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*)CompileBroker::compiler(CompLevel_full_optimization);
75 bool is_available = false;
76
77 {
78 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
79 // the compiler must transition to '_thread_in_vm' state because both
80 // methods access VM-internal data.
81 VM_ENTRY_MARK;
82 methodHandle mh(THREAD, m->get_Method());
83 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
84 if (is_available && is_virtual) {
85 is_available = vmIntrinsics::does_virtual_dispatch(id);
86 }
87 }
88
89 if (is_available) {
90 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
91 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
92 return new LibraryIntrinsic(m, is_virtual,
93 vmIntrinsics::predicates_needed(id),
94 vmIntrinsics::does_virtual_dispatch(id),
95 id);
96 } else {
97 return nullptr;
98 }
99 }
100
101 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
102 LibraryCallKit kit(jvms, this);
103 Compile* C = kit.C;
104 int nodes = C->unique();
105 #ifndef PRODUCT
106 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
107 char buf[1000];
108 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
109 tty->print_cr("Intrinsic %s", str);
110 }
111 #endif
112 ciMethod* callee = kit.callee();
113 const int bci = kit.bci();
114 #ifdef ASSERT
115 Node* ctrl = kit.control();
116 #endif
117 // Try to inline the intrinsic.
118 if (callee->check_intrinsic_candidate() &&
119 kit.try_to_inline(_last_predicate)) {
120 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
121 : "(intrinsic)";
122 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
123 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 case vmIntrinsics::_double_keccak:
598 return inline_double_keccak();
599
600 case vmIntrinsics::_digestBase_implCompressMB:
601 return inline_digestBase_implCompressMB(predicate);
602
603 case vmIntrinsics::_multiplyToLen:
604 return inline_multiplyToLen();
605
606 case vmIntrinsics::_squareToLen:
607 return inline_squareToLen();
608
609 case vmIntrinsics::_mulAdd:
610 return inline_mulAdd();
611
612 case vmIntrinsics::_montgomeryMultiply:
613 return inline_montgomeryMultiply();
614 case vmIntrinsics::_montgomerySquare:
615 return inline_montgomerySquare();
616
617 case vmIntrinsics::_bigIntegerRightShiftWorker:
618 return inline_bigIntegerShift(true);
619 case vmIntrinsics::_bigIntegerLeftShiftWorker:
620 return inline_bigIntegerShift(false);
621
622 case vmIntrinsics::_vectorizedMismatch:
623 return inline_vectorizedMismatch();
624
625 case vmIntrinsics::_ghash_processBlocks:
626 return inline_ghash_processBlocks();
627 case vmIntrinsics::_chacha20Block:
628 return inline_chacha20Block();
629 case vmIntrinsics::_dilithiumAlmostNtt:
630 return inline_dilithiumAlmostNtt();
631 case vmIntrinsics::_dilithiumAlmostInverseNtt:
632 return inline_dilithiumAlmostInverseNtt();
633 case vmIntrinsics::_dilithiumNttMult:
634 return inline_dilithiumNttMult();
635 case vmIntrinsics::_dilithiumMontMulByConstant:
636 return inline_dilithiumMontMulByConstant();
637 case vmIntrinsics::_dilithiumDecomposePoly:
638 return inline_dilithiumDecomposePoly();
639 case vmIntrinsics::_base64_encodeBlock:
640 return inline_base64_encodeBlock();
641 case vmIntrinsics::_base64_decodeBlock:
642 return inline_base64_decodeBlock();
643 case vmIntrinsics::_poly1305_processBlocks:
644 return inline_poly1305_processBlocks();
645 case vmIntrinsics::_intpoly_montgomeryMult_P256:
646 return inline_intpoly_montgomeryMult_P256();
647 case vmIntrinsics::_intpoly_assign:
648 return inline_intpoly_assign();
649 case vmIntrinsics::_encodeISOArray:
650 case vmIntrinsics::_encodeByteISOArray:
651 return inline_encodeISOArray(false);
652 case vmIntrinsics::_encodeAsciiArray:
653 return inline_encodeISOArray(true);
654
655 case vmIntrinsics::_updateCRC32:
656 return inline_updateCRC32();
657 case vmIntrinsics::_updateBytesCRC32:
658 return inline_updateBytesCRC32();
659 case vmIntrinsics::_updateByteBufferCRC32:
660 return inline_updateByteBufferCRC32();
661
662 case vmIntrinsics::_updateBytesCRC32C:
663 return inline_updateBytesCRC32C();
664 case vmIntrinsics::_updateDirectByteBufferCRC32C:
665 return inline_updateDirectByteBufferCRC32C();
666
667 case vmIntrinsics::_updateBytesAdler32:
668 return inline_updateBytesAdler32();
669 case vmIntrinsics::_updateByteBufferAdler32:
670 return inline_updateByteBufferAdler32();
671
672 case vmIntrinsics::_profileBoolean:
673 return inline_profileBoolean();
674 case vmIntrinsics::_isCompileConstant:
675 return inline_isCompileConstant();
676
677 case vmIntrinsics::_countPositives:
678 return inline_countPositives();
679
680 case vmIntrinsics::_fmaD:
681 case vmIntrinsics::_fmaF:
682 return inline_fma(intrinsic_id());
683
684 case vmIntrinsics::_isDigit:
685 case vmIntrinsics::_isLowerCase:
686 case vmIntrinsics::_isUpperCase:
687 case vmIntrinsics::_isWhitespace:
688 return inline_character_compare(intrinsic_id());
689
690 case vmIntrinsics::_min:
691 case vmIntrinsics::_max:
692 case vmIntrinsics::_min_strict:
693 case vmIntrinsics::_max_strict:
694 case vmIntrinsics::_minL:
695 case vmIntrinsics::_maxL:
696 case vmIntrinsics::_minF:
697 case vmIntrinsics::_maxF:
698 case vmIntrinsics::_minD:
699 case vmIntrinsics::_maxD:
700 case vmIntrinsics::_minF_strict:
701 case vmIntrinsics::_maxF_strict:
702 case vmIntrinsics::_minD_strict:
703 case vmIntrinsics::_maxD_strict:
704 return inline_min_max(intrinsic_id());
705
706 case vmIntrinsics::_VectorUnaryOp:
707 return inline_vector_nary_operation(1);
708 case vmIntrinsics::_VectorBinaryOp:
709 return inline_vector_nary_operation(2);
710 case vmIntrinsics::_VectorTernaryOp:
711 return inline_vector_nary_operation(3);
712 case vmIntrinsics::_VectorFromBitsCoerced:
713 return inline_vector_frombits_coerced();
714 case vmIntrinsics::_VectorMaskOp:
715 return inline_vector_mask_operation();
716 case vmIntrinsics::_VectorLoadOp:
717 return inline_vector_mem_operation(/*is_store=*/false);
718 case vmIntrinsics::_VectorLoadMaskedOp:
719 return inline_vector_mem_masked_operation(/*is_store*/false);
720 case vmIntrinsics::_VectorStoreOp:
721 return inline_vector_mem_operation(/*is_store=*/true);
722 case vmIntrinsics::_VectorStoreMaskedOp:
723 return inline_vector_mem_masked_operation(/*is_store=*/true);
724 case vmIntrinsics::_VectorGatherOp:
725 return inline_vector_gather_scatter(/*is_scatter*/ false);
726 case vmIntrinsics::_VectorScatterOp:
727 return inline_vector_gather_scatter(/*is_scatter*/ true);
728 case vmIntrinsics::_VectorReductionCoerced:
729 return inline_vector_reduction();
730 case vmIntrinsics::_VectorTest:
731 return inline_vector_test();
732 case vmIntrinsics::_VectorBlend:
733 return inline_vector_blend();
734 case vmIntrinsics::_VectorRearrange:
735 return inline_vector_rearrange();
736 case vmIntrinsics::_VectorSelectFrom:
737 return inline_vector_select_from();
738 case vmIntrinsics::_VectorCompare:
739 return inline_vector_compare();
740 case vmIntrinsics::_VectorBroadcastInt:
741 return inline_vector_broadcast_int();
742 case vmIntrinsics::_VectorConvert:
743 return inline_vector_convert();
744 case vmIntrinsics::_VectorInsert:
745 return inline_vector_insert();
746 case vmIntrinsics::_VectorExtract:
747 return inline_vector_extract();
748 case vmIntrinsics::_VectorCompressExpand:
749 return inline_vector_compress_expand();
750 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
751 return inline_vector_select_from_two_vectors();
752 case vmIntrinsics::_IndexVector:
753 return inline_index_vector();
754 case vmIntrinsics::_IndexPartiallyInUpperRange:
755 return inline_index_partially_in_upper_range();
756
757 case vmIntrinsics::_getObjectSize:
758 return inline_getObjectSize();
759
760 case vmIntrinsics::_blackhole:
761 return inline_blackhole();
762
763 default:
764 // If you get here, it may be that someone has added a new intrinsic
765 // to the list in vmIntrinsics.hpp without implementing it here.
766 #ifndef PRODUCT
767 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
768 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
769 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
770 }
771 #endif
772 return false;
773 }
774 }
775
776 Node* LibraryCallKit::try_to_predicate(int predicate) {
777 if (!jvms()->has_method()) {
778 // Root JVMState has a null method.
779 assert(map()->memory()->Opcode() == Op_Parm, "");
780 // Insert the memory aliasing node
781 set_all_memory(reset_memory());
782 }
783 assert(merged_memory(), "");
784
785 switch (intrinsic_id()) {
786 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
787 return inline_cipherBlockChaining_AESCrypt_predicate(false);
788 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
789 return inline_cipherBlockChaining_AESCrypt_predicate(true);
790 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
791 return inline_electronicCodeBook_AESCrypt_predicate(false);
792 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
793 return inline_electronicCodeBook_AESCrypt_predicate(true);
794 case vmIntrinsics::_counterMode_AESCrypt:
795 return inline_counterMode_AESCrypt_predicate();
796 case vmIntrinsics::_digestBase_implCompressMB:
797 return inline_digestBase_implCompressMB_predicate(predicate);
798 case vmIntrinsics::_galoisCounterMode_AESCrypt:
799 return inline_galoisCounterMode_AESCrypt_predicate();
800
801 default:
802 // If you get here, it may be that someone has added a new intrinsic
803 // to the list in vmIntrinsics.hpp without implementing it here.
804 #ifndef PRODUCT
805 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
806 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
807 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
808 }
809 #endif
810 Node* slow_ctl = control();
811 set_control(top()); // No fast path intrinsic
812 return slow_ctl;
813 }
814 }
815
816 //------------------------------set_result-------------------------------
817 // Helper function for finishing intrinsics.
818 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
819 record_for_igvn(region);
820 set_control(_gvn.transform(region));
821 set_result( _gvn.transform(value));
822 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
823 }
824
825 //------------------------------generate_guard---------------------------
826 // Helper function for generating guarded fast-slow graph structures.
827 // The given 'test', if true, guards a slow path. If the test fails
828 // then a fast path can be taken. (We generally hope it fails.)
829 // In all cases, GraphKit::control() is updated to the fast path.
830 // The returned value represents the control for the slow path.
831 // The return value is never 'top'; it is either a valid control
832 // or null if it is obvious that the slow path can never be taken.
833 // Also, if region and the slow control are not null, the slow edge
834 // is appended to the region.
835 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
836 if (stopped()) {
837 // Already short circuited.
838 return nullptr;
839 }
840
841 // Build an if node and its projections.
842 // If test is true we take the slow path, which we assume is uncommon.
843 if (_gvn.type(test) == TypeInt::ZERO) {
844 // The slow branch is never taken. No need to build this guard.
845 return nullptr;
846 }
847
848 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
849
850 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
851 if (if_slow == top()) {
852 // The slow branch is never taken. No need to build this guard.
853 return nullptr;
854 }
855
856 if (region != nullptr)
857 region->add_req(if_slow);
858
859 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
860 set_control(if_fast);
861
862 return if_slow;
863 }
864
865 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
866 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
867 }
868 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
869 return generate_guard(test, region, PROB_FAIR);
870 }
871
872 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
873 Node* *pos_index) {
874 if (stopped())
875 return nullptr; // already stopped
876 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
877 return nullptr; // index is already adequately typed
878 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
879 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
880 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
881 if (is_neg != nullptr && pos_index != nullptr) {
882 // Emulate effect of Parse::adjust_map_after_if.
883 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
884 (*pos_index) = _gvn.transform(ccast);
885 }
886 return is_neg;
887 }
888
889 // Make sure that 'position' is a valid limit index, in [0..length].
890 // There are two equivalent plans for checking this:
891 // A. (offset + copyLength) unsigned<= arrayLength
892 // B. offset <= (arrayLength - copyLength)
893 // We require that all of the values above, except for the sum and
894 // difference, are already known to be non-negative.
895 // Plan A is robust in the face of overflow, if offset and copyLength
896 // are both hugely positive.
897 //
898 // Plan B is less direct and intuitive, but it does not overflow at
899 // all, since the difference of two non-negatives is always
900 // representable. Whenever Java methods must perform the equivalent
901 // check they generally use Plan B instead of Plan A.
902 // For the moment we use Plan A.
903 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
904 Node* subseq_length,
905 Node* array_length,
906 RegionNode* region) {
907 if (stopped())
908 return nullptr; // already stopped
909 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
910 if (zero_offset && subseq_length->eqv_uncast(array_length))
911 return nullptr; // common case of whole-array copy
912 Node* last = subseq_length;
913 if (!zero_offset) // last += offset
914 last = _gvn.transform(new AddINode(last, offset));
915 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
916 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
917 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
918 return is_over;
919 }
920
921 // Emit range checks for the given String.value byte array
922 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
923 if (stopped()) {
924 return; // already stopped
925 }
926 RegionNode* bailout = new RegionNode(1);
927 record_for_igvn(bailout);
928 if (char_count) {
929 // Convert char count to byte count
930 count = _gvn.transform(new LShiftINode(count, intcon(1)));
931 }
932
933 // Offset and count must not be negative
934 generate_negative_guard(offset, bailout);
935 generate_negative_guard(count, bailout);
936 // Offset + count must not exceed length of array
937 generate_limit_guard(offset, count, load_array_length(array), bailout);
938
939 if (bailout->req() > 1) {
940 PreserveJVMState pjvms(this);
941 set_control(_gvn.transform(bailout));
942 uncommon_trap(Deoptimization::Reason_intrinsic,
943 Deoptimization::Action_maybe_recompile);
944 }
945 }
946
947 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
948 bool is_immutable) {
949 ciKlass* thread_klass = env()->Thread_klass();
950 const Type* thread_type
951 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
952
953 Node* thread = _gvn.transform(new ThreadLocalNode());
954 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
955 tls_output = thread;
956
957 Node* thread_obj_handle
958 = (is_immutable
959 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
960 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
961 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
962 thread_obj_handle = _gvn.transform(thread_obj_handle);
963
964 DecoratorSet decorators = IN_NATIVE;
965 if (is_immutable) {
966 decorators |= C2_IMMUTABLE_MEMORY;
967 }
968 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
969 }
970
971 //--------------------------generate_current_thread--------------------
972 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
973 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
974 /*is_immutable*/false);
975 }
976
977 //--------------------------generate_virtual_thread--------------------
978 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
979 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
980 !C->method()->changes_current_thread());
981 }
982
983 //------------------------------make_string_method_node------------------------
984 // Helper method for String intrinsic functions. This version is called with
985 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
986 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
987 // containing the lengths of str1 and str2.
988 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
989 Node* result = nullptr;
990 switch (opcode) {
991 case Op_StrIndexOf:
992 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
993 str1_start, cnt1, str2_start, cnt2, ae);
994 break;
995 case Op_StrComp:
996 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
997 str1_start, cnt1, str2_start, cnt2, ae);
998 break;
999 case Op_StrEquals:
1000 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1001 // Use the constant length if there is one because optimized match rule may exist.
1002 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1003 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1004 break;
1005 default:
1006 ShouldNotReachHere();
1007 return nullptr;
1008 }
1009
1010 // All these intrinsics have checks.
1011 C->set_has_split_ifs(true); // Has chance for split-if optimization
1012 clear_upper_avx();
1013
1014 return _gvn.transform(result);
1015 }
1016
1017 //------------------------------inline_string_compareTo------------------------
1018 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1019 Node* arg1 = argument(0);
1020 Node* arg2 = argument(1);
1021
1022 arg1 = must_be_not_null(arg1, true);
1023 arg2 = must_be_not_null(arg2, true);
1024
1025 // Get start addr and length of first argument
1026 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1027 Node* arg1_cnt = load_array_length(arg1);
1028
1029 // Get start addr and length of second argument
1030 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1031 Node* arg2_cnt = load_array_length(arg2);
1032
1033 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1034 set_result(result);
1035 return true;
1036 }
1037
1038 //------------------------------inline_string_equals------------------------
1039 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1040 Node* arg1 = argument(0);
1041 Node* arg2 = argument(1);
1042
1043 // paths (plus control) merge
1044 RegionNode* region = new RegionNode(3);
1045 Node* phi = new PhiNode(region, TypeInt::BOOL);
1046
1047 if (!stopped()) {
1048
1049 arg1 = must_be_not_null(arg1, true);
1050 arg2 = must_be_not_null(arg2, true);
1051
1052 // Get start addr and length of first argument
1053 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1054 Node* arg1_cnt = load_array_length(arg1);
1055
1056 // Get start addr and length of second argument
1057 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1058 Node* arg2_cnt = load_array_length(arg2);
1059
1060 // Check for arg1_cnt != arg2_cnt
1061 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1062 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1063 Node* if_ne = generate_slow_guard(bol, nullptr);
1064 if (if_ne != nullptr) {
1065 phi->init_req(2, intcon(0));
1066 region->init_req(2, if_ne);
1067 }
1068
1069 // Check for count == 0 is done by assembler code for StrEquals.
1070
1071 if (!stopped()) {
1072 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1073 phi->init_req(1, equals);
1074 region->init_req(1, control());
1075 }
1076 }
1077
1078 // post merge
1079 set_control(_gvn.transform(region));
1080 record_for_igvn(region);
1081
1082 set_result(_gvn.transform(phi));
1083 return true;
1084 }
1085
1086 //------------------------------inline_array_equals----------------------------
1087 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1088 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1089 Node* arg1 = argument(0);
1090 Node* arg2 = argument(1);
1091
1092 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1093 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1094 clear_upper_avx();
1095
1096 return true;
1097 }
1098
1099
1100 //------------------------------inline_countPositives------------------------------
1101 bool LibraryCallKit::inline_countPositives() {
1102 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1103 return false;
1104 }
1105
1106 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1107 // no receiver since it is static method
1108 Node* ba = argument(0);
1109 Node* offset = argument(1);
1110 Node* len = argument(2);
1111
1112 ba = must_be_not_null(ba, true);
1113
1114 // Range checks
1115 generate_string_range_check(ba, offset, len, false);
1116 if (stopped()) {
1117 return true;
1118 }
1119 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1120 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1121 set_result(_gvn.transform(result));
1122 clear_upper_avx();
1123 return true;
1124 }
1125
1126 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1127 Node* index = argument(0);
1128 Node* length = bt == T_INT ? argument(1) : argument(2);
1129 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1130 return false;
1131 }
1132
1133 // check that length is positive
1134 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1135 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1136
1137 {
1138 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1139 uncommon_trap(Deoptimization::Reason_intrinsic,
1140 Deoptimization::Action_make_not_entrant);
1141 }
1142
1143 if (stopped()) {
1144 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1145 return true;
1146 }
1147
1148 // length is now known positive, add a cast node to make this explicit
1149 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1150 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1151 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1152 ConstraintCastNode::RegularDependency, bt);
1153 casted_length = _gvn.transform(casted_length);
1154 replace_in_map(length, casted_length);
1155 length = casted_length;
1156
1157 // Use an unsigned comparison for the range check itself
1158 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1159 BoolTest::mask btest = BoolTest::lt;
1160 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1161 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1162 _gvn.set_type(rc, rc->Value(&_gvn));
1163 if (!rc_bool->is_Con()) {
1164 record_for_igvn(rc);
1165 }
1166 set_control(_gvn.transform(new IfTrueNode(rc)));
1167 {
1168 PreserveJVMState pjvms(this);
1169 set_control(_gvn.transform(new IfFalseNode(rc)));
1170 uncommon_trap(Deoptimization::Reason_range_check,
1171 Deoptimization::Action_make_not_entrant);
1172 }
1173
1174 if (stopped()) {
1175 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1176 return true;
1177 }
1178
1179 // index is now known to be >= 0 and < length, cast it
1180 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1181 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1182 ConstraintCastNode::RegularDependency, bt);
1183 result = _gvn.transform(result);
1184 set_result(result);
1185 replace_in_map(index, result);
1186 return true;
1187 }
1188
1189 //------------------------------inline_string_indexOf------------------------
1190 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1191 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1192 return false;
1193 }
1194 Node* src = argument(0);
1195 Node* tgt = argument(1);
1196
1197 // Make the merge point
1198 RegionNode* result_rgn = new RegionNode(4);
1199 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1200
1201 src = must_be_not_null(src, true);
1202 tgt = must_be_not_null(tgt, true);
1203
1204 // Get start addr and length of source string
1205 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1206 Node* src_count = load_array_length(src);
1207
1208 // Get start addr and length of substring
1209 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1210 Node* tgt_count = load_array_length(tgt);
1211
1212 Node* result = nullptr;
1213 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1214
1215 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1216 // Divide src size by 2 if String is UTF16 encoded
1217 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1218 }
1219 if (ae == StrIntrinsicNode::UU) {
1220 // Divide substring size by 2 if String is UTF16 encoded
1221 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1222 }
1223
1224 if (call_opt_stub) {
1225 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1226 StubRoutines::_string_indexof_array[ae],
1227 "stringIndexOf", TypePtr::BOTTOM, src_start,
1228 src_count, tgt_start, tgt_count);
1229 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1230 } else {
1231 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1232 result_rgn, result_phi, ae);
1233 }
1234 if (result != nullptr) {
1235 result_phi->init_req(3, result);
1236 result_rgn->init_req(3, control());
1237 }
1238 set_control(_gvn.transform(result_rgn));
1239 record_for_igvn(result_rgn);
1240 set_result(_gvn.transform(result_phi));
1241
1242 return true;
1243 }
1244
1245 //-----------------------------inline_string_indexOfI-----------------------
1246 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1247 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1248 return false;
1249 }
1250 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1251 return false;
1252 }
1253
1254 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1255 Node* src = argument(0); // byte[]
1256 Node* src_count = argument(1); // char count
1257 Node* tgt = argument(2); // byte[]
1258 Node* tgt_count = argument(3); // char count
1259 Node* from_index = argument(4); // char index
1260
1261 src = must_be_not_null(src, true);
1262 tgt = must_be_not_null(tgt, true);
1263
1264 // Multiply byte array index by 2 if String is UTF16 encoded
1265 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1266 src_count = _gvn.transform(new SubINode(src_count, from_index));
1267 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1268 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1269
1270 // Range checks
1271 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1272 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1273 if (stopped()) {
1274 return true;
1275 }
1276
1277 RegionNode* region = new RegionNode(5);
1278 Node* phi = new PhiNode(region, TypeInt::INT);
1279 Node* result = nullptr;
1280
1281 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1282
1283 if (call_opt_stub) {
1284 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1285 StubRoutines::_string_indexof_array[ae],
1286 "stringIndexOf", TypePtr::BOTTOM, src_start,
1287 src_count, tgt_start, tgt_count);
1288 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1289 } else {
1290 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1291 region, phi, ae);
1292 }
1293 if (result != nullptr) {
1294 // The result is index relative to from_index if substring was found, -1 otherwise.
1295 // Generate code which will fold into cmove.
1296 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1297 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1298
1299 Node* if_lt = generate_slow_guard(bol, nullptr);
1300 if (if_lt != nullptr) {
1301 // result == -1
1302 phi->init_req(3, result);
1303 region->init_req(3, if_lt);
1304 }
1305 if (!stopped()) {
1306 result = _gvn.transform(new AddINode(result, from_index));
1307 phi->init_req(4, result);
1308 region->init_req(4, control());
1309 }
1310 }
1311
1312 set_control(_gvn.transform(region));
1313 record_for_igvn(region);
1314 set_result(_gvn.transform(phi));
1315 clear_upper_avx();
1316
1317 return true;
1318 }
1319
1320 // Create StrIndexOfNode with fast path checks
1321 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1322 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1323 // Check for substr count > string count
1324 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1325 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1326 Node* if_gt = generate_slow_guard(bol, nullptr);
1327 if (if_gt != nullptr) {
1328 phi->init_req(1, intcon(-1));
1329 region->init_req(1, if_gt);
1330 }
1331 if (!stopped()) {
1332 // Check for substr count == 0
1333 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1334 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1335 Node* if_zero = generate_slow_guard(bol, nullptr);
1336 if (if_zero != nullptr) {
1337 phi->init_req(2, intcon(0));
1338 region->init_req(2, if_zero);
1339 }
1340 }
1341 if (!stopped()) {
1342 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1343 }
1344 return nullptr;
1345 }
1346
1347 //-----------------------------inline_string_indexOfChar-----------------------
1348 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1349 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1350 return false;
1351 }
1352 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1353 return false;
1354 }
1355 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1356 Node* src = argument(0); // byte[]
1357 Node* int_ch = argument(1);
1358 Node* from_index = argument(2);
1359 Node* max = argument(3);
1360
1361 src = must_be_not_null(src, true);
1362
1363 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1364 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1365 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1366
1367 // Range checks
1368 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1369
1370 // Check for int_ch >= 0
1371 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1372 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1373 {
1374 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1375 uncommon_trap(Deoptimization::Reason_intrinsic,
1376 Deoptimization::Action_maybe_recompile);
1377 }
1378 if (stopped()) {
1379 return true;
1380 }
1381
1382 RegionNode* region = new RegionNode(3);
1383 Node* phi = new PhiNode(region, TypeInt::INT);
1384
1385 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1386 C->set_has_split_ifs(true); // Has chance for split-if optimization
1387 _gvn.transform(result);
1388
1389 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1390 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1391
1392 Node* if_lt = generate_slow_guard(bol, nullptr);
1393 if (if_lt != nullptr) {
1394 // result == -1
1395 phi->init_req(2, result);
1396 region->init_req(2, if_lt);
1397 }
1398 if (!stopped()) {
1399 result = _gvn.transform(new AddINode(result, from_index));
1400 phi->init_req(1, result);
1401 region->init_req(1, control());
1402 }
1403 set_control(_gvn.transform(region));
1404 record_for_igvn(region);
1405 set_result(_gvn.transform(phi));
1406 clear_upper_avx();
1407
1408 return true;
1409 }
1410 //---------------------------inline_string_copy---------------------
1411 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1412 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1413 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1414 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1415 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1416 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1417 bool LibraryCallKit::inline_string_copy(bool compress) {
1418 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1419 return false;
1420 }
1421 int nargs = 5; // 2 oops, 3 ints
1422 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1423
1424 Node* src = argument(0);
1425 Node* src_offset = argument(1);
1426 Node* dst = argument(2);
1427 Node* dst_offset = argument(3);
1428 Node* length = argument(4);
1429
1430 // Check for allocation before we add nodes that would confuse
1431 // tightly_coupled_allocation()
1432 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1433
1434 // Figure out the size and type of the elements we will be copying.
1435 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1436 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1437 if (src_type == nullptr || dst_type == nullptr) {
1438 return false;
1439 }
1440 BasicType src_elem = src_type->elem()->array_element_basic_type();
1441 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1442 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1443 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1444 "Unsupported array types for inline_string_copy");
1445
1446 src = must_be_not_null(src, true);
1447 dst = must_be_not_null(dst, true);
1448
1449 // Convert char[] offsets to byte[] offsets
1450 bool convert_src = (compress && src_elem == T_BYTE);
1451 bool convert_dst = (!compress && dst_elem == T_BYTE);
1452 if (convert_src) {
1453 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1454 } else if (convert_dst) {
1455 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1456 }
1457
1458 // Range checks
1459 generate_string_range_check(src, src_offset, length, convert_src);
1460 generate_string_range_check(dst, dst_offset, length, convert_dst);
1461 if (stopped()) {
1462 return true;
1463 }
1464
1465 Node* src_start = array_element_address(src, src_offset, src_elem);
1466 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1467 // 'src_start' points to src array + scaled offset
1468 // 'dst_start' points to dst array + scaled offset
1469 Node* count = nullptr;
1470 if (compress) {
1471 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1472 } else {
1473 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1474 }
1475
1476 if (alloc != nullptr) {
1477 if (alloc->maybe_set_complete(&_gvn)) {
1478 // "You break it, you buy it."
1479 InitializeNode* init = alloc->initialization();
1480 assert(init->is_complete(), "we just did this");
1481 init->set_complete_with_arraycopy();
1482 assert(dst->is_CheckCastPP(), "sanity");
1483 assert(dst->in(0)->in(0) == init, "dest pinned");
1484 }
1485 // Do not let stores that initialize this object be reordered with
1486 // a subsequent store that would make this object accessible by
1487 // other threads.
1488 // Record what AllocateNode this StoreStore protects so that
1489 // escape analysis can go from the MemBarStoreStoreNode to the
1490 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1491 // based on the escape status of the AllocateNode.
1492 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1493 }
1494 if (compress) {
1495 set_result(_gvn.transform(count));
1496 }
1497 clear_upper_avx();
1498
1499 return true;
1500 }
1501
1502 #ifdef _LP64
1503 #define XTOP ,top() /*additional argument*/
1504 #else //_LP64
1505 #define XTOP /*no additional argument*/
1506 #endif //_LP64
1507
1508 //------------------------inline_string_toBytesU--------------------------
1509 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1510 bool LibraryCallKit::inline_string_toBytesU() {
1511 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1512 return false;
1513 }
1514 // Get the arguments.
1515 Node* value = argument(0);
1516 Node* offset = argument(1);
1517 Node* length = argument(2);
1518
1519 Node* newcopy = nullptr;
1520
1521 // Set the original stack and the reexecute bit for the interpreter to reexecute
1522 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1523 { PreserveReexecuteState preexecs(this);
1524 jvms()->set_should_reexecute(true);
1525
1526 // Check if a null path was taken unconditionally.
1527 value = null_check(value);
1528
1529 RegionNode* bailout = new RegionNode(1);
1530 record_for_igvn(bailout);
1531
1532 // Range checks
1533 generate_negative_guard(offset, bailout);
1534 generate_negative_guard(length, bailout);
1535 generate_limit_guard(offset, length, load_array_length(value), bailout);
1536 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1537 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1538
1539 if (bailout->req() > 1) {
1540 PreserveJVMState pjvms(this);
1541 set_control(_gvn.transform(bailout));
1542 uncommon_trap(Deoptimization::Reason_intrinsic,
1543 Deoptimization::Action_maybe_recompile);
1544 }
1545 if (stopped()) {
1546 return true;
1547 }
1548
1549 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1550 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1551 newcopy = new_array(klass_node, size, 0); // no arguments to push
1552 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1553 guarantee(alloc != nullptr, "created above");
1554
1555 // Calculate starting addresses.
