655 }
5656
5657
5658 //------------------------------inline_array_sort-----------------------
5659 bool LibraryCallKit::inline_array_sort() {
5660 address stubAddr = StubRoutines::select_arraysort_function();
5661 if (stubAddr == nullptr) {
5662 return false; // Intrinsic's stub is not implemented on this platform
5663 }
5664 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
5665
5666 // no receiver because it is a static method
5667 Node* elementType = argument(0);
5668 Node* obj = argument(1);
5669 Node* offset = argument(2); // long
5670 Node* fromIndex = argument(4);
5671 Node* toIndex = argument(5);
5672 // SortOperation: argument(6) is ignored
5673
5674 BasicType bt = T_ILLEGAL;
5675
5676 if (!check_array_sort_arguments(elementType, obj, bt)) {
5677 return false;
5678 }
5679 null_check(obj);
5680 // If obj is dead, only null-path is taken.
5681 if (stopped()) {
5682 return true;
5683 }
5684 Node* obj_adr = make_unsafe_address(obj, offset);
5685
5686 // pass the basic type enum to the stub
5687 Node* elemType = intcon(bt);
5688
5689 // Call the stub.
5690 const char *stubName = "arraysort_stub";
5691 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
5692 stubAddr, stubName, TypePtr::BOTTOM,
5693 obj_adr, elemType, fromIndex, toIndex);
5694
5695 return true;
5696 }
5697
5698
5699 //------------------------------inline_arraycopy-----------------------
5700 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5701 // Object dest, int destPos,
5702 // int length);
5703 bool LibraryCallKit::inline_arraycopy() {
5704 // Get the arguments.
5705 Node* src = argument(0); // type: oop
5706 Node* src_offset = argument(1); // type: int
5707 Node* dest = argument(2); // type: oop
5708 Node* dest_offset = argument(3); // type: int
5709 Node* length = argument(4); // type: int
5710
5711 uint new_idx = C->unique();
5712
5713 // Check for allocation before we add nodes that would confuse
5714 // tightly_coupled_allocation()
5715 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5716
5717 int saved_reexecute_sp = -1;
5718 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5719 // See arraycopy_restore_alloc_state() comment
5720 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5721 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5722 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5723 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5724
5725 // The following tests must be performed
5726 // (1) src and dest are arrays.
5727 // (2) src and dest arrays must have elements of the same BasicType
5728 // (3) src and dest must not be null.
5729 // (4) src_offset must not be negative.
5730 // (5) dest_offset must not be negative.
5731 // (6) length must not be negative.
5732 // (7) src_offset + length must not exceed length of src.
5733 // (8) dest_offset + length must not exceed length of dest.
5734 // (9) each element of an oop array must be assignable
5735
5736 // (3) src and dest must not be null.
5737 // always do this here because we need the JVM state for uncommon traps
5738 Node* null_ctl = top();
5739 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5740 assert(null_ctl->is_top(), "no null control here");
5741 dest = null_check(dest, T_ARRAY);
5742
5743 if (!can_emit_guards) {
5744 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5745 // guards but the arraycopy node could still take advantage of a
5746 // tightly allocated allocation. tightly_coupled_allocation() is
5747 // called again to make sure it takes the null check above into
5748 // account: the null check is mandatory and if it caused an
5749 // uncommon trap to be emitted then the allocation can't be
5750 // considered tightly coupled in this context.
5751 alloc = tightly_coupled_allocation(dest);
5752 }
5753
5754 bool validated = false;
5755
5756 const Type* src_type = _gvn.type(src);
5757 const Type* dest_type = _gvn.type(dest);
5758 const TypeAryPtr* top_src = src_type->isa_aryptr();
5759 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5760
5761 // Do we have the type of src?
5762 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5763 // Do we have the type of dest?
5764 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5765 // Is the type for src from speculation?
5766 bool src_spec = false;
5767 // Is the type for dest from speculation?
5768 bool dest_spec = false;
5769
5770 if ((!has_src || !has_dest) && can_emit_guards) {
5771 // We don't have sufficient type information, let's see if
5772 // speculative types can help. We need to have types for both src
5773 // and dest so that it pays off.
5774
5775 // Do we already have or could we have type information for src
5776 bool could_have_src = has_src;
5777 // Do we already have or could we have type information for dest
5778 bool could_have_dest = has_dest;
5779
5780 ciKlass* src_k = nullptr;
5781 if (!has_src) {
5782 src_k = src_type->speculative_type_not_null();
5783 if (src_k != nullptr && src_k->is_array_klass()) {
5784 could_have_src = true;
5785 }
5786 }
5787
5788 ciKlass* dest_k = nullptr;
5789 if (!has_dest) {
5790 dest_k = dest_type->speculative_type_not_null();
5791 if (dest_k != nullptr && dest_k->is_array_klass()) {
5792 could_have_dest = true;
5793 }
5794 }
5795
5796 if (could_have_src && could_have_dest) {
5797 // This is going to pay off so emit the required guards
5798 if (!has_src) {
5799 src = maybe_cast_profiled_obj(src, src_k, true);
5800 src_type = _gvn.type(src);
5801 top_src = src_type->isa_aryptr();
5802 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5803 src_spec = true;
5804 }
5805 if (!has_dest) {
5806 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5807 dest_type = _gvn.type(dest);
5808 top_dest = dest_type->isa_aryptr();
5809 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5810 dest_spec = true;
5811 }
5812 }
5813 }
5814
5815 if (has_src && has_dest && can_emit_guards) {
5816 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5817 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5818 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5819 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5820
5821 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5822 // If both arrays are object arrays then having the exact types
5823 // for both will remove the need for a subtype check at runtime
5824 // before the call and may make it possible to pick a faster copy
5825 // routine (without a subtype check on every element)
5826 // Do we have the exact type of src?
5827 bool could_have_src = src_spec;
5828 // Do we have the exact type of dest?
5829 bool could_have_dest = dest_spec;
5830 ciKlass* src_k = nullptr;
5831 ciKlass* dest_k = nullptr;
5832 if (!src_spec) {
5833 src_k = src_type->speculative_type_not_null();
5834 if (src_k != nullptr && src_k->is_array_klass()) {
5835 could_have_src = true;
5836 }
5837 }
5838 if (!dest_spec) {
5839 dest_k = dest_type->speculative_type_not_null();
5840 if (dest_k != nullptr && dest_k->is_array_klass()) {
5841 could_have_dest = true;
5842 }
5843 }
5844 if (could_have_src && could_have_dest) {
5845 // If we can have both exact types, emit the missing guards
5846 if (could_have_src && !src_spec) {
5847 src = maybe_cast_profiled_obj(src, src_k, true);
5848 }
5849 if (could_have_dest && !dest_spec) {
5850 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5851 }
5852 }
5853 }
5854 }
5855
5856 ciMethod* trap_method = method();
5857 int trap_bci = bci();
5858 if (saved_jvms_before_guards != nullptr) {
5859 trap_method = alloc->jvms()->method();
5860 trap_bci = alloc->jvms()->bci();
5861 }
5862
5863 bool negative_length_guard_generated = false;
5864
5865 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5866 can_emit_guards &&
5867 !src->is_top() && !dest->is_top()) {
5868 // validate arguments: enables transformation the ArrayCopyNode
5869 validated = true;
5870
5871 RegionNode* slow_region = new RegionNode(1);
5872 record_for_igvn(slow_region);
5873
5874 // (1) src and dest are arrays.
5875 generate_non_array_guard(load_object_klass(src), slow_region, &src);
5876 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
5877
5878 // (2) src and dest arrays must have elements of the same BasicType
5879 // done at macro expansion or at Ideal transformation time
5880
5881 // (4) src_offset must not be negative.
5882 generate_negative_guard(src_offset, slow_region);
5883
5884 // (5) dest_offset must not be negative.
5885 generate_negative_guard(dest_offset, slow_region);
5886
5887 // (7) src_offset + length must not exceed length of src.
5888 generate_limit_guard(src_offset, length,
5889 load_array_length(src),
5890 slow_region);
5891
5892 // (8) dest_offset + length must not exceed length of dest.
5893 generate_limit_guard(dest_offset, length,
5894 load_array_length(dest),
5895 slow_region);
5896
5897 // (6) length must not be negative.
5898 // This is also checked in generate_arraycopy() during macro expansion, but
5899 // we also have to check it here for the case where the ArrayCopyNode will
5900 // be eliminated by Escape Analysis.
5901 if (EliminateAllocations) {
5902 generate_negative_guard(length, slow_region);
5903 negative_length_guard_generated = true;
5904 }
5905
5906 // (9) each element of an oop array must be assignable
5907 Node* dest_klass = load_object_klass(dest);
5908 if (src != dest) {
5909 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
5910
5911 if (not_subtype_ctrl != top()) {
5912 PreserveJVMState pjvms(this);
5913 set_control(not_subtype_ctrl);
5914 uncommon_trap(Deoptimization::Reason_intrinsic,
5915 Deoptimization::Action_make_not_entrant);
5916 assert(stopped(), "Should be stopped");
5917 }
5918 }
5919 {
5920 PreserveJVMState pjvms(this);
5921 set_control(_gvn.transform(slow_region));
5922 uncommon_trap(Deoptimization::Reason_intrinsic,
5923 Deoptimization::Action_make_not_entrant);
5924 assert(stopped(), "Should be stopped");
5925 }
5926
5927 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5928 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
5929 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5930 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
5931 }
5932
5933 if (stopped()) {
5934 return true;
5935 }
5936
5937 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
5938 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5939 // so the compiler has a chance to eliminate them: during macro expansion,
5940 // we have to set their control (CastPP nodes are eliminated).
5941 load_object_klass(src), load_object_klass(dest),
5942 load_array_length(src), load_array_length(dest));
5943
5944 ac->set_arraycopy(validated);
5945
5946 Node* n = _gvn.transform(ac);
5947 if (n == ac) {
5948 ac->connect_outputs(this);
5949 } else {
5950 assert(validated, "shouldn't transform if all arguments not validated");
5951 set_all_memory(n);
5952 }
5953 clear_upper_avx();
5954
5955
5956 return true;
5957 }
5958
5959
5960 // Helper function which determines if an arraycopy immediately follows
5961 // an allocation, with no intervening tests or other escapes for the object.
5962 AllocateArrayNode*
5963 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
5964 if (stopped()) return nullptr; // no fast path
5965 if (!C->do_aliasing()) return nullptr; // no MergeMems around
5966
5967 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
5968 if (alloc == nullptr) return nullptr;
5969
5970 Node* rawmem = memory(Compile::AliasIdxRaw);
5971 // Is the allocation's memory state untouched?
5972 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5973 // Bail out if there have been raw-memory effects since the allocation.
5974 // (Example: There might have been a call or safepoint.)
5975 return nullptr;
5976 }
5977 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5978 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5979 return nullptr;
5980 }
5981
5982 // There must be no unexpected observers of this allocation.
5983 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5984 Node* obs = ptr->fast_out(i);
5985 if (obs != this->map()) {
5986 return nullptr;
5987 }
5988 }
5989
5990 // This arraycopy must unconditionally follow the allocation of the ptr.
5991 Node* alloc_ctl = ptr->in(0);
5992 Node* ctl = control();
5993 while (ctl != alloc_ctl) {
5994 // There may be guards which feed into the slow_region.
5995 // Any other control flow means that we might not get a chance
5996 // to finish initializing the allocated object.
