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 "classfile/vmIntrinsics.hpp"
28 #include "compiler/compileBroker.hpp"
29 #include "compiler/compileLog.hpp"
30 #include "gc/shared/barrierSet.hpp"
31 #include "jfr/support/jfrIntrinsics.hpp"
32 #include "memory/resourceArea.hpp"
33 #include "oops/klass.inline.hpp"
34 #include "oops/objArrayKlass.hpp"
35 #include "opto/addnode.hpp"
36 #include "opto/arraycopynode.hpp"
37 #include "opto/c2compiler.hpp"
38 #include "opto/castnode.hpp"
39 #include "opto/cfgnode.hpp"
40 #include "opto/convertnode.hpp"
41 #include "opto/countbitsnode.hpp"
42 #include "opto/idealKit.hpp"
43 #include "opto/library_call.hpp"
44 #include "opto/mathexactnode.hpp"
45 #include "opto/mulnode.hpp"
46 #include "opto/narrowptrnode.hpp"
47 #include "opto/opaquenode.hpp"
48 #include "opto/parse.hpp"
49 #include "opto/runtime.hpp"
50 #include "opto/rootnode.hpp"
51 #include "opto/subnode.hpp"
52 #include "opto/vectornode.hpp"
53 #include "prims/jvmtiExport.hpp"
54 #include "prims/jvmtiThreadState.hpp"
55 #include "prims/unsafe.hpp"
56 #include "runtime/jniHandles.inline.hpp"
57 #include "runtime/objectMonitor.hpp"
58 #include "runtime/sharedRuntime.hpp"
59 #include "runtime/stubRoutines.hpp"
60 #include "utilities/macros.hpp"
61 #include "utilities/powerOfTwo.hpp"
62
63 //---------------------------make_vm_intrinsic----------------------------
64 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
65 vmIntrinsicID id = m->intrinsic_id();
66 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
67
68 if (!m->is_loaded()) {
69 // Do not attempt to inline unloaded methods.
70 return nullptr;
71 }
72
73 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
74 bool is_available = false;
75
76 {
77 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
78 // the compiler must transition to '_thread_in_vm' state because both
79 // methods access VM-internal data.
80 VM_ENTRY_MARK;
81 methodHandle mh(THREAD, m->get_Method());
82 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
83 if (is_available && is_virtual) {
84 is_available = vmIntrinsics::does_virtual_dispatch(id);
85 }
86 }
87
88 if (is_available) {
89 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
90 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
91 return new LibraryIntrinsic(m, is_virtual,
92 vmIntrinsics::predicates_needed(id),
93 vmIntrinsics::does_virtual_dispatch(id),
94 id);
95 } else {
96 return nullptr;
97 }
98 }
99
100 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
101 LibraryCallKit kit(jvms, this);
102 Compile* C = kit.C;
103 int nodes = C->unique();
104 #ifndef PRODUCT
105 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
106 char buf[1000];
107 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
108 tty->print_cr("Intrinsic %s", str);
109 }
110 #endif
111 ciMethod* callee = kit.callee();
112 const int bci = kit.bci();
113 #ifdef ASSERT
114 Node* ctrl = kit.control();
115 #endif
116 // Try to inline the intrinsic.
117 if (callee->check_intrinsic_candidate() &&
118 kit.try_to_inline(_last_predicate)) {
119 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
120 : "(intrinsic)";
121 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
122 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
123 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
124 if (C->log()) {
125 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
126 vmIntrinsics::name_at(intrinsic_id()),
127 (is_virtual() ? " virtual='1'" : ""),
128 C->unique() - nodes);
129 }
130 // Push the result from the inlined method onto the stack.
131 kit.push_result();
132 return kit.transfer_exceptions_into_jvms();
133 }
134
135 // The intrinsic bailed out
136 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
137 if (jvms->has_method()) {
138 // Not a root compile.
139 const char* msg;
140 if (callee->intrinsic_candidate()) {
141 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
142 } else {
143 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
144 : "failed to inline (intrinsic), method not annotated";
145 }
146 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
147 C->inline_printer()->record(callee, jvms, InliningResult::FAILURE, msg);
148 } else {
149 // Root compile
150 ResourceMark rm;
151 stringStream msg_stream;
152 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
153 vmIntrinsics::name_at(intrinsic_id()),
154 is_virtual() ? " (virtual)" : "", bci);
155 const char *msg = msg_stream.freeze();
156 log_debug(jit, inlining)("%s", msg);
157 if (C->print_intrinsics() || C->print_inlining()) {
158 tty->print("%s", msg);
159 }
160 }
161 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
162
163 return nullptr;
164 }
165
166 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
167 LibraryCallKit kit(jvms, this);
168 Compile* C = kit.C;
169 int nodes = C->unique();
170 _last_predicate = predicate;
171 #ifndef PRODUCT
172 assert(is_predicated() && predicate < predicates_count(), "sanity");
173 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
174 char buf[1000];
175 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
176 tty->print_cr("Predicate for intrinsic %s", str);
177 }
178 #endif
179 ciMethod* callee = kit.callee();
180 const int bci = kit.bci();
181
182 Node* slow_ctl = kit.try_to_predicate(predicate);
183 if (!kit.failing()) {
184 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
185 : "(intrinsic, predicate)";
186 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
187 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
188
189 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
190 if (C->log()) {
191 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
192 vmIntrinsics::name_at(intrinsic_id()),
193 (is_virtual() ? " virtual='1'" : ""),
194 C->unique() - nodes);
195 }
196 return slow_ctl; // Could be null if the check folds.
197 }
198
199 // The intrinsic bailed out
200 if (jvms->has_method()) {
201 // Not a root compile.
202 const char* msg = "failed to generate predicate for intrinsic";
203 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
204 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
205 } else {
206 // Root compile
207 ResourceMark rm;
208 stringStream msg_stream;
209 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
210 vmIntrinsics::name_at(intrinsic_id()),
211 is_virtual() ? " (virtual)" : "", bci);
212 const char *msg = msg_stream.freeze();
213 log_debug(jit, inlining)("%s", msg);
214 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
215 }
216 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
217 return nullptr;
218 }
219
220 bool LibraryCallKit::try_to_inline(int predicate) {
221 // Handle symbolic names for otherwise undistinguished boolean switches:
222 const bool is_store = true;
223 const bool is_compress = true;
224 const bool is_static = true;
225 const bool is_volatile = true;
226
227 if (!jvms()->has_method()) {
228 // Root JVMState has a null method.
229 assert(map()->memory()->Opcode() == Op_Parm, "");
230 // Insert the memory aliasing node
231 set_all_memory(reset_memory());
232 }
233 assert(merged_memory(), "");
234
235 switch (intrinsic_id()) {
236 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
237 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
238 case vmIntrinsics::_getClass: return inline_native_getClass();
239
240 case vmIntrinsics::_ceil:
241 case vmIntrinsics::_floor:
242 case vmIntrinsics::_rint:
243 case vmIntrinsics::_dsin:
244 case vmIntrinsics::_dcos:
245 case vmIntrinsics::_dtan:
246 case vmIntrinsics::_dtanh:
247 case vmIntrinsics::_dabs:
248 case vmIntrinsics::_fabs:
249 case vmIntrinsics::_iabs:
250 case vmIntrinsics::_labs:
251 case vmIntrinsics::_datan2:
252 case vmIntrinsics::_dsqrt:
253 case vmIntrinsics::_dsqrt_strict:
254 case vmIntrinsics::_dexp:
255 case vmIntrinsics::_dlog:
256 case vmIntrinsics::_dlog10:
257 case vmIntrinsics::_dpow:
258 case vmIntrinsics::_dcopySign:
259 case vmIntrinsics::_fcopySign:
260 case vmIntrinsics::_dsignum:
261 case vmIntrinsics::_roundF:
262 case vmIntrinsics::_roundD:
263 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
264
265 case vmIntrinsics::_notify:
266 case vmIntrinsics::_notifyAll:
267 return inline_notify(intrinsic_id());
268
269 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
270 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
271 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
272 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
273 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
274 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
275 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
276 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
277 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
278 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
279 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
280 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
281 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
282 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
283
284 case vmIntrinsics::_arraycopy: return inline_arraycopy();
285
286 case vmIntrinsics::_arraySort: return inline_array_sort();
287 case vmIntrinsics::_arrayPartition: return inline_array_partition();
288
289 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
290 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
291 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
292 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
293
294 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
295 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
296 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
297 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
298 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
299 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
300 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
301 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
302
303 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
304
305 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
306
307 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
308 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
309 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
310 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
311
312 case vmIntrinsics::_compressStringC:
313 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
314 case vmIntrinsics::_inflateStringC:
315 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
316
317 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
318 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
319 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
320 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
321 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
322 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
323 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
324 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
325 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
326
327 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
328 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
329 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
330 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
331 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
332 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
333 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
334 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
335 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
336
337 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
338 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
339 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
340 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
341 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
342 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
343 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
344 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
345 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
346
347 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
348 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
349 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
350 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
351 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
352 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
353 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
354 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
355 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
356
357 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
358 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
359 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
360 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
361
362 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
363 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
364 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
365 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
366
367 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
368 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
369 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
370 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
371 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
372 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
373 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
374 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
375 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
376
377 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
378 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
379 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
380 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
381 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
382 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
383 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
384 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
385 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
386
387 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
388 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
389 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
390 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
391 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
392 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
393 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
394 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
395 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
396
397 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
398 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
399 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
400 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
401 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
402 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
403 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
404 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
405 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
406
407 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
408 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
409 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
410 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
411 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
412
413 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
414 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
415 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
416 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
417 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
418 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
419 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
420 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
421 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
422 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
423 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
424 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
425 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
426 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
427 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
428 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
429 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
430 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
431 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
432 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
433
434 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
435 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
436 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
437 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
438 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
439 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
440 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
441 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
442 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
443 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
444 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
445 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
446 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
447 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
448 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
449
450 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
451 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
452 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
453 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
454
455 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
456 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
457 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
458 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
459 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
460
461 case vmIntrinsics::_loadFence:
462 case vmIntrinsics::_storeFence:
463 case vmIntrinsics::_storeStoreFence:
464 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
465
466 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
467
468 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
469 case vmIntrinsics::_currentThread: return inline_native_currentThread();
470 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
471
472 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
473 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
474
475 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
476 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
477
478 #if INCLUDE_JVMTI
479 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
480 "notifyJvmtiStart", true, false);
481 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
482 "notifyJvmtiEnd", false, true);
483 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
484 "notifyJvmtiMount", false, false);
485 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
486 "notifyJvmtiUnmount", false, false);
487 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
488 #endif
489
490 #ifdef JFR_HAVE_INTRINSICS
491 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
492 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
493 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
494 #endif
495 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
496 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
497 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
498 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
499 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
500 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
501 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
502 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
503 case vmIntrinsics::_getLength: return inline_native_getLength();
504 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
505 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
506 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
507 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
508 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
509 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
510 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
511
512 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
513 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
514
515 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
516
517 case vmIntrinsics::_isInstance:
518 case vmIntrinsics::_getModifiers:
519 case vmIntrinsics::_isInterface:
520 case vmIntrinsics::_isArray:
521 case vmIntrinsics::_isPrimitive:
522 case vmIntrinsics::_isHidden:
523 case vmIntrinsics::_getSuperclass:
524 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
525
526 case vmIntrinsics::_floatToRawIntBits:
527 case vmIntrinsics::_floatToIntBits:
528 case vmIntrinsics::_intBitsToFloat:
529 case vmIntrinsics::_doubleToRawLongBits:
530 case vmIntrinsics::_doubleToLongBits:
531 case vmIntrinsics::_longBitsToDouble:
532 case vmIntrinsics::_floatToFloat16:
533 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
534
535 case vmIntrinsics::_floatIsFinite:
536 case vmIntrinsics::_floatIsInfinite:
537 case vmIntrinsics::_doubleIsFinite:
538 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
539
540 case vmIntrinsics::_numberOfLeadingZeros_i:
541 case vmIntrinsics::_numberOfLeadingZeros_l:
542 case vmIntrinsics::_numberOfTrailingZeros_i:
543 case vmIntrinsics::_numberOfTrailingZeros_l:
544 case vmIntrinsics::_bitCount_i:
545 case vmIntrinsics::_bitCount_l:
546 case vmIntrinsics::_reverse_i:
547 case vmIntrinsics::_reverse_l:
548 case vmIntrinsics::_reverseBytes_i:
549 case vmIntrinsics::_reverseBytes_l:
550 case vmIntrinsics::_reverseBytes_s:
551 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
552
553 case vmIntrinsics::_compress_i:
554 case vmIntrinsics::_compress_l:
555 case vmIntrinsics::_expand_i:
556 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
557
558 case vmIntrinsics::_compareUnsigned_i:
559 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
560
561 case vmIntrinsics::_divideUnsigned_i:
562 case vmIntrinsics::_divideUnsigned_l:
563 case vmIntrinsics::_remainderUnsigned_i:
564 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
565
566 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
567
568 case vmIntrinsics::_Reference_get: return inline_reference_get();
569 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
570 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
571 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
572 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
573
574 case vmIntrinsics::_Class_cast: return inline_Class_cast();
575
576 case vmIntrinsics::_aescrypt_encryptBlock:
577 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
578
579 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
580 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
581 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
582
583 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
584 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
585 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
586
587 case vmIntrinsics::_counterMode_AESCrypt:
588 return inline_counterMode_AESCrypt(intrinsic_id());
589
590 case vmIntrinsics::_galoisCounterMode_AESCrypt:
591 return inline_galoisCounterMode_AESCrypt();
592
593 case vmIntrinsics::_md5_implCompress:
594 case vmIntrinsics::_sha_implCompress:
595 case vmIntrinsics::_sha2_implCompress:
596 case vmIntrinsics::_sha5_implCompress:
597 case vmIntrinsics::_sha3_implCompress:
598 return inline_digestBase_implCompress(intrinsic_id());
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::_base64_encodeBlock:
630 return inline_base64_encodeBlock();
631 case vmIntrinsics::_base64_decodeBlock:
632 return inline_base64_decodeBlock();
633 case vmIntrinsics::_poly1305_processBlocks:
634 return inline_poly1305_processBlocks();
635 case vmIntrinsics::_intpoly_montgomeryMult_P256:
636 return inline_intpoly_montgomeryMult_P256();
637 case vmIntrinsics::_intpoly_assign:
638 return inline_intpoly_assign();
639 case vmIntrinsics::_encodeISOArray:
640 case vmIntrinsics::_encodeByteISOArray:
641 return inline_encodeISOArray(false);
642 case vmIntrinsics::_encodeAsciiArray:
643 return inline_encodeISOArray(true);
644
645 case vmIntrinsics::_updateCRC32:
646 return inline_updateCRC32();
647 case vmIntrinsics::_updateBytesCRC32:
648 return inline_updateBytesCRC32();
649 case vmIntrinsics::_updateByteBufferCRC32:
650 return inline_updateByteBufferCRC32();
651
652 case vmIntrinsics::_updateBytesCRC32C:
653 return inline_updateBytesCRC32C();
654 case vmIntrinsics::_updateDirectByteBufferCRC32C:
655 return inline_updateDirectByteBufferCRC32C();
656
657 case vmIntrinsics::_updateBytesAdler32:
658 return inline_updateBytesAdler32();
659 case vmIntrinsics::_updateByteBufferAdler32:
660 return inline_updateByteBufferAdler32();
661
662 case vmIntrinsics::_profileBoolean:
663 return inline_profileBoolean();
664 case vmIntrinsics::_isCompileConstant:
665 return inline_isCompileConstant();
666
667 case vmIntrinsics::_countPositives:
668 return inline_countPositives();
669
670 case vmIntrinsics::_fmaD:
671 case vmIntrinsics::_fmaF:
672 return inline_fma(intrinsic_id());
673
674 case vmIntrinsics::_isDigit:
675 case vmIntrinsics::_isLowerCase:
676 case vmIntrinsics::_isUpperCase:
677 case vmIntrinsics::_isWhitespace:
678 return inline_character_compare(intrinsic_id());
679
680 case vmIntrinsics::_min:
681 case vmIntrinsics::_max:
682 case vmIntrinsics::_min_strict:
683 case vmIntrinsics::_max_strict:
684 return inline_min_max(intrinsic_id());
685
686 case vmIntrinsics::_maxF:
687 case vmIntrinsics::_minF:
688 case vmIntrinsics::_maxD:
689 case vmIntrinsics::_minD:
690 case vmIntrinsics::_maxF_strict:
691 case vmIntrinsics::_minF_strict:
692 case vmIntrinsics::_maxD_strict:
693 case vmIntrinsics::_minD_strict:
694 return inline_fp_min_max(intrinsic_id());
695
696 case vmIntrinsics::_VectorUnaryOp:
697 return inline_vector_nary_operation(1);
698 case vmIntrinsics::_VectorBinaryOp:
699 return inline_vector_nary_operation(2);
700 case vmIntrinsics::_VectorTernaryOp:
701 return inline_vector_nary_operation(3);
702 case vmIntrinsics::_VectorFromBitsCoerced:
703 return inline_vector_frombits_coerced();
704 case vmIntrinsics::_VectorMaskOp:
705 return inline_vector_mask_operation();
706 case vmIntrinsics::_VectorLoadOp:
707 return inline_vector_mem_operation(/*is_store=*/false);
708 case vmIntrinsics::_VectorLoadMaskedOp:
709 return inline_vector_mem_masked_operation(/*is_store*/false);
710 case vmIntrinsics::_VectorStoreOp:
711 return inline_vector_mem_operation(/*is_store=*/true);
712 case vmIntrinsics::_VectorStoreMaskedOp:
713 return inline_vector_mem_masked_operation(/*is_store=*/true);
714 case vmIntrinsics::_VectorGatherOp:
715 return inline_vector_gather_scatter(/*is_scatter*/ false);
716 case vmIntrinsics::_VectorScatterOp:
717 return inline_vector_gather_scatter(/*is_scatter*/ true);
718 case vmIntrinsics::_VectorReductionCoerced:
719 return inline_vector_reduction();
720 case vmIntrinsics::_VectorTest:
721 return inline_vector_test();
722 case vmIntrinsics::_VectorBlend:
723 return inline_vector_blend();
724 case vmIntrinsics::_VectorRearrange:
725 return inline_vector_rearrange();
726 case vmIntrinsics::_VectorSelectFrom:
727 return inline_vector_select_from();
728 case vmIntrinsics::_VectorCompare:
729 return inline_vector_compare();
730 case vmIntrinsics::_VectorBroadcastInt:
731 return inline_vector_broadcast_int();
732 case vmIntrinsics::_VectorConvert:
733 return inline_vector_convert();
734 case vmIntrinsics::_VectorInsert:
735 return inline_vector_insert();
736 case vmIntrinsics::_VectorExtract:
737 return inline_vector_extract();
738 case vmIntrinsics::_VectorCompressExpand:
739 return inline_vector_compress_expand();
740 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
741 return inline_vector_select_from_two_vectors();
742 case vmIntrinsics::_IndexVector:
743 return inline_index_vector();
744 case vmIntrinsics::_IndexPartiallyInUpperRange:
745 return inline_index_partially_in_upper_range();
746
747 case vmIntrinsics::_getObjectSize:
748 return inline_getObjectSize();
749
750 case vmIntrinsics::_blackhole:
751 return inline_blackhole();
752
753 default:
754 // If you get here, it may be that someone has added a new intrinsic
755 // to the list in vmIntrinsics.hpp without implementing it here.
756 #ifndef PRODUCT
757 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
758 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
759 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
760 }
761 #endif
762 return false;
763 }
764 }
765
766 Node* LibraryCallKit::try_to_predicate(int predicate) {
767 if (!jvms()->has_method()) {
768 // Root JVMState has a null method.
769 assert(map()->memory()->Opcode() == Op_Parm, "");
770 // Insert the memory aliasing node
771 set_all_memory(reset_memory());
772 }
773 assert(merged_memory(), "");
774
775 switch (intrinsic_id()) {
776 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
777 return inline_cipherBlockChaining_AESCrypt_predicate(false);
778 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
779 return inline_cipherBlockChaining_AESCrypt_predicate(true);
780 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
781 return inline_electronicCodeBook_AESCrypt_predicate(false);
782 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
783 return inline_electronicCodeBook_AESCrypt_predicate(true);
784 case vmIntrinsics::_counterMode_AESCrypt:
785 return inline_counterMode_AESCrypt_predicate();
786 case vmIntrinsics::_digestBase_implCompressMB:
787 return inline_digestBase_implCompressMB_predicate(predicate);
788 case vmIntrinsics::_galoisCounterMode_AESCrypt:
789 return inline_galoisCounterMode_AESCrypt_predicate();
790
791 default:
792 // If you get here, it may be that someone has added a new intrinsic
793 // to the list in vmIntrinsics.hpp without implementing it here.
794 #ifndef PRODUCT
795 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
796 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
797 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
798 }
799 #endif
800 Node* slow_ctl = control();
801 set_control(top()); // No fast path intrinsic
802 return slow_ctl;
803 }
804 }
805
806 //------------------------------set_result-------------------------------
807 // Helper function for finishing intrinsics.
808 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
809 record_for_igvn(region);
810 set_control(_gvn.transform(region));
811 set_result( _gvn.transform(value));
812 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
813 }
814
815 //------------------------------generate_guard---------------------------
816 // Helper function for generating guarded fast-slow graph structures.
817 // The given 'test', if true, guards a slow path. If the test fails
818 // then a fast path can be taken. (We generally hope it fails.)
819 // In all cases, GraphKit::control() is updated to the fast path.
820 // The returned value represents the control for the slow path.
821 // The return value is never 'top'; it is either a valid control
822 // or null if it is obvious that the slow path can never be taken.
823 // Also, if region and the slow control are not null, the slow edge
824 // is appended to the region.
825 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
826 if (stopped()) {
827 // Already short circuited.
828 return nullptr;
829 }
830
831 // Build an if node and its projections.
832 // If test is true we take the slow path, which we assume is uncommon.
833 if (_gvn.type(test) == TypeInt::ZERO) {
834 // The slow branch is never taken. No need to build this guard.
835 return nullptr;
836 }
837
838 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
839
840 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
841 if (if_slow == top()) {
842 // The slow branch is never taken. No need to build this guard.
843 return nullptr;
844 }
845
846 if (region != nullptr)
847 region->add_req(if_slow);
848
849 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
850 set_control(if_fast);
851
852 return if_slow;
853 }
854
855 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
856 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
857 }
858 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
859 return generate_guard(test, region, PROB_FAIR);
860 }
861
862 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
863 Node* *pos_index) {
864 if (stopped())
865 return nullptr; // already stopped
866 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
867 return nullptr; // index is already adequately typed
868 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
869 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
870 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
871 if (is_neg != nullptr && pos_index != nullptr) {
872 // Emulate effect of Parse::adjust_map_after_if.
873 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
874 (*pos_index) = _gvn.transform(ccast);
875 }
876 return is_neg;
877 }
878
879 // Make sure that 'position' is a valid limit index, in [0..length].
880 // There are two equivalent plans for checking this:
881 // A. (offset + copyLength) unsigned<= arrayLength
882 // B. offset <= (arrayLength - copyLength)
883 // We require that all of the values above, except for the sum and
884 // difference, are already known to be non-negative.
885 // Plan A is robust in the face of overflow, if offset and copyLength
886 // are both hugely positive.
887 //
888 // Plan B is less direct and intuitive, but it does not overflow at
889 // all, since the difference of two non-negatives is always
890 // representable. Whenever Java methods must perform the equivalent
891 // check they generally use Plan B instead of Plan A.
892 // For the moment we use Plan A.
893 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
894 Node* subseq_length,
895 Node* array_length,
896 RegionNode* region) {
897 if (stopped())
898 return nullptr; // already stopped
899 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
900 if (zero_offset && subseq_length->eqv_uncast(array_length))
901 return nullptr; // common case of whole-array copy
902 Node* last = subseq_length;
903 if (!zero_offset) // last += offset
904 last = _gvn.transform(new AddINode(last, offset));
905 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
906 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
907 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
908 return is_over;
909 }
910
911 // Emit range checks for the given String.value byte array
912 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
913 if (stopped()) {
914 return; // already stopped
915 }
916 RegionNode* bailout = new RegionNode(1);
917 record_for_igvn(bailout);
918 if (char_count) {
919 // Convert char count to byte count
920 count = _gvn.transform(new LShiftINode(count, intcon(1)));
921 }
922
923 // Offset and count must not be negative
924 generate_negative_guard(offset, bailout);
925 generate_negative_guard(count, bailout);
926 // Offset + count must not exceed length of array
927 generate_limit_guard(offset, count, load_array_length(array), bailout);
928
929 if (bailout->req() > 1) {
930 PreserveJVMState pjvms(this);
931 set_control(_gvn.transform(bailout));
932 uncommon_trap(Deoptimization::Reason_intrinsic,
933 Deoptimization::Action_maybe_recompile);
934 }
935 }
936
937 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
938 bool is_immutable) {
939 ciKlass* thread_klass = env()->Thread_klass();
940 const Type* thread_type
941 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
942
943 Node* thread = _gvn.transform(new ThreadLocalNode());
944 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
945 tls_output = thread;
946
947 Node* thread_obj_handle
948 = (is_immutable
949 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
950 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
951 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
952 thread_obj_handle = _gvn.transform(thread_obj_handle);
953
954 DecoratorSet decorators = IN_NATIVE;
955 if (is_immutable) {
956 decorators |= C2_IMMUTABLE_MEMORY;
957 }
958 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
959 }
960
961 //--------------------------generate_current_thread--------------------
962 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
963 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
964 /*is_immutable*/false);
965 }
966
967 //--------------------------generate_virtual_thread--------------------
968 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
969 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
970 !C->method()->changes_current_thread());
971 }
972
973 //------------------------------make_string_method_node------------------------
974 // Helper method for String intrinsic functions. This version is called with
975 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
976 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
977 // containing the lengths of str1 and str2.
978 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
979 Node* result = nullptr;
980 switch (opcode) {
981 case Op_StrIndexOf:
982 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
983 str1_start, cnt1, str2_start, cnt2, ae);
984 break;
985 case Op_StrComp:
986 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
987 str1_start, cnt1, str2_start, cnt2, ae);
988 break;
989 case Op_StrEquals:
990 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
991 // Use the constant length if there is one because optimized match rule may exist.
992 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
993 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
994 break;
995 default:
996 ShouldNotReachHere();
997 return nullptr;
998 }
999
1000 // All these intrinsics have checks.
1001 C->set_has_split_ifs(true); // Has chance for split-if optimization
1002 clear_upper_avx();
1003
1004 return _gvn.transform(result);
1005 }
1006
1007 //------------------------------inline_string_compareTo------------------------
1008 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1009 Node* arg1 = argument(0);
1010 Node* arg2 = argument(1);
1011
1012 arg1 = must_be_not_null(arg1, true);
1013 arg2 = must_be_not_null(arg2, true);
1014
1015 // Get start addr and length of first argument
1016 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1017 Node* arg1_cnt = load_array_length(arg1);
1018
1019 // Get start addr and length of second argument
1020 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1021 Node* arg2_cnt = load_array_length(arg2);
1022
1023 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1024 set_result(result);
1025 return true;
1026 }
1027
1028 //------------------------------inline_string_equals------------------------
1029 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1030 Node* arg1 = argument(0);
1031 Node* arg2 = argument(1);
1032
1033 // paths (plus control) merge
1034 RegionNode* region = new RegionNode(3);
1035 Node* phi = new PhiNode(region, TypeInt::BOOL);
1036
1037 if (!stopped()) {
1038
1039 arg1 = must_be_not_null(arg1, true);
1040 arg2 = must_be_not_null(arg2, true);
1041
1042 // Get start addr and length of first argument
1043 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1044 Node* arg1_cnt = load_array_length(arg1);
1045
1046 // Get start addr and length of second argument
1047 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1048 Node* arg2_cnt = load_array_length(arg2);
1049
1050 // Check for arg1_cnt != arg2_cnt
1051 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1052 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1053 Node* if_ne = generate_slow_guard(bol, nullptr);
1054 if (if_ne != nullptr) {
1055 phi->init_req(2, intcon(0));
1056 region->init_req(2, if_ne);
1057 }
1058
1059 // Check for count == 0 is done by assembler code for StrEquals.
1060
1061 if (!stopped()) {
1062 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1063 phi->init_req(1, equals);
1064 region->init_req(1, control());
1065 }
1066 }
1067
1068 // post merge
1069 set_control(_gvn.transform(region));
1070 record_for_igvn(region);
1071
1072 set_result(_gvn.transform(phi));
1073 return true;
1074 }
1075
1076 //------------------------------inline_array_equals----------------------------
1077 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1078 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1079 Node* arg1 = argument(0);
1080 Node* arg2 = argument(1);
1081
1082 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1083 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1084 clear_upper_avx();
1085
1086 return true;
1087 }
1088
1089
1090 //------------------------------inline_countPositives------------------------------
1091 bool LibraryCallKit::inline_countPositives() {
1092 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1093 return false;
1094 }
1095
1096 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1097 // no receiver since it is static method
1098 Node* ba = argument(0);
1099 Node* offset = argument(1);
1100 Node* len = argument(2);
1101
1102 ba = must_be_not_null(ba, true);
1103
1104 // Range checks
1105 generate_string_range_check(ba, offset, len, false);
1106 if (stopped()) {
1107 return true;
1108 }
1109 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1110 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1111 set_result(_gvn.transform(result));
1112 clear_upper_avx();
1113 return true;
1114 }
1115
1116 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1117 Node* index = argument(0);
1118 Node* length = bt == T_INT ? argument(1) : argument(2);
1119 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1120 return false;
1121 }
1122
1123 // check that length is positive
1124 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1125 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1126
1127 {
1128 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1129 uncommon_trap(Deoptimization::Reason_intrinsic,
1130 Deoptimization::Action_make_not_entrant);
1131 }
1132
1133 if (stopped()) {
1134 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1135 return true;
1136 }
1137
1138 // length is now known positive, add a cast node to make this explicit
1139 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1140 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1141 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1142 ConstraintCastNode::RegularDependency, bt);
1143 casted_length = _gvn.transform(casted_length);
1144 replace_in_map(length, casted_length);
1145 length = casted_length;
1146
1147 // Use an unsigned comparison for the range check itself
1148 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1149 BoolTest::mask btest = BoolTest::lt;
1150 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1151 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1152 _gvn.set_type(rc, rc->Value(&_gvn));
1153 if (!rc_bool->is_Con()) {
1154 record_for_igvn(rc);
1155 }
1156 set_control(_gvn.transform(new IfTrueNode(rc)));
1157 {
1158 PreserveJVMState pjvms(this);
1159 set_control(_gvn.transform(new IfFalseNode(rc)));
1160 uncommon_trap(Deoptimization::Reason_range_check,
1161 Deoptimization::Action_make_not_entrant);
1162 }
1163
1164 if (stopped()) {
1165 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1166 return true;
1167 }
1168
1169 // index is now known to be >= 0 and < length, cast it
1170 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1171 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1172 ConstraintCastNode::RegularDependency, bt);
1173 result = _gvn.transform(result);
1174 set_result(result);
1175 replace_in_map(index, result);
1176 return true;
1177 }
1178
1179 //------------------------------inline_string_indexOf------------------------
1180 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1181 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1182 return false;
1183 }
1184 Node* src = argument(0);
1185 Node* tgt = argument(1);
1186
1187 // Make the merge point
1188 RegionNode* result_rgn = new RegionNode(4);
1189 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1190
1191 src = must_be_not_null(src, true);
1192 tgt = must_be_not_null(tgt, true);
1193
1194 // Get start addr and length of source string
1195 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1196 Node* src_count = load_array_length(src);
1197
1198 // Get start addr and length of substring
1199 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1200 Node* tgt_count = load_array_length(tgt);
1201
1202 Node* result = nullptr;
1203 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1204
1205 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1206 // Divide src size by 2 if String is UTF16 encoded
1207 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1208 }
1209 if (ae == StrIntrinsicNode::UU) {
1210 // Divide substring size by 2 if String is UTF16 encoded
1211 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1212 }
1213
1214 if (call_opt_stub) {
1215 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1216 StubRoutines::_string_indexof_array[ae],
1217 "stringIndexOf", TypePtr::BOTTOM, src_start,
1218 src_count, tgt_start, tgt_count);
1219 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1220 } else {
1221 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1222 result_rgn, result_phi, ae);
1223 }
1224 if (result != nullptr) {
1225 result_phi->init_req(3, result);
1226 result_rgn->init_req(3, control());
1227 }
1228 set_control(_gvn.transform(result_rgn));
1229 record_for_igvn(result_rgn);
1230 set_result(_gvn.transform(result_phi));
1231
1232 return true;
1233 }
1234
1235 //-----------------------------inline_string_indexOfI-----------------------
1236 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1237 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1238 return false;
1239 }
1240 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1241 return false;
1242 }
1243
1244 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1245 Node* src = argument(0); // byte[]
1246 Node* src_count = argument(1); // char count
1247 Node* tgt = argument(2); // byte[]
1248 Node* tgt_count = argument(3); // char count
1249 Node* from_index = argument(4); // char index
1250
1251 src = must_be_not_null(src, true);
1252 tgt = must_be_not_null(tgt, true);
1253
1254 // Multiply byte array index by 2 if String is UTF16 encoded
1255 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1256 src_count = _gvn.transform(new SubINode(src_count, from_index));
1257 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1258 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1259
1260 // Range checks
1261 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1262 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1263 if (stopped()) {
1264 return true;
1265 }
1266
1267 RegionNode* region = new RegionNode(5);
1268 Node* phi = new PhiNode(region, TypeInt::INT);
1269 Node* result = nullptr;
1270
1271 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1272
1273 if (call_opt_stub) {
1274 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1275 StubRoutines::_string_indexof_array[ae],
1276 "stringIndexOf", TypePtr::BOTTOM, src_start,
1277 src_count, tgt_start, tgt_count);
1278 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1279 } else {
1280 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1281 region, phi, ae);
1282 }
1283 if (result != nullptr) {
1284 // The result is index relative to from_index if substring was found, -1 otherwise.