1556 Node* src_start = array_element_address(value, offset, T_CHAR);
1557 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1558
1559 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1560 const TypeInt* toffset = gvn().type(offset)->is_int();
1561 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1562
1563 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1564 const char* copyfunc_name = "arraycopy";
1565 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1566 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1567 OptoRuntime::fast_arraycopy_Type(),
1568 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1569 src_start, dst_start, ConvI2X(length) XTOP);
1570 // Do not let reads from the cloned object float above the arraycopy.
1571 if (alloc->maybe_set_complete(&_gvn)) {
1572 // "You break it, you buy it."
1573 InitializeNode* init = alloc->initialization();
1574 assert(init->is_complete(), "we just did this");
1575 init->set_complete_with_arraycopy();
1576 assert(newcopy->is_CheckCastPP(), "sanity");
1577 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1578 }
1579 // Do not let stores that initialize this object be reordered with
1580 // a subsequent store that would make this object accessible by
1581 // other threads.
1582 // Record what AllocateNode this StoreStore protects so that
1583 // escape analysis can go from the MemBarStoreStoreNode to the
1584 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1585 // based on the escape status of the AllocateNode.
1586 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1587 } // original reexecute is set back here
1588
1589 C->set_has_split_ifs(true); // Has chance for split-if optimization
1590 if (!stopped()) {
1591 set_result(newcopy);
1592 }
1593 clear_upper_avx();
1594
1595 return true;
1596 }
1597
1598 //------------------------inline_string_getCharsU--------------------------
1599 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1600 bool LibraryCallKit::inline_string_getCharsU() {
1601 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1602 return false;
1603 }
1604
1605 // Get the arguments.
1606 Node* src = argument(0);
1607 Node* src_begin = argument(1);
1608 Node* src_end = argument(2); // exclusive offset (i < src_end)
1609 Node* dst = argument(3);
1610 Node* dst_begin = argument(4);
1611
1612 // Check for allocation before we add nodes that would confuse
1613 // tightly_coupled_allocation()
1614 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1615
1616 // Check if a null path was taken unconditionally.
1617 src = null_check(src);
1618 dst = null_check(dst);
1619 if (stopped()) {
1620 return true;
1621 }
1622
1623 // Get length and convert char[] offset to byte[] offset
1624 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1625 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1626
1627 // Range checks
1628 generate_string_range_check(src, src_begin, length, true);
1629 generate_string_range_check(dst, dst_begin, length, false);
1630 if (stopped()) {
1631 return true;
1632 }
1633
1634 if (!stopped()) {
1635 // Calculate starting addresses.
1636 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1637 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1638
1639 // Check if array addresses are aligned to HeapWordSize
1640 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1641 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1642 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1643 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1644
1645 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1646 const char* copyfunc_name = "arraycopy";
1647 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1648 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1649 OptoRuntime::fast_arraycopy_Type(),
1650 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1651 src_start, dst_start, ConvI2X(length) XTOP);
1652 // Do not let reads from the cloned object float above the arraycopy.
1653 if (alloc != nullptr) {
1654 if (alloc->maybe_set_complete(&_gvn)) {
1655 // "You break it, you buy it."
1656 InitializeNode* init = alloc->initialization();
1657 assert(init->is_complete(), "we just did this");
1658 init->set_complete_with_arraycopy();
1659 assert(dst->is_CheckCastPP(), "sanity");
1660 assert(dst->in(0)->in(0) == init, "dest pinned");
1661 }
1662 // Do not let stores that initialize this object be reordered with
1663 // a subsequent store that would make this object accessible by
1664 // other threads.
1665 // Record what AllocateNode this StoreStore protects so that
1666 // escape analysis can go from the MemBarStoreStoreNode to the
1667 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1668 // based on the escape status of the AllocateNode.
1669 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1670 } else {
1671 insert_mem_bar(Op_MemBarCPUOrder);
1672 }
1673 }
1674
1675 C->set_has_split_ifs(true); // Has chance for split-if optimization
1676 return true;
1677 }
1678
1679 //----------------------inline_string_char_access----------------------------
1680 // Store/Load char to/from byte[] array.
1681 // static void StringUTF16.putChar(byte[] val, int index, int c)
1682 // static char StringUTF16.getChar(byte[] val, int index)
1683 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1684 Node* value = argument(0);
1685 Node* index = argument(1);
1686 Node* ch = is_store ? argument(2) : nullptr;
1687
1688 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1689 // correctly requires matched array shapes.
1690 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1691 "sanity: byte[] and char[] bases agree");
1692 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1693 "sanity: byte[] and char[] scales agree");
1694
1695 // Bail when getChar over constants is requested: constant folding would
1696 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1697 // Java method would constant fold nicely instead.
1698 if (!is_store && value->is_Con() && index->is_Con()) {
1699 return false;
1700 }
1701
1702 // Save state and restore on bailout
1703 uint old_sp = sp();
1704 SafePointNode* old_map = clone_map();
1705
1706 value = must_be_not_null(value, true);
1707
1708 Node* adr = array_element_address(value, index, T_CHAR);
1709 if (adr->is_top()) {
1710 set_map(old_map);
1711 set_sp(old_sp);
1712 return false;
1713 }
1714 destruct_map_clone(old_map);
1715 if (is_store) {
1716 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1717 } else {
1718 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);
1719 set_result(ch);
1720 }
1721 return true;
1722 }
1723
1724
1725 //------------------------------inline_math-----------------------------------
1726 // public static double Math.abs(double)
1727 // public static double Math.sqrt(double)
1728 // public static double Math.log(double)
1729 // public static double Math.log10(double)
1730 // public static double Math.round(double)
1731 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1732 Node* arg = argument(0);
1733 Node* n = nullptr;
1734 switch (id) {
1735 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1736 case vmIntrinsics::_dsqrt:
1737 case vmIntrinsics::_dsqrt_strict:
1738 n = new SqrtDNode(C, control(), arg); break;
1739 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1740 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1741 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1742 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1743 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1744 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1745 default: fatal_unexpected_iid(id); break;
1746 }
1747 set_result(_gvn.transform(n));
1748 return true;
1749 }
1750
1751 //------------------------------inline_math-----------------------------------
1752 // public static float Math.abs(float)
1753 // public static int Math.abs(int)
1754 // public static long Math.abs(long)
1755 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1756 Node* arg = argument(0);
1757 Node* n = nullptr;
1758 switch (id) {
1759 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1760 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1761 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1762 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1763 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1764 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1765 default: fatal_unexpected_iid(id); break;
1766 }
1767 set_result(_gvn.transform(n));
1768 return true;
1769 }
1770
1771 //------------------------------runtime_math-----------------------------
1772 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1773 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1774 "must be (DD)D or (D)D type");
1775
1776 // Inputs
1777 Node* a = argument(0);
1778 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1779
1780 const TypePtr* no_memory_effects = nullptr;
1781 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1782 no_memory_effects,
1783 a, top(), b, b ? top() : nullptr);
1784 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1785 #ifdef ASSERT
1786 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1787 assert(value_top == top(), "second value must be top");
1788 #endif
1789
1790 set_result(value);
1791 return true;
1792 }
1793
1794 //------------------------------inline_math_pow-----------------------------
1795 bool LibraryCallKit::inline_math_pow() {
1796 Node* exp = argument(2);
1797 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1798 if (d != nullptr) {
1799 if (d->getd() == 2.0) {
1800 // Special case: pow(x, 2.0) => x * x
1801 Node* base = argument(0);
1802 set_result(_gvn.transform(new MulDNode(base, base)));
1803 return true;
1804 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1805 // Special case: pow(x, 0.5) => sqrt(x)
1806 Node* base = argument(0);
1807 Node* zero = _gvn.zerocon(T_DOUBLE);
1808
1809 RegionNode* region = new RegionNode(3);
1810 Node* phi = new PhiNode(region, Type::DOUBLE);
1811
1812 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1813 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1814 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1815 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1816 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1817
1818 Node* if_pow = generate_slow_guard(test, nullptr);
1819 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1820 phi->init_req(1, value_sqrt);
1821 region->init_req(1, control());
1822
1823 if (if_pow != nullptr) {
1824 set_control(if_pow);
1825 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1826 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1827 const TypePtr* no_memory_effects = nullptr;
1828 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1829 no_memory_effects, base, top(), exp, top());
1830 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1831 #ifdef ASSERT
1832 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1833 assert(value_top == top(), "second value must be top");
1834 #endif
1835 phi->init_req(2, value_pow);
1836 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1837 }
1838
1839 C->set_has_split_ifs(true); // Has chance for split-if optimization
1840 set_control(_gvn.transform(region));
1841 record_for_igvn(region);
1842 set_result(_gvn.transform(phi));
1843
1844 return true;
1845 }
1846 }
1847
1848 return StubRoutines::dpow() != nullptr ?
1849 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1850 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1851 }
1852
1853 //------------------------------inline_math_native-----------------------------
1854 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1855 switch (id) {
1856 case vmIntrinsics::_dsin:
1857 return StubRoutines::dsin() != nullptr ?
1858 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1859 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1860 case vmIntrinsics::_dcos:
1861 return StubRoutines::dcos() != nullptr ?
1862 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1863 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1864 case vmIntrinsics::_dtan:
1865 return StubRoutines::dtan() != nullptr ?
1866 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1867 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1868 case vmIntrinsics::_dtanh:
1869 return StubRoutines::dtanh() != nullptr ?
1870 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1871 case vmIntrinsics::_dexp:
1872 return StubRoutines::dexp() != nullptr ?
1873 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1874 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1875 case vmIntrinsics::_dlog:
1876 return StubRoutines::dlog() != nullptr ?
1877 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1878 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1879 case vmIntrinsics::_dlog10:
1880 return StubRoutines::dlog10() != nullptr ?
1881 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1882 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1883
1884 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1885 case vmIntrinsics::_ceil:
1886 case vmIntrinsics::_floor:
1887 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1888
1889 case vmIntrinsics::_dsqrt:
1890 case vmIntrinsics::_dsqrt_strict:
1891 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1892 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1893 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1894 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1895 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1896
1897 case vmIntrinsics::_dpow: return inline_math_pow();
1898 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1899 case vmIntrinsics::_fcopySign: return inline_math(id);
1900 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1901 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1902 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1903
1904 // These intrinsics are not yet correctly implemented
1905 case vmIntrinsics::_datan2:
1906 return false;
1907
1908 default:
1909 fatal_unexpected_iid(id);
1910 return false;
1911 }
1912 }
1913
1914 //----------------------------inline_notify-----------------------------------*
1915 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1916 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1917 address func;
1918 if (id == vmIntrinsics::_notify) {
1919 func = OptoRuntime::monitor_notify_Java();
1920 } else {
1921 func = OptoRuntime::monitor_notifyAll_Java();
1922 }
1923 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1924 make_slow_call_ex(call, env()->Throwable_klass(), false);
1925 return true;
1926 }
1927
1928
1929 //----------------------------inline_min_max-----------------------------------
1930 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1931 Node* a = nullptr;
1932 Node* b = nullptr;
1933 Node* n = nullptr;
1934 switch (id) {
1935 case vmIntrinsics::_min:
1936 case vmIntrinsics::_max:
1937 case vmIntrinsics::_minF:
1938 case vmIntrinsics::_maxF:
1939 case vmIntrinsics::_minF_strict:
1940 case vmIntrinsics::_maxF_strict:
1941 case vmIntrinsics::_min_strict:
1942 case vmIntrinsics::_max_strict:
1943 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
1944 a = argument(0);
1945 b = argument(1);
1946 break;
1947 case vmIntrinsics::_minD:
1948 case vmIntrinsics::_maxD:
1949 case vmIntrinsics::_minD_strict:
1950 case vmIntrinsics::_maxD_strict:
1951 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
1952 a = argument(0);
1953 b = argument(2);
1954 break;
1955 case vmIntrinsics::_minL:
1956 case vmIntrinsics::_maxL:
1957 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
1958 a = argument(0);
1959 b = argument(2);
1960 break;
1961 default:
1962 fatal_unexpected_iid(id);
1963 break;
1964 }
1965
1966 switch (id) {
1967 case vmIntrinsics::_min:
1968 case vmIntrinsics::_min_strict:
1969 n = new MinINode(a, b);
1970 break;
1971 case vmIntrinsics::_max:
1972 case vmIntrinsics::_max_strict:
1973 n = new MaxINode(a, b);
1974 break;
1975 case vmIntrinsics::_minF:
1976 case vmIntrinsics::_minF_strict:
1977 n = new MinFNode(a, b);
1978 break;
1979 case vmIntrinsics::_maxF:
1980 case vmIntrinsics::_maxF_strict:
1981 n = new MaxFNode(a, b);
1982 break;
1983 case vmIntrinsics::_minD:
1984 case vmIntrinsics::_minD_strict:
1985 n = new MinDNode(a, b);
1986 break;
1987 case vmIntrinsics::_maxD:
1988 case vmIntrinsics::_maxD_strict:
1989 n = new MaxDNode(a, b);
1990 break;
1991 case vmIntrinsics::_minL:
1992 n = new MinLNode(_gvn.C, a, b);
1993 break;
1994 case vmIntrinsics::_maxL:
1995 n = new MaxLNode(_gvn.C, a, b);
1996 break;
1997 default:
1998 fatal_unexpected_iid(id);
1999 break;
2000 }
2001
2002 set_result(_gvn.transform(n));
2003 return true;
2004 }
2005
2006 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
2007 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
2008 env()->ArithmeticException_instance())) {
2009 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
2010 // so let's bail out intrinsic rather than risking deopting again.
2011 return false;
2012 }
2013
2014 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2015 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2016 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2017 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2018
2019 {
2020 PreserveJVMState pjvms(this);
2021 PreserveReexecuteState preexecs(this);
2022 jvms()->set_should_reexecute(true);
2023
2024 set_control(slow_path);
2025 set_i_o(i_o());
2026
2027 builtin_throw(Deoptimization::Reason_intrinsic,
2028 env()->ArithmeticException_instance(),
2029 /*allow_too_many_traps*/ false);
2030 }
2031
2032 set_control(fast_path);
2033 set_result(math);
2034 return true;
2035 }
2036
2037 template <typename OverflowOp>
2038 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2039 typedef typename OverflowOp::MathOp MathOp;
2040
2041 MathOp* mathOp = new MathOp(arg1, arg2);
2042 Node* operation = _gvn.transform( mathOp );
2043 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2044 return inline_math_mathExact(operation, ofcheck);
2045 }
2046
2047 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2048 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2049 }
2050
2051 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2052 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2053 }
2054
2055 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2056 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2057 }
2058
2059 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2060 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2061 }
2062
2063 bool LibraryCallKit::inline_math_negateExactI() {
2064 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2065 }
2066
2067 bool LibraryCallKit::inline_math_negateExactL() {
2068 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2069 }
2070
2071 bool LibraryCallKit::inline_math_multiplyExactI() {
2072 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2073 }
2074
2075 bool LibraryCallKit::inline_math_multiplyExactL() {
2076 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2077 }
2078
2079 bool LibraryCallKit::inline_math_multiplyHigh() {
2080 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2081 return true;
2082 }
2083
2084 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2085 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2086 return true;
2087 }
2088
2089 inline int
2090 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2091 const TypePtr* base_type = TypePtr::NULL_PTR;
2092 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2093 if (base_type == nullptr) {
2094 // Unknown type.
2095 return Type::AnyPtr;
2096 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2097 // Since this is a null+long form, we have to switch to a rawptr.
2098 base = _gvn.transform(new CastX2PNode(offset));
2099 offset = MakeConX(0);
2100 return Type::RawPtr;
2101 } else if (base_type->base() == Type::RawPtr) {
2102 return Type::RawPtr;
2103 } else if (base_type->isa_oopptr()) {
2104 // Base is never null => always a heap address.
2105 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2106 return Type::OopPtr;
2107 }
2108 // Offset is small => always a heap address.
2109 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2110 if (offset_type != nullptr &&
2111 base_type->offset() == 0 && // (should always be?)
2112 offset_type->_lo >= 0 &&
2113 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2114 return Type::OopPtr;
2115 } else if (type == T_OBJECT) {
2116 // off heap access to an oop doesn't make any sense. Has to be on
2117 // heap.
2118 return Type::OopPtr;
2119 }
2120 // Otherwise, it might either be oop+off or null+addr.
2121 return Type::AnyPtr;
2122 } else {
2123 // No information:
2124 return Type::AnyPtr;
2125 }
2126 }
2127
2128 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2129 Node* uncasted_base = base;
2130 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2131 if (kind == Type::RawPtr) {
2132 return basic_plus_adr(top(), uncasted_base, offset);
2133 } else if (kind == Type::AnyPtr) {
2134 assert(base == uncasted_base, "unexpected base change");
2135 if (can_cast) {
2136 if (!_gvn.type(base)->speculative_maybe_null() &&
2137 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2138 // According to profiling, this access is always on
2139 // heap. Casting the base to not null and thus avoiding membars
2140 // around the access should allow better optimizations
2141 Node* null_ctl = top();
2142 base = null_check_oop(base, &null_ctl, true, true, true);
2143 assert(null_ctl->is_top(), "no null control here");
2144 return basic_plus_adr(base, offset);
2145 } else if (_gvn.type(base)->speculative_always_null() &&
2146 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2147 // According to profiling, this access is always off
2148 // heap.
2149 base = null_assert(base);
2150 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2151 offset = MakeConX(0);
2152 return basic_plus_adr(top(), raw_base, offset);
2153 }
2154 }
2155 // We don't know if it's an on heap or off heap access. Fall back
2156 // to raw memory access.
2157 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2158 return basic_plus_adr(top(), raw, offset);
2159 } else {
2160 assert(base == uncasted_base, "unexpected base change");
2161 // We know it's an on heap access so base can't be null
2162 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2163 base = must_be_not_null(base, true);
2164 }
2165 return basic_plus_adr(base, offset);
2166 }
2167 }
2168
2169 //--------------------------inline_number_methods-----------------------------
2170 // inline int Integer.numberOfLeadingZeros(int)
2171 // inline int Long.numberOfLeadingZeros(long)
2172 //
2173 // inline int Integer.numberOfTrailingZeros(int)
2174 // inline int Long.numberOfTrailingZeros(long)
2175 //
2176 // inline int Integer.bitCount(int)
2177 // inline int Long.bitCount(long)
2178 //
2179 // inline char Character.reverseBytes(char)
2180 // inline short Short.reverseBytes(short)
2181 // inline int Integer.reverseBytes(int)
2182 // inline long Long.reverseBytes(long)
2183 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2184 Node* arg = argument(0);
2185 Node* n = nullptr;
2186 switch (id) {
2187 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2188 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2189 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2190 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2191 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2192 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2193 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2194 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2195 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2196 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2197 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2198 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2199 default: fatal_unexpected_iid(id); break;
2200 }
2201 set_result(_gvn.transform(n));
2202 return true;
2203 }
2204
2205 //--------------------------inline_bitshuffle_methods-----------------------------
2206 // inline int Integer.compress(int, int)
2207 // inline int Integer.expand(int, int)
2208 // inline long Long.compress(long, long)
2209 // inline long Long.expand(long, long)
2210 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2211 Node* n = nullptr;
2212 switch (id) {
2213 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2214 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2215 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2216 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2217 default: fatal_unexpected_iid(id); break;
2218 }
2219 set_result(_gvn.transform(n));
2220 return true;
2221 }
2222
2223 //--------------------------inline_number_methods-----------------------------
2224 // inline int Integer.compareUnsigned(int, int)
2225 // inline int Long.compareUnsigned(long, long)
2226 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2227 Node* arg1 = argument(0);
2228 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2229 Node* n = nullptr;
2230 switch (id) {
2231 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2232 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2233 default: fatal_unexpected_iid(id); break;
2234 }
2235 set_result(_gvn.transform(n));
2236 return true;
2237 }
2238
2239 //--------------------------inline_unsigned_divmod_methods-----------------------------
2240 // inline int Integer.divideUnsigned(int, int)
2241 // inline int Integer.remainderUnsigned(int, int)
2242 // inline long Long.divideUnsigned(long, long)
2243 // inline long Long.remainderUnsigned(long, long)
2244 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2245 Node* n = nullptr;
2246 switch (id) {
2247 case vmIntrinsics::_divideUnsigned_i: {
2248 zero_check_int(argument(1));
2249 // Compile-time detect of null-exception
2250 if (stopped()) {
2251 return true; // keep the graph constructed so far
2252 }
2253 n = new UDivINode(control(), argument(0), argument(1));
2254 break;
2255 }
2256 case vmIntrinsics::_divideUnsigned_l: {
2257 zero_check_long(argument(2));
2258 // Compile-time detect of null-exception
2259 if (stopped()) {
2260 return true; // keep the graph constructed so far
2261 }
2262 n = new UDivLNode(control(), argument(0), argument(2));
2263 break;
2264 }
2265 case vmIntrinsics::_remainderUnsigned_i: {
2266 zero_check_int(argument(1));
2267 // Compile-time detect of null-exception
2268 if (stopped()) {
2269 return true; // keep the graph constructed so far
2270 }
2271 n = new UModINode(control(), argument(0), argument(1));
2272 break;
2273 }
2274 case vmIntrinsics::_remainderUnsigned_l: {
2275 zero_check_long(argument(2));
2276 // Compile-time detect of null-exception
2277 if (stopped()) {
2278 return true; // keep the graph constructed so far
2279 }
2280 n = new UModLNode(control(), argument(0), argument(2));
2281 break;
2282 }
2283 default: fatal_unexpected_iid(id); break;
2284 }
2285 set_result(_gvn.transform(n));
2286 return true;
2287 }
2288
2289 //----------------------------inline_unsafe_access----------------------------
2290
2291 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2292 // Attempt to infer a sharper value type from the offset and base type.
2293 ciKlass* sharpened_klass = nullptr;
2294
2295 // See if it is an instance field, with an object type.
2296 if (alias_type->field() != nullptr) {
2297 if (alias_type->field()->type()->is_klass()) {
2298 sharpened_klass = alias_type->field()->type()->as_klass();
2299 }
2300 }
2301
2302 const TypeOopPtr* result = nullptr;
2303 // See if it is a narrow oop array.
2304 if (adr_type->isa_aryptr()) {
2305 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2306 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2307 if (elem_type != nullptr && elem_type->is_loaded()) {
2308 // Sharpen the value type.
2309 result = elem_type;
2310 }
2311 }
2312 }
2313
2314 // The sharpened class might be unloaded if there is no class loader
2315 // contraint in place.
2316 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2317 // Sharpen the value type.
2318 result = TypeOopPtr::make_from_klass(sharpened_klass);
2319 }
2320 if (result != nullptr) {
2321 #ifndef PRODUCT
2322 if (C->print_intrinsics() || C->print_inlining()) {
2323 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2324 tty->print(" sharpened value: "); result->dump(); tty->cr();
2325 }
2326 #endif
2327 }
2328 return result;
2329 }
2330
2331 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2332 switch (kind) {
2333 case Relaxed:
2334 return MO_UNORDERED;
2335 case Opaque:
2336 return MO_RELAXED;
2337 case Acquire:
2338 return MO_ACQUIRE;
2339 case Release:
2340 return MO_RELEASE;
2341 case Volatile:
2342 return MO_SEQ_CST;
2343 default:
2344 ShouldNotReachHere();
2345 return 0;
2346 }
2347 }
2348
2349 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2350 if (callee()->is_static()) return false; // caller must have the capability!
2351 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2352 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2353 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2354 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2355
2356 if (is_reference_type(type)) {
2357 decorators |= ON_UNKNOWN_OOP_REF;
2358 }
2359
2360 if (unaligned) {
2361 decorators |= C2_UNALIGNED;
2362 }
2363
2364 #ifndef PRODUCT
2365 {
2366 ResourceMark rm;
2367 // Check the signatures.
2368 ciSignature* sig = callee()->signature();
2369 #ifdef ASSERT
2370 if (!is_store) {
2371 // Object getReference(Object base, int/long offset), etc.
2372 BasicType rtype = sig->return_type()->basic_type();
2373 assert(rtype == type, "getter must return the expected value");
2374 assert(sig->count() == 2, "oop getter has 2 arguments");
2375 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2376 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2377 } else {
2378 // void putReference(Object base, int/long offset, Object x), etc.
2379 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2380 assert(sig->count() == 3, "oop putter has 3 arguments");
2381 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2382 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2383 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2384 assert(vtype == type, "putter must accept the expected value");
2385 }
2386 #endif // ASSERT
2387 }
2388 #endif //PRODUCT
2389
2390 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2391
2392 Node* receiver = argument(0); // type: oop
2393
2394 // Build address expression.
2395 Node* heap_base_oop = top();
2396
2397 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2398 Node* base = argument(1); // type: oop
2399 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2400 Node* offset = argument(2); // type: long
2401 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2402 // to be plain byte offsets, which are also the same as those accepted
2403 // by oopDesc::field_addr.
2404 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2405 "fieldOffset must be byte-scaled");
2406 // 32-bit machines ignore the high half!
2407 offset = ConvL2X(offset);
2408
2409 // Save state and restore on bailout
2410 uint old_sp = sp();
2411 SafePointNode* old_map = clone_map();
2412
2413 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2414 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2415
2416 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2417 if (type != T_OBJECT) {
2418 decorators |= IN_NATIVE; // off-heap primitive access
2419 } else {
2420 set_map(old_map);
2421 set_sp(old_sp);
2422 return false; // off-heap oop accesses are not supported
2423 }
2424 } else {
2425 heap_base_oop = base; // on-heap or mixed access
2426 }
2427
2428 // Can base be null? Otherwise, always on-heap access.
2429 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2430
2431 if (!can_access_non_heap) {
2432 decorators |= IN_HEAP;
2433 }
2434
2435 Node* val = is_store ? argument(4) : nullptr;
2436
2437 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2438 if (adr_type == TypePtr::NULL_PTR) {
2439 set_map(old_map);
2440 set_sp(old_sp);
2441 return false; // off-heap access with zero address
2442 }
2443
2444 // Try to categorize the address.
2445 Compile::AliasType* alias_type = C->alias_type(adr_type);
2446 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2447
2448 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2449 alias_type->adr_type() == TypeAryPtr::RANGE) {
2450 set_map(old_map);
2451 set_sp(old_sp);
2452 return false; // not supported
2453 }
2454
2455 bool mismatched = false;
2456 BasicType bt = alias_type->basic_type();
2457 if (bt != T_ILLEGAL) {
2458 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2459 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2460 // Alias type doesn't differentiate between byte[] and boolean[]).
2461 // Use address type to get the element type.
2462 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2463 }
2464 if (is_reference_type(bt, true)) {
2465 // accessing an array field with getReference is not a mismatch
2466 bt = T_OBJECT;
2467 }
2468 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2469 // Don't intrinsify mismatched object accesses
2470 set_map(old_map);
2471 set_sp(old_sp);
2472 return false;
2473 }
2474 mismatched = (bt != type);
2475 } else if (alias_type->adr_type()->isa_oopptr()) {
2476 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2477 }
2478
2479 destruct_map_clone(old_map);
2480 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2481
2482 if (mismatched) {
2483 decorators |= C2_MISMATCHED;
2484 }
2485
2486 // First guess at the value type.
2487 const Type *value_type = Type::get_const_basic_type(type);
2488
2489 // Figure out the memory ordering.
2490 decorators |= mo_decorator_for_access_kind(kind);
2491
2492 if (!is_store && type == T_OBJECT) {
2493 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2494 if (tjp != nullptr) {
2495 value_type = tjp;
2496 }
2497 }
2498
2499 receiver = null_check(receiver);
2500 if (stopped()) {
2501 return true;
2502 }
2503 // Heap pointers get a null-check from the interpreter,
2504 // as a courtesy. However, this is not guaranteed by Unsafe,
2505 // and it is not possible to fully distinguish unintended nulls
2506 // from intended ones in this API.
2507
2508 if (!is_store) {
2509 Node* p = nullptr;
2510 // Try to constant fold a load from a constant field
2511 ciField* field = alias_type->field();
2512 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) {
2513 // final or stable field
2514 p = make_constant_from_field(field, heap_base_oop);
2515 }
2516
2517 if (p == nullptr) { // Could not constant fold the load
2518 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2519 // Normalize the value returned by getBoolean in the following cases
2520 if (type == T_BOOLEAN &&
2521 (mismatched ||
2522 heap_base_oop == top() || // - heap_base_oop is null or
2523 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2524 // and the unsafe access is made to large offset
2525 // (i.e., larger than the maximum offset necessary for any
2526 // field access)
2527 ) {
2528 IdealKit ideal = IdealKit(this);
2529 #define __ ideal.
2530 IdealVariable normalized_result(ideal);
2531 __ declarations_done();
2532 __ set(normalized_result, p);
2533 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2534 __ set(normalized_result, ideal.ConI(1));
2535 ideal.end_if();
2536 final_sync(ideal);
2537 p = __ value(normalized_result);
2538 #undef __
2539 }
2540 }
2541 if (type == T_ADDRESS) {
2542 p = gvn().transform(new CastP2XNode(nullptr, p));
2543 p = ConvX2UL(p);
2544 }
2545 // The load node has the control of the preceding MemBarCPUOrder. All
2546 // following nodes will have the control of the MemBarCPUOrder inserted at
2547 // the end of this method. So, pushing the load onto the stack at a later
2548 // point is fine.
2549 set_result(p);
2550 } else {
2551 if (bt == T_ADDRESS) {
2552 // Repackage the long as a pointer.
2553 val = ConvL2X(val);
2554 val = gvn().transform(new CastX2PNode(val));
2555 }
2556 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2557 }
2558
2559 return true;
2560 }
2561
2562 //----------------------------inline_unsafe_load_store----------------------------
2563 // This method serves a couple of different customers (depending on LoadStoreKind):
2564 //
2565 // LS_cmp_swap:
2566 //
2567 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2568 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2569 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2570 //
2571 // LS_cmp_swap_weak:
2572 //
2573 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2574 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2575 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2576 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2577 //
2578 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2579 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2580 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2581 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2582 //
2583 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2584 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2585 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2586 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2587 //
2588 // LS_cmp_exchange:
2589 //
2590 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2591 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2592 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2593 //
2594 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2595 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2596 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2597 //
2598 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2599 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2600 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2601 //
2602 // LS_get_add:
2603 //
2604 // int getAndAddInt( Object o, long offset, int delta)
2605 // long getAndAddLong(Object o, long offset, long delta)
2606 //
2607 // LS_get_set:
2608 //
2609 // int getAndSet(Object o, long offset, int newValue)
2610 // long getAndSet(Object o, long offset, long newValue)
2611 // Object getAndSet(Object o, long offset, Object newValue)
2612 //
2613 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2614 // This basic scheme here is the same as inline_unsafe_access, but
2615 // differs in enough details that combining them would make the code
2616 // overly confusing. (This is a true fact! I originally combined
2617 // them, but even I was confused by it!) As much code/comments as
2618 // possible are retained from inline_unsafe_access though to make
2619 // the correspondences clearer. - dl
2620
2621 if (callee()->is_static()) return false; // caller must have the capability!