5997 // Various low-level checks bottom out in uncommon traps. These
5998 // are considered safe since we've already checked above that
5999 // there is no unexpected observer of this allocation.
6000 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6001 assert(ctl->in(0)->is_If(), "must be If");
6002 ctl = ctl->in(0)->in(0);
6003 } else {
6004 return nullptr;
6005 }
6006 }
6007
6008 // If we get this far, we have an allocation which immediately
6009 // precedes the arraycopy, and we can take over zeroing the new object.
6010 // The arraycopy will finish the initialization, and provide
6011 // a new control state to which we will anchor the destination pointer.
6012
6013 return alloc;
6014 }
6015
6016 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6017 if (node->is_IfProj()) {
6018 Node* other_proj = node->as_IfProj()->other_if_proj();
6019 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6020 Node* obs = other_proj->fast_out(j);
6021 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6022 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6023 return obs->as_CallStaticJava();
6024 }
6025 }
6026 }
6027 return nullptr;
6028 }
6029
6030 //-------------inline_encodeISOArray-----------------------------------
6031 // encode char[] to byte[] in ISO_8859_1 or ASCII
6032 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6033 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6034 // no receiver since it is static method
6035 Node *src = argument(0);
6036 Node *src_offset = argument(1);
6037 Node *dst = argument(2);
6038 Node *dst_offset = argument(3);
6039 Node *length = argument(4);
6040
6041 src = must_be_not_null(src, true);
6042 dst = must_be_not_null(dst, true);
6043
6044 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6045 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6046 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6047 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6048 // failed array check
6049 return false;
6050 }
6051
6052 // Figure out the size and type of the elements we will be copying.
6053 BasicType src_elem = src_type->elem()->array_element_basic_type();
6054 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6055 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6056 return false;
6057 }
6058
6059 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6060 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6061 // 'src_start' points to src array + scaled offset
6062 // 'dst_start' points to dst array + scaled offset
6063
6064 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6065 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6066 enc = _gvn.transform(enc);
6067 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6068 set_memory(res_mem, mtype);
6069 set_result(enc);
6070 clear_upper_avx();
6071
6072 return true;
6073 }
6074
6075 //-------------inline_multiplyToLen-----------------------------------
6076 bool LibraryCallKit::inline_multiplyToLen() {
6077 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6078
6079 address stubAddr = StubRoutines::multiplyToLen();
6080 if (stubAddr == nullptr) {
6081 return false; // Intrinsic's stub is not implemented on this platform
6082 }
6083 const char* stubName = "multiplyToLen";
6084
6085 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6086
6087 // no receiver because it is a static method
6088 Node* x = argument(0);
6089 Node* xlen = argument(1);
6090 Node* y = argument(2);
6091 Node* ylen = argument(3);
6092 Node* z = argument(4);
6093
6094 x = must_be_not_null(x, true);
6095 y = must_be_not_null(y, true);
6096
6097 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6098 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6099 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6100 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6101 // failed array check
6102 return false;
6103 }
6104
6105 BasicType x_elem = x_type->elem()->array_element_basic_type();
6106 BasicType y_elem = y_type->elem()->array_element_basic_type();
6107 if (x_elem != T_INT || y_elem != T_INT) {
6108 return false;
6109 }
6110
6111 Node* x_start = array_element_address(x, intcon(0), x_elem);
6112 Node* y_start = array_element_address(y, intcon(0), y_elem);
6113 // 'x_start' points to x array + scaled xlen
6114 // 'y_start' points to y array + scaled ylen
6115
6116 Node* z_start = array_element_address(z, intcon(0), T_INT);
6117
6118 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6119 OptoRuntime::multiplyToLen_Type(),
6120 stubAddr, stubName, TypePtr::BOTTOM,
6121 x_start, xlen, y_start, ylen, z_start);
6122
6123 C->set_has_split_ifs(true); // Has chance for split-if optimization
6124 set_result(z);
6125 return true;
6126 }
6127
6128 //-------------inline_squareToLen------------------------------------
6129 bool LibraryCallKit::inline_squareToLen() {
6130 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6131
6132 address stubAddr = StubRoutines::squareToLen();
6133 if (stubAddr == nullptr) {
6134 return false; // Intrinsic's stub is not implemented on this platform
6135 }
6136 const char* stubName = "squareToLen";
6137
6138 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6139
6140 Node* x = argument(0);
6141 Node* len = argument(1);
6142 Node* z = argument(2);
6143 Node* zlen = argument(3);
6144
6145 x = must_be_not_null(x, true);
6146 z = must_be_not_null(z, true);
6147
6148 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6149 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6150 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6151 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6152 // failed array check
6153 return false;
6154 }
6155
6156 BasicType x_elem = x_type->elem()->array_element_basic_type();
6157 BasicType z_elem = z_type->elem()->array_element_basic_type();
6158 if (x_elem != T_INT || z_elem != T_INT) {
6159 return false;
6160 }
6161
6162
6163 Node* x_start = array_element_address(x, intcon(0), x_elem);
6164 Node* z_start = array_element_address(z, intcon(0), z_elem);
6165
6166 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6167 OptoRuntime::squareToLen_Type(),
6168 stubAddr, stubName, TypePtr::BOTTOM,
6169 x_start, len, z_start, zlen);
6170
6171 set_result(z);
6172 return true;
6173 }
6174
6175 //-------------inline_mulAdd------------------------------------------
6176 bool LibraryCallKit::inline_mulAdd() {
6177 assert(UseMulAddIntrinsic, "not implemented on this platform");
6178
6179 address stubAddr = StubRoutines::mulAdd();
6180 if (stubAddr == nullptr) {
6181 return false; // Intrinsic's stub is not implemented on this platform
6182 }
6183 const char* stubName = "mulAdd";
6184
6185 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6186
6187 Node* out = argument(0);
6188 Node* in = argument(1);
6189 Node* offset = argument(2);
6190 Node* len = argument(3);
6191 Node* k = argument(4);
6192
6193 in = must_be_not_null(in, true);
6194 out = must_be_not_null(out, true);
6195
6196 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6197 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6198 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6199 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6200 // failed array check
6201 return false;
6202 }
6203
6204 BasicType out_elem = out_type->elem()->array_element_basic_type();
6205 BasicType in_elem = in_type->elem()->array_element_basic_type();
6206 if (out_elem != T_INT || in_elem != T_INT) {
6207 return false;
6208 }
6209
6210 Node* outlen = load_array_length(out);
6211 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6212 Node* out_start = array_element_address(out, intcon(0), out_elem);
6213 Node* in_start = array_element_address(in, intcon(0), in_elem);
6214
6215 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6216 OptoRuntime::mulAdd_Type(),
6217 stubAddr, stubName, TypePtr::BOTTOM,
6218 out_start,in_start, new_offset, len, k);
6219 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6220 set_result(result);
6221 return true;
6222 }
6223
6224 //-------------inline_montgomeryMultiply-----------------------------------
6225 bool LibraryCallKit::inline_montgomeryMultiply() {
6226 address stubAddr = StubRoutines::montgomeryMultiply();
6227 if (stubAddr == nullptr) {
6228 return false; // Intrinsic's stub is not implemented on this platform
6229 }
6230
6231 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6232 const char* stubName = "montgomery_multiply";
6233
6234 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6235
6236 Node* a = argument(0);
6237 Node* b = argument(1);
6238 Node* n = argument(2);
6239 Node* len = argument(3);
6240 Node* inv = argument(4);
6241 Node* m = argument(6);
6242
6243 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6244 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6245 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6246 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6247 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6248 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6249 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6250 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6251 // failed array check
6252 return false;
6253 }
6254
6255 BasicType a_elem = a_type->elem()->array_element_basic_type();
6256 BasicType b_elem = b_type->elem()->array_element_basic_type();
6257 BasicType n_elem = n_type->elem()->array_element_basic_type();
6258 BasicType m_elem = m_type->elem()->array_element_basic_type();
6259 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6260 return false;
6261 }
6262
6263 // Make the call
6264 {
6265 Node* a_start = array_element_address(a, intcon(0), a_elem);
6266 Node* b_start = array_element_address(b, intcon(0), b_elem);
6267 Node* n_start = array_element_address(n, intcon(0), n_elem);
6268 Node* m_start = array_element_address(m, intcon(0), m_elem);
6269
6270 Node* call = make_runtime_call(RC_LEAF,
6271 OptoRuntime::montgomeryMultiply_Type(),
6272 stubAddr, stubName, TypePtr::BOTTOM,
6273 a_start, b_start, n_start, len, inv, top(),
6274 m_start);
6275 set_result(m);
6276 }
6277
6278 return true;
6279 }
6280
6281 bool LibraryCallKit::inline_montgomerySquare() {
6282 address stubAddr = StubRoutines::montgomerySquare();
6283 if (stubAddr == nullptr) {
6284 return false; // Intrinsic's stub is not implemented on this platform
6285 }
6286
6287 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6288 const char* stubName = "montgomery_square";
6289
6290 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6291
6292 Node* a = argument(0);
6293 Node* n = argument(1);
6294 Node* len = argument(2);
6295 Node* inv = argument(3);
6296 Node* m = argument(5);
6297
6298 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6299 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6300 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6301 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6302 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6303 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6304 // failed array check
6305 return false;
6306 }
6307
6308 BasicType a_elem = a_type->elem()->array_element_basic_type();
6309 BasicType n_elem = n_type->elem()->array_element_basic_type();
6310 BasicType m_elem = m_type->elem()->array_element_basic_type();
6311 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6312 return false;
6313 }
6314
6315 // Make the call
6316 {
6317 Node* a_start = array_element_address(a, intcon(0), a_elem);
6318 Node* n_start = array_element_address(n, intcon(0), n_elem);
6319 Node* m_start = array_element_address(m, intcon(0), m_elem);
6320
6321 Node* call = make_runtime_call(RC_LEAF,
6322 OptoRuntime::montgomerySquare_Type(),
6323 stubAddr, stubName, TypePtr::BOTTOM,
6324 a_start, n_start, len, inv, top(),
6325 m_start);
6326 set_result(m);
6327 }
6328
6329 return true;
6330 }
6331
6332 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6333 address stubAddr = nullptr;
6334 const char* stubName = nullptr;
6335
6336 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6337 if (stubAddr == nullptr) {
6338 return false; // Intrinsic's stub is not implemented on this platform
6339 }
6340
6341 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6342
6343 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6344
6345 Node* newArr = argument(0);
6346 Node* oldArr = argument(1);
6347 Node* newIdx = argument(2);
6348 Node* shiftCount = argument(3);
6349 Node* numIter = argument(4);
6350
6351 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6352 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6353 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6354 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6355 return false;
6356 }
6357
6358 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6359 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6360 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6361 return false;
6362 }
6363
6364 // Make the call
6365 {
6366 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6367 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6368
6369 Node* call = make_runtime_call(RC_LEAF,
6370 OptoRuntime::bigIntegerShift_Type(),
6371 stubAddr,
6372 stubName,
6373 TypePtr::BOTTOM,
6374 newArr_start,
6375 oldArr_start,
6376 newIdx,
6377 shiftCount,
6378 numIter);
6379 }
6380
6381 return true;
6382 }
6383
6384 //-------------inline_vectorizedMismatch------------------------------
6385 bool LibraryCallKit::inline_vectorizedMismatch() {
6386 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6387
6388 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6389 Node* obja = argument(0); // Object
6390 Node* aoffset = argument(1); // long
6391 Node* objb = argument(3); // Object
6392 Node* boffset = argument(4); // long
6393 Node* length = argument(6); // int
6394 Node* scale = argument(7); // int
6395
