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/ciFlatArrayKlass.hpp"
27 #include "ci/ciUtilities.inline.hpp"
28 #include "ci/ciSymbols.hpp"
29 #include "classfile/vmIntrinsics.hpp"
30 #include "compiler/compileBroker.hpp"
31 #include "compiler/compileLog.hpp"
32 #include "gc/shared/barrierSet.hpp"
33 #include "jfr/support/jfrIntrinsics.hpp"
34 #include "memory/resourceArea.hpp"
35 #include "oops/klass.inline.hpp"
36 #include "oops/objArrayKlass.hpp"
37 #include "opto/addnode.hpp"
38 #include "opto/arraycopynode.hpp"
39 #include "opto/c2compiler.hpp"
40 #include "opto/castnode.hpp"
41 #include "opto/cfgnode.hpp"
42 #include "opto/convertnode.hpp"
43 #include "opto/countbitsnode.hpp"
44 #include "opto/idealKit.hpp"
45 #include "opto/library_call.hpp"
46 #include "opto/mathexactnode.hpp"
47 #include "opto/mulnode.hpp"
48 #include "opto/narrowptrnode.hpp"
49 #include "opto/opaquenode.hpp"
50 #include "opto/parse.hpp"
51 #include "opto/runtime.hpp"
52 #include "opto/rootnode.hpp"
53 #include "opto/subnode.hpp"
54 #include "opto/vectornode.hpp"
55 #include "prims/jvmtiExport.hpp"
56 #include "prims/jvmtiThreadState.hpp"
57 #include "prims/unsafe.hpp"
58 #include "runtime/jniHandles.inline.hpp"
59 #include "runtime/objectMonitor.hpp"
60 #include "runtime/sharedRuntime.hpp"
61 #include "runtime/stubRoutines.hpp"
62 #include "utilities/macros.hpp"
63 #include "utilities/powerOfTwo.hpp"
64
65 //---------------------------make_vm_intrinsic----------------------------
66 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
67 vmIntrinsicID id = m->intrinsic_id();
68 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
69
70 if (!m->is_loaded()) {
71 // Do not attempt to inline unloaded methods.
72 return nullptr;
73 }
74
75 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
76 bool is_available = false;
77
78 {
79 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
80 // the compiler must transition to '_thread_in_vm' state because both
81 // methods access VM-internal data.
82 VM_ENTRY_MARK;
83 methodHandle mh(THREAD, m->get_Method());
84 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
85 if (is_available && is_virtual) {
86 is_available = vmIntrinsics::does_virtual_dispatch(id);
87 }
88 }
89
90 if (is_available) {
91 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
92 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
93 return new LibraryIntrinsic(m, is_virtual,
94 vmIntrinsics::predicates_needed(id),
95 vmIntrinsics::does_virtual_dispatch(id),
96 id);
97 } else {
98 return nullptr;
99 }
100 }
101
102 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
103 LibraryCallKit kit(jvms, this);
104 Compile* C = kit.C;
105 int nodes = C->unique();
106 #ifndef PRODUCT
107 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
108 char buf[1000];
109 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
110 tty->print_cr("Intrinsic %s", str);
111 }
112 #endif
113 ciMethod* callee = kit.callee();
114 const int bci = kit.bci();
115 #ifdef ASSERT
116 Node* ctrl = kit.control();
117 #endif
118 // Try to inline the intrinsic.
119 if (callee->check_intrinsic_candidate() &&
120 kit.try_to_inline(_last_predicate)) {
121 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
122 : "(intrinsic)";
123 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
124 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
125 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
126 if (C->log()) {
127 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
128 vmIntrinsics::name_at(intrinsic_id()),
129 (is_virtual() ? " virtual='1'" : ""),
130 C->unique() - nodes);
131 }
132 // Push the result from the inlined method onto the stack.
133 kit.push_result();
134 return kit.transfer_exceptions_into_jvms();
135 }
136
137 // The intrinsic bailed out
138 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
139 if (jvms->has_method()) {
140 // Not a root compile.
141 const char* msg;
142 if (callee->intrinsic_candidate()) {
143 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
144 } else {
145 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
146 : "failed to inline (intrinsic), method not annotated";
147 }
148 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
149 C->inline_printer()->record(callee, jvms, InliningResult::FAILURE, msg);
150 } else {
151 // Root compile
152 ResourceMark rm;
153 stringStream msg_stream;
154 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
155 vmIntrinsics::name_at(intrinsic_id()),
156 is_virtual() ? " (virtual)" : "", bci);
157 const char *msg = msg_stream.freeze();
158 log_debug(jit, inlining)("%s", msg);
159 if (C->print_intrinsics() || C->print_inlining()) {
160 tty->print("%s", msg);
161 }
162 }
163 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
164
165 return nullptr;
166 }
167
168 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
169 LibraryCallKit kit(jvms, this);
170 Compile* C = kit.C;
171 int nodes = C->unique();
172 _last_predicate = predicate;
173 #ifndef PRODUCT
174 assert(is_predicated() && predicate < predicates_count(), "sanity");
175 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
176 char buf[1000];
177 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
178 tty->print_cr("Predicate for intrinsic %s", str);
179 }
180 #endif
181 ciMethod* callee = kit.callee();
182 const int bci = kit.bci();
183
184 Node* slow_ctl = kit.try_to_predicate(predicate);
185 if (!kit.failing()) {
186 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
187 : "(intrinsic, predicate)";
188 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
189 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
190
191 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
192 if (C->log()) {
193 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
194 vmIntrinsics::name_at(intrinsic_id()),
195 (is_virtual() ? " virtual='1'" : ""),
196 C->unique() - nodes);
197 }
198 return slow_ctl; // Could be null if the check folds.