1285 // Generate code which will fold into cmove.
1286 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1287 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1288
1289 Node* if_lt = generate_slow_guard(bol, nullptr);
1290 if (if_lt != nullptr) {
1291 // result == -1
1292 phi->init_req(3, result);
1293 region->init_req(3, if_lt);
1294 }
1295 if (!stopped()) {
1296 result = _gvn.transform(new AddINode(result, from_index));
1297 phi->init_req(4, result);
1298 region->init_req(4, control());
1299 }
1300 }
1301
1302 set_control(_gvn.transform(region));
1303 record_for_igvn(region);
1304 set_result(_gvn.transform(phi));
1305 clear_upper_avx();
1306
1307 return true;
1308 }
1309
1310 // Create StrIndexOfNode with fast path checks
1311 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1312 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1313 // Check for substr count > string count
1314 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1315 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1316 Node* if_gt = generate_slow_guard(bol, nullptr);
1317 if (if_gt != nullptr) {
1318 phi->init_req(1, intcon(-1));
1319 region->init_req(1, if_gt);
1320 }
1321 if (!stopped()) {
1322 // Check for substr count == 0
1323 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1324 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1325 Node* if_zero = generate_slow_guard(bol, nullptr);
1326 if (if_zero != nullptr) {
1327 phi->init_req(2, intcon(0));
1328 region->init_req(2, if_zero);
1329 }
1330 }
1331 if (!stopped()) {
1332 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1333 }
1334 return nullptr;
1335 }
1336
1337 //-----------------------------inline_string_indexOfChar-----------------------
1338 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1339 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1340 return false;
1341 }
1342 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1343 return false;
1344 }
1345 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1346 Node* src = argument(0); // byte[]
1347 Node* int_ch = argument(1);
1348 Node* from_index = argument(2);
1349 Node* max = argument(3);
1350
1351 src = must_be_not_null(src, true);
1352
1353 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1354 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1355 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1356
1357 // Range checks
1358 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1359
1360 // Check for int_ch >= 0
1361 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1362 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1363 {
1364 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1365 uncommon_trap(Deoptimization::Reason_intrinsic,
1366 Deoptimization::Action_maybe_recompile);
1367 }
1368 if (stopped()) {
1369 return true;
1370 }
1371
1372 RegionNode* region = new RegionNode(3);
1373 Node* phi = new PhiNode(region, TypeInt::INT);
1374
1375 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1376 C->set_has_split_ifs(true); // Has chance for split-if optimization
1377 _gvn.transform(result);
1378
1379 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1380 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1381
1382 Node* if_lt = generate_slow_guard(bol, nullptr);
1383 if (if_lt != nullptr) {
1384 // result == -1
1385 phi->init_req(2, result);
1386 region->init_req(2, if_lt);
1387 }
1388 if (!stopped()) {
1389 result = _gvn.transform(new AddINode(result, from_index));
1390 phi->init_req(1, result);
1391 region->init_req(1, control());
1392 }
1393 set_control(_gvn.transform(region));
1394 record_for_igvn(region);
1395 set_result(_gvn.transform(phi));
1396 clear_upper_avx();
1397
1398 return true;
1399 }
1400 //---------------------------inline_string_copy---------------------
1401 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1402 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1403 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1404 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1405 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1406 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1407 bool LibraryCallKit::inline_string_copy(bool compress) {
1408 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1409 return false;
1410 }
1411 int nargs = 5; // 2 oops, 3 ints
1412 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1413
1414 Node* src = argument(0);
1415 Node* src_offset = argument(1);
1416 Node* dst = argument(2);
1417 Node* dst_offset = argument(3);
1418 Node* length = argument(4);
1419
1420 // Check for allocation before we add nodes that would confuse
1421 // tightly_coupled_allocation()
1422 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1423
1424 // Figure out the size and type of the elements we will be copying.
1425 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1426 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1427 if (src_type == nullptr || dst_type == nullptr) {
1428 return false;
1429 }
1430 BasicType src_elem = src_type->elem()->array_element_basic_type();
1431 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1432 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1433 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1434 "Unsupported array types for inline_string_copy");
1435
1436 src = must_be_not_null(src, true);
1437 dst = must_be_not_null(dst, true);
1438
1439 // Convert char[] offsets to byte[] offsets
1440 bool convert_src = (compress && src_elem == T_BYTE);
1441 bool convert_dst = (!compress && dst_elem == T_BYTE);
1442 if (convert_src) {
1443 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1444 } else if (convert_dst) {
1445 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1446 }
1447
1448 // Range checks
1449 generate_string_range_check(src, src_offset, length, convert_src);
1450 generate_string_range_check(dst, dst_offset, length, convert_dst);
1451 if (stopped()) {
1452 return true;
1453 }
1454
1455 Node* src_start = array_element_address(src, src_offset, src_elem);
1456 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1457 // 'src_start' points to src array + scaled offset
1458 // 'dst_start' points to dst array + scaled offset
1459 Node* count = nullptr;
1460 if (compress) {
1461 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1462 } else {
1463 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1464 }
1465
1466 if (alloc != nullptr) {
1467 if (alloc->maybe_set_complete(&_gvn)) {
1468 // "You break it, you buy it."
1469 InitializeNode* init = alloc->initialization();
1470 assert(init->is_complete(), "we just did this");
1471 init->set_complete_with_arraycopy();
1472 assert(dst->is_CheckCastPP(), "sanity");
1473 assert(dst->in(0)->in(0) == init, "dest pinned");
1474 }
1475 // Do not let stores that initialize this object be reordered with
1476 // a subsequent store that would make this object accessible by
1477 // other threads.
1478 // Record what AllocateNode this StoreStore protects so that
1479 // escape analysis can go from the MemBarStoreStoreNode to the
1480 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1481 // based on the escape status of the AllocateNode.
1482 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1483 }
1484 if (compress) {
1485 set_result(_gvn.transform(count));
1486 }
1487 clear_upper_avx();
1488
1489 return true;
1490 }
1491
1492 #ifdef _LP64
1493 #define XTOP ,top() /*additional argument*/
1494 #else //_LP64
1495 #define XTOP /*no additional argument*/
1496 #endif //_LP64
1497
1498 //------------------------inline_string_toBytesU--------------------------
1499 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1500 bool LibraryCallKit::inline_string_toBytesU() {
1501 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1502 return false;
1503 }
1504 // Get the arguments.
1505 Node* value = argument(0);
1506 Node* offset = argument(1);
1507 Node* length = argument(2);
1508
1509 Node* newcopy = nullptr;
1510
1511 // Set the original stack and the reexecute bit for the interpreter to reexecute
1512 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1513 { PreserveReexecuteState preexecs(this);
1514 jvms()->set_should_reexecute(true);
1515
1516 // Check if a null path was taken unconditionally.
1517 value = null_check(value);
1518
1519 RegionNode* bailout = new RegionNode(1);
1520 record_for_igvn(bailout);
1521
1522 // Range checks
1523 generate_negative_guard(offset, bailout);
1524 generate_negative_guard(length, bailout);
1525 generate_limit_guard(offset, length, load_array_length(value), bailout);
1526 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1527 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1528
1529 if (bailout->req() > 1) {
1530 PreserveJVMState pjvms(this);
1531 set_control(_gvn.transform(bailout));
1532 uncommon_trap(Deoptimization::Reason_intrinsic,
1533 Deoptimization::Action_maybe_recompile);
1534 }
1535 if (stopped()) {
1536 return true;
1537 }
1538
1539 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1540 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1541 newcopy = new_array(klass_node, size, 0); // no arguments to push
1542 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1543 guarantee(alloc != nullptr, "created above");
1544
1545 // Calculate starting addresses.
1546 Node* src_start = array_element_address(value, offset, T_CHAR);
1547 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1548
1549 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1550 const TypeInt* toffset = gvn().type(offset)->is_int();
1551 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1552
1553 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1554 const char* copyfunc_name = "arraycopy";
1555 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1556 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1557 OptoRuntime::fast_arraycopy_Type(),
1558 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1559 src_start, dst_start, ConvI2X(length) XTOP);
1560 // Do not let reads from the cloned object float above the arraycopy.
1561 if (alloc->maybe_set_complete(&_gvn)) {
1562 // "You break it, you buy it."
1563 InitializeNode* init = alloc->initialization();
1564 assert(init->is_complete(), "we just did this");
1565 init->set_complete_with_arraycopy();
1566 assert(newcopy->is_CheckCastPP(), "sanity");
1567 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1568 }
1569 // Do not let stores that initialize this object be reordered with
1570 // a subsequent store that would make this object accessible by
1571 // other threads.
1572 // Record what AllocateNode this StoreStore protects so that
1573 // escape analysis can go from the MemBarStoreStoreNode to the
1574 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1575 // based on the escape status of the AllocateNode.
1576 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1577 } // original reexecute is set back here
1578
1579 C->set_has_split_ifs(true); // Has chance for split-if optimization
1580 if (!stopped()) {
1581 set_result(newcopy);
1582 }
1583 clear_upper_avx();
1584
1585 return true;
1586 }
1587
1588 //------------------------inline_string_getCharsU--------------------------
1589 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1590 bool LibraryCallKit::inline_string_getCharsU() {
1591 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1592 return false;
1593 }
1594
1595 // Get the arguments.
1596 Node* src = argument(0);
1597 Node* src_begin = argument(1);
1598 Node* src_end = argument(2); // exclusive offset (i < src_end)
1599 Node* dst = argument(3);
1600 Node* dst_begin = argument(4);
1601
1602 // Check for allocation before we add nodes that would confuse
1603 // tightly_coupled_allocation()
1604 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1605
1606 // Check if a null path was taken unconditionally.
1607 src = null_check(src);
1608 dst = null_check(dst);
1609 if (stopped()) {
1610 return true;
1611 }
1612
1613 // Get length and convert char[] offset to byte[] offset
1614 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1615 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1616
1617 // Range checks
1618 generate_string_range_check(src, src_begin, length, true);
1619 generate_string_range_check(dst, dst_begin, length, false);
1620 if (stopped()) {
1621 return true;
1622 }
1623
1624 if (!stopped()) {
1625 // Calculate starting addresses.
1626 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1627 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1628
1629 // Check if array addresses are aligned to HeapWordSize
1630 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1631 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1632 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1633 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1634
1635 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1636 const char* copyfunc_name = "arraycopy";
1637 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1638 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1639 OptoRuntime::fast_arraycopy_Type(),
1640 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1641 src_start, dst_start, ConvI2X(length) XTOP);
1642 // Do not let reads from the cloned object float above the arraycopy.
1643 if (alloc != nullptr) {
1644 if (alloc->maybe_set_complete(&_gvn)) {
1645 // "You break it, you buy it."
1646 InitializeNode* init = alloc->initialization();
1647 assert(init->is_complete(), "we just did this");
1648 init->set_complete_with_arraycopy();
1649 assert(dst->is_CheckCastPP(), "sanity");
1650 assert(dst->in(0)->in(0) == init, "dest pinned");
1651 }
1652 // Do not let stores that initialize this object be reordered with
1653 // a subsequent store that would make this object accessible by
1654 // other threads.
1655 // Record what AllocateNode this StoreStore protects so that
1656 // escape analysis can go from the MemBarStoreStoreNode to the
1657 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1658 // based on the escape status of the AllocateNode.
1659 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1660 } else {
1661 insert_mem_bar(Op_MemBarCPUOrder);
1662 }
1663 }
1664
1665 C->set_has_split_ifs(true); // Has chance for split-if optimization
1666 return true;
1667 }
1668
1669 //----------------------inline_string_char_access----------------------------
1670 // Store/Load char to/from byte[] array.
1671 // static void StringUTF16.putChar(byte[] val, int index, int c)
1672 // static char StringUTF16.getChar(byte[] val, int index)
1673 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1674 Node* value = argument(0);
1675 Node* index = argument(1);
1676 Node* ch = is_store ? argument(2) : nullptr;
1677
1678 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1679 // correctly requires matched array shapes.
1680 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1681 "sanity: byte[] and char[] bases agree");
1682 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1683 "sanity: byte[] and char[] scales agree");
1684
1685 // Bail when getChar over constants is requested: constant folding would
1686 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1687 // Java method would constant fold nicely instead.
1688 if (!is_store && value->is_Con() && index->is_Con()) {
1689 return false;
1690 }
1691
1692 // Save state and restore on bailout
1693 uint old_sp = sp();
1694 SafePointNode* old_map = clone_map();
1695
1696 value = must_be_not_null(value, true);
1697
1698 Node* adr = array_element_address(value, index, T_CHAR);
1699 if (adr->is_top()) {
1700 set_map(old_map);
1701 set_sp(old_sp);
1702 return false;
1703 }
1704 destruct_map_clone(old_map);
1705 if (is_store) {
1706 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1707 } else {
1708 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);
1709 set_result(ch);
1710 }
1711 return true;
1712 }
1713
1714 //--------------------------round_double_node--------------------------------
1715 // Round a double node if necessary.
1716 Node* LibraryCallKit::round_double_node(Node* n) {
1717 if (Matcher::strict_fp_requires_explicit_rounding) {
1718 #ifdef IA32
1719 if (UseSSE < 2) {
1720 n = _gvn.transform(new RoundDoubleNode(nullptr, n));
1721 }
1722 #else
1723 Unimplemented();
1724 #endif // IA32
1725 }
1726 return n;
1727 }
1728
1729 //------------------------------inline_math-----------------------------------
1730 // public static double Math.abs(double)
1731 // public static double Math.sqrt(double)
1732 // public static double Math.log(double)
1733 // public static double Math.log10(double)
1734 // public static double Math.round(double)
1735 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1736 Node* arg = round_double_node(argument(0));
1737 Node* n = nullptr;
1738 switch (id) {
1739 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1740 case vmIntrinsics::_dsqrt:
1741 case vmIntrinsics::_dsqrt_strict:
1742 n = new SqrtDNode(C, control(), arg); break;
1743 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1744 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1745 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1746 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1747 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1748 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1749 default: fatal_unexpected_iid(id); break;
1750 }
1751 set_result(_gvn.transform(n));
1752 return true;
1753 }
1754
1755 //------------------------------inline_math-----------------------------------
1756 // public static float Math.abs(float)
1757 // public static int Math.abs(int)
1758 // public static long Math.abs(long)
1759 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1760 Node* arg = argument(0);
1761 Node* n = nullptr;
1762 switch (id) {
1763 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1764 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1765 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1766 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1767 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1768 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1769 default: fatal_unexpected_iid(id); break;
1770 }
1771 set_result(_gvn.transform(n));
1772 return true;
1773 }
1774
1775 //------------------------------runtime_math-----------------------------
1776 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1777 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1778 "must be (DD)D or (D)D type");
1779
1780 // Inputs
1781 Node* a = round_double_node(argument(0));
1782 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : nullptr;
1783
1784 const TypePtr* no_memory_effects = nullptr;
1785 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1786 no_memory_effects,
1787 a, top(), b, b ? top() : nullptr);
1788 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1789 #ifdef ASSERT
1790 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1791 assert(value_top == top(), "second value must be top");
1792 #endif
1793
1794 set_result(value);
1795 return true;
1796 }
1797
1798 //------------------------------inline_math_pow-----------------------------
1799 bool LibraryCallKit::inline_math_pow() {
1800 Node* exp = round_double_node(argument(2));
1801 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1802 if (d != nullptr) {
1803 if (d->getd() == 2.0) {
1804 // Special case: pow(x, 2.0) => x * x
1805 Node* base = round_double_node(argument(0));
1806 set_result(_gvn.transform(new MulDNode(base, base)));
1807 return true;
1808 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1809 // Special case: pow(x, 0.5) => sqrt(x)
1810 Node* base = round_double_node(argument(0));
1811 Node* zero = _gvn.zerocon(T_DOUBLE);
1812
1813 RegionNode* region = new RegionNode(3);
1814 Node* phi = new PhiNode(region, Type::DOUBLE);
1815
1816 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1817 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1818 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1819 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1820 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1821
1822 Node* if_pow = generate_slow_guard(test, nullptr);
1823 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1824 phi->init_req(1, value_sqrt);
1825 region->init_req(1, control());
1826
1827 if (if_pow != nullptr) {
1828 set_control(if_pow);
1829 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1830 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1831 const TypePtr* no_memory_effects = nullptr;
1832 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1833 no_memory_effects, base, top(), exp, top());
1834 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1835 #ifdef ASSERT
1836 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1837 assert(value_top == top(), "second value must be top");
1838 #endif
1839 phi->init_req(2, value_pow);
1840 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1841 }
1842
1843 C->set_has_split_ifs(true); // Has chance for split-if optimization
1844 set_control(_gvn.transform(region));
1845 record_for_igvn(region);
1846 set_result(_gvn.transform(phi));
1847
1848 return true;
1849 }
1850 }
1851
1852 return StubRoutines::dpow() != nullptr ?
1853 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1854 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1855 }
1856
1857 //------------------------------inline_math_native-----------------------------
1858 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1859 switch (id) {
1860 case vmIntrinsics::_dsin:
1861 return StubRoutines::dsin() != nullptr ?
1862 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1863 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1864 case vmIntrinsics::_dcos:
1865 return StubRoutines::dcos() != nullptr ?
1866 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1867 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1868 case vmIntrinsics::_dtan:
1869 return StubRoutines::dtan() != nullptr ?
1870 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1871 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1872 case vmIntrinsics::_dtanh:
1873 return StubRoutines::dtanh() != nullptr ?
1874 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1875 case vmIntrinsics::_dexp:
1876 return StubRoutines::dexp() != nullptr ?
1877 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1878 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1879 case vmIntrinsics::_dlog:
1880 return StubRoutines::dlog() != nullptr ?
1881 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1882 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1883 case vmIntrinsics::_dlog10:
1884 return StubRoutines::dlog10() != nullptr ?
1885 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1886 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1887
1888 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1889 case vmIntrinsics::_ceil:
1890 case vmIntrinsics::_floor:
1891 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1892
1893 case vmIntrinsics::_dsqrt:
1894 case vmIntrinsics::_dsqrt_strict:
1895 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1896 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1897 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1898 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1899 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1900
1901 case vmIntrinsics::_dpow: return inline_math_pow();
1902 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1903 case vmIntrinsics::_fcopySign: return inline_math(id);
1904 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1905 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1906 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1907
1908 // These intrinsics are not yet correctly implemented
1909 case vmIntrinsics::_datan2:
1910 return false;
1911
1912 default:
1913 fatal_unexpected_iid(id);
1914 return false;
1915 }
1916 }
1917
1918 //----------------------------inline_notify-----------------------------------*
1919 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1920 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1921 address func;
1922 if (id == vmIntrinsics::_notify) {
1923 func = OptoRuntime::monitor_notify_Java();
1924 } else {
1925 func = OptoRuntime::monitor_notifyAll_Java();
1926 }
1927 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1928 make_slow_call_ex(call, env()->Throwable_klass(), false);
1929 return true;
1930 }
1931
1932
1933 //----------------------------inline_min_max-----------------------------------
1934 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1935 set_result(generate_min_max(id, argument(0), argument(1)));
1936 return true;
1937 }
1938
1939 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1940 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1941 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1942 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1943 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1944
1945 {
1946 PreserveJVMState pjvms(this);
1947 PreserveReexecuteState preexecs(this);
1948 jvms()->set_should_reexecute(true);
1949
1950 set_control(slow_path);
1951 set_i_o(i_o());
1952
1953 uncommon_trap(Deoptimization::Reason_intrinsic,
1954 Deoptimization::Action_none);
1955 }
1956
1957 set_control(fast_path);
1958 set_result(math);
1959 }
1960
1961 template <typename OverflowOp>
1962 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1963 typedef typename OverflowOp::MathOp MathOp;
1964
1965 MathOp* mathOp = new MathOp(arg1, arg2);
1966 Node* operation = _gvn.transform( mathOp );
1967 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1968 inline_math_mathExact(operation, ofcheck);
1969 return true;
1970 }
1971
1972 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1973 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1974 }
1975
1976 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1977 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1978 }
1979
1980 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1981 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1982 }
1983
1984 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1985 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1986 }
1987
1988 bool LibraryCallKit::inline_math_negateExactI() {
1989 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1990 }
1991
1992 bool LibraryCallKit::inline_math_negateExactL() {
1993 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1994 }
1995
1996 bool LibraryCallKit::inline_math_multiplyExactI() {
1997 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1998 }
1999
2000 bool LibraryCallKit::inline_math_multiplyExactL() {
2001 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2002 }
2003
2004 bool LibraryCallKit::inline_math_multiplyHigh() {
2005 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2006 return true;
2007 }
2008
2009 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2010 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2011 return true;
2012 }
2013
2014 Node*
2015 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
2016 Node* result_val = nullptr;
2017 switch (id) {
2018 case vmIntrinsics::_min:
2019 case vmIntrinsics::_min_strict:
2020 result_val = _gvn.transform(new MinINode(x0, y0));
2021 break;
2022 case vmIntrinsics::_max:
2023 case vmIntrinsics::_max_strict:
2024 result_val = _gvn.transform(new MaxINode(x0, y0));
2025 break;
2026 default:
2027 fatal_unexpected_iid(id);
2028 break;
2029 }
2030 return result_val;
2031 }
2032
2033 inline int
2034 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2035 const TypePtr* base_type = TypePtr::NULL_PTR;
2036 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2037 if (base_type == nullptr) {
2038 // Unknown type.
2039 return Type::AnyPtr;
2040 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2041 // Since this is a null+long form, we have to switch to a rawptr.
2042 base = _gvn.transform(new CastX2PNode(offset));
2043 offset = MakeConX(0);
2044 return Type::RawPtr;
2045 } else if (base_type->base() == Type::RawPtr) {
2046 return Type::RawPtr;
2047 } else if (base_type->isa_oopptr()) {
2048 // Base is never null => always a heap address.
2049 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2050 return Type::OopPtr;
2051 }
2052 // Offset is small => always a heap address.
2053 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2054 if (offset_type != nullptr &&
2055 base_type->offset() == 0 && // (should always be?)
2056 offset_type->_lo >= 0 &&
2057 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2058 return Type::OopPtr;
2059 } else if (type == T_OBJECT) {
2060 // off heap access to an oop doesn't make any sense. Has to be on
2061 // heap.
2062 return Type::OopPtr;
2063 }
2064 // Otherwise, it might either be oop+off or null+addr.
2065 return Type::AnyPtr;
2066 } else {
2067 // No information:
2068 return Type::AnyPtr;
2069 }
2070 }
2071
2072 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2073 Node* uncasted_base = base;
2074 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2075 if (kind == Type::RawPtr) {
2076 return basic_plus_adr(top(), uncasted_base, offset);
2077 } else if (kind == Type::AnyPtr) {
2078 assert(base == uncasted_base, "unexpected base change");
2079 if (can_cast) {
2080 if (!_gvn.type(base)->speculative_maybe_null() &&
2081 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2082 // According to profiling, this access is always on
2083 // heap. Casting the base to not null and thus avoiding membars
2084 // around the access should allow better optimizations
2085 Node* null_ctl = top();
2086 base = null_check_oop(base, &null_ctl, true, true, true);
2087 assert(null_ctl->is_top(), "no null control here");
2088 return basic_plus_adr(base, offset);
2089 } else if (_gvn.type(base)->speculative_always_null() &&
2090 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2091 // According to profiling, this access is always off
2092 // heap.
2093 base = null_assert(base);
2094 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2095 offset = MakeConX(0);
2096 return basic_plus_adr(top(), raw_base, offset);
2097 }
2098 }
2099 // We don't know if it's an on heap or off heap access. Fall back
2100 // to raw memory access.
2101 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2102 return basic_plus_adr(top(), raw, offset);
2103 } else {
2104 assert(base == uncasted_base, "unexpected base change");
2105 // We know it's an on heap access so base can't be null
2106 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2107 base = must_be_not_null(base, true);
2108 }
2109 return basic_plus_adr(base, offset);
2110 }
2111 }
2112
2113 //--------------------------inline_number_methods-----------------------------
2114 // inline int Integer.numberOfLeadingZeros(int)
2115 // inline int Long.numberOfLeadingZeros(long)
2116 //
2117 // inline int Integer.numberOfTrailingZeros(int)
2118 // inline int Long.numberOfTrailingZeros(long)
2119 //
2120 // inline int Integer.bitCount(int)
2121 // inline int Long.bitCount(long)
2122 //
2123 // inline char Character.reverseBytes(char)
2124 // inline short Short.reverseBytes(short)
2125 // inline int Integer.reverseBytes(int)
2126 // inline long Long.reverseBytes(long)
2127 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2128 Node* arg = argument(0);
2129 Node* n = nullptr;
2130 switch (id) {
2131 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2132 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2133 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2134 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2135 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2136 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2137 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2138 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2139 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2140 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2141 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2142 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2143 default: fatal_unexpected_iid(id); break;
2144 }
2145 set_result(_gvn.transform(n));
2146 return true;
2147 }
2148
2149 //--------------------------inline_bitshuffle_methods-----------------------------
2150 // inline int Integer.compress(int, int)
2151 // inline int Integer.expand(int, int)
2152 // inline long Long.compress(long, long)
2153 // inline long Long.expand(long, long)
2154 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2155 Node* n = nullptr;
2156 switch (id) {
2157 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2158 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2159 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2160 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2161 default: fatal_unexpected_iid(id); break;
2162 }
2163 set_result(_gvn.transform(n));
2164 return true;
2165 }
2166
2167 //--------------------------inline_number_methods-----------------------------
2168 // inline int Integer.compareUnsigned(int, int)
2169 // inline int Long.compareUnsigned(long, long)
2170 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2171 Node* arg1 = argument(0);
2172 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2173 Node* n = nullptr;
2174 switch (id) {
2175 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2176 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2177 default: fatal_unexpected_iid(id); break;
2178 }
2179 set_result(_gvn.transform(n));
2180 return true;
2181 }
2182
2183 //--------------------------inline_unsigned_divmod_methods-----------------------------
2184 // inline int Integer.divideUnsigned(int, int)
2185 // inline int Integer.remainderUnsigned(int, int)
2186 // inline long Long.divideUnsigned(long, long)
2187 // inline long Long.remainderUnsigned(long, long)
2188 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2189 Node* n = nullptr;
2190 switch (id) {
2191 case vmIntrinsics::_divideUnsigned_i: {
2192 zero_check_int(argument(1));
2193 // Compile-time detect of null-exception
2194 if (stopped()) {
2195 return true; // keep the graph constructed so far
2196 }
2197 n = new UDivINode(control(), argument(0), argument(1));
2198 break;
2199 }
2200 case vmIntrinsics::_divideUnsigned_l: {
2201 zero_check_long(argument(2));
2202 // Compile-time detect of null-exception
2203 if (stopped()) {
2204 return true; // keep the graph constructed so far
2205 }
2206 n = new UDivLNode(control(), argument(0), argument(2));
2207 break;
2208 }
2209 case vmIntrinsics::_remainderUnsigned_i: {
2210 zero_check_int(argument(1));
2211 // Compile-time detect of null-exception
2212 if (stopped()) {
2213 return true; // keep the graph constructed so far
2214 }
2215 n = new UModINode(control(), argument(0), argument(1));
2216 break;
2217 }
2218 case vmIntrinsics::_remainderUnsigned_l: {
2219 zero_check_long(argument(2));
2220 // Compile-time detect of null-exception
2221 if (stopped()) {
2222 return true; // keep the graph constructed so far
2223 }
2224 n = new UModLNode(control(), argument(0), argument(2));
2225 break;
2226 }
2227 default: fatal_unexpected_iid(id); break;
2228 }
2229 set_result(_gvn.transform(n));
2230 return true;
2231 }
2232
2233 //----------------------------inline_unsafe_access----------------------------
2234
2235 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2236 // Attempt to infer a sharper value type from the offset and base type.
2237 ciKlass* sharpened_klass = nullptr;
2238
2239 // See if it is an instance field, with an object type.
2240 if (alias_type->field() != nullptr) {
2241 if (alias_type->field()->type()->is_klass()) {
2242 sharpened_klass = alias_type->field()->type()->as_klass();
2243 }
2244 }
2245
2246 const TypeOopPtr* result = nullptr;
2247 // See if it is a narrow oop array.
2248 if (adr_type->isa_aryptr()) {
2249 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2250 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2251 if (elem_type != nullptr && elem_type->is_loaded()) {
2252 // Sharpen the value type.
2253 result = elem_type;
2254 }
2255 }
2256 }
2257
2258 // The sharpened class might be unloaded if there is no class loader
2259 // contraint in place.
2260 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2261 // Sharpen the value type.
2262 result = TypeOopPtr::make_from_klass(sharpened_klass);
2263 }
2264 if (result != nullptr) {
2265 #ifndef PRODUCT
2266 if (C->print_intrinsics() || C->print_inlining()) {
2267 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2268 tty->print(" sharpened value: "); result->dump(); tty->cr();
2269 }
2270 #endif
2271 }
2272 return result;
2273 }
2274
2275 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2276 switch (kind) {
2277 case Relaxed:
2278 return MO_UNORDERED;
2279 case Opaque:
2280 return MO_RELAXED;
2281 case Acquire:
2282 return MO_ACQUIRE;
2283 case Release:
2284 return MO_RELEASE;
2285 case Volatile:
2286 return MO_SEQ_CST;
2287 default:
2288 ShouldNotReachHere();
2289 return 0;
2290 }
2291 }
2292
2293 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2294 if (callee()->is_static()) return false; // caller must have the capability!
2295 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2296 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2297 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2298 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2299
2300 if (is_reference_type(type)) {
2301 decorators |= ON_UNKNOWN_OOP_REF;
2302 }
2303
2304 if (unaligned) {
2305 decorators |= C2_UNALIGNED;
2306 }
2307
2308 #ifndef PRODUCT
2309 {
2310 ResourceMark rm;
2311 // Check the signatures.
2312 ciSignature* sig = callee()->signature();
2313 #ifdef ASSERT
2314 if (!is_store) {
2315 // Object getReference(Object base, int/long offset), etc.
2316 BasicType rtype = sig->return_type()->basic_type();
2317 assert(rtype == type, "getter must return the expected value");
2318 assert(sig->count() == 2, "oop getter has 2 arguments");
2319 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2320 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2321 } else {
2322 // void putReference(Object base, int/long offset, Object x), etc.
2323 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2324 assert(sig->count() == 3, "oop putter has 3 arguments");
2325 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2326 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2327 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2328 assert(vtype == type, "putter must accept the expected value");
2329 }
2330 #endif // ASSERT
2331 }
2332 #endif //PRODUCT
2333
2334 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2335
2336 Node* receiver = argument(0); // type: oop
2337
2338 // Build address expression.
2339 Node* heap_base_oop = top();
2340
2341 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2342 Node* base = argument(1); // type: oop
2343 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2344 Node* offset = argument(2); // type: long
2345 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2346 // to be plain byte offsets, which are also the same as those accepted
2347 // by oopDesc::field_addr.
2348 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2349 "fieldOffset must be byte-scaled");
2350 // 32-bit machines ignore the high half!