6396 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6397 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6398 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6399 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6400 scale == top()) {
6401 return false; // failed input validation
6402 }
6403
6404 Node* obja_adr = make_unsafe_address(obja, aoffset);
6405 Node* objb_adr = make_unsafe_address(objb, boffset);
6406
6407 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6408 //
6409 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6410 // if (length <= inline_limit) {
6411 // inline_path:
6412 // vmask = VectorMaskGen length
6413 // vload1 = LoadVectorMasked obja, vmask
6414 // vload2 = LoadVectorMasked objb, vmask
6415 // result1 = VectorCmpMasked vload1, vload2, vmask
6416 // } else {
6417 // call_stub_path:
6418 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6419 // }
6420 // exit_block:
6421 // return Phi(result1, result2);
6422 //
6423 enum { inline_path = 1, // input is small enough to process it all at once
6424 stub_path = 2, // input is too large; call into the VM
6425 PATH_LIMIT = 3
6426 };
6427
6428 Node* exit_block = new RegionNode(PATH_LIMIT);
6429 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6430 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6431
6432 Node* call_stub_path = control();
6433
6434 BasicType elem_bt = T_ILLEGAL;
6435
6436 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6437 if (scale_t->is_con()) {
6438 switch (scale_t->get_con()) {
6439 case 0: elem_bt = T_BYTE; break;
6440 case 1: elem_bt = T_SHORT; break;
6441 case 2: elem_bt = T_INT; break;
6442 case 3: elem_bt = T_LONG; break;
6443
6444 default: elem_bt = T_ILLEGAL; break; // not supported
6445 }
6446 }
6447
6448 int inline_limit = 0;
6449 bool do_partial_inline = false;
6450
6451 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6452 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6453 do_partial_inline = inline_limit >= 16;
6454 }
6455
6456 if (do_partial_inline) {
6457 assert(elem_bt != T_ILLEGAL, "sanity");
6458
6459 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6460 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6461 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6462
6463 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6464 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6465 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6466
6467 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6468
6469 if (!stopped()) {
6470 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6471
6472 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6473 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6474 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6475 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6476
6477 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6478 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6479 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6480 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6481
6482 exit_block->init_req(inline_path, control());
6483 memory_phi->init_req(inline_path, map()->memory());
6484 result_phi->init_req(inline_path, result);
6485
6486 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6487 clear_upper_avx();
6488 }
6489 }
6490 }
6491
6492 if (call_stub_path != nullptr) {
6493 set_control(call_stub_path);
6494
6495 Node* call = make_runtime_call(RC_LEAF,
6496 OptoRuntime::vectorizedMismatch_Type(),
6497 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6498 obja_adr, objb_adr, length, scale);
6499
6500 exit_block->init_req(stub_path, control());
6501 memory_phi->init_req(stub_path, map()->memory());
6502 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6503 }
6504
6505 exit_block = _gvn.transform(exit_block);
6506 memory_phi = _gvn.transform(memory_phi);
6507 result_phi = _gvn.transform(result_phi);
6508
6509 set_control(exit_block);
6510 set_all_memory(memory_phi);
6511 set_result(result_phi);
6512
6513 return true;
6514 }
6515
6516 //------------------------------inline_vectorizedHashcode----------------------------
6517 bool LibraryCallKit::inline_vectorizedHashCode() {
6518 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6519
6520 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6521 Node* array = argument(0);
6522 Node* offset = argument(1);
6523 Node* length = argument(2);
6524 Node* initialValue = argument(3);
6525 Node* basic_type = argument(4);
6526
6527 if (basic_type == top()) {
6528 return false; // failed input validation
6529 }
6530
6531 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6532 if (!basic_type_t->is_con()) {
6533 return false; // Only intrinsify if mode argument is constant
6534 }
6535
6536 array = must_be_not_null(array, true);
6537
6538 BasicType bt = (BasicType)basic_type_t->get_con();
6539
6540 // Resolve address of first element
6541 Node* array_start = array_element_address(array, offset, bt);
6542
6543 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6544 array_start, length, initialValue, basic_type)));
6545 clear_upper_avx();
6546
6547 return true;
6548 }
6549
6550 /**
6551 * Calculate CRC32 for byte.
6552 * int java.util.zip.CRC32.update(int crc, int b)
6553 */
6554 bool LibraryCallKit::inline_updateCRC32() {
6555 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6556 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6557 // no receiver since it is static method
6558 Node* crc = argument(0); // type: int
6559 Node* b = argument(1); // type: int
6560
6561 /*
6562 * int c = ~ crc;
6563 * b = timesXtoThe32[(b ^ c) & 0xFF];
6564 * b = b ^ (c >>> 8);
6565 * crc = ~b;
6566 */
6567
6568 Node* M1 = intcon(-1);
6569 crc = _gvn.transform(new XorINode(crc, M1));
6570 Node* result = _gvn.transform(new XorINode(crc, b));
6571 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6572
6573 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6574 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6575 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6576 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6577
6578 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6579 result = _gvn.transform(new XorINode(crc, result));
6580 result = _gvn.transform(new XorINode(result, M1));
6581 set_result(result);
6582 return true;
6583 }
6584
6585 /**
6586 * Calculate CRC32 for byte[] array.
6587 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6588 */
6589 bool LibraryCallKit::inline_updateBytesCRC32() {
6590 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6591 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6592 // no receiver since it is static method
6593 Node* crc = argument(0); // type: int
6594 Node* src = argument(1); // type: oop
6595 Node* offset = argument(2); // type: int
6596 Node* length = argument(3); // type: int
6597
6598 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6599 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6600 // failed array check
6601 return false;
6602 }
6603
6604 // Figure out the size and type of the elements we will be copying.
6605 BasicType src_elem = src_type->elem()->array_element_basic_type();
6606 if (src_elem != T_BYTE) {
6607 return false;
6608 }
6609
6610 // 'src_start' points to src array + scaled offset
6611 src = must_be_not_null(src, true);
6612 Node* src_start = array_element_address(src, offset, src_elem);
6613
6614 // We assume that range check is done by caller.
6615 // TODO: generate range check (offset+length < src.length) in debug VM.
6616
6617 // Call the stub.
6618 address stubAddr = StubRoutines::updateBytesCRC32();
6619 const char *stubName = "updateBytesCRC32";
6620
6621 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6622 stubAddr, stubName, TypePtr::BOTTOM,
6623 crc, src_start, length);
6624 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6625 set_result(result);
6626 return true;
6627 }
6628
6629 /**
6630 * Calculate CRC32 for ByteBuffer.
6631 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6632 */
6633 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6634 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6635 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6636 // no receiver since it is static method
6637 Node* crc = argument(0); // type: int
6638 Node* src = argument(1); // type: long
6639 Node* offset = argument(3); // type: int
6640 Node* length = argument(4); // type: int
6641
6642 src = ConvL2X(src); // adjust Java long to machine word
6643 Node* base = _gvn.transform(new CastX2PNode(src));
6644 offset = ConvI2X(offset);
6645
6646 // 'src_start' points to src array + scaled offset
6647 Node* src_start = basic_plus_adr(top(), base, offset);
6648
6649 // Call the stub.
6650 address stubAddr = StubRoutines::updateBytesCRC32();
6651 const char *stubName = "updateBytesCRC32";
6652
6653 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6654 stubAddr, stubName, TypePtr::BOTTOM,
6655 crc, src_start, length);
6656 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6657 set_result(result);
6658 return true;
6659 }
6660
6661 //------------------------------get_table_from_crc32c_class-----------------------
6662 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6663 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6664 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6665
6666 return table;
6667 }
6668
6669 //------------------------------inline_updateBytesCRC32C-----------------------
6670 //
6671 // Calculate CRC32C for byte[] array.
6672 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6673 //
6674 bool LibraryCallKit::inline_updateBytesCRC32C() {
6675 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6676 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6677 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6678 // no receiver since it is a static method
6679 Node* crc = argument(0); // type: int
6680 Node* src = argument(1); // type: oop
6681 Node* offset = argument(2); // type: int
6682 Node* end = argument(3); // type: int
6683
6684 Node* length = _gvn.transform(new SubINode(end, offset));
6685
6686 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6687 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6688 // failed array check
6689 return false;
6690 }
6691
6692 // Figure out the size and type of the elements we will be copying.
6693 BasicType src_elem = src_type->elem()->array_element_basic_type();
6694 if (src_elem != T_BYTE) {
6695 return false;
6696 }
6697
6698 // 'src_start' points to src array + scaled offset
6699 src = must_be_not_null(src, true);
6700 Node* src_start = array_element_address(src, offset, src_elem);
6701
6702 // static final int[] byteTable in class CRC32C
6703 Node* table = get_table_from_crc32c_class(callee()->holder());
6704 table = must_be_not_null(table, true);
6705 Node* table_start = array_element_address(table, intcon(0), T_INT);
6706
6707 // We assume that range check is done by caller.
6708 // TODO: generate range check (offset+length < src.length) in debug VM.
6709
6710 // Call the stub.
6711 address stubAddr = StubRoutines::updateBytesCRC32C();
6712 const char *stubName = "updateBytesCRC32C";
6713
6714 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6715 stubAddr, stubName, TypePtr::BOTTOM,
6716 crc, src_start, length, table_start);
6717 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6718 set_result(result);
6719 return true;
6720 }
6721
6722 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6723 //
6724 // Calculate CRC32C for DirectByteBuffer.
6725 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6726 //
6727 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6728 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6729 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6730 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6731 // no receiver since it is a static method
6732 Node* crc = argument(0); // type: int
6733 Node* src = argument(1); // type: long
6734 Node* offset = argument(3); // type: int
6735 Node* end = argument(4); // type: int
6736
6737 Node* length = _gvn.transform(new SubINode(end, offset));
6738
6739 src = ConvL2X(src); // adjust Java long to machine word
6740 Node* base = _gvn.transform(new CastX2PNode(src));
6741 offset = ConvI2X(offset);
6742
6743 // 'src_start' points to src array + scaled offset
6744 Node* src_start = basic_plus_adr(top(), base, offset);
6745
6746 // static final int[] byteTable in class CRC32C
6747 Node* table = get_table_from_crc32c_class(callee()->holder());
6748 table = must_be_not_null(table, true);
6749 Node* table_start = array_element_address(table, intcon(0), T_INT);
6750
6751 // Call the stub.
6752 address stubAddr = StubRoutines::updateBytesCRC32C();
6753 const char *stubName = "updateBytesCRC32C";
6754
6755 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6756 stubAddr, stubName, TypePtr::BOTTOM,
6757 crc, src_start, length, table_start);
6758 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6759 set_result(result);
6760 return true;
6761 }
6762
6763 //------------------------------inline_updateBytesAdler32----------------------
6764 //
6765 // Calculate Adler32 checksum for byte[] array.
6766 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6767 //
6768 bool LibraryCallKit::inline_updateBytesAdler32() {
6769 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6770 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6771 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6772 // no receiver since it is static method
6773 Node* crc = argument(0); // type: int
6774 Node* src = argument(1); // type: oop
6775 Node* offset = argument(2); // type: int
6776 Node* length = argument(3); // type: int
6777
6778 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6779 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6780 // failed array check
6781 return false;
6782 }
6783
6784 // Figure out the size and type of the elements we will be copying.