199 }
200
201 // The intrinsic bailed out
202 if (jvms->has_method()) {
203 // Not a root compile.
204 const char* msg = "failed to generate predicate for intrinsic";
205 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
206 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
207 } else {
208 // Root compile
209 ResourceMark rm;
210 stringStream msg_stream;
211 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
212 vmIntrinsics::name_at(intrinsic_id()),
213 is_virtual() ? " (virtual)" : "", bci);
214 const char *msg = msg_stream.freeze();
215 log_debug(jit, inlining)("%s", msg);
216 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
217 }
218 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
219 return nullptr;
220 }
221
222 bool LibraryCallKit::try_to_inline(int predicate) {
223 // Handle symbolic names for otherwise undistinguished boolean switches:
224 const bool is_store = true;
225 const bool is_compress = true;
226 const bool is_static = true;
227 const bool is_volatile = true;
228
229 if (!jvms()->has_method()) {
230 // Root JVMState has a null method.
231 assert(map()->memory()->Opcode() == Op_Parm, "");
232 // Insert the memory aliasing node
233 set_all_memory(reset_memory());
234 }
235 assert(merged_memory(), "");
236
237 switch (intrinsic_id()) {
238 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
239 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
240 case vmIntrinsics::_getClass: return inline_native_getClass();
241
242 case vmIntrinsics::_ceil:
243 case vmIntrinsics::_floor:
244 case vmIntrinsics::_rint:
245 case vmIntrinsics::_dsin:
246 case vmIntrinsics::_dcos:
247 case vmIntrinsics::_dtan:
248 case vmIntrinsics::_dtanh:
249 case vmIntrinsics::_dabs:
250 case vmIntrinsics::_fabs:
251 case vmIntrinsics::_iabs:
252 case vmIntrinsics::_labs:
253 case vmIntrinsics::_datan2:
254 case vmIntrinsics::_dsqrt:
255 case vmIntrinsics::_dsqrt_strict:
256 case vmIntrinsics::_dexp:
257 case vmIntrinsics::_dlog:
258 case vmIntrinsics::_dlog10:
259 case vmIntrinsics::_dpow:
260 case vmIntrinsics::_dcopySign:
261 case vmIntrinsics::_fcopySign:
262 case vmIntrinsics::_dsignum:
263 case vmIntrinsics::_roundF:
264 case vmIntrinsics::_roundD:
265 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
266
267 case vmIntrinsics::_notify:
268 case vmIntrinsics::_notifyAll:
269 return inline_notify(intrinsic_id());
270
271 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
272 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
273 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
274 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
275 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
276 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
277 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
278 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
279 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
280 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
281 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
282 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
283 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
284 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
285
286 case vmIntrinsics::_arraycopy: return inline_arraycopy();
287
288 case vmIntrinsics::_arraySort: return inline_array_sort();
289 case vmIntrinsics::_arrayPartition: return inline_array_partition();
290
291 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
292 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
293 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
294 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
295
296 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
297 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
298 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
299 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
300 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
301 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
302 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
303 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
304
305 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
306
307 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
308
309 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
310 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
311 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
312 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
313
314 case vmIntrinsics::_compressStringC:
315 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
316 case vmIntrinsics::_inflateStringC:
317 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
318
319 case vmIntrinsics::_makePrivateBuffer: return inline_unsafe_make_private_buffer();
320 case vmIntrinsics::_finishPrivateBuffer: return inline_unsafe_finish_private_buffer();
321 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
322 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
323 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
324 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
325 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
326 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
327 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
328 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
329 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
330 case vmIntrinsics::_getValue: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false, true);
331
332 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
333 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
334 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
335 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
336 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
337 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
338 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
339 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
340 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
341 case vmIntrinsics::_putValue: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false, true);
342
343 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
344 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
345 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
346 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
347 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
348 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
349 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
350 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
351 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
352