2622
2623 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2624 decorators |= mo_decorator_for_access_kind(access_kind);
2625
2626 #ifndef PRODUCT
2627 BasicType rtype;
2628 {
2629 ResourceMark rm;
2630 // Check the signatures.
2631 ciSignature* sig = callee()->signature();
2632 rtype = sig->return_type()->basic_type();
2633 switch(kind) {
2634 case LS_get_add:
2635 case LS_get_set: {
2636 // Check the signatures.
2637 #ifdef ASSERT
2638 assert(rtype == type, "get and set must return the expected type");
2639 assert(sig->count() == 3, "get and set has 3 arguments");
2640 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2641 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2642 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2643 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2644 #endif // ASSERT
2645 break;
2646 }
2647 case LS_cmp_swap:
2648 case LS_cmp_swap_weak: {
2649 // Check the signatures.
2650 #ifdef ASSERT
2651 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2652 assert(sig->count() == 4, "CAS has 4 arguments");
2653 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2654 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2655 #endif // ASSERT
2656 break;
2657 }
2658 case LS_cmp_exchange: {
2659 // Check the signatures.
2660 #ifdef ASSERT
2661 assert(rtype == type, "CAS must return the expected type");
2662 assert(sig->count() == 4, "CAS has 4 arguments");
2663 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2664 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2665 #endif // ASSERT
2666 break;
2667 }
2668 default:
2669 ShouldNotReachHere();
2670 }
2671 }
2672 #endif //PRODUCT
2673
2674 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2675
2676 // Get arguments:
2677 Node* receiver = nullptr;
2678 Node* base = nullptr;
2679 Node* offset = nullptr;
2680 Node* oldval = nullptr;
2681 Node* newval = nullptr;
2682 switch(kind) {
2683 case LS_cmp_swap:
2684 case LS_cmp_swap_weak:
2685 case LS_cmp_exchange: {
2686 const bool two_slot_type = type2size[type] == 2;
2687 receiver = argument(0); // type: oop
2688 base = argument(1); // type: oop
2689 offset = argument(2); // type: long
2690 oldval = argument(4); // type: oop, int, or long
2691 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2692 break;
2693 }
2694 case LS_get_add:
2695 case LS_get_set: {
2696 receiver = argument(0); // type: oop
2697 base = argument(1); // type: oop
2698 offset = argument(2); // type: long
2699 oldval = nullptr;
2700 newval = argument(4); // type: oop, int, or long
2701 break;
2702 }
2703 default:
2704 ShouldNotReachHere();
2705 }
2706
2707 // Build field offset expression.
2708 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2709 // to be plain byte offsets, which are also the same as those accepted
2710 // by oopDesc::field_addr.
2711 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2712 // 32-bit machines ignore the high half of long offsets
2713 offset = ConvL2X(offset);
2714 // Save state and restore on bailout
2715 uint old_sp = sp();
2716 SafePointNode* old_map = clone_map();
2717 Node* adr = make_unsafe_address(base, offset,type, false);
2718 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2719
2720 Compile::AliasType* alias_type = C->alias_type(adr_type);
2721 BasicType bt = alias_type->basic_type();
2722 if (bt != T_ILLEGAL &&
2723 (is_reference_type(bt) != (type == T_OBJECT))) {
2724 // Don't intrinsify mismatched object accesses.
2725 set_map(old_map);
2726 set_sp(old_sp);
2727 return false;
2728 }
2729
2730 destruct_map_clone(old_map);
2731
2732 // For CAS, unlike inline_unsafe_access, there seems no point in
2733 // trying to refine types. Just use the coarse types here.
2734 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2735 const Type *value_type = Type::get_const_basic_type(type);
2736
2737 switch (kind) {
2738 case LS_get_set:
2739 case LS_cmp_exchange: {
2740 if (type == T_OBJECT) {
2741 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2742 if (tjp != nullptr) {
2743 value_type = tjp;
2744 }
2745 }
2746 break;
2747 }
2748 case LS_cmp_swap:
2749 case LS_cmp_swap_weak:
2750 case LS_get_add:
2751 break;
2752 default:
2753 ShouldNotReachHere();
2754 }
2755
2756 // Null check receiver.
2757 receiver = null_check(receiver);
2758 if (stopped()) {
2759 return true;
2760 }
2761
2762 int alias_idx = C->get_alias_index(adr_type);
2763
2764 if (is_reference_type(type)) {
2765 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2766
2767 // Transformation of a value which could be null pointer (CastPP #null)
2768 // could be delayed during Parse (for example, in adjust_map_after_if()).
2769 // Execute transformation here to avoid barrier generation in such case.
2770 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2771 newval = _gvn.makecon(TypePtr::NULL_PTR);
2772
2773 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2774 // Refine the value to a null constant, when it is known to be null
2775 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2776 }
2777 }
2778
2779 Node* result = nullptr;
2780 switch (kind) {
2781 case LS_cmp_exchange: {
2782 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2783 oldval, newval, value_type, type, decorators);
2784 break;
2785 }
2786 case LS_cmp_swap_weak:
2787 decorators |= C2_WEAK_CMPXCHG;
2788 case LS_cmp_swap: {
2789 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2790 oldval, newval, value_type, type, decorators);
2791 break;
2792 }
2793 case LS_get_set: {
2794 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2795 newval, value_type, type, decorators);
2796 break;
2797 }
2798 case LS_get_add: {
2799 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2800 newval, value_type, type, decorators);
2801 break;
2802 }
2803 default:
2804 ShouldNotReachHere();
2805 }
2806
2807 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2808 set_result(result);
2809 return true;
2810 }
2811
2812 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2813 // Regardless of form, don't allow previous ld/st to move down,
2814 // then issue acquire, release, or volatile mem_bar.
2815 insert_mem_bar(Op_MemBarCPUOrder);
2816 switch(id) {
2817 case vmIntrinsics::_loadFence:
2818 insert_mem_bar(Op_LoadFence);
2819 return true;
2820 case vmIntrinsics::_storeFence:
2821 insert_mem_bar(Op_StoreFence);
2822 return true;
2823 case vmIntrinsics::_storeStoreFence:
2824 insert_mem_bar(Op_StoreStoreFence);
2825 return true;
2826 case vmIntrinsics::_fullFence:
2827 insert_mem_bar(Op_MemBarVolatile);
2828 return true;
2829 default:
2830 fatal_unexpected_iid(id);
2831 return false;
2832 }
2833 }
2834
2835 bool LibraryCallKit::inline_onspinwait() {
2836 insert_mem_bar(Op_OnSpinWait);
2837 return true;
2838 }
2839
2840 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2841 if (!kls->is_Con()) {
2842 return true;
2843 }
2844 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2845 if (klsptr == nullptr) {
2846 return true;
2847 }
2848 ciInstanceKlass* ik = klsptr->instance_klass();
2849 // don't need a guard for a klass that is already initialized
2850 return !ik->is_initialized();
2851 }
2852
2853 //----------------------------inline_unsafe_writeback0-------------------------
2854 // public native void Unsafe.writeback0(long address)
2855 bool LibraryCallKit::inline_unsafe_writeback0() {
2856 if (!Matcher::has_match_rule(Op_CacheWB)) {
2857 return false;
2858 }
2859 #ifndef PRODUCT
2860 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2861 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2862 ciSignature* sig = callee()->signature();
2863 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2864 #endif
2865 null_check_receiver(); // null-check, then ignore
2866 Node *addr = argument(1);
2867 addr = new CastX2PNode(addr);
2868 addr = _gvn.transform(addr);
2869 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2870 flush = _gvn.transform(flush);
2871 set_memory(flush, TypeRawPtr::BOTTOM);
2872 return true;
2873 }
2874
2875 //----------------------------inline_unsafe_writeback0-------------------------
2876 // public native void Unsafe.writeback0(long address)
2877 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2878 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2879 return false;
2880 }
2881 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2882 return false;
2883 }
2884 #ifndef PRODUCT
2885 assert(Matcher::has_match_rule(Op_CacheWB),
2886 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2887 : "found match rule for CacheWBPostSync but not CacheWB"));
2888
2889 #endif
2890 null_check_receiver(); // null-check, then ignore
2891 Node *sync;
2892 if (is_pre) {
2893 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2894 } else {
2895 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2896 }
2897 sync = _gvn.transform(sync);
2898 set_memory(sync, TypeRawPtr::BOTTOM);
2899 return true;
2900 }
2901
2902 //----------------------------inline_unsafe_allocate---------------------------
2903 // public native Object Unsafe.allocateInstance(Class<?> cls);
2904 bool LibraryCallKit::inline_unsafe_allocate() {
2905
2906 #if INCLUDE_JVMTI
2907 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
2908 return false;
2909 }
2910 #endif //INCLUDE_JVMTI
2911
2912 if (callee()->is_static()) return false; // caller must have the capability!
2913
2914 null_check_receiver(); // null-check, then ignore
2915 Node* cls = null_check(argument(1));
2916 if (stopped()) return true;
2917
2918 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
2919 kls = null_check(kls);
2920 if (stopped()) return true; // argument was like int.class
2921
2922 #if INCLUDE_JVMTI
2923 // Don't try to access new allocated obj in the intrinsic.
2924 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
2925 // Deoptimize and allocate in interpreter instead.
2926 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
2927 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
2928 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
2929 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
2930 {
2931 BuildCutout unless(this, tst, PROB_MAX);
2932 uncommon_trap(Deoptimization::Reason_intrinsic,
2933 Deoptimization::Action_make_not_entrant);
2934 }
2935 if (stopped()) {
2936 return true;
2937 }
2938 #endif //INCLUDE_JVMTI
2939
2940 Node* test = nullptr;
2941 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2942 // Note: The argument might still be an illegal value like
2943 // Serializable.class or Object[].class. The runtime will handle it.
2944 // But we must make an explicit check for initialization.
2945 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2946 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2947 // can generate code to load it as unsigned byte.
2948 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
2949 Node* bits = intcon(InstanceKlass::fully_initialized);
2950 test = _gvn.transform(new SubINode(inst, bits));
2951 // The 'test' is non-zero if we need to take a slow path.
2952 }
2953
2954 Node* obj = new_instance(kls, test);
2955 set_result(obj);
2956 return true;
2957 }
2958
2959 //------------------------inline_native_time_funcs--------------
2960 // inline code for System.currentTimeMillis() and System.nanoTime()
2961 // these have the same type and signature
2962 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2963 const TypeFunc* tf = OptoRuntime::void_long_Type();
2964 const TypePtr* no_memory_effects = nullptr;
2965 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2966 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2967 #ifdef ASSERT
2968 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2969 assert(value_top == top(), "second value must be top");
2970 #endif
2971 set_result(value);
2972 return true;
2973 }
2974
2975
2976 #if INCLUDE_JVMTI
2977
2978 // When notifications are disabled then just update the VTMS transition bit and return.
2979 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
2980 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
2981 if (!DoJVMTIVirtualThreadTransitions) {
2982 return true;
2983 }
2984 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
2985 IdealKit ideal(this);
2986
2987 Node* ONE = ideal.ConI(1);
2988 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
2989 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
2990 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
2991
2992 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
2993 sync_kit(ideal);
2994 // if notifyJvmti enabled then make a call to the given SharedRuntime function
2995 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
2996 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
2997 ideal.sync_kit(this);
2998 } ideal.else_(); {
2999 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
3000 Node* thread = ideal.thread();
3001 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
3002 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
3003
3004 sync_kit(ideal);
3005 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3006 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3007
3008 ideal.sync_kit(this);
3009 } ideal.end_if();
3010 final_sync(ideal);
3011
3012 return true;
3013 }
3014
3015 // Always update the is_disable_suspend bit.
3016 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3017 if (!DoJVMTIVirtualThreadTransitions) {
3018 return true;
3019 }
3020 IdealKit ideal(this);
3021
3022 {
3023 // unconditionally update the is_disable_suspend bit in current JavaThread
3024 Node* thread = ideal.thread();
3025 Node* arg = _gvn.transform(argument(0)); // argument for notification
3026 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3027 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3028
3029 sync_kit(ideal);
3030 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3031 ideal.sync_kit(this);
3032 }
3033 final_sync(ideal);
3034
3035 return true;
3036 }
3037
3038 #endif // INCLUDE_JVMTI
3039
3040 #ifdef JFR_HAVE_INTRINSICS
3041
3042 /**
3043 * if oop->klass != null
3044 * // normal class
3045 * epoch = _epoch_state ? 2 : 1
3046 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3047 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3048 * }
3049 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3050 * else
3051 * // primitive class
3052 * if oop->array_klass != null
3053 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3054 * else
3055 * id = LAST_TYPE_ID + 1 // void class path
3056 * if (!signaled)
3057 * signaled = true
3058 */
3059 bool LibraryCallKit::inline_native_classID() {
3060 Node* cls = argument(0);
3061
3062 IdealKit ideal(this);
3063 #define __ ideal.
3064 IdealVariable result(ideal); __ declarations_done();
3065 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3066 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3067 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3068
3069
3070 __ if_then(kls, BoolTest::ne, null()); {
3071 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3072 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3073
3074 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3075 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3076 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3077 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3078 mask = _gvn.transform(new OrLNode(mask, epoch));
3079 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3080
3081 float unlikely = PROB_UNLIKELY(0.999);
3082 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3083 sync_kit(ideal);
3084 make_runtime_call(RC_LEAF,
3085 OptoRuntime::class_id_load_barrier_Type(),
3086 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3087 "class id load barrier",
3088 TypePtr::BOTTOM,
3089 kls);
3090 ideal.sync_kit(this);
3091 } __ end_if();
3092
3093 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3094 } __ else_(); {
3095 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3096 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3097 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3098 __ if_then(array_kls, BoolTest::ne, null()); {
3099 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3100 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3101 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3102 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3103 } __ else_(); {
3104 // void class case
3105 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3106 } __ end_if();
3107
3108 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3109 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3110 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3111 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3112 } __ end_if();
3113 } __ end_if();
3114
3115 final_sync(ideal);
3116 set_result(ideal.value(result));
3117 #undef __
3118 return true;
3119 }
3120
3121 //------------------------inline_native_jvm_commit------------------
3122 bool LibraryCallKit::inline_native_jvm_commit() {
3123 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3124
3125 // Save input memory and i_o state.
3126 Node* input_memory_state = reset_memory();
3127 set_all_memory(input_memory_state);
3128 Node* input_io_state = i_o();
3129
3130 // TLS.
3131 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3132 // Jfr java buffer.
3133 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3134 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3135 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3136
3137 // Load the current value of the notified field in the JfrThreadLocal.
3138 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3139 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3140
3141 // Test for notification.
3142 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3143 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3144 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3145
3146 // True branch, is notified.
3147 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3148 set_control(is_notified);
3149
3150 // Reset notified state.
3151 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3152 Node* notified_reset_memory = reset_memory();
3153
3154 // 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.
3155 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3156 // Convert the machine-word to a long.
3157 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3158
3159 // False branch, not notified.
3160 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3161 set_control(not_notified);
3162 set_all_memory(input_memory_state);
3163
3164 // Arg is the next position as a long.
3165 Node* arg = argument(0);
3166 // Convert long to machine-word.
3167 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3168
3169 // Store the next_position to the underlying jfr java buffer.
3170 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3171
3172 Node* commit_memory = reset_memory();
3173 set_all_memory(commit_memory);
3174
3175 // 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.
3176 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3177 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3178 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3179
3180 // And flags with lease constant.
3181 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3182
3183 // Branch on lease to conditionalize returning the leased java buffer.
3184 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3185 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3186 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3187
3188 // False branch, not a lease.
3189 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3190
3191 // True branch, is lease.
3192 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3193 set_control(is_lease);
3194
3195 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3196 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3197 OptoRuntime::void_void_Type(),
3198 SharedRuntime::jfr_return_lease(),
3199 "return_lease", TypePtr::BOTTOM);
3200 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3201
3202 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3203 record_for_igvn(lease_compare_rgn);
3204 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3205 record_for_igvn(lease_compare_mem);
3206 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3207 record_for_igvn(lease_compare_io);
3208 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3209 record_for_igvn(lease_result_value);
3210
3211 // Update control and phi nodes.
3212 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3213 lease_compare_rgn->init_req(_false_path, not_lease);
3214
3215 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3216 lease_compare_mem->init_req(_false_path, commit_memory);
3217
3218 lease_compare_io->init_req(_true_path, i_o());
3219 lease_compare_io->init_req(_false_path, input_io_state);
3220
3221 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3222 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3223
3224 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3225 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3226 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3227 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3228
3229 // Update control and phi nodes.
3230 result_rgn->init_req(_true_path, is_notified);
3231 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3232
3233 result_mem->init_req(_true_path, notified_reset_memory);
3234 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3235
3236 result_io->init_req(_true_path, input_io_state);
3237 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3238
3239 result_value->init_req(_true_path, current_pos);
3240 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3241
3242 // Set output state.
3243 set_control(_gvn.transform(result_rgn));
3244 set_all_memory(_gvn.transform(result_mem));
3245 set_i_o(_gvn.transform(result_io));
3246 set_result(result_rgn, result_value);
3247 return true;
3248 }
3249
3250 /*
3251 * The intrinsic is a model of this pseudo-code:
3252 *
3253 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3254 * jobject h_event_writer = tl->java_event_writer();
3255 * if (h_event_writer == nullptr) {
3256 * return nullptr;
3257 * }
3258 * oop threadObj = Thread::threadObj();
3259 * oop vthread = java_lang_Thread::vthread(threadObj);
3260 * traceid tid;
3261 * bool pinVirtualThread;
3262 * bool excluded;
3263 * if (vthread != threadObj) { // i.e. current thread is virtual
3264 * tid = java_lang_Thread::tid(vthread);
3265 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3266 * pinVirtualThread = VMContinuations;
3267 * excluded = vthread_epoch_raw & excluded_mask;
3268 * if (!excluded) {
3269 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3270 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3271 * if (vthread_epoch != current_epoch) {
3272 * write_checkpoint();
3273 * }
3274 * }
3275 * } else {
3276 * tid = java_lang_Thread::tid(threadObj);
3277 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3278 * pinVirtualThread = false;
3279 * excluded = thread_epoch_raw & excluded_mask;
3280 * }
3281 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3282 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3283 * if (tid_in_event_writer != tid) {
3284 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3285 * setField(event_writer, "excluded", excluded);
3286 * setField(event_writer, "threadID", tid);
3287 * }
3288 * return event_writer
3289 */
3290 bool LibraryCallKit::inline_native_getEventWriter() {
3291 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3292
3293 // Save input memory and i_o state.
3294 Node* input_memory_state = reset_memory();
3295 set_all_memory(input_memory_state);
3296 Node* input_io_state = i_o();
3297
3298 // The most significant bit of the u2 is used to denote thread exclusion
3299 Node* excluded_shift = _gvn.intcon(15);
3300 Node* excluded_mask = _gvn.intcon(1 << 15);
3301 // The epoch generation is the range [1-32767]
3302 Node* epoch_mask = _gvn.intcon(32767);
3303
3304 // TLS
3305 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3306
3307 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3308 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3309
3310 // Load the eventwriter jobject handle.
3311 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3312
3313 // Null check the jobject handle.
3314 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3315 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3316 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3317
3318 // False path, jobj is null.
3319 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3320
3321 // True path, jobj is not null.
3322 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3323
3324 set_control(jobj_is_not_null);
3325
3326 // Load the threadObj for the CarrierThread.
3327 Node* threadObj = generate_current_thread(tls_ptr);
3328
3329 // Load the vthread.
3330 Node* vthread = generate_virtual_thread(tls_ptr);
3331
3332 // If vthread != threadObj, this is a virtual thread.
3333 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3334 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3335 IfNode* iff_vthread_not_equal_threadObj =
3336 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3337
3338 // False branch, fallback to threadObj.
3339 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3340 set_control(vthread_equal_threadObj);
3341
3342 // Load the tid field from the vthread object.
3343 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3344
3345 // Load the raw epoch value from the threadObj.
3346 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3347 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3348 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3349 TypeInt::CHAR, T_CHAR,
3350 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3351
3352 // Mask off the excluded information from the epoch.
3353 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3354
3355 // True branch, this is a virtual thread.
3356 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3357 set_control(vthread_not_equal_threadObj);
3358
3359 // Load the tid field from the vthread object.
3360 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3361
3362 // Continuation support determines if a virtual thread should be pinned.
3363 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3364 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3365
3366 // Load the raw epoch value from the vthread.
3367 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3368 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3369 TypeInt::CHAR, T_CHAR,
3370 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3371
3372 // Mask off the excluded information from the epoch.
3373 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3374
3375 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3376 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3377 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3378 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3379
3380 // False branch, vthread is excluded, no need to write epoch info.
3381 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3382
3383 // True branch, vthread is included, update epoch info.
3384 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3385 set_control(included);
3386
3387 // Get epoch value.
3388 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3389
3390 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3391 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3392 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3393
3394 // Compare the epoch in the vthread to the current epoch generation.
3395 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3396 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3397 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3398
3399 // False path, epoch is equal, checkpoint information is valid.
3400 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3401
3402 // True path, epoch is not equal, write a checkpoint for the vthread.
3403 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3404
3405 set_control(epoch_is_not_equal);
3406
3407 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3408 // The call also updates the native thread local thread id and the vthread with the current epoch.
3409 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3410 OptoRuntime::jfr_write_checkpoint_Type(),
3411 SharedRuntime::jfr_write_checkpoint(),
3412 "write_checkpoint", TypePtr::BOTTOM);
3413 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3414
3415 // vthread epoch != current epoch
3416 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3417 record_for_igvn(epoch_compare_rgn);
3418 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3419 record_for_igvn(epoch_compare_mem);
3420 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3421 record_for_igvn(epoch_compare_io);
3422
3423 // Update control and phi nodes.
3424 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3425 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3426 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3427 epoch_compare_mem->init_req(_false_path, input_memory_state);
3428 epoch_compare_io->init_req(_true_path, i_o());
3429 epoch_compare_io->init_req(_false_path, input_io_state);
3430
3431 // excluded != true
3432 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3433 record_for_igvn(exclude_compare_rgn);
3434 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3435 record_for_igvn(exclude_compare_mem);
3436 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3437 record_for_igvn(exclude_compare_io);
3438
3439 // Update control and phi nodes.
3440 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3441 exclude_compare_rgn->init_req(_false_path, excluded);
3442 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3443 exclude_compare_mem->init_req(_false_path, input_memory_state);
3444 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3445 exclude_compare_io->init_req(_false_path, input_io_state);
3446
3447 // vthread != threadObj
3448 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3449 record_for_igvn(vthread_compare_rgn);
3450 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3451 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3452 record_for_igvn(vthread_compare_io);
3453 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3454 record_for_igvn(tid);
3455 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3456 record_for_igvn(exclusion);
3457 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3458 record_for_igvn(pinVirtualThread);
3459
3460 // Update control and phi nodes.
3461 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3462 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3463 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3464 vthread_compare_mem->init_req(_false_path, input_memory_state);
3465 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3466 vthread_compare_io->init_req(_false_path, input_io_state);
3467 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3468 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3469 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3470 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3471 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3472 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3473
3474 // Update branch state.
3475 set_control(_gvn.transform(vthread_compare_rgn));
3476 set_all_memory(_gvn.transform(vthread_compare_mem));
3477 set_i_o(_gvn.transform(vthread_compare_io));
3478
3479 // Load the event writer oop by dereferencing the jobject handle.
3480 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3481 assert(klass_EventWriter->is_loaded(), "invariant");
3482 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3483 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3484 const TypeOopPtr* const xtype = aklass->as_instance_type();
3485 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3486 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3487
3488 // Load the current thread id from the event writer object.
3489 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3490 // Get the field offset to, conditionally, store an updated tid value later.
3491 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3492 // Get the field offset to, conditionally, store an updated exclusion value later.
3493 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3494 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3495 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3496
3497 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3498 record_for_igvn(event_writer_tid_compare_rgn);
3499 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3500 record_for_igvn(event_writer_tid_compare_mem);
3501 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3502 record_for_igvn(event_writer_tid_compare_io);
3503
3504 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3505 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3506 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3507 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3508
3509 // False path, tids are the same.
3510 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3511
3512 // True path, tid is not equal, need to update the tid in the event writer.
3513 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3514 record_for_igvn(tid_is_not_equal);
3515
3516 // Store the pin state to the event writer.
3517 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3518
3519 // Store the exclusion state to the event writer.
3520 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3521 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3522
3523 // Store the tid to the event writer.
3524 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3525
3526 // Update control and phi nodes.
3527 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3528 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3529 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3530 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3531 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3532 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3533
3534 // Result of top level CFG, Memory, IO and Value.
3535 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3536 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3537 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3538 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3539
3540 // Result control.
3541 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3542 result_rgn->init_req(_false_path, jobj_is_null);
3543
3544 // Result memory.
3545 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3546 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3547
3548 // Result IO.
3549 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3550 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3551
3552 // Result value.
3553 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3554 result_value->init_req(_false_path, null()); // return null
3555
3556 // Set output state.
3557 set_control(_gvn.transform(result_rgn));
3558 set_all_memory(_gvn.transform(result_mem));
3559 set_i_o(_gvn.transform(result_io));
3560 set_result(result_rgn, result_value);
3561 return true;
3562 }
3563
3564 /*
3565 * The intrinsic is a model of this pseudo-code:
3566 *
3567 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3568 * if (carrierThread != thread) { // is virtual thread
3569 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3570 * bool excluded = vthread_epoch_raw & excluded_mask;
3571 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3572 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3573 * if (!excluded) {
3574 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3575 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3576 * }
3577 * Atomic::release_store(&tl->_vthread, true);
3578 * return;
3579 * }
3580 * Atomic::release_store(&tl->_vthread, false);
3581 */
3582 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3583 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3584
3585 Node* input_memory_state = reset_memory();
3586 set_all_memory(input_memory_state);
3587
3588 // The most significant bit of the u2 is used to denote thread exclusion
3589 Node* excluded_mask = _gvn.intcon(1 << 15);
3590 // The epoch generation is the range [1-32767]
3591 Node* epoch_mask = _gvn.intcon(32767);
3592
3593 Node* const carrierThread = generate_current_thread(jt);
3594 // If thread != carrierThread, this is a virtual thread.
3595 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3596 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3597 IfNode* iff_thread_not_equal_carrierThread =
3598 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3599
3600 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3601
3602 // False branch, is carrierThread.
3603 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3604 // Store release
3605 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3606
3607 set_all_memory(input_memory_state);
3608
3609 // True branch, is virtual thread.
3610 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3611 set_control(thread_not_equal_carrierThread);
3612
3613 // Load the raw epoch value from the vthread.
3614 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3615 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
3616 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3617
3618 // Mask off the excluded information from the epoch.
3619 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3620
3621 // Load the tid field from the thread.
3622 Node* tid = load_field_from_object(thread, "tid", "J");
3623
3624 // Store the vthread tid to the jfr thread local.
3625 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3626 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
3627
3628 // Branch is_excluded to conditionalize updating the epoch .
3629 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3630 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3631 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3632
3633 // True branch, vthread is excluded, no need to write epoch info.
3634 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3635 set_control(excluded);
3636 Node* vthread_is_excluded = _gvn.intcon(1);
3637
3638 // False branch, vthread is included, update epoch info.
3639 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3640 set_control(included);
3641 Node* vthread_is_included = _gvn.intcon(0);
3642
3643 // Get epoch value.
3644 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3645
3646 // Store the vthread epoch to the jfr thread local.
3647 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3648 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
3649
3650 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3651 record_for_igvn(excluded_rgn);
3652 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3653 record_for_igvn(excluded_mem);
3654 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3655 record_for_igvn(exclusion);
3656
3657 // Merge the excluded control and memory.
3658 excluded_rgn->init_req(_true_path, excluded);
3659 excluded_rgn->init_req(_false_path, included);
3660 excluded_mem->init_req(_true_path, tid_memory);
3661 excluded_mem->init_req(_false_path, included_memory);
3662 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3663 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3664
3665 // Set intermediate state.
3666 set_control(_gvn.transform(excluded_rgn));
3667 set_all_memory(excluded_mem);
3668
3669 // Store the vthread exclusion state to the jfr thread local.
3670 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3671 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
3672
3673 // Store release
3674 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
3675
3676 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3677 record_for_igvn(thread_compare_rgn);
3678 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3679 record_for_igvn(thread_compare_mem);
3680 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3681 record_for_igvn(vthread);
3682
3683 // Merge the thread_compare control and memory.
3684 thread_compare_rgn->init_req(_true_path, control());
3685 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3686 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3687 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3688
3689 // Set output state.
3690 set_control(_gvn.transform(thread_compare_rgn));
3691 set_all_memory(_gvn.transform(thread_compare_mem));
3692 }
3693
3694 #endif // JFR_HAVE_INTRINSICS
3695
3696 //------------------------inline_native_currentCarrierThread------------------
3697 bool LibraryCallKit::inline_native_currentCarrierThread() {
3698 Node* junk = nullptr;
3699 set_result(generate_current_thread(junk));
3700 return true;
3701 }
3702
3703 //------------------------inline_native_currentThread------------------
3704 bool LibraryCallKit::inline_native_currentThread() {
3705 Node* junk = nullptr;
3706 set_result(generate_virtual_thread(junk));
3707 return true;
3708 }
3709
3710 //------------------------inline_native_setVthread------------------
3711 bool LibraryCallKit::inline_native_setCurrentThread() {
3712 assert(C->method()->changes_current_thread(),
3713 "method changes current Thread but is not annotated ChangesCurrentThread");
3714 Node* arr = argument(1);
3715 Node* thread = _gvn.transform(new ThreadLocalNode());
3716 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3717 Node* thread_obj_handle
3718 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3719 thread_obj_handle = _gvn.transform(thread_obj_handle);
3720 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3721 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3722
3723 // Change the _monitor_owner_id of the JavaThread
3724 Node* tid = load_field_from_object(arr, "tid", "J");
3725 Node* monitor_owner_id_offset = basic_plus_adr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
3726 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
3727
3728 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3729 return true;
3730 }
3731
3732 const Type* LibraryCallKit::scopedValueCache_type() {
3733 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3734 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3735 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3736
3737 // Because we create the scopedValue cache lazily we have to make the
3738 // type of the result BotPTR.
3739 bool xk = etype->klass_is_exact();
3740 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3741 return objects_type;
3742 }
3743
3744 Node* LibraryCallKit::scopedValueCache_helper() {
3745 Node* thread = _gvn.transform(new ThreadLocalNode());
3746 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3747 // We cannot use immutable_memory() because we might flip onto a
3748 // different carrier thread, at which point we'll need to use that
3749 // carrier thread's cache.