6785 BasicType src_elem = src_type->elem()->array_element_basic_type();
6786 if (src_elem != T_BYTE) {
6787 return false;
6788 }
6789
6790 // 'src_start' points to src array + scaled offset
6791 Node* src_start = array_element_address(src, offset, src_elem);
6792
6793 // We assume that range check is done by caller.
6794 // TODO: generate range check (offset+length < src.length) in debug VM.
6795
6796 // Call the stub.
6797 address stubAddr = StubRoutines::updateBytesAdler32();
6798 const char *stubName = "updateBytesAdler32";
6799
6800 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6801 stubAddr, stubName, TypePtr::BOTTOM,
6802 crc, src_start, length);
6803 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6804 set_result(result);
6805 return true;
6806 }
6807
6808 //------------------------------inline_updateByteBufferAdler32---------------
6809 //
6810 // Calculate Adler32 checksum for DirectByteBuffer.
6811 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6812 //
6813 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6814 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6815 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6816 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6817 // no receiver since it is static method
6818 Node* crc = argument(0); // type: int
6819 Node* src = argument(1); // type: long
6820 Node* offset = argument(3); // type: int
6821 Node* length = argument(4); // type: int
6822
6823 src = ConvL2X(src); // adjust Java long to machine word
6824 Node* base = _gvn.transform(new CastX2PNode(src));
6825 offset = ConvI2X(offset);
6826
6827 // 'src_start' points to src array + scaled offset
6828 Node* src_start = basic_plus_adr(top(), base, offset);
6829
6830 // Call the stub.
6831 address stubAddr = StubRoutines::updateBytesAdler32();
6832 const char *stubName = "updateBytesAdler32";
6833
6834 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6835 stubAddr, stubName, TypePtr::BOTTOM,
6836 crc, src_start, length);
6837
6838 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6839 set_result(result);
6840 return true;
6841 }
6842
6843 //----------------------------inline_reference_get----------------------------
6844 // public T java.lang.ref.Reference.get();
6845 bool LibraryCallKit::inline_reference_get() {
6846 const int referent_offset = java_lang_ref_Reference::referent_offset();
6847
6848 // Get the argument:
6849 Node* reference_obj = null_check_receiver();
6850 if (stopped()) return true;
6851
6852 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6853 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6854 decorators, /*is_static*/ false, nullptr);
6855 if (result == nullptr) return false;
6856
6857 // Add memory barrier to prevent commoning reads from this field
6858 // across safepoint since GC can change its value.
6859 insert_mem_bar(Op_MemBarCPUOrder);
6860
6861 set_result(result);
6862 return true;
6863 }
6864
6865 //----------------------------inline_reference_refersTo0----------------------------
6866 // bool java.lang.ref.Reference.refersTo0();
6867 // bool java.lang.ref.PhantomReference.refersTo0();
6868 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6869 // Get arguments:
6870 Node* reference_obj = null_check_receiver();
6871 Node* other_obj = argument(1);
6872 if (stopped()) return true;
6873
6874 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6875 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6876 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6877 decorators, /*is_static*/ false, nullptr);
6878 if (referent == nullptr) return false;
6879
6880 // Add memory barrier to prevent commoning reads from this field
6881 // across safepoint since GC can change its value.
6882 insert_mem_bar(Op_MemBarCPUOrder);
6883
6884 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
6885 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6886 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
6887
6888 RegionNode* region = new RegionNode(3);
6889 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
6890
6891 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
6892 region->init_req(1, if_true);
6893 phi->init_req(1, intcon(1));
6894
6895 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
6896 region->init_req(2, if_false);
6897 phi->init_req(2, intcon(0));
6898
6899 set_control(_gvn.transform(region));
6900 record_for_igvn(region);
6901 set_result(_gvn.transform(phi));
6902 return true;
6903 }
6904
6905 //----------------------------inline_reference_clear0----------------------------
6906 // void java.lang.ref.Reference.clear0();
6907 // void java.lang.ref.PhantomReference.clear0();
6908 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
6909 // This matches the implementation in JVM_ReferenceClear, see the comments there.
6910
6911 // Get arguments
6912 Node* reference_obj = null_check_receiver();
6913 if (stopped()) return true;
6914
6915 // Common access parameters
6916 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6917 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6918 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
6919 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
6920 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
6921
6922 Node* referent = access_load_at(reference_obj,
6923 referent_field_addr,
6924 referent_field_addr_type,
6925 val_type,
6926 T_OBJECT,
6927 decorators);
6928
6929 IdealKit ideal(this);
6930 #define __ ideal.
6931 __ if_then(referent, BoolTest::ne, null());
6932 sync_kit(ideal);
6933 access_store_at(reference_obj,
6934 referent_field_addr,
6935 referent_field_addr_type,
6936 null(),
6937 val_type,
6938 T_OBJECT,
6939 decorators);
6940 __ sync_kit(this);
6941 __ end_if();
6942 final_sync(ideal);
6943 #undef __
6944
6945 return true;
6946 }
6947
6948 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
6949 DecoratorSet decorators, bool is_static,
6950 ciInstanceKlass* fromKls) {
6951 if (fromKls == nullptr) {
6952 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6953 assert(tinst != nullptr, "obj is null");
6954 assert(tinst->is_loaded(), "obj is not loaded");
6955 fromKls = tinst->instance_klass();
6956 } else {
6957 assert(is_static, "only for static field access");
6958 }
6959 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6960 ciSymbol::make(fieldTypeString),
6961 is_static);
6962
6963 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
6964 if (field == nullptr) return (Node *) nullptr;
6965
6966 if (is_static) {
6967 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6968 fromObj = makecon(tip);
6969 }
6970
6971 // Next code copied from Parse::do_get_xxx():
6972
6973 // Compute address and memory type.
6974 int offset = field->offset_in_bytes();
6975 bool is_vol = field->is_volatile();
6976 ciType* field_klass = field->type();
6977 assert(field_klass->is_loaded(), "should be loaded");
6978 const TypePtr* adr_type = C->alias_type(field)->adr_type();
6979 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6980 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
6981 "slice of address and input slice don't match");
6982 BasicType bt = field->layout_type();
6983
6984 // Build the resultant type of the load
6985 const Type *type;
6986 if (bt == T_OBJECT) {
6987 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
6988 } else {
6989 type = Type::get_const_basic_type(bt);
6990 }
6991
6992 if (is_vol) {
6993 decorators |= MO_SEQ_CST;
6994 }
6995
6996 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
6997 }
6998
6999 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7000 bool is_exact /* true */, bool is_static /* false */,
7001 ciInstanceKlass * fromKls /* nullptr */) {
7002 if (fromKls == nullptr) {
7003 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7004 assert(tinst != nullptr, "obj is null");
7005 assert(tinst->is_loaded(), "obj is not loaded");
7006 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7007 fromKls = tinst->instance_klass();
7008 }
7009 else {
7010 assert(is_static, "only for static field access");
7011 }
7012 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7013 ciSymbol::make(fieldTypeString),
7014 is_static);
7015
7016 assert(field != nullptr, "undefined field");
7017 assert(!field->is_volatile(), "not defined for volatile fields");
7018
7019 if (is_static) {
7020 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7021 fromObj = makecon(tip);
7022 }
7023
7024 // Next code copied from Parse::do_get_xxx():
7025
7026 // Compute address and memory type.
7027 int offset = field->offset_in_bytes();
7028 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7029
7030 return adr;
7031 }
7032
7033 //------------------------------inline_aescrypt_Block-----------------------
7034 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7035 address stubAddr = nullptr;
7036 const char *stubName;
7037 assert(UseAES, "need AES instruction support");
7038
7039 switch(id) {
7040 case vmIntrinsics::_aescrypt_encryptBlock:
7041 stubAddr = StubRoutines::aescrypt_encryptBlock();
7042 stubName = "aescrypt_encryptBlock";
7043 break;
7044 case vmIntrinsics::_aescrypt_decryptBlock:
7045 stubAddr = StubRoutines::aescrypt_decryptBlock();
7046 stubName = "aescrypt_decryptBlock";
7047 break;
7048 default:
7049 break;
7050 }
7051 if (stubAddr == nullptr) return false;
7052
7053 Node* aescrypt_object = argument(0);
7054 Node* src = argument(1);
7055 Node* src_offset = argument(2);
7056 Node* dest = argument(3);
7057 Node* dest_offset = argument(4);
7058
7059 src = must_be_not_null(src, true);
7060 dest = must_be_not_null(dest, true);
7061
7062 // (1) src and dest are arrays.
7063 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7064 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7065 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7066 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7067
7068 // for the quick and dirty code we will skip all the checks.
7069 // we are just trying to get the call to be generated.
7070 Node* src_start = src;
7071 Node* dest_start = dest;
7072 if (src_offset != nullptr || dest_offset != nullptr) {
7073 assert(src_offset != nullptr && dest_offset != nullptr, "");
7074 src_start = array_element_address(src, src_offset, T_BYTE);
7075 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7076 }
7077
7078 // now need to get the start of its expanded key array
7079 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7080 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7081 if (k_start == nullptr) return false;
7082
7083 // Call the stub.
7084 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7085 stubAddr, stubName, TypePtr::BOTTOM,
7086 src_start, dest_start, k_start);
7087
7088 return true;
7089 }
7090
7091 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7092 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7093 address stubAddr = nullptr;
7094 const char *stubName = nullptr;
7095
7096 assert(UseAES, "need AES instruction support");
7097
7098 switch(id) {
7099 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7100 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7101 stubName = "cipherBlockChaining_encryptAESCrypt";
7102 break;
7103 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7104 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7105 stubName = "cipherBlockChaining_decryptAESCrypt";
7106 break;
7107 default:
7108 break;
7109 }
7110 if (stubAddr == nullptr) return false;
7111
7112 Node* cipherBlockChaining_object = argument(0);
7113 Node* src = argument(1);
7114 Node* src_offset = argument(2);
7115 Node* len = argument(3);
7116 Node* dest = argument(4);
7117 Node* dest_offset = argument(5);
7118
7119 src = must_be_not_null(src, false);
7120 dest = must_be_not_null(dest, false);
7121
7122 // (1) src and dest are arrays.
7123 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7124 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7125 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7126 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7127
7128 // checks are the responsibility of the caller
7129 Node* src_start = src;
7130 Node* dest_start = dest;
7131 if (src_offset != nullptr || dest_offset != nullptr) {
7132 assert(src_offset != nullptr && dest_offset != nullptr, "");
7133 src_start = array_element_address(src, src_offset, T_BYTE);
7134 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7135 }
7136
7137 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7138 // (because of the predicated logic executed earlier).
7139 // so we cast it here safely.