353 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
354 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
355 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
356 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
357 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
358 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
359 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
360 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
361 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
362
363 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
364 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
365 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
366 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
367
368 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
369 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
370 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
371 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
372
373 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
374 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
375 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
376 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
377 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
378 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
379 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
380 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
381 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
382
383 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
384 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
385 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
386 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
387 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
388 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
389 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
390 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
391 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
392
393 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
394 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
395 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
396 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
397 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
398 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
399 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
400 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
401 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
402
403 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
404 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
405 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
406 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
407 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
408 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
409 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
410 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
411 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
412
413 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
414 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
415 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
416 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
417 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
418
419 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
420 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
421 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
422 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
423 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
424 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
425 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
426 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
427 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
428 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
429 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
430 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
431 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
432 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
433 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
434 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
435 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
436 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
437 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
438 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
439
440 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
441 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
442 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
443 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
444 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
445 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
446 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
447 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
448 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
449 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
450 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
451 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
452 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
453 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
454 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
455
456 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
457 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
458 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
459 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
460
461 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
462 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
463 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
464 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
465 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
466
467 case vmIntrinsics::_loadFence:
468 case vmIntrinsics::_storeFence:
469 case vmIntrinsics::_storeStoreFence:
470 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
471
472 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
473
474 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
475 case vmIntrinsics::_currentThread: return inline_native_currentThread();
476 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
477
478 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
479 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
480
481 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
482 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
483
484 #if INCLUDE_JVMTI