2351 offset = ConvL2X(offset);
2352
2353 // Save state and restore on bailout
2354 uint old_sp = sp();
2355 SafePointNode* old_map = clone_map();
2356
2357 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2358 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2359
2360 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2361 if (type != T_OBJECT) {
2362 decorators |= IN_NATIVE; // off-heap primitive access
2363 } else {
2364 set_map(old_map);
2365 set_sp(old_sp);
2366 return false; // off-heap oop accesses are not supported
2367 }
2368 } else {
2369 heap_base_oop = base; // on-heap or mixed access
2370 }
2371
2372 // Can base be null? Otherwise, always on-heap access.
2373 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2374
2375 if (!can_access_non_heap) {
2376 decorators |= IN_HEAP;
2377 }
2378
2379 Node* val = is_store ? argument(4) : nullptr;
2380
2381 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2382 if (adr_type == TypePtr::NULL_PTR) {
2383 set_map(old_map);
2384 set_sp(old_sp);
2385 return false; // off-heap access with zero address
2386 }
2387
2388 // Try to categorize the address.
2389 Compile::AliasType* alias_type = C->alias_type(adr_type);
2390 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2391
2392 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2393 alias_type->adr_type() == TypeAryPtr::RANGE) {
2394 set_map(old_map);
2395 set_sp(old_sp);
2396 return false; // not supported
2397 }
2398
2399 bool mismatched = false;
2400 BasicType bt = alias_type->basic_type();
2401 if (bt != T_ILLEGAL) {
2402 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2403 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2404 // Alias type doesn't differentiate between byte[] and boolean[]).
2405 // Use address type to get the element type.
2406 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2407 }
2408 if (is_reference_type(bt, true)) {
2409 // accessing an array field with getReference is not a mismatch
2410 bt = T_OBJECT;
2411 }
2412 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2413 // Don't intrinsify mismatched object accesses
2414 set_map(old_map);
2415 set_sp(old_sp);
2416 return false;
2417 }
2418 mismatched = (bt != type);
2419 } else if (alias_type->adr_type()->isa_oopptr()) {
2420 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2421 }
2422
2423 destruct_map_clone(old_map);
2424 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2425
2426 if (mismatched) {
2427 decorators |= C2_MISMATCHED;
2428 }
2429
2430 // First guess at the value type.
2431 const Type *value_type = Type::get_const_basic_type(type);
2432
2433 // Figure out the memory ordering.
2434 decorators |= mo_decorator_for_access_kind(kind);
2435
2436 if (!is_store && type == T_OBJECT) {
2437 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2438 if (tjp != nullptr) {
2439 value_type = tjp;
2440 }
2441 }
2442
2443 receiver = null_check(receiver);
2444 if (stopped()) {
2445 return true;
2446 }
2447 // Heap pointers get a null-check from the interpreter,
2448 // as a courtesy. However, this is not guaranteed by Unsafe,
2449 // and it is not possible to fully distinguish unintended nulls
2450 // from intended ones in this API.
2451
2452 if (!is_store) {
2453 Node* p = nullptr;
2454 // Try to constant fold a load from a constant field
2455 ciField* field = alias_type->field();
2456 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) {
2457 // final or stable field
2458 p = make_constant_from_field(field, heap_base_oop);
2459 }
2460
2461 if (p == nullptr) { // Could not constant fold the load
2462 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2463 // Normalize the value returned by getBoolean in the following cases
2464 if (type == T_BOOLEAN &&
2465 (mismatched ||
2466 heap_base_oop == top() || // - heap_base_oop is null or
2467 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2468 // and the unsafe access is made to large offset
2469 // (i.e., larger than the maximum offset necessary for any
2470 // field access)
2471 ) {
2472 IdealKit ideal = IdealKit(this);
2473 #define __ ideal.
2474 IdealVariable normalized_result(ideal);
2475 __ declarations_done();
2476 __ set(normalized_result, p);
2477 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2478 __ set(normalized_result, ideal.ConI(1));
2479 ideal.end_if();
2480 final_sync(ideal);
2481 p = __ value(normalized_result);
2482 #undef __
2483 }
2484 }
2485 if (type == T_ADDRESS) {
2486 p = gvn().transform(new CastP2XNode(nullptr, p));
2487 p = ConvX2UL(p);
2488 }
2489 // The load node has the control of the preceding MemBarCPUOrder. All
2490 // following nodes will have the control of the MemBarCPUOrder inserted at
2491 // the end of this method. So, pushing the load onto the stack at a later
2492 // point is fine.
2493 set_result(p);
2494 } else {
2495 if (bt == T_ADDRESS) {
2496 // Repackage the long as a pointer.
2497 val = ConvL2X(val);
2498 val = gvn().transform(new CastX2PNode(val));
2499 }
2500 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2501 }
2502
2503 return true;
2504 }
2505
2506 //----------------------------inline_unsafe_load_store----------------------------
2507 // This method serves a couple of different customers (depending on LoadStoreKind):
2508 //
2509 // LS_cmp_swap:
2510 //
2511 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2512 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2513 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2514 //
2515 // LS_cmp_swap_weak:
2516 //
2517 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2518 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2519 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2520 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2521 //
2522 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2523 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2524 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2525 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2526 //
2527 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2528 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2529 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2530 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2531 //
2532 // LS_cmp_exchange:
2533 //
2534 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2535 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2536 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2537 //
2538 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2539 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2540 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2541 //
2542 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2543 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2544 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2545 //
2546 // LS_get_add:
2547 //
2548 // int getAndAddInt( Object o, long offset, int delta)
2549 // long getAndAddLong(Object o, long offset, long delta)
2550 //
2551 // LS_get_set:
2552 //
2553 // int getAndSet(Object o, long offset, int newValue)
2554 // long getAndSet(Object o, long offset, long newValue)
2555 // Object getAndSet(Object o, long offset, Object newValue)
2556 //
2557 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2558 // This basic scheme here is the same as inline_unsafe_access, but
2559 // differs in enough details that combining them would make the code
2560 // overly confusing. (This is a true fact! I originally combined
2561 // them, but even I was confused by it!) As much code/comments as
2562 // possible are retained from inline_unsafe_access though to make
2563 // the correspondences clearer. - dl
2564
2565 if (callee()->is_static()) return false; // caller must have the capability!
2566
2567 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2568 decorators |= mo_decorator_for_access_kind(access_kind);
2569
2570 #ifndef PRODUCT
2571 BasicType rtype;
2572 {
2573 ResourceMark rm;
2574 // Check the signatures.
2575 ciSignature* sig = callee()->signature();
2576 rtype = sig->return_type()->basic_type();
2577 switch(kind) {
2578 case LS_get_add:
2579 case LS_get_set: {
2580 // Check the signatures.
2581 #ifdef ASSERT
2582 assert(rtype == type, "get and set must return the expected type");
2583 assert(sig->count() == 3, "get and set has 3 arguments");
2584 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2585 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2586 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2587 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2588 #endif // ASSERT
2589 break;
2590 }
2591 case LS_cmp_swap:
2592 case LS_cmp_swap_weak: {
2593 // Check the signatures.
2594 #ifdef ASSERT
2595 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2596 assert(sig->count() == 4, "CAS has 4 arguments");
2597 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2598 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2599 #endif // ASSERT
2600 break;
2601 }
2602 case LS_cmp_exchange: {
2603 // Check the signatures.
2604 #ifdef ASSERT
2605 assert(rtype == type, "CAS must return the expected type");
2606 assert(sig->count() == 4, "CAS has 4 arguments");
2607 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2608 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2609 #endif // ASSERT
2610 break;
2611 }
2612 default:
2613 ShouldNotReachHere();
2614 }
2615 }
2616 #endif //PRODUCT
2617
2618 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2619
2620 // Get arguments:
2621 Node* receiver = nullptr;
2622 Node* base = nullptr;
2623 Node* offset = nullptr;
2624 Node* oldval = nullptr;
2625 Node* newval = nullptr;
2626 switch(kind) {
2627 case LS_cmp_swap:
2628 case LS_cmp_swap_weak:
2629 case LS_cmp_exchange: {
2630 const bool two_slot_type = type2size[type] == 2;
2631 receiver = argument(0); // type: oop
2632 base = argument(1); // type: oop
2633 offset = argument(2); // type: long
2634 oldval = argument(4); // type: oop, int, or long
2635 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2636 break;
2637 }
2638 case LS_get_add:
2639 case LS_get_set: {
2640 receiver = argument(0); // type: oop
2641 base = argument(1); // type: oop
2642 offset = argument(2); // type: long
2643 oldval = nullptr;
2644 newval = argument(4); // type: oop, int, or long
2645 break;
2646 }
2647 default:
2648 ShouldNotReachHere();
2649 }
2650
2651 // Build field offset expression.
2652 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2653 // to be plain byte offsets, which are also the same as those accepted
2654 // by oopDesc::field_addr.
2655 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2656 // 32-bit machines ignore the high half of long offsets
2657 offset = ConvL2X(offset);
2658 // Save state and restore on bailout
2659 uint old_sp = sp();
2660 SafePointNode* old_map = clone_map();
2661 Node* adr = make_unsafe_address(base, offset,type, false);
2662 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2663
2664 Compile::AliasType* alias_type = C->alias_type(adr_type);
2665 BasicType bt = alias_type->basic_type();
2666 if (bt != T_ILLEGAL &&
2667 (is_reference_type(bt) != (type == T_OBJECT))) {
2668 // Don't intrinsify mismatched object accesses.
2669 set_map(old_map);
2670 set_sp(old_sp);
2671 return false;
2672 }
2673
2674 destruct_map_clone(old_map);
2675
2676 // For CAS, unlike inline_unsafe_access, there seems no point in
2677 // trying to refine types. Just use the coarse types here.
2678 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2679 const Type *value_type = Type::get_const_basic_type(type);
2680
2681 switch (kind) {
2682 case LS_get_set:
2683 case LS_cmp_exchange: {
2684 if (type == T_OBJECT) {
2685 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2686 if (tjp != nullptr) {
2687 value_type = tjp;
2688 }
2689 }
2690 break;
2691 }
2692 case LS_cmp_swap:
2693 case LS_cmp_swap_weak:
2694 case LS_get_add:
2695 break;
2696 default:
2697 ShouldNotReachHere();
2698 }
2699
2700 // Null check receiver.
2701 receiver = null_check(receiver);
2702 if (stopped()) {
2703 return true;
2704 }
2705
2706 int alias_idx = C->get_alias_index(adr_type);
2707
2708 if (is_reference_type(type)) {
2709 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2710
2711 // Transformation of a value which could be null pointer (CastPP #null)
2712 // could be delayed during Parse (for example, in adjust_map_after_if()).
2713 // Execute transformation here to avoid barrier generation in such case.
2714 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2715 newval = _gvn.makecon(TypePtr::NULL_PTR);
2716
2717 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2718 // Refine the value to a null constant, when it is known to be null
2719 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2720 }
2721 }
2722
2723 Node* result = nullptr;
2724 switch (kind) {
2725 case LS_cmp_exchange: {
2726 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2727 oldval, newval, value_type, type, decorators);
2728 break;
2729 }
2730 case LS_cmp_swap_weak:
2731 decorators |= C2_WEAK_CMPXCHG;
2732 case LS_cmp_swap: {
2733 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2734 oldval, newval, value_type, type, decorators);
2735 break;
2736 }
2737 case LS_get_set: {
2738 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2739 newval, value_type, type, decorators);
2740 break;
2741 }
2742 case LS_get_add: {
2743 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2744 newval, value_type, type, decorators);
2745 break;
2746 }
2747 default:
2748 ShouldNotReachHere();
2749 }
2750
2751 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2752 set_result(result);
2753 return true;
2754 }
2755
2756 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2757 // Regardless of form, don't allow previous ld/st to move down,
2758 // then issue acquire, release, or volatile mem_bar.
2759 insert_mem_bar(Op_MemBarCPUOrder);
2760 switch(id) {
2761 case vmIntrinsics::_loadFence:
2762 insert_mem_bar(Op_LoadFence);
2763 return true;
2764 case vmIntrinsics::_storeFence:
2765 insert_mem_bar(Op_StoreFence);
2766 return true;
2767 case vmIntrinsics::_storeStoreFence:
2768 insert_mem_bar(Op_StoreStoreFence);
2769 return true;
2770 case vmIntrinsics::_fullFence:
2771 insert_mem_bar(Op_MemBarVolatile);
2772 return true;
2773 default:
2774 fatal_unexpected_iid(id);
2775 return false;
2776 }
2777 }
2778
2779 bool LibraryCallKit::inline_onspinwait() {
2780 insert_mem_bar(Op_OnSpinWait);
2781 return true;
2782 }
2783
2784 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2785 if (!kls->is_Con()) {
2786 return true;
2787 }
2788 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2789 if (klsptr == nullptr) {
2790 return true;
2791 }
2792 ciInstanceKlass* ik = klsptr->instance_klass();
2793 // don't need a guard for a klass that is already initialized
2794 return !ik->is_initialized();
2795 }
2796
2797 //----------------------------inline_unsafe_writeback0-------------------------
2798 // public native void Unsafe.writeback0(long address)
2799 bool LibraryCallKit::inline_unsafe_writeback0() {
2800 if (!Matcher::has_match_rule(Op_CacheWB)) {
2801 return false;
2802 }
2803 #ifndef PRODUCT
2804 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2805 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2806 ciSignature* sig = callee()->signature();
2807 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2808 #endif
2809 null_check_receiver(); // null-check, then ignore
2810 Node *addr = argument(1);
2811 addr = new CastX2PNode(addr);
2812 addr = _gvn.transform(addr);
2813 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2814 flush = _gvn.transform(flush);
2815 set_memory(flush, TypeRawPtr::BOTTOM);
2816 return true;
2817 }
2818
2819 //----------------------------inline_unsafe_writeback0-------------------------
2820 // public native void Unsafe.writeback0(long address)
2821 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2822 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2823 return false;
2824 }
2825 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2826 return false;
2827 }
2828 #ifndef PRODUCT
2829 assert(Matcher::has_match_rule(Op_CacheWB),
2830 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2831 : "found match rule for CacheWBPostSync but not CacheWB"));
2832
2833 #endif
2834 null_check_receiver(); // null-check, then ignore
2835 Node *sync;
2836 if (is_pre) {
2837 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2838 } else {
2839 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2840 }
2841 sync = _gvn.transform(sync);
2842 set_memory(sync, TypeRawPtr::BOTTOM);
2843 return true;
2844 }
2845
2846 //----------------------------inline_unsafe_allocate---------------------------
2847 // public native Object Unsafe.allocateInstance(Class<?> cls);
2848 bool LibraryCallKit::inline_unsafe_allocate() {
2849
2850 #if INCLUDE_JVMTI
2851 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
2852 return false;
2853 }
2854 #endif //INCLUDE_JVMTI
2855
2856 if (callee()->is_static()) return false; // caller must have the capability!
2857
2858 null_check_receiver(); // null-check, then ignore
2859 Node* cls = null_check(argument(1));
2860 if (stopped()) return true;
2861
2862 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
2863 kls = null_check(kls);
2864 if (stopped()) return true; // argument was like int.class
2865
2866 #if INCLUDE_JVMTI
2867 // Don't try to access new allocated obj in the intrinsic.
2868 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
2869 // Deoptimize and allocate in interpreter instead.
2870 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
2871 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
2872 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
2873 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
2874 {
2875 BuildCutout unless(this, tst, PROB_MAX);
2876 uncommon_trap(Deoptimization::Reason_intrinsic,
2877 Deoptimization::Action_make_not_entrant);
2878 }
2879 if (stopped()) {
2880 return true;
2881 }
2882 #endif //INCLUDE_JVMTI
2883
2884 Node* test = nullptr;
2885 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2886 // Note: The argument might still be an illegal value like
2887 // Serializable.class or Object[].class. The runtime will handle it.
2888 // But we must make an explicit check for initialization.
2889 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2890 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2891 // can generate code to load it as unsigned byte.
2892 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
2893 Node* bits = intcon(InstanceKlass::fully_initialized);
2894 test = _gvn.transform(new SubINode(inst, bits));
2895 // The 'test' is non-zero if we need to take a slow path.
2896 }
2897
2898 Node* obj = new_instance(kls, test);
2899 set_result(obj);
2900 return true;
2901 }
2902
2903 //------------------------inline_native_time_funcs--------------
2904 // inline code for System.currentTimeMillis() and System.nanoTime()
2905 // these have the same type and signature
2906 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2907 const TypeFunc* tf = OptoRuntime::void_long_Type();
2908 const TypePtr* no_memory_effects = nullptr;
2909 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2910 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2911 #ifdef ASSERT
2912 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2913 assert(value_top == top(), "second value must be top");
2914 #endif
2915 set_result(value);
2916 return true;
2917 }
2918
2919
2920 #if INCLUDE_JVMTI
2921
2922 // When notifications are disabled then just update the VTMS transition bit and return.
2923 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
2924 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
2925 if (!DoJVMTIVirtualThreadTransitions) {
2926 return true;
2927 }
2928 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
2929 IdealKit ideal(this);
2930
2931 Node* ONE = ideal.ConI(1);
2932 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
2933 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
2934 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
2935
2936 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
2937 sync_kit(ideal);
2938 // if notifyJvmti enabled then make a call to the given SharedRuntime function
2939 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
2940 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
2941 ideal.sync_kit(this);
2942 } ideal.else_(); {
2943 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
2944 Node* thread = ideal.thread();
2945 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
2946 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
2947
2948 sync_kit(ideal);
2949 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2950 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2951
2952 ideal.sync_kit(this);
2953 } ideal.end_if();
2954 final_sync(ideal);
2955
2956 return true;
2957 }
2958
2959 // Always update the is_disable_suspend bit.
2960 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
2961 if (!DoJVMTIVirtualThreadTransitions) {
2962 return true;
2963 }
2964 IdealKit ideal(this);
2965
2966 {
2967 // unconditionally update the is_disable_suspend bit in current JavaThread
2968 Node* thread = ideal.thread();
2969 Node* arg = _gvn.transform(argument(0)); // argument for notification
2970 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
2971 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
2972
2973 sync_kit(ideal);
2974 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2975 ideal.sync_kit(this);
2976 }
2977 final_sync(ideal);
2978
2979 return true;
2980 }
2981
2982 #endif // INCLUDE_JVMTI
2983
2984 #ifdef JFR_HAVE_INTRINSICS
2985
2986 /**
2987 * if oop->klass != null
2988 * // normal class
2989 * epoch = _epoch_state ? 2 : 1
2990 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
2991 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
2992 * }
2993 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
2994 * else
2995 * // primitive class
2996 * if oop->array_klass != null
2997 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
2998 * else
2999 * id = LAST_TYPE_ID + 1 // void class path
3000 * if (!signaled)
3001 * signaled = true
3002 */
3003 bool LibraryCallKit::inline_native_classID() {
3004 Node* cls = argument(0);
3005
3006 IdealKit ideal(this);
3007 #define __ ideal.
3008 IdealVariable result(ideal); __ declarations_done();
3009 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3010 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3011 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3012
3013
3014 __ if_then(kls, BoolTest::ne, null()); {
3015 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3016 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3017
3018 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3019 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3020 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3021 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3022 mask = _gvn.transform(new OrLNode(mask, epoch));
3023 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3024
3025 float unlikely = PROB_UNLIKELY(0.999);
3026 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3027 sync_kit(ideal);
3028 make_runtime_call(RC_LEAF,
3029 OptoRuntime::class_id_load_barrier_Type(),
3030 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3031 "class id load barrier",
3032 TypePtr::BOTTOM,
3033 kls);
3034 ideal.sync_kit(this);
3035 } __ end_if();
3036
3037 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3038 } __ else_(); {
3039 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3040 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3041 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3042 __ if_then(array_kls, BoolTest::ne, null()); {
3043 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3044 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3045 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3046 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3047 } __ else_(); {
3048 // void class case
3049 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3050 } __ end_if();
3051
3052 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3053 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3054 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3055 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3056 } __ end_if();
3057 } __ end_if();
3058
3059 final_sync(ideal);
3060 set_result(ideal.value(result));
3061 #undef __
3062 return true;
3063 }
3064
3065 //------------------------inline_native_jvm_commit------------------
3066 bool LibraryCallKit::inline_native_jvm_commit() {
3067 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3068
3069 // Save input memory and i_o state.
3070 Node* input_memory_state = reset_memory();
3071 set_all_memory(input_memory_state);
3072 Node* input_io_state = i_o();
3073
3074 // TLS.
3075 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3076 // Jfr java buffer.
3077 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3078 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3079 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3080
3081 // Load the current value of the notified field in the JfrThreadLocal.
3082 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3083 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3084
3085 // Test for notification.
3086 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3087 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3088 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3089
3090 // True branch, is notified.
3091 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3092 set_control(is_notified);
3093
3094 // Reset notified state.
3095 Node* notified_reset_memory = store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3096
3097 // 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.
3098 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3099 // Convert the machine-word to a long.
3100 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3101
3102 // False branch, not notified.
3103 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3104 set_control(not_notified);
3105 set_all_memory(input_memory_state);
3106
3107 // Arg is the next position as a long.
3108 Node* arg = argument(0);
3109 // Convert long to machine-word.
3110 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3111
3112 // Store the next_position to the underlying jfr java buffer.
3113 Node* commit_memory;
3114 #ifdef _LP64
3115 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_LONG, MemNode::release);
3116 #else
3117 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_INT, MemNode::release);
3118 #endif
3119
3120 // 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.
3121 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3122 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3123 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3124
3125 // And flags with lease constant.
3126 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3127
3128 // Branch on lease to conditionalize returning the leased java buffer.
3129 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3130 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3131 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3132
3133 // False branch, not a lease.
3134 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3135
3136 // True branch, is lease.
3137 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3138 set_control(is_lease);
3139
3140 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3141 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3142 OptoRuntime::void_void_Type(),
3143 SharedRuntime::jfr_return_lease(),
3144 "return_lease", TypePtr::BOTTOM);
3145 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3146
3147 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3148 record_for_igvn(lease_compare_rgn);
3149 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3150 record_for_igvn(lease_compare_mem);
3151 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3152 record_for_igvn(lease_compare_io);
3153 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3154 record_for_igvn(lease_result_value);
3155
3156 // Update control and phi nodes.
3157 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3158 lease_compare_rgn->init_req(_false_path, not_lease);
3159
3160 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3161 lease_compare_mem->init_req(_false_path, commit_memory);
3162
3163 lease_compare_io->init_req(_true_path, i_o());
3164 lease_compare_io->init_req(_false_path, input_io_state);
3165
3166 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3167 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3168
3169 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3170 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3171 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3172 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3173
3174 // Update control and phi nodes.
3175 result_rgn->init_req(_true_path, is_notified);
3176 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3177
3178 result_mem->init_req(_true_path, notified_reset_memory);
3179 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3180
3181 result_io->init_req(_true_path, input_io_state);
3182 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3183
3184 result_value->init_req(_true_path, current_pos);
3185 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3186
3187 // Set output state.
3188 set_control(_gvn.transform(result_rgn));
3189 set_all_memory(_gvn.transform(result_mem));
3190 set_i_o(_gvn.transform(result_io));
3191 set_result(result_rgn, result_value);
3192 return true;
3193 }
3194
3195 /*
3196 * The intrinsic is a model of this pseudo-code:
3197 *
3198 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3199 * jobject h_event_writer = tl->java_event_writer();
3200 * if (h_event_writer == nullptr) {
3201 * return nullptr;
3202 * }
3203 * oop threadObj = Thread::threadObj();
3204 * oop vthread = java_lang_Thread::vthread(threadObj);
3205 * traceid tid;
3206 * bool pinVirtualThread;
3207 * bool excluded;
3208 * if (vthread != threadObj) { // i.e. current thread is virtual
3209 * tid = java_lang_Thread::tid(vthread);
3210 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3211 * pinVirtualThread = VMContinuations;
3212 * excluded = vthread_epoch_raw & excluded_mask;
3213 * if (!excluded) {
3214 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3215 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3216 * if (vthread_epoch != current_epoch) {
3217 * write_checkpoint();
3218 * }
3219 * }
3220 * } else {
3221 * tid = java_lang_Thread::tid(threadObj);
3222 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3223 * pinVirtualThread = false;
3224 * excluded = thread_epoch_raw & excluded_mask;
3225 * }
3226 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3227 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3228 * if (tid_in_event_writer != tid) {
3229 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3230 * setField(event_writer, "excluded", excluded);
3231 * setField(event_writer, "threadID", tid);
3232 * }
3233 * return event_writer
3234 */
3235 bool LibraryCallKit::inline_native_getEventWriter() {
3236 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3237
3238 // Save input memory and i_o state.
3239 Node* input_memory_state = reset_memory();
3240 set_all_memory(input_memory_state);
3241 Node* input_io_state = i_o();
3242
3243 // The most significant bit of the u2 is used to denote thread exclusion
3244 Node* excluded_shift = _gvn.intcon(15);
3245 Node* excluded_mask = _gvn.intcon(1 << 15);
3246 // The epoch generation is the range [1-32767]
3247 Node* epoch_mask = _gvn.intcon(32767);
3248
3249 // TLS
3250 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3251
3252 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3253 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3254
3255 // Load the eventwriter jobject handle.
3256 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3257
3258 // Null check the jobject handle.
3259 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3260 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3261 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3262
3263 // False path, jobj is null.
3264 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3265
3266 // True path, jobj is not null.
3267 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3268
3269 set_control(jobj_is_not_null);
3270
3271 // Load the threadObj for the CarrierThread.
3272 Node* threadObj = generate_current_thread(tls_ptr);
3273
3274 // Load the vthread.
3275 Node* vthread = generate_virtual_thread(tls_ptr);
3276
3277 // If vthread != threadObj, this is a virtual thread.
3278 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3279 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3280 IfNode* iff_vthread_not_equal_threadObj =
3281 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3282
3283 // False branch, fallback to threadObj.
3284 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3285 set_control(vthread_equal_threadObj);
3286
3287 // Load the tid field from the vthread object.
3288 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3289
3290 // Load the raw epoch value from the threadObj.
3291 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3292 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3293 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3294 TypeInt::CHAR, T_CHAR,
3295 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3296
3297 // Mask off the excluded information from the epoch.
3298 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3299
3300 // True branch, this is a virtual thread.
3301 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3302 set_control(vthread_not_equal_threadObj);
3303
3304 // Load the tid field from the vthread object.
3305 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3306
3307 // Continuation support determines if a virtual thread should be pinned.
3308 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3309 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3310
3311 // Load the raw epoch value from the vthread.
3312 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3313 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3314 TypeInt::CHAR, T_CHAR,
3315 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3316
3317 // Mask off the excluded information from the epoch.
3318 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3319
3320 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3321 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3322 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3323 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3324
3325 // False branch, vthread is excluded, no need to write epoch info.
3326 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3327
3328 // True branch, vthread is included, update epoch info.
3329 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3330 set_control(included);
3331
3332 // Get epoch value.
3333 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3334
3335 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3336 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3337 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3338
3339 // Compare the epoch in the vthread to the current epoch generation.
3340 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3341 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3342 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3343
3344 // False path, epoch is equal, checkpoint information is valid.
3345 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3346
3347 // True path, epoch is not equal, write a checkpoint for the vthread.
3348 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3349
3350 set_control(epoch_is_not_equal);
3351
3352 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3353 // The call also updates the native thread local thread id and the vthread with the current epoch.
3354 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3355 OptoRuntime::jfr_write_checkpoint_Type(),
3356 SharedRuntime::jfr_write_checkpoint(),
3357 "write_checkpoint", TypePtr::BOTTOM);
3358 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3359
3360 // vthread epoch != current epoch
3361 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3362 record_for_igvn(epoch_compare_rgn);
3363 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3364 record_for_igvn(epoch_compare_mem);
3365 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3366 record_for_igvn(epoch_compare_io);
3367
3368 // Update control and phi nodes.
3369 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3370 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3371 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3372 epoch_compare_mem->init_req(_false_path, input_memory_state);
3373 epoch_compare_io->init_req(_true_path, i_o());
3374 epoch_compare_io->init_req(_false_path, input_io_state);
3375
3376 // excluded != true
3377 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3378 record_for_igvn(exclude_compare_rgn);
3379 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3380 record_for_igvn(exclude_compare_mem);
3381 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3382 record_for_igvn(exclude_compare_io);
3383
3384 // Update control and phi nodes.
3385 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3386 exclude_compare_rgn->init_req(_false_path, excluded);
3387 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3388 exclude_compare_mem->init_req(_false_path, input_memory_state);
3389 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3390 exclude_compare_io->init_req(_false_path, input_io_state);
3391
3392 // vthread != threadObj
3393 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3394 record_for_igvn(vthread_compare_rgn);
3395 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3396 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3397 record_for_igvn(vthread_compare_io);
3398 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3399 record_for_igvn(tid);
3400 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3401 record_for_igvn(exclusion);
3402 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3403 record_for_igvn(pinVirtualThread);
3404
3405 // Update control and phi nodes.
3406 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3407 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3408 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3409 vthread_compare_mem->init_req(_false_path, input_memory_state);
3410 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3411 vthread_compare_io->init_req(_false_path, input_io_state);
3412 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3413 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3414 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3415 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3416 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3417 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3418
3419 // Update branch state.
3420 set_control(_gvn.transform(vthread_compare_rgn));
3421 set_all_memory(_gvn.transform(vthread_compare_mem));
3422 set_i_o(_gvn.transform(vthread_compare_io));
3423
3424 // Load the event writer oop by dereferencing the jobject handle.
3425 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3426 assert(klass_EventWriter->is_loaded(), "invariant");
3427 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3428 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3429 const TypeOopPtr* const xtype = aklass->as_instance_type();
3430 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3431 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3432
3433 // Load the current thread id from the event writer object.
3434 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3435 // Get the field offset to, conditionally, store an updated tid value later.
3436 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3437 // Get the field offset to, conditionally, store an updated exclusion value later.
3438 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3439 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3440 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3441
3442 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3443 record_for_igvn(event_writer_tid_compare_rgn);
3444 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3445 record_for_igvn(event_writer_tid_compare_mem);
3446 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3447 record_for_igvn(event_writer_tid_compare_io);
3448
3449 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3450 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3451 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3452 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3453
3454 // False path, tids are the same.
3455 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3456
3457 // True path, tid is not equal, need to update the tid in the event writer.
3458 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3459 record_for_igvn(tid_is_not_equal);
3460
3461 // Store the pin state to the event writer.
3462 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3463
3464 // Store the exclusion state to the event writer.
3465 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3466 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3467
3468 // Store the tid to the event writer.
3469 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3470
3471 // Update control and phi nodes.
3472 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3473 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3474 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3475 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3476 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3477 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3478
3479 // Result of top level CFG, Memory, IO and Value.
3480 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3481 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3482 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3483 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3484
3485 // Result control.
3486 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3487 result_rgn->init_req(_false_path, jobj_is_null);
3488
3489 // Result memory.
3490 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3491 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3492
3493 // Result IO.
3494 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3495 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3496
3497 // Result value.
3498 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3499 result_value->init_req(_false_path, null()); // return null
3500
3501 // Set output state.
3502 set_control(_gvn.transform(result_rgn));
3503 set_all_memory(_gvn.transform(result_mem));
3504 set_i_o(_gvn.transform(result_io));
3505 set_result(result_rgn, result_value);
3506 return true;
3507 }
3508
3509 /*
3510 * The intrinsic is a model of this pseudo-code:
3511 *
3512 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3513 * if (carrierThread != thread) { // is virtual thread
3514 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3515 * bool excluded = vthread_epoch_raw & excluded_mask;
3516 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3517 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3518 * if (!excluded) {
3519 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3520 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3521 * }
3522 * Atomic::release_store(&tl->_vthread, true);
3523 * return;
3524 * }
3525 * Atomic::release_store(&tl->_vthread, false);
3526 */
3527 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3528 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3529
3530 Node* input_memory_state = reset_memory();
3531 set_all_memory(input_memory_state);
3532
3533 // The most significant bit of the u2 is used to denote thread exclusion
3534 Node* excluded_mask = _gvn.intcon(1 << 15);
3535 // The epoch generation is the range [1-32767]
3536 Node* epoch_mask = _gvn.intcon(32767);
3537
3538 Node* const carrierThread = generate_current_thread(jt);
3539 // If thread != carrierThread, this is a virtual thread.
3540 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3541 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3542 IfNode* iff_thread_not_equal_carrierThread =
3543 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3544
3545 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3546
3547 // False branch, is carrierThread.
3548 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3549 // Store release
3550 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3551
3552 set_all_memory(input_memory_state);
3553
3554 // True branch, is virtual thread.
3555 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3556 set_control(thread_not_equal_carrierThread);
3557
3558 // Load the raw epoch value from the vthread.
3559 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3560 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
3561 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3562
3563 // Mask off the excluded information from the epoch.
3564 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3565
3566 // Load the tid field from the thread.
3567 Node* tid = load_field_from_object(thread, "tid", "J");
3568
3569 // Store the vthread tid to the jfr thread local.