3750 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3751 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3752 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3753 }
3754
3755 //------------------------inline_native_scopedValueCache------------------
3756 bool LibraryCallKit::inline_native_scopedValueCache() {
3757 Node* cache_obj_handle = scopedValueCache_helper();
3758 const Type* objects_type = scopedValueCache_type();
3759 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3760
3761 return true;
3762 }
3763
3764 //------------------------inline_native_setScopedValueCache------------------
3765 bool LibraryCallKit::inline_native_setScopedValueCache() {
3766 Node* arr = argument(0);
3767 Node* cache_obj_handle = scopedValueCache_helper();
3768 const Type* objects_type = scopedValueCache_type();
3769
3770 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3771 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3772
3773 return true;
3774 }
3775
3776 //------------------------inline_native_Continuation_pin and unpin-----------
3777
3778 // Shared implementation routine for both pin and unpin.
3779 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
3780 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3781
3782 // Save input memory.
3783 Node* input_memory_state = reset_memory();
3784 set_all_memory(input_memory_state);
3785
3786 // TLS
3787 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3788 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
3789 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
3790
3791 // Null check the last continuation object.
3792 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
3793 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
3794 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3795
3796 // False path, last continuation is null.
3797 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
3798
3799 // True path, last continuation is not null.
3800 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
3801
3802 set_control(continuation_is_not_null);
3803
3804 // Load the pin count from the last continuation.
3805 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
3806 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
3807
3808 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
3809 Node* pin_count_rhs;
3810 if (unpin) {
3811 pin_count_rhs = _gvn.intcon(0);
3812 } else {
3813 pin_count_rhs = _gvn.intcon(UINT32_MAX);
3814 }
3815 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
3816 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
3817 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
3818
3819 // True branch, pin count over/underflow.
3820 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
3821 {
3822 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
3823 // which will throw IllegalStateException for pin count over/underflow.
3824 // No memory changed so far - we can use memory create by reset_memory()
3825 // at the beginning of this intrinsic. No need to call reset_memory() again.
3826 PreserveJVMState pjvms(this);
3827 set_control(pin_count_over_underflow);
3828 uncommon_trap(Deoptimization::Reason_intrinsic,
3829 Deoptimization::Action_none);
3830 assert(stopped(), "invariant");
3831 }
3832
3833 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
3834 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
3835 set_control(valid_pin_count);
3836
3837 Node* next_pin_count;
3838 if (unpin) {
3839 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
3840 } else {
3841 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
3842 }
3843
3844 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
3845
3846 // Result of top level CFG and Memory.
3847 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3848 record_for_igvn(result_rgn);
3849 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3850 record_for_igvn(result_mem);
3851
3852 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
3853 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
3854 result_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3855 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3856
3857 // Set output state.
3858 set_control(_gvn.transform(result_rgn));
3859 set_all_memory(_gvn.transform(result_mem));
3860
3861 return true;
3862 }
3863
3864 //---------------------------load_mirror_from_klass----------------------------
3865 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3866 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3867 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3868 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3869 // mirror = ((OopHandle)mirror)->resolve();
3870 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3871 }
3872
3873 //-----------------------load_klass_from_mirror_common-------------------------
3874 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3875 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3876 // and branch to the given path on the region.
3877 // If never_see_null, take an uncommon trap on null, so we can optimistically
3878 // compile for the non-null case.
3879 // If the region is null, force never_see_null = true.
3880 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3881 bool never_see_null,
3882 RegionNode* region,
3883 int null_path,
3884 int offset) {
3885 if (region == nullptr) never_see_null = true;
3886 Node* p = basic_plus_adr(mirror, offset);
3887 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3888 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3889 Node* null_ctl = top();
3890 kls = null_check_oop(kls, &null_ctl, never_see_null);
3891 if (region != nullptr) {
3892 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3893 region->init_req(null_path, null_ctl);
3894 } else {
3895 assert(null_ctl == top(), "no loose ends");
3896 }
3897 return kls;
3898 }
3899
3900 //--------------------(inline_native_Class_query helpers)---------------------
3901 // Use this for JVM_ACC_INTERFACE.
3902 // Fall through if (mods & mask) == bits, take the guard otherwise.
3903 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
3904 ByteSize offset, const Type* type, BasicType bt) {
3905 // Branch around if the given klass has the given modifier bit set.
3906 // Like generate_guard, adds a new path onto the region.
3907 Node* modp = basic_plus_adr(kls, in_bytes(offset));
3908 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
3909 Node* mask = intcon(modifier_mask);
3910 Node* bits = intcon(modifier_bits);
3911 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3912 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3913 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3914 return generate_fair_guard(bol, region);
3915 }
3916 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3917 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
3918 Klass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
3919 }
3920
3921 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
3922 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3923 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
3924 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
3925 }
3926
3927 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3928 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
3929 }
3930
3931 //-------------------------inline_native_Class_query-------------------
3932 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3933 const Type* return_type = TypeInt::BOOL;
3934 Node* prim_return_value = top(); // what happens if it's a primitive class?
3935 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3936 bool expect_prim = false; // most of these guys expect to work on refs
3937
3938 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3939
3940 Node* mirror = argument(0);
3941 Node* obj = top();
3942
3943 switch (id) {
3944 case vmIntrinsics::_isInstance:
3945 // nothing is an instance of a primitive type
3946 prim_return_value = intcon(0);
3947 obj = argument(1);
3948 break;
3949 case vmIntrinsics::_isHidden:
3950 prim_return_value = intcon(0);
3951 break;
3952 case vmIntrinsics::_getSuperclass:
3953 prim_return_value = null();
3954 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3955 break;
3956 case vmIntrinsics::_getClassAccessFlags:
3957 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3958 return_type = TypeInt::CHAR;
3959 break;
3960 default:
3961 fatal_unexpected_iid(id);
3962 break;
3963 }
3964
3965 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3966 if (mirror_con == nullptr) return false; // cannot happen?
3967
3968 #ifndef PRODUCT
3969 if (C->print_intrinsics() || C->print_inlining()) {
3970 ciType* k = mirror_con->java_mirror_type();
3971 if (k) {
3972 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3973 k->print_name();
3974 tty->cr();
3975 }
3976 }
3977 #endif
3978
3979 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3980 RegionNode* region = new RegionNode(PATH_LIMIT);
3981 record_for_igvn(region);
3982 PhiNode* phi = new PhiNode(region, return_type);
3983
3984 // The mirror will never be null of Reflection.getClassAccessFlags, however
3985 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3986 // if it is. See bug 4774291.
3987
3988 // For Reflection.getClassAccessFlags(), the null check occurs in
3989 // the wrong place; see inline_unsafe_access(), above, for a similar
3990 // situation.
3991 mirror = null_check(mirror);
3992 // If mirror or obj is dead, only null-path is taken.
3993 if (stopped()) return true;
3994
3995 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3996
3997 // Now load the mirror's klass metaobject, and null-check it.
3998 // Side-effects region with the control path if the klass is null.
3999 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4000 // If kls is null, we have a primitive mirror.
4001 phi->init_req(_prim_path, prim_return_value);
4002 if (stopped()) { set_result(region, phi); return true; }
4003 bool safe_for_replace = (region->in(_prim_path) == top());
4004
4005 Node* p; // handy temp
4006 Node* null_ctl;
4007
4008 // Now that we have the non-null klass, we can perform the real query.
4009 // For constant classes, the query will constant-fold in LoadNode::Value.
4010 Node* query_value = top();
4011 switch (id) {
4012 case vmIntrinsics::_isInstance:
4013 // nothing is an instance of a primitive type
4014 query_value = gen_instanceof(obj, kls, safe_for_replace);
4015 break;
4016
4017 case vmIntrinsics::_isHidden:
4018 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4019 if (generate_hidden_class_guard(kls, region) != nullptr)
4020 // A guard was added. If the guard is taken, it was an hidden class.
4021 phi->add_req(intcon(1));
4022 // If we fall through, it's a plain class.
4023 query_value = intcon(0);
4024 break;
4025
4026
4027 case vmIntrinsics::_getSuperclass:
4028 // The rules here are somewhat unfortunate, but we can still do better
4029 // with random logic than with a JNI call.
4030 // Interfaces store null or Object as _super, but must report null.
4031 // Arrays store an intermediate super as _super, but must report Object.
4032 // Other types can report the actual _super.
4033 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4034 if (generate_interface_guard(kls, region) != nullptr)
4035 // A guard was added. If the guard is taken, it was an interface.
4036 phi->add_req(null());
4037 if (generate_array_guard(kls, region) != nullptr)
4038 // A guard was added. If the guard is taken, it was an array.
4039 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4040 // If we fall through, it's a plain class. Get its _super.
4041 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4042 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4043 null_ctl = top();
4044 kls = null_check_oop(kls, &null_ctl);
4045 if (null_ctl != top()) {
4046 // If the guard is taken, Object.superClass is null (both klass and mirror).
4047 region->add_req(null_ctl);
4048 phi ->add_req(null());
4049 }
4050 if (!stopped()) {
4051 query_value = load_mirror_from_klass(kls);
4052 }
4053 break;
4054
4055 case vmIntrinsics::_getClassAccessFlags:
4056 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
4057 query_value = make_load(nullptr, p, TypeInt::CHAR, T_CHAR, MemNode::unordered);
4058 break;
4059
4060 default:
4061 fatal_unexpected_iid(id);
4062 break;
4063 }
4064
4065 // Fall-through is the normal case of a query to a real class.
4066 phi->init_req(1, query_value);
4067 region->init_req(1, control());
4068
4069 C->set_has_split_ifs(true); // Has chance for split-if optimization
4070 set_result(region, phi);
4071 return true;
4072 }
4073
4074 //-------------------------inline_Class_cast-------------------
4075 bool LibraryCallKit::inline_Class_cast() {
4076 Node* mirror = argument(0); // Class
4077 Node* obj = argument(1);
4078 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4079 if (mirror_con == nullptr) {
4080 return false; // dead path (mirror->is_top()).
4081 }
4082 if (obj == nullptr || obj->is_top()) {
4083 return false; // dead path
4084 }
4085 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4086
4087 // First, see if Class.cast() can be folded statically.
4088 // java_mirror_type() returns non-null for compile-time Class constants.
4089 ciType* tm = mirror_con->java_mirror_type();
4090 if (tm != nullptr && tm->is_klass() &&
4091 tp != nullptr) {
4092 if (!tp->is_loaded()) {
4093 // Don't use intrinsic when class is not loaded.
4094 return false;
4095 } else {
4096 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
4097 if (static_res == Compile::SSC_always_true) {
4098 // isInstance() is true - fold the code.
4099 set_result(obj);
4100 return true;
4101 } else if (static_res == Compile::SSC_always_false) {
4102 // Don't use intrinsic, have to throw ClassCastException.
4103 // If the reference is null, the non-intrinsic bytecode will
4104 // be optimized appropriately.
4105 return false;
4106 }
4107 }
4108 }
4109
4110 // Bailout intrinsic and do normal inlining if exception path is frequent.
4111 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4112 return false;
4113 }
4114
4115 // Generate dynamic checks.
4116 // Class.cast() is java implementation of _checkcast bytecode.
4117 // Do checkcast (Parse::do_checkcast()) optimizations here.
4118
4119 mirror = null_check(mirror);
4120 // If mirror is dead, only null-path is taken.
4121 if (stopped()) {
4122 return true;
4123 }
4124
4125 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive).
4126 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
4127 RegionNode* region = new RegionNode(PATH_LIMIT);
4128 record_for_igvn(region);
4129
4130 // Now load the mirror's klass metaobject, and null-check it.
4131 // If kls is null, we have a primitive mirror and
4132 // nothing is an instance of a primitive type.
4133 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4134
4135 Node* res = top();
4136 if (!stopped()) {
4137 Node* bad_type_ctrl = top();
4138 // Do checkcast optimizations.
4139 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4140 region->init_req(_bad_type_path, bad_type_ctrl);
4141 }
4142 if (region->in(_prim_path) != top() ||
4143 region->in(_bad_type_path) != top()) {
4144 // Let Interpreter throw ClassCastException.
4145 PreserveJVMState pjvms(this);
4146 set_control(_gvn.transform(region));
4147 uncommon_trap(Deoptimization::Reason_intrinsic,
4148 Deoptimization::Action_maybe_recompile);
4149 }
4150 if (!stopped()) {
4151 set_result(res);
4152 }
4153 return true;
4154 }
4155
4156
4157 //--------------------------inline_native_subtype_check------------------------
4158 // This intrinsic takes the JNI calls out of the heart of
4159 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4160 bool LibraryCallKit::inline_native_subtype_check() {
4161 // Pull both arguments off the stack.
4162 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4163 args[0] = argument(0);
4164 args[1] = argument(1);
4165 Node* klasses[2]; // corresponding Klasses: superk, subk
4166 klasses[0] = klasses[1] = top();
4167
4168 enum {
4169 // A full decision tree on {superc is prim, subc is prim}:
4170 _prim_0_path = 1, // {P,N} => false
4171 // {P,P} & superc!=subc => false
4172 _prim_same_path, // {P,P} & superc==subc => true
4173 _prim_1_path, // {N,P} => false
4174 _ref_subtype_path, // {N,N} & subtype check wins => true
4175 _both_ref_path, // {N,N} & subtype check loses => false
4176 PATH_LIMIT
4177 };
4178
4179 RegionNode* region = new RegionNode(PATH_LIMIT);
4180 Node* phi = new PhiNode(region, TypeInt::BOOL);
4181 record_for_igvn(region);
4182
4183 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4184 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4185 int class_klass_offset = java_lang_Class::klass_offset();
4186
4187 // First null-check both mirrors and load each mirror's klass metaobject.
4188 int which_arg;
4189 for (which_arg = 0; which_arg <= 1; which_arg++) {
4190 Node* arg = args[which_arg];
4191 arg = null_check(arg);
4192 if (stopped()) break;
4193 args[which_arg] = arg;
4194
4195 Node* p = basic_plus_adr(arg, class_klass_offset);
4196 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4197 klasses[which_arg] = _gvn.transform(kls);
4198 }
4199
4200 // Having loaded both klasses, test each for null.
4201 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4202 for (which_arg = 0; which_arg <= 1; which_arg++) {
4203 Node* kls = klasses[which_arg];
4204 Node* null_ctl = top();
4205 kls = null_check_oop(kls, &null_ctl, never_see_null);
4206 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
4207 region->init_req(prim_path, null_ctl);
4208 if (stopped()) break;
4209 klasses[which_arg] = kls;
4210 }
4211
4212 if (!stopped()) {
4213 // now we have two reference types, in klasses[0..1]
4214 Node* subk = klasses[1]; // the argument to isAssignableFrom
4215 Node* superk = klasses[0]; // the receiver
4216 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4217 // now we have a successful reference subtype check
4218 region->set_req(_ref_subtype_path, control());
4219 }
4220
4221 // If both operands are primitive (both klasses null), then
4222 // we must return true when they are identical primitives.
4223 // It is convenient to test this after the first null klass check.
4224 set_control(region->in(_prim_0_path)); // go back to first null check
4225 if (!stopped()) {
4226 // Since superc is primitive, make a guard for the superc==subc case.
4227 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4228 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4229 generate_guard(bol_eq, region, PROB_FAIR);
4230 if (region->req() == PATH_LIMIT+1) {
4231 // A guard was added. If the added guard is taken, superc==subc.
4232 region->swap_edges(PATH_LIMIT, _prim_same_path);
4233 region->del_req(PATH_LIMIT);
4234 }
4235 region->set_req(_prim_0_path, control()); // Not equal after all.
4236 }
4237
4238 // these are the only paths that produce 'true':
4239 phi->set_req(_prim_same_path, intcon(1));
4240 phi->set_req(_ref_subtype_path, intcon(1));
4241
4242 // pull together the cases:
4243 assert(region->req() == PATH_LIMIT, "sane region");
4244 for (uint i = 1; i < region->req(); i++) {
4245 Node* ctl = region->in(i);
4246 if (ctl == nullptr || ctl == top()) {
4247 region->set_req(i, top());
4248 phi ->set_req(i, top());
4249 } else if (phi->in(i) == nullptr) {
4250 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4251 }
4252 }
4253
4254 set_control(_gvn.transform(region));
4255 set_result(_gvn.transform(phi));
4256 return true;
4257 }
4258
4259 //---------------------generate_array_guard_common------------------------
4260 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
4261 bool obj_array, bool not_array, Node** obj) {
4262
4263 if (stopped()) {
4264 return nullptr;
4265 }
4266
4267 // If obj_array/non_array==false/false:
4268 // Branch around if the given klass is in fact an array (either obj or prim).
4269 // If obj_array/non_array==false/true:
4270 // Branch around if the given klass is not an array klass of any kind.
4271 // If obj_array/non_array==true/true:
4272 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
4273 // If obj_array/non_array==true/false:
4274 // Branch around if the kls is an oop array (Object[] or subtype)
4275 //
4276 // Like generate_guard, adds a new path onto the region.
4277 jint layout_con = 0;
4278 Node* layout_val = get_layout_helper(kls, layout_con);
4279 if (layout_val == nullptr) {
4280 bool query = (obj_array
4281 ? Klass::layout_helper_is_objArray(layout_con)
4282 : Klass::layout_helper_is_array(layout_con));
4283 if (query == not_array) {
4284 return nullptr; // never a branch
4285 } else { // always a branch
4286 Node* always_branch = control();
4287 if (region != nullptr)
4288 region->add_req(always_branch);
4289 set_control(top());
4290 return always_branch;
4291 }
4292 }
4293 // Now test the correct condition.
4294 jint nval = (obj_array
4295 ? (jint)(Klass::_lh_array_tag_type_value
4296 << Klass::_lh_array_tag_shift)
4297 : Klass::_lh_neutral_value);
4298 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4299 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
4300 // invert the test if we are looking for a non-array
4301 if (not_array) btest = BoolTest(btest).negate();
4302 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4303 Node* ctrl = generate_fair_guard(bol, region);
4304 Node* is_array_ctrl = not_array ? control() : ctrl;
4305 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4306 // Keep track of the fact that 'obj' is an array to prevent
4307 // array specific accesses from floating above the guard.
4308 *obj = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4309 }
4310 return ctrl;
4311 }
4312
4313
4314 //-----------------------inline_native_newArray--------------------------
4315 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4316 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4317 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4318 Node* mirror;
4319 Node* count_val;
4320 if (uninitialized) {
4321 null_check_receiver();
4322 mirror = argument(1);
4323 count_val = argument(2);
4324 } else {
4325 mirror = argument(0);
4326 count_val = argument(1);
4327 }
4328
4329 mirror = null_check(mirror);
4330 // If mirror or obj is dead, only null-path is taken.
4331 if (stopped()) return true;
4332
4333 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4334 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4335 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4336 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4337 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4338
4339 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4340 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4341 result_reg, _slow_path);
4342 Node* normal_ctl = control();
4343 Node* no_array_ctl = result_reg->in(_slow_path);
4344
4345 // Generate code for the slow case. We make a call to newArray().
4346 set_control(no_array_ctl);
4347 if (!stopped()) {
4348 // Either the input type is void.class, or else the
4349 // array klass has not yet been cached. Either the
4350 // ensuing call will throw an exception, or else it
4351 // will cache the array klass for next time.
4352 PreserveJVMState pjvms(this);
4353 CallJavaNode* slow_call = nullptr;
4354 if (uninitialized) {
4355 // Generate optimized virtual call (holder class 'Unsafe' is final)
4356 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4357 } else {
4358 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4359 }
4360 Node* slow_result = set_results_for_java_call(slow_call);
4361 // this->control() comes from set_results_for_java_call
4362 result_reg->set_req(_slow_path, control());
4363 result_val->set_req(_slow_path, slow_result);
4364 result_io ->set_req(_slow_path, i_o());
4365 result_mem->set_req(_slow_path, reset_memory());
4366 }
4367
4368 set_control(normal_ctl);
4369 if (!stopped()) {
4370 // Normal case: The array type has been cached in the java.lang.Class.
4371 // The following call works fine even if the array type is polymorphic.
4372 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4373 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4374 result_reg->init_req(_normal_path, control());
4375 result_val->init_req(_normal_path, obj);
4376 result_io ->init_req(_normal_path, i_o());
4377 result_mem->init_req(_normal_path, reset_memory());
4378
4379 if (uninitialized) {
4380 // Mark the allocation so that zeroing is skipped
4381 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4382 alloc->maybe_set_complete(&_gvn);
4383 }
4384 }
4385
4386 // Return the combined state.
4387 set_i_o( _gvn.transform(result_io) );
4388 set_all_memory( _gvn.transform(result_mem));
4389
4390 C->set_has_split_ifs(true); // Has chance for split-if optimization
4391 set_result(result_reg, result_val);
4392 return true;
4393 }
4394
4395 //----------------------inline_native_getLength--------------------------
4396 // public static native int java.lang.reflect.Array.getLength(Object array);
4397 bool LibraryCallKit::inline_native_getLength() {
4398 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4399
4400 Node* array = null_check(argument(0));
4401 // If array is dead, only null-path is taken.
4402 if (stopped()) return true;
4403
4404 // Deoptimize if it is a non-array.
4405 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
4406
4407 if (non_array != nullptr) {
4408 PreserveJVMState pjvms(this);
4409 set_control(non_array);
4410 uncommon_trap(Deoptimization::Reason_intrinsic,
4411 Deoptimization::Action_maybe_recompile);
4412 }
4413
4414 // If control is dead, only non-array-path is taken.
4415 if (stopped()) return true;
4416
4417 // The works fine even if the array type is polymorphic.
4418 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4419 Node* result = load_array_length(array);
4420
4421 C->set_has_split_ifs(true); // Has chance for split-if optimization
4422 set_result(result);
4423 return true;
4424 }
4425
4426 //------------------------inline_array_copyOf----------------------------
4427 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4428 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4429 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4430 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4431
4432 // Get the arguments.
4433 Node* original = argument(0);
4434 Node* start = is_copyOfRange? argument(1): intcon(0);
4435 Node* end = is_copyOfRange? argument(2): argument(1);
4436 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4437
4438 Node* newcopy = nullptr;
4439
4440 // Set the original stack and the reexecute bit for the interpreter to reexecute
4441 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4442 { PreserveReexecuteState preexecs(this);
4443 jvms()->set_should_reexecute(true);
4444
4445 array_type_mirror = null_check(array_type_mirror);
4446 original = null_check(original);
4447
4448 // Check if a null path was taken unconditionally.
4449 if (stopped()) return true;
4450
4451 Node* orig_length = load_array_length(original);
4452
4453 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4454 klass_node = null_check(klass_node);
4455
4456 RegionNode* bailout = new RegionNode(1);
4457 record_for_igvn(bailout);
4458
4459 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4460 // Bail out if that is so.
4461 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4462 if (not_objArray != nullptr) {
4463 // Improve the klass node's type from the new optimistic assumption:
4464 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4465 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4466 Node* cast = new CastPPNode(control(), klass_node, akls);
4467 klass_node = _gvn.transform(cast);
4468 }
4469
4470 // Bail out if either start or end is negative.
4471 generate_negative_guard(start, bailout, &start);
4472 generate_negative_guard(end, bailout, &end);
4473
4474 Node* length = end;
4475 if (_gvn.type(start) != TypeInt::ZERO) {
4476 length = _gvn.transform(new SubINode(end, start));
4477 }
4478
4479 // Bail out if length is negative (i.e., if start > end).
4480 // Without this the new_array would throw
4481 // NegativeArraySizeException but IllegalArgumentException is what
4482 // should be thrown
4483 generate_negative_guard(length, bailout, &length);
4484
4485 // Bail out if start is larger than the original length
4486 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4487 generate_negative_guard(orig_tail, bailout, &orig_tail);
4488
4489 if (bailout->req() > 1) {
4490 PreserveJVMState pjvms(this);
4491 set_control(_gvn.transform(bailout));
4492 uncommon_trap(Deoptimization::Reason_intrinsic,
4493 Deoptimization::Action_maybe_recompile);
4494 }
4495
4496 if (!stopped()) {
4497 // How many elements will we copy from the original?
4498 // The answer is MinI(orig_tail, length).
4499 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
4500
4501 // Generate a direct call to the right arraycopy function(s).
4502 // We know the copy is disjoint but we might not know if the
4503 // oop stores need checking.
4504 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4505 // This will fail a store-check if x contains any non-nulls.
4506
4507 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4508 // loads/stores but it is legal only if we're sure the
4509 // Arrays.copyOf would succeed. So we need all input arguments
4510 // to the copyOf to be validated, including that the copy to the
4511 // new array won't trigger an ArrayStoreException. That subtype
4512 // check can be optimized if we know something on the type of
4513 // the input array from type speculation.
4514 if (_gvn.type(klass_node)->singleton()) {
4515 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4516 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4517
4518 int test = C->static_subtype_check(superk, subk);
4519 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4520 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4521 if (t_original->speculative_type() != nullptr) {
4522 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4523 }
4524 }
4525 }
4526
4527 bool validated = false;
4528 // Reason_class_check rather than Reason_intrinsic because we
4529 // want to intrinsify even if this traps.
4530 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4531 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4532
4533 if (not_subtype_ctrl != top()) {
4534 PreserveJVMState pjvms(this);
4535 set_control(not_subtype_ctrl);
4536 uncommon_trap(Deoptimization::Reason_class_check,
4537 Deoptimization::Action_make_not_entrant);
4538 assert(stopped(), "Should be stopped");
4539 }
4540 validated = true;
4541 }
4542
4543 if (!stopped()) {
4544 newcopy = new_array(klass_node, length, 0); // no arguments to push
4545
4546 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4547 load_object_klass(original), klass_node);
4548 if (!is_copyOfRange) {
4549 ac->set_copyof(validated);
4550 } else {
4551 ac->set_copyofrange(validated);
4552 }
4553 Node* n = _gvn.transform(ac);
4554 if (n == ac) {
4555 ac->connect_outputs(this);
4556 } else {
4557 assert(validated, "shouldn't transform if all arguments not validated");
4558 set_all_memory(n);
4559 }
4560 }
4561 }
4562 } // original reexecute is set back here
4563
4564 C->set_has_split_ifs(true); // Has chance for split-if optimization
4565 if (!stopped()) {
4566 set_result(newcopy);
4567 }
4568 return true;
4569 }
4570
4571
4572 //----------------------generate_virtual_guard---------------------------
4573 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4574 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4575 RegionNode* slow_region) {
4576 ciMethod* method = callee();
4577 int vtable_index = method->vtable_index();
4578 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4579 "bad index %d", vtable_index);
4580 // Get the Method* out of the appropriate vtable entry.
4581 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4582 vtable_index*vtableEntry::size_in_bytes() +
4583 in_bytes(vtableEntry::method_offset());
4584 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4585 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4586
4587 // Compare the target method with the expected method (e.g., Object.hashCode).
4588 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4589
4590 Node* native_call = makecon(native_call_addr);
4591 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4592 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4593
4594 return generate_slow_guard(test_native, slow_region);
4595 }
4596
4597 //-----------------------generate_method_call----------------------------
4598 // Use generate_method_call to make a slow-call to the real
4599 // method if the fast path fails. An alternative would be to
4600 // use a stub like OptoRuntime::slow_arraycopy_Java.
4601 // This only works for expanding the current library call,
4602 // not another intrinsic. (E.g., don't use this for making an
4603 // arraycopy call inside of the copyOf intrinsic.)
4604 CallJavaNode*
4605 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4606 // When compiling the intrinsic method itself, do not use this technique.
4607 guarantee(callee() != C->method(), "cannot make slow-call to self");
4608
4609 ciMethod* method = callee();
4610 // ensure the JVMS we have will be correct for this call
4611 guarantee(method_id == method->intrinsic_id(), "must match");
4612
4613 const TypeFunc* tf = TypeFunc::make(method);
4614 if (res_not_null) {
4615 assert(tf->return_type() == T_OBJECT, "");
4616 const TypeTuple* range = tf->range();
4617 const Type** fields = TypeTuple::fields(range->cnt());
4618 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4619 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4620 tf = TypeFunc::make(tf->domain(), new_range);
4621 }
4622 CallJavaNode* slow_call;
4623 if (is_static) {
4624 assert(!is_virtual, "");
4625 slow_call = new CallStaticJavaNode(C, tf,
4626 SharedRuntime::get_resolve_static_call_stub(), method);
4627 } else if (is_virtual) {
4628 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4629 int vtable_index = Method::invalid_vtable_index;
4630 if (UseInlineCaches) {
4631 // Suppress the vtable call
4632 } else {
4633 // hashCode and clone are not a miranda methods,
4634 // so the vtable index is fixed.
4635 // No need to use the linkResolver to get it.
4636 vtable_index = method->vtable_index();
4637 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4638 "bad index %d", vtable_index);
4639 }
4640 slow_call = new CallDynamicJavaNode(tf,
4641 SharedRuntime::get_resolve_virtual_call_stub(),
4642 method, vtable_index);
4643 } else { // neither virtual nor static: opt_virtual
4644 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4645 slow_call = new CallStaticJavaNode(C, tf,
4646 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4647 slow_call->set_optimized_virtual(true);
4648 }
4649 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4650 // To be able to issue a direct call (optimized virtual or virtual)
4651 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4652 // about the method being invoked should be attached to the call site to
4653 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4654 slow_call->set_override_symbolic_info(true);
4655 }
4656 set_arguments_for_java_call(slow_call);
4657 set_edges_for_java_call(slow_call);
4658 return slow_call;
4659 }
4660
4661
4662 /**
4663 * Build special case code for calls to hashCode on an object. This call may
4664 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4665 * slightly different code.
4666 */
4667 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4668 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4669 assert(!(is_virtual && is_static), "either virtual, special, or static");
4670
4671 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4672
4673 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4674 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4675 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4676 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4677 Node* obj = nullptr;
4678 if (!is_static) {
4679 // Check for hashing null object
4680 obj = null_check_receiver();
4681 if (stopped()) return true; // unconditionally null
4682 result_reg->init_req(_null_path, top());
4683 result_val->init_req(_null_path, top());
4684 } else {
4685 // Do a null check, and return zero if null.
4686 // System.identityHashCode(null) == 0
4687 obj = argument(0);
4688 Node* null_ctl = top();
4689 obj = null_check_oop(obj, &null_ctl);
4690 result_reg->init_req(_null_path, null_ctl);
4691 result_val->init_req(_null_path, _gvn.intcon(0));
4692 }
4693
4694 // Unconditionally null? Then return right away.
4695 if (stopped()) {
4696 set_control( result_reg->in(_null_path));
4697 if (!stopped())
4698 set_result(result_val->in(_null_path));
4699 return true;
4700 }
4701
4702 // We only go to the fast case code if we pass a number of guards. The
4703 // paths which do not pass are accumulated in the slow_region.
4704 RegionNode* slow_region = new RegionNode(1);
4705 record_for_igvn(slow_region);
4706
4707 // If this is a virtual call, we generate a funny guard. We pull out
4708 // the vtable entry corresponding to hashCode() from the target object.