485 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
486 "notifyJvmtiStart", true, false);
487 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
488 "notifyJvmtiEnd", false, true);
489 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
490 "notifyJvmtiMount", false, false);
491 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
492 "notifyJvmtiUnmount", false, false);
493 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
494 #endif
495
496 #ifdef JFR_HAVE_INTRINSICS
497 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
498 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
499 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
500 #endif
501 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
502 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
503 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
504 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
505 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
506 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
507 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
508 case vmIntrinsics::_isFlatArray: return inline_unsafe_isFlatArray();
509 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
510 case vmIntrinsics::_getLength: return inline_native_getLength();
511 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
512 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
513 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
514 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
515 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
516 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
517 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
518
519 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
520 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
521 case vmIntrinsics::_newNullRestrictedNonAtomicArray: return inline_newArray(/* null_free */ true, /* atomic */ false);
522 case vmIntrinsics::_newNullRestrictedAtomicArray: return inline_newArray(/* null_free */ true, /* atomic */ true);
523 case vmIntrinsics::_newNullableAtomicArray: return inline_newArray(/* null_free */ false, /* atomic */ true);
524
525 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
526
527 case vmIntrinsics::_isInstance:
528 case vmIntrinsics::_isHidden:
529 case vmIntrinsics::_getSuperclass:
530 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
531
532 case vmIntrinsics::_floatToRawIntBits:
533 case vmIntrinsics::_floatToIntBits:
534 case vmIntrinsics::_intBitsToFloat:
535 case vmIntrinsics::_doubleToRawLongBits:
536 case vmIntrinsics::_doubleToLongBits:
537 case vmIntrinsics::_longBitsToDouble:
538 case vmIntrinsics::_floatToFloat16:
539 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
540 case vmIntrinsics::_sqrt_float16: return inline_fp16_operations(intrinsic_id(), 1);
541 case vmIntrinsics::_fma_float16: return inline_fp16_operations(intrinsic_id(), 3);
542 case vmIntrinsics::_floatIsFinite:
543 case vmIntrinsics::_floatIsInfinite:
544 case vmIntrinsics::_doubleIsFinite:
545 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
546
547 case vmIntrinsics::_numberOfLeadingZeros_i:
548 case vmIntrinsics::_numberOfLeadingZeros_l:
549 case vmIntrinsics::_numberOfTrailingZeros_i:
550 case vmIntrinsics::_numberOfTrailingZeros_l:
551 case vmIntrinsics::_bitCount_i:
552 case vmIntrinsics::_bitCount_l:
553 case vmIntrinsics::_reverse_i:
554 case vmIntrinsics::_reverse_l:
555 case vmIntrinsics::_reverseBytes_i:
556 case vmIntrinsics::_reverseBytes_l:
557 case vmIntrinsics::_reverseBytes_s:
558 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
559
560 case vmIntrinsics::_compress_i:
561 case vmIntrinsics::_compress_l:
562 case vmIntrinsics::_expand_i:
563 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
564
565 case vmIntrinsics::_compareUnsigned_i:
566 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
567
568 case vmIntrinsics::_divideUnsigned_i:
569 case vmIntrinsics::_divideUnsigned_l:
570 case vmIntrinsics::_remainderUnsigned_i:
571 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
572
573 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
574
575 case vmIntrinsics::_Reference_get: return inline_reference_get();
576 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
577 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
578 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
579 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
580
581 case vmIntrinsics::_Class_cast: return inline_Class_cast();
582
583 case vmIntrinsics::_aescrypt_encryptBlock:
584 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
585
586 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
587 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
588 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
589
590 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
591 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
592 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
593
594 case vmIntrinsics::_counterMode_AESCrypt:
595 return inline_counterMode_AESCrypt(intrinsic_id());
596
597 case vmIntrinsics::_galoisCounterMode_AESCrypt:
598 return inline_galoisCounterMode_AESCrypt();
599
600 case vmIntrinsics::_md5_implCompress:
601 case vmIntrinsics::_sha_implCompress:
602 case vmIntrinsics::_sha2_implCompress:
603 case vmIntrinsics::_sha5_implCompress:
604 case vmIntrinsics::_sha3_implCompress:
605 return inline_digestBase_implCompress(intrinsic_id());
606 case vmIntrinsics::_double_keccak:
607 return inline_double_keccak();
608
609 case vmIntrinsics::_digestBase_implCompressMB:
610 return inline_digestBase_implCompressMB(predicate);
611
612 case vmIntrinsics::_multiplyToLen:
613 return inline_multiplyToLen();
614
615 case vmIntrinsics::_squareToLen:
616 return inline_squareToLen();
617
618 case vmIntrinsics::_mulAdd:
619 return inline_mulAdd();
620
621 case vmIntrinsics::_montgomeryMultiply:
622 return inline_montgomeryMultiply();
623 case vmIntrinsics::_montgomerySquare:
624 return inline_montgomerySquare();
625
626 case vmIntrinsics::_bigIntegerRightShiftWorker:
627 return inline_bigIntegerShift(true);
628 case vmIntrinsics::_bigIntegerLeftShiftWorker:
629 return inline_bigIntegerShift(false);
630
631 case vmIntrinsics::_vectorizedMismatch:
632 return inline_vectorizedMismatch();
633
634 case vmIntrinsics::_ghash_processBlocks:
635 return inline_ghash_processBlocks();
636 case vmIntrinsics::_chacha20Block:
637 return inline_chacha20Block();
638 case vmIntrinsics::_dilithiumAlmostNtt:
639 return inline_dilithiumAlmostNtt();
640 case vmIntrinsics::_dilithiumAlmostInverseNtt:
641 return inline_dilithiumAlmostInverseNtt();