3570 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3571 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
3572
3573 // Branch is_excluded to conditionalize updating the epoch .
3574 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3575 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3576 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3577
3578 // True branch, vthread is excluded, no need to write epoch info.
3579 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3580 set_control(excluded);
3581 Node* vthread_is_excluded = _gvn.intcon(1);
3582
3583 // False branch, vthread is included, update epoch info.
3584 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3585 set_control(included);
3586 Node* vthread_is_included = _gvn.intcon(0);
3587
3588 // Get epoch value.
3589 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3590
3591 // Store the vthread epoch to the jfr thread local.
3592 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3593 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
3594
3595 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3596 record_for_igvn(excluded_rgn);
3597 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3598 record_for_igvn(excluded_mem);
3599 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3600 record_for_igvn(exclusion);
3601
3602 // Merge the excluded control and memory.
3603 excluded_rgn->init_req(_true_path, excluded);
3604 excluded_rgn->init_req(_false_path, included);
3605 excluded_mem->init_req(_true_path, tid_memory);
3606 excluded_mem->init_req(_false_path, included_memory);
3607 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3608 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3609
3610 // Set intermediate state.
3611 set_control(_gvn.transform(excluded_rgn));
3612 set_all_memory(excluded_mem);
3613
3614 // Store the vthread exclusion state to the jfr thread local.
3615 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3616 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
3617
3618 // Store release
3619 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
3620
3621 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3622 record_for_igvn(thread_compare_rgn);
3623 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3624 record_for_igvn(thread_compare_mem);
3625 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3626 record_for_igvn(vthread);
3627
3628 // Merge the thread_compare control and memory.
3629 thread_compare_rgn->init_req(_true_path, control());
3630 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3631 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3632 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3633
3634 // Set output state.
3635 set_control(_gvn.transform(thread_compare_rgn));
3636 set_all_memory(_gvn.transform(thread_compare_mem));
3637 }
3638
3639 #endif // JFR_HAVE_INTRINSICS
3640
3641 //------------------------inline_native_currentCarrierThread------------------
3642 bool LibraryCallKit::inline_native_currentCarrierThread() {
3643 Node* junk = nullptr;
3644 set_result(generate_current_thread(junk));
3645 return true;
3646 }
3647
3648 //------------------------inline_native_currentThread------------------
3649 bool LibraryCallKit::inline_native_currentThread() {
3650 Node* junk = nullptr;
3651 set_result(generate_virtual_thread(junk));
3652 return true;
3653 }
3654
3655 //------------------------inline_native_setVthread------------------
3656 bool LibraryCallKit::inline_native_setCurrentThread() {
3657 assert(C->method()->changes_current_thread(),
3658 "method changes current Thread but is not annotated ChangesCurrentThread");
3659 Node* arr = argument(1);
3660 Node* thread = _gvn.transform(new ThreadLocalNode());
3661 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3662 Node* thread_obj_handle
3663 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3664 thread_obj_handle = _gvn.transform(thread_obj_handle);
3665 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3666 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3667
3668 // Change the _monitor_owner_id of the JavaThread
3669 Node* tid = load_field_from_object(arr, "tid", "J");
3670 Node* monitor_owner_id_offset = basic_plus_adr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
3671 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
3672
3673 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3674 return true;
3675 }
3676
3677 const Type* LibraryCallKit::scopedValueCache_type() {
3678 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3679 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3680 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3681
3682 // Because we create the scopedValue cache lazily we have to make the
3683 // type of the result BotPTR.
3684 bool xk = etype->klass_is_exact();
3685 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3686 return objects_type;
3687 }
3688
3689 Node* LibraryCallKit::scopedValueCache_helper() {
3690 Node* thread = _gvn.transform(new ThreadLocalNode());
3691 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3692 // We cannot use immutable_memory() because we might flip onto a
3693 // different carrier thread, at which point we'll need to use that
3694 // carrier thread's cache.
3695 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3696 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3697 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3698 }
3699
3700 //------------------------inline_native_scopedValueCache------------------
3701 bool LibraryCallKit::inline_native_scopedValueCache() {
3702 Node* cache_obj_handle = scopedValueCache_helper();
3703 const Type* objects_type = scopedValueCache_type();
3704 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3705
3706 return true;
3707 }
3708
3709 //------------------------inline_native_setScopedValueCache------------------
3710 bool LibraryCallKit::inline_native_setScopedValueCache() {
3711 Node* arr = argument(0);
3712 Node* cache_obj_handle = scopedValueCache_helper();
3713 const Type* objects_type = scopedValueCache_type();
3714
3715 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3716 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3717
3718 return true;
3719 }
3720
3721 //------------------------inline_native_Continuation_pin and unpin-----------
3722
3723 // Shared implementation routine for both pin and unpin.
3724 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
3725 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3726
3727 // Save input memory.
3728 Node* input_memory_state = reset_memory();
3729 set_all_memory(input_memory_state);
3730
3731 // TLS
3732 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3733 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
3734 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
3735
3736 // Null check the last continuation object.
3737 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
3738 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
3739 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3740
3741 // False path, last continuation is null.
3742 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
3743
3744 // True path, last continuation is not null.
3745 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
3746
3747 set_control(continuation_is_not_null);
3748
3749 // Load the pin count from the last continuation.
3750 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
3751 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
3752
3753 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
3754 Node* pin_count_rhs;
3755 if (unpin) {
3756 pin_count_rhs = _gvn.intcon(0);
3757 } else {
3758 pin_count_rhs = _gvn.intcon(UINT32_MAX);
3759 }
3760 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
3761 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
3762 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
3763
3764 // True branch, pin count over/underflow.
3765 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
3766 {
3767 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
3768 // which will throw IllegalStateException for pin count over/underflow.
3769 // No memory changed so far - we can use memory create by reset_memory()
3770 // at the beginning of this intrinsic. No need to call reset_memory() again.
3771 PreserveJVMState pjvms(this);
3772 set_control(pin_count_over_underflow);
3773 uncommon_trap(Deoptimization::Reason_intrinsic,
3774 Deoptimization::Action_none);
3775 assert(stopped(), "invariant");
3776 }
3777
3778 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
3779 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
3780 set_control(valid_pin_count);
3781
3782 Node* next_pin_count;
3783 if (unpin) {
3784 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
3785 } else {
3786 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
3787 }
3788
3789 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
3790
3791 // Result of top level CFG and Memory.
3792 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3793 record_for_igvn(result_rgn);
3794 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3795 record_for_igvn(result_mem);
3796
3797 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
3798 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
3799 result_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3800 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3801
3802 // Set output state.
3803 set_control(_gvn.transform(result_rgn));
3804 set_all_memory(_gvn.transform(result_mem));
3805
3806 return true;
3807 }
3808
3809 //---------------------------load_mirror_from_klass----------------------------
3810 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3811 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3812 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3813 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3814 // mirror = ((OopHandle)mirror)->resolve();
3815 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3816 }
3817
3818 //-----------------------load_klass_from_mirror_common-------------------------
3819 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3820 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3821 // and branch to the given path on the region.
3822 // If never_see_null, take an uncommon trap on null, so we can optimistically
3823 // compile for the non-null case.
3824 // If the region is null, force never_see_null = true.
3825 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3826 bool never_see_null,
3827 RegionNode* region,
3828 int null_path,
3829 int offset) {
3830 if (region == nullptr) never_see_null = true;
3831 Node* p = basic_plus_adr(mirror, offset);
3832 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3833 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3834 Node* null_ctl = top();
3835 kls = null_check_oop(kls, &null_ctl, never_see_null);
3836 if (region != nullptr) {
3837 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3838 region->init_req(null_path, null_ctl);
3839 } else {
3840 assert(null_ctl == top(), "no loose ends");
3841 }
3842 return kls;
3843 }
3844
3845 //--------------------(inline_native_Class_query helpers)---------------------
3846 // Use this for JVM_ACC_INTERFACE.
3847 // Fall through if (mods & mask) == bits, take the guard otherwise.
3848 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
3849 ByteSize offset, const Type* type, BasicType bt) {
3850 // Branch around if the given klass has the given modifier bit set.
3851 // Like generate_guard, adds a new path onto the region.
3852 Node* modp = basic_plus_adr(kls, in_bytes(offset));
3853 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
3854 Node* mask = intcon(modifier_mask);
3855 Node* bits = intcon(modifier_bits);
3856 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3857 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3858 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3859 return generate_fair_guard(bol, region);
3860 }
3861 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3862 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
3863 Klass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
3864 }
3865
3866 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
3867 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3868 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
3869 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
3870 }
3871
3872 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3873 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
3874 }
3875
3876 //-------------------------inline_native_Class_query-------------------
3877 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3878 const Type* return_type = TypeInt::BOOL;
3879 Node* prim_return_value = top(); // what happens if it's a primitive class?
3880 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3881 bool expect_prim = false; // most of these guys expect to work on refs
3882
3883 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3884
3885 Node* mirror = argument(0);
3886 Node* obj = top();
3887
3888 switch (id) {
3889 case vmIntrinsics::_isInstance:
3890 // nothing is an instance of a primitive type
3891 prim_return_value = intcon(0);
3892 obj = argument(1);
3893 break;
3894 case vmIntrinsics::_getModifiers:
3895 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3896 return_type = TypeInt::CHAR;
3897 break;
3898 case vmIntrinsics::_isInterface:
3899 prim_return_value = intcon(0);
3900 break;
3901 case vmIntrinsics::_isArray:
3902 prim_return_value = intcon(0);
3903 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3904 break;
3905 case vmIntrinsics::_isPrimitive:
3906 prim_return_value = intcon(1);
3907 expect_prim = true; // obviously
3908 break;
3909 case vmIntrinsics::_isHidden:
3910 prim_return_value = intcon(0);
3911 break;
3912 case vmIntrinsics::_getSuperclass:
3913 prim_return_value = null();
3914 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3915 break;
3916 case vmIntrinsics::_getClassAccessFlags:
3917 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3918 return_type = TypeInt::CHAR;
3919 break;
3920 default:
3921 fatal_unexpected_iid(id);
3922 break;
3923 }
3924
3925 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3926 if (mirror_con == nullptr) return false; // cannot happen?
3927
3928 #ifndef PRODUCT
3929 if (C->print_intrinsics() || C->print_inlining()) {
3930 ciType* k = mirror_con->java_mirror_type();
3931 if (k) {
3932 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3933 k->print_name();
3934 tty->cr();
3935 }
3936 }
3937 #endif
3938
3939 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3940 RegionNode* region = new RegionNode(PATH_LIMIT);
3941 record_for_igvn(region);
3942 PhiNode* phi = new PhiNode(region, return_type);
3943
3944 // The mirror will never be null of Reflection.getClassAccessFlags, however
3945 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3946 // if it is. See bug 4774291.
3947
3948 // For Reflection.getClassAccessFlags(), the null check occurs in
3949 // the wrong place; see inline_unsafe_access(), above, for a similar
3950 // situation.
3951 mirror = null_check(mirror);
3952 // If mirror or obj is dead, only null-path is taken.
3953 if (stopped()) return true;
3954
3955 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3956
3957 // Now load the mirror's klass metaobject, and null-check it.
3958 // Side-effects region with the control path if the klass is null.
3959 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3960 // If kls is null, we have a primitive mirror.
3961 phi->init_req(_prim_path, prim_return_value);
3962 if (stopped()) { set_result(region, phi); return true; }
3963 bool safe_for_replace = (region->in(_prim_path) == top());
3964
3965 Node* p; // handy temp
3966 Node* null_ctl;
3967
3968 // Now that we have the non-null klass, we can perform the real query.
3969 // For constant classes, the query will constant-fold in LoadNode::Value.
3970 Node* query_value = top();
3971 switch (id) {
3972 case vmIntrinsics::_isInstance:
3973 // nothing is an instance of a primitive type
3974 query_value = gen_instanceof(obj, kls, safe_for_replace);
3975 break;
3976
3977 case vmIntrinsics::_getModifiers:
3978 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3979 query_value = make_load(nullptr, p, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3980 break;
3981
3982 case vmIntrinsics::_isInterface:
3983 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3984 if (generate_interface_guard(kls, region) != nullptr)
3985 // A guard was added. If the guard is taken, it was an interface.
3986 phi->add_req(intcon(1));
3987 // If we fall through, it's a plain class.
3988 query_value = intcon(0);
3989 break;
3990
3991 case vmIntrinsics::_isArray:
3992 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3993 if (generate_array_guard(kls, region) != nullptr)
3994 // A guard was added. If the guard is taken, it was an array.
3995 phi->add_req(intcon(1));
3996 // If we fall through, it's a plain class.
3997 query_value = intcon(0);
3998 break;
3999
4000 case vmIntrinsics::_isPrimitive:
4001 query_value = intcon(0); // "normal" path produces false
4002 break;
4003
4004 case vmIntrinsics::_isHidden:
4005 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4006 if (generate_hidden_class_guard(kls, region) != nullptr)
4007 // A guard was added. If the guard is taken, it was an hidden class.
4008 phi->add_req(intcon(1));
4009 // If we fall through, it's a plain class.
4010 query_value = intcon(0);
4011 break;
4012
4013
4014 case vmIntrinsics::_getSuperclass:
4015 // The rules here are somewhat unfortunate, but we can still do better
4016 // with random logic than with a JNI call.
4017 // Interfaces store null or Object as _super, but must report null.
4018 // Arrays store an intermediate super as _super, but must report Object.
4019 // Other types can report the actual _super.
4020 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4021 if (generate_interface_guard(kls, region) != nullptr)
4022 // A guard was added. If the guard is taken, it was an interface.
4023 phi->add_req(null());
4024 if (generate_array_guard(kls, region) != nullptr)
4025 // A guard was added. If the guard is taken, it was an array.
4026 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4027 // If we fall through, it's a plain class. Get its _super.
4028 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4029 kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4030 null_ctl = top();
4031 kls = null_check_oop(kls, &null_ctl);
4032 if (null_ctl != top()) {
4033 // If the guard is taken, Object.superClass is null (both klass and mirror).
4034 region->add_req(null_ctl);
4035 phi ->add_req(null());
4036 }
4037 if (!stopped()) {
4038 query_value = load_mirror_from_klass(kls);
4039 }
4040 break;
4041
4042 case vmIntrinsics::_getClassAccessFlags:
4043 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
4044 query_value = make_load(nullptr, p, TypeInt::CHAR, T_CHAR, MemNode::unordered);
4045 break;
4046
4047 default:
4048 fatal_unexpected_iid(id);
4049 break;
4050 }
4051
4052 // Fall-through is the normal case of a query to a real class.
4053 phi->init_req(1, query_value);
4054 region->init_req(1, control());
4055
4056 C->set_has_split_ifs(true); // Has chance for split-if optimization
4057 set_result(region, phi);
4058 return true;
4059 }
4060
4061 //-------------------------inline_Class_cast-------------------
4062 bool LibraryCallKit::inline_Class_cast() {
4063 Node* mirror = argument(0); // Class
4064 Node* obj = argument(1);
4065 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4066 if (mirror_con == nullptr) {
4067 return false; // dead path (mirror->is_top()).
4068 }
4069 if (obj == nullptr || obj->is_top()) {
4070 return false; // dead path
4071 }
4072 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4073
4074 // First, see if Class.cast() can be folded statically.
4075 // java_mirror_type() returns non-null for compile-time Class constants.
4076 ciType* tm = mirror_con->java_mirror_type();
4077 if (tm != nullptr && tm->is_klass() &&
4078 tp != nullptr) {
4079 if (!tp->is_loaded()) {
4080 // Don't use intrinsic when class is not loaded.
4081 return false;
4082 } else {
4083 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
4084 if (static_res == Compile::SSC_always_true) {
4085 // isInstance() is true - fold the code.
4086 set_result(obj);
4087 return true;
4088 } else if (static_res == Compile::SSC_always_false) {
4089 // Don't use intrinsic, have to throw ClassCastException.
4090 // If the reference is null, the non-intrinsic bytecode will
4091 // be optimized appropriately.
4092 return false;
4093 }
4094 }
4095 }
4096
4097 // Bailout intrinsic and do normal inlining if exception path is frequent.
4098 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4099 return false;
4100 }
4101
4102 // Generate dynamic checks.
4103 // Class.cast() is java implementation of _checkcast bytecode.
4104 // Do checkcast (Parse::do_checkcast()) optimizations here.
4105
4106 mirror = null_check(mirror);
4107 // If mirror is dead, only null-path is taken.
4108 if (stopped()) {
4109 return true;
4110 }
4111
4112 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive).
4113 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
4114 RegionNode* region = new RegionNode(PATH_LIMIT);
4115 record_for_igvn(region);
4116
4117 // Now load the mirror's klass metaobject, and null-check it.
4118 // If kls is null, we have a primitive mirror and
4119 // nothing is an instance of a primitive type.
4120 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4121
4122 Node* res = top();
4123 if (!stopped()) {
4124 Node* bad_type_ctrl = top();
4125 // Do checkcast optimizations.
4126 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4127 region->init_req(_bad_type_path, bad_type_ctrl);
4128 }
4129 if (region->in(_prim_path) != top() ||
4130 region->in(_bad_type_path) != top()) {
4131 // Let Interpreter throw ClassCastException.
4132 PreserveJVMState pjvms(this);
4133 set_control(_gvn.transform(region));
4134 uncommon_trap(Deoptimization::Reason_intrinsic,
4135 Deoptimization::Action_maybe_recompile);
4136 }
4137 if (!stopped()) {
4138 set_result(res);
4139 }
4140 return true;
4141 }
4142
4143
4144 //--------------------------inline_native_subtype_check------------------------
4145 // This intrinsic takes the JNI calls out of the heart of
4146 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4147 bool LibraryCallKit::inline_native_subtype_check() {
4148 // Pull both arguments off the stack.
4149 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4150 args[0] = argument(0);
4151 args[1] = argument(1);
4152 Node* klasses[2]; // corresponding Klasses: superk, subk
4153 klasses[0] = klasses[1] = top();
4154
4155 enum {
4156 // A full decision tree on {superc is prim, subc is prim}:
4157 _prim_0_path = 1, // {P,N} => false
4158 // {P,P} & superc!=subc => false
4159 _prim_same_path, // {P,P} & superc==subc => true
4160 _prim_1_path, // {N,P} => false
4161 _ref_subtype_path, // {N,N} & subtype check wins => true
4162 _both_ref_path, // {N,N} & subtype check loses => false
4163 PATH_LIMIT
4164 };
4165
4166 RegionNode* region = new RegionNode(PATH_LIMIT);
4167 Node* phi = new PhiNode(region, TypeInt::BOOL);
4168 record_for_igvn(region);
4169
4170 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4171 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4172 int class_klass_offset = java_lang_Class::klass_offset();
4173
4174 // First null-check both mirrors and load each mirror's klass metaobject.
4175 int which_arg;
4176 for (which_arg = 0; which_arg <= 1; which_arg++) {
4177 Node* arg = args[which_arg];
4178 arg = null_check(arg);
4179 if (stopped()) break;
4180 args[which_arg] = arg;
4181
4182 Node* p = basic_plus_adr(arg, class_klass_offset);
4183 Node* kls = LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, adr_type, kls_type);
4184 klasses[which_arg] = _gvn.transform(kls);
4185 }
4186
4187 // Having loaded both klasses, test each for null.
4188 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4189 for (which_arg = 0; which_arg <= 1; which_arg++) {
4190 Node* kls = klasses[which_arg];
4191 Node* null_ctl = top();
4192 kls = null_check_oop(kls, &null_ctl, never_see_null);
4193 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
4194 region->init_req(prim_path, null_ctl);
4195 if (stopped()) break;
4196 klasses[which_arg] = kls;
4197 }
4198
4199 if (!stopped()) {
4200 // now we have two reference types, in klasses[0..1]
4201 Node* subk = klasses[1]; // the argument to isAssignableFrom
4202 Node* superk = klasses[0]; // the receiver
4203 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4204 // now we have a successful reference subtype check
4205 region->set_req(_ref_subtype_path, control());
4206 }
4207
4208 // If both operands are primitive (both klasses null), then
4209 // we must return true when they are identical primitives.
4210 // It is convenient to test this after the first null klass check.
4211 set_control(region->in(_prim_0_path)); // go back to first null check
4212 if (!stopped()) {
4213 // Since superc is primitive, make a guard for the superc==subc case.
4214 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4215 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4216 generate_guard(bol_eq, region, PROB_FAIR);
4217 if (region->req() == PATH_LIMIT+1) {
4218 // A guard was added. If the added guard is taken, superc==subc.
4219 region->swap_edges(PATH_LIMIT, _prim_same_path);
4220 region->del_req(PATH_LIMIT);
4221 }
4222 region->set_req(_prim_0_path, control()); // Not equal after all.
4223 }
4224
4225 // these are the only paths that produce 'true':
4226 phi->set_req(_prim_same_path, intcon(1));
4227 phi->set_req(_ref_subtype_path, intcon(1));
4228
4229 // pull together the cases:
4230 assert(region->req() == PATH_LIMIT, "sane region");
4231 for (uint i = 1; i < region->req(); i++) {
4232 Node* ctl = region->in(i);
4233 if (ctl == nullptr || ctl == top()) {
4234 region->set_req(i, top());
4235 phi ->set_req(i, top());
4236 } else if (phi->in(i) == nullptr) {
4237 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4238 }
4239 }
4240
4241 set_control(_gvn.transform(region));
4242 set_result(_gvn.transform(phi));
4243 return true;
4244 }
4245
4246 //---------------------generate_array_guard_common------------------------
4247 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
4248 bool obj_array, bool not_array, Node** obj) {
4249
4250 if (stopped()) {
4251 return nullptr;
4252 }
4253
4254 // If obj_array/non_array==false/false:
4255 // Branch around if the given klass is in fact an array (either obj or prim).
4256 // If obj_array/non_array==false/true:
4257 // Branch around if the given klass is not an array klass of any kind.
4258 // If obj_array/non_array==true/true:
4259 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
4260 // If obj_array/non_array==true/false:
4261 // Branch around if the kls is an oop array (Object[] or subtype)
4262 //
4263 // Like generate_guard, adds a new path onto the region.
4264 jint layout_con = 0;
4265 Node* layout_val = get_layout_helper(kls, layout_con);
4266 if (layout_val == nullptr) {
4267 bool query = (obj_array
4268 ? Klass::layout_helper_is_objArray(layout_con)
4269 : Klass::layout_helper_is_array(layout_con));
4270 if (query == not_array) {
4271 return nullptr; // never a branch
4272 } else { // always a branch
4273 Node* always_branch = control();
4274 if (region != nullptr)
4275 region->add_req(always_branch);
4276 set_control(top());
4277 return always_branch;
4278 }
4279 }
4280 // Now test the correct condition.
4281 jint nval = (obj_array
4282 ? (jint)(Klass::_lh_array_tag_type_value
4283 << Klass::_lh_array_tag_shift)
4284 : Klass::_lh_neutral_value);
4285 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4286 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
4287 // invert the test if we are looking for a non-array
4288 if (not_array) btest = BoolTest(btest).negate();
4289 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4290 Node* ctrl = generate_fair_guard(bol, region);
4291 Node* is_array_ctrl = not_array ? control() : ctrl;
4292 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4293 // Keep track of the fact that 'obj' is an array to prevent
4294 // array specific accesses from floating above the guard.
4295 Node* cast = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4296 // Check for top because in rare cases, the type system can determine that
4297 // the object can't be an array but the layout helper check is not folded.
4298 if (!cast->is_top()) {
4299 *obj = cast;
4300 }
4301 }
4302 return ctrl;
4303 }
4304
4305
4306 //-----------------------inline_native_newArray--------------------------
4307 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4308 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4309 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4310 Node* mirror;
4311 Node* count_val;
4312 if (uninitialized) {
4313 null_check_receiver();
4314 mirror = argument(1);
4315 count_val = argument(2);
4316 } else {
4317 mirror = argument(0);
4318 count_val = argument(1);
4319 }
4320
4321 mirror = null_check(mirror);
4322 // If mirror or obj is dead, only null-path is taken.
4323 if (stopped()) return true;
4324
4325 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4326 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4327 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4328 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4329 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4330
4331 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4332 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4333 result_reg, _slow_path);
4334 Node* normal_ctl = control();
4335 Node* no_array_ctl = result_reg->in(_slow_path);
4336
4337 // Generate code for the slow case. We make a call to newArray().
4338 set_control(no_array_ctl);
4339 if (!stopped()) {
4340 // Either the input type is void.class, or else the
4341 // array klass has not yet been cached. Either the
4342 // ensuing call will throw an exception, or else it
4343 // will cache the array klass for next time.
4344 PreserveJVMState pjvms(this);
4345 CallJavaNode* slow_call = nullptr;
4346 if (uninitialized) {
4347 // Generate optimized virtual call (holder class 'Unsafe' is final)
4348 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4349 } else {
4350 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4351 }
4352 Node* slow_result = set_results_for_java_call(slow_call);
4353 // this->control() comes from set_results_for_java_call
4354 result_reg->set_req(_slow_path, control());
4355 result_val->set_req(_slow_path, slow_result);
4356 result_io ->set_req(_slow_path, i_o());
4357 result_mem->set_req(_slow_path, reset_memory());
4358 }
4359
4360 set_control(normal_ctl);
4361 if (!stopped()) {
4362 // Normal case: The array type has been cached in the java.lang.Class.
4363 // The following call works fine even if the array type is polymorphic.
4364 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4365 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4366 result_reg->init_req(_normal_path, control());
4367 result_val->init_req(_normal_path, obj);
4368 result_io ->init_req(_normal_path, i_o());
4369 result_mem->init_req(_normal_path, reset_memory());
4370
4371 if (uninitialized) {
4372 // Mark the allocation so that zeroing is skipped
4373 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4374 alloc->maybe_set_complete(&_gvn);
4375 }
4376 }
4377
4378 // Return the combined state.
4379 set_i_o( _gvn.transform(result_io) );
4380 set_all_memory( _gvn.transform(result_mem));
4381
4382 C->set_has_split_ifs(true); // Has chance for split-if optimization
4383 set_result(result_reg, result_val);
4384 return true;
4385 }
4386
4387 //----------------------inline_native_getLength--------------------------
4388 // public static native int java.lang.reflect.Array.getLength(Object array);
4389 bool LibraryCallKit::inline_native_getLength() {
4390 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4391
4392 Node* array = null_check(argument(0));
4393 // If array is dead, only null-path is taken.
4394 if (stopped()) return true;
4395
4396 // Deoptimize if it is a non-array.
4397 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
4398
4399 if (non_array != nullptr) {
4400 PreserveJVMState pjvms(this);
4401 set_control(non_array);
4402 uncommon_trap(Deoptimization::Reason_intrinsic,
4403 Deoptimization::Action_maybe_recompile);
4404 }
4405
4406 // If control is dead, only non-array-path is taken.
4407 if (stopped()) return true;
4408
4409 // The works fine even if the array type is polymorphic.
4410 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4411 Node* result = load_array_length(array);
4412
4413 C->set_has_split_ifs(true); // Has chance for split-if optimization
4414 set_result(result);
4415 return true;
4416 }
4417
4418 //------------------------inline_array_copyOf----------------------------
4419 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4420 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4421 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4422 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4423
4424 // Get the arguments.
4425 Node* original = argument(0);
4426 Node* start = is_copyOfRange? argument(1): intcon(0);
4427 Node* end = is_copyOfRange? argument(2): argument(1);
4428 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4429
4430 Node* newcopy = nullptr;
4431
4432 // Set the original stack and the reexecute bit for the interpreter to reexecute
4433 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4434 { PreserveReexecuteState preexecs(this);
4435 jvms()->set_should_reexecute(true);
4436
4437 array_type_mirror = null_check(array_type_mirror);
4438 original = null_check(original);
4439
4440 // Check if a null path was taken unconditionally.
4441 if (stopped()) return true;
4442
4443 Node* orig_length = load_array_length(original);
4444
4445 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4446 klass_node = null_check(klass_node);
4447
4448 RegionNode* bailout = new RegionNode(1);
4449 record_for_igvn(bailout);
4450
4451 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4452 // Bail out if that is so.
4453 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4454 if (not_objArray != nullptr) {
4455 // Improve the klass node's type from the new optimistic assumption:
4456 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4457 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4458 Node* cast = new CastPPNode(control(), klass_node, akls);
4459 klass_node = _gvn.transform(cast);
4460 }
4461
4462 // Bail out if either start or end is negative.
4463 generate_negative_guard(start, bailout, &start);
4464 generate_negative_guard(end, bailout, &end);
4465
4466 Node* length = end;
4467 if (_gvn.type(start) != TypeInt::ZERO) {
4468 length = _gvn.transform(new SubINode(end, start));
4469 }
4470
4471 // Bail out if length is negative (i.e., if start > end).
4472 // Without this the new_array would throw
4473 // NegativeArraySizeException but IllegalArgumentException is what
4474 // should be thrown
4475 generate_negative_guard(length, bailout, &length);
4476
4477 // Bail out if start is larger than the original length
4478 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4479 generate_negative_guard(orig_tail, bailout, &orig_tail);
4480
4481 if (bailout->req() > 1) {
4482 PreserveJVMState pjvms(this);
4483 set_control(_gvn.transform(bailout));
4484 uncommon_trap(Deoptimization::Reason_intrinsic,
4485 Deoptimization::Action_maybe_recompile);
4486 }
4487
4488 if (!stopped()) {
4489 // How many elements will we copy from the original?
4490 // The answer is MinI(orig_tail, length).
4491 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4492
4493 // Generate a direct call to the right arraycopy function(s).
4494 // We know the copy is disjoint but we might not know if the
4495 // oop stores need checking.
4496 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4497 // This will fail a store-check if x contains any non-nulls.
4498
4499 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4500 // loads/stores but it is legal only if we're sure the
4501 // Arrays.copyOf would succeed. So we need all input arguments
4502 // to the copyOf to be validated, including that the copy to the
4503 // new array won't trigger an ArrayStoreException. That subtype
4504 // check can be optimized if we know something on the type of
4505 // the input array from type speculation.
4506 if (_gvn.type(klass_node)->singleton()) {
4507 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4508 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4509
4510 int test = C->static_subtype_check(superk, subk);
4511 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4512 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4513 if (t_original->speculative_type() != nullptr) {
4514 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4515 }
4516 }
4517 }
4518
4519 bool validated = false;
4520 // Reason_class_check rather than Reason_intrinsic because we
4521 // want to intrinsify even if this traps.
4522 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4523 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4524
4525 if (not_subtype_ctrl != top()) {
4526 PreserveJVMState pjvms(this);
4527 set_control(not_subtype_ctrl);
4528 uncommon_trap(Deoptimization::Reason_class_check,
4529 Deoptimization::Action_make_not_entrant);
4530 assert(stopped(), "Should be stopped");
4531 }
4532 validated = true;
4533 }
4534
4535 if (!stopped()) {
4536 newcopy = new_array(klass_node, length, 0); // no arguments to push
4537
4538 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4539 load_object_klass(original), klass_node);
4540 if (!is_copyOfRange) {
4541 ac->set_copyof(validated);
4542 } else {
4543 ac->set_copyofrange(validated);
4544 }
4545 Node* n = _gvn.transform(ac);
4546 if (n == ac) {
4547 ac->connect_outputs(this);
4548 } else {
4549 assert(validated, "shouldn't transform if all arguments not validated");
4550 set_all_memory(n);
4551 }
4552 }
4553 }
4554 } // original reexecute is set back here
4555
4556 C->set_has_split_ifs(true); // Has chance for split-if optimization
4557 if (!stopped()) {
4558 set_result(newcopy);
4559 }
4560 return true;
4561 }
4562
4563
4564 //----------------------generate_virtual_guard---------------------------
4565 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4566 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4567 RegionNode* slow_region) {
4568 ciMethod* method = callee();
4569 int vtable_index = method->vtable_index();
4570 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4571 "bad index %d", vtable_index);
4572 // Get the Method* out of the appropriate vtable entry.
4573 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4574 vtable_index*vtableEntry::size_in_bytes() +
4575 in_bytes(vtableEntry::method_offset());
4576 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4577 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4578
4579 // Compare the target method with the expected method (e.g., Object.hashCode).
4580 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4581
4582 Node* native_call = makecon(native_call_addr);
4583 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4584 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4585
4586 return generate_slow_guard(test_native, slow_region);
4587 }
4588
4589 //-----------------------generate_method_call----------------------------
4590 // Use generate_method_call to make a slow-call to the real
4591 // method if the fast path fails. An alternative would be to
4592 // use a stub like OptoRuntime::slow_arraycopy_Java.
4593 // This only works for expanding the current library call,
4594 // not another intrinsic. (E.g., don't use this for making an
4595 // arraycopy call inside of the copyOf intrinsic.)