4709 // If the target method which we are calling happens to be the native
4710 // Object hashCode() method, we pass the guard. We do not need this
4711 // guard for non-virtual calls -- the caller is known to be the native
4712 // Object hashCode().
4713 if (is_virtual) {
4714 // After null check, get the object's klass.
4715 Node* obj_klass = load_object_klass(obj);
4716 generate_virtual_guard(obj_klass, slow_region);
4717 }
4718
4719 // Get the header out of the object, use LoadMarkNode when available
4720 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4721 // The control of the load must be null. Otherwise, the load can move before
4722 // the null check after castPP removal.
4723 Node* no_ctrl = nullptr;
4724 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4725
4726 if (!UseObjectMonitorTable) {
4727 // Test the header to see if it is safe to read w.r.t. locking.
4728 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4729 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4730 if (LockingMode == LM_LIGHTWEIGHT) {
4731 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
4732 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
4733 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
4734
4735 generate_slow_guard(test_monitor, slow_region);
4736 } else {
4737 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4738 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val));
4739 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne));
4740
4741 generate_slow_guard(test_not_unlocked, slow_region);
4742 }
4743 }
4744
4745 // Get the hash value and check to see that it has been properly assigned.
4746 // We depend on hash_mask being at most 32 bits and avoid the use of
4747 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4748 // vm: see markWord.hpp.
4749 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4750 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4751 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4752 // This hack lets the hash bits live anywhere in the mark object now, as long
4753 // as the shift drops the relevant bits into the low 32 bits. Note that
4754 // Java spec says that HashCode is an int so there's no point in capturing
4755 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4756 hshifted_header = ConvX2I(hshifted_header);
4757 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4758
4759 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4760 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4761 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4762
4763 generate_slow_guard(test_assigned, slow_region);
4764
4765 Node* init_mem = reset_memory();
4766 // fill in the rest of the null path:
4767 result_io ->init_req(_null_path, i_o());
4768 result_mem->init_req(_null_path, init_mem);
4769
4770 result_val->init_req(_fast_path, hash_val);
4771 result_reg->init_req(_fast_path, control());
4772 result_io ->init_req(_fast_path, i_o());
4773 result_mem->init_req(_fast_path, init_mem);
4774
4775 // Generate code for the slow case. We make a call to hashCode().
4776 set_control(_gvn.transform(slow_region));
4777 if (!stopped()) {
4778 // No need for PreserveJVMState, because we're using up the present state.
4779 set_all_memory(init_mem);
4780 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4781 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4782 Node* slow_result = set_results_for_java_call(slow_call);
4783 // this->control() comes from set_results_for_java_call
4784 result_reg->init_req(_slow_path, control());
4785 result_val->init_req(_slow_path, slow_result);
4786 result_io ->set_req(_slow_path, i_o());
4787 result_mem ->set_req(_slow_path, reset_memory());
4788 }
4789
4790 // Return the combined state.
4791 set_i_o( _gvn.transform(result_io) );
4792 set_all_memory( _gvn.transform(result_mem));
4793
4794 set_result(result_reg, result_val);
4795 return true;
4796 }
4797
4798 //---------------------------inline_native_getClass----------------------------
4799 // public final native Class<?> java.lang.Object.getClass();
4800 //
4801 // Build special case code for calls to getClass on an object.
4802 bool LibraryCallKit::inline_native_getClass() {
4803 Node* obj = null_check_receiver();
4804 if (stopped()) return true;
4805 set_result(load_mirror_from_klass(load_object_klass(obj)));
4806 return true;
4807 }
4808
4809 //-----------------inline_native_Reflection_getCallerClass---------------------
4810 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4811 //
4812 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4813 //
4814 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4815 // in that it must skip particular security frames and checks for
4816 // caller sensitive methods.
4817 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4818 #ifndef PRODUCT
4819 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4820 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4821 }
4822 #endif
4823
4824 if (!jvms()->has_method()) {
4825 #ifndef PRODUCT
4826 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4827 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4828 }
4829 #endif
4830 return false;
4831 }
4832
4833 // Walk back up the JVM state to find the caller at the required
4834 // depth.
4835 JVMState* caller_jvms = jvms();
4836
4837 // Cf. JVM_GetCallerClass
4838 // NOTE: Start the loop at depth 1 because the current JVM state does
4839 // not include the Reflection.getCallerClass() frame.
4840 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4841 ciMethod* m = caller_jvms->method();
4842 switch (n) {
4843 case 0:
4844 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4845 break;
4846 case 1:
4847 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4848 if (!m->caller_sensitive()) {
4849 #ifndef PRODUCT
4850 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4851 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4852 }
4853 #endif
4854 return false; // bail-out; let JVM_GetCallerClass do the work
4855 }
4856 break;
4857 default:
4858 if (!m->is_ignored_by_security_stack_walk()) {
4859 // We have reached the desired frame; return the holder class.
4860 // Acquire method holder as java.lang.Class and push as constant.
4861 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4862 ciInstance* caller_mirror = caller_klass->java_mirror();
4863 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4864
4865 #ifndef PRODUCT
4866 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4867 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());
4868 tty->print_cr(" JVM state at this point:");
4869 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4870 ciMethod* m = jvms()->of_depth(i)->method();
4871 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4872 }
4873 }
4874 #endif
4875 return true;
4876 }
4877 break;
4878 }
4879 }
4880
4881 #ifndef PRODUCT
4882 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4883 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4884 tty->print_cr(" JVM state at this point:");
4885 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4886 ciMethod* m = jvms()->of_depth(i)->method();
4887 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4888 }
4889 }
4890 #endif
4891
4892 return false; // bail-out; let JVM_GetCallerClass do the work
4893 }
4894
4895 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4896 Node* arg = argument(0);
4897 Node* result = nullptr;
4898
4899 switch (id) {
4900 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4901 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4902 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4903 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4904 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
4905 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
4906
4907 case vmIntrinsics::_doubleToLongBits: {
4908 // two paths (plus control) merge in a wood
4909 RegionNode *r = new RegionNode(3);
4910 Node *phi = new PhiNode(r, TypeLong::LONG);
4911
4912 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4913 // Build the boolean node
4914 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4915
4916 // Branch either way.
4917 // NaN case is less traveled, which makes all the difference.
4918 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4919 Node *opt_isnan = _gvn.transform(ifisnan);
4920 assert( opt_isnan->is_If(), "Expect an IfNode");
4921 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4922 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4923
4924 set_control(iftrue);
4925
4926 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4927 Node *slow_result = longcon(nan_bits); // return NaN
4928 phi->init_req(1, _gvn.transform( slow_result ));
4929 r->init_req(1, iftrue);
4930
4931 // Else fall through
4932 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4933 set_control(iffalse);
4934
4935 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4936 r->init_req(2, iffalse);
4937
4938 // Post merge
4939 set_control(_gvn.transform(r));
4940 record_for_igvn(r);
4941
4942 C->set_has_split_ifs(true); // Has chance for split-if optimization
4943 result = phi;
4944 assert(result->bottom_type()->isa_long(), "must be");
4945 break;
4946 }
4947
4948 case vmIntrinsics::_floatToIntBits: {
4949 // two paths (plus control) merge in a wood
4950 RegionNode *r = new RegionNode(3);
4951 Node *phi = new PhiNode(r, TypeInt::INT);
4952
4953 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4954 // Build the boolean node
4955 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4956
4957 // Branch either way.
4958 // NaN case is less traveled, which makes all the difference.
4959 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4960 Node *opt_isnan = _gvn.transform(ifisnan);
4961 assert( opt_isnan->is_If(), "Expect an IfNode");
4962 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4963 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4964
4965 set_control(iftrue);
4966
4967 static const jint nan_bits = 0x7fc00000;
4968 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4969 phi->init_req(1, _gvn.transform( slow_result ));
4970 r->init_req(1, iftrue);
4971
4972 // Else fall through
4973 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4974 set_control(iffalse);
4975
4976 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4977 r->init_req(2, iffalse);
4978
4979 // Post merge
4980 set_control(_gvn.transform(r));
4981 record_for_igvn(r);
4982
4983 C->set_has_split_ifs(true); // Has chance for split-if optimization
4984 result = phi;
4985 assert(result->bottom_type()->isa_int(), "must be");
4986 break;
4987 }
4988
4989 default:
4990 fatal_unexpected_iid(id);
4991 break;
4992 }
4993 set_result(_gvn.transform(result));
4994 return true;
4995 }
4996
4997 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
4998 Node* arg = argument(0);
4999 Node* result = nullptr;
5000
5001 switch (id) {
5002 case vmIntrinsics::_floatIsInfinite:
5003 result = new IsInfiniteFNode(arg);
5004 break;
5005 case vmIntrinsics::_floatIsFinite:
5006 result = new IsFiniteFNode(arg);
5007 break;
5008 case vmIntrinsics::_doubleIsInfinite:
5009 result = new IsInfiniteDNode(arg);
5010 break;
5011 case vmIntrinsics::_doubleIsFinite:
5012 result = new IsFiniteDNode(arg);
5013 break;
5014 default:
5015 fatal_unexpected_iid(id);
5016 break;
5017 }
5018 set_result(_gvn.transform(result));
5019 return true;
5020 }
5021
5022 //----------------------inline_unsafe_copyMemory-------------------------
5023 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5024
5025 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5026 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5027 const Type* base_t = gvn.type(base);
5028
5029 bool in_native = (base_t == TypePtr::NULL_PTR);
5030 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5031 bool is_mixed = !in_heap && !in_native;
5032
5033 if (is_mixed) {
5034 return true; // mixed accesses can touch both on-heap and off-heap memory
5035 }
5036 if (in_heap) {
5037 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5038 if (!is_prim_array) {
5039 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5040 // there's not enough type information available to determine proper memory slice for it.
5041 return true;
5042 }
5043 }
5044 return false;
5045 }
5046
5047 bool LibraryCallKit::inline_unsafe_copyMemory() {
5048 if (callee()->is_static()) return false; // caller must have the capability!
5049 null_check_receiver(); // null-check receiver
5050 if (stopped()) return true;
5051
5052 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5053
5054 Node* src_base = argument(1); // type: oop
5055 Node* src_off = ConvL2X(argument(2)); // type: long
5056 Node* dst_base = argument(4); // type: oop
5057 Node* dst_off = ConvL2X(argument(5)); // type: long
5058 Node* size = ConvL2X(argument(7)); // type: long
5059
5060 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5061 "fieldOffset must be byte-scaled");
5062
5063 Node* src_addr = make_unsafe_address(src_base, src_off);
5064 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5065
5066 Node* thread = _gvn.transform(new ThreadLocalNode());
5067 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5068 BasicType doing_unsafe_access_bt = T_BYTE;
5069 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5070
5071 // update volatile field
5072 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5073
5074 int flags = RC_LEAF | RC_NO_FP;
5075
5076 const TypePtr* dst_type = TypePtr::BOTTOM;
5077
5078 // Adjust memory effects of the runtime call based on input values.
5079 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5080 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5081 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5082
5083 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5084 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5085 flags |= RC_NARROW_MEM; // narrow in memory
5086 }
5087 }
5088
5089 // Call it. Note that the length argument is not scaled.
5090 make_runtime_call(flags,
5091 OptoRuntime::fast_arraycopy_Type(),
5092 StubRoutines::unsafe_arraycopy(),
5093 "unsafe_arraycopy",
5094 dst_type,
5095 src_addr, dst_addr, size XTOP);
5096
5097 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5098
5099 return true;
5100 }
5101
5102 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5103 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5104 bool LibraryCallKit::inline_unsafe_setMemory() {
5105 if (callee()->is_static()) return false; // caller must have the capability!
5106 null_check_receiver(); // null-check receiver
5107 if (stopped()) return true;
5108
5109 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5110
5111 Node* dst_base = argument(1); // type: oop
5112 Node* dst_off = ConvL2X(argument(2)); // type: long
5113 Node* size = ConvL2X(argument(4)); // type: long
5114 Node* byte = argument(6); // type: byte
5115
5116 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5117 "fieldOffset must be byte-scaled");
5118
5119 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5120
5121 Node* thread = _gvn.transform(new ThreadLocalNode());
5122 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5123 BasicType doing_unsafe_access_bt = T_BYTE;
5124 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5125
5126 // update volatile field
5127 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5128
5129 int flags = RC_LEAF | RC_NO_FP;
5130
5131 const TypePtr* dst_type = TypePtr::BOTTOM;
5132
5133 // Adjust memory effects of the runtime call based on input values.
5134 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5135 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5136
5137 flags |= RC_NARROW_MEM; // narrow in memory
5138 }
5139
5140 // Call it. Note that the length argument is not scaled.
5141 make_runtime_call(flags,
5142 OptoRuntime::unsafe_setmemory_Type(),
5143 StubRoutines::unsafe_setmemory(),
5144 "unsafe_setmemory",
5145 dst_type,
5146 dst_addr, size XTOP, byte);
5147
5148 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5149
5150 return true;
5151 }
5152
5153 #undef XTOP
5154
5155 //------------------------clone_coping-----------------------------------
5156 // Helper function for inline_native_clone.
5157 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5158 assert(obj_size != nullptr, "");
5159 Node* raw_obj = alloc_obj->in(1);
5160 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5161
5162 AllocateNode* alloc = nullptr;
5163 if (ReduceBulkZeroing &&
5164 // If we are implementing an array clone without knowing its source type
5165 // (can happen when compiling the array-guarded branch of a reflective
5166 // Object.clone() invocation), initialize the array within the allocation.
5167 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5168 // to a runtime clone call that assumes fully initialized source arrays.
5169 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5170 // We will be completely responsible for initializing this object -
5171 // mark Initialize node as complete.
5172 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5173 // The object was just allocated - there should be no any stores!
5174 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5175 // Mark as complete_with_arraycopy so that on AllocateNode
5176 // expansion, we know this AllocateNode is initialized by an array
5177 // copy and a StoreStore barrier exists after the array copy.
5178 alloc->initialization()->set_complete_with_arraycopy();
5179 }
5180
5181 Node* size = _gvn.transform(obj_size);
5182 access_clone(obj, alloc_obj, size, is_array);
5183
5184 // Do not let reads from the cloned object float above the arraycopy.
5185 if (alloc != nullptr) {
5186 // Do not let stores that initialize this object be reordered with
5187 // a subsequent store that would make this object accessible by
5188 // other threads.
5189 // Record what AllocateNode this StoreStore protects so that
5190 // escape analysis can go from the MemBarStoreStoreNode to the
5191 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5192 // based on the escape status of the AllocateNode.
5193 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5194 } else {
5195 insert_mem_bar(Op_MemBarCPUOrder);
5196 }
5197 }
5198
5199 //------------------------inline_native_clone----------------------------
5200 // protected native Object java.lang.Object.clone();
5201 //
5202 // Here are the simple edge cases:
5203 // null receiver => normal trap
5204 // virtual and clone was overridden => slow path to out-of-line clone
5205 // not cloneable or finalizer => slow path to out-of-line Object.clone
5206 //
5207 // The general case has two steps, allocation and copying.
5208 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5209 //
5210 // Copying also has two cases, oop arrays and everything else.
5211 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5212 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5213 //
5214 // These steps fold up nicely if and when the cloned object's klass
5215 // can be sharply typed as an object array, a type array, or an instance.
5216 //
5217 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5218 PhiNode* result_val;
5219
5220 // Set the reexecute bit for the interpreter to reexecute
5221 // the bytecode that invokes Object.clone if deoptimization happens.
5222 { PreserveReexecuteState preexecs(this);
5223 jvms()->set_should_reexecute(true);
5224
5225 Node* obj = null_check_receiver();
5226 if (stopped()) return true;
5227
5228 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5229
5230 // If we are going to clone an instance, we need its exact type to
5231 // know the number and types of fields to convert the clone to
5232 // loads/stores. Maybe a speculative type can help us.
5233 if (!obj_type->klass_is_exact() &&
5234 obj_type->speculative_type() != nullptr &&
5235 obj_type->speculative_type()->is_instance_klass()) {
5236 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5237 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5238 !spec_ik->has_injected_fields()) {
5239 if (!obj_type->isa_instptr() ||
5240 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5241 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5242 }
5243 }
5244 }
5245
5246 // Conservatively insert a memory barrier on all memory slices.
5247 // Do not let writes into the original float below the clone.
5248 insert_mem_bar(Op_MemBarCPUOrder);
5249
5250 // paths into result_reg:
5251 enum {
5252 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5253 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5254 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5255 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5256 PATH_LIMIT
5257 };
5258 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5259 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5260 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5261 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5262 record_for_igvn(result_reg);
5263
5264 Node* obj_klass = load_object_klass(obj);
5265 Node* array_obj = obj;
5266 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5267 if (array_ctl != nullptr) {
5268 // It's an array.
5269 PreserveJVMState pjvms(this);
5270 set_control(array_ctl);
5271 Node* obj_length = load_array_length(array_obj);
5272 Node* array_size = nullptr; // Size of the array without object alignment padding.
5273 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5274
5275 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5276 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5277 // If it is an oop array, it requires very special treatment,
5278 // because gc barriers are required when accessing the array.
5279 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5280 if (is_obja != nullptr) {
5281 PreserveJVMState pjvms2(this);
5282 set_control(is_obja);
5283 // Generate a direct call to the right arraycopy function(s).
5284 // Clones are always tightly coupled.
5285 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5286 ac->set_clone_oop_array();
5287 Node* n = _gvn.transform(ac);
5288 assert(n == ac, "cannot disappear");
5289 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5290
5291 result_reg->init_req(_objArray_path, control());
5292 result_val->init_req(_objArray_path, alloc_obj);
5293 result_i_o ->set_req(_objArray_path, i_o());
5294 result_mem ->set_req(_objArray_path, reset_memory());
5295 }
5296 }
5297 // Otherwise, there are no barriers to worry about.
5298 // (We can dispense with card marks if we know the allocation
5299 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5300 // causes the non-eden paths to take compensating steps to
5301 // simulate a fresh allocation, so that no further
5302 // card marks are required in compiled code to initialize
5303 // the object.)
5304
5305 if (!stopped()) {
5306 copy_to_clone(array_obj, alloc_obj, array_size, true);
5307
5308 // Present the results of the copy.
5309 result_reg->init_req(_array_path, control());
5310 result_val->init_req(_array_path, alloc_obj);
5311 result_i_o ->set_req(_array_path, i_o());
5312 result_mem ->set_req(_array_path, reset_memory());
5313 }
5314 }
5315
5316 // We only go to the instance fast case code if we pass a number of guards.
5317 // The paths which do not pass are accumulated in the slow_region.
5318 RegionNode* slow_region = new RegionNode(1);
5319 record_for_igvn(slow_region);
5320 if (!stopped()) {
5321 // It's an instance (we did array above). Make the slow-path tests.
5322 // If this is a virtual call, we generate a funny guard. We grab
5323 // the vtable entry corresponding to clone() from the target object.
5324 // If the target method which we are calling happens to be the
5325 // Object clone() method, we pass the guard. We do not need this
5326 // guard for non-virtual calls; the caller is known to be the native
5327 // Object clone().
5328 if (is_virtual) {
5329 generate_virtual_guard(obj_klass, slow_region);
5330 }
5331
5332 // The object must be easily cloneable and must not have a finalizer.
5333 // Both of these conditions may be checked in a single test.
5334 // We could optimize the test further, but we don't care.
5335 generate_misc_flags_guard(obj_klass,
5336 // Test both conditions:
5337 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5338 // Must be cloneable but not finalizer:
5339 KlassFlags::_misc_is_cloneable_fast,
5340 slow_region);
5341 }
5342
5343 if (!stopped()) {
5344 // It's an instance, and it passed the slow-path tests.
5345 PreserveJVMState pjvms(this);
5346 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5347 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5348 // is reexecuted if deoptimization occurs and there could be problems when merging
5349 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5350 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5351
5352 copy_to_clone(obj, alloc_obj, obj_size, false);
5353
5354 // Present the results of the slow call.
5355 result_reg->init_req(_instance_path, control());
5356 result_val->init_req(_instance_path, alloc_obj);
5357 result_i_o ->set_req(_instance_path, i_o());
5358 result_mem ->set_req(_instance_path, reset_memory());
5359 }
5360
5361 // Generate code for the slow case. We make a call to clone().
5362 set_control(_gvn.transform(slow_region));
5363 if (!stopped()) {
5364 PreserveJVMState pjvms(this);
5365 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5366 // We need to deoptimize on exception (see comment above)
5367 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5368 // this->control() comes from set_results_for_java_call
5369 result_reg->init_req(_slow_path, control());
5370 result_val->init_req(_slow_path, slow_result);
5371 result_i_o ->set_req(_slow_path, i_o());
5372 result_mem ->set_req(_slow_path, reset_memory());
5373 }
5374
5375 // Return the combined state.
5376 set_control( _gvn.transform(result_reg));
5377 set_i_o( _gvn.transform(result_i_o));
5378 set_all_memory( _gvn.transform(result_mem));
5379 } // original reexecute is set back here
5380
5381 set_result(_gvn.transform(result_val));
5382 return true;
5383 }
5384
5385 // If we have a tightly coupled allocation, the arraycopy may take care
5386 // of the array initialization. If one of the guards we insert between
5387 // the allocation and the arraycopy causes a deoptimization, an
5388 // uninitialized array will escape the compiled method. To prevent that
5389 // we set the JVM state for uncommon traps between the allocation and
5390 // the arraycopy to the state before the allocation so, in case of
5391 // deoptimization, we'll reexecute the allocation and the
5392 // initialization.
5393 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5394 if (alloc != nullptr) {
5395 ciMethod* trap_method = alloc->jvms()->method();
5396 int trap_bci = alloc->jvms()->bci();
5397
5398 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5399 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5400 // Make sure there's no store between the allocation and the
5401 // arraycopy otherwise visible side effects could be rexecuted
5402 // in case of deoptimization and cause incorrect execution.
5403 bool no_interfering_store = true;
5404 Node* mem = alloc->in(TypeFunc::Memory);
5405 if (mem->is_MergeMem()) {
5406 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5407 Node* n = mms.memory();
5408 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5409 assert(n->is_Store(), "what else?");
5410 no_interfering_store = false;
5411 break;
5412 }
5413 }
5414 } else {
5415 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5416 Node* n = mms.memory();
5417 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5418 assert(n->is_Store(), "what else?");
5419 no_interfering_store = false;
5420 break;
5421 }
5422 }
5423 }
5424
5425 if (no_interfering_store) {
5426 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5427
5428 JVMState* saved_jvms = jvms();
5429 saved_reexecute_sp = _reexecute_sp;
5430
5431 set_jvms(sfpt->jvms());
5432 _reexecute_sp = jvms()->sp();
5433
5434 return saved_jvms;
5435 }
5436 }
5437 }
5438 return nullptr;
5439 }
5440
5441 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5442 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5443 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5444 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5445 uint size = alloc->req();
5446 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5447 old_jvms->set_map(sfpt);
5448 for (uint i = 0; i < size; i++) {
5449 sfpt->init_req(i, alloc->in(i));
5450 }
5451 // re-push array length for deoptimization
5452 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5453 old_jvms->set_sp(old_jvms->sp()+1);
5454 old_jvms->set_monoff(old_jvms->monoff()+1);
5455 old_jvms->set_scloff(old_jvms->scloff()+1);
5456 old_jvms->set_endoff(old_jvms->endoff()+1);
5457 old_jvms->set_should_reexecute(true);
5458
5459 sfpt->set_i_o(map()->i_o());
5460 sfpt->set_memory(map()->memory());
5461 sfpt->set_control(map()->control());
5462 return sfpt;
5463 }
5464
5465 // In case of a deoptimization, we restart execution at the
5466 // allocation, allocating a new array. We would leave an uninitialized
5467 // array in the heap that GCs wouldn't expect. Move the allocation
5468 // after the traps so we don't allocate the array if we
5469 // deoptimize. This is possible because tightly_coupled_allocation()
5470 // guarantees there's no observer of the allocated array at this point
5471 // and the control flow is simple enough.
5472 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5473 int saved_reexecute_sp, uint new_idx) {
5474 if (saved_jvms_before_guards != nullptr && !stopped()) {
5475 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5476
5477 assert(alloc != nullptr, "only with a tightly coupled allocation");
5478 // restore JVM state to the state at the arraycopy
5479 saved_jvms_before_guards->map()->set_control(map()->control());
5480 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5481 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5482 // If we've improved the types of some nodes (null check) while
5483 // emitting the guards, propagate them to the current state
5484 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5485 set_jvms(saved_jvms_before_guards);
5486 _reexecute_sp = saved_reexecute_sp;
5487
5488 // Remove the allocation from above the guards
5489 CallProjections callprojs;
5490 alloc->extract_projections(&callprojs, true);
5491 InitializeNode* init = alloc->initialization();
5492 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5493 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5494 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5495
5496 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5497 // the allocation (i.e. is only valid if the allocation succeeds):
5498 // 1) replace CastIINode with AllocateArrayNode's length here
5499 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5500 //
5501 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5502 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5503 Node* init_control = init->proj_out(TypeFunc::Control);
5504 Node* alloc_length = alloc->Ideal_length();
5505 #ifdef ASSERT
5506 Node* prev_cast = nullptr;
5507 #endif
5508 for (uint i = 0; i < init_control->outcnt(); i++) {
5509 Node* init_out = init_control->raw_out(i);
5510 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5511 #ifdef ASSERT
5512 if (prev_cast == nullptr) {
5513 prev_cast = init_out;
5514 } else {
5515 if (prev_cast->cmp(*init_out) == false) {
5516 prev_cast->dump();
5517 init_out->dump();
5518 assert(false, "not equal CastIINode");
5519 }
5520 }
5521 #endif
5522 C->gvn_replace_by(init_out, alloc_length);
5523 }
5524 }
5525 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5526
5527 // move the allocation here (after the guards)
5528 _gvn.hash_delete(alloc);
5529 alloc->set_req(TypeFunc::Control, control());
5530 alloc->set_req(TypeFunc::I_O, i_o());
5531 Node *mem = reset_memory();
5532 set_all_memory(mem);
5533 alloc->set_req(TypeFunc::Memory, mem);
5534 set_control(init->proj_out_or_null(TypeFunc::Control));
5535 set_i_o(callprojs.fallthrough_ioproj);
5536
5537 // Update memory as done in GraphKit::set_output_for_allocation()
5538 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5539 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5540 if (ary_type->isa_aryptr() && length_type != nullptr) {
5541 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5542 }
5543 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5544 int elemidx = C->get_alias_index(telemref);
5545 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5546 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5547
5548 Node* allocx = _gvn.transform(alloc);
5549 assert(allocx == alloc, "where has the allocation gone?");
5550 assert(dest->is_CheckCastPP(), "not an allocation result?");
5551
5552 _gvn.hash_delete(dest);
5553 dest->set_req(0, control());
5554 Node* destx = _gvn.transform(dest);
5555 assert(destx == dest, "where has the allocation result gone?");
5556
5557 array_ideal_length(alloc, ary_type, true);
5558 }
5559 }
5560
5561 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5562 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5563 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5564 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5565 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5566 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5567 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5568 JVMState* saved_jvms_before_guards) {
5569 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5570 // There is at least one unrelated uncommon trap which needs to be replaced.
5571 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5572
5573 JVMState* saved_jvms = jvms();
5574 const int saved_reexecute_sp = _reexecute_sp;
5575 set_jvms(sfpt->jvms());
5576 _reexecute_sp = jvms()->sp();
5577
5578 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5579
5580 // Restore state
5581 set_jvms(saved_jvms);
5582 _reexecute_sp = saved_reexecute_sp;
5583 }
5584 }
5585
5586 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5587 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5588 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5589 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5590 while (if_proj->is_IfProj()) {
5591 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5592 if (uncommon_trap != nullptr) {
5593 create_new_uncommon_trap(uncommon_trap);
5594 }
5595 assert(if_proj->in(0)->is_If(), "must be If");
5596 if_proj = if_proj->in(0)->in(0);
5597 }
5598 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5599 "must have reached control projection of init node");
5600 }
5601
5602 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5603 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5604 assert(trap_request != 0, "no valid UCT trap request");
5605 PreserveJVMState pjvms(this);
5606 set_control(uncommon_trap_call->in(0));
5607 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5608 Deoptimization::trap_request_action(trap_request));
5609 assert(stopped(), "Should be stopped");
5610 _gvn.hash_delete(uncommon_trap_call);
5611 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5612 }
5613
5614 // Common checks for array sorting intrinsics arguments.
5615 // Returns `true` if checks passed.
5616 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
5617 // check address of the class
5618 if (elementType == nullptr || elementType->is_top()) {
5619 return false; // dead path
5620 }
5621 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5622 if (elem_klass == nullptr) {
5623 return false; // dead path
5624 }
5625 // java_mirror_type() returns non-null for compile-time Class constants only
5626 ciType* elem_type = elem_klass->java_mirror_type();
5627 if (elem_type == nullptr) {
5628 return false;
5629 }
5630 bt = elem_type->basic_type();
5631 // Disable the intrinsic if the CPU does not support SIMD sort
5632 if (!Matcher::supports_simd_sort(bt)) {
5633 return false;
5634 }
5635 // check address of the array
5636 if (obj == nullptr || obj->is_top()) {
5637 return false; // dead path
5638 }
5639 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5640 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
5641 return false; // failed input validation
5642 }
5643 return true;
5644 }
5645
5646 //------------------------------inline_array_partition-----------------------
5647 bool LibraryCallKit::inline_array_partition() {
5648 address stubAddr = StubRoutines::select_array_partition_function();
5649 if (stubAddr == nullptr) {
5650 return false; // Intrinsic's stub is not implemented on this platform
5651 }
5652 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
5653
5654 // no receiver because it is a static method
5655 Node* elementType = argument(0);
5656 Node* obj = argument(1);
5657 Node* offset = argument(2); // long
5658 Node* fromIndex = argument(4);
5659 Node* toIndex = argument(5);
5660 Node* indexPivot1 = argument(6);
5661 Node* indexPivot2 = argument(7);
5662 // PartitionOperation: argument(8) is ignored
5663
5664 Node* pivotIndices = nullptr;
5665 BasicType bt = T_ILLEGAL;
5666
5667 if (!check_array_sort_arguments(elementType, obj, bt)) {
5668 return false;
5669 }
5670 null_check(obj);
5671 // If obj is dead, only null-path is taken.