7140 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7141
7142 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7143 if (embeddedCipherObj == nullptr) return false;
7144
7145 // cast it to what we know it will be at runtime
7146 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7147 assert(tinst != nullptr, "CBC obj is null");
7148 assert(tinst->is_loaded(), "CBC obj is not loaded");
7149 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7150 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7151
7152 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7153 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7154 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7155 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7156 aescrypt_object = _gvn.transform(aescrypt_object);
7157
7158 // we need to get the start of the aescrypt_object's expanded key array
7159 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7160 if (k_start == nullptr) return false;
7161
7162 // similarly, get the start address of the r vector
7163 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7164 if (objRvec == nullptr) return false;
7165 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7166
7167 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7168 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7169 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7170 stubAddr, stubName, TypePtr::BOTTOM,
7171 src_start, dest_start, k_start, r_start, len);
7172
7173 // return cipher length (int)
7174 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7175 set_result(retvalue);
7176 return true;
7177 }
7178
7179 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7180 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7181 address stubAddr = nullptr;
7182 const char *stubName = nullptr;
7183
7184 assert(UseAES, "need AES instruction support");
7185
7186 switch (id) {
7187 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7188 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7189 stubName = "electronicCodeBook_encryptAESCrypt";
7190 break;
7191 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7192 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7193 stubName = "electronicCodeBook_decryptAESCrypt";
7194 break;
7195 default:
7196 break;
7197 }
7198
7199 if (stubAddr == nullptr) return false;
7200
7201 Node* electronicCodeBook_object = argument(0);
7202 Node* src = argument(1);
7203 Node* src_offset = argument(2);
7204 Node* len = argument(3);
7205 Node* dest = argument(4);
7206 Node* dest_offset = argument(5);
7207
7208 // (1) src and dest are arrays.
7209 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7210 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7211 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7212 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7213
7214 // checks are the responsibility of the caller
7215 Node* src_start = src;
7216 Node* dest_start = dest;
7217 if (src_offset != nullptr || dest_offset != nullptr) {
7218 assert(src_offset != nullptr && dest_offset != nullptr, "");
7219 src_start = array_element_address(src, src_offset, T_BYTE);
7220 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7221 }
7222
7223 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7224 // (because of the predicated logic executed earlier).
7225 // so we cast it here safely.
7226 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7227
7228 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7229 if (embeddedCipherObj == nullptr) return false;
7230
7231 // cast it to what we know it will be at runtime
7232 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7233 assert(tinst != nullptr, "ECB obj is null");
7234 assert(tinst->is_loaded(), "ECB obj is not loaded");
7235 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7236 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7237
7238 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7239 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7240 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7241 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7242 aescrypt_object = _gvn.transform(aescrypt_object);
7243
7244 // we need to get the start of the aescrypt_object's expanded key array
7245 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7246 if (k_start == nullptr) return false;
7247
7248 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7249 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7250 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7251 stubAddr, stubName, TypePtr::BOTTOM,
7252 src_start, dest_start, k_start, len);
7253
7254 // return cipher length (int)
7255 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7256 set_result(retvalue);
7257 return true;
7258 }
7259
7260 //------------------------------inline_counterMode_AESCrypt-----------------------
7261 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7262 assert(UseAES, "need AES instruction support");
7263 if (!UseAESCTRIntrinsics) return false;
7264
7265 address stubAddr = nullptr;
7266 const char *stubName = nullptr;
7267 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7268 stubAddr = StubRoutines::counterMode_AESCrypt();
7269 stubName = "counterMode_AESCrypt";
7270 }
7271 if (stubAddr == nullptr) return false;
7272
7273 Node* counterMode_object = argument(0);
7274 Node* src = argument(1);
7275 Node* src_offset = argument(2);
7276 Node* len = argument(3);
7277 Node* dest = argument(4);
7278 Node* dest_offset = argument(5);
7279
7280 // (1) src and dest are arrays.
7281 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7282 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7283 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7284 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7285
7286 // checks are the responsibility of the caller
7287 Node* src_start = src;
7288 Node* dest_start = dest;
7289 if (src_offset != nullptr || dest_offset != nullptr) {
7290 assert(src_offset != nullptr && dest_offset != nullptr, "");
7291 src_start = array_element_address(src, src_offset, T_BYTE);
7292 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7293 }
7294
7295 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7296 // (because of the predicated logic executed earlier).
7297 // so we cast it here safely.
7298 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7299 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7300 if (embeddedCipherObj == nullptr) return false;
7301 // cast it to what we know it will be at runtime
7302 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7303 assert(tinst != nullptr, "CTR obj is null");
7304 assert(tinst->is_loaded(), "CTR obj is not loaded");
7305 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7306 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7307 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7308 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7309 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7310 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7311 aescrypt_object = _gvn.transform(aescrypt_object);
7312 // we need to get the start of the aescrypt_object's expanded key array
7313 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7314 if (k_start == nullptr) return false;
7315 // similarly, get the start address of the r vector
7316 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7317 if (obj_counter == nullptr) return false;
7318 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7319
7320 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7321 if (saved_encCounter == nullptr) return false;
7322 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7323 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7324
7325 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7326 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7327 OptoRuntime::counterMode_aescrypt_Type(),
7328 stubAddr, stubName, TypePtr::BOTTOM,
7329 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7330
7331 // return cipher length (int)
7332 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7333 set_result(retvalue);
7334 return true;
7335 }
7336
7337 //------------------------------get_key_start_from_aescrypt_object-----------------------
7338 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7339 #if defined(PPC64) || defined(S390) || defined(RISCV64)
7340 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7341 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7342 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7343 // The ppc64 and riscv64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7344 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7345 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7346 if (objSessionK == nullptr) {
7347 return (Node *) nullptr;
7348 }
7349 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7350 #else
7351 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7352 #endif // PPC64
7353 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7354 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7355
7356 // now have the array, need to get the start address of the K array
7357 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7358 return k_start;
7359 }
7360
7361 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7362 // Return node representing slow path of predicate check.
7363 // the pseudo code we want to emulate with this predicate is:
7364 // for encryption:
7365 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7366 // for decryption:
7367 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7368 // note cipher==plain is more conservative than the original java code but that's OK
7369 //
7370 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7371 // The receiver was checked for null already.
7372 Node* objCBC = argument(0);
7373
7374 Node* src = argument(1);
7375 Node* dest = argument(4);
7376
7377 // Load embeddedCipher field of CipherBlockChaining object.
7378 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7379
7380 // get AESCrypt klass for instanceOf check
7381 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7382 // will have same classloader as CipherBlockChaining object
7383 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7384 assert(tinst != nullptr, "CBCobj is null");
7385 assert(tinst->is_loaded(), "CBCobj is not loaded");
7386
7387 // we want to do an instanceof comparison against the AESCrypt class
7388 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7389 if (!klass_AESCrypt->is_loaded()) {
7390 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7391 Node* ctrl = control();
7392 set_control(top()); // no regular fast path
7393 return ctrl;
7394 }
7395
7396 src = must_be_not_null(src, true);
7397 dest = must_be_not_null(dest, true);
7398
7399 // Resolve oops to stable for CmpP below.
7400 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7401
7402 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7403 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7404 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7405
7406 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7407
7408 // for encryption, we are done
7409 if (!decrypting)
7410 return instof_false; // even if it is null
7411
7412 // for decryption, we need to add a further check to avoid
7413 // taking the intrinsic path when cipher and plain are the same
7414 // see the original java code for why.
7415 RegionNode* region = new RegionNode(3);
7416 region->init_req(1, instof_false);
7417
7418 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7419 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7420 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7421 region->init_req(2, src_dest_conjoint);
7422
7423 record_for_igvn(region);
7424 return _gvn.transform(region);
7425 }
7426
7427 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7428 // Return node representing slow path of predicate check.
7429 // the pseudo code we want to emulate with this predicate is:
7430 // for encryption:
7431 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7432 // for decryption:
7433 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7434 // note cipher==plain is more conservative than the original java code but that's OK
7435 //
7436 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7437 // The receiver was checked for null already.
7438 Node* objECB = argument(0);
7439
7440 // Load embeddedCipher field of ElectronicCodeBook object.
7441 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7442
7443 // get AESCrypt klass for instanceOf check
7444 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7445 // will have same classloader as ElectronicCodeBook object
7446 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7447 assert(tinst != nullptr, "ECBobj is null");
7448 assert(tinst->is_loaded(), "ECBobj is not loaded");
7449
7450 // we want to do an instanceof comparison against the AESCrypt class
7451 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7452 if (!klass_AESCrypt->is_loaded()) {
7453 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7454 Node* ctrl = control();
7455 set_control(top()); // no regular fast path
7456 return ctrl;
7457 }
7458 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7459
7460 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7461 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7462 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7463
7464 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7465
7466 // for encryption, we are done
7467 if (!decrypting)
7468 return instof_false; // even if it is null
7469
7470 // for decryption, we need to add a further check to avoid
7471 // taking the intrinsic path when cipher and plain are the same
7472 // see the original java code for why.
7473 RegionNode* region = new RegionNode(3);
7474 region->init_req(1, instof_false);
7475 Node* src = argument(1);
7476 Node* dest = argument(4);
7477 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7478 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7479 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7480 region->init_req(2, src_dest_conjoint);
7481
7482 record_for_igvn(region);
7483 return _gvn.transform(region);
7484 }
7485
7486 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7487 // Return node representing slow path of predicate check.
7488 // the pseudo code we want to emulate with this predicate is:
7489 // for encryption:
7490 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7491 // for decryption:
7492 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7493 // note cipher==plain is more conservative than the original java code but that's OK
7494 //
7495
7496 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7497 // The receiver was checked for null already.
7498 Node* objCTR = argument(0);
7499
7500 // Load embeddedCipher field of CipherBlockChaining object.