642 case vmIntrinsics::_dilithiumNttMult:
643 return inline_dilithiumNttMult();
644 case vmIntrinsics::_dilithiumMontMulByConstant:
645 return inline_dilithiumMontMulByConstant();
646 case vmIntrinsics::_dilithiumDecomposePoly:
647 return inline_dilithiumDecomposePoly();
648 case vmIntrinsics::_base64_encodeBlock:
649 return inline_base64_encodeBlock();
650 case vmIntrinsics::_base64_decodeBlock:
651 return inline_base64_decodeBlock();
652 case vmIntrinsics::_poly1305_processBlocks:
653 return inline_poly1305_processBlocks();
654 case vmIntrinsics::_intpoly_montgomeryMult_P256:
655 return inline_intpoly_montgomeryMult_P256();
656 case vmIntrinsics::_intpoly_assign:
657 return inline_intpoly_assign();
658 case vmIntrinsics::_encodeISOArray:
659 case vmIntrinsics::_encodeByteISOArray:
660 return inline_encodeISOArray(false);
661 case vmIntrinsics::_encodeAsciiArray:
662 return inline_encodeISOArray(true);
663
664 case vmIntrinsics::_updateCRC32:
665 return inline_updateCRC32();
666 case vmIntrinsics::_updateBytesCRC32:
667 return inline_updateBytesCRC32();
668 case vmIntrinsics::_updateByteBufferCRC32:
669 return inline_updateByteBufferCRC32();
670
671 case vmIntrinsics::_updateBytesCRC32C:
672 return inline_updateBytesCRC32C();
673 case vmIntrinsics::_updateDirectByteBufferCRC32C:
674 return inline_updateDirectByteBufferCRC32C();
675
676 case vmIntrinsics::_updateBytesAdler32:
677 return inline_updateBytesAdler32();
678 case vmIntrinsics::_updateByteBufferAdler32:
679 return inline_updateByteBufferAdler32();
680
681 case vmIntrinsics::_profileBoolean:
682 return inline_profileBoolean();
683 case vmIntrinsics::_isCompileConstant:
684 return inline_isCompileConstant();
685
686 case vmIntrinsics::_countPositives:
687 return inline_countPositives();
688
689 case vmIntrinsics::_fmaD:
690 case vmIntrinsics::_fmaF:
691 return inline_fma(intrinsic_id());
692
693 case vmIntrinsics::_isDigit:
694 case vmIntrinsics::_isLowerCase:
695 case vmIntrinsics::_isUpperCase:
696 case vmIntrinsics::_isWhitespace:
697 return inline_character_compare(intrinsic_id());
698
699 case vmIntrinsics::_min:
700 case vmIntrinsics::_max:
701 case vmIntrinsics::_min_strict:
702 case vmIntrinsics::_max_strict:
703 case vmIntrinsics::_minL:
704 case vmIntrinsics::_maxL:
705 case vmIntrinsics::_minF:
706 case vmIntrinsics::_maxF:
707 case vmIntrinsics::_minD:
708 case vmIntrinsics::_maxD:
709 case vmIntrinsics::_minF_strict:
710 case vmIntrinsics::_maxF_strict:
711 case vmIntrinsics::_minD_strict:
712 case vmIntrinsics::_maxD_strict:
713 return inline_min_max(intrinsic_id());
714
715 case vmIntrinsics::_VectorUnaryOp:
716 return inline_vector_nary_operation(1);
717 case vmIntrinsics::_VectorBinaryOp:
718 return inline_vector_nary_operation(2);
719 case vmIntrinsics::_VectorTernaryOp:
720 return inline_vector_nary_operation(3);
721 case vmIntrinsics::_VectorFromBitsCoerced:
722 return inline_vector_frombits_coerced();
723 case vmIntrinsics::_VectorMaskOp:
724 return inline_vector_mask_operation();
725 case vmIntrinsics::_VectorLoadOp:
726 return inline_vector_mem_operation(/*is_store=*/false);
727 case vmIntrinsics::_VectorLoadMaskedOp:
728 return inline_vector_mem_masked_operation(/*is_store*/false);
729 case vmIntrinsics::_VectorStoreOp:
730 return inline_vector_mem_operation(/*is_store=*/true);
731 case vmIntrinsics::_VectorStoreMaskedOp:
732 return inline_vector_mem_masked_operation(/*is_store=*/true);
733 case vmIntrinsics::_VectorGatherOp:
734 return inline_vector_gather_scatter(/*is_scatter*/ false);
735 case vmIntrinsics::_VectorScatterOp:
736 return inline_vector_gather_scatter(/*is_scatter*/ true);
737 case vmIntrinsics::_VectorReductionCoerced:
738 return inline_vector_reduction();
739 case vmIntrinsics::_VectorTest:
740 return inline_vector_test();
741 case vmIntrinsics::_VectorBlend:
742 return inline_vector_blend();
743 case vmIntrinsics::_VectorRearrange:
744 return inline_vector_rearrange();
745 case vmIntrinsics::_VectorSelectFrom:
746 return inline_vector_select_from();
747 case vmIntrinsics::_VectorCompare:
748 return inline_vector_compare();
749 case vmIntrinsics::_VectorBroadcastInt:
750 return inline_vector_broadcast_int();
751 case vmIntrinsics::_VectorConvert:
752 return inline_vector_convert();
753 case vmIntrinsics::_VectorInsert:
754 return inline_vector_insert();
755 case vmIntrinsics::_VectorExtract:
756 return inline_vector_extract();
757 case vmIntrinsics::_VectorCompressExpand:
758 return inline_vector_compress_expand();
759 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
760 return inline_vector_select_from_two_vectors();
761 case vmIntrinsics::_IndexVector:
762 return inline_index_vector();
763 case vmIntrinsics::_IndexPartiallyInUpperRange:
764 return inline_index_partially_in_upper_range();
765
766 case vmIntrinsics::_getObjectSize:
767 return inline_getObjectSize();
768
769 case vmIntrinsics::_blackhole:
770 return inline_blackhole();
771
772 default:
773 // If you get here, it may be that someone has added a new intrinsic
774 // to the list in vmIntrinsics.hpp without implementing it here.
775 #ifndef PRODUCT
776 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
777 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
778 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
779 }
780 #endif
781 return false;
782 }
783 }
784
785 Node* LibraryCallKit::try_to_predicate(int predicate) {
786 if (!jvms()->has_method()) {
787 // Root JVMState has a null method.
788 assert(map()->memory()->Opcode() == Op_Parm, "");
789 // Insert the memory aliasing node
790 set_all_memory(reset_memory());
791 }
792 assert(merged_memory(), "");
793
794 switch (intrinsic_id()) {
795 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
796 return inline_cipherBlockChaining_AESCrypt_predicate(false);
797 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
798 return inline_cipherBlockChaining_AESCrypt_predicate(true);
799 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
800 return inline_electronicCodeBook_AESCrypt_predicate(false);
801 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
802 return inline_electronicCodeBook_AESCrypt_predicate(true);
803 case vmIntrinsics::_counterMode_AESCrypt:
804 return inline_counterMode_AESCrypt_predicate();
805 case vmIntrinsics::_digestBase_implCompressMB:
806 return inline_digestBase_implCompressMB_predicate(predicate);
807 case vmIntrinsics::_galoisCounterMode_AESCrypt:
808 return inline_galoisCounterMode_AESCrypt_predicate();
809
810 default:
811 // If you get here, it may be that someone has added a new intrinsic
812 // to the list in vmIntrinsics.hpp without implementing it here.