5672 if (stopped()) {
5673 return true;
5674 }
5675 // Set the original stack and the reexecute bit for the interpreter to reexecute
5676 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
5677 { PreserveReexecuteState preexecs(this);
5678 jvms()->set_should_reexecute(true);
5679
5680 Node* obj_adr = make_unsafe_address(obj, offset);
5681
5682 // create the pivotIndices array of type int and size = 2
5683 Node* size = intcon(2);
5684 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
5685 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
5686 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
5687 guarantee(alloc != nullptr, "created above");
5688 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
5689
5690 // pass the basic type enum to the stub
5691 Node* elemType = intcon(bt);
5692
5693 // Call the stub
5694 const char *stubName = "array_partition_stub";
5695 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
5696 stubAddr, stubName, TypePtr::BOTTOM,
5697 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
5698 indexPivot1, indexPivot2);
5699
5700 } // original reexecute is set back here
5701
5702 if (!stopped()) {
5703 set_result(pivotIndices);
5704 }
5705
5706 return true;
5707 }
5708
5709
5710 //------------------------------inline_array_sort-----------------------
5711 bool LibraryCallKit::inline_array_sort() {
5712 address stubAddr = StubRoutines::select_arraysort_function();
5713 if (stubAddr == nullptr) {
5714 return false; // Intrinsic's stub is not implemented on this platform
5715 }
5716 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
5717
5718 // no receiver because it is a static method
5719 Node* elementType = argument(0);
5720 Node* obj = argument(1);
5721 Node* offset = argument(2); // long
5722 Node* fromIndex = argument(4);
5723 Node* toIndex = argument(5);
5724 // SortOperation: argument(6) is ignored
5725
5726 BasicType bt = T_ILLEGAL;
5727
5728 if (!check_array_sort_arguments(elementType, obj, bt)) {
5729 return false;
5730 }
5731 null_check(obj);
5732 // If obj is dead, only null-path is taken.
5733 if (stopped()) {
5734 return true;
5735 }
5736 Node* obj_adr = make_unsafe_address(obj, offset);
5737
5738 // pass the basic type enum to the stub
5739 Node* elemType = intcon(bt);
5740
5741 // Call the stub.
5742 const char *stubName = "arraysort_stub";
5743 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
5744 stubAddr, stubName, TypePtr::BOTTOM,
5745 obj_adr, elemType, fromIndex, toIndex);
5746
5747 return true;
5748 }
5749
5750
5751 //------------------------------inline_arraycopy-----------------------
5752 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5753 // Object dest, int destPos,
5754 // int length);
5755 bool LibraryCallKit::inline_arraycopy() {
5756 // Get the arguments.
5757 Node* src = argument(0); // type: oop
5758 Node* src_offset = argument(1); // type: int
5759 Node* dest = argument(2); // type: oop
5760 Node* dest_offset = argument(3); // type: int
5761 Node* length = argument(4); // type: int
5762
5763 uint new_idx = C->unique();
5764
5765 // Check for allocation before we add nodes that would confuse
5766 // tightly_coupled_allocation()
5767 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5768
5769 int saved_reexecute_sp = -1;
5770 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5771 // See arraycopy_restore_alloc_state() comment
5772 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5773 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5774 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5775 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5776
5777 // The following tests must be performed
5778 // (1) src and dest are arrays.
5779 // (2) src and dest arrays must have elements of the same BasicType
5780 // (3) src and dest must not be null.
5781 // (4) src_offset must not be negative.
5782 // (5) dest_offset must not be negative.
5783 // (6) length must not be negative.
5784 // (7) src_offset + length must not exceed length of src.
5785 // (8) dest_offset + length must not exceed length of dest.
5786 // (9) each element of an oop array must be assignable
5787
5788 // (3) src and dest must not be null.
5789 // always do this here because we need the JVM state for uncommon traps
5790 Node* null_ctl = top();
5791 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5792 assert(null_ctl->is_top(), "no null control here");
5793 dest = null_check(dest, T_ARRAY);
5794
5795 if (!can_emit_guards) {
5796 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5797 // guards but the arraycopy node could still take advantage of a
5798 // tightly allocated allocation. tightly_coupled_allocation() is
5799 // called again to make sure it takes the null check above into
5800 // account: the null check is mandatory and if it caused an
5801 // uncommon trap to be emitted then the allocation can't be
5802 // considered tightly coupled in this context.
5803 alloc = tightly_coupled_allocation(dest);
5804 }
5805
5806 bool validated = false;
5807
5808 const Type* src_type = _gvn.type(src);
5809 const Type* dest_type = _gvn.type(dest);
5810 const TypeAryPtr* top_src = src_type->isa_aryptr();
5811 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5812
5813 // Do we have the type of src?
5814 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5815 // Do we have the type of dest?
5816 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5817 // Is the type for src from speculation?
5818 bool src_spec = false;
5819 // Is the type for dest from speculation?
5820 bool dest_spec = false;
5821
5822 if ((!has_src || !has_dest) && can_emit_guards) {
5823 // We don't have sufficient type information, let's see if
5824 // speculative types can help. We need to have types for both src
5825 // and dest so that it pays off.
5826
5827 // Do we already have or could we have type information for src
5828 bool could_have_src = has_src;
5829 // Do we already have or could we have type information for dest
5830 bool could_have_dest = has_dest;
5831
5832 ciKlass* src_k = nullptr;
5833 if (!has_src) {
5834 src_k = src_type->speculative_type_not_null();
5835 if (src_k != nullptr && src_k->is_array_klass()) {
5836 could_have_src = true;
5837 }
5838 }
5839
5840 ciKlass* dest_k = nullptr;
5841 if (!has_dest) {
5842 dest_k = dest_type->speculative_type_not_null();
5843 if (dest_k != nullptr && dest_k->is_array_klass()) {
5844 could_have_dest = true;
5845 }
5846 }
5847
5848 if (could_have_src && could_have_dest) {
5849 // This is going to pay off so emit the required guards
5850 if (!has_src) {
5851 src = maybe_cast_profiled_obj(src, src_k, true);
5852 src_type = _gvn.type(src);
5853 top_src = src_type->isa_aryptr();
5854 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5855 src_spec = true;
5856 }
5857 if (!has_dest) {
5858 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5859 dest_type = _gvn.type(dest);
5860 top_dest = dest_type->isa_aryptr();
5861 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5862 dest_spec = true;
5863 }
5864 }
5865 }
5866
5867 if (has_src && has_dest && can_emit_guards) {
5868 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5869 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5870 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5871 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5872
5873 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5874 // If both arrays are object arrays then having the exact types
5875 // for both will remove the need for a subtype check at runtime
5876 // before the call and may make it possible to pick a faster copy
5877 // routine (without a subtype check on every element)
5878 // Do we have the exact type of src?
5879 bool could_have_src = src_spec;
5880 // Do we have the exact type of dest?
5881 bool could_have_dest = dest_spec;
5882 ciKlass* src_k = nullptr;
5883 ciKlass* dest_k = nullptr;
5884 if (!src_spec) {
5885 src_k = src_type->speculative_type_not_null();
5886 if (src_k != nullptr && src_k->is_array_klass()) {
5887 could_have_src = true;
5888 }
5889 }
5890 if (!dest_spec) {
5891 dest_k = dest_type->speculative_type_not_null();
5892 if (dest_k != nullptr && dest_k->is_array_klass()) {
5893 could_have_dest = true;
5894 }
5895 }
5896 if (could_have_src && could_have_dest) {
5897 // If we can have both exact types, emit the missing guards
5898 if (could_have_src && !src_spec) {
5899 src = maybe_cast_profiled_obj(src, src_k, true);
5900 }
5901 if (could_have_dest && !dest_spec) {
5902 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5903 }
5904 }
5905 }
5906 }
5907
5908 ciMethod* trap_method = method();
5909 int trap_bci = bci();
5910 if (saved_jvms_before_guards != nullptr) {
5911 trap_method = alloc->jvms()->method();
5912 trap_bci = alloc->jvms()->bci();
5913 }
5914
5915 bool negative_length_guard_generated = false;
5916
5917 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5918 can_emit_guards &&
5919 !src->is_top() && !dest->is_top()) {
5920 // validate arguments: enables transformation the ArrayCopyNode
5921 validated = true;
5922
5923 RegionNode* slow_region = new RegionNode(1);
5924 record_for_igvn(slow_region);
5925
5926 // (1) src and dest are arrays.
5927 generate_non_array_guard(load_object_klass(src), slow_region, &src);
5928 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
5929
5930 // (2) src and dest arrays must have elements of the same BasicType
5931 // done at macro expansion or at Ideal transformation time
5932
5933 // (4) src_offset must not be negative.
5934 generate_negative_guard(src_offset, slow_region);
5935
5936 // (5) dest_offset must not be negative.
5937 generate_negative_guard(dest_offset, slow_region);
5938
5939 // (7) src_offset + length must not exceed length of src.
5940 generate_limit_guard(src_offset, length,
5941 load_array_length(src),
5942 slow_region);
5943
5944 // (8) dest_offset + length must not exceed length of dest.
5945 generate_limit_guard(dest_offset, length,
5946 load_array_length(dest),
5947 slow_region);
5948
5949 // (6) length must not be negative.
5950 // This is also checked in generate_arraycopy() during macro expansion, but
5951 // we also have to check it here for the case where the ArrayCopyNode will
5952 // be eliminated by Escape Analysis.
5953 if (EliminateAllocations) {
5954 generate_negative_guard(length, slow_region);
5955 negative_length_guard_generated = true;
5956 }
5957
5958 // (9) each element of an oop array must be assignable
5959 Node* dest_klass = load_object_klass(dest);
5960 if (src != dest) {
5961 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
5962
5963 if (not_subtype_ctrl != top()) {
5964 PreserveJVMState pjvms(this);
5965 set_control(not_subtype_ctrl);
5966 uncommon_trap(Deoptimization::Reason_intrinsic,
5967 Deoptimization::Action_make_not_entrant);
5968 assert(stopped(), "Should be stopped");
5969 }
5970 }
5971 {
5972 PreserveJVMState pjvms(this);
5973 set_control(_gvn.transform(slow_region));
5974 uncommon_trap(Deoptimization::Reason_intrinsic,
5975 Deoptimization::Action_make_not_entrant);
5976 assert(stopped(), "Should be stopped");
5977 }
5978
5979 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5980 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
5981 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5982 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
5983 }
5984
5985 if (stopped()) {
5986 return true;
5987 }
5988
5989 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
5990 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5991 // so the compiler has a chance to eliminate them: during macro expansion,
5992 // we have to set their control (CastPP nodes are eliminated).
5993 load_object_klass(src), load_object_klass(dest),
5994 load_array_length(src), load_array_length(dest));
5995
5996 ac->set_arraycopy(validated);
5997
5998 Node* n = _gvn.transform(ac);
5999 if (n == ac) {
6000 ac->connect_outputs(this);
6001 } else {
6002 assert(validated, "shouldn't transform if all arguments not validated");
6003 set_all_memory(n);
6004 }
6005 clear_upper_avx();
6006
6007
6008 return true;
6009 }
6010
6011
6012 // Helper function which determines if an arraycopy immediately follows
6013 // an allocation, with no intervening tests or other escapes for the object.
6014 AllocateArrayNode*
6015 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6016 if (stopped()) return nullptr; // no fast path
6017 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6018
6019 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6020 if (alloc == nullptr) return nullptr;
6021
6022 Node* rawmem = memory(Compile::AliasIdxRaw);
6023 // Is the allocation's memory state untouched?
6024 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6025 // Bail out if there have been raw-memory effects since the allocation.
6026 // (Example: There might have been a call or safepoint.)
6027 return nullptr;
6028 }
6029 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6030 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6031 return nullptr;
6032 }
6033
6034 // There must be no unexpected observers of this allocation.
6035 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6036 Node* obs = ptr->fast_out(i);
6037 if (obs != this->map()) {
6038 return nullptr;
6039 }
6040 }
6041
6042 // This arraycopy must unconditionally follow the allocation of the ptr.
6043 Node* alloc_ctl = ptr->in(0);
6044 Node* ctl = control();
6045 while (ctl != alloc_ctl) {
6046 // There may be guards which feed into the slow_region.
6047 // Any other control flow means that we might not get a chance
6048 // to finish initializing the allocated object.
6049 // Various low-level checks bottom out in uncommon traps. These
6050 // are considered safe since we've already checked above that
6051 // there is no unexpected observer of this allocation.
6052 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6053 assert(ctl->in(0)->is_If(), "must be If");
6054 ctl = ctl->in(0)->in(0);
6055 } else {
6056 return nullptr;
6057 }
6058 }
6059
6060 // If we get this far, we have an allocation which immediately
6061 // precedes the arraycopy, and we can take over zeroing the new object.
6062 // The arraycopy will finish the initialization, and provide
6063 // a new control state to which we will anchor the destination pointer.
6064
6065 return alloc;
6066 }
6067
6068 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6069 if (node->is_IfProj()) {
6070 Node* other_proj = node->as_IfProj()->other_if_proj();
6071 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6072 Node* obs = other_proj->fast_out(j);
6073 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6074 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6075 return obs->as_CallStaticJava();
6076 }
6077 }
6078 }
6079 return nullptr;
6080 }
6081
6082 //-------------inline_encodeISOArray-----------------------------------
6083 // encode char[] to byte[] in ISO_8859_1 or ASCII
6084 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6085 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6086 // no receiver since it is static method
6087 Node *src = argument(0);
6088 Node *src_offset = argument(1);
6089 Node *dst = argument(2);
6090 Node *dst_offset = argument(3);
6091 Node *length = argument(4);
6092
6093 src = must_be_not_null(src, true);
6094 dst = must_be_not_null(dst, true);
6095
6096 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6097 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6098 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6099 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6100 // failed array check
6101 return false;
6102 }
6103
6104 // Figure out the size and type of the elements we will be copying.
6105 BasicType src_elem = src_type->elem()->array_element_basic_type();
6106 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6107 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6108 return false;
6109 }
6110
6111 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6112 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6113 // 'src_start' points to src array + scaled offset
6114 // 'dst_start' points to dst array + scaled offset
6115
6116 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6117 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6118 enc = _gvn.transform(enc);
6119 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6120 set_memory(res_mem, mtype);
6121 set_result(enc);
6122 clear_upper_avx();
6123
6124 return true;
6125 }
6126
6127 //-------------inline_multiplyToLen-----------------------------------
6128 bool LibraryCallKit::inline_multiplyToLen() {
6129 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6130
6131 address stubAddr = StubRoutines::multiplyToLen();
6132 if (stubAddr == nullptr) {
6133 return false; // Intrinsic's stub is not implemented on this platform
6134 }
6135 const char* stubName = "multiplyToLen";
6136
6137 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6138
6139 // no receiver because it is a static method
6140 Node* x = argument(0);
6141 Node* xlen = argument(1);
6142 Node* y = argument(2);
6143 Node* ylen = argument(3);
6144 Node* z = argument(4);
6145
6146 x = must_be_not_null(x, true);
6147 y = must_be_not_null(y, true);
6148
6149 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6150 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6151 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6152 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6153 // failed array check
6154 return false;
6155 }
6156
6157 BasicType x_elem = x_type->elem()->array_element_basic_type();
6158 BasicType y_elem = y_type->elem()->array_element_basic_type();
6159 if (x_elem != T_INT || y_elem != T_INT) {
6160 return false;
6161 }
6162
6163 Node* x_start = array_element_address(x, intcon(0), x_elem);
6164 Node* y_start = array_element_address(y, intcon(0), y_elem);
6165 // 'x_start' points to x array + scaled xlen
6166 // 'y_start' points to y array + scaled ylen
6167
6168 Node* z_start = array_element_address(z, intcon(0), T_INT);
6169
6170 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6171 OptoRuntime::multiplyToLen_Type(),
6172 stubAddr, stubName, TypePtr::BOTTOM,
6173 x_start, xlen, y_start, ylen, z_start);
6174
6175 C->set_has_split_ifs(true); // Has chance for split-if optimization
6176 set_result(z);
6177 return true;
6178 }
6179
6180 //-------------inline_squareToLen------------------------------------
6181 bool LibraryCallKit::inline_squareToLen() {
6182 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6183
6184 address stubAddr = StubRoutines::squareToLen();
6185 if (stubAddr == nullptr) {
6186 return false; // Intrinsic's stub is not implemented on this platform
6187 }
6188 const char* stubName = "squareToLen";
6189
6190 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6191
6192 Node* x = argument(0);
6193 Node* len = argument(1);
6194 Node* z = argument(2);
6195 Node* zlen = argument(3);
6196
6197 x = must_be_not_null(x, true);
6198 z = must_be_not_null(z, true);
6199
6200 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6201 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6202 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6203 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6204 // failed array check
6205 return false;
6206 }
6207
6208 BasicType x_elem = x_type->elem()->array_element_basic_type();
6209 BasicType z_elem = z_type->elem()->array_element_basic_type();
6210 if (x_elem != T_INT || z_elem != T_INT) {
6211 return false;
6212 }
6213
6214
6215 Node* x_start = array_element_address(x, intcon(0), x_elem);
6216 Node* z_start = array_element_address(z, intcon(0), z_elem);
6217
6218 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6219 OptoRuntime::squareToLen_Type(),
6220 stubAddr, stubName, TypePtr::BOTTOM,
6221 x_start, len, z_start, zlen);
6222
6223 set_result(z);
6224 return true;
6225 }
6226
6227 //-------------inline_mulAdd------------------------------------------
6228 bool LibraryCallKit::inline_mulAdd() {
6229 assert(UseMulAddIntrinsic, "not implemented on this platform");
6230
6231 address stubAddr = StubRoutines::mulAdd();
6232 if (stubAddr == nullptr) {
6233 return false; // Intrinsic's stub is not implemented on this platform
6234 }
6235 const char* stubName = "mulAdd";
6236
6237 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6238
6239 Node* out = argument(0);
6240 Node* in = argument(1);
6241 Node* offset = argument(2);
6242 Node* len = argument(3);
6243 Node* k = argument(4);
6244
6245 in = must_be_not_null(in, true);
6246 out = must_be_not_null(out, true);
6247
6248 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6249 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6250 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6251 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6252 // failed array check
6253 return false;
6254 }
6255
6256 BasicType out_elem = out_type->elem()->array_element_basic_type();
6257 BasicType in_elem = in_type->elem()->array_element_basic_type();
6258 if (out_elem != T_INT || in_elem != T_INT) {
6259 return false;
6260 }
6261
6262 Node* outlen = load_array_length(out);
6263 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6264 Node* out_start = array_element_address(out, intcon(0), out_elem);
6265 Node* in_start = array_element_address(in, intcon(0), in_elem);
6266
6267 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6268 OptoRuntime::mulAdd_Type(),
6269 stubAddr, stubName, TypePtr::BOTTOM,
6270 out_start,in_start, new_offset, len, k);
6271 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6272 set_result(result);
6273 return true;
6274 }
6275
6276 //-------------inline_montgomeryMultiply-----------------------------------
6277 bool LibraryCallKit::inline_montgomeryMultiply() {
6278 address stubAddr = StubRoutines::montgomeryMultiply();
6279 if (stubAddr == nullptr) {
6280 return false; // Intrinsic's stub is not implemented on this platform
6281 }
6282
6283 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6284 const char* stubName = "montgomery_multiply";
6285
6286 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6287
6288 Node* a = argument(0);
6289 Node* b = argument(1);
6290 Node* n = argument(2);
6291 Node* len = argument(3);
6292 Node* inv = argument(4);
6293 Node* m = argument(6);
6294
6295 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6296 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6297 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6298 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6299 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6300 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6301 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6302 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6303 // failed array check
6304 return false;
6305 }
6306
6307 BasicType a_elem = a_type->elem()->array_element_basic_type();
6308 BasicType b_elem = b_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 || b_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* b_start = array_element_address(b, intcon(0), b_elem);
6319 Node* n_start = array_element_address(n, intcon(0), n_elem);
6320 Node* m_start = array_element_address(m, intcon(0), m_elem);
6321
6322 Node* call = make_runtime_call(RC_LEAF,
6323 OptoRuntime::montgomeryMultiply_Type(),
6324 stubAddr, stubName, TypePtr::BOTTOM,
6325 a_start, b_start, n_start, len, inv, top(),
6326 m_start);
6327 set_result(m);
6328 }
6329
6330 return true;
6331 }
6332
6333 bool LibraryCallKit::inline_montgomerySquare() {
6334 address stubAddr = StubRoutines::montgomerySquare();
6335 if (stubAddr == nullptr) {
6336 return false; // Intrinsic's stub is not implemented on this platform
6337 }
6338
6339 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6340 const char* stubName = "montgomery_square";
6341
6342 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6343
6344 Node* a = argument(0);
6345 Node* n = argument(1);
6346 Node* len = argument(2);
6347 Node* inv = argument(3);
6348 Node* m = argument(5);
6349
6350 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6351 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6352 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6353 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6354 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6355 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6356 // failed array check
6357 return false;
6358 }
6359
6360 BasicType a_elem = a_type->elem()->array_element_basic_type();
6361 BasicType n_elem = n_type->elem()->array_element_basic_type();
6362 BasicType m_elem = m_type->elem()->array_element_basic_type();
6363 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6364 return false;
6365 }
6366
6367 // Make the call
6368 {
6369 Node* a_start = array_element_address(a, intcon(0), a_elem);
6370 Node* n_start = array_element_address(n, intcon(0), n_elem);
6371 Node* m_start = array_element_address(m, intcon(0), m_elem);
6372
6373 Node* call = make_runtime_call(RC_LEAF,
6374 OptoRuntime::montgomerySquare_Type(),
6375 stubAddr, stubName, TypePtr::BOTTOM,
6376 a_start, n_start, len, inv, top(),
6377 m_start);
6378 set_result(m);
6379 }
6380
6381 return true;
6382 }
6383
6384 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6385 address stubAddr = nullptr;
6386 const char* stubName = nullptr;
6387
6388 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6389 if (stubAddr == nullptr) {
6390 return false; // Intrinsic's stub is not implemented on this platform
6391 }
6392
6393 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6394
6395 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6396
6397 Node* newArr = argument(0);
6398 Node* oldArr = argument(1);
6399 Node* newIdx = argument(2);
6400 Node* shiftCount = argument(3);
6401 Node* numIter = argument(4);
6402
6403 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6404 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6405 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6406 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6407 return false;
6408 }
6409
6410 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6411 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6412 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6413 return false;
6414 }
6415
6416 // Make the call
6417 {
6418 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6419 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6420
6421 Node* call = make_runtime_call(RC_LEAF,
6422 OptoRuntime::bigIntegerShift_Type(),
6423 stubAddr,
6424 stubName,
6425 TypePtr::BOTTOM,
6426 newArr_start,
6427 oldArr_start,
6428 newIdx,
6429 shiftCount,
6430 numIter);
6431 }
6432
6433 return true;
6434 }
6435
6436 //-------------inline_vectorizedMismatch------------------------------
6437 bool LibraryCallKit::inline_vectorizedMismatch() {
6438 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6439
6440 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6441 Node* obja = argument(0); // Object
6442 Node* aoffset = argument(1); // long
6443 Node* objb = argument(3); // Object
6444 Node* boffset = argument(4); // long
6445 Node* length = argument(6); // int
6446 Node* scale = argument(7); // int
6447
6448 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6449 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6450 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6451 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6452 scale == top()) {
6453 return false; // failed input validation
6454 }
6455
6456 Node* obja_adr = make_unsafe_address(obja, aoffset);
6457 Node* objb_adr = make_unsafe_address(objb, boffset);
6458
6459 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6460 //
6461 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6462 // if (length <= inline_limit) {
6463 // inline_path:
6464 // vmask = VectorMaskGen length
6465 // vload1 = LoadVectorMasked obja, vmask
6466 // vload2 = LoadVectorMasked objb, vmask
6467 // result1 = VectorCmpMasked vload1, vload2, vmask
6468 // } else {
6469 // call_stub_path:
6470 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6471 // }
6472 // exit_block:
6473 // return Phi(result1, result2);
6474 //
6475 enum { inline_path = 1, // input is small enough to process it all at once
6476 stub_path = 2, // input is too large; call into the VM
6477 PATH_LIMIT = 3
6478 };
6479
6480 Node* exit_block = new RegionNode(PATH_LIMIT);
6481 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6482 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6483
6484 Node* call_stub_path = control();
6485
6486 BasicType elem_bt = T_ILLEGAL;
6487
6488 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6489 if (scale_t->is_con()) {
6490 switch (scale_t->get_con()) {
6491 case 0: elem_bt = T_BYTE; break;
6492 case 1: elem_bt = T_SHORT; break;
6493 case 2: elem_bt = T_INT; break;
6494 case 3: elem_bt = T_LONG; break;
6495
6496 default: elem_bt = T_ILLEGAL; break; // not supported
6497 }
6498 }
6499
6500 int inline_limit = 0;
6501 bool do_partial_inline = false;
6502
6503 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6504 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6505 do_partial_inline = inline_limit >= 16;
6506 }
6507
6508 if (do_partial_inline) {
6509 assert(elem_bt != T_ILLEGAL, "sanity");
6510
6511 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6512 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6513 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6514
6515 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6516 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6517 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6518
6519 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6520
6521 if (!stopped()) {
6522 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6523
6524 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6525 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6526 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6527 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6528
6529 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6530 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6531 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6532 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6533
6534 exit_block->init_req(inline_path, control());
6535 memory_phi->init_req(inline_path, map()->memory());
6536 result_phi->init_req(inline_path, result);
6537
6538 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6539 clear_upper_avx();
6540 }
6541 }
6542 }
6543
6544 if (call_stub_path != nullptr) {
6545 set_control(call_stub_path);
6546
6547 Node* call = make_runtime_call(RC_LEAF,
6548 OptoRuntime::vectorizedMismatch_Type(),
6549 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6550 obja_adr, objb_adr, length, scale);
6551
6552 exit_block->init_req(stub_path, control());
6553 memory_phi->init_req(stub_path, map()->memory());
6554 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6555 }
6556
6557 exit_block = _gvn.transform(exit_block);
6558 memory_phi = _gvn.transform(memory_phi);
6559 result_phi = _gvn.transform(result_phi);
6560
6561 set_control(exit_block);
6562 set_all_memory(memory_phi);
6563 set_result(result_phi);
6564
6565 return true;
6566 }
6567
6568 //------------------------------inline_vectorizedHashcode----------------------------
6569 bool LibraryCallKit::inline_vectorizedHashCode() {
6570 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6571
6572 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6573 Node* array = argument(0);
6574 Node* offset = argument(1);
6575 Node* length = argument(2);
6576 Node* initialValue = argument(3);
6577 Node* basic_type = argument(4);
6578
6579 if (basic_type == top()) {
6580 return false; // failed input validation
6581 }
6582
6583 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6584 if (!basic_type_t->is_con()) {
6585 return false; // Only intrinsify if mode argument is constant
6586 }
6587
6588 array = must_be_not_null(array, true);
6589
6590 BasicType bt = (BasicType)basic_type_t->get_con();
6591
6592 // Resolve address of first element
6593 Node* array_start = array_element_address(array, offset, bt);
6594
6595 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6596 array_start, length, initialValue, basic_type)));
6597 clear_upper_avx();
6598
6599 return true;
6600 }
6601
6602 /**
6603 * Calculate CRC32 for byte.
6604 * int java.util.zip.CRC32.update(int crc, int b)
6605 */
6606 bool LibraryCallKit::inline_updateCRC32() {
6607 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6608 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6609 // no receiver since it is static method
6610 Node* crc = argument(0); // type: int
6611 Node* b = argument(1); // type: int
6612
6613 /*
6614 * int c = ~ crc;
6615 * b = timesXtoThe32[(b ^ c) & 0xFF];
6616 * b = b ^ (c >>> 8);
6617 * crc = ~b;
6618 */
6619
6620 Node* M1 = intcon(-1);
6621 crc = _gvn.transform(new XorINode(crc, M1));
6622 Node* result = _gvn.transform(new XorINode(crc, b));
6623 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6624
6625 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6626 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6627 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6628 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6629
6630 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6631 result = _gvn.transform(new XorINode(crc, result));
6632 result = _gvn.transform(new XorINode(result, M1));
6633 set_result(result);
6634 return true;
6635 }
6636
6637 /**
6638 * Calculate CRC32 for byte[] array.
6639 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6640 */
6641 bool LibraryCallKit::inline_updateBytesCRC32() {
6642 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6643 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6644 // no receiver since it is static method
6645 Node* crc = argument(0); // type: int
6646 Node* src = argument(1); // type: oop
6647 Node* offset = argument(2); // type: int
6648 Node* length = argument(3); // type: int
6649
6650 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6651 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6652 // failed array check
6653 return false;
6654 }
6655
6656 // Figure out the size and type of the elements we will be copying.
6657 BasicType src_elem = src_type->elem()->array_element_basic_type();
6658 if (src_elem != T_BYTE) {
6659 return false;
6660 }
6661
6662 // 'src_start' points to src array + scaled offset
6663 src = must_be_not_null(src, true);
6664 Node* src_start = array_element_address(src, offset, src_elem);
6665
6666 // We assume that range check is done by caller.
6667 // TODO: generate range check (offset+length < src.length) in debug VM.
6668
6669 // Call the stub.
6670 address stubAddr = StubRoutines::updateBytesCRC32();
6671 const char *stubName = "updateBytesCRC32";
6672
6673 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6674 stubAddr, stubName, TypePtr::BOTTOM,
6675 crc, src_start, length);
6676 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6677 set_result(result);
6678 return true;
6679 }
6680
6681 /**
6682 * Calculate CRC32 for ByteBuffer.
6683 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6684 */
6685 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6686 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6687 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6688 // no receiver since it is static method
6689 Node* crc = argument(0); // type: int
6690 Node* src = argument(1); // type: long
6691 Node* offset = argument(3); // type: int
6692 Node* length = argument(4); // type: int
6693
6694 src = ConvL2X(src); // adjust Java long to machine word
6695 Node* base = _gvn.transform(new CastX2PNode(src));
6696 offset = ConvI2X(offset);
6697
6698 // 'src_start' points to src array + scaled offset
6699 Node* src_start = basic_plus_adr(top(), base, offset);
6700
6701 // Call the stub.
6702 address stubAddr = StubRoutines::updateBytesCRC32();
6703 const char *stubName = "updateBytesCRC32";
6704
6705 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6706 stubAddr, stubName, TypePtr::BOTTOM,
6707 crc, src_start, length);
6708 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6709 set_result(result);
6710 return true;
6711 }
6712
6713 //------------------------------get_table_from_crc32c_class-----------------------
6714 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6715 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6716 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6717
6718 return table;
6719 }
6720
6721 //------------------------------inline_updateBytesCRC32C-----------------------
6722 //
6723 // Calculate CRC32C for byte[] array.
6724 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6725 //
6726 bool LibraryCallKit::inline_updateBytesCRC32C() {
6727 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6728 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6729 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6730 // no receiver since it is a static method
6731 Node* crc = argument(0); // type: int
6732 Node* src = argument(1); // type: oop
6733 Node* offset = argument(2); // type: int
6734 Node* end = argument(3); // type: int
6735
6736 Node* length = _gvn.transform(new SubINode(end, offset));
6737
6738 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6739 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6740 // failed array check
6741 return false;
6742 }
6743
6744 // Figure out the size and type of the elements we will be copying.