4596 CallJavaNode*
4597 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4598 // When compiling the intrinsic method itself, do not use this technique.
4599 guarantee(callee() != C->method(), "cannot make slow-call to self");
4600
4601 ciMethod* method = callee();
4602 // ensure the JVMS we have will be correct for this call
4603 guarantee(method_id == method->intrinsic_id(), "must match");
4604
4605 const TypeFunc* tf = TypeFunc::make(method);
4606 if (res_not_null) {
4607 assert(tf->return_type() == T_OBJECT, "");
4608 const TypeTuple* range = tf->range();
4609 const Type** fields = TypeTuple::fields(range->cnt());
4610 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4611 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4612 tf = TypeFunc::make(tf->domain(), new_range);
4613 }
4614 CallJavaNode* slow_call;
4615 if (is_static) {
4616 assert(!is_virtual, "");
4617 slow_call = new CallStaticJavaNode(C, tf,
4618 SharedRuntime::get_resolve_static_call_stub(), method);
4619 } else if (is_virtual) {
4620 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4621 int vtable_index = Method::invalid_vtable_index;
4622 if (UseInlineCaches) {
4623 // Suppress the vtable call
4624 } else {
4625 // hashCode and clone are not a miranda methods,
4626 // so the vtable index is fixed.
4627 // No need to use the linkResolver to get it.
4628 vtable_index = method->vtable_index();
4629 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4630 "bad index %d", vtable_index);
4631 }
4632 slow_call = new CallDynamicJavaNode(tf,
4633 SharedRuntime::get_resolve_virtual_call_stub(),
4634 method, vtable_index);
4635 } else { // neither virtual nor static: opt_virtual
4636 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4637 slow_call = new CallStaticJavaNode(C, tf,
4638 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4639 slow_call->set_optimized_virtual(true);
4640 }
4641 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4642 // To be able to issue a direct call (optimized virtual or virtual)
4643 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4644 // about the method being invoked should be attached to the call site to
4645 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4646 slow_call->set_override_symbolic_info(true);
4647 }
4648 set_arguments_for_java_call(slow_call);
4649 set_edges_for_java_call(slow_call);
4650 return slow_call;
4651 }
4652
4653
4654 /**
4655 * Build special case code for calls to hashCode on an object. This call may
4656 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4657 * slightly different code.
4658 */
4659 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4660 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4661 assert(!(is_virtual && is_static), "either virtual, special, or static");
4662
4663 enum { _slow_path = 1, _null_path, _fast_path, _fast_path2, PATH_LIMIT };
4664
4665 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4666 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4667 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4668 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4669 Node* obj = nullptr;
4670 if (!is_static) {
4671 // Check for hashing null object
4672 obj = null_check_receiver();
4673 if (stopped()) return true; // unconditionally null
4674 result_reg->init_req(_null_path, top());
4675 result_val->init_req(_null_path, top());
4676 } else {
4677 // Do a null check, and return zero if null.
4678 // System.identityHashCode(null) == 0
4679 obj = argument(0);
4680 Node* null_ctl = top();
4681 obj = null_check_oop(obj, &null_ctl);
4682 result_reg->init_req(_null_path, null_ctl);
4683 result_val->init_req(_null_path, _gvn.intcon(0));
4684 }
4685
4686 // Unconditionally null? Then return right away.
4687 if (stopped()) {
4688 set_control( result_reg->in(_null_path));
4689 if (!stopped())
4690 set_result(result_val->in(_null_path));
4691 return true;
4692 }
4693
4694 // We only go to the fast case code if we pass a number of guards. The
4695 // paths which do not pass are accumulated in the slow_region.
4696 RegionNode* slow_region = new RegionNode(1);
4697 record_for_igvn(slow_region);
4698
4699 // If this is a virtual call, we generate a funny guard. We pull out
4700 // the vtable entry corresponding to hashCode() from the target object.
4701 // If the target method which we are calling happens to be the native
4702 // Object hashCode() method, we pass the guard. We do not need this
4703 // guard for non-virtual calls -- the caller is known to be the native
4704 // Object hashCode().
4705 if (is_virtual) {
4706 // After null check, get the object's klass.
4707 Node* obj_klass = load_object_klass(obj);
4708 generate_virtual_guard(obj_klass, slow_region);
4709 }
4710
4711 if (UseCompactObjectHeaders) {
4712 // Get the header out of the object.
4713 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4714 // The control of the load must be null. Otherwise, the load can move before
4715 // the null check after castPP removal.
4716 Node* no_ctrl = nullptr;
4717 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4718
4719 // Test the header to see if the object is in hashed or copied state.
4720 Node* hashctrl_mask = _gvn.MakeConX(markWord::hashctrl_mask_in_place);
4721 Node* masked_header = _gvn.transform(new AndXNode(header, hashctrl_mask));
4722
4723 // Take slow-path when the object has not been hashed.
4724 Node* not_hashed_val = _gvn.MakeConX(0);
4725 Node* chk_hashed = _gvn.transform(new CmpXNode(masked_header, not_hashed_val));
4726 Node* test_hashed = _gvn.transform(new BoolNode(chk_hashed, BoolTest::eq));
4727
4728 generate_slow_guard(test_hashed, slow_region);
4729
4730 // Test whether the object is hashed or hashed&copied.
4731 Node* hashed_copied = _gvn.MakeConX(markWord::hashctrl_expanded_mask_in_place | markWord::hashctrl_hashed_mask_in_place);
4732 Node* chk_copied = _gvn.transform(new CmpXNode(masked_header, hashed_copied));
4733 // If true, then object has been hashed&copied, otherwise it's only hashed.
4734 Node* test_copied = _gvn.transform(new BoolNode(chk_copied, BoolTest::eq));
4735 IfNode* if_copied = create_and_map_if(control(), test_copied, PROB_FAIR, COUNT_UNKNOWN);
4736 Node* if_true = _gvn.transform(new IfTrueNode(if_copied));
4737 Node* if_false = _gvn.transform(new IfFalseNode(if_copied));
4738
4739 // Hashed&Copied path: read hash-code out of the object.
4740 set_control(if_true);
4741 // result_val->del_req(_fast_path2);
4742 // result_reg->del_req(_fast_path2);
4743 // result_io->del_req(_fast_path2);
4744 // result_mem->del_req(_fast_path2);
4745
4746 Node* obj_klass = load_object_klass(obj);
4747 Node* hash_addr;
4748 const TypeKlassPtr* klass_t = _gvn.type(obj_klass)->isa_klassptr();
4749 bool load_offset_runtime = true;
4750
4751 if (klass_t != nullptr) {
4752 if (klass_t->klass_is_exact() && klass_t->isa_instklassptr()) {
4753 ciInstanceKlass* ciKlass = reinterpret_cast<ciInstanceKlass*>(klass_t->is_instklassptr()->exact_klass());
4754 if (!ciKlass->is_mirror_instance_klass() && !ciKlass->is_reference_instance_klass()) {
4755 // We know the InstanceKlass, load hash_offset from there at compile-time.
4756 int hash_offset = ciKlass->hash_offset_in_bytes();
4757 hash_addr = basic_plus_adr(obj, hash_offset);
4758 Node* loaded_hash = make_load(control(), hash_addr, TypeInt::INT, T_INT, MemNode::unordered);
4759 result_val->init_req(_fast_path2, loaded_hash);
4760 result_reg->init_req(_fast_path2, control());
4761 load_offset_runtime = false;
4762 }
4763 }
4764 }
4765
4766 //tty->print_cr("Load hash-offset at runtime: %s", BOOL_TO_STR(load_offset_runtime));
4767
4768 if (load_offset_runtime) {
4769 // We don't know if it is an array or an exact type, figure it out at run-time.
4770 // If not an ordinary instance, then we need to take slow-path.
4771 Node* kind_addr = basic_plus_adr(obj_klass, Klass::kind_offset_in_bytes());
4772 Node* kind = make_load(control(), kind_addr, TypeInt::INT, T_INT, MemNode::unordered);
4773 Node* instance_val = _gvn.intcon(Klass::InstanceKlassKind);
4774 Node* chk_inst = _gvn.transform(new CmpINode(kind, instance_val));
4775 Node* test_inst = _gvn.transform(new BoolNode(chk_inst, BoolTest::ne));
4776 generate_slow_guard(test_inst, slow_region);
4777
4778 // Otherwise it's an instance and we can read the hash_offset from the InstanceKlass.
4779 Node* hash_offset_addr = basic_plus_adr(obj_klass, InstanceKlass::hash_offset_offset_in_bytes());
4780 Node* hash_offset = make_load(control(), hash_offset_addr, TypeInt::INT, T_INT, MemNode::unordered);
4781 // hash_offset->dump();
4782 Node* hash_addr = basic_plus_adr(obj, ConvI2X(hash_offset));
4783 Compile::current()->set_has_unsafe_access(true);
4784 Node* loaded_hash = make_load(control(), hash_addr, TypeInt::INT, T_INT, MemNode::unordered);
4785 result_val->init_req(_fast_path2, loaded_hash);
4786 result_reg->init_req(_fast_path2, control());
4787 }
4788
4789 // Hashed-only path: recompute hash-code from object address.
4790 set_control(if_false);
4791 // Our constants.
4792 Node* M = _gvn.intcon(0x337954D5);
4793 Node* A = _gvn.intcon(0xAAAAAAAA);
4794 // Split object address into lo and hi 32 bits.
4795 Node* obj_addr = _gvn.transform(new CastP2XNode(nullptr, obj));
4796 Node* x = _gvn.transform(new ConvL2INode(obj_addr));
4797 Node* upper_addr = _gvn.transform(new URShiftLNode(obj_addr, _gvn.intcon(32)));
4798 Node* y = _gvn.transform(new ConvL2INode(upper_addr));
4799
4800 Node* H0 = _gvn.transform(new XorINode(x, y));
4801 Node* L0 = _gvn.transform(new XorINode(x, A));
4802
4803 // Full multiplication of two 32 bit values L0 and M into a hi/lo result in two 32 bit values V0 and U0.
4804 Node* L0_64 = _gvn.transform(new ConvI2LNode(L0));
4805 L0_64 = _gvn.transform(new AndLNode(L0_64, _gvn.longcon(0xFFFFFFFF)));
4806 Node* M_64 = _gvn.transform(new ConvI2LNode(M));
4807 // M_64 = _gvn.transform(new AndLNode(M_64, _gvn.longcon(0xFFFFFFFF)));
4808 Node* prod64 = _gvn.transform(new MulLNode(L0_64, M_64));
4809 Node* V0 = _gvn.transform(new ConvL2INode(prod64));
4810 Node* prod_upper = _gvn.transform(new URShiftLNode(prod64, _gvn.intcon(32)));
4811 Node* U0 = _gvn.transform(new ConvL2INode(prod_upper));
4812
4813 Node* Q0 = _gvn.transform(new MulINode(H0, M));
4814 Node* L1 = _gvn.transform(new XorINode(Q0, U0));
4815
4816 // Full multiplication of two 32 bit values L1 and M into a hi/lo result in two 32 bit values V1 and U1.
4817 Node* L1_64 = _gvn.transform(new ConvI2LNode(L1));
4818 L1_64 = _gvn.transform(new AndLNode(L1_64, _gvn.longcon(0xFFFFFFFF)));
4819 prod64 = _gvn.transform(new MulLNode(L1_64, M_64));
4820 Node* V1 = _gvn.transform(new ConvL2INode(prod64));
4821 prod_upper = _gvn.transform(new URShiftLNode(prod64, _gvn.intcon(32)));
4822 Node* U1 = _gvn.transform(new ConvL2INode(prod_upper));
4823
4824 Node* P1 = _gvn.transform(new XorINode(V0, M));
4825
4826 // Right rotate P1 by distance L1.
4827 Node* distance = _gvn.transform(new AndINode(L1, _gvn.intcon(32 - 1)));
4828 Node* inverse_distance = _gvn.transform(new SubINode(_gvn.intcon(32), distance));
4829 Node* ror_part1 = _gvn.transform(new URShiftINode(P1, distance));
4830 Node* ror_part2 = _gvn.transform(new LShiftINode(P1, inverse_distance));
4831 Node* Q1 = _gvn.transform(new OrINode(ror_part1, ror_part2));
4832
4833 Node* L2 = _gvn.transform(new XorINode(Q1, U1));
4834 Node* hash = _gvn.transform(new XorINode(V1, L2));
4835 Node* hash_truncated = _gvn.transform(new AndINode(hash, _gvn.intcon(markWord::hash_mask)));
4836
4837 // TODO: We could generate a fast case here under the following conditions:
4838 // - The hashctrl is set to hash_is_copied (see markWord::hash_is_copied())
4839 // - The type of the object is known
4840 // Then we can load the identity hashcode from the int field at Klass::hash_offset_in_bytes() of the object.
4841 result_val->init_req(_fast_path, hash_truncated);
4842 } else {
4843 // Get the header out of the object, use LoadMarkNode when available
4844 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4845 // The control of the load must be null. Otherwise, the load can move before
4846 // the null check after castPP removal.
4847 Node* no_ctrl = nullptr;
4848 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4849
4850 if (!UseObjectMonitorTable) {
4851 // Test the header to see if it is safe to read w.r.t. locking.
4852 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4853 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4854 if (LockingMode == LM_LIGHTWEIGHT) {
4855 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
4856 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
4857 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
4858
4859 generate_slow_guard(test_monitor, slow_region);
4860 } else {
4861 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4862 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val));
4863 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne));
4864
4865 generate_slow_guard(test_not_unlocked, slow_region);
4866 }
4867 }
4868
4869 // Get the hash value and check to see that it has been properly assigned.
4870 // We depend on hash_mask being at most 32 bits and avoid the use of
4871 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4872 // vm: see markWord.hpp.
4873 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4874 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4875 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4876 // This hack lets the hash bits live anywhere in the mark object now, as long
4877 // as the shift drops the relevant bits into the low 32 bits. Note that
4878 // Java spec says that HashCode is an int so there's no point in capturing
4879 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4880 hshifted_header = ConvX2I(hshifted_header);
4881 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4882
4883 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4884 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4885 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4886
4887 generate_slow_guard(test_assigned, slow_region);
4888
4889 result_val->init_req(_fast_path, hash_val);
4890
4891 // _fast_path2 is not used here.
4892 result_val->del_req(_fast_path2);
4893 result_reg->del_req(_fast_path2);
4894 result_io->del_req(_fast_path2);
4895 result_mem->del_req(_fast_path2);
4896 }
4897
4898 Node* init_mem = reset_memory();
4899 // fill in the rest of the null path:
4900 result_io ->init_req(_null_path, i_o());
4901 result_mem->init_req(_null_path, init_mem);
4902
4903 result_reg->init_req(_fast_path, control());
4904 result_io ->init_req(_fast_path, i_o());
4905 result_mem->init_req(_fast_path, init_mem);
4906
4907 if (UseCompactObjectHeaders) {
4908 result_io->init_req(_fast_path2, i_o());
4909 result_mem->init_req(_fast_path2, init_mem);
4910 }
4911
4912 // Generate code for the slow case. We make a call to hashCode().
4913 assert(slow_region != nullptr, "must have slow_region");
4914 set_control(_gvn.transform(slow_region));
4915 if (!stopped()) {
4916 // No need for PreserveJVMState, because we're using up the present state.
4917 set_all_memory(init_mem);
4918 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4919 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4920 Node* slow_result = set_results_for_java_call(slow_call);
4921 // this->control() comes from set_results_for_java_call
4922 result_reg->init_req(_slow_path, control());
4923 result_val->init_req(_slow_path, slow_result);
4924 result_io ->set_req(_slow_path, i_o());
4925 result_mem ->set_req(_slow_path, reset_memory());
4926 }
4927
4928 // Return the combined state.
4929 set_i_o( _gvn.transform(result_io) );
4930 set_all_memory( _gvn.transform(result_mem));
4931
4932 set_result(result_reg, result_val);
4933 return true;
4934 }
4935
4936 //---------------------------inline_native_getClass----------------------------
4937 // public final native Class<?> java.lang.Object.getClass();
4938 //
4939 // Build special case code for calls to getClass on an object.
4940 bool LibraryCallKit::inline_native_getClass() {
4941 Node* obj = null_check_receiver();
4942 if (stopped()) return true;
4943 set_result(load_mirror_from_klass(load_object_klass(obj)));
4944 return true;
4945 }
4946
4947 //-----------------inline_native_Reflection_getCallerClass---------------------
4948 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4949 //
4950 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4951 //
4952 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4953 // in that it must skip particular security frames and checks for
4954 // caller sensitive methods.
4955 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4956 #ifndef PRODUCT
4957 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4958 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4959 }
4960 #endif
4961
4962 if (!jvms()->has_method()) {
4963 #ifndef PRODUCT
4964 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4965 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4966 }
4967 #endif
4968 return false;
4969 }
4970
4971 // Walk back up the JVM state to find the caller at the required
4972 // depth.
4973 JVMState* caller_jvms = jvms();
4974
4975 // Cf. JVM_GetCallerClass
4976 // NOTE: Start the loop at depth 1 because the current JVM state does
4977 // not include the Reflection.getCallerClass() frame.
4978 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4979 ciMethod* m = caller_jvms->method();
4980 switch (n) {
4981 case 0:
4982 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4983 break;
4984 case 1:
4985 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4986 if (!m->caller_sensitive()) {
4987 #ifndef PRODUCT
4988 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4989 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4990 }
4991 #endif
4992 return false; // bail-out; let JVM_GetCallerClass do the work
4993 }
4994 break;
4995 default:
4996 if (!m->is_ignored_by_security_stack_walk()) {
4997 // We have reached the desired frame; return the holder class.
4998 // Acquire method holder as java.lang.Class and push as constant.
4999 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
5000 ciInstance* caller_mirror = caller_klass->java_mirror();
5001 set_result(makecon(TypeInstPtr::make(caller_mirror)));
5002
5003 #ifndef PRODUCT
5004 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5005 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());
5006 tty->print_cr(" JVM state at this point:");
5007 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5008 ciMethod* m = jvms()->of_depth(i)->method();
5009 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5010 }
5011 }
5012 #endif
5013 return true;
5014 }
5015 break;
5016 }
5017 }
5018
5019 #ifndef PRODUCT
5020 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5021 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
5022 tty->print_cr(" JVM state at this point:");
5023 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5024 ciMethod* m = jvms()->of_depth(i)->method();
5025 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5026 }
5027 }
5028 #endif
5029
5030 return false; // bail-out; let JVM_GetCallerClass do the work
5031 }
5032
5033 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
5034 Node* arg = argument(0);
5035 Node* result = nullptr;
5036
5037 switch (id) {
5038 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
5039 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
5040 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5041 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5042 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5043 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5044
5045 case vmIntrinsics::_doubleToLongBits: {
5046 // two paths (plus control) merge in a wood
5047 RegionNode *r = new RegionNode(3);
5048 Node *phi = new PhiNode(r, TypeLong::LONG);
5049
5050 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5051 // Build the boolean node
5052 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5053
5054 // Branch either way.
5055 // NaN case is less traveled, which makes all the difference.
5056 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5057 Node *opt_isnan = _gvn.transform(ifisnan);
5058 assert( opt_isnan->is_If(), "Expect an IfNode");
5059 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5060 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5061
5062 set_control(iftrue);
5063
5064 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5065 Node *slow_result = longcon(nan_bits); // return NaN
5066 phi->init_req(1, _gvn.transform( slow_result ));
5067 r->init_req(1, iftrue);
5068
5069 // Else fall through
5070 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5071 set_control(iffalse);
5072
5073 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5074 r->init_req(2, iffalse);
5075
5076 // Post merge
5077 set_control(_gvn.transform(r));
5078 record_for_igvn(r);
5079
5080 C->set_has_split_ifs(true); // Has chance for split-if optimization
5081 result = phi;
5082 assert(result->bottom_type()->isa_long(), "must be");
5083 break;
5084 }
5085
5086 case vmIntrinsics::_floatToIntBits: {
5087 // two paths (plus control) merge in a wood
5088 RegionNode *r = new RegionNode(3);
5089 Node *phi = new PhiNode(r, TypeInt::INT);
5090
5091 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5092 // Build the boolean node
5093 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5094
5095 // Branch either way.
5096 // NaN case is less traveled, which makes all the difference.
5097 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5098 Node *opt_isnan = _gvn.transform(ifisnan);
5099 assert( opt_isnan->is_If(), "Expect an IfNode");
5100 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5101 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5102
5103 set_control(iftrue);
5104
5105 static const jint nan_bits = 0x7fc00000;
5106 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5107 phi->init_req(1, _gvn.transform( slow_result ));
5108 r->init_req(1, iftrue);
5109
5110 // Else fall through
5111 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5112 set_control(iffalse);
5113
5114 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5115 r->init_req(2, iffalse);
5116
5117 // Post merge
5118 set_control(_gvn.transform(r));
5119 record_for_igvn(r);
5120
5121 C->set_has_split_ifs(true); // Has chance for split-if optimization
5122 result = phi;
5123 assert(result->bottom_type()->isa_int(), "must be");
5124 break;
5125 }
5126
5127 default:
5128 fatal_unexpected_iid(id);
5129 break;
5130 }
5131 set_result(_gvn.transform(result));
5132 return true;
5133 }
5134
5135 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5136 Node* arg = argument(0);
5137 Node* result = nullptr;
5138
5139 switch (id) {
5140 case vmIntrinsics::_floatIsInfinite:
5141 result = new IsInfiniteFNode(arg);
5142 break;
5143 case vmIntrinsics::_floatIsFinite:
5144 result = new IsFiniteFNode(arg);
5145 break;
5146 case vmIntrinsics::_doubleIsInfinite:
5147 result = new IsInfiniteDNode(arg);
5148 break;
5149 case vmIntrinsics::_doubleIsFinite:
5150 result = new IsFiniteDNode(arg);
5151 break;
5152 default:
5153 fatal_unexpected_iid(id);
5154 break;
5155 }
5156 set_result(_gvn.transform(result));
5157 return true;
5158 }
5159
5160 //----------------------inline_unsafe_copyMemory-------------------------
5161 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5162
5163 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5164 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5165 const Type* base_t = gvn.type(base);
5166
5167 bool in_native = (base_t == TypePtr::NULL_PTR);
5168 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5169 bool is_mixed = !in_heap && !in_native;
5170
5171 if (is_mixed) {
5172 return true; // mixed accesses can touch both on-heap and off-heap memory
5173 }
5174 if (in_heap) {
5175 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5176 if (!is_prim_array) {
5177 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5178 // there's not enough type information available to determine proper memory slice for it.
5179 return true;
5180 }
5181 }
5182 return false;
5183 }
5184
5185 bool LibraryCallKit::inline_unsafe_copyMemory() {
5186 if (callee()->is_static()) return false; // caller must have the capability!
5187 null_check_receiver(); // null-check receiver
5188 if (stopped()) return true;
5189
5190 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5191
5192 Node* src_base = argument(1); // type: oop
5193 Node* src_off = ConvL2X(argument(2)); // type: long
5194 Node* dst_base = argument(4); // type: oop
5195 Node* dst_off = ConvL2X(argument(5)); // type: long
5196 Node* size = ConvL2X(argument(7)); // type: long
5197
5198 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5199 "fieldOffset must be byte-scaled");
5200
5201 Node* src_addr = make_unsafe_address(src_base, src_off);
5202 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5203
5204 Node* thread = _gvn.transform(new ThreadLocalNode());
5205 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5206 BasicType doing_unsafe_access_bt = T_BYTE;
5207 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5208
5209 // update volatile field
5210 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5211
5212 int flags = RC_LEAF | RC_NO_FP;
5213
5214 const TypePtr* dst_type = TypePtr::BOTTOM;
5215
5216 // Adjust memory effects of the runtime call based on input values.
5217 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5218 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5219 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5220
5221 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5222 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5223 flags |= RC_NARROW_MEM; // narrow in memory
5224 }
5225 }
5226
5227 // Call it. Note that the length argument is not scaled.
5228 make_runtime_call(flags,
5229 OptoRuntime::fast_arraycopy_Type(),
5230 StubRoutines::unsafe_arraycopy(),
5231 "unsafe_arraycopy",
5232 dst_type,
5233 src_addr, dst_addr, size XTOP);
5234
5235 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5236
5237 return true;
5238 }
5239
5240 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5241 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5242 bool LibraryCallKit::inline_unsafe_setMemory() {
5243 if (callee()->is_static()) return false; // caller must have the capability!
5244 null_check_receiver(); // null-check receiver
5245 if (stopped()) return true;
5246
5247 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5248
5249 Node* dst_base = argument(1); // type: oop
5250 Node* dst_off = ConvL2X(argument(2)); // type: long
5251 Node* size = ConvL2X(argument(4)); // type: long
5252 Node* byte = argument(6); // type: byte
5253
5254 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5255 "fieldOffset must be byte-scaled");
5256
5257 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5258
5259 Node* thread = _gvn.transform(new ThreadLocalNode());
5260 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5261 BasicType doing_unsafe_access_bt = T_BYTE;
5262 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5263
5264 // update volatile field
5265 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5266
5267 int flags = RC_LEAF | RC_NO_FP;
5268
5269 const TypePtr* dst_type = TypePtr::BOTTOM;
5270
5271 // Adjust memory effects of the runtime call based on input values.
5272 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5273 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5274
5275 flags |= RC_NARROW_MEM; // narrow in memory
5276 }
5277
5278 // Call it. Note that the length argument is not scaled.
5279 make_runtime_call(flags,
5280 OptoRuntime::unsafe_setmemory_Type(),
5281 StubRoutines::unsafe_setmemory(),
5282 "unsafe_setmemory",
5283 dst_type,
5284 dst_addr, size XTOP, byte);
5285
5286 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5287
5288 return true;
5289 }
5290
5291 #undef XTOP
5292
5293 //------------------------clone_coping-----------------------------------
5294 // Helper function for inline_native_clone.
5295 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5296 assert(obj_size != nullptr, "");
5297 Node* raw_obj = alloc_obj->in(1);
5298 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5299
5300 AllocateNode* alloc = nullptr;
5301 if (ReduceBulkZeroing &&
5302 // If we are implementing an array clone without knowing its source type
5303 // (can happen when compiling the array-guarded branch of a reflective
5304 // Object.clone() invocation), initialize the array within the allocation.
5305 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5306 // to a runtime clone call that assumes fully initialized source arrays.
5307 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5308 // We will be completely responsible for initializing this object -
5309 // mark Initialize node as complete.
5310 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5311 // The object was just allocated - there should be no any stores!
5312 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5313 // Mark as complete_with_arraycopy so that on AllocateNode
5314 // expansion, we know this AllocateNode is initialized by an array
5315 // copy and a StoreStore barrier exists after the array copy.
5316 alloc->initialization()->set_complete_with_arraycopy();
5317 }
5318
5319 Node* size = _gvn.transform(obj_size);
5320 access_clone(obj, alloc_obj, size, is_array);
5321
5322 // Do not let reads from the cloned object float above the arraycopy.
5323 if (alloc != nullptr) {
5324 // Do not let stores that initialize this object be reordered with
5325 // a subsequent store that would make this object accessible by
5326 // other threads.
5327 // Record what AllocateNode this StoreStore protects so that
5328 // escape analysis can go from the MemBarStoreStoreNode to the
5329 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5330 // based on the escape status of the AllocateNode.
5331 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5332 } else {
5333 insert_mem_bar(Op_MemBarCPUOrder);
5334 }
5335 }
5336
5337 //------------------------inline_native_clone----------------------------
5338 // protected native Object java.lang.Object.clone();
5339 //
5340 // Here are the simple edge cases:
5341 // null receiver => normal trap
5342 // virtual and clone was overridden => slow path to out-of-line clone
5343 // not cloneable or finalizer => slow path to out-of-line Object.clone
5344 //
5345 // The general case has two steps, allocation and copying.
5346 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5347 //
5348 // Copying also has two cases, oop arrays and everything else.
5349 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5350 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5351 //
5352 // These steps fold up nicely if and when the cloned object's klass
5353 // can be sharply typed as an object array, a type array, or an instance.
5354 //
5355 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5356 PhiNode* result_val;
5357
5358 // Set the reexecute bit for the interpreter to reexecute
5359 // the bytecode that invokes Object.clone if deoptimization happens.
5360 { PreserveReexecuteState preexecs(this);
5361 jvms()->set_should_reexecute(true);
5362
5363 Node* obj = null_check_receiver();
5364 if (stopped()) return true;
5365
5366 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5367
5368 // If we are going to clone an instance, we need its exact type to
5369 // know the number and types of fields to convert the clone to
5370 // loads/stores. Maybe a speculative type can help us.
5371 if (!obj_type->klass_is_exact() &&
5372 obj_type->speculative_type() != nullptr &&
5373 obj_type->speculative_type()->is_instance_klass()) {
5374 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5375 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5376 !spec_ik->has_injected_fields()) {
5377 if (!obj_type->isa_instptr() ||
5378 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5379 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5380 }
5381 }
5382 }
5383
5384 // Conservatively insert a memory barrier on all memory slices.
5385 // Do not let writes into the original float below the clone.
5386 insert_mem_bar(Op_MemBarCPUOrder);
5387
5388 // paths into result_reg:
5389 enum {
5390 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5391 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5392 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5393 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5394 PATH_LIMIT
5395 };
5396 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5397 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5398 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5399 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5400 record_for_igvn(result_reg);
5401
5402 Node* obj_klass = load_object_klass(obj);
5403 Node* array_obj = obj;
5404 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5405 if (array_ctl != nullptr) {
5406 // It's an array.
5407 PreserveJVMState pjvms(this);
5408 set_control(array_ctl);
5409 Node* obj_length = load_array_length(array_obj);
5410 Node* array_size = nullptr; // Size of the array without object alignment padding.
5411 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5412
5413 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5414 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5415 // If it is an oop array, it requires very special treatment,
5416 // because gc barriers are required when accessing the array.
5417 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5418 if (is_obja != nullptr) {
5419 PreserveJVMState pjvms2(this);
5420 set_control(is_obja);
5421 // Generate a direct call to the right arraycopy function(s).
5422 // Clones are always tightly coupled.
5423 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5424 ac->set_clone_oop_array();
5425 Node* n = _gvn.transform(ac);
5426 assert(n == ac, "cannot disappear");
5427 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5428
5429 result_reg->init_req(_objArray_path, control());
5430 result_val->init_req(_objArray_path, alloc_obj);
5431 result_i_o ->set_req(_objArray_path, i_o());
5432 result_mem ->set_req(_objArray_path, reset_memory());
5433 }
5434 }
5435 // Otherwise, there are no barriers to worry about.
5436 // (We can dispense with card marks if we know the allocation
5437 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5438 // causes the non-eden paths to take compensating steps to
5439 // simulate a fresh allocation, so that no further
5440 // card marks are required in compiled code to initialize
5441 // the object.)
5442
5443 if (!stopped()) {
5444 copy_to_clone(array_obj, alloc_obj, array_size, true);
5445
5446 // Present the results of the copy.
5447 result_reg->init_req(_array_path, control());
5448 result_val->init_req(_array_path, alloc_obj);
5449 result_i_o ->set_req(_array_path, i_o());
5450 result_mem ->set_req(_array_path, reset_memory());
5451 }
5452 }
5453
5454 // We only go to the instance fast case code if we pass a number of guards.
5455 // The paths which do not pass are accumulated in the slow_region.
5456 RegionNode* slow_region = new RegionNode(1);
5457 record_for_igvn(slow_region);
5458 if (!stopped()) {
5459 // It's an instance (we did array above). Make the slow-path tests.
5460 // If this is a virtual call, we generate a funny guard. We grab
5461 // the vtable entry corresponding to clone() from the target object.
5462 // If the target method which we are calling happens to be the
5463 // Object clone() method, we pass the guard. We do not need this
5464 // guard for non-virtual calls; the caller is known to be the native
5465 // Object clone().
5466 if (is_virtual) {
5467 generate_virtual_guard(obj_klass, slow_region);
5468 }
5469
5470 // The object must be easily cloneable and must not have a finalizer.
5471 // Both of these conditions may be checked in a single test.
5472 // We could optimize the test further, but we don't care.
5473 generate_misc_flags_guard(obj_klass,
5474 // Test both conditions:
5475 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5476 // Must be cloneable but not finalizer:
5477 KlassFlags::_misc_is_cloneable_fast,
5478 slow_region);
5479 }
5480
5481 if (!stopped()) {
5482 // It's an instance, and it passed the slow-path tests.
5483 PreserveJVMState pjvms(this);
5484 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5485 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5486 // is reexecuted if deoptimization occurs and there could be problems when merging
5487 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5488 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5489
5490 copy_to_clone(obj, alloc_obj, obj_size, false);
5491
5492 // Present the results of the slow call.