813 #ifndef PRODUCT
814 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
815 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
816 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
817 }
818 #endif
819 Node* slow_ctl = control();
820 set_control(top()); // No fast path intrinsic
821 return slow_ctl;
822 }
823 }
824
825 //------------------------------set_result-------------------------------
826 // Helper function for finishing intrinsics.
827 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
828 record_for_igvn(region);
829 set_control(_gvn.transform(region));
830 set_result( _gvn.transform(value));
831 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
832 }
833
834 //------------------------------generate_guard---------------------------
835 // Helper function for generating guarded fast-slow graph structures.
836 // The given 'test', if true, guards a slow path. If the test fails
837 // then a fast path can be taken. (We generally hope it fails.)
838 // In all cases, GraphKit::control() is updated to the fast path.
839 // The returned value represents the control for the slow path.
840 // The return value is never 'top'; it is either a valid control
841 // or null if it is obvious that the slow path can never be taken.
842 // Also, if region and the slow control are not null, the slow edge
843 // is appended to the region.
844 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
845 if (stopped()) {
846 // Already short circuited.
847 return nullptr;
848 }
849
850 // Build an if node and its projections.
851 // If test is true we take the slow path, which we assume is uncommon.
852 if (_gvn.type(test) == TypeInt::ZERO) {
853 // The slow branch is never taken. No need to build this guard.
854 return nullptr;
855 }
856
857 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
858
859 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
860 if (if_slow == top()) {
861 // The slow branch is never taken. No need to build this guard.
862 return nullptr;
863 }
864
865 if (region != nullptr)
866 region->add_req(if_slow);
867
868 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
869 set_control(if_fast);
870
871 return if_slow;
872 }
873
874 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
875 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
876 }
877 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
878 return generate_guard(test, region, PROB_FAIR);
879 }
880
881 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
882 Node* *pos_index) {
883 if (stopped())
884 return nullptr; // already stopped
885 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
886 return nullptr; // index is already adequately typed
887 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
888 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
889 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
890 if (is_neg != nullptr && pos_index != nullptr) {
891 // Emulate effect of Parse::adjust_map_after_if.
892 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
893 (*pos_index) = _gvn.transform(ccast);
894 }
895 return is_neg;
896 }
897
898 // Make sure that 'position' is a valid limit index, in [0..length].
899 // There are two equivalent plans for checking this:
900 // A. (offset + copyLength) unsigned<= arrayLength
901 // B. offset <= (arrayLength - copyLength)
902 // We require that all of the values above, except for the sum and
903 // difference, are already known to be non-negative.
904 // Plan A is robust in the face of overflow, if offset and copyLength
905 // are both hugely positive.
906 //
907 // Plan B is less direct and intuitive, but it does not overflow at
908 // all, since the difference of two non-negatives is always
909 // representable. Whenever Java methods must perform the equivalent
910 // check they generally use Plan B instead of Plan A.
911 // For the moment we use Plan A.
912 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
913 Node* subseq_length,
914 Node* array_length,
915 RegionNode* region) {
916 if (stopped())
917 return nullptr; // already stopped
918 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
919 if (zero_offset && subseq_length->eqv_uncast(array_length))
920 return nullptr; // common case of whole-array copy
921 Node* last = subseq_length;
922 if (!zero_offset) // last += offset
923 last = _gvn.transform(new AddINode(last, offset));
924 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
925 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
926 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
927 return is_over;
928 }
929
930 // Emit range checks for the given String.value byte array
931 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
932 if (stopped()) {
933 return; // already stopped
934 }
935 RegionNode* bailout = new RegionNode(1);
936 record_for_igvn(bailout);
937 if (char_count) {
938 // Convert char count to byte count
939 count = _gvn.transform(new LShiftINode(count, intcon(1)));
940 }
941
942 // Offset and count must not be negative
943 generate_negative_guard(offset, bailout);
944 generate_negative_guard(count, bailout);
945 // Offset + count must not exceed length of array
946 generate_limit_guard(offset, count, load_array_length(array), bailout);
947
948 if (bailout->req() > 1) {
949 PreserveJVMState pjvms(this);
950 set_control(_gvn.transform(bailout));
951 uncommon_trap(Deoptimization::Reason_intrinsic,
952 Deoptimization::Action_maybe_recompile);
953 }
954 }
955
956 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
957 bool is_immutable) {
958 ciKlass* thread_klass = env()->Thread_klass();
959 const Type* thread_type
960 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
961
962 Node* thread = _gvn.transform(new ThreadLocalNode());
963 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
964 tls_output = thread;
965
966 Node* thread_obj_handle
967 = (is_immutable
968 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
969 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
970 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
971 thread_obj_handle = _gvn.transform(thread_obj_handle);
972
973 DecoratorSet decorators = IN_NATIVE;
974 if (is_immutable) {
975 decorators |= C2_IMMUTABLE_MEMORY;
976 }
977 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
978 }
979
980 //--------------------------generate_current_thread--------------------
981 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
982 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
983 /*is_immutable*/false);
984 }
985
986 //--------------------------generate_virtual_thread--------------------
987 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
988 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
989 !C->method()->changes_current_thread());
990 }
991
992 //------------------------------make_string_method_node------------------------
993 // Helper method for String intrinsic functions. This version is called with
994 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
995 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
996 // containing the lengths of str1 and str2.