7501 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7502
7503 // get AESCrypt klass for instanceOf check
7504 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7505 // will have same classloader as CipherBlockChaining object
7506 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7507 assert(tinst != nullptr, "CTRobj is null");
7508 assert(tinst->is_loaded(), "CTRobj is not loaded");
7509
7510 // we want to do an instanceof comparison against the AESCrypt class
7511 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7512 if (!klass_AESCrypt->is_loaded()) {
7513 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7514 Node* ctrl = control();
7515 set_control(top()); // no regular fast path
7516 return ctrl;
7517 }
7518
7519 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7520 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7521 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7522 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7523 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7524
7525 return instof_false; // even if it is null
7526 }
7527
7528 //------------------------------inline_ghash_processBlocks
7529 bool LibraryCallKit::inline_ghash_processBlocks() {
7530 address stubAddr;
7531 const char *stubName;
7532 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7533
7534 stubAddr = StubRoutines::ghash_processBlocks();
7535 stubName = "ghash_processBlocks";
7536
7537 Node* data = argument(0);
7538 Node* offset = argument(1);
7539 Node* len = argument(2);
7540 Node* state = argument(3);
7541 Node* subkeyH = argument(4);
7542
7543 state = must_be_not_null(state, true);
7544 subkeyH = must_be_not_null(subkeyH, true);
7545 data = must_be_not_null(data, true);
7546
7547 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7548 assert(state_start, "state is null");
7549 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7550 assert(subkeyH_start, "subkeyH is null");
7551 Node* data_start = array_element_address(data, offset, T_BYTE);
7552 assert(data_start, "data is null");
7553
7554 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7555 OptoRuntime::ghash_processBlocks_Type(),
7556 stubAddr, stubName, TypePtr::BOTTOM,
7557 state_start, subkeyH_start, data_start, len);
7558 return true;
7559 }
7560
7561 //------------------------------inline_chacha20Block
7562 bool LibraryCallKit::inline_chacha20Block() {
7563 address stubAddr;
7564 const char *stubName;
7565 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7566
7567 stubAddr = StubRoutines::chacha20Block();
7568 stubName = "chacha20Block";
7569
7570 Node* state = argument(0);
7571 Node* result = argument(1);
7572
7573 state = must_be_not_null(state, true);
7574 result = must_be_not_null(result, true);
7575
7576 Node* state_start = array_element_address(state, intcon(0), T_INT);
7577 assert(state_start, "state is null");
7578 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7579 assert(result_start, "result is null");
7580
7581 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7582 OptoRuntime::chacha20Block_Type(),
7583 stubAddr, stubName, TypePtr::BOTTOM,
7584 state_start, result_start);
7585 // return key stream length (int)
7586 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7587 set_result(retvalue);
7588 return true;
7589 }
7590
7591 bool LibraryCallKit::inline_base64_encodeBlock() {
7592 address stubAddr;
7593 const char *stubName;
7594 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7595 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
7596 stubAddr = StubRoutines::base64_encodeBlock();
7597 stubName = "encodeBlock";
7598
7599 if (!stubAddr) return false;
7600 Node* base64obj = argument(0);
7601 Node* src = argument(1);
7602 Node* offset = argument(2);
7603 Node* len = argument(3);
7604 Node* dest = argument(4);
7605 Node* dp = argument(5);
7606 Node* isURL = argument(6);
7607
7608 src = must_be_not_null(src, true);
7609 dest = must_be_not_null(dest, true);
7610
7611 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7612 assert(src_start, "source array is null");
7613 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7614 assert(dest_start, "destination array is null");
7615
7616 Node* base64 = make_runtime_call(RC_LEAF,
7617 OptoRuntime::base64_encodeBlock_Type(),
7618 stubAddr, stubName, TypePtr::BOTTOM,
7619 src_start, offset, len, dest_start, dp, isURL);
7620 return true;
7621 }
7622
7623 bool LibraryCallKit::inline_base64_decodeBlock() {
7624 address stubAddr;
7625 const char *stubName;
7626 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7627 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
7628 stubAddr = StubRoutines::base64_decodeBlock();
7629 stubName = "decodeBlock";
7630
7631 if (!stubAddr) return false;
7632 Node* base64obj = argument(0);
7633 Node* src = argument(1);
7634 Node* src_offset = argument(2);
7635 Node* len = argument(3);
7636 Node* dest = argument(4);
7637 Node* dest_offset = argument(5);
7638 Node* isURL = argument(6);
7639 Node* isMIME = argument(7);
7640
7641 src = must_be_not_null(src, true);
7642 dest = must_be_not_null(dest, true);
7643
7644 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7645 assert(src_start, "source array is null");
7646 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7647 assert(dest_start, "destination array is null");
7648
7649 Node* call = make_runtime_call(RC_LEAF,
7650 OptoRuntime::base64_decodeBlock_Type(),
7651 stubAddr, stubName, TypePtr::BOTTOM,
7652 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
7653 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7654 set_result(result);
7655 return true;
7656 }
7657
7658 bool LibraryCallKit::inline_poly1305_processBlocks() {
7659 address stubAddr;
7660 const char *stubName;
7661 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
7662 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
7663 stubAddr = StubRoutines::poly1305_processBlocks();
7664 stubName = "poly1305_processBlocks";
7665
7666 if (!stubAddr) return false;
7667 null_check_receiver(); // null-check receiver
7668 if (stopped()) return true;
7669
7670 Node* input = argument(1);
7671 Node* input_offset = argument(2);
7672 Node* len = argument(3);
7673 Node* alimbs = argument(4);
7674 Node* rlimbs = argument(5);
7675
7676 input = must_be_not_null(input, true);
7677 alimbs = must_be_not_null(alimbs, true);
7678 rlimbs = must_be_not_null(rlimbs, true);
7679
7680 Node* input_start = array_element_address(input, input_offset, T_BYTE);
7681 assert(input_start, "input array is null");
7682 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
7683 assert(acc_start, "acc array is null");
7684 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
7685 assert(r_start, "r array is null");
7686
7687 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7688 OptoRuntime::poly1305_processBlocks_Type(),
7689 stubAddr, stubName, TypePtr::BOTTOM,
7690 input_start, len, acc_start, r_start);
7691 return true;
7692 }
7693
7694 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
7695 address stubAddr;
7696 const char *stubName;
7697 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
7698 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
7699 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
7700 stubName = "intpoly_montgomeryMult_P256";
7701
7702 if (!stubAddr) return false;
7703 null_check_receiver(); // null-check receiver
7704 if (stopped()) return true;
7705
7706 Node* a = argument(1);
7707 Node* b = argument(2);
7708 Node* r = argument(3);
7709
7710 a = must_be_not_null(a, true);
7711 b = must_be_not_null(b, true);
7712 r = must_be_not_null(r, true);
7713
7714 Node* a_start = array_element_address(a, intcon(0), T_LONG);
7715 assert(a_start, "a array is null");
7716 Node* b_start = array_element_address(b, intcon(0), T_LONG);
7717 assert(b_start, "b array is null");
7718 Node* r_start = array_element_address(r, intcon(0), T_LONG);
7719 assert(r_start, "r array is null");
7720
7721 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7722 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
7723 stubAddr, stubName, TypePtr::BOTTOM,
7724 a_start, b_start, r_start);
7725 return true;
7726 }
7727
7728 bool LibraryCallKit::inline_intpoly_assign() {
7729 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
7730 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
7731 const char *stubName = "intpoly_assign";
7732 address stubAddr = StubRoutines::intpoly_assign();
7733 if (!stubAddr) return false;
7734
7735 Node* set = argument(0);
7736 Node* a = argument(1);
7737 Node* b = argument(2);
7738 Node* arr_length = load_array_length(a);
7739
7740 a = must_be_not_null(a, true);
7741 b = must_be_not_null(b, true);
7742
7743 Node* a_start = array_element_address(a, intcon(0), T_LONG);
7744 assert(a_start, "a array is null");
7745 Node* b_start = array_element_address(b, intcon(0), T_LONG);
7746 assert(b_start, "b array is null");
7747
7748 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7749 OptoRuntime::intpoly_assign_Type(),
7750 stubAddr, stubName, TypePtr::BOTTOM,
7751 set, a_start, b_start, arr_length);
7752 return true;
7753 }
7754
7755 //------------------------------inline_digestBase_implCompress-----------------------
7756 //
7757 // Calculate MD5 for single-block byte[] array.
7758 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
7759 //
7760 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
7761 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
7762 //
7763 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
7764 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
7765 //
7766 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
7767 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
7768 //
7769 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
7770 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
7771 //
7772 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
7773 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
7774
7775 Node* digestBase_obj = argument(0);
7776 Node* src = argument(1); // type oop
7777 Node* ofs = argument(2); // type int
7778
7779 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7780 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7781 // failed array check
7782 return false;
7783 }
7784 // Figure out the size and type of the elements we will be copying.
7785 BasicType src_elem = src_type->elem()->array_element_basic_type();
7786 if (src_elem != T_BYTE) {
7787 return false;
7788 }
7789 // 'src_start' points to src array + offset
7790 src = must_be_not_null(src, true);
7791 Node* src_start = array_element_address(src, ofs, src_elem);
7792 Node* state = nullptr;
7793 Node* block_size = nullptr;
7794 address stubAddr;
7795 const char *stubName;
7796
7797 switch(id) {
7798 case vmIntrinsics::_md5_implCompress:
7799 assert(UseMD5Intrinsics, "need MD5 instruction support");
7800 state = get_state_from_digest_object(digestBase_obj, T_INT);
7801 stubAddr = StubRoutines::md5_implCompress();
7802 stubName = "md5_implCompress";
7803 break;
7804 case vmIntrinsics::_sha_implCompress:
7805 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
7806 state = get_state_from_digest_object(digestBase_obj, T_INT);
7807 stubAddr = StubRoutines::sha1_implCompress();
7808 stubName = "sha1_implCompress";
7809 break;
7810 case vmIntrinsics::_sha2_implCompress:
7811 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
7812 state = get_state_from_digest_object(digestBase_obj, T_INT);
7813 stubAddr = StubRoutines::sha256_implCompress();
7814 stubName = "sha256_implCompress";
7815 break;
7816 case vmIntrinsics::_sha5_implCompress:
7817 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
7818 state = get_state_from_digest_object(digestBase_obj, T_LONG);
7819 stubAddr = StubRoutines::sha512_implCompress();
7820 stubName = "sha512_implCompress";
7821 break;
7822 case vmIntrinsics::_sha3_implCompress:
7823 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
7824 state = get_state_from_digest_object(digestBase_obj, T_LONG);
7825 stubAddr = StubRoutines::sha3_implCompress();
7826 stubName = "sha3_implCompress";
7827 block_size = get_block_size_from_digest_object(digestBase_obj);
7828 if (block_size == nullptr) return false;
7829 break;
7830 default:
7831 fatal_unexpected_iid(id);
7832 return false;
7833 }
7834 if (state == nullptr) return false;
7835
7836 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
7837 if (stubAddr == nullptr) return false;
7838
7839 // Call the stub.
7840 Node* call;
7841 if (block_size == nullptr) {
7842 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
7843 stubAddr, stubName, TypePtr::BOTTOM,
7844 src_start, state);
7845 } else {
7846 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
7847 stubAddr, stubName, TypePtr::BOTTOM,
7848 src_start, state, block_size);
7849 }
7850
7851 return true;
7852 }
7853
7854 //------------------------------inline_digestBase_implCompressMB-----------------------
7855 //
7856 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
7857 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
7858 //
7859 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
7860 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7861 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7862 assert((uint)predicate < 5, "sanity");
7863 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
7864
7865 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
7866 Node* src = argument(1); // byte[] array
7867 Node* ofs = argument(2); // type int
7868 Node* limit = argument(3); // type int
7869
7870 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7871 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7872 // failed array check
7873 return false;
7874 }
7875 // Figure out the size and type of the elements we will be copying.
7876 BasicType src_elem = src_type->elem()->array_element_basic_type();
7877 if (src_elem != T_BYTE) {
7878 return false;
7879 }
7880 // 'src_start' points to src array + offset
7881 src = must_be_not_null(src, false);
7882 Node* src_start = array_element_address(src, ofs, src_elem);
7883
7884 const char* klass_digestBase_name = nullptr;
7885 const char* stub_name = nullptr;
7886 address stub_addr = nullptr;
7887 BasicType elem_type = T_INT;
7888
7889 switch (predicate) {
7890 case 0:
7891 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
7892 klass_digestBase_name = "sun/security/provider/MD5";
7893 stub_name = "md5_implCompressMB";
7894 stub_addr = StubRoutines::md5_implCompressMB();
7895 }
7896 break;
7897 case 1:
7898 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
7899 klass_digestBase_name = "sun/security/provider/SHA";
7900 stub_name = "sha1_implCompressMB";
7901 stub_addr = StubRoutines::sha1_implCompressMB();
7902 }
7903 break;
7904 case 2:
7905 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
7906 klass_digestBase_name = "sun/security/provider/SHA2";
7907 stub_name = "sha256_implCompressMB";
7908 stub_addr = StubRoutines::sha256_implCompressMB();
7909 }
7910 break;
7911 case 3:
7912 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
7913 klass_digestBase_name = "sun/security/provider/SHA5";
7914 stub_name = "sha512_implCompressMB";
7915 stub_addr = StubRoutines::sha512_implCompressMB();
7916 elem_type = T_LONG;
7917 }
7918 break;
7919 case 4:
7920 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
7921 klass_digestBase_name = "sun/security/provider/SHA3";
7922 stub_name = "sha3_implCompressMB";
7923 stub_addr = StubRoutines::sha3_implCompressMB();
7924 elem_type = T_LONG;
7925 }
7926 break;
7927 default:
7928 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
7929 }
7930 if (klass_digestBase_name != nullptr) {
7931 assert(stub_addr != nullptr, "Stub is generated");
7932 if (stub_addr == nullptr) return false;
7933
7934 // get DigestBase klass to lookup for SHA klass
7935 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
7936 assert(tinst != nullptr, "digestBase_obj is not instance???");
7937 assert(tinst->is_loaded(), "DigestBase is not loaded");
7938
7939 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
7940 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
7941 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
7942 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
7943 }
7944 return false;
7945 }
7946
7947 //------------------------------inline_digestBase_implCompressMB-----------------------
7948 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
7949 BasicType elem_type, address stubAddr, const char *stubName,
7950 Node* src_start, Node* ofs, Node* limit) {
7951 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
7952 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7953 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
7954 digest_obj = _gvn.transform(digest_obj);
7955
7956 Node* state = get_state_from_digest_object(digest_obj, elem_type);
7957 if (state == nullptr) return false;
7958
7959 Node* block_size = nullptr;
7960 if (strcmp("sha3_implCompressMB", stubName) == 0) {
7961 block_size = get_block_size_from_digest_object(digest_obj);
7962 if (block_size == nullptr) return false;
7963 }
7964
7965 // Call the stub.