5493 result_reg->init_req(_instance_path, control());
5494 result_val->init_req(_instance_path, alloc_obj);
5495 result_i_o ->set_req(_instance_path, i_o());
5496 result_mem ->set_req(_instance_path, reset_memory());
5497 }
5498
5499 // Generate code for the slow case. We make a call to clone().
5500 set_control(_gvn.transform(slow_region));
5501 if (!stopped()) {
5502 PreserveJVMState pjvms(this);
5503 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5504 // We need to deoptimize on exception (see comment above)
5505 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5506 // this->control() comes from set_results_for_java_call
5507 result_reg->init_req(_slow_path, control());
5508 result_val->init_req(_slow_path, slow_result);
5509 result_i_o ->set_req(_slow_path, i_o());
5510 result_mem ->set_req(_slow_path, reset_memory());
5511 }
5512
5513 // Return the combined state.
5514 set_control( _gvn.transform(result_reg));
5515 set_i_o( _gvn.transform(result_i_o));
5516 set_all_memory( _gvn.transform(result_mem));
5517 } // original reexecute is set back here
5518
5519 set_result(_gvn.transform(result_val));
5520 return true;
5521 }
5522
5523 // If we have a tightly coupled allocation, the arraycopy may take care
5524 // of the array initialization. If one of the guards we insert between
5525 // the allocation and the arraycopy causes a deoptimization, an
5526 // uninitialized array will escape the compiled method. To prevent that
5527 // we set the JVM state for uncommon traps between the allocation and
5528 // the arraycopy to the state before the allocation so, in case of
5529 // deoptimization, we'll reexecute the allocation and the
5530 // initialization.
5531 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5532 if (alloc != nullptr) {
5533 ciMethod* trap_method = alloc->jvms()->method();
5534 int trap_bci = alloc->jvms()->bci();
5535
5536 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5537 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5538 // Make sure there's no store between the allocation and the
5539 // arraycopy otherwise visible side effects could be rexecuted
5540 // in case of deoptimization and cause incorrect execution.
5541 bool no_interfering_store = true;
5542 Node* mem = alloc->in(TypeFunc::Memory);
5543 if (mem->is_MergeMem()) {
5544 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5545 Node* n = mms.memory();
5546 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5547 assert(n->is_Store(), "what else?");
5548 no_interfering_store = false;
5549 break;
5550 }
5551 }
5552 } else {
5553 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5554 Node* n = mms.memory();
5555 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5556 assert(n->is_Store(), "what else?");
5557 no_interfering_store = false;
5558 break;
5559 }
5560 }
5561 }
5562
5563 if (no_interfering_store) {
5564 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5565
5566 JVMState* saved_jvms = jvms();
5567 saved_reexecute_sp = _reexecute_sp;
5568
5569 set_jvms(sfpt->jvms());
5570 _reexecute_sp = jvms()->sp();
5571
5572 return saved_jvms;
5573 }
5574 }
5575 }
5576 return nullptr;
5577 }
5578
5579 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5580 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5581 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5582 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5583 uint size = alloc->req();
5584 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5585 old_jvms->set_map(sfpt);
5586 for (uint i = 0; i < size; i++) {
5587 sfpt->init_req(i, alloc->in(i));
5588 }
5589 // re-push array length for deoptimization
5590 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5591 old_jvms->set_sp(old_jvms->sp()+1);
5592 old_jvms->set_monoff(old_jvms->monoff()+1);
5593 old_jvms->set_scloff(old_jvms->scloff()+1);
5594 old_jvms->set_endoff(old_jvms->endoff()+1);
5595 old_jvms->set_should_reexecute(true);
5596
5597 sfpt->set_i_o(map()->i_o());
5598 sfpt->set_memory(map()->memory());
5599 sfpt->set_control(map()->control());
5600 return sfpt;
5601 }
5602
5603 // In case of a deoptimization, we restart execution at the
5604 // allocation, allocating a new array. We would leave an uninitialized
5605 // array in the heap that GCs wouldn't expect. Move the allocation
5606 // after the traps so we don't allocate the array if we
5607 // deoptimize. This is possible because tightly_coupled_allocation()
5608 // guarantees there's no observer of the allocated array at this point
5609 // and the control flow is simple enough.
5610 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5611 int saved_reexecute_sp, uint new_idx) {
5612 if (saved_jvms_before_guards != nullptr && !stopped()) {
5613 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5614
5615 assert(alloc != nullptr, "only with a tightly coupled allocation");
5616 // restore JVM state to the state at the arraycopy
5617 saved_jvms_before_guards->map()->set_control(map()->control());
5618 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5619 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5620 // If we've improved the types of some nodes (null check) while
5621 // emitting the guards, propagate them to the current state
5622 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5623 set_jvms(saved_jvms_before_guards);
5624 _reexecute_sp = saved_reexecute_sp;
5625
5626 // Remove the allocation from above the guards
5627 CallProjections callprojs;
5628 alloc->extract_projections(&callprojs, true);
5629 InitializeNode* init = alloc->initialization();
5630 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5631 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5632 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5633
5634 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5635 // the allocation (i.e. is only valid if the allocation succeeds):
5636 // 1) replace CastIINode with AllocateArrayNode's length here
5637 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5638 //
5639 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5640 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5641 Node* init_control = init->proj_out(TypeFunc::Control);
5642 Node* alloc_length = alloc->Ideal_length();
5643 #ifdef ASSERT
5644 Node* prev_cast = nullptr;
5645 #endif
5646 for (uint i = 0; i < init_control->outcnt(); i++) {
5647 Node* init_out = init_control->raw_out(i);
5648 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5649 #ifdef ASSERT
5650 if (prev_cast == nullptr) {
5651 prev_cast = init_out;
5652 } else {
5653 if (prev_cast->cmp(*init_out) == false) {
5654 prev_cast->dump();
5655 init_out->dump();
5656 assert(false, "not equal CastIINode");
5657 }
5658 }
5659 #endif
5660 C->gvn_replace_by(init_out, alloc_length);
5661 }
5662 }
5663 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5664
5665 // move the allocation here (after the guards)
5666 _gvn.hash_delete(alloc);
5667 alloc->set_req(TypeFunc::Control, control());
5668 alloc->set_req(TypeFunc::I_O, i_o());
5669 Node *mem = reset_memory();
5670 set_all_memory(mem);
5671 alloc->set_req(TypeFunc::Memory, mem);
5672 set_control(init->proj_out_or_null(TypeFunc::Control));
5673 set_i_o(callprojs.fallthrough_ioproj);
5674
5675 // Update memory as done in GraphKit::set_output_for_allocation()
5676 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5677 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5678 if (ary_type->isa_aryptr() && length_type != nullptr) {
5679 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5680 }
5681 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5682 int elemidx = C->get_alias_index(telemref);
5683 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5684 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5685
5686 Node* allocx = _gvn.transform(alloc);
5687 assert(allocx == alloc, "where has the allocation gone?");
5688 assert(dest->is_CheckCastPP(), "not an allocation result?");
5689
5690 _gvn.hash_delete(dest);
5691 dest->set_req(0, control());
5692 Node* destx = _gvn.transform(dest);
5693 assert(destx == dest, "where has the allocation result gone?");
5694
5695 array_ideal_length(alloc, ary_type, true);
5696 }
5697 }
5698
5699 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5700 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5701 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5702 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5703 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5704 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5705 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5706 JVMState* saved_jvms_before_guards) {
5707 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5708 // There is at least one unrelated uncommon trap which needs to be replaced.
5709 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5710
5711 JVMState* saved_jvms = jvms();
5712 const int saved_reexecute_sp = _reexecute_sp;
5713 set_jvms(sfpt->jvms());
5714 _reexecute_sp = jvms()->sp();
5715
5716 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5717
5718 // Restore state
5719 set_jvms(saved_jvms);
5720 _reexecute_sp = saved_reexecute_sp;
5721 }
5722 }
5723
5724 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5725 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5726 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5727 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5728 while (if_proj->is_IfProj()) {
5729 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5730 if (uncommon_trap != nullptr) {
5731 create_new_uncommon_trap(uncommon_trap);
5732 }
5733 assert(if_proj->in(0)->is_If(), "must be If");
5734 if_proj = if_proj->in(0)->in(0);
5735 }
5736 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5737 "must have reached control projection of init node");
5738 }
5739
5740 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5741 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5742 assert(trap_request != 0, "no valid UCT trap request");
5743 PreserveJVMState pjvms(this);
5744 set_control(uncommon_trap_call->in(0));
5745 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5746 Deoptimization::trap_request_action(trap_request));
5747 assert(stopped(), "Should be stopped");
5748 _gvn.hash_delete(uncommon_trap_call);
5749 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5750 }
5751
5752 // Common checks for array sorting intrinsics arguments.
5753 // Returns `true` if checks passed.
5754 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
5755 // check address of the class
5756 if (elementType == nullptr || elementType->is_top()) {
5757 return false; // dead path
5758 }
5759 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5760 if (elem_klass == nullptr) {
5761 return false; // dead path
5762 }
5763 // java_mirror_type() returns non-null for compile-time Class constants only
5764 ciType* elem_type = elem_klass->java_mirror_type();
5765 if (elem_type == nullptr) {
5766 return false;
5767 }
5768 bt = elem_type->basic_type();
5769 // Disable the intrinsic if the CPU does not support SIMD sort
5770 if (!Matcher::supports_simd_sort(bt)) {
5771 return false;
5772 }
5773 // check address of the array
5774 if (obj == nullptr || obj->is_top()) {
5775 return false; // dead path
5776 }
5777 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5778 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
5779 return false; // failed input validation
5780 }
5781 return true;
5782 }
5783
5784 //------------------------------inline_array_partition-----------------------
5785 bool LibraryCallKit::inline_array_partition() {
5786 address stubAddr = StubRoutines::select_array_partition_function();
5787 if (stubAddr == nullptr) {
5788 return false; // Intrinsic's stub is not implemented on this platform
5789 }
5790 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
5791
5792 // no receiver because it is a static method
5793 Node* elementType = argument(0);
5794 Node* obj = argument(1);
5795 Node* offset = argument(2); // long
5796 Node* fromIndex = argument(4);
5797 Node* toIndex = argument(5);
5798 Node* indexPivot1 = argument(6);
5799 Node* indexPivot2 = argument(7);
5800 // PartitionOperation: argument(8) is ignored
5801
5802 Node* pivotIndices = nullptr;
5803 BasicType bt = T_ILLEGAL;
5804
5805 if (!check_array_sort_arguments(elementType, obj, bt)) {
5806 return false;
5807 }
5808 null_check(obj);
5809 // If obj is dead, only null-path is taken.
5810 if (stopped()) {
5811 return true;
5812 }
5813 // Set the original stack and the reexecute bit for the interpreter to reexecute
5814 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
5815 { PreserveReexecuteState preexecs(this);
5816 jvms()->set_should_reexecute(true);
5817
5818 Node* obj_adr = make_unsafe_address(obj, offset);
5819
5820 // create the pivotIndices array of type int and size = 2
5821 Node* size = intcon(2);
5822 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
5823 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
5824 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
5825 guarantee(alloc != nullptr, "created above");
5826 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
5827
5828 // pass the basic type enum to the stub
5829 Node* elemType = intcon(bt);
5830
5831 // Call the stub
5832 const char *stubName = "array_partition_stub";
5833 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
5834 stubAddr, stubName, TypePtr::BOTTOM,
5835 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
5836 indexPivot1, indexPivot2);
5837
5838 } // original reexecute is set back here
5839
5840 if (!stopped()) {
5841 set_result(pivotIndices);
5842 }
5843
5844 return true;
5845 }
5846
5847
5848 //------------------------------inline_array_sort-----------------------
5849 bool LibraryCallKit::inline_array_sort() {
5850 address stubAddr = StubRoutines::select_arraysort_function();
5851 if (stubAddr == nullptr) {
5852 return false; // Intrinsic's stub is not implemented on this platform
5853 }
5854 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
5855
5856 // no receiver because it is a static method
5857 Node* elementType = argument(0);
5858 Node* obj = argument(1);
5859 Node* offset = argument(2); // long
5860 Node* fromIndex = argument(4);
5861 Node* toIndex = argument(5);
5862 // SortOperation: argument(6) is ignored
5863
5864 BasicType bt = T_ILLEGAL;
5865
5866 if (!check_array_sort_arguments(elementType, obj, bt)) {
5867 return false;
5868 }
5869 null_check(obj);
5870 // If obj is dead, only null-path is taken.
5871 if (stopped()) {
5872 return true;
5873 }
5874 Node* obj_adr = make_unsafe_address(obj, offset);
5875
5876 // pass the basic type enum to the stub
5877 Node* elemType = intcon(bt);
5878
5879 // Call the stub.
5880 const char *stubName = "arraysort_stub";
5881 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
5882 stubAddr, stubName, TypePtr::BOTTOM,
5883 obj_adr, elemType, fromIndex, toIndex);
5884
5885 return true;
5886 }
5887
5888
5889 //------------------------------inline_arraycopy-----------------------
5890 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5891 // Object dest, int destPos,
5892 // int length);
5893 bool LibraryCallKit::inline_arraycopy() {
5894 // Get the arguments.
5895 Node* src = argument(0); // type: oop
5896 Node* src_offset = argument(1); // type: int
5897 Node* dest = argument(2); // type: oop
5898 Node* dest_offset = argument(3); // type: int
5899 Node* length = argument(4); // type: int
5900
5901 uint new_idx = C->unique();
5902
5903 // Check for allocation before we add nodes that would confuse
5904 // tightly_coupled_allocation()
5905 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5906
5907 int saved_reexecute_sp = -1;
5908 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5909 // See arraycopy_restore_alloc_state() comment
5910 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5911 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5912 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5913 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5914
5915 // The following tests must be performed
5916 // (1) src and dest are arrays.
5917 // (2) src and dest arrays must have elements of the same BasicType
5918 // (3) src and dest must not be null.
5919 // (4) src_offset must not be negative.
5920 // (5) dest_offset must not be negative.
5921 // (6) length must not be negative.
5922 // (7) src_offset + length must not exceed length of src.
5923 // (8) dest_offset + length must not exceed length of dest.
5924 // (9) each element of an oop array must be assignable
5925
5926 // (3) src and dest must not be null.
5927 // always do this here because we need the JVM state for uncommon traps
5928 Node* null_ctl = top();
5929 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5930 assert(null_ctl->is_top(), "no null control here");
5931 dest = null_check(dest, T_ARRAY);
5932
5933 if (!can_emit_guards) {
5934 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5935 // guards but the arraycopy node could still take advantage of a
5936 // tightly allocated allocation. tightly_coupled_allocation() is
5937 // called again to make sure it takes the null check above into
5938 // account: the null check is mandatory and if it caused an
5939 // uncommon trap to be emitted then the allocation can't be
5940 // considered tightly coupled in this context.
5941 alloc = tightly_coupled_allocation(dest);
5942 }
5943
5944 bool validated = false;
5945
5946 const Type* src_type = _gvn.type(src);
5947 const Type* dest_type = _gvn.type(dest);
5948 const TypeAryPtr* top_src = src_type->isa_aryptr();
5949 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5950
5951 // Do we have the type of src?
5952 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5953 // Do we have the type of dest?
5954 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5955 // Is the type for src from speculation?
5956 bool src_spec = false;
5957 // Is the type for dest from speculation?
5958 bool dest_spec = false;
5959
5960 if ((!has_src || !has_dest) && can_emit_guards) {
5961 // We don't have sufficient type information, let's see if
5962 // speculative types can help. We need to have types for both src
5963 // and dest so that it pays off.
5964
5965 // Do we already have or could we have type information for src
5966 bool could_have_src = has_src;
5967 // Do we already have or could we have type information for dest
5968 bool could_have_dest = has_dest;
5969
5970 ciKlass* src_k = nullptr;
5971 if (!has_src) {
5972 src_k = src_type->speculative_type_not_null();
5973 if (src_k != nullptr && src_k->is_array_klass()) {
5974 could_have_src = true;
5975 }
5976 }
5977
5978 ciKlass* dest_k = nullptr;
5979 if (!has_dest) {
5980 dest_k = dest_type->speculative_type_not_null();
5981 if (dest_k != nullptr && dest_k->is_array_klass()) {
5982 could_have_dest = true;
5983 }
5984 }
5985
5986 if (could_have_src && could_have_dest) {
5987 // This is going to pay off so emit the required guards
5988 if (!has_src) {
5989 src = maybe_cast_profiled_obj(src, src_k, true);
5990 src_type = _gvn.type(src);
5991 top_src = src_type->isa_aryptr();
5992 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5993 src_spec = true;
5994 }
5995 if (!has_dest) {
5996 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5997 dest_type = _gvn.type(dest);
5998 top_dest = dest_type->isa_aryptr();
5999 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6000 dest_spec = true;
6001 }
6002 }
6003 }
6004
6005 if (has_src && has_dest && can_emit_guards) {
6006 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
6007 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
6008 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
6009 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
6010
6011 if (src_elem == dest_elem && src_elem == T_OBJECT) {
6012 // If both arrays are object arrays then having the exact types
6013 // for both will remove the need for a subtype check at runtime
6014 // before the call and may make it possible to pick a faster copy
6015 // routine (without a subtype check on every element)
6016 // Do we have the exact type of src?
6017 bool could_have_src = src_spec;
6018 // Do we have the exact type of dest?
6019 bool could_have_dest = dest_spec;
6020 ciKlass* src_k = nullptr;
6021 ciKlass* dest_k = nullptr;
6022 if (!src_spec) {
6023 src_k = src_type->speculative_type_not_null();
6024 if (src_k != nullptr && src_k->is_array_klass()) {
6025 could_have_src = true;
6026 }
6027 }
6028 if (!dest_spec) {
6029 dest_k = dest_type->speculative_type_not_null();
6030 if (dest_k != nullptr && dest_k->is_array_klass()) {
6031 could_have_dest = true;
6032 }
6033 }
6034 if (could_have_src && could_have_dest) {
6035 // If we can have both exact types, emit the missing guards
6036 if (could_have_src && !src_spec) {
6037 src = maybe_cast_profiled_obj(src, src_k, true);
6038 }
6039 if (could_have_dest && !dest_spec) {
6040 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6041 }
6042 }
6043 }
6044 }
6045
6046 ciMethod* trap_method = method();
6047 int trap_bci = bci();
6048 if (saved_jvms_before_guards != nullptr) {
6049 trap_method = alloc->jvms()->method();
6050 trap_bci = alloc->jvms()->bci();
6051 }
6052
6053 bool negative_length_guard_generated = false;
6054
6055 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6056 can_emit_guards &&
6057 !src->is_top() && !dest->is_top()) {
6058 // validate arguments: enables transformation the ArrayCopyNode
6059 validated = true;
6060
6061 RegionNode* slow_region = new RegionNode(1);
6062 record_for_igvn(slow_region);
6063
6064 // (1) src and dest are arrays.
6065 generate_non_array_guard(load_object_klass(src), slow_region, &src);
6066 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
6067
6068 // (2) src and dest arrays must have elements of the same BasicType
6069 // done at macro expansion or at Ideal transformation time
6070
6071 // (4) src_offset must not be negative.
6072 generate_negative_guard(src_offset, slow_region);
6073
6074 // (5) dest_offset must not be negative.
6075 generate_negative_guard(dest_offset, slow_region);
6076
6077 // (7) src_offset + length must not exceed length of src.
6078 generate_limit_guard(src_offset, length,
6079 load_array_length(src),
6080 slow_region);
6081
6082 // (8) dest_offset + length must not exceed length of dest.
6083 generate_limit_guard(dest_offset, length,
6084 load_array_length(dest),
6085 slow_region);
6086
6087 // (6) length must not be negative.
6088 // This is also checked in generate_arraycopy() during macro expansion, but
6089 // we also have to check it here for the case where the ArrayCopyNode will
6090 // be eliminated by Escape Analysis.
6091 if (EliminateAllocations) {
6092 generate_negative_guard(length, slow_region);
6093 negative_length_guard_generated = true;
6094 }
6095
6096 // (9) each element of an oop array must be assignable
6097 Node* dest_klass = load_object_klass(dest);
6098 if (src != dest) {
6099 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6100
6101 if (not_subtype_ctrl != top()) {
6102 PreserveJVMState pjvms(this);
6103 set_control(not_subtype_ctrl);
6104 uncommon_trap(Deoptimization::Reason_intrinsic,
6105 Deoptimization::Action_make_not_entrant);
6106 assert(stopped(), "Should be stopped");
6107 }
6108 }
6109 {
6110 PreserveJVMState pjvms(this);
6111 set_control(_gvn.transform(slow_region));
6112 uncommon_trap(Deoptimization::Reason_intrinsic,
6113 Deoptimization::Action_make_not_entrant);
6114 assert(stopped(), "Should be stopped");
6115 }
6116
6117 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
6118 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6119 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6120 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6121 }
6122
6123 if (stopped()) {
6124 return true;
6125 }
6126
6127 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6128 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6129 // so the compiler has a chance to eliminate them: during macro expansion,
6130 // we have to set their control (CastPP nodes are eliminated).
6131 load_object_klass(src), load_object_klass(dest),
6132 load_array_length(src), load_array_length(dest));
6133
6134 ac->set_arraycopy(validated);
6135
6136 Node* n = _gvn.transform(ac);
6137 if (n == ac) {
6138 ac->connect_outputs(this);
6139 } else {
6140 assert(validated, "shouldn't transform if all arguments not validated");
6141 set_all_memory(n);
6142 }
6143 clear_upper_avx();
6144
6145
6146 return true;
6147 }
6148
6149
6150 // Helper function which determines if an arraycopy immediately follows
6151 // an allocation, with no intervening tests or other escapes for the object.
6152 AllocateArrayNode*
6153 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6154 if (stopped()) return nullptr; // no fast path
6155 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6156
6157 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6158 if (alloc == nullptr) return nullptr;
6159
6160 Node* rawmem = memory(Compile::AliasIdxRaw);
6161 // Is the allocation's memory state untouched?
6162 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6163 // Bail out if there have been raw-memory effects since the allocation.
6164 // (Example: There might have been a call or safepoint.)
6165 return nullptr;
6166 }
6167 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6168 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6169 return nullptr;
6170 }
6171
6172 // There must be no unexpected observers of this allocation.
6173 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6174 Node* obs = ptr->fast_out(i);
6175 if (obs != this->map()) {
6176 return nullptr;
6177 }
6178 }
6179
6180 // This arraycopy must unconditionally follow the allocation of the ptr.
6181 Node* alloc_ctl = ptr->in(0);
6182 Node* ctl = control();
6183 while (ctl != alloc_ctl) {
6184 // There may be guards which feed into the slow_region.
6185 // Any other control flow means that we might not get a chance
6186 // to finish initializing the allocated object.
6187 // Various low-level checks bottom out in uncommon traps. These
6188 // are considered safe since we've already checked above that
6189 // there is no unexpected observer of this allocation.
6190 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6191 assert(ctl->in(0)->is_If(), "must be If");
6192 ctl = ctl->in(0)->in(0);
6193 } else {
6194 return nullptr;
6195 }
6196 }
6197
6198 // If we get this far, we have an allocation which immediately
6199 // precedes the arraycopy, and we can take over zeroing the new object.
6200 // The arraycopy will finish the initialization, and provide
6201 // a new control state to which we will anchor the destination pointer.
6202
6203 return alloc;
6204 }
6205
6206 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6207 if (node->is_IfProj()) {
6208 Node* other_proj = node->as_IfProj()->other_if_proj();
6209 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6210 Node* obs = other_proj->fast_out(j);
6211 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6212 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6213 return obs->as_CallStaticJava();
6214 }
6215 }
6216 }
6217 return nullptr;
6218 }
6219
6220 //-------------inline_encodeISOArray-----------------------------------
6221 // encode char[] to byte[] in ISO_8859_1 or ASCII
6222 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6223 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6224 // no receiver since it is static method
6225 Node *src = argument(0);
6226 Node *src_offset = argument(1);
6227 Node *dst = argument(2);
6228 Node *dst_offset = argument(3);
6229 Node *length = argument(4);
6230
6231 src = must_be_not_null(src, true);
6232 dst = must_be_not_null(dst, true);
6233
6234 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6235 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6236 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6237 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6238 // failed array check
6239 return false;
6240 }
6241
6242 // Figure out the size and type of the elements we will be copying.
6243 BasicType src_elem = src_type->elem()->array_element_basic_type();
6244 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6245 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6246 return false;
6247 }
6248
6249 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6250 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6251 // 'src_start' points to src array + scaled offset
6252 // 'dst_start' points to dst array + scaled offset
6253
6254 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6255 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6256 enc = _gvn.transform(enc);
6257 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6258 set_memory(res_mem, mtype);
6259 set_result(enc);
6260 clear_upper_avx();
6261
6262 return true;
6263 }
6264
6265 //-------------inline_multiplyToLen-----------------------------------
6266 bool LibraryCallKit::inline_multiplyToLen() {
6267 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6268
6269 address stubAddr = StubRoutines::multiplyToLen();
6270 if (stubAddr == nullptr) {
6271 return false; // Intrinsic's stub is not implemented on this platform
6272 }
6273 const char* stubName = "multiplyToLen";
6274
6275 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6276
6277 // no receiver because it is a static method
6278 Node* x = argument(0);
6279 Node* xlen = argument(1);
6280 Node* y = argument(2);
6281 Node* ylen = argument(3);
6282 Node* z = argument(4);
6283
6284 x = must_be_not_null(x, true);
6285 y = must_be_not_null(y, true);
6286
6287 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6288 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6289 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6290 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6291 // failed array check
6292 return false;
6293 }
6294
6295 BasicType x_elem = x_type->elem()->array_element_basic_type();
6296 BasicType y_elem = y_type->elem()->array_element_basic_type();
6297 if (x_elem != T_INT || y_elem != T_INT) {
6298 return false;
6299 }
6300
6301 Node* x_start = array_element_address(x, intcon(0), x_elem);
6302 Node* y_start = array_element_address(y, intcon(0), y_elem);
6303 // 'x_start' points to x array + scaled xlen
6304 // 'y_start' points to y array + scaled ylen
6305
6306 Node* z_start = array_element_address(z, intcon(0), T_INT);
6307
6308 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6309 OptoRuntime::multiplyToLen_Type(),
6310 stubAddr, stubName, TypePtr::BOTTOM,
6311 x_start, xlen, y_start, ylen, z_start);
6312
6313 C->set_has_split_ifs(true); // Has chance for split-if optimization
6314 set_result(z);
6315 return true;
6316 }
6317
6318 //-------------inline_squareToLen------------------------------------
6319 bool LibraryCallKit::inline_squareToLen() {
6320 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6321
6322 address stubAddr = StubRoutines::squareToLen();
6323 if (stubAddr == nullptr) {
6324 return false; // Intrinsic's stub is not implemented on this platform
6325 }
6326 const char* stubName = "squareToLen";
6327
6328 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6329
6330 Node* x = argument(0);
6331 Node* len = argument(1);
6332 Node* z = argument(2);
6333 Node* zlen = argument(3);
6334
6335 x = must_be_not_null(x, true);
6336 z = must_be_not_null(z, true);
6337
6338 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6339 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6340 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6341 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6342 // failed array check
6343 return false;
6344 }
6345
6346 BasicType x_elem = x_type->elem()->array_element_basic_type();
6347 BasicType z_elem = z_type->elem()->array_element_basic_type();
6348 if (x_elem != T_INT || z_elem != T_INT) {
6349 return false;
6350 }
6351
6352
6353 Node* x_start = array_element_address(x, intcon(0), x_elem);
6354 Node* z_start = array_element_address(z, intcon(0), z_elem);
6355
6356 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6357 OptoRuntime::squareToLen_Type(),
6358 stubAddr, stubName, TypePtr::BOTTOM,
6359 x_start, len, z_start, zlen);
6360
6361 set_result(z);
6362 return true;
6363 }
6364
6365 //-------------inline_mulAdd------------------------------------------
6366 bool LibraryCallKit::inline_mulAdd() {
6367 assert(UseMulAddIntrinsic, "not implemented on this platform");
6368
6369 address stubAddr = StubRoutines::mulAdd();
6370 if (stubAddr == nullptr) {
6371 return false; // Intrinsic's stub is not implemented on this platform
6372 }
6373 const char* stubName = "mulAdd";
6374
6375 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6376
6377 Node* out = argument(0);
6378 Node* in = argument(1);
6379 Node* offset = argument(2);
6380 Node* len = argument(3);
6381 Node* k = argument(4);
6382
6383 in = must_be_not_null(in, true);
6384 out = must_be_not_null(out, true);
6385
6386 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6387 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6388 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6389 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6390 // failed array check
6391 return false;
6392 }
6393
6394 BasicType out_elem = out_type->elem()->array_element_basic_type();
6395 BasicType in_elem = in_type->elem()->array_element_basic_type();
6396 if (out_elem != T_INT || in_elem != T_INT) {
6397 return false;
6398 }
6399
6400 Node* outlen = load_array_length(out);
6401 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6402 Node* out_start = array_element_address(out, intcon(0), out_elem);
6403 Node* in_start = array_element_address(in, intcon(0), in_elem);
6404
6405 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6406 OptoRuntime::mulAdd_Type(),
6407 stubAddr, stubName, TypePtr::BOTTOM,
6408 out_start,in_start, new_offset, len, k);
6409 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6410 set_result(result);
6411 return true;
6412 }
6413
6414 //-------------inline_montgomeryMultiply-----------------------------------
6415 bool LibraryCallKit::inline_montgomeryMultiply() {
6416 address stubAddr = StubRoutines::montgomeryMultiply();
6417 if (stubAddr == nullptr) {
6418 return false; // Intrinsic's stub is not implemented on this platform
6419 }
6420
6421 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6422 const char* stubName = "montgomery_multiply";
6423
6424 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6425
6426 Node* a = argument(0);
6427 Node* b = argument(1);
6428 Node* n = argument(2);
6429 Node* len = argument(3);
6430 Node* inv = argument(4);
6431 Node* m = argument(6);
6432
6433 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6434 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6435 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6436 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6437 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6438 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6439 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6440 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6441 // failed array check
6442 return false;
6443 }
6444
6445 BasicType a_elem = a_type->elem()->array_element_basic_type();
6446 BasicType b_elem = b_type->elem()->array_element_basic_type();
6447 BasicType n_elem = n_type->elem()->array_element_basic_type();
6448 BasicType m_elem = m_type->elem()->array_element_basic_type();
6449 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6450 return false;
6451 }
6452
6453 // Make the call
6454 {
6455 Node* a_start = array_element_address(a, intcon(0), a_elem);
6456 Node* b_start = array_element_address(b, intcon(0), b_elem);
6457 Node* n_start = array_element_address(n, intcon(0), n_elem);
6458 Node* m_start = array_element_address(m, intcon(0), m_elem);
6459
6460 Node* call = make_runtime_call(RC_LEAF,
6461 OptoRuntime::montgomeryMultiply_Type(),
6462 stubAddr, stubName, TypePtr::BOTTOM,
6463 a_start, b_start, n_start, len, inv, top(),
6464 m_start);
6465 set_result(m);
6466 }
6467
6468 return true;
6469 }
6470
6471 bool LibraryCallKit::inline_montgomerySquare() {
6472 address stubAddr = StubRoutines::montgomerySquare();
6473 if (stubAddr == nullptr) {
6474 return false; // Intrinsic's stub is not implemented on this platform
6475 }
6476
6477 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6478 const char* stubName = "montgomery_square";
6479
6480 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6481
6482 Node* a = argument(0);
6483 Node* n = argument(1);
6484 Node* len = argument(2);
6485 Node* inv = argument(3);
6486 Node* m = argument(5);
6487
6488 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6489 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6490 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6491 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6492 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6493 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6494 // failed array check
6495 return false;
6496 }
6497
6498 BasicType a_elem = a_type->elem()->array_element_basic_type();
6499 BasicType n_elem = n_type->elem()->array_element_basic_type();
6500 BasicType m_elem = m_type->elem()->array_element_basic_type();
6501 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6502 return false;
6503 }
6504
6505 // Make the call
6506 {
6507 Node* a_start = array_element_address(a, intcon(0), a_elem);
6508 Node* n_start = array_element_address(n, intcon(0), n_elem);
6509 Node* m_start = array_element_address(m, intcon(0), m_elem);
6510
6511 Node* call = make_runtime_call(RC_LEAF,
6512 OptoRuntime::montgomerySquare_Type(),
6513 stubAddr, stubName, TypePtr::BOTTOM,
6514 a_start, n_start, len, inv, top(),
6515 m_start);
6516 set_result(m);
6517 }
6518
6519 return true;
6520 }
6521
6522 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6523 address stubAddr = nullptr;
6524 const char* stubName = nullptr;
6525
6526 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6527 if (stubAddr == nullptr) {
6528 return false; // Intrinsic's stub is not implemented on this platform
6529 }
6530
6531 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6532
6533 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6534
6535 Node* newArr = argument(0);
6536 Node* oldArr = argument(1);
6537 Node* newIdx = argument(2);
6538 Node* shiftCount = argument(3);
6539 Node* numIter = argument(4);
6540
6541 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6542 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6543 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6544 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6545 return false;
6546 }
6547
6548 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6549 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6550 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6551 return false;
6552 }
6553
6554 // Make the call
6555 {
6556 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6557 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6558
6559 Node* call = make_runtime_call(RC_LEAF,
6560 OptoRuntime::bigIntegerShift_Type(),
6561 stubAddr,
6562 stubName,
6563 TypePtr::BOTTOM,
6564 newArr_start,
6565 oldArr_start,
6566 newIdx,
6567 shiftCount,
6568 numIter);
6569 }
6570
6571 return true;
6572 }
6573
6574 //-------------inline_vectorizedMismatch------------------------------
6575 bool LibraryCallKit::inline_vectorizedMismatch() {
6576 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6577
6578 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6579 Node* obja = argument(0); // Object
6580 Node* aoffset = argument(1); // long
6581 Node* objb = argument(3); // Object
6582 Node* boffset = argument(4); // long
6583 Node* length = argument(6); // int
6584 Node* scale = argument(7); // int
6585
6586 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6587 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6588 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6589 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6590 scale == top()) {
6591 return false; // failed input validation
6592 }
6593
6594 Node* obja_adr = make_unsafe_address(obja, aoffset);
6595 Node* objb_adr = make_unsafe_address(objb, boffset);
6596
6597 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6598 //
6599 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6600 // if (length <= inline_limit) {
6601 // inline_path:
6602 // vmask = VectorMaskGen length
6603 // vload1 = LoadVectorMasked obja, vmask
6604 // vload2 = LoadVectorMasked objb, vmask
6605 // result1 = VectorCmpMasked vload1, vload2, vmask
6606 // } else {
6607 // call_stub_path:
6608 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6609 // }
6610 // exit_block:
6611 // return Phi(result1, result2);
6612 //
6613 enum { inline_path = 1, // input is small enough to process it all at once
6614 stub_path = 2, // input is too large; call into the VM
6615 PATH_LIMIT = 3
6616 };
6617
6618 Node* exit_block = new RegionNode(PATH_LIMIT);
6619 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6620 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6621
6622 Node* call_stub_path = control();
6623
6624 BasicType elem_bt = T_ILLEGAL;
6625
6626 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6627 if (scale_t->is_con()) {
6628 switch (scale_t->get_con()) {
6629 case 0: elem_bt = T_BYTE; break;
6630 case 1: elem_bt = T_SHORT; break;
6631 case 2: elem_bt = T_INT; break;
6632 case 3: elem_bt = T_LONG; break;
6633
6634 default: elem_bt = T_ILLEGAL; break; // not supported
6635 }
6636 }
6637
6638 int inline_limit = 0;
6639 bool do_partial_inline = false;
6640
6641 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6642 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6643 do_partial_inline = inline_limit >= 16;
6644 }
6645
6646 if (do_partial_inline) {
6647 assert(elem_bt != T_ILLEGAL, "sanity");
6648
6649 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6650 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6651 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6652
6653 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6654 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6655 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6656
6657 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6658
6659 if (!stopped()) {
6660 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6661
6662 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6663 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6664 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6665 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6666
6667 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6668 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6669 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6670 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6671
6672 exit_block->init_req(inline_path, control());
6673 memory_phi->init_req(inline_path, map()->memory());
6674 result_phi->init_req(inline_path, result);
6675
6676 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6677 clear_upper_avx();
6678 }
6679 }
6680 }
6681
6682 if (call_stub_path != nullptr) {
6683 set_control(call_stub_path);
6684
6685 Node* call = make_runtime_call(RC_LEAF,
6686 OptoRuntime::vectorizedMismatch_Type(),
6687 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6688 obja_adr, objb_adr, length, scale);
6689
6690 exit_block->init_req(stub_path, control());
6691 memory_phi->init_req(stub_path, map()->memory());
6692 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6693 }
6694
6695 exit_block = _gvn.transform(exit_block);
6696 memory_phi = _gvn.transform(memory_phi);
6697 result_phi = _gvn.transform(result_phi);
6698
6699 set_control(exit_block);
6700 set_all_memory(memory_phi);
6701 set_result(result_phi);
6702
6703 return true;
6704 }
6705
6706 //------------------------------inline_vectorizedHashcode----------------------------
6707 bool LibraryCallKit::inline_vectorizedHashCode() {
6708 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6709
6710 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6711 Node* array = argument(0);
6712 Node* offset = argument(1);
6713 Node* length = argument(2);
6714 Node* initialValue = argument(3);
6715 Node* basic_type = argument(4);
6716
6717 if (basic_type == top()) {
6718 return false; // failed input validation
6719 }
6720
6721 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6722 if (!basic_type_t->is_con()) {
6723 return false; // Only intrinsify if mode argument is constant
6724 }
6725
6726 array = must_be_not_null(array, true);
6727
6728 BasicType bt = (BasicType)basic_type_t->get_con();
6729
6730 // Resolve address of first element
6731 Node* array_start = array_element_address(array, offset, bt);
6732
6733 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6734 array_start, length, initialValue, basic_type)));
6735 clear_upper_avx();
6736
6737 return true;
6738 }
6739
6740 /**
6741 * Calculate CRC32 for byte.