6745 BasicType src_elem = src_type->elem()->array_element_basic_type();
6746 if (src_elem != T_BYTE) {
6747 return false;
6748 }
6749
6750 // 'src_start' points to src array + scaled offset
6751 src = must_be_not_null(src, true);
6752 Node* src_start = array_element_address(src, offset, src_elem);
6753
6754 // static final int[] byteTable in class CRC32C
6755 Node* table = get_table_from_crc32c_class(callee()->holder());
6756 table = must_be_not_null(table, true);
6757 Node* table_start = array_element_address(table, intcon(0), T_INT);
6758
6759 // We assume that range check is done by caller.
6760 // TODO: generate range check (offset+length < src.length) in debug VM.
6761
6762 // Call the stub.
6763 address stubAddr = StubRoutines::updateBytesCRC32C();
6764 const char *stubName = "updateBytesCRC32C";
6765
6766 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6767 stubAddr, stubName, TypePtr::BOTTOM,
6768 crc, src_start, length, table_start);
6769 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6770 set_result(result);
6771 return true;
6772 }
6773
6774 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6775 //
6776 // Calculate CRC32C for DirectByteBuffer.
6777 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6778 //
6779 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6780 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6781 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6782 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6783 // no receiver since it is a static method
6784 Node* crc = argument(0); // type: int
6785 Node* src = argument(1); // type: long
6786 Node* offset = argument(3); // type: int
6787 Node* end = argument(4); // type: int
6788
6789 Node* length = _gvn.transform(new SubINode(end, offset));
6790
6791 src = ConvL2X(src); // adjust Java long to machine word
6792 Node* base = _gvn.transform(new CastX2PNode(src));
6793 offset = ConvI2X(offset);
6794
6795 // 'src_start' points to src array + scaled offset
6796 Node* src_start = basic_plus_adr(top(), base, offset);
6797
6798 // static final int[] byteTable in class CRC32C
6799 Node* table = get_table_from_crc32c_class(callee()->holder());
6800 table = must_be_not_null(table, true);
6801 Node* table_start = array_element_address(table, intcon(0), T_INT);
6802
6803 // Call the stub.
6804 address stubAddr = StubRoutines::updateBytesCRC32C();
6805 const char *stubName = "updateBytesCRC32C";
6806
6807 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6808 stubAddr, stubName, TypePtr::BOTTOM,
6809 crc, src_start, length, table_start);
6810 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6811 set_result(result);
6812 return true;
6813 }
6814
6815 //------------------------------inline_updateBytesAdler32----------------------
6816 //
6817 // Calculate Adler32 checksum for byte[] array.
6818 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6819 //
6820 bool LibraryCallKit::inline_updateBytesAdler32() {
6821 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6822 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6823 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6824 // no receiver since it is static method
6825 Node* crc = argument(0); // type: int
6826 Node* src = argument(1); // type: oop
6827 Node* offset = argument(2); // type: int
6828 Node* length = argument(3); // type: int
6829
6830 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6831 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6832 // failed array check
6833 return false;
6834 }
6835
6836 // Figure out the size and type of the elements we will be copying.
6837 BasicType src_elem = src_type->elem()->array_element_basic_type();
6838 if (src_elem != T_BYTE) {
6839 return false;
6840 }
6841
6842 // 'src_start' points to src array + scaled offset
6843 Node* src_start = array_element_address(src, offset, src_elem);
6844
6845 // We assume that range check is done by caller.
6846 // TODO: generate range check (offset+length < src.length) in debug VM.
6847
6848 // Call the stub.
6849 address stubAddr = StubRoutines::updateBytesAdler32();
6850 const char *stubName = "updateBytesAdler32";
6851
6852 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6853 stubAddr, stubName, TypePtr::BOTTOM,
6854 crc, src_start, length);
6855 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6856 set_result(result);
6857 return true;
6858 }
6859
6860 //------------------------------inline_updateByteBufferAdler32---------------
6861 //
6862 // Calculate Adler32 checksum for DirectByteBuffer.
6863 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6864 //
6865 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6866 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6867 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6868 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6869 // no receiver since it is static method
6870 Node* crc = argument(0); // type: int
6871 Node* src = argument(1); // type: long
6872 Node* offset = argument(3); // type: int
6873 Node* length = argument(4); // type: int
6874
6875 src = ConvL2X(src); // adjust Java long to machine word
6876 Node* base = _gvn.transform(new CastX2PNode(src));
6877 offset = ConvI2X(offset);
6878
6879 // 'src_start' points to src array + scaled offset
6880 Node* src_start = basic_plus_adr(top(), base, offset);
6881
6882 // Call the stub.
6883 address stubAddr = StubRoutines::updateBytesAdler32();
6884 const char *stubName = "updateBytesAdler32";
6885
6886 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6887 stubAddr, stubName, TypePtr::BOTTOM,
6888 crc, src_start, length);
6889
6890 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6891 set_result(result);
6892 return true;
6893 }
6894
6895 //----------------------------inline_reference_get----------------------------
6896 // public T java.lang.ref.Reference.get();
6897 bool LibraryCallKit::inline_reference_get() {
6898 const int referent_offset = java_lang_ref_Reference::referent_offset();
6899
6900 // Get the argument:
6901 Node* reference_obj = null_check_receiver();
6902 if (stopped()) return true;
6903
6904 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6905 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6906 decorators, /*is_static*/ false, nullptr);
6907 if (result == nullptr) return false;
6908
6909 // Add memory barrier to prevent commoning reads from this field
6910 // across safepoint since GC can change its value.
6911 insert_mem_bar(Op_MemBarCPUOrder);
6912
6913 set_result(result);
6914 return true;
6915 }
6916
6917 //----------------------------inline_reference_refersTo0----------------------------
6918 // bool java.lang.ref.Reference.refersTo0();
6919 // bool java.lang.ref.PhantomReference.refersTo0();
6920 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6921 // Get arguments:
6922 Node* reference_obj = null_check_receiver();
6923 Node* other_obj = argument(1);
6924 if (stopped()) return true;
6925
6926 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6927 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6928 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6929 decorators, /*is_static*/ false, nullptr);
6930 if (referent == nullptr) return false;
6931
6932 // Add memory barrier to prevent commoning reads from this field
6933 // across safepoint since GC can change its value.
6934 insert_mem_bar(Op_MemBarCPUOrder);
6935
6936 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
6937 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6938 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
6939
6940 RegionNode* region = new RegionNode(3);
6941 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
6942
6943 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
6944 region->init_req(1, if_true);
6945 phi->init_req(1, intcon(1));
6946
6947 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
6948 region->init_req(2, if_false);
6949 phi->init_req(2, intcon(0));
6950
6951 set_control(_gvn.transform(region));
6952 record_for_igvn(region);
6953 set_result(_gvn.transform(phi));
6954 return true;
6955 }
6956
6957 //----------------------------inline_reference_clear0----------------------------
6958 // void java.lang.ref.Reference.clear0();
6959 // void java.lang.ref.PhantomReference.clear0();
6960 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
6961 // This matches the implementation in JVM_ReferenceClear, see the comments there.
6962
6963 // Get arguments
6964 Node* reference_obj = null_check_receiver();
6965 if (stopped()) return true;
6966
6967 // Common access parameters
6968 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6969 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6970 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
6971 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
6972 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
6973
6974 Node* referent = access_load_at(reference_obj,
6975 referent_field_addr,
6976 referent_field_addr_type,
6977 val_type,
6978 T_OBJECT,
6979 decorators);
6980
6981 IdealKit ideal(this);
6982 #define __ ideal.
6983 __ if_then(referent, BoolTest::ne, null());
6984 sync_kit(ideal);
6985 access_store_at(reference_obj,
6986 referent_field_addr,
6987 referent_field_addr_type,
6988 null(),
6989 val_type,
6990 T_OBJECT,
6991 decorators);
6992 __ sync_kit(this);
6993 __ end_if();
6994 final_sync(ideal);
6995 #undef __
6996
6997 return true;
6998 }
6999
7000 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7001 DecoratorSet decorators, bool is_static,
7002 ciInstanceKlass* fromKls) {
7003 if (fromKls == nullptr) {
7004 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7005 assert(tinst != nullptr, "obj is null");
7006 assert(tinst->is_loaded(), "obj is not loaded");
7007 fromKls = tinst->instance_klass();
7008 } else {
7009 assert(is_static, "only for static field access");
7010 }
7011 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7012 ciSymbol::make(fieldTypeString),
7013 is_static);
7014
7015 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7016 if (field == nullptr) return (Node *) nullptr;
7017
7018 if (is_static) {
7019 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7020 fromObj = makecon(tip);
7021 }
7022
7023 // Next code copied from Parse::do_get_xxx():
7024
7025 // Compute address and memory type.
7026 int offset = field->offset_in_bytes();
7027 bool is_vol = field->is_volatile();
7028 ciType* field_klass = field->type();
7029 assert(field_klass->is_loaded(), "should be loaded");
7030 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7031 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7032 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7033 "slice of address and input slice don't match");
7034 BasicType bt = field->layout_type();
7035
7036 // Build the resultant type of the load
7037 const Type *type;
7038 if (bt == T_OBJECT) {
7039 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7040 } else {
7041 type = Type::get_const_basic_type(bt);
7042 }
7043
7044 if (is_vol) {
7045 decorators |= MO_SEQ_CST;
7046 }
7047
7048 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7049 }
7050
7051 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7052 bool is_exact /* true */, bool is_static /* false */,
7053 ciInstanceKlass * fromKls /* nullptr */) {
7054 if (fromKls == nullptr) {
7055 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7056 assert(tinst != nullptr, "obj is null");
7057 assert(tinst->is_loaded(), "obj is not loaded");
7058 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7059 fromKls = tinst->instance_klass();
7060 }
7061 else {
7062 assert(is_static, "only for static field access");
7063 }
7064 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7065 ciSymbol::make(fieldTypeString),
7066 is_static);
7067
7068 assert(field != nullptr, "undefined field");
7069 assert(!field->is_volatile(), "not defined for volatile fields");
7070
7071 if (is_static) {
7072 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7073 fromObj = makecon(tip);
7074 }
7075
7076 // Next code copied from Parse::do_get_xxx():
7077
7078 // Compute address and memory type.
7079 int offset = field->offset_in_bytes();
7080 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7081
7082 return adr;
7083 }
7084
7085 //------------------------------inline_aescrypt_Block-----------------------
7086 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7087 address stubAddr = nullptr;
7088 const char *stubName;
7089 assert(UseAES, "need AES instruction support");
7090
7091 switch(id) {
7092 case vmIntrinsics::_aescrypt_encryptBlock:
7093 stubAddr = StubRoutines::aescrypt_encryptBlock();
7094 stubName = "aescrypt_encryptBlock";
7095 break;
7096 case vmIntrinsics::_aescrypt_decryptBlock:
7097 stubAddr = StubRoutines::aescrypt_decryptBlock();
7098 stubName = "aescrypt_decryptBlock";
7099 break;
7100 default:
7101 break;
7102 }
7103 if (stubAddr == nullptr) return false;
7104
7105 Node* aescrypt_object = argument(0);
7106 Node* src = argument(1);
7107 Node* src_offset = argument(2);
7108 Node* dest = argument(3);
7109 Node* dest_offset = argument(4);
7110
7111 src = must_be_not_null(src, true);
7112 dest = must_be_not_null(dest, true);
7113
7114 // (1) src and dest are arrays.
7115 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7116 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7117 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7118 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7119
7120 // for the quick and dirty code we will skip all the checks.
7121 // we are just trying to get the call to be generated.
7122 Node* src_start = src;
7123 Node* dest_start = dest;
7124 if (src_offset != nullptr || dest_offset != nullptr) {
7125 assert(src_offset != nullptr && dest_offset != nullptr, "");
7126 src_start = array_element_address(src, src_offset, T_BYTE);
7127 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7128 }
7129
7130 // now need to get the start of its expanded key array
7131 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7132 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7133 if (k_start == nullptr) return false;
7134
7135 // Call the stub.
7136 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7137 stubAddr, stubName, TypePtr::BOTTOM,
7138 src_start, dest_start, k_start);
7139
7140 return true;
7141 }
7142
7143 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7144 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7145 address stubAddr = nullptr;
7146 const char *stubName = nullptr;
7147
7148 assert(UseAES, "need AES instruction support");
7149
7150 switch(id) {
7151 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7152 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7153 stubName = "cipherBlockChaining_encryptAESCrypt";
7154 break;
7155 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7156 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7157 stubName = "cipherBlockChaining_decryptAESCrypt";
7158 break;
7159 default:
7160 break;
7161 }
7162 if (stubAddr == nullptr) return false;
7163
7164 Node* cipherBlockChaining_object = argument(0);
7165 Node* src = argument(1);
7166 Node* src_offset = argument(2);
7167 Node* len = argument(3);
7168 Node* dest = argument(4);
7169 Node* dest_offset = argument(5);
7170
7171 src = must_be_not_null(src, false);
7172 dest = must_be_not_null(dest, false);
7173
7174 // (1) src and dest are arrays.
7175 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7176 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7177 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7178 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7179
7180 // checks are the responsibility of the caller
7181 Node* src_start = src;
7182 Node* dest_start = dest;
7183 if (src_offset != nullptr || dest_offset != nullptr) {
7184 assert(src_offset != nullptr && dest_offset != nullptr, "");
7185 src_start = array_element_address(src, src_offset, T_BYTE);
7186 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7187 }
7188
7189 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7190 // (because of the predicated logic executed earlier).
7191 // so we cast it here safely.
7192 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7193
7194 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7195 if (embeddedCipherObj == nullptr) return false;
7196
7197 // cast it to what we know it will be at runtime
7198 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7199 assert(tinst != nullptr, "CBC obj is null");
7200 assert(tinst->is_loaded(), "CBC obj is not loaded");
7201 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7202 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7203
7204 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7205 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7206 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7207 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7208 aescrypt_object = _gvn.transform(aescrypt_object);
7209
7210 // we need to get the start of the aescrypt_object's expanded key array
7211 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7212 if (k_start == nullptr) return false;
7213
7214 // similarly, get the start address of the r vector
7215 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7216 if (objRvec == nullptr) return false;
7217 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7218
7219 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7220 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7221 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7222 stubAddr, stubName, TypePtr::BOTTOM,
7223 src_start, dest_start, k_start, r_start, len);
7224
7225 // return cipher length (int)
7226 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7227 set_result(retvalue);
7228 return true;
7229 }
7230
7231 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7232 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7233 address stubAddr = nullptr;
7234 const char *stubName = nullptr;
7235
7236 assert(UseAES, "need AES instruction support");
7237
7238 switch (id) {
7239 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7240 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7241 stubName = "electronicCodeBook_encryptAESCrypt";
7242 break;
7243 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7244 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7245 stubName = "electronicCodeBook_decryptAESCrypt";
7246 break;
7247 default:
7248 break;
7249 }
7250
7251 if (stubAddr == nullptr) return false;
7252
7253 Node* electronicCodeBook_object = argument(0);
7254 Node* src = argument(1);
7255 Node* src_offset = argument(2);
7256 Node* len = argument(3);
7257 Node* dest = argument(4);
7258 Node* dest_offset = argument(5);
7259
7260 // (1) src and dest are arrays.
7261 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7262 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7263 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7264 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7265
7266 // checks are the responsibility of the caller
7267 Node* src_start = src;
7268 Node* dest_start = dest;
7269 if (src_offset != nullptr || dest_offset != nullptr) {
7270 assert(src_offset != nullptr && dest_offset != nullptr, "");
7271 src_start = array_element_address(src, src_offset, T_BYTE);
7272 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7273 }
7274
7275 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7276 // (because of the predicated logic executed earlier).
7277 // so we cast it here safely.
7278 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7279
7280 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7281 if (embeddedCipherObj == nullptr) return false;
7282
7283 // cast it to what we know it will be at runtime
7284 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7285 assert(tinst != nullptr, "ECB obj is null");
7286 assert(tinst->is_loaded(), "ECB obj is not loaded");
7287 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7288 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7289
7290 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7291 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7292 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7293 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7294 aescrypt_object = _gvn.transform(aescrypt_object);
7295
7296 // we need to get the start of the aescrypt_object's expanded key array
7297 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7298 if (k_start == nullptr) return false;
7299
7300 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7301 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7302 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7303 stubAddr, stubName, TypePtr::BOTTOM,
7304 src_start, dest_start, k_start, len);
7305
7306 // return cipher length (int)
7307 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7308 set_result(retvalue);
7309 return true;
7310 }
7311
7312 //------------------------------inline_counterMode_AESCrypt-----------------------
7313 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7314 assert(UseAES, "need AES instruction support");
7315 if (!UseAESCTRIntrinsics) return false;
7316
7317 address stubAddr = nullptr;
7318 const char *stubName = nullptr;
7319 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7320 stubAddr = StubRoutines::counterMode_AESCrypt();
7321 stubName = "counterMode_AESCrypt";
7322 }
7323 if (stubAddr == nullptr) return false;
7324
7325 Node* counterMode_object = argument(0);
7326 Node* src = argument(1);
7327 Node* src_offset = argument(2);
7328 Node* len = argument(3);
7329 Node* dest = argument(4);
7330 Node* dest_offset = argument(5);
7331
7332 // (1) src and dest are arrays.
7333 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7334 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7335 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7336 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7337
7338 // checks are the responsibility of the caller
7339 Node* src_start = src;
7340 Node* dest_start = dest;
7341 if (src_offset != nullptr || dest_offset != nullptr) {
7342 assert(src_offset != nullptr && dest_offset != nullptr, "");
7343 src_start = array_element_address(src, src_offset, T_BYTE);
7344 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7345 }
7346
7347 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7348 // (because of the predicated logic executed earlier).
7349 // so we cast it here safely.
7350 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7351 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7352 if (embeddedCipherObj == nullptr) return false;
7353 // cast it to what we know it will be at runtime
7354 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7355 assert(tinst != nullptr, "CTR obj is null");
7356 assert(tinst->is_loaded(), "CTR obj is not loaded");
7357 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7358 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7359 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7360 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7361 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7362 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7363 aescrypt_object = _gvn.transform(aescrypt_object);
7364 // we need to get the start of the aescrypt_object's expanded key array
7365 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7366 if (k_start == nullptr) return false;
7367 // similarly, get the start address of the r vector
7368 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7369 if (obj_counter == nullptr) return false;
7370 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7371
7372 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7373 if (saved_encCounter == nullptr) return false;
7374 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7375 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7376
7377 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7378 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7379 OptoRuntime::counterMode_aescrypt_Type(),
7380 stubAddr, stubName, TypePtr::BOTTOM,
7381 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7382
7383 // return cipher length (int)
7384 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7385 set_result(retvalue);
7386 return true;
7387 }
7388
7389 //------------------------------get_key_start_from_aescrypt_object-----------------------
7390 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7391 #if defined(PPC64) || defined(S390) || defined(RISCV64)
7392 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7393 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7394 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7395 // The ppc64 and riscv64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7396 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7397 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7398 if (objSessionK == nullptr) {
7399 return (Node *) nullptr;
7400 }
7401 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7402 #else
7403 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7404 #endif // PPC64
7405 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7406 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7407
7408 // now have the array, need to get the start address of the K array
7409 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7410 return k_start;
7411 }
7412
7413 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7414 // Return node representing slow path of predicate check.
7415 // the pseudo code we want to emulate with this predicate is:
7416 // for encryption:
7417 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7418 // for decryption:
7419 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7420 // note cipher==plain is more conservative than the original java code but that's OK
7421 //
7422 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7423 // The receiver was checked for null already.
7424 Node* objCBC = argument(0);
7425
7426 Node* src = argument(1);
7427 Node* dest = argument(4);
7428
7429 // Load embeddedCipher field of CipherBlockChaining object.
7430 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7431
7432 // get AESCrypt klass for instanceOf check
7433 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7434 // will have same classloader as CipherBlockChaining object
7435 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7436 assert(tinst != nullptr, "CBCobj is null");
7437 assert(tinst->is_loaded(), "CBCobj is not loaded");
7438
7439 // we want to do an instanceof comparison against the AESCrypt class
7440 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7441 if (!klass_AESCrypt->is_loaded()) {
7442 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7443 Node* ctrl = control();
7444 set_control(top()); // no regular fast path
7445 return ctrl;
7446 }
7447
7448 src = must_be_not_null(src, true);
7449 dest = must_be_not_null(dest, true);
7450
7451 // Resolve oops to stable for CmpP below.
7452 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7453
7454 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7455 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7456 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7457
7458 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7459
7460 // for encryption, we are done
7461 if (!decrypting)
7462 return instof_false; // even if it is null
7463
7464 // for decryption, we need to add a further check to avoid
7465 // taking the intrinsic path when cipher and plain are the same
7466 // see the original java code for why.
7467 RegionNode* region = new RegionNode(3);
7468 region->init_req(1, instof_false);
7469
7470 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7471 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7472 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7473 region->init_req(2, src_dest_conjoint);
7474
7475 record_for_igvn(region);
7476 return _gvn.transform(region);
7477 }
7478
7479 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7480 // Return node representing slow path of predicate check.
7481 // the pseudo code we want to emulate with this predicate is:
7482 // for encryption:
7483 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7484 // for decryption:
7485 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7486 // note cipher==plain is more conservative than the original java code but that's OK
7487 //
7488 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7489 // The receiver was checked for null already.
7490 Node* objECB = argument(0);
7491
7492 // Load embeddedCipher field of ElectronicCodeBook object.
7493 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7494
7495 // get AESCrypt klass for instanceOf check
7496 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7497 // will have same classloader as ElectronicCodeBook object
7498 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7499 assert(tinst != nullptr, "ECBobj is null");
7500 assert(tinst->is_loaded(), "ECBobj is not loaded");
7501
7502 // we want to do an instanceof comparison against the AESCrypt class
7503 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7504 if (!klass_AESCrypt->is_loaded()) {
7505 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7506 Node* ctrl = control();
7507 set_control(top()); // no regular fast path
7508 return ctrl;
7509 }
7510 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7511
7512 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7513 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7514 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7515
7516 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7517
7518 // for encryption, we are done
7519 if (!decrypting)
7520 return instof_false; // even if it is null
7521
7522 // for decryption, we need to add a further check to avoid
7523 // taking the intrinsic path when cipher and plain are the same
7524 // see the original java code for why.
7525 RegionNode* region = new RegionNode(3);
7526 region->init_req(1, instof_false);
7527 Node* src = argument(1);
7528 Node* dest = argument(4);
7529 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7530 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7531 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7532 region->init_req(2, src_dest_conjoint);
7533
7534 record_for_igvn(region);
7535 return _gvn.transform(region);
7536 }
7537
7538 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7539 // Return node representing slow path of predicate check.
7540 // the pseudo code we want to emulate with this predicate is:
7541 // for encryption:
7542 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7543 // for decryption:
7544 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7545 // note cipher==plain is more conservative than the original java code but that's OK
7546 //
7547
7548 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7549 // The receiver was checked for null already.
7550 Node* objCTR = argument(0);
7551
7552 // Load embeddedCipher field of CipherBlockChaining object.
7553 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7554
7555 // get AESCrypt klass for instanceOf check
7556 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7557 // will have same classloader as CipherBlockChaining object
7558 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7559 assert(tinst != nullptr, "CTRobj is null");
7560 assert(tinst->is_loaded(), "CTRobj is not loaded");
7561
7562 // we want to do an instanceof comparison against the AESCrypt class
7563 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7564 if (!klass_AESCrypt->is_loaded()) {
7565 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7566 Node* ctrl = control();
7567 set_control(top()); // no regular fast path
7568 return ctrl;
7569 }
7570
7571 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7572 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7573 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7574 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7575 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7576
7577 return instof_false; // even if it is null
7578 }
7579
7580 //------------------------------inline_ghash_processBlocks
7581 bool LibraryCallKit::inline_ghash_processBlocks() {
7582 address stubAddr;
7583 const char *stubName;
7584 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7585
7586 stubAddr = StubRoutines::ghash_processBlocks();
7587 stubName = "ghash_processBlocks";
7588
7589 Node* data = argument(0);
7590 Node* offset = argument(1);
7591 Node* len = argument(2);
7592 Node* state = argument(3);
7593 Node* subkeyH = argument(4);
7594
7595 state = must_be_not_null(state, true);
7596 subkeyH = must_be_not_null(subkeyH, true);
7597 data = must_be_not_null(data, true);
7598
7599 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7600 assert(state_start, "state is null");
7601 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7602 assert(subkeyH_start, "subkeyH is null");
7603 Node* data_start = array_element_address(data, offset, T_BYTE);
7604 assert(data_start, "data is null");
7605
7606 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7607 OptoRuntime::ghash_processBlocks_Type(),
7608 stubAddr, stubName, TypePtr::BOTTOM,
7609 state_start, subkeyH_start, data_start, len);
7610 return true;
7611 }
7612
7613 //------------------------------inline_chacha20Block
7614 bool LibraryCallKit::inline_chacha20Block() {
7615 address stubAddr;
7616 const char *stubName;
7617 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7618
7619 stubAddr = StubRoutines::chacha20Block();
7620 stubName = "chacha20Block";
7621
7622 Node* state = argument(0);
7623 Node* result = argument(1);
7624
7625 state = must_be_not_null(state, true);
7626 result = must_be_not_null(result, true);
7627
7628 Node* state_start = array_element_address(state, intcon(0), T_INT);
7629 assert(state_start, "state is null");
7630 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7631 assert(result_start, "result is null");
7632
7633 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7634 OptoRuntime::chacha20Block_Type(),
7635 stubAddr, stubName, TypePtr::BOTTOM,
7636 state_start, result_start);
7637 // return key stream length (int)
7638 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7639 set_result(retvalue);
7640 return true;
7641 }
7642
7643 //------------------------------inline_dilithiumAlmostNtt
7644 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
7645 address stubAddr;
7646 const char *stubName;
7647 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
7648 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
7649
7650 stubAddr = StubRoutines::dilithiumAlmostNtt();
7651 stubName = "dilithiumAlmostNtt";
7652 if (!stubAddr) return false;
7653
7654 Node* coeffs = argument(0);
7655 Node* ntt_zetas = argument(1);
7656
7657 coeffs = must_be_not_null(coeffs, true);
7658 ntt_zetas = must_be_not_null(ntt_zetas, true);
7659
7660 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
7661 assert(coeffs_start, "coeffs is null");
7662 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
7663 assert(ntt_zetas_start, "ntt_zetas is null");
7664 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7665 OptoRuntime::dilithiumAlmostNtt_Type(),
7666 stubAddr, stubName, TypePtr::BOTTOM,
7667 coeffs_start, ntt_zetas_start);
7668 // return an int
7669 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
7670 set_result(retvalue);
7671 return true;
7672 }
7673
7674 //------------------------------inline_dilithiumAlmostInverseNtt
7675 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
7676 address stubAddr;
7677 const char *stubName;
7678 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
7679 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
7680
7681 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
7682 stubName = "dilithiumAlmostInverseNtt";
7683 if (!stubAddr) return false;
7684
7685 Node* coeffs = argument(0);
7686 Node* zetas = argument(1);
7687
7688 coeffs = must_be_not_null(coeffs, true);
7689 zetas = must_be_not_null(zetas, true);
7690
7691 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
7692 assert(coeffs_start, "coeffs is null");
7693 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
7694 assert(zetas_start, "inverseNtt_zetas is null");
7695 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7696 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
7697 stubAddr, stubName, TypePtr::BOTTOM,
7698 coeffs_start, zetas_start);
7699
7700 // return an int
7701 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
7702 set_result(retvalue);
7703 return true;
7704 }
7705
7706 //------------------------------inline_dilithiumNttMult
7707 bool LibraryCallKit::inline_dilithiumNttMult() {
7708 address stubAddr;
7709 const char *stubName;
7710 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
7711 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
7712
7713 stubAddr = StubRoutines::dilithiumNttMult();
7714 stubName = "dilithiumNttMult";
7715 if (!stubAddr) return false;
7716
7717 Node* result = argument(0);
7718 Node* ntta = argument(1);
7719 Node* nttb = argument(2);
7720
7721 result = must_be_not_null(result, true);
7722 ntta = must_be_not_null(ntta, true);
7723 nttb = must_be_not_null(nttb, true);
7724
7725 Node* result_start = array_element_address(result, intcon(0), T_INT);
7726 assert(result_start, "result is null");
7727 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
7728 assert(ntta_start, "ntta is null");
7729 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
7730 assert(nttb_start, "nttb is null");
7731 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
7732 OptoRuntime::dilithiumNttMult_Type(),
7733 stubAddr, stubName, TypePtr::BOTTOM,
7734 result_start, ntta_start, nttb_start);
7735
7736 // return an int
7737 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
7738 set_result(retvalue);
7739
7740 return true;
7741 }
7742
7743 //------------------------------inline_dilithiumMontMulByConstant
7744 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
7745 address stubAddr;
7746 const char *stubName;
7747 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
7748 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
7749
7750 stubAddr = StubRoutines::dilithiumMontMulByConstant();
7751 stubName = "dilithiumMontMulByConstant";
7752 if (!stubAddr) return false;
7753
7754 Node* coeffs = argument(0);
7755 Node* constant = argument(1);
7756
7757 coeffs = must_be_not_null(coeffs, true);
7758
7759 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
7760 assert(coeffs_start, "coeffs is null");
7761 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
7762 OptoRuntime::dilithiumMontMulByConstant_Type(),
7763 stubAddr, stubName, TypePtr::BOTTOM,
7764 coeffs_start, constant);
7765
7766 // return an int
7767 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
7768 set_result(retvalue);
7769 return true;
7770 }
7771
7772
7773 //------------------------------inline_dilithiumDecomposePoly
7774 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
7775 address stubAddr;
7776 const char *stubName;
7777 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
7778 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
7779
7780 stubAddr = StubRoutines::dilithiumDecomposePoly();
7781 stubName = "dilithiumDecomposePoly";
7782 if (!stubAddr) return false;
7783
7784 Node* input = argument(0);
7785 Node* lowPart = argument(1);
7786 Node* highPart = argument(2);
7787 Node* twoGamma2 = argument(3);
7788 Node* multiplier = argument(4);
7789
7790 input = must_be_not_null(input, true);
7791 lowPart = must_be_not_null(lowPart, true);
7792 highPart = must_be_not_null(highPart, true);
7793
7794 Node* input_start = array_element_address(input, intcon(0), T_INT);
7795 assert(input_start, "input is null");
7796 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
7797 assert(lowPart_start, "lowPart is null");
7798 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
7799 assert(highPart_start, "highPart is null");
7800
7801 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
7802 OptoRuntime::dilithiumDecomposePoly_Type(),
7803 stubAddr, stubName, TypePtr::BOTTOM,
7804 input_start, lowPart_start, highPart_start,
7805 twoGamma2, multiplier);
7806
7807 // return an int
7808 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
7809 set_result(retvalue);
7810 return true;
7811 }
7812
7813 bool LibraryCallKit::inline_base64_encodeBlock() {
7814 address stubAddr;
7815 const char *stubName;
7816 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7817 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
7818 stubAddr = StubRoutines::base64_encodeBlock();
7819 stubName = "encodeBlock";
7820
7821 if (!stubAddr) return false;
7822 Node* base64obj = argument(0);
7823 Node* src = argument(1);
7824 Node* offset = argument(2);
7825 Node* len = argument(3);
7826 Node* dest = argument(4);
7827 Node* dp = argument(5);
7828 Node* isURL = argument(6);
7829
7830 src = must_be_not_null(src, true);
7831 dest = must_be_not_null(dest, true);
7832
7833 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7834 assert(src_start, "source array is null");
7835 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7836 assert(dest_start, "destination array is null");
7837
7838 Node* base64 = make_runtime_call(RC_LEAF,
7839 OptoRuntime::base64_encodeBlock_Type(),
7840 stubAddr, stubName, TypePtr::BOTTOM,
7841 src_start, offset, len, dest_start, dp, isURL);
7842 return true;
7843 }
7844
7845 bool LibraryCallKit::inline_base64_decodeBlock() {
7846 address stubAddr;
7847 const char *stubName;
7848 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7849 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
7850 stubAddr = StubRoutines::base64_decodeBlock();
7851 stubName = "decodeBlock";
7852
7853 if (!stubAddr) return false;
7854 Node* base64obj = argument(0);
7855 Node* src = argument(1);
7856 Node* src_offset = argument(2);
7857 Node* len = argument(3);
7858 Node* dest = argument(4);
7859 Node* dest_offset = argument(5);
7860 Node* isURL = argument(6);
7861 Node* isMIME = argument(7);
7862
7863 src = must_be_not_null(src, true);
7864 dest = must_be_not_null(dest, true);
7865
7866 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7867 assert(src_start, "source array is null");
7868 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7869 assert(dest_start, "destination array is null");
7870
7871 Node* call = make_runtime_call(RC_LEAF,
7872 OptoRuntime::base64_decodeBlock_Type(),
7873 stubAddr, stubName, TypePtr::BOTTOM,
7874 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
7875 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7876 set_result(result);
7877 return true;
7878 }
7879
7880 bool LibraryCallKit::inline_poly1305_processBlocks() {
7881 address stubAddr;
7882 const char *stubName;
7883 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
7884 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
7885 stubAddr = StubRoutines::poly1305_processBlocks();
7886 stubName = "poly1305_processBlocks";
7887
7888 if (!stubAddr) return false;
7889 null_check_receiver(); // null-check receiver
7890 if (stopped()) return true;
7891
7892 Node* input = argument(1);
7893 Node* input_offset = argument(2);
7894 Node* len = argument(3);
7895 Node* alimbs = argument(4);
7896 Node* rlimbs = argument(5);
7897
7898 input = must_be_not_null(input, true);
7899 alimbs = must_be_not_null(alimbs, true);
7900 rlimbs = must_be_not_null(rlimbs, true);
7901
7902 Node* input_start = array_element_address(input, input_offset, T_BYTE);
7903 assert(input_start, "input array is null");
7904 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
7905 assert(acc_start, "acc array is null");
7906 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
7907 assert(r_start, "r array is null");
7908
7909 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7910 OptoRuntime::poly1305_processBlocks_Type(),
7911 stubAddr, stubName, TypePtr::BOTTOM,
7912 input_start, len, acc_start, r_start);
7913 return true;
7914 }
7915
7916 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
7917 address stubAddr;
7918 const char *stubName;
7919 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
7920 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
7921 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
7922 stubName = "intpoly_montgomeryMult_P256";
7923
7924 if (!stubAddr) return false;
7925 null_check_receiver(); // null-check receiver
7926 if (stopped()) return true;
7927
7928 Node* a = argument(1);
7929 Node* b = argument(2);
7930 Node* r = argument(3);
7931
7932 a = must_be_not_null(a, true);
7933 b = must_be_not_null(b, true);
7934 r = must_be_not_null(r, true);
7935
7936 Node* a_start = array_element_address(a, intcon(0), T_LONG);
7937 assert(a_start, "a array is null");
7938 Node* b_start = array_element_address(b, intcon(0), T_LONG);
7939 assert(b_start, "b array is null");
7940 Node* r_start = array_element_address(r, intcon(0), T_LONG);
7941 assert(r_start, "r array is null");
7942
7943 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7944 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
7945 stubAddr, stubName, TypePtr::BOTTOM,
7946 a_start, b_start, r_start);
7947 return true;
7948 }
7949
7950 bool LibraryCallKit::inline_intpoly_assign() {
7951 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
7952 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
7953 const char *stubName = "intpoly_assign";
7954 address stubAddr = StubRoutines::intpoly_assign();
7955 if (!stubAddr) return false;
7956
7957 Node* set = argument(0);
7958 Node* a = argument(1);
7959 Node* b = argument(2);
7960 Node* arr_length = load_array_length(a);
7961
7962 a = must_be_not_null(a, true);
7963 b = must_be_not_null(b, true);
7964
7965 Node* a_start = array_element_address(a, intcon(0), T_LONG);
7966 assert(a_start, "a array is null");
7967 Node* b_start = array_element_address(b, intcon(0), T_LONG);
7968 assert(b_start, "b array is null");
7969
7970 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7971 OptoRuntime::intpoly_assign_Type(),
7972 stubAddr, stubName, TypePtr::BOTTOM,
7973 set, a_start, b_start, arr_length);
7974 return true;
7975 }
7976
7977 //------------------------------inline_digestBase_implCompress-----------------------
7978 //
7979 // Calculate MD5 for single-block byte[] array.