7966 Node* call;
7967 if (block_size == nullptr) {
7968 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7969 OptoRuntime::digestBase_implCompressMB_Type(false),
7970 stubAddr, stubName, TypePtr::BOTTOM,
7971 src_start, state, ofs, limit);
7972 } else {
7973 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7974 OptoRuntime::digestBase_implCompressMB_Type(true),
7975 stubAddr, stubName, TypePtr::BOTTOM,
7976 src_start, state, block_size, ofs, limit);
7977 }
7978
7979 // return ofs (int)
7980 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7981 set_result(result);
7982
7983 return true;
7984 }
7985
7986 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
7987 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
7988 assert(UseAES, "need AES instruction support");
7989 address stubAddr = nullptr;
7990 const char *stubName = nullptr;
7991 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
7992 stubName = "galoisCounterMode_AESCrypt";
7993
7994 if (stubAddr == nullptr) return false;
7995
7996 Node* in = argument(0);
7997 Node* inOfs = argument(1);
7998 Node* len = argument(2);
7999 Node* ct = argument(3);
8000 Node* ctOfs = argument(4);
8001 Node* out = argument(5);
8002 Node* outOfs = argument(6);
8003 Node* gctr_object = argument(7);
8004 Node* ghash_object = argument(8);
8005
8006 // (1) in, ct and out are arrays.
8007 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8008 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8009 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8010 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8011 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8012 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8013
8014 // checks are the responsibility of the caller
8015 Node* in_start = in;
8016 Node* ct_start = ct;
8017 Node* out_start = out;
8018 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8019 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8020 in_start = array_element_address(in, inOfs, T_BYTE);
8021 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8022 out_start = array_element_address(out, outOfs, T_BYTE);
8023 }
8024
8025 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8026 // (because of the predicated logic executed earlier).
8027 // so we cast it here safely.
8028 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8029 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8030 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8031 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8032 Node* state = load_field_from_object(ghash_object, "state", "[J");
8033
8034 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8035 return false;
8036 }
8037 // cast it to what we know it will be at runtime
8038 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8039 assert(tinst != nullptr, "GCTR obj is null");
8040 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8041 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8042 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8043 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8044 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8045 const TypeOopPtr* xtype = aklass->as_instance_type();
8046 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8047 aescrypt_object = _gvn.transform(aescrypt_object);
8048 // we need to get the start of the aescrypt_object's expanded key array
8049 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8050 if (k_start == nullptr) return false;
8051 // similarly, get the start address of the r vector
8052 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8053 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8054 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8055
8056
8057 // Call the stub, passing params
8058 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8059 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8060 stubAddr, stubName, TypePtr::BOTTOM,
8061 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8062
8063 // return cipher length (int)
8064 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8065 set_result(retvalue);
8066
8067 return true;
8068 }
8069
8070 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8071 // Return node representing slow path of predicate check.
8072 // the pseudo code we want to emulate with this predicate is:
8073 // for encryption:
8074 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8075 // for decryption:
8076 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8077 // note cipher==plain is more conservative than the original java code but that's OK
8078 //
8079
8080 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8081 // The receiver was checked for null already.
8082 Node* objGCTR = argument(7);
8083 // Load embeddedCipher field of GCTR object.
8084 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8085 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8086
8087 // get AESCrypt klass for instanceOf check
8088 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8089 // will have same classloader as CipherBlockChaining object
8090 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8091 assert(tinst != nullptr, "GCTR obj is null");
8092 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8093
8094 // we want to do an instanceof comparison against the AESCrypt class
8095 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8096 if (!klass_AESCrypt->is_loaded()) {
8097 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8098 Node* ctrl = control();
8099 set_control(top()); // no regular fast path
8100 return ctrl;
8101 }
8102
8103 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8104 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8105 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8106 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8107 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8108
8109 return instof_false; // even if it is null
8110 }
8111
8112 //------------------------------get_state_from_digest_object-----------------------
8113 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8114 const char* state_type;
8115 switch (elem_type) {
8116 case T_BYTE: state_type = "[B"; break;
8117 case T_INT: state_type = "[I"; break;
8118 case T_LONG: state_type = "[J"; break;
8119 default: ShouldNotReachHere();
8120 }
8121 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8122 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8123 if (digest_state == nullptr) return (Node *) nullptr;
8124
8125 // now have the array, need to get the start address of the state array
8126 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8127 return state;
8128 }
8129
8130 //------------------------------get_block_size_from_sha3_object----------------------------------
8131 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8132 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8133 assert (block_size != nullptr, "sanity");
8134 return block_size;
8135 }
8136
8137 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8138 // Return node representing slow path of predicate check.
8139 // the pseudo code we want to emulate with this predicate is:
8140 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8141 //
8142 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8143 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8144 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8145 assert((uint)predicate < 5, "sanity");
8146
8147 // The receiver was checked for null already.
8148 Node* digestBaseObj = argument(0);
8149
8150 // get DigestBase klass for instanceOf check
8151 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8152 assert(tinst != nullptr, "digestBaseObj is null");
8153 assert(tinst->is_loaded(), "DigestBase is not loaded");
8154
8155 const char* klass_name = nullptr;
8156 switch (predicate) {
8157 case 0:
8158 if (UseMD5Intrinsics) {
8159 // we want to do an instanceof comparison against the MD5 class
8160 klass_name = "sun/security/provider/MD5";
8161 }
8162 break;
8163 case 1:
8164 if (UseSHA1Intrinsics) {
8165 // we want to do an instanceof comparison against the SHA class
8166 klass_name = "sun/security/provider/SHA";
8167 }
8168 break;
8169 case 2:
8170 if (UseSHA256Intrinsics) {
8171 // we want to do an instanceof comparison against the SHA2 class
8172 klass_name = "sun/security/provider/SHA2";
8173 }
8174 break;
8175 case 3:
8176 if (UseSHA512Intrinsics) {
8177 // we want to do an instanceof comparison against the SHA5 class
8178 klass_name = "sun/security/provider/SHA5";
8179 }
8180 break;
8181 case 4:
8182 if (UseSHA3Intrinsics) {
8183 // we want to do an instanceof comparison against the SHA3 class
8184 klass_name = "sun/security/provider/SHA3";
8185 }
8186 break;
8187 default:
8188 fatal("unknown SHA intrinsic predicate: %d", predicate);
8189 }
8190
8191 ciKlass* klass = nullptr;
8192 if (klass_name != nullptr) {
8193 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8194 }
8195 if ((klass == nullptr) || !klass->is_loaded()) {
8196 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8197 Node* ctrl = control();
8198 set_control(top()); // no intrinsic path
8199 return ctrl;
8200 }
8201 ciInstanceKlass* instklass = klass->as_instance_klass();
8202
8203 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8204 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8205 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8206 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8207
8208 return instof_false; // even if it is null
8209 }
8210
8211 //-------------inline_fma-----------------------------------
8212 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8213 Node *a = nullptr;
8214 Node *b = nullptr;
8215 Node *c = nullptr;
8216 Node* result = nullptr;
8217 switch (id) {
8218 case vmIntrinsics::_fmaD:
8219 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8220 // no receiver since it is static method
8221 a = round_double_node(argument(0));
8222 b = round_double_node(argument(2));
8223 c = round_double_node(argument(4));
8224 result = _gvn.transform(new FmaDNode(a, b, c));
8225 break;
8226 case vmIntrinsics::_fmaF:
8227 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8228 a = argument(0);
8229 b = argument(1);
8230 c = argument(2);
8231 result = _gvn.transform(new FmaFNode(a, b, c));
8232 break;
8233 default:
8234 fatal_unexpected_iid(id); break;
8235 }
8236 set_result(result);
8237 return true;
8238 }
8239
8240 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8241 // argument(0) is receiver
8242 Node* codePoint = argument(1);
8243 Node* n = nullptr;
8244
8245 switch (id) {
8246 case vmIntrinsics::_isDigit :
8247 n = new DigitNode(control(), codePoint);
8248 break;
8249 case vmIntrinsics::_isLowerCase :
8250 n = new LowerCaseNode(control(), codePoint);
8251 break;
8252 case vmIntrinsics::_isUpperCase :
8253 n = new UpperCaseNode(control(), codePoint);
8254 break;
8255 case vmIntrinsics::_isWhitespace :
8256 n = new WhitespaceNode(control(), codePoint);
8257 break;
8258 default:
8259 fatal_unexpected_iid(id);
8260 }
8261
8262 set_result(_gvn.transform(n));
8263 return true;
8264 }
8265
8266 //------------------------------inline_fp_min_max------------------------------
8267 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
8268 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
8269
8270 // The intrinsic should be used only when the API branches aren't predictable,
8271 // the last one performing the most important comparison. The following heuristic
8272 // uses the branch statistics to eventually bail out if necessary.
8273
8274 ciMethodData *md = callee()->method_data();
8275
8276 if ( md != nullptr && md->is_mature() && md->invocation_count() > 0 ) {
8277 ciCallProfile cp = caller()->call_profile_at_bci(bci());
8278
8279 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
8280 // Bail out if the call-site didn't contribute enough to the statistics.
8281 return false;
8282 }
8283
8284 uint taken = 0, not_taken = 0;
8285
8286 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
8287 if (p->is_BranchData()) {
8288 taken = ((ciBranchData*)p)->taken();
8289 not_taken = ((ciBranchData*)p)->not_taken();
8290 }
8291 }
8292
8293 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
8294 balance = balance < 0 ? -balance : balance;
8295 if ( balance > 0.2 ) {
8296 // Bail out if the most important branch is predictable enough.