6742 * int java.util.zip.CRC32.update(int crc, int b)
6743 */
6744 bool LibraryCallKit::inline_updateCRC32() {
6745 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6746 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6747 // no receiver since it is static method
6748 Node* crc = argument(0); // type: int
6749 Node* b = argument(1); // type: int
6750
6751 /*
6752 * int c = ~ crc;
6753 * b = timesXtoThe32[(b ^ c) & 0xFF];
6754 * b = b ^ (c >>> 8);
6755 * crc = ~b;
6756 */
6757
6758 Node* M1 = intcon(-1);
6759 crc = _gvn.transform(new XorINode(crc, M1));
6760 Node* result = _gvn.transform(new XorINode(crc, b));
6761 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6762
6763 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6764 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6765 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6766 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6767
6768 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6769 result = _gvn.transform(new XorINode(crc, result));
6770 result = _gvn.transform(new XorINode(result, M1));
6771 set_result(result);
6772 return true;
6773 }
6774
6775 /**
6776 * Calculate CRC32 for byte[] array.
6777 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6778 */
6779 bool LibraryCallKit::inline_updateBytesCRC32() {
6780 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6781 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6782 // no receiver since it is static method
6783 Node* crc = argument(0); // type: int
6784 Node* src = argument(1); // type: oop
6785 Node* offset = argument(2); // type: int
6786 Node* length = argument(3); // type: int
6787
6788 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6789 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6790 // failed array check
6791 return false;
6792 }
6793
6794 // Figure out the size and type of the elements we will be copying.
6795 BasicType src_elem = src_type->elem()->array_element_basic_type();
6796 if (src_elem != T_BYTE) {
6797 return false;
6798 }
6799
6800 // 'src_start' points to src array + scaled offset
6801 src = must_be_not_null(src, true);
6802 Node* src_start = array_element_address(src, offset, src_elem);
6803
6804 // We assume that range check is done by caller.
6805 // TODO: generate range check (offset+length < src.length) in debug VM.
6806
6807 // Call the stub.
6808 address stubAddr = StubRoutines::updateBytesCRC32();
6809 const char *stubName = "updateBytesCRC32";
6810
6811 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6812 stubAddr, stubName, TypePtr::BOTTOM,
6813 crc, src_start, length);
6814 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6815 set_result(result);
6816 return true;
6817 }
6818
6819 /**
6820 * Calculate CRC32 for ByteBuffer.
6821 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6822 */
6823 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6824 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6825 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6826 // no receiver since it is static method
6827 Node* crc = argument(0); // type: int
6828 Node* src = argument(1); // type: long
6829 Node* offset = argument(3); // type: int
6830 Node* length = argument(4); // type: int
6831
6832 src = ConvL2X(src); // adjust Java long to machine word
6833 Node* base = _gvn.transform(new CastX2PNode(src));
6834 offset = ConvI2X(offset);
6835
6836 // 'src_start' points to src array + scaled offset
6837 Node* src_start = basic_plus_adr(top(), base, offset);
6838
6839 // Call the stub.
6840 address stubAddr = StubRoutines::updateBytesCRC32();
6841 const char *stubName = "updateBytesCRC32";
6842
6843 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6844 stubAddr, stubName, TypePtr::BOTTOM,
6845 crc, src_start, length);
6846 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6847 set_result(result);
6848 return true;
6849 }
6850
6851 //------------------------------get_table_from_crc32c_class-----------------------
6852 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6853 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6854 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6855
6856 return table;
6857 }
6858
6859 //------------------------------inline_updateBytesCRC32C-----------------------
6860 //
6861 // Calculate CRC32C for byte[] array.
6862 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6863 //
6864 bool LibraryCallKit::inline_updateBytesCRC32C() {
6865 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6866 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6867 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6868 // no receiver since it is a static method
6869 Node* crc = argument(0); // type: int
6870 Node* src = argument(1); // type: oop
6871 Node* offset = argument(2); // type: int
6872 Node* end = argument(3); // type: int
6873
6874 Node* length = _gvn.transform(new SubINode(end, offset));
6875
6876 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6877 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6878 // failed array check
6879 return false;
6880 }
6881
6882 // Figure out the size and type of the elements we will be copying.
6883 BasicType src_elem = src_type->elem()->array_element_basic_type();
6884 if (src_elem != T_BYTE) {
6885 return false;
6886 }
6887
6888 // 'src_start' points to src array + scaled offset
6889 src = must_be_not_null(src, true);
6890 Node* src_start = array_element_address(src, offset, src_elem);
6891
6892 // static final int[] byteTable in class CRC32C
6893 Node* table = get_table_from_crc32c_class(callee()->holder());
6894 table = must_be_not_null(table, true);
6895 Node* table_start = array_element_address(table, intcon(0), T_INT);
6896
6897 // We assume that range check is done by caller.
6898 // TODO: generate range check (offset+length < src.length) in debug VM.
6899
6900 // Call the stub.
6901 address stubAddr = StubRoutines::updateBytesCRC32C();
6902 const char *stubName = "updateBytesCRC32C";
6903
6904 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6905 stubAddr, stubName, TypePtr::BOTTOM,
6906 crc, src_start, length, table_start);
6907 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6908 set_result(result);
6909 return true;
6910 }
6911
6912 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6913 //
6914 // Calculate CRC32C for DirectByteBuffer.
6915 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6916 //
6917 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6918 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6919 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6920 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6921 // no receiver since it is a static method
6922 Node* crc = argument(0); // type: int
6923 Node* src = argument(1); // type: long
6924 Node* offset = argument(3); // type: int
6925 Node* end = argument(4); // type: int
6926
6927 Node* length = _gvn.transform(new SubINode(end, offset));
6928
6929 src = ConvL2X(src); // adjust Java long to machine word
6930 Node* base = _gvn.transform(new CastX2PNode(src));
6931 offset = ConvI2X(offset);
6932
6933 // 'src_start' points to src array + scaled offset
6934 Node* src_start = basic_plus_adr(top(), base, offset);
6935
6936 // static final int[] byteTable in class CRC32C
6937 Node* table = get_table_from_crc32c_class(callee()->holder());
6938 table = must_be_not_null(table, true);
6939 Node* table_start = array_element_address(table, intcon(0), T_INT);
6940
6941 // Call the stub.
6942 address stubAddr = StubRoutines::updateBytesCRC32C();
6943 const char *stubName = "updateBytesCRC32C";
6944
6945 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6946 stubAddr, stubName, TypePtr::BOTTOM,
6947 crc, src_start, length, table_start);
6948 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6949 set_result(result);
6950 return true;
6951 }
6952
6953 //------------------------------inline_updateBytesAdler32----------------------
6954 //
6955 // Calculate Adler32 checksum for byte[] array.
6956 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6957 //
6958 bool LibraryCallKit::inline_updateBytesAdler32() {
6959 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6960 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6961 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6962 // no receiver since it is static method
6963 Node* crc = argument(0); // type: int
6964 Node* src = argument(1); // type: oop
6965 Node* offset = argument(2); // type: int
6966 Node* length = argument(3); // type: int
6967
6968 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6969 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6970 // failed array check
6971 return false;
6972 }
6973
6974 // Figure out the size and type of the elements we will be copying.
6975 BasicType src_elem = src_type->elem()->array_element_basic_type();
6976 if (src_elem != T_BYTE) {
6977 return false;
6978 }
6979
6980 // 'src_start' points to src array + scaled offset
6981 Node* src_start = array_element_address(src, offset, src_elem);
6982
6983 // We assume that range check is done by caller.
6984 // TODO: generate range check (offset+length < src.length) in debug VM.
6985
6986 // Call the stub.
6987 address stubAddr = StubRoutines::updateBytesAdler32();
6988 const char *stubName = "updateBytesAdler32";
6989
6990 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6991 stubAddr, stubName, TypePtr::BOTTOM,
6992 crc, src_start, length);
6993 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6994 set_result(result);
6995 return true;
6996 }
6997
6998 //------------------------------inline_updateByteBufferAdler32---------------
6999 //
7000 // Calculate Adler32 checksum for DirectByteBuffer.
7001 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
7002 //
7003 bool LibraryCallKit::inline_updateByteBufferAdler32() {
7004 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7005 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7006 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7007 // no receiver since it is static method
7008 Node* crc = argument(0); // type: int
7009 Node* src = argument(1); // type: long
7010 Node* offset = argument(3); // type: int
7011 Node* length = argument(4); // type: int
7012
7013 src = ConvL2X(src); // adjust Java long to machine word
7014 Node* base = _gvn.transform(new CastX2PNode(src));
7015 offset = ConvI2X(offset);
7016
7017 // 'src_start' points to src array + scaled offset
7018 Node* src_start = basic_plus_adr(top(), base, offset);
7019
7020 // Call the stub.
7021 address stubAddr = StubRoutines::updateBytesAdler32();
7022 const char *stubName = "updateBytesAdler32";
7023
7024 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7025 stubAddr, stubName, TypePtr::BOTTOM,
7026 crc, src_start, length);
7027
7028 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7029 set_result(result);
7030 return true;
7031 }
7032
7033 //----------------------------inline_reference_get----------------------------
7034 // public T java.lang.ref.Reference.get();
7035 bool LibraryCallKit::inline_reference_get() {
7036 const int referent_offset = java_lang_ref_Reference::referent_offset();
7037
7038 // Get the argument:
7039 Node* reference_obj = null_check_receiver();
7040 if (stopped()) return true;
7041
7042 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7043 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7044 decorators, /*is_static*/ false, nullptr);
7045 if (result == nullptr) return false;
7046
7047 // Add memory barrier to prevent commoning reads from this field
7048 // across safepoint since GC can change its value.
7049 insert_mem_bar(Op_MemBarCPUOrder);
7050
7051 set_result(result);
7052 return true;
7053 }
7054
7055 //----------------------------inline_reference_refersTo0----------------------------
7056 // bool java.lang.ref.Reference.refersTo0();
7057 // bool java.lang.ref.PhantomReference.refersTo0();
7058 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7059 // Get arguments:
7060 Node* reference_obj = null_check_receiver();
7061 Node* other_obj = argument(1);
7062 if (stopped()) return true;
7063
7064 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7065 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7066 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7067 decorators, /*is_static*/ false, nullptr);
7068 if (referent == nullptr) return false;
7069
7070 // Add memory barrier to prevent commoning reads from this field
7071 // across safepoint since GC can change its value.
7072 insert_mem_bar(Op_MemBarCPUOrder);
7073
7074 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7075 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7076 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7077
7078 RegionNode* region = new RegionNode(3);
7079 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7080
7081 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7082 region->init_req(1, if_true);
7083 phi->init_req(1, intcon(1));
7084
7085 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7086 region->init_req(2, if_false);
7087 phi->init_req(2, intcon(0));
7088
7089 set_control(_gvn.transform(region));
7090 record_for_igvn(region);
7091 set_result(_gvn.transform(phi));
7092 return true;
7093 }
7094
7095 //----------------------------inline_reference_clear0----------------------------
7096 // void java.lang.ref.Reference.clear0();
7097 // void java.lang.ref.PhantomReference.clear0();
7098 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7099 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7100
7101 // Get arguments
7102 Node* reference_obj = null_check_receiver();
7103 if (stopped()) return true;
7104
7105 // Common access parameters
7106 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7107 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7108 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7109 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7110 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7111
7112 Node* referent = access_load_at(reference_obj,
7113 referent_field_addr,
7114 referent_field_addr_type,
7115 val_type,
7116 T_OBJECT,
7117 decorators);
7118
7119 IdealKit ideal(this);
7120 #define __ ideal.
7121 __ if_then(referent, BoolTest::ne, null());
7122 sync_kit(ideal);
7123 access_store_at(reference_obj,
7124 referent_field_addr,
7125 referent_field_addr_type,
7126 null(),
7127 val_type,
7128 T_OBJECT,
7129 decorators);
7130 __ sync_kit(this);
7131 __ end_if();
7132 final_sync(ideal);
7133 #undef __
7134
7135 return true;
7136 }
7137
7138 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7139 DecoratorSet decorators, bool is_static,
7140 ciInstanceKlass* fromKls) {
7141 if (fromKls == nullptr) {
7142 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7143 assert(tinst != nullptr, "obj is null");
7144 assert(tinst->is_loaded(), "obj is not loaded");
7145 fromKls = tinst->instance_klass();
7146 } else {
7147 assert(is_static, "only for static field access");
7148 }
7149 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7150 ciSymbol::make(fieldTypeString),
7151 is_static);
7152
7153 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7154 if (field == nullptr) return (Node *) nullptr;
7155
7156 if (is_static) {
7157 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7158 fromObj = makecon(tip);
7159 }
7160
7161 // Next code copied from Parse::do_get_xxx():
7162
7163 // Compute address and memory type.
7164 int offset = field->offset_in_bytes();
7165 bool is_vol = field->is_volatile();
7166 ciType* field_klass = field->type();
7167 assert(field_klass->is_loaded(), "should be loaded");
7168 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7169 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7170 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7171 "slice of address and input slice don't match");
7172 BasicType bt = field->layout_type();
7173
7174 // Build the resultant type of the load
7175 const Type *type;
7176 if (bt == T_OBJECT) {
7177 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7178 } else {
7179 type = Type::get_const_basic_type(bt);
7180 }
7181
7182 if (is_vol) {
7183 decorators |= MO_SEQ_CST;
7184 }
7185
7186 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7187 }
7188
7189 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7190 bool is_exact /* true */, bool is_static /* false */,
7191 ciInstanceKlass * fromKls /* nullptr */) {
7192 if (fromKls == nullptr) {
7193 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7194 assert(tinst != nullptr, "obj is null");
7195 assert(tinst->is_loaded(), "obj is not loaded");
7196 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7197 fromKls = tinst->instance_klass();
7198 }
7199 else {
7200 assert(is_static, "only for static field access");
7201 }
7202 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7203 ciSymbol::make(fieldTypeString),
7204 is_static);
7205
7206 assert(field != nullptr, "undefined field");
7207 assert(!field->is_volatile(), "not defined for volatile fields");
7208
7209 if (is_static) {
7210 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7211 fromObj = makecon(tip);
7212 }
7213
7214 // Next code copied from Parse::do_get_xxx():
7215
7216 // Compute address and memory type.
7217 int offset = field->offset_in_bytes();
7218 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7219
7220 return adr;
7221 }
7222
7223 //------------------------------inline_aescrypt_Block-----------------------
7224 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7225 address stubAddr = nullptr;
7226 const char *stubName;
7227 assert(UseAES, "need AES instruction support");
7228
7229 switch(id) {
7230 case vmIntrinsics::_aescrypt_encryptBlock:
7231 stubAddr = StubRoutines::aescrypt_encryptBlock();
7232 stubName = "aescrypt_encryptBlock";
7233 break;
7234 case vmIntrinsics::_aescrypt_decryptBlock:
7235 stubAddr = StubRoutines::aescrypt_decryptBlock();
7236 stubName = "aescrypt_decryptBlock";
7237 break;
7238 default:
7239 break;
7240 }
7241 if (stubAddr == nullptr) return false;
7242
7243 Node* aescrypt_object = argument(0);
7244 Node* src = argument(1);
7245 Node* src_offset = argument(2);
7246 Node* dest = argument(3);
7247 Node* dest_offset = argument(4);
7248
7249 src = must_be_not_null(src, true);
7250 dest = must_be_not_null(dest, true);
7251
7252 // (1) src and dest are arrays.
7253 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7254 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7255 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7256 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7257
7258 // for the quick and dirty code we will skip all the checks.
7259 // we are just trying to get the call to be generated.
7260 Node* src_start = src;
7261 Node* dest_start = dest;
7262 if (src_offset != nullptr || dest_offset != nullptr) {
7263 assert(src_offset != nullptr && dest_offset != nullptr, "");
7264 src_start = array_element_address(src, src_offset, T_BYTE);
7265 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7266 }
7267
7268 // now need to get the start of its expanded key array
7269 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7270 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7271 if (k_start == nullptr) return false;
7272
7273 // Call the stub.
7274 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7275 stubAddr, stubName, TypePtr::BOTTOM,
7276 src_start, dest_start, k_start);
7277
7278 return true;
7279 }
7280
7281 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7282 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7283 address stubAddr = nullptr;
7284 const char *stubName = nullptr;
7285
7286 assert(UseAES, "need AES instruction support");
7287
7288 switch(id) {
7289 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7290 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7291 stubName = "cipherBlockChaining_encryptAESCrypt";
7292 break;
7293 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7294 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7295 stubName = "cipherBlockChaining_decryptAESCrypt";
7296 break;
7297 default:
7298 break;
7299 }
7300 if (stubAddr == nullptr) return false;
7301
7302 Node* cipherBlockChaining_object = argument(0);
7303 Node* src = argument(1);
7304 Node* src_offset = argument(2);
7305 Node* len = argument(3);
7306 Node* dest = argument(4);
7307 Node* dest_offset = argument(5);
7308
7309 src = must_be_not_null(src, false);
7310 dest = must_be_not_null(dest, false);
7311
7312 // (1) src and dest are arrays.
7313 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7314 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7315 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7316 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7317
7318 // checks are the responsibility of the caller
7319 Node* src_start = src;
7320 Node* dest_start = dest;
7321 if (src_offset != nullptr || dest_offset != nullptr) {
7322 assert(src_offset != nullptr && dest_offset != nullptr, "");
7323 src_start = array_element_address(src, src_offset, T_BYTE);
7324 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7325 }
7326
7327 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7328 // (because of the predicated logic executed earlier).
7329 // so we cast it here safely.
7330 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7331
7332 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7333 if (embeddedCipherObj == nullptr) return false;
7334
7335 // cast it to what we know it will be at runtime
7336 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7337 assert(tinst != nullptr, "CBC obj is null");
7338 assert(tinst->is_loaded(), "CBC obj is not loaded");
7339 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7340 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7341
7342 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7343 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7344 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7345 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7346 aescrypt_object = _gvn.transform(aescrypt_object);
7347
7348 // we need to get the start of the aescrypt_object's expanded key array
7349 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7350 if (k_start == nullptr) return false;
7351
7352 // similarly, get the start address of the r vector
7353 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7354 if (objRvec == nullptr) return false;
7355 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7356
7357 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7358 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7359 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7360 stubAddr, stubName, TypePtr::BOTTOM,
7361 src_start, dest_start, k_start, r_start, len);
7362
7363 // return cipher length (int)
7364 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7365 set_result(retvalue);
7366 return true;
7367 }
7368
7369 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7370 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7371 address stubAddr = nullptr;
7372 const char *stubName = nullptr;
7373
7374 assert(UseAES, "need AES instruction support");
7375
7376 switch (id) {
7377 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7378 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7379 stubName = "electronicCodeBook_encryptAESCrypt";
7380 break;
7381 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7382 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7383 stubName = "electronicCodeBook_decryptAESCrypt";
7384 break;
7385 default:
7386 break;
7387 }
7388
7389 if (stubAddr == nullptr) return false;
7390
7391 Node* electronicCodeBook_object = argument(0);
7392 Node* src = argument(1);
7393 Node* src_offset = argument(2);
7394 Node* len = argument(3);
7395 Node* dest = argument(4);
7396 Node* dest_offset = argument(5);
7397
7398 // (1) src and dest are arrays.
7399 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7400 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7401 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7402 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7403
7404 // checks are the responsibility of the caller
7405 Node* src_start = src;
7406 Node* dest_start = dest;
7407 if (src_offset != nullptr || dest_offset != nullptr) {
7408 assert(src_offset != nullptr && dest_offset != nullptr, "");
7409 src_start = array_element_address(src, src_offset, T_BYTE);
7410 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7411 }
7412
7413 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7414 // (because of the predicated logic executed earlier).
7415 // so we cast it here safely.
7416 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7417
7418 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7419 if (embeddedCipherObj == nullptr) return false;
7420
7421 // cast it to what we know it will be at runtime
7422 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7423 assert(tinst != nullptr, "ECB obj is null");
7424 assert(tinst->is_loaded(), "ECB obj is not loaded");
7425 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7426 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7427
7428 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7429 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7430 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7431 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7432 aescrypt_object = _gvn.transform(aescrypt_object);
7433
7434 // we need to get the start of the aescrypt_object's expanded key array
7435 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7436 if (k_start == nullptr) return false;
7437
7438 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7439 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7440 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7441 stubAddr, stubName, TypePtr::BOTTOM,
7442 src_start, dest_start, k_start, len);
7443
7444 // return cipher length (int)
7445 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7446 set_result(retvalue);
7447 return true;
7448 }
7449
7450 //------------------------------inline_counterMode_AESCrypt-----------------------
7451 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7452 assert(UseAES, "need AES instruction support");
7453 if (!UseAESCTRIntrinsics) return false;
7454
7455 address stubAddr = nullptr;
7456 const char *stubName = nullptr;
7457 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7458 stubAddr = StubRoutines::counterMode_AESCrypt();
7459 stubName = "counterMode_AESCrypt";
7460 }
7461 if (stubAddr == nullptr) return false;
7462
7463 Node* counterMode_object = argument(0);
7464 Node* src = argument(1);
7465 Node* src_offset = argument(2);
7466 Node* len = argument(3);
7467 Node* dest = argument(4);
7468 Node* dest_offset = argument(5);
7469
7470 // (1) src and dest are arrays.
7471 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7472 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7473 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7474 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7475
7476 // checks are the responsibility of the caller
7477 Node* src_start = src;
7478 Node* dest_start = dest;
7479 if (src_offset != nullptr || dest_offset != nullptr) {
7480 assert(src_offset != nullptr && dest_offset != nullptr, "");
7481 src_start = array_element_address(src, src_offset, T_BYTE);
7482 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7483 }
7484
7485 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7486 // (because of the predicated logic executed earlier).
7487 // so we cast it here safely.
7488 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7489 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7490 if (embeddedCipherObj == nullptr) return false;
7491 // cast it to what we know it will be at runtime
7492 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7493 assert(tinst != nullptr, "CTR obj is null");
7494 assert(tinst->is_loaded(), "CTR obj is not loaded");
7495 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7496 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7497 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7498 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7499 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7500 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7501 aescrypt_object = _gvn.transform(aescrypt_object);
7502 // we need to get the start of the aescrypt_object's expanded key array
7503 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7504 if (k_start == nullptr) return false;
7505 // similarly, get the start address of the r vector
7506 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7507 if (obj_counter == nullptr) return false;
7508 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7509
7510 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7511 if (saved_encCounter == nullptr) return false;
7512 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7513 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7514
7515 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7516 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7517 OptoRuntime::counterMode_aescrypt_Type(),
7518 stubAddr, stubName, TypePtr::BOTTOM,
7519 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7520
7521 // return cipher length (int)
7522 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7523 set_result(retvalue);
7524 return true;
7525 }
7526
7527 //------------------------------get_key_start_from_aescrypt_object-----------------------
7528 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7529 #if defined(PPC64) || defined(S390) || defined(RISCV64)
7530 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7531 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7532 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7533 // The ppc64 and riscv64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7534 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7535 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7536 if (objSessionK == nullptr) {
7537 return (Node *) nullptr;
7538 }
7539 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7540 #else
7541 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7542 #endif // PPC64
7543 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7544 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7545
7546 // now have the array, need to get the start address of the K array
7547 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7548 return k_start;
7549 }
7550
7551 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7552 // Return node representing slow path of predicate check.
7553 // the pseudo code we want to emulate with this predicate is:
7554 // for encryption:
7555 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7556 // for decryption:
7557 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7558 // note cipher==plain is more conservative than the original java code but that's OK
7559 //
7560 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7561 // The receiver was checked for null already.
7562 Node* objCBC = argument(0);
7563
7564 Node* src = argument(1);
7565 Node* dest = argument(4);
7566
7567 // Load embeddedCipher field of CipherBlockChaining object.
7568 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7569
7570 // get AESCrypt klass for instanceOf check
7571 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7572 // will have same classloader as CipherBlockChaining object
7573 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7574 assert(tinst != nullptr, "CBCobj is null");
7575 assert(tinst->is_loaded(), "CBCobj is not loaded");
7576
7577 // we want to do an instanceof comparison against the AESCrypt class
7578 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7579 if (!klass_AESCrypt->is_loaded()) {
7580 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7581 Node* ctrl = control();
7582 set_control(top()); // no regular fast path
7583 return ctrl;
7584 }
7585
7586 src = must_be_not_null(src, true);
7587 dest = must_be_not_null(dest, true);
7588
7589 // Resolve oops to stable for CmpP below.
7590 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7591
7592 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7593 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7594 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7595
7596 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7597
7598 // for encryption, we are done
7599 if (!decrypting)
7600 return instof_false; // even if it is null
7601
7602 // for decryption, we need to add a further check to avoid
7603 // taking the intrinsic path when cipher and plain are the same
7604 // see the original java code for why.
7605 RegionNode* region = new RegionNode(3);
7606 region->init_req(1, instof_false);
7607
7608 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7609 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7610 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7611 region->init_req(2, src_dest_conjoint);
7612
7613 record_for_igvn(region);
7614 return _gvn.transform(region);
7615 }
7616
7617 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7618 // Return node representing slow path of predicate check.
7619 // the pseudo code we want to emulate with this predicate is:
7620 // for encryption:
7621 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7622 // for decryption:
7623 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7624 // note cipher==plain is more conservative than the original java code but that's OK
7625 //
7626 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7627 // The receiver was checked for null already.
7628 Node* objECB = argument(0);
7629
7630 // Load embeddedCipher field of ElectronicCodeBook object.
7631 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7632
7633 // get AESCrypt klass for instanceOf check
7634 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7635 // will have same classloader as ElectronicCodeBook object
7636 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7637 assert(tinst != nullptr, "ECBobj is null");
7638 assert(tinst->is_loaded(), "ECBobj is not loaded");
7639
7640 // we want to do an instanceof comparison against the AESCrypt class
7641 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7642 if (!klass_AESCrypt->is_loaded()) {
7643 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7644 Node* ctrl = control();
7645 set_control(top()); // no regular fast path
7646 return ctrl;
7647 }
7648 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7649
7650 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7651 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7652 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7653
7654 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7655
7656 // for encryption, we are done
7657 if (!decrypting)
7658 return instof_false; // even if it is null
7659
7660 // for decryption, we need to add a further check to avoid
7661 // taking the intrinsic path when cipher and plain are the same
7662 // see the original java code for why.
7663 RegionNode* region = new RegionNode(3);
7664 region->init_req(1, instof_false);
7665 Node* src = argument(1);
7666 Node* dest = argument(4);
7667 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7668 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7669 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7670 region->init_req(2, src_dest_conjoint);
7671
7672 record_for_igvn(region);
7673 return _gvn.transform(region);
7674 }
7675
7676 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7677 // Return node representing slow path of predicate check.
7678 // the pseudo code we want to emulate with this predicate is:
7679 // for encryption:
7680 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7681 // for decryption:
7682 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7683 // note cipher==plain is more conservative than the original java code but that's OK
7684 //
7685
7686 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7687 // The receiver was checked for null already.
7688 Node* objCTR = argument(0);
7689
7690 // Load embeddedCipher field of CipherBlockChaining object.