997 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
998 Node* result = nullptr;
999 switch (opcode) {
1000 case Op_StrIndexOf:
1001 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1002 str1_start, cnt1, str2_start, cnt2, ae);
1003 break;
1004 case Op_StrComp:
1005 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1006 str1_start, cnt1, str2_start, cnt2, ae);
1007 break;
1008 case Op_StrEquals:
1009 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1010 // Use the constant length if there is one because optimized match rule may exist.
1011 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1012 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1013 break;
1014 default:
1015 ShouldNotReachHere();
1016 return nullptr;
1017 }
1018
1019 // All these intrinsics have checks.
1020 C->set_has_split_ifs(true); // Has chance for split-if optimization
1021 clear_upper_avx();
1022
1023 return _gvn.transform(result);
1024 }
1025
1026 //------------------------------inline_string_compareTo------------------------
1027 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1028 Node* arg1 = argument(0);
1029 Node* arg2 = argument(1);
1030
1031 arg1 = must_be_not_null(arg1, true);
1032 arg2 = must_be_not_null(arg2, true);
1033
1034 // Get start addr and length of first argument
1035 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1036 Node* arg1_cnt = load_array_length(arg1);
1037
1038 // Get start addr and length of second argument
1039 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1040 Node* arg2_cnt = load_array_length(arg2);
1041
1042 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1043 set_result(result);
1044 return true;
1045 }
1046
1047 //------------------------------inline_string_equals------------------------
1048 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1049 Node* arg1 = argument(0);
1050 Node* arg2 = argument(1);
1051
1052 // paths (plus control) merge
1053 RegionNode* region = new RegionNode(3);
1054 Node* phi = new PhiNode(region, TypeInt::BOOL);
1055
1056 if (!stopped()) {
1057
1058 arg1 = must_be_not_null(arg1, true);
1059 arg2 = must_be_not_null(arg2, true);
1060
1061 // Get start addr and length of first argument
1062 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1063 Node* arg1_cnt = load_array_length(arg1);
1064
1065 // Get start addr and length of second argument
1066 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1067 Node* arg2_cnt = load_array_length(arg2);
1068
1069 // Check for arg1_cnt != arg2_cnt
1070 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1071 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1072 Node* if_ne = generate_slow_guard(bol, nullptr);
1073 if (if_ne != nullptr) {
1074 phi->init_req(2, intcon(0));
1075 region->init_req(2, if_ne);
1076 }
1077
1078 // Check for count == 0 is done by assembler code for StrEquals.
1079
1080 if (!stopped()) {
1081 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1082 phi->init_req(1, equals);
1083 region->init_req(1, control());
1084 }
1085 }
1086
1087 // post merge
1088 set_control(_gvn.transform(region));
1089 record_for_igvn(region);
1090
1091 set_result(_gvn.transform(phi));
1092 return true;
1093 }
1094
1095 //------------------------------inline_array_equals----------------------------
1096 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1097 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1098 Node* arg1 = argument(0);
1099 Node* arg2 = argument(1);
1100
1101 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1102 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1103 clear_upper_avx();
1104
1105 return true;
1106 }
1107
1108
1109 //------------------------------inline_countPositives------------------------------
1110 bool LibraryCallKit::inline_countPositives() {
1111 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1112 return false;
1113 }
1114
1115 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1116 // no receiver since it is static method
1117 Node* ba = argument(0);
1118 Node* offset = argument(1);
1119 Node* len = argument(2);
1120
1121 ba = must_be_not_null(ba, true);
1122
1123 // Range checks
1124 generate_string_range_check(ba, offset, len, false);
1125 if (stopped()) {
1126 return true;
1127 }
1128 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1129 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1130 set_result(_gvn.transform(result));
1131 clear_upper_avx();
1132 return true;
1133 }
1134
1135 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1136 Node* index = argument(0);
1137 Node* length = bt == T_INT ? argument(1) : argument(2);
1138 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1139 return false;
1140 }
1141
1142 // check that length is positive
1143 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1144 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1145
1146 {
1147 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1148 uncommon_trap(Deoptimization::Reason_intrinsic,
1149 Deoptimization::Action_make_not_entrant);
1150 }
1151
1152 if (stopped()) {
1153 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1154 return true;
1155 }
1156
1157 // length is now known positive, add a cast node to make this explicit
1158 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1159 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1160 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1161 ConstraintCastNode::RegularDependency, bt);
1162 casted_length = _gvn.transform(casted_length);
1163 replace_in_map(length, casted_length);
1164 length = casted_length;
1165
1166 // Use an unsigned comparison for the range check itself