8297 return false;
8298 }
8299 }
8300 */
8301
8302 Node *a = nullptr;
8303 Node *b = nullptr;
8304 Node *n = nullptr;
8305 switch (id) {
8306 case vmIntrinsics::_maxF:
8307 case vmIntrinsics::_minF:
8308 case vmIntrinsics::_maxF_strict:
8309 case vmIntrinsics::_minF_strict:
8310 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
8311 a = argument(0);
8312 b = argument(1);
8313 break;
8314 case vmIntrinsics::_maxD:
8315 case vmIntrinsics::_minD:
8316 case vmIntrinsics::_maxD_strict:
8317 case vmIntrinsics::_minD_strict:
8318 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
8319 a = round_double_node(argument(0));
8320 b = round_double_node(argument(2));
8321 break;
8322 default:
8323 fatal_unexpected_iid(id);
8324 break;
8325 }
8326 switch (id) {
8327 case vmIntrinsics::_maxF:
8328 case vmIntrinsics::_maxF_strict:
8329 n = new MaxFNode(a, b);
8330 break;
8331 case vmIntrinsics::_minF:
8332 case vmIntrinsics::_minF_strict:
8333 n = new MinFNode(a, b);
8334 break;
8335 case vmIntrinsics::_maxD:
8336 case vmIntrinsics::_maxD_strict:
8337 n = new MaxDNode(a, b);
8338 break;
8339 case vmIntrinsics::_minD:
8340 case vmIntrinsics::_minD_strict:
8341 n = new MinDNode(a, b);
8342 break;
8343 default:
8344 fatal_unexpected_iid(id);
8345 break;
8346 }
8347 set_result(_gvn.transform(n));
8348 return true;
8349 }
8350
8351 bool LibraryCallKit::inline_profileBoolean() {
8352 Node* counts = argument(1);
8353 const TypeAryPtr* ary = nullptr;
8354 ciArray* aobj = nullptr;
8355 if (counts->is_Con()
8356 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8357 && (aobj = ary->const_oop()->as_array()) != nullptr
8358 && (aobj->length() == 2)) {
8359 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8360 jint false_cnt = aobj->element_value(0).as_int();
8361 jint true_cnt = aobj->element_value(1).as_int();
8362
8363 if (C->log() != nullptr) {
8364 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8365 false_cnt, true_cnt);
8366 }
8367
8368 if (false_cnt + true_cnt == 0) {
8369 // According to profile, never executed.
8370 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8371 Deoptimization::Action_reinterpret);
8372 return true;
8373 }
8374
8375 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8376 // is a number of each value occurrences.
8377 Node* result = argument(0);
8378 if (false_cnt == 0 || true_cnt == 0) {
8379 // According to profile, one value has been never seen.
8380 int expected_val = (false_cnt == 0) ? 1 : 0;
8381
8382 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8383 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8384
8385 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8386 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8387 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8388
8389 { // Slow path: uncommon trap for never seen value and then reexecute
8390 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8391 // the value has been seen at least once.
8392 PreserveJVMState pjvms(this);
8393 PreserveReexecuteState preexecs(this);
8394 jvms()->set_should_reexecute(true);
8395
8396 set_control(slow_path);
8397 set_i_o(i_o());
8398
8399 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8400 Deoptimization::Action_reinterpret);
8401 }
8402 // The guard for never seen value enables sharpening of the result and
8403 // returning a constant. It allows to eliminate branches on the same value
8404 // later on.
8405 set_control(fast_path);
8406 result = intcon(expected_val);
8407 }
8408 // Stop profiling.
8409 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8410 // By replacing method body with profile data (represented as ProfileBooleanNode
8411 // on IR level) we effectively disable profiling.
8412 // It enables full speed execution once optimized code is generated.
8413 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8414 C->record_for_igvn(profile);
8415 set_result(profile);
8416 return true;
8417 } else {
8418 // Continue profiling.
8419 // Profile data isn't available at the moment. So, execute method's bytecode version.
8420 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8421 // is compiled and counters aren't available since corresponding MethodHandle
8422 // isn't a compile-time constant.
8423 return false;
8424 }
8425 }
8426
8427 bool LibraryCallKit::inline_isCompileConstant() {
8428 Node* n = argument(0);
8429 set_result(n->is_Con() ? intcon(1) : intcon(0));
8430 return true;
8431 }
8432
8433 //------------------------------- inline_getObjectSize --------------------------------------
8434 //
8435 // Calculate the runtime size of the object/array.
8436 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8437 //
8438 bool LibraryCallKit::inline_getObjectSize() {
8439 Node* obj = argument(3);
8440 Node* klass_node = load_object_klass(obj);
8441
8442 jint layout_con = Klass::_lh_neutral_value;
8443 Node* layout_val = get_layout_helper(klass_node, layout_con);
8444 int layout_is_con = (layout_val == nullptr);
8445
8446 if (layout_is_con) {
8447 // Layout helper is constant, can figure out things at compile time.
8448
8449 if (Klass::layout_helper_is_instance(layout_con)) {
8450 // Instance case: layout_con contains the size itself.
8451 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8452 set_result(size);
8453 } else {
8454 // Array case: size is round(header + element_size*arraylength).
8455 // Since arraylength is different for every array instance, we have to
8456 // compute the whole thing at runtime.
8457
8458 Node* arr_length = load_array_length(obj);
8459
8460 int round_mask = MinObjAlignmentInBytes - 1;
8461 int hsize = Klass::layout_helper_header_size(layout_con);
8462 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8463
8464 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8465 round_mask = 0; // strength-reduce it if it goes away completely
8466 }
8467 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8468 Node* header_size = intcon(hsize + round_mask);
8469
8470 Node* lengthx = ConvI2X(arr_length);
8471 Node* headerx = ConvI2X(header_size);
8472
8473 Node* abody = lengthx;
8474 if (eshift != 0) {
8475 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8476 }
8477 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8478 if (round_mask != 0) {
8479 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8480 }
8481 size = ConvX2L(size);
8482 set_result(size);
8483 }
8484 } else {
8485 // Layout helper is not constant, need to test for array-ness at runtime.
8486
8487 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8488 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8489 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8490 record_for_igvn(result_reg);
8491
8492 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
8493 if (array_ctl != nullptr) {
8494 // Array case: size is round(header + element_size*arraylength).
8495 // Since arraylength is different for every array instance, we have to
8496 // compute the whole thing at runtime.
8497
8498 PreserveJVMState pjvms(this);
8499 set_control(array_ctl);
8500 Node* arr_length = load_array_length(obj);
8501
8502 int round_mask = MinObjAlignmentInBytes - 1;
8503 Node* mask = intcon(round_mask);
8504
8505 Node* hss = intcon(Klass::_lh_header_size_shift);
8506 Node* hsm = intcon(Klass::_lh_header_size_mask);
8507 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8508 header_size = _gvn.transform(new AndINode(header_size, hsm));
8509 header_size = _gvn.transform(new AddINode(header_size, mask));
8510
8511 // There is no need to mask or shift this value.
8512 // The semantics of LShiftINode include an implicit mask to 0x1F.
8513 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8514 Node* elem_shift = layout_val;
8515
8516 Node* lengthx = ConvI2X(arr_length);
8517 Node* headerx = ConvI2X(header_size);
8518
8519 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
8520 Node* size = _gvn.transform(new AddXNode(headerx, abody));
8521 if (round_mask != 0) {
8522 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
8523 }
8524 size = ConvX2L(size);
8525
8526 result_reg->init_req(_array_path, control());
8527 result_val->init_req(_array_path, size);
8528 }
8529
8530 if (!stopped()) {
8531 // Instance case: the layout helper gives us instance size almost directly,
8532 // but we need to mask out the _lh_instance_slow_path_bit.
8533 Node* size = ConvI2X(layout_val);
8534 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
8535 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
8536 size = _gvn.transform(new AndXNode(size, mask));
8537 size = ConvX2L(size);
8538
8539 result_reg->init_req(_instance_path, control());
8540 result_val->init_req(_instance_path, size);
8541 }
8542
8543 set_result(result_reg, result_val);
8544 }
8545
8546 return true;
8547 }
8548
8549 //------------------------------- inline_blackhole --------------------------------------
8550 //
8551 // Make sure all arguments to this node are alive.
8552 // This matches methods that were requested to be blackholed through compile commands.
8553 //
8554 bool LibraryCallKit::inline_blackhole() {
8555 assert(callee()->is_static(), "Should have been checked before: only static methods here");
8556 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
8557 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
8558
8559 // Blackhole node pinches only the control, not memory. This allows
8560 // the blackhole to be pinned in the loop that computes blackholed
8561 // values, but have no other side effects, like breaking the optimizations
8562 // across the blackhole.
8563
8564 Node* bh = _gvn.transform(new BlackholeNode(control()));
8565 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
8566
8567 // Bind call arguments as blackhole arguments to keep them alive
8568 uint nargs = callee()->arg_size();
8569 for (uint i = 0; i < nargs; i++) {
8570 bh->add_req(argument(i));
8571 }
8572
8573 return true;
8574 }
8575
8576 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
8577 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
8578 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
8579 return nullptr; // box klass is not Float16
8580 }
8581
8582 // Null check; get notnull casted pointer
8583 Node* null_ctl = top();
8584 Node* not_null_box = null_check_oop(box, &null_ctl, true);
8585 // If not_null_box is dead, only null-path is taken
8586 if (stopped()) {
8587 set_control(null_ctl);
8588 return nullptr;
8589 }
8590 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
8591 const TypePtr* adr_type = C->alias_type(field)->adr_type();
8592 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
8593 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
8594 }
8595
8596 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
8597 PreserveReexecuteState preexecs(this);
8598 jvms()->set_should_reexecute(true);
8599
8600 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
8601 Node* klass_node = makecon(klass_type);
8602 Node* box = new_instance(klass_node);
8603
8604 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
8605 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
8606
8607 Node* field_store = _gvn.transform(access_store_at(box,
8608 value_field,
8609 value_adr_type,
8610 value,
8611 TypeInt::SHORT,
8612 T_SHORT,
8613 IN_HEAP));
8614 set_memory(field_store, value_adr_type);
8615 return box;
8616 }
8617
8618 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
8619 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
8620 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
8621 return false;
8622 }
8623
8624 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
8625 if (box_type == nullptr || box_type->const_oop() == nullptr) {
8626 return false;
8627 }
8628
8629 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
8630 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
8631 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
8632 ciSymbols::short_signature(),
8633 false);
8634 assert(field != nullptr, "");
8635
8636 // Transformed nodes
8637 Node* fld1 = nullptr;
8638 Node* fld2 = nullptr;
8639 Node* fld3 = nullptr;
8640 switch(num_args) {
8641 case 3:
8642 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
8643 if (fld3 == nullptr) {
8644 return false;
8645 }
8646 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
8647 // fall-through
8648 case 2:
8649 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
8650 if (fld2 == nullptr) {
8651 return false;
8652 }
8653 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
8654 // fall-through
8655 case 1:
8656 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
8657 if (fld1 == nullptr) {
8658 return false;
8659 }
8660 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
8661 break;
8662 default: fatal("Unsupported number of arguments %d", num_args);
8663 }
8664
8665 Node* result = nullptr;
8666 switch (id) {
8667 // Unary operations
8668 case vmIntrinsics::_sqrt_float16:
8669 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
8670 break;
8671 // Ternary operations
8672 case vmIntrinsics::_fma_float16:
8673 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
8674 break;
8675 default:
8676 fatal_unexpected_iid(id);
8677 break;
8678 }
8679 result = _gvn.transform(new ReinterpretHF2SNode(result));
8680 set_result(box_fp16_value(float16_box_type, field, result));
8681 return true;
8682 }
8683