7691 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7692
7693 // get AESCrypt klass for instanceOf check
7694 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7695 // will have same classloader as CipherBlockChaining object
7696 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7697 assert(tinst != nullptr, "CTRobj is null");
7698 assert(tinst->is_loaded(), "CTRobj is not loaded");
7699
7700 // we want to do an instanceof comparison against the AESCrypt class
7701 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7702 if (!klass_AESCrypt->is_loaded()) {
7703 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7704 Node* ctrl = control();
7705 set_control(top()); // no regular fast path
7706 return ctrl;
7707 }
7708
7709 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7710 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7711 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7712 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7713 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7714
7715 return instof_false; // even if it is null
7716 }
7717
7718 //------------------------------inline_ghash_processBlocks
7719 bool LibraryCallKit::inline_ghash_processBlocks() {
7720 address stubAddr;
7721 const char *stubName;
7722 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7723
7724 stubAddr = StubRoutines::ghash_processBlocks();
7725 stubName = "ghash_processBlocks";
7726
7727 Node* data = argument(0);
7728 Node* offset = argument(1);
7729 Node* len = argument(2);
7730 Node* state = argument(3);
7731 Node* subkeyH = argument(4);
7732
7733 state = must_be_not_null(state, true);
7734 subkeyH = must_be_not_null(subkeyH, true);
7735 data = must_be_not_null(data, true);
7736
7737 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7738 assert(state_start, "state is null");
7739 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7740 assert(subkeyH_start, "subkeyH is null");
7741 Node* data_start = array_element_address(data, offset, T_BYTE);
7742 assert(data_start, "data is null");
7743
7744 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7745 OptoRuntime::ghash_processBlocks_Type(),
7746 stubAddr, stubName, TypePtr::BOTTOM,
7747 state_start, subkeyH_start, data_start, len);
7748 return true;
7749 }
7750
7751 //------------------------------inline_chacha20Block
7752 bool LibraryCallKit::inline_chacha20Block() {
7753 address stubAddr;
7754 const char *stubName;
7755 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7756
7757 stubAddr = StubRoutines::chacha20Block();
7758 stubName = "chacha20Block";
7759
7760 Node* state = argument(0);
7761 Node* result = argument(1);
7762
7763 state = must_be_not_null(state, true);
7764 result = must_be_not_null(result, true);
7765
7766 Node* state_start = array_element_address(state, intcon(0), T_INT);
7767 assert(state_start, "state is null");
7768 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7769 assert(result_start, "result is null");
7770
7771 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7772 OptoRuntime::chacha20Block_Type(),
7773 stubAddr, stubName, TypePtr::BOTTOM,
7774 state_start, result_start);
7775 // return key stream length (int)
7776 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7777 set_result(retvalue);
7778 return true;
7779 }
7780
7781 bool LibraryCallKit::inline_base64_encodeBlock() {
7782 address stubAddr;
7783 const char *stubName;
7784 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7785 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
7786 stubAddr = StubRoutines::base64_encodeBlock();
7787 stubName = "encodeBlock";
7788
7789 if (!stubAddr) return false;
7790 Node* base64obj = argument(0);
7791 Node* src = argument(1);
7792 Node* offset = argument(2);
7793 Node* len = argument(3);
7794 Node* dest = argument(4);
7795 Node* dp = argument(5);
7796 Node* isURL = argument(6);
7797
7798 src = must_be_not_null(src, true);
7799 dest = must_be_not_null(dest, true);
7800
7801 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7802 assert(src_start, "source array is null");
7803 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7804 assert(dest_start, "destination array is null");
7805
7806 Node* base64 = make_runtime_call(RC_LEAF,
7807 OptoRuntime::base64_encodeBlock_Type(),
7808 stubAddr, stubName, TypePtr::BOTTOM,
7809 src_start, offset, len, dest_start, dp, isURL);
7810 return true;
7811 }
7812
7813 bool LibraryCallKit::inline_base64_decodeBlock() {
7814 address stubAddr;
7815 const char *stubName;
7816 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7817 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
7818 stubAddr = StubRoutines::base64_decodeBlock();
7819 stubName = "decodeBlock";
7820
7821 if (!stubAddr) return false;
7822 Node* base64obj = argument(0);
7823 Node* src = argument(1);
7824 Node* src_offset = argument(2);
7825 Node* len = argument(3);
7826 Node* dest = argument(4);
7827 Node* dest_offset = argument(5);
7828 Node* isURL = argument(6);
7829 Node* isMIME = argument(7);
7830
7831 src = must_be_not_null(src, true);
7832 dest = must_be_not_null(dest, true);
7833
7834 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7835 assert(src_start, "source array is null");
7836 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7837 assert(dest_start, "destination array is null");
7838
7839 Node* call = make_runtime_call(RC_LEAF,
7840 OptoRuntime::base64_decodeBlock_Type(),
7841 stubAddr, stubName, TypePtr::BOTTOM,
7842 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
7843 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7844 set_result(result);
7845 return true;
7846 }
7847
7848 bool LibraryCallKit::inline_poly1305_processBlocks() {
7849 address stubAddr;
7850 const char *stubName;
7851 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
7852 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
7853 stubAddr = StubRoutines::poly1305_processBlocks();
7854 stubName = "poly1305_processBlocks";
7855
7856 if (!stubAddr) return false;
7857 null_check_receiver(); // null-check receiver
7858 if (stopped()) return true;
7859
7860 Node* input = argument(1);
7861 Node* input_offset = argument(2);
7862 Node* len = argument(3);
7863 Node* alimbs = argument(4);
7864 Node* rlimbs = argument(5);
7865
7866 input = must_be_not_null(input, true);
7867 alimbs = must_be_not_null(alimbs, true);
7868 rlimbs = must_be_not_null(rlimbs, true);
7869
7870 Node* input_start = array_element_address(input, input_offset, T_BYTE);
7871 assert(input_start, "input array is null");
7872 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
7873 assert(acc_start, "acc array is null");
7874 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
7875 assert(r_start, "r array is null");
7876
7877 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7878 OptoRuntime::poly1305_processBlocks_Type(),
7879 stubAddr, stubName, TypePtr::BOTTOM,
7880 input_start, len, acc_start, r_start);
7881 return true;
7882 }
7883
7884 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
7885 address stubAddr;
7886 const char *stubName;
7887 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
7888 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
7889 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
7890 stubName = "intpoly_montgomeryMult_P256";
7891
7892 if (!stubAddr) return false;
7893 null_check_receiver(); // null-check receiver
7894 if (stopped()) return true;
7895
7896 Node* a = argument(1);
7897 Node* b = argument(2);
7898 Node* r = argument(3);
7899
7900 a = must_be_not_null(a, true);
7901 b = must_be_not_null(b, true);
7902 r = must_be_not_null(r, true);
7903
7904 Node* a_start = array_element_address(a, intcon(0), T_LONG);
7905 assert(a_start, "a array is null");
7906 Node* b_start = array_element_address(b, intcon(0), T_LONG);
7907 assert(b_start, "b array is null");
7908 Node* r_start = array_element_address(r, intcon(0), T_LONG);
7909 assert(r_start, "r array is null");
7910
7911 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7912 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
7913 stubAddr, stubName, TypePtr::BOTTOM,
7914 a_start, b_start, r_start);
7915 return true;
7916 }
7917
7918 bool LibraryCallKit::inline_intpoly_assign() {
7919 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
7920 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
7921 const char *stubName = "intpoly_assign";
7922 address stubAddr = StubRoutines::intpoly_assign();
7923 if (!stubAddr) return false;
7924
7925 Node* set = argument(0);
7926 Node* a = argument(1);
7927 Node* b = argument(2);
7928 Node* arr_length = load_array_length(a);
7929
7930 a = must_be_not_null(a, true);
7931 b = must_be_not_null(b, true);
7932
7933 Node* a_start = array_element_address(a, intcon(0), T_LONG);
7934 assert(a_start, "a array is null");
7935 Node* b_start = array_element_address(b, intcon(0), T_LONG);
7936 assert(b_start, "b array is null");
7937
7938 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7939 OptoRuntime::intpoly_assign_Type(),
7940 stubAddr, stubName, TypePtr::BOTTOM,
7941 set, a_start, b_start, arr_length);
7942 return true;
7943 }
7944
7945 //------------------------------inline_digestBase_implCompress-----------------------
7946 //
7947 // Calculate MD5 for single-block byte[] array.
7948 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
7949 //
7950 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
7951 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
7952 //
7953 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
7954 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
7955 //
7956 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
7957 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
7958 //
7959 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
7960 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
7961 //
7962 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
7963 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
7964
7965 Node* digestBase_obj = argument(0);
7966 Node* src = argument(1); // type oop
7967 Node* ofs = argument(2); // type int
7968
7969 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7970 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7971 // failed array check
7972 return false;
7973 }
7974 // Figure out the size and type of the elements we will be copying.
7975 BasicType src_elem = src_type->elem()->array_element_basic_type();
7976 if (src_elem != T_BYTE) {
7977 return false;
7978 }
7979 // 'src_start' points to src array + offset
7980 src = must_be_not_null(src, true);
7981 Node* src_start = array_element_address(src, ofs, src_elem);
7982 Node* state = nullptr;
7983 Node* block_size = nullptr;
7984 address stubAddr;
7985 const char *stubName;
7986
7987 switch(id) {
7988 case vmIntrinsics::_md5_implCompress:
7989 assert(UseMD5Intrinsics, "need MD5 instruction support");
7990 state = get_state_from_digest_object(digestBase_obj, T_INT);
7991 stubAddr = StubRoutines::md5_implCompress();
7992 stubName = "md5_implCompress";
7993 break;
7994 case vmIntrinsics::_sha_implCompress:
7995 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
7996 state = get_state_from_digest_object(digestBase_obj, T_INT);
7997 stubAddr = StubRoutines::sha1_implCompress();
7998 stubName = "sha1_implCompress";
7999 break;
8000 case vmIntrinsics::_sha2_implCompress:
8001 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
8002 state = get_state_from_digest_object(digestBase_obj, T_INT);
8003 stubAddr = StubRoutines::sha256_implCompress();
8004 stubName = "sha256_implCompress";
8005 break;
8006 case vmIntrinsics::_sha5_implCompress:
8007 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
8008 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8009 stubAddr = StubRoutines::sha512_implCompress();
8010 stubName = "sha512_implCompress";
8011 break;
8012 case vmIntrinsics::_sha3_implCompress:
8013 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
8014 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8015 stubAddr = StubRoutines::sha3_implCompress();
8016 stubName = "sha3_implCompress";
8017 block_size = get_block_size_from_digest_object(digestBase_obj);
8018 if (block_size == nullptr) return false;
8019 break;
8020 default:
8021 fatal_unexpected_iid(id);
8022 return false;
8023 }
8024 if (state == nullptr) return false;
8025
8026 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
8027 if (stubAddr == nullptr) return false;
8028
8029 // Call the stub.
8030 Node* call;
8031 if (block_size == nullptr) {
8032 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
8033 stubAddr, stubName, TypePtr::BOTTOM,
8034 src_start, state);
8035 } else {
8036 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
8037 stubAddr, stubName, TypePtr::BOTTOM,
8038 src_start, state, block_size);
8039 }
8040
8041 return true;
8042 }
8043
8044 //------------------------------inline_digestBase_implCompressMB-----------------------
8045 //
8046 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
8047 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
8048 //
8049 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
8050 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8051 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8052 assert((uint)predicate < 5, "sanity");
8053 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
8054
8055 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
8056 Node* src = argument(1); // byte[] array
8057 Node* ofs = argument(2); // type int
8058 Node* limit = argument(3); // type int
8059
8060 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8061 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8062 // failed array check
8063 return false;
8064 }
8065 // Figure out the size and type of the elements we will be copying.
8066 BasicType src_elem = src_type->elem()->array_element_basic_type();
8067 if (src_elem != T_BYTE) {
8068 return false;
8069 }
8070 // 'src_start' points to src array + offset
8071 src = must_be_not_null(src, false);
8072 Node* src_start = array_element_address(src, ofs, src_elem);
8073
8074 const char* klass_digestBase_name = nullptr;
8075 const char* stub_name = nullptr;
8076 address stub_addr = nullptr;
8077 BasicType elem_type = T_INT;
8078
8079 switch (predicate) {
8080 case 0:
8081 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
8082 klass_digestBase_name = "sun/security/provider/MD5";
8083 stub_name = "md5_implCompressMB";
8084 stub_addr = StubRoutines::md5_implCompressMB();
8085 }
8086 break;
8087 case 1:
8088 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
8089 klass_digestBase_name = "sun/security/provider/SHA";
8090 stub_name = "sha1_implCompressMB";
8091 stub_addr = StubRoutines::sha1_implCompressMB();
8092 }
8093 break;
8094 case 2:
8095 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
8096 klass_digestBase_name = "sun/security/provider/SHA2";
8097 stub_name = "sha256_implCompressMB";
8098 stub_addr = StubRoutines::sha256_implCompressMB();
8099 }
8100 break;
8101 case 3:
8102 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
8103 klass_digestBase_name = "sun/security/provider/SHA5";
8104 stub_name = "sha512_implCompressMB";
8105 stub_addr = StubRoutines::sha512_implCompressMB();
8106 elem_type = T_LONG;
8107 }
8108 break;
8109 case 4:
8110 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
8111 klass_digestBase_name = "sun/security/provider/SHA3";
8112 stub_name = "sha3_implCompressMB";
8113 stub_addr = StubRoutines::sha3_implCompressMB();
8114 elem_type = T_LONG;
8115 }
8116 break;
8117 default:
8118 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
8119 }
8120 if (klass_digestBase_name != nullptr) {
8121 assert(stub_addr != nullptr, "Stub is generated");
8122 if (stub_addr == nullptr) return false;
8123
8124 // get DigestBase klass to lookup for SHA klass
8125 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
8126 assert(tinst != nullptr, "digestBase_obj is not instance???");
8127 assert(tinst->is_loaded(), "DigestBase is not loaded");
8128
8129 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
8130 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
8131 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
8132 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
8133 }
8134 return false;
8135 }
8136
8137 //------------------------------inline_digestBase_implCompressMB-----------------------
8138 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
8139 BasicType elem_type, address stubAddr, const char *stubName,
8140 Node* src_start, Node* ofs, Node* limit) {
8141 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
8142 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8143 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
8144 digest_obj = _gvn.transform(digest_obj);
8145
8146 Node* state = get_state_from_digest_object(digest_obj, elem_type);
8147 if (state == nullptr) return false;
8148
8149 Node* block_size = nullptr;
8150 if (strcmp("sha3_implCompressMB", stubName) == 0) {
8151 block_size = get_block_size_from_digest_object(digest_obj);
8152 if (block_size == nullptr) return false;
8153 }
8154
8155 // Call the stub.
8156 Node* call;
8157 if (block_size == nullptr) {
8158 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8159 OptoRuntime::digestBase_implCompressMB_Type(false),
8160 stubAddr, stubName, TypePtr::BOTTOM,
8161 src_start, state, ofs, limit);
8162 } else {
8163 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8164 OptoRuntime::digestBase_implCompressMB_Type(true),
8165 stubAddr, stubName, TypePtr::BOTTOM,
8166 src_start, state, block_size, ofs, limit);
8167 }
8168
8169 // return ofs (int)
8170 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8171 set_result(result);
8172
8173 return true;
8174 }
8175
8176 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8177 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8178 assert(UseAES, "need AES instruction support");
8179 address stubAddr = nullptr;
8180 const char *stubName = nullptr;
8181 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8182 stubName = "galoisCounterMode_AESCrypt";
8183
8184 if (stubAddr == nullptr) return false;
8185
8186 Node* in = argument(0);
8187 Node* inOfs = argument(1);
8188 Node* len = argument(2);
8189 Node* ct = argument(3);
8190 Node* ctOfs = argument(4);
8191 Node* out = argument(5);
8192 Node* outOfs = argument(6);
8193 Node* gctr_object = argument(7);
8194 Node* ghash_object = argument(8);
8195
8196 // (1) in, ct and out are arrays.
8197 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8198 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8199 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8200 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8201 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8202 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8203
8204 // checks are the responsibility of the caller
8205 Node* in_start = in;
8206 Node* ct_start = ct;
8207 Node* out_start = out;
8208 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8209 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8210 in_start = array_element_address(in, inOfs, T_BYTE);
8211 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8212 out_start = array_element_address(out, outOfs, T_BYTE);
8213 }
8214
8215 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8216 // (because of the predicated logic executed earlier).
8217 // so we cast it here safely.
8218 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8219 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8220 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8221 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8222 Node* state = load_field_from_object(ghash_object, "state", "[J");
8223
8224 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8225 return false;
8226 }
8227 // cast it to what we know it will be at runtime
8228 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8229 assert(tinst != nullptr, "GCTR obj is null");
8230 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8231 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8232 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8233 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8234 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8235 const TypeOopPtr* xtype = aklass->as_instance_type();
8236 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8237 aescrypt_object = _gvn.transform(aescrypt_object);
8238 // we need to get the start of the aescrypt_object's expanded key array
8239 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8240 if (k_start == nullptr) return false;
8241 // similarly, get the start address of the r vector
8242 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8243 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8244 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8245
8246
8247 // Call the stub, passing params
8248 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8249 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8250 stubAddr, stubName, TypePtr::BOTTOM,
8251 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8252
8253 // return cipher length (int)
8254 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8255 set_result(retvalue);
8256
8257 return true;
8258 }
8259
8260 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8261 // Return node representing slow path of predicate check.
8262 // the pseudo code we want to emulate with this predicate is:
8263 // for encryption:
8264 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8265 // for decryption:
8266 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8267 // note cipher==plain is more conservative than the original java code but that's OK
8268 //
8269
8270 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8271 // The receiver was checked for null already.
8272 Node* objGCTR = argument(7);
8273 // Load embeddedCipher field of GCTR object.
8274 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8275 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8276
8277 // get AESCrypt klass for instanceOf check
8278 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8279 // will have same classloader as CipherBlockChaining object
8280 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8281 assert(tinst != nullptr, "GCTR obj is null");
8282 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8283
8284 // we want to do an instanceof comparison against the AESCrypt class
8285 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8286 if (!klass_AESCrypt->is_loaded()) {
8287 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8288 Node* ctrl = control();
8289 set_control(top()); // no regular fast path
8290 return ctrl;
8291 }
8292
8293 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8294 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8295 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8296 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8297 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8298
8299 return instof_false; // even if it is null
8300 }
8301
8302 //------------------------------get_state_from_digest_object-----------------------
8303 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8304 const char* state_type;
8305 switch (elem_type) {
8306 case T_BYTE: state_type = "[B"; break;
8307 case T_INT: state_type = "[I"; break;
8308 case T_LONG: state_type = "[J"; break;
8309 default: ShouldNotReachHere();
8310 }
8311 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8312 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8313 if (digest_state == nullptr) return (Node *) nullptr;
8314
8315 // now have the array, need to get the start address of the state array
8316 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8317 return state;
8318 }
8319
8320 //------------------------------get_block_size_from_sha3_object----------------------------------
8321 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8322 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8323 assert (block_size != nullptr, "sanity");
8324 return block_size;
8325 }
8326
8327 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8328 // Return node representing slow path of predicate check.
8329 // the pseudo code we want to emulate with this predicate is:
8330 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8331 //
8332 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8333 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8334 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8335 assert((uint)predicate < 5, "sanity");
8336
8337 // The receiver was checked for null already.
8338 Node* digestBaseObj = argument(0);
8339
8340 // get DigestBase klass for instanceOf check
8341 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8342 assert(tinst != nullptr, "digestBaseObj is null");
8343 assert(tinst->is_loaded(), "DigestBase is not loaded");
8344
8345 const char* klass_name = nullptr;
8346 switch (predicate) {
8347 case 0:
8348 if (UseMD5Intrinsics) {
8349 // we want to do an instanceof comparison against the MD5 class
8350 klass_name = "sun/security/provider/MD5";
8351 }
8352 break;
8353 case 1:
8354 if (UseSHA1Intrinsics) {
8355 // we want to do an instanceof comparison against the SHA class
8356 klass_name = "sun/security/provider/SHA";
8357 }
8358 break;
8359 case 2:
8360 if (UseSHA256Intrinsics) {
8361 // we want to do an instanceof comparison against the SHA2 class
8362 klass_name = "sun/security/provider/SHA2";
8363 }
8364 break;
8365 case 3:
8366 if (UseSHA512Intrinsics) {
8367 // we want to do an instanceof comparison against the SHA5 class
8368 klass_name = "sun/security/provider/SHA5";
8369 }
8370 break;
8371 case 4:
8372 if (UseSHA3Intrinsics) {
8373 // we want to do an instanceof comparison against the SHA3 class
8374 klass_name = "sun/security/provider/SHA3";
8375 }
8376 break;
8377 default:
8378 fatal("unknown SHA intrinsic predicate: %d", predicate);
8379 }
8380
8381 ciKlass* klass = nullptr;
8382 if (klass_name != nullptr) {
8383 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8384 }
8385 if ((klass == nullptr) || !klass->is_loaded()) {
8386 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8387 Node* ctrl = control();
8388 set_control(top()); // no intrinsic path
8389 return ctrl;
8390 }
8391 ciInstanceKlass* instklass = klass->as_instance_klass();
8392
8393 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8394 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8395 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8396 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8397
8398 return instof_false; // even if it is null
8399 }
8400
8401 //-------------inline_fma-----------------------------------
8402 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8403 Node *a = nullptr;
8404 Node *b = nullptr;
8405 Node *c = nullptr;
8406 Node* result = nullptr;
8407 switch (id) {
8408 case vmIntrinsics::_fmaD:
8409 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8410 // no receiver since it is static method
8411 a = round_double_node(argument(0));
8412 b = round_double_node(argument(2));
8413 c = round_double_node(argument(4));
8414 result = _gvn.transform(new FmaDNode(a, b, c));
8415 break;
8416 case vmIntrinsics::_fmaF:
8417 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8418 a = argument(0);
8419 b = argument(1);
8420 c = argument(2);
8421 result = _gvn.transform(new FmaFNode(a, b, c));
8422 break;
8423 default:
8424 fatal_unexpected_iid(id); break;
8425 }
8426 set_result(result);
8427 return true;
8428 }
8429
8430 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8431 // argument(0) is receiver
8432 Node* codePoint = argument(1);
8433 Node* n = nullptr;
8434
8435 switch (id) {
8436 case vmIntrinsics::_isDigit :
8437 n = new DigitNode(control(), codePoint);
8438 break;
8439 case vmIntrinsics::_isLowerCase :
8440 n = new LowerCaseNode(control(), codePoint);
8441 break;
8442 case vmIntrinsics::_isUpperCase :
8443 n = new UpperCaseNode(control(), codePoint);
8444 break;
8445 case vmIntrinsics::_isWhitespace :
8446 n = new WhitespaceNode(control(), codePoint);
8447 break;
8448 default:
8449 fatal_unexpected_iid(id);
8450 }
8451
8452 set_result(_gvn.transform(n));
8453 return true;
8454 }
8455
8456 //------------------------------inline_fp_min_max------------------------------
8457 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
8458 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
8459
8460 // The intrinsic should be used only when the API branches aren't predictable,
8461 // the last one performing the most important comparison. The following heuristic
8462 // uses the branch statistics to eventually bail out if necessary.
8463
8464 ciMethodData *md = callee()->method_data();
8465
8466 if ( md != nullptr && md->is_mature() && md->invocation_count() > 0 ) {
8467 ciCallProfile cp = caller()->call_profile_at_bci(bci());
8468
8469 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
8470 // Bail out if the call-site didn't contribute enough to the statistics.
8471 return false;
8472 }
8473
8474 uint taken = 0, not_taken = 0;
8475
8476 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
8477 if (p->is_BranchData()) {
8478 taken = ((ciBranchData*)p)->taken();
8479 not_taken = ((ciBranchData*)p)->not_taken();
8480 }
8481 }
8482
8483 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
8484 balance = balance < 0 ? -balance : balance;
8485 if ( balance > 0.2 ) {
8486 // Bail out if the most important branch is predictable enough.
8487 return false;
8488 }
8489 }
8490 */
8491
8492 Node *a = nullptr;
8493 Node *b = nullptr;
8494 Node *n = nullptr;
8495 switch (id) {
8496 case vmIntrinsics::_maxF:
8497 case vmIntrinsics::_minF:
8498 case vmIntrinsics::_maxF_strict:
8499 case vmIntrinsics::_minF_strict:
8500 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
8501 a = argument(0);
8502 b = argument(1);
8503 break;
8504 case vmIntrinsics::_maxD:
8505 case vmIntrinsics::_minD:
8506 case vmIntrinsics::_maxD_strict:
8507 case vmIntrinsics::_minD_strict:
8508 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
8509 a = round_double_node(argument(0));
8510 b = round_double_node(argument(2));
8511 break;
8512 default:
8513 fatal_unexpected_iid(id);
8514 break;
8515 }
8516 switch (id) {
8517 case vmIntrinsics::_maxF:
8518 case vmIntrinsics::_maxF_strict:
8519 n = new MaxFNode(a, b);
8520 break;
8521 case vmIntrinsics::_minF:
8522 case vmIntrinsics::_minF_strict:
8523 n = new MinFNode(a, b);
8524 break;
8525 case vmIntrinsics::_maxD:
8526 case vmIntrinsics::_maxD_strict:
8527 n = new MaxDNode(a, b);
8528 break;
8529 case vmIntrinsics::_minD:
8530 case vmIntrinsics::_minD_strict:
8531 n = new MinDNode(a, b);
8532 break;
8533 default:
8534 fatal_unexpected_iid(id);
8535 break;
8536 }
8537 set_result(_gvn.transform(n));
8538 return true;
8539 }
8540
8541 bool LibraryCallKit::inline_profileBoolean() {
8542 Node* counts = argument(1);
8543 const TypeAryPtr* ary = nullptr;
8544 ciArray* aobj = nullptr;
8545 if (counts->is_Con()
8546 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8547 && (aobj = ary->const_oop()->as_array()) != nullptr
8548 && (aobj->length() == 2)) {
8549 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8550 jint false_cnt = aobj->element_value(0).as_int();
8551 jint true_cnt = aobj->element_value(1).as_int();
8552
8553 if (C->log() != nullptr) {
8554 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8555 false_cnt, true_cnt);
8556 }
8557
8558 if (false_cnt + true_cnt == 0) {
8559 // According to profile, never executed.
8560 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8561 Deoptimization::Action_reinterpret);
8562 return true;
8563 }
8564
8565 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8566 // is a number of each value occurrences.
8567 Node* result = argument(0);
8568 if (false_cnt == 0 || true_cnt == 0) {
8569 // According to profile, one value has been never seen.
8570 int expected_val = (false_cnt == 0) ? 1 : 0;
8571
8572 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8573 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8574
8575 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8576 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8577 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8578
8579 { // Slow path: uncommon trap for never seen value and then reexecute
8580 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8581 // the value has been seen at least once.
8582 PreserveJVMState pjvms(this);
8583 PreserveReexecuteState preexecs(this);
8584 jvms()->set_should_reexecute(true);
8585
8586 set_control(slow_path);
8587 set_i_o(i_o());
8588
8589 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8590 Deoptimization::Action_reinterpret);
8591 }
8592 // The guard for never seen value enables sharpening of the result and
8593 // returning a constant. It allows to eliminate branches on the same value
8594 // later on.
8595 set_control(fast_path);
8596 result = intcon(expected_val);
8597 }
8598 // Stop profiling.
8599 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8600 // By replacing method body with profile data (represented as ProfileBooleanNode
8601 // on IR level) we effectively disable profiling.
8602 // It enables full speed execution once optimized code is generated.
8603 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8604 C->record_for_igvn(profile);
8605 set_result(profile);
8606 return true;
8607 } else {
8608 // Continue profiling.
8609 // Profile data isn't available at the moment. So, execute method's bytecode version.
8610 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8611 // is compiled and counters aren't available since corresponding MethodHandle
8612 // isn't a compile-time constant.
8613 return false;
8614 }
8615 }
8616
8617 bool LibraryCallKit::inline_isCompileConstant() {
8618 Node* n = argument(0);
8619 set_result(n->is_Con() ? intcon(1) : intcon(0));
8620 return true;
8621 }
8622
8623 //------------------------------- inline_getObjectSize --------------------------------------
8624 //
8625 // Calculate the runtime size of the object/array.
8626 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8627 //
8628 bool LibraryCallKit::inline_getObjectSize() {
8629 Node* obj = argument(3);
8630 Node* klass_node = load_object_klass(obj);
8631
8632 jint layout_con = Klass::_lh_neutral_value;
8633 Node* layout_val = get_layout_helper(klass_node, layout_con);
8634 int layout_is_con = (layout_val == nullptr);
8635
8636 if (layout_is_con) {
8637 // Layout helper is constant, can figure out things at compile time.
8638
8639 if (Klass::layout_helper_is_instance(layout_con)) {
8640 // Instance case: layout_con contains the size itself.
8641 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8642 set_result(size);
8643 } else {
8644 // Array case: size is round(header + element_size*arraylength).
8645 // Since arraylength is different for every array instance, we have to
8646 // compute the whole thing at runtime.
8647
8648 Node* arr_length = load_array_length(obj);
8649
8650 int round_mask = MinObjAlignmentInBytes - 1;
8651 int hsize = Klass::layout_helper_header_size(layout_con);
8652 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8653
8654 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8655 round_mask = 0; // strength-reduce it if it goes away completely
8656 }
8657 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8658 Node* header_size = intcon(hsize + round_mask);
8659
8660 Node* lengthx = ConvI2X(arr_length);
8661 Node* headerx = ConvI2X(header_size);
8662
8663 Node* abody = lengthx;
8664 if (eshift != 0) {
8665 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8666 }
8667 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8668 if (round_mask != 0) {
8669 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8670 }
8671 size = ConvX2L(size);
8672 set_result(size);
8673 }
8674 } else {
8675 // Layout helper is not constant, need to test for array-ness at runtime.
8676
8677 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8678 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8679 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8680 record_for_igvn(result_reg);
8681
8682 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
8683 if (array_ctl != nullptr) {
8684 // Array case: size is round(header + element_size*arraylength).
8685 // Since arraylength is different for every array instance, we have to
8686 // compute the whole thing at runtime.
8687
8688 PreserveJVMState pjvms(this);
8689 set_control(array_ctl);
8690 Node* arr_length = load_array_length(obj);
8691
8692 int round_mask = MinObjAlignmentInBytes - 1;
8693 Node* mask = intcon(round_mask);
8694
8695 Node* hss = intcon(Klass::_lh_header_size_shift);
8696 Node* hsm = intcon(Klass::_lh_header_size_mask);
8697 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8698 header_size = _gvn.transform(new AndINode(header_size, hsm));
8699 header_size = _gvn.transform(new AddINode(header_size, mask));
8700
8701 // There is no need to mask or shift this value.
8702 // The semantics of LShiftINode include an implicit mask to 0x1F.
8703 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8704 Node* elem_shift = layout_val;
8705
8706 Node* lengthx = ConvI2X(arr_length);
8707 Node* headerx = ConvI2X(header_size);
8708
8709 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
8710 Node* size = _gvn.transform(new AddXNode(headerx, abody));
8711 if (round_mask != 0) {
8712 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
8713 }
8714 size = ConvX2L(size);
8715
8716 result_reg->init_req(_array_path, control());
8717 result_val->init_req(_array_path, size);
8718 }
8719
8720 if (!stopped()) {
8721 // Instance case: the layout helper gives us instance size almost directly,
8722 // but we need to mask out the _lh_instance_slow_path_bit.
8723 Node* size = ConvI2X(layout_val);
8724 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
8725 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
8726 size = _gvn.transform(new AndXNode(size, mask));
8727 size = ConvX2L(size);
8728
8729 result_reg->init_req(_instance_path, control());
8730 result_val->init_req(_instance_path, size);
8731 }
8732
8733 set_result(result_reg, result_val);
8734 }
8735
8736 return true;
8737 }
8738
8739 //------------------------------- inline_blackhole --------------------------------------
8740 //
8741 // Make sure all arguments to this node are alive.
8742 // This matches methods that were requested to be blackholed through compile commands.
8743 //
8744 bool LibraryCallKit::inline_blackhole() {
8745 assert(callee()->is_static(), "Should have been checked before: only static methods here");
8746 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
8747 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
8748
8749 // Blackhole node pinches only the control, not memory. This allows
8750 // the blackhole to be pinned in the loop that computes blackholed
8751 // values, but have no other side effects, like breaking the optimizations
8752 // across the blackhole.
8753
8754 Node* bh = _gvn.transform(new BlackholeNode(control()));
8755 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
8756
8757 // Bind call arguments as blackhole arguments to keep them alive
8758 uint nargs = callee()->arg_size();
8759 for (uint i = 0; i < nargs; i++) {
8760 bh->add_req(argument(i));
8761 }
8762
8763 return true;
8764 }