7980 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
7981 //
7982 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
7983 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
7984 //
7985 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
7986 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
7987 //
7988 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
7989 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
7990 //
7991 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
7992 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
7993 //
7994 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
7995 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
7996
7997 Node* digestBase_obj = argument(0);
7998 Node* src = argument(1); // type oop
7999 Node* ofs = argument(2); // type int
8000
8001 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8002 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8003 // failed array check
8004 return false;
8005 }
8006 // Figure out the size and type of the elements we will be copying.
8007 BasicType src_elem = src_type->elem()->array_element_basic_type();
8008 if (src_elem != T_BYTE) {
8009 return false;
8010 }
8011 // 'src_start' points to src array + offset
8012 src = must_be_not_null(src, true);
8013 Node* src_start = array_element_address(src, ofs, src_elem);
8014 Node* state = nullptr;
8015 Node* block_size = nullptr;
8016 address stubAddr;
8017 const char *stubName;
8018
8019 switch(id) {
8020 case vmIntrinsics::_md5_implCompress:
8021 assert(UseMD5Intrinsics, "need MD5 instruction support");
8022 state = get_state_from_digest_object(digestBase_obj, T_INT);
8023 stubAddr = StubRoutines::md5_implCompress();
8024 stubName = "md5_implCompress";
8025 break;
8026 case vmIntrinsics::_sha_implCompress:
8027 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
8028 state = get_state_from_digest_object(digestBase_obj, T_INT);
8029 stubAddr = StubRoutines::sha1_implCompress();
8030 stubName = "sha1_implCompress";
8031 break;
8032 case vmIntrinsics::_sha2_implCompress:
8033 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
8034 state = get_state_from_digest_object(digestBase_obj, T_INT);
8035 stubAddr = StubRoutines::sha256_implCompress();
8036 stubName = "sha256_implCompress";
8037 break;
8038 case vmIntrinsics::_sha5_implCompress:
8039 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
8040 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8041 stubAddr = StubRoutines::sha512_implCompress();
8042 stubName = "sha512_implCompress";
8043 break;
8044 case vmIntrinsics::_sha3_implCompress:
8045 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
8046 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8047 stubAddr = StubRoutines::sha3_implCompress();
8048 stubName = "sha3_implCompress";
8049 block_size = get_block_size_from_digest_object(digestBase_obj);
8050 if (block_size == nullptr) return false;
8051 break;
8052 default:
8053 fatal_unexpected_iid(id);
8054 return false;
8055 }
8056 if (state == nullptr) return false;
8057
8058 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
8059 if (stubAddr == nullptr) return false;
8060
8061 // Call the stub.
8062 Node* call;
8063 if (block_size == nullptr) {
8064 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
8065 stubAddr, stubName, TypePtr::BOTTOM,
8066 src_start, state);
8067 } else {
8068 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
8069 stubAddr, stubName, TypePtr::BOTTOM,
8070 src_start, state, block_size);
8071 }
8072
8073 return true;
8074 }
8075
8076 //------------------------------inline_double_keccak
8077 bool LibraryCallKit::inline_double_keccak() {
8078 address stubAddr;
8079 const char *stubName;
8080 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
8081 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
8082
8083 stubAddr = StubRoutines::double_keccak();
8084 stubName = "double_keccak";
8085 if (!stubAddr) return false;
8086
8087 Node* status0 = argument(0);
8088 Node* status1 = argument(1);
8089
8090 status0 = must_be_not_null(status0, true);
8091 status1 = must_be_not_null(status1, true);
8092
8093 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
8094 assert(status0_start, "status0 is null");
8095 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
8096 assert(status1_start, "status1 is null");
8097 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
8098 OptoRuntime::double_keccak_Type(),
8099 stubAddr, stubName, TypePtr::BOTTOM,
8100 status0_start, status1_start);
8101 // return an int
8102 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
8103 set_result(retvalue);
8104 return true;
8105 }
8106
8107
8108 //------------------------------inline_digestBase_implCompressMB-----------------------
8109 //
8110 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
8111 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
8112 //
8113 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
8114 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8115 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8116 assert((uint)predicate < 5, "sanity");
8117 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
8118
8119 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
8120 Node* src = argument(1); // byte[] array
8121 Node* ofs = argument(2); // type int
8122 Node* limit = argument(3); // type int
8123
8124 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8125 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8126 // failed array check
8127 return false;
8128 }
8129 // Figure out the size and type of the elements we will be copying.
8130 BasicType src_elem = src_type->elem()->array_element_basic_type();
8131 if (src_elem != T_BYTE) {
8132 return false;
8133 }
8134 // 'src_start' points to src array + offset
8135 src = must_be_not_null(src, false);
8136 Node* src_start = array_element_address(src, ofs, src_elem);
8137
8138 const char* klass_digestBase_name = nullptr;
8139 const char* stub_name = nullptr;
8140 address stub_addr = nullptr;
8141 BasicType elem_type = T_INT;
8142
8143 switch (predicate) {
8144 case 0:
8145 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
8146 klass_digestBase_name = "sun/security/provider/MD5";
8147 stub_name = "md5_implCompressMB";
8148 stub_addr = StubRoutines::md5_implCompressMB();
8149 }
8150 break;
8151 case 1:
8152 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
8153 klass_digestBase_name = "sun/security/provider/SHA";
8154 stub_name = "sha1_implCompressMB";
8155 stub_addr = StubRoutines::sha1_implCompressMB();
8156 }
8157 break;
8158 case 2:
8159 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
8160 klass_digestBase_name = "sun/security/provider/SHA2";
8161 stub_name = "sha256_implCompressMB";
8162 stub_addr = StubRoutines::sha256_implCompressMB();
8163 }
8164 break;
8165 case 3:
8166 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
8167 klass_digestBase_name = "sun/security/provider/SHA5";
8168 stub_name = "sha512_implCompressMB";
8169 stub_addr = StubRoutines::sha512_implCompressMB();
8170 elem_type = T_LONG;
8171 }
8172 break;
8173 case 4:
8174 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
8175 klass_digestBase_name = "sun/security/provider/SHA3";
8176 stub_name = "sha3_implCompressMB";
8177 stub_addr = StubRoutines::sha3_implCompressMB();
8178 elem_type = T_LONG;
8179 }
8180 break;
8181 default:
8182 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
8183 }
8184 if (klass_digestBase_name != nullptr) {
8185 assert(stub_addr != nullptr, "Stub is generated");
8186 if (stub_addr == nullptr) return false;
8187
8188 // get DigestBase klass to lookup for SHA klass
8189 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
8190 assert(tinst != nullptr, "digestBase_obj is not instance???");
8191 assert(tinst->is_loaded(), "DigestBase is not loaded");
8192
8193 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
8194 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
8195 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
8196 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
8197 }
8198 return false;
8199 }
8200
8201 //------------------------------inline_digestBase_implCompressMB-----------------------
8202 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
8203 BasicType elem_type, address stubAddr, const char *stubName,
8204 Node* src_start, Node* ofs, Node* limit) {
8205 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
8206 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8207 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
8208 digest_obj = _gvn.transform(digest_obj);
8209
8210 Node* state = get_state_from_digest_object(digest_obj, elem_type);
8211 if (state == nullptr) return false;
8212
8213 Node* block_size = nullptr;
8214 if (strcmp("sha3_implCompressMB", stubName) == 0) {
8215 block_size = get_block_size_from_digest_object(digest_obj);
8216 if (block_size == nullptr) return false;
8217 }
8218
8219 // Call the stub.
8220 Node* call;
8221 if (block_size == nullptr) {
8222 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8223 OptoRuntime::digestBase_implCompressMB_Type(false),
8224 stubAddr, stubName, TypePtr::BOTTOM,
8225 src_start, state, ofs, limit);
8226 } else {
8227 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8228 OptoRuntime::digestBase_implCompressMB_Type(true),
8229 stubAddr, stubName, TypePtr::BOTTOM,
8230 src_start, state, block_size, ofs, limit);
8231 }
8232
8233 // return ofs (int)
8234 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8235 set_result(result);
8236
8237 return true;
8238 }
8239
8240 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8241 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8242 assert(UseAES, "need AES instruction support");
8243 address stubAddr = nullptr;
8244 const char *stubName = nullptr;
8245 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8246 stubName = "galoisCounterMode_AESCrypt";
8247
8248 if (stubAddr == nullptr) return false;
8249
8250 Node* in = argument(0);
8251 Node* inOfs = argument(1);
8252 Node* len = argument(2);
8253 Node* ct = argument(3);
8254 Node* ctOfs = argument(4);
8255 Node* out = argument(5);
8256 Node* outOfs = argument(6);
8257 Node* gctr_object = argument(7);
8258 Node* ghash_object = argument(8);
8259
8260 // (1) in, ct and out are arrays.
8261 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8262 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8263 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8264 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8265 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8266 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8267
8268 // checks are the responsibility of the caller
8269 Node* in_start = in;
8270 Node* ct_start = ct;
8271 Node* out_start = out;
8272 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8273 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8274 in_start = array_element_address(in, inOfs, T_BYTE);
8275 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8276 out_start = array_element_address(out, outOfs, T_BYTE);
8277 }
8278
8279 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8280 // (because of the predicated logic executed earlier).
8281 // so we cast it here safely.
8282 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8283 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8284 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8285 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8286 Node* state = load_field_from_object(ghash_object, "state", "[J");
8287
8288 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8289 return false;
8290 }
8291 // cast it to what we know it will be at runtime
8292 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8293 assert(tinst != nullptr, "GCTR obj is null");
8294 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8295 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8296 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8297 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8298 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8299 const TypeOopPtr* xtype = aklass->as_instance_type();
8300 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8301 aescrypt_object = _gvn.transform(aescrypt_object);
8302 // we need to get the start of the aescrypt_object's expanded key array
8303 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8304 if (k_start == nullptr) return false;
8305 // similarly, get the start address of the r vector
8306 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8307 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8308 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8309
8310
8311 // Call the stub, passing params
8312 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8313 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8314 stubAddr, stubName, TypePtr::BOTTOM,
8315 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8316
8317 // return cipher length (int)
8318 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8319 set_result(retvalue);
8320
8321 return true;
8322 }
8323
8324 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8325 // Return node representing slow path of predicate check.
8326 // the pseudo code we want to emulate with this predicate is:
8327 // for encryption:
8328 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8329 // for decryption:
8330 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8331 // note cipher==plain is more conservative than the original java code but that's OK
8332 //
8333
8334 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8335 // The receiver was checked for null already.
8336 Node* objGCTR = argument(7);
8337 // Load embeddedCipher field of GCTR object.
8338 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8339 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8340
8341 // get AESCrypt klass for instanceOf check
8342 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8343 // will have same classloader as CipherBlockChaining object
8344 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8345 assert(tinst != nullptr, "GCTR obj is null");
8346 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8347
8348 // we want to do an instanceof comparison against the AESCrypt class
8349 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8350 if (!klass_AESCrypt->is_loaded()) {
8351 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8352 Node* ctrl = control();
8353 set_control(top()); // no regular fast path
8354 return ctrl;
8355 }
8356
8357 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8358 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8359 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8360 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8361 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8362
8363 return instof_false; // even if it is null
8364 }
8365
8366 //------------------------------get_state_from_digest_object-----------------------
8367 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8368 const char* state_type;
8369 switch (elem_type) {
8370 case T_BYTE: state_type = "[B"; break;
8371 case T_INT: state_type = "[I"; break;
8372 case T_LONG: state_type = "[J"; break;
8373 default: ShouldNotReachHere();
8374 }
8375 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8376 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8377 if (digest_state == nullptr) return (Node *) nullptr;
8378
8379 // now have the array, need to get the start address of the state array
8380 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8381 return state;
8382 }
8383
8384 //------------------------------get_block_size_from_sha3_object----------------------------------
8385 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8386 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8387 assert (block_size != nullptr, "sanity");
8388 return block_size;
8389 }
8390
8391 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8392 // Return node representing slow path of predicate check.
8393 // the pseudo code we want to emulate with this predicate is:
8394 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8395 //
8396 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8397 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8398 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8399 assert((uint)predicate < 5, "sanity");
8400
8401 // The receiver was checked for null already.
8402 Node* digestBaseObj = argument(0);
8403
8404 // get DigestBase klass for instanceOf check
8405 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8406 assert(tinst != nullptr, "digestBaseObj is null");
8407 assert(tinst->is_loaded(), "DigestBase is not loaded");
8408
8409 const char* klass_name = nullptr;
8410 switch (predicate) {
8411 case 0:
8412 if (UseMD5Intrinsics) {
8413 // we want to do an instanceof comparison against the MD5 class
8414 klass_name = "sun/security/provider/MD5";
8415 }
8416 break;
8417 case 1:
8418 if (UseSHA1Intrinsics) {
8419 // we want to do an instanceof comparison against the SHA class
8420 klass_name = "sun/security/provider/SHA";
8421 }
8422 break;
8423 case 2:
8424 if (UseSHA256Intrinsics) {
8425 // we want to do an instanceof comparison against the SHA2 class
8426 klass_name = "sun/security/provider/SHA2";
8427 }
8428 break;
8429 case 3:
8430 if (UseSHA512Intrinsics) {
8431 // we want to do an instanceof comparison against the SHA5 class
8432 klass_name = "sun/security/provider/SHA5";
8433 }
8434 break;
8435 case 4:
8436 if (UseSHA3Intrinsics) {
8437 // we want to do an instanceof comparison against the SHA3 class
8438 klass_name = "sun/security/provider/SHA3";
8439 }
8440 break;
8441 default:
8442 fatal("unknown SHA intrinsic predicate: %d", predicate);
8443 }
8444
8445 ciKlass* klass = nullptr;
8446 if (klass_name != nullptr) {
8447 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8448 }
8449 if ((klass == nullptr) || !klass->is_loaded()) {
8450 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8451 Node* ctrl = control();
8452 set_control(top()); // no intrinsic path
8453 return ctrl;
8454 }
8455 ciInstanceKlass* instklass = klass->as_instance_klass();
8456
8457 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8458 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8459 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8460 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8461
8462 return instof_false; // even if it is null
8463 }
8464
8465 //-------------inline_fma-----------------------------------
8466 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8467 Node *a = nullptr;
8468 Node *b = nullptr;
8469 Node *c = nullptr;
8470 Node* result = nullptr;
8471 switch (id) {
8472 case vmIntrinsics::_fmaD:
8473 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8474 // no receiver since it is static method
8475 a = argument(0);
8476 b = argument(2);
8477 c = argument(4);
8478 result = _gvn.transform(new FmaDNode(a, b, c));
8479 break;
8480 case vmIntrinsics::_fmaF:
8481 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8482 a = argument(0);
8483 b = argument(1);
8484 c = argument(2);
8485 result = _gvn.transform(new FmaFNode(a, b, c));
8486 break;
8487 default:
8488 fatal_unexpected_iid(id); break;
8489 }
8490 set_result(result);
8491 return true;
8492 }
8493
8494 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8495 // argument(0) is receiver
8496 Node* codePoint = argument(1);
8497 Node* n = nullptr;
8498
8499 switch (id) {
8500 case vmIntrinsics::_isDigit :
8501 n = new DigitNode(control(), codePoint);
8502 break;
8503 case vmIntrinsics::_isLowerCase :
8504 n = new LowerCaseNode(control(), codePoint);
8505 break;
8506 case vmIntrinsics::_isUpperCase :
8507 n = new UpperCaseNode(control(), codePoint);
8508 break;
8509 case vmIntrinsics::_isWhitespace :
8510 n = new WhitespaceNode(control(), codePoint);
8511 break;
8512 default:
8513 fatal_unexpected_iid(id);
8514 }
8515
8516 set_result(_gvn.transform(n));
8517 return true;
8518 }
8519
8520 bool LibraryCallKit::inline_profileBoolean() {
8521 Node* counts = argument(1);
8522 const TypeAryPtr* ary = nullptr;
8523 ciArray* aobj = nullptr;
8524 if (counts->is_Con()
8525 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8526 && (aobj = ary->const_oop()->as_array()) != nullptr
8527 && (aobj->length() == 2)) {
8528 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8529 jint false_cnt = aobj->element_value(0).as_int();
8530 jint true_cnt = aobj->element_value(1).as_int();
8531
8532 if (C->log() != nullptr) {
8533 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8534 false_cnt, true_cnt);
8535 }
8536
8537 if (false_cnt + true_cnt == 0) {
8538 // According to profile, never executed.
8539 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8540 Deoptimization::Action_reinterpret);
8541 return true;
8542 }
8543
8544 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8545 // is a number of each value occurrences.
8546 Node* result = argument(0);
8547 if (false_cnt == 0 || true_cnt == 0) {
8548 // According to profile, one value has been never seen.
8549 int expected_val = (false_cnt == 0) ? 1 : 0;
8550
8551 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8552 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8553
8554 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8555 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8556 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8557
8558 { // Slow path: uncommon trap for never seen value and then reexecute
8559 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8560 // the value has been seen at least once.
8561 PreserveJVMState pjvms(this);
8562 PreserveReexecuteState preexecs(this);
8563 jvms()->set_should_reexecute(true);
8564
8565 set_control(slow_path);
8566 set_i_o(i_o());
8567
8568 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8569 Deoptimization::Action_reinterpret);
8570 }
8571 // The guard for never seen value enables sharpening of the result and
8572 // returning a constant. It allows to eliminate branches on the same value
8573 // later on.
8574 set_control(fast_path);
8575 result = intcon(expected_val);
8576 }
8577 // Stop profiling.
8578 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8579 // By replacing method body with profile data (represented as ProfileBooleanNode
8580 // on IR level) we effectively disable profiling.
8581 // It enables full speed execution once optimized code is generated.
8582 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8583 C->record_for_igvn(profile);
8584 set_result(profile);
8585 return true;
8586 } else {
8587 // Continue profiling.
8588 // Profile data isn't available at the moment. So, execute method's bytecode version.
8589 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8590 // is compiled and counters aren't available since corresponding MethodHandle
8591 // isn't a compile-time constant.
8592 return false;
8593 }
8594 }
8595
8596 bool LibraryCallKit::inline_isCompileConstant() {
8597 Node* n = argument(0);
8598 set_result(n->is_Con() ? intcon(1) : intcon(0));
8599 return true;
8600 }
8601
8602 //------------------------------- inline_getObjectSize --------------------------------------
8603 //
8604 // Calculate the runtime size of the object/array.
8605 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8606 //
8607 bool LibraryCallKit::inline_getObjectSize() {
8608 Node* obj = argument(3);
8609 Node* klass_node = load_object_klass(obj);
8610
8611 jint layout_con = Klass::_lh_neutral_value;
8612 Node* layout_val = get_layout_helper(klass_node, layout_con);
8613 int layout_is_con = (layout_val == nullptr);
8614
8615 if (layout_is_con) {
8616 // Layout helper is constant, can figure out things at compile time.
8617
8618 if (Klass::layout_helper_is_instance(layout_con)) {
8619 // Instance case: layout_con contains the size itself.
8620 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8621 set_result(size);
8622 } else {
8623 // Array case: size is round(header + element_size*arraylength).
8624 // Since arraylength is different for every array instance, we have to
8625 // compute the whole thing at runtime.
8626
8627 Node* arr_length = load_array_length(obj);
8628
8629 int round_mask = MinObjAlignmentInBytes - 1;
8630 int hsize = Klass::layout_helper_header_size(layout_con);
8631 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8632
8633 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8634 round_mask = 0; // strength-reduce it if it goes away completely
8635 }
8636 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8637 Node* header_size = intcon(hsize + round_mask);
8638
8639 Node* lengthx = ConvI2X(arr_length);
8640 Node* headerx = ConvI2X(header_size);
8641
8642 Node* abody = lengthx;
8643 if (eshift != 0) {
8644 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8645 }
8646 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8647 if (round_mask != 0) {
8648 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8649 }
8650 size = ConvX2L(size);
8651 set_result(size);
8652 }
8653 } else {
8654 // Layout helper is not constant, need to test for array-ness at runtime.
8655
8656 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8657 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8658 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8659 record_for_igvn(result_reg);
8660
8661 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
8662 if (array_ctl != nullptr) {
8663 // Array case: size is round(header + element_size*arraylength).
8664 // Since arraylength is different for every array instance, we have to
8665 // compute the whole thing at runtime.
8666
8667 PreserveJVMState pjvms(this);
8668 set_control(array_ctl);
8669 Node* arr_length = load_array_length(obj);
8670
8671 int round_mask = MinObjAlignmentInBytes - 1;
8672 Node* mask = intcon(round_mask);
8673
8674 Node* hss = intcon(Klass::_lh_header_size_shift);
8675 Node* hsm = intcon(Klass::_lh_header_size_mask);
8676 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8677 header_size = _gvn.transform(new AndINode(header_size, hsm));
8678 header_size = _gvn.transform(new AddINode(header_size, mask));
8679
8680 // There is no need to mask or shift this value.
8681 // The semantics of LShiftINode include an implicit mask to 0x1F.
8682 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8683 Node* elem_shift = layout_val;
8684
8685 Node* lengthx = ConvI2X(arr_length);
8686 Node* headerx = ConvI2X(header_size);
8687
8688 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
8689 Node* size = _gvn.transform(new AddXNode(headerx, abody));
8690 if (round_mask != 0) {
8691 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
8692 }
8693 size = ConvX2L(size);
8694
8695 result_reg->init_req(_array_path, control());
8696 result_val->init_req(_array_path, size);
8697 }
8698
8699 if (!stopped()) {
8700 // Instance case: the layout helper gives us instance size almost directly,
8701 // but we need to mask out the _lh_instance_slow_path_bit.
8702 Node* size = ConvI2X(layout_val);
8703 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
8704 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
8705 size = _gvn.transform(new AndXNode(size, mask));
8706 size = ConvX2L(size);
8707
8708 result_reg->init_req(_instance_path, control());
8709 result_val->init_req(_instance_path, size);
8710 }
8711
8712 set_result(result_reg, result_val);
8713 }
8714
8715 return true;
8716 }
8717
8718 //------------------------------- inline_blackhole --------------------------------------
8719 //
8720 // Make sure all arguments to this node are alive.
8721 // This matches methods that were requested to be blackholed through compile commands.
8722 //
8723 bool LibraryCallKit::inline_blackhole() {
8724 assert(callee()->is_static(), "Should have been checked before: only static methods here");
8725 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
8726 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
8727
8728 // Blackhole node pinches only the control, not memory. This allows
8729 // the blackhole to be pinned in the loop that computes blackholed
8730 // values, but have no other side effects, like breaking the optimizations
8731 // across the blackhole.
8732
8733 Node* bh = _gvn.transform(new BlackholeNode(control()));
8734 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
8735
8736 // Bind call arguments as blackhole arguments to keep them alive
8737 uint nargs = callee()->arg_size();
8738 for (uint i = 0; i < nargs; i++) {
8739 bh->add_req(argument(i));
8740 }
8741
8742 return true;
8743 }
8744
8745 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
8746 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
8747 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
8748 return nullptr; // box klass is not Float16
8749 }
8750
8751 // Null check; get notnull casted pointer
8752 Node* null_ctl = top();
8753 Node* not_null_box = null_check_oop(box, &null_ctl, true);
8754 // If not_null_box is dead, only null-path is taken
8755 if (stopped()) {
8756 set_control(null_ctl);
8757 return nullptr;
8758 }
8759 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
8760 const TypePtr* adr_type = C->alias_type(field)->adr_type();
8761 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
8762 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
8763 }
8764
8765 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
8766 PreserveReexecuteState preexecs(this);
8767 jvms()->set_should_reexecute(true);
8768
8769 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
8770 Node* klass_node = makecon(klass_type);
8771 Node* box = new_instance(klass_node);
8772
8773 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
8774 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
8775
8776 Node* field_store = _gvn.transform(access_store_at(box,
8777 value_field,
8778 value_adr_type,
8779 value,
8780 TypeInt::SHORT,
8781 T_SHORT,
8782 IN_HEAP));
8783 set_memory(field_store, value_adr_type);
8784 return box;
8785 }
8786
8787 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
8788 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
8789 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
8790 return false;
8791 }
8792
8793 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
8794 if (box_type == nullptr || box_type->const_oop() == nullptr) {
8795 return false;
8796 }
8797
8798 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
8799 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
8800 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
8801 ciSymbols::short_signature(),
8802 false);
8803 assert(field != nullptr, "");
8804
8805 // Transformed nodes
8806 Node* fld1 = nullptr;
8807 Node* fld2 = nullptr;
8808 Node* fld3 = nullptr;
8809 switch(num_args) {
8810 case 3:
8811 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
8812 if (fld3 == nullptr) {
8813 return false;
8814 }
8815 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
8816 // fall-through
8817 case 2:
8818 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
8819 if (fld2 == nullptr) {
8820 return false;
8821 }
8822 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
8823 // fall-through
8824 case 1:
8825 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
8826 if (fld1 == nullptr) {
8827 return false;
8828 }
8829 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
8830 break;
8831 default: fatal("Unsupported number of arguments %d", num_args);
8832 }
8833
8834 Node* result = nullptr;
8835 switch (id) {
8836 // Unary operations
8837 case vmIntrinsics::_sqrt_float16:
8838 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
8839 break;
8840 // Ternary operations
8841 case vmIntrinsics::_fma_float16:
8842 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
8843 break;
8844 default:
8845 fatal_unexpected_iid(id);
8846 break;
8847 }
8848 result = _gvn.transform(new ReinterpretHF2SNode(result));
8849 set_result(box_fp16_value(float16_box_type, field, result));
8850 return true;
8851 }
8852