1167 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1168 BoolTest::mask btest = BoolTest::lt;
1169 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1170 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1171 _gvn.set_type(rc, rc->Value(&_gvn));
1172 if (!rc_bool->is_Con()) {
1173 record_for_igvn(rc);
1174 }
1175 set_control(_gvn.transform(new IfTrueNode(rc)));
1176 {
1177 PreserveJVMState pjvms(this);
1178 set_control(_gvn.transform(new IfFalseNode(rc)));
1179 uncommon_trap(Deoptimization::Reason_range_check,
1180 Deoptimization::Action_make_not_entrant);
1181 }
1182
1183 if (stopped()) {
1184 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1185 return true;
1186 }
1187
1188 // index is now known to be >= 0 and < length, cast it
1189 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1190 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1191 ConstraintCastNode::RegularDependency, bt);
1192 result = _gvn.transform(result);
1193 set_result(result);
1194 replace_in_map(index, result);
1195 return true;
1196 }
1197
1198 //------------------------------inline_string_indexOf------------------------
1199 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1200 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1201 return false;
1202 }
1203 Node* src = argument(0);
1204 Node* tgt = argument(1);
1205
1206 // Make the merge point
1207 RegionNode* result_rgn = new RegionNode(4);
1208 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1209
1210 src = must_be_not_null(src, true);
1211 tgt = must_be_not_null(tgt, true);
1212
1213 // Get start addr and length of source string
1214 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1215 Node* src_count = load_array_length(src);
1216
1217 // Get start addr and length of substring
1218 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1219 Node* tgt_count = load_array_length(tgt);
1220
1221 Node* result = nullptr;
1222 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1223
1224 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1225 // Divide src size by 2 if String is UTF16 encoded
1226 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1227 }
1228 if (ae == StrIntrinsicNode::UU) {
1229 // Divide substring size by 2 if String is UTF16 encoded
1230 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1231 }
1232
1233 if (call_opt_stub) {
1234 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1235 StubRoutines::_string_indexof_array[ae],
1236 "stringIndexOf", TypePtr::BOTTOM, src_start,
1237 src_count, tgt_start, tgt_count);
1238 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1239 } else {
1240 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1241 result_rgn, result_phi, ae);
1242 }
1243 if (result != nullptr) {
1244 result_phi->init_req(3, result);
1245 result_rgn->init_req(3, control());
1246 }
1247 set_control(_gvn.transform(result_rgn));
1248 record_for_igvn(result_rgn);
1249 set_result(_gvn.transform(result_phi));
1250
1251 return true;
1252 }
1253
1254 //-----------------------------inline_string_indexOfI-----------------------
1255 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1256 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1257 return false;
1258 }
1259 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1260 return false;
1261 }
1262
1263 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1264 Node* src = argument(0); // byte[]
1265 Node* src_count = argument(1); // char count
1266 Node* tgt = argument(2); // byte[]
1267 Node* tgt_count = argument(3); // char count
1268 Node* from_index = argument(4); // char index
1269
1270 src = must_be_not_null(src, true);
1271 tgt = must_be_not_null(tgt, true);
1272
1273 // Multiply byte array index by 2 if String is UTF16 encoded
1274 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1275 src_count = _gvn.transform(new SubINode(src_count, from_index));
1276 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1277 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1278
1279 // Range checks
1280 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1281 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1282 if (stopped()) {
1283 return true;
1284 }
1285
1286 RegionNode* region = new RegionNode(5);
1287 Node* phi = new PhiNode(region, TypeInt::INT);
1288 Node* result = nullptr;
1289
1290 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1291
1292 if (call_opt_stub) {
1293 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1294 StubRoutines::_string_indexof_array[ae],
1295 "stringIndexOf", TypePtr::BOTTOM, src_start,
1296 src_count, tgt_start, tgt_count);
1297 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1298 } else {
1299 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1300 region, phi, ae);
1301 }
1302 if (result != nullptr) {
1303 // The result is index relative to from_index if substring was found, -1 otherwise.
1304 // Generate code which will fold into cmove.
1305 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1306 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1307
1308 Node* if_lt = generate_slow_guard(bol, nullptr);
1309 if (if_lt != nullptr) {
1310 // result == -1
1311 phi->init_req(3, result);
1312 region->init_req(3, if_lt);
1313 }
1314 if (!stopped()) {
1315 result = _gvn.transform(new AddINode(result, from_index));
1316 phi->init_req(4, result);
1317 region->init_req(4, control());
1318 }
1319 }
1320
1321 set_control(_gvn.transform(region));
1322 record_for_igvn(region);
1323 set_result(_gvn.transform(phi));
1324 clear_upper_avx();
1325
1326 return true;
1327 }
1328
1329 // Create StrIndexOfNode with fast path checks
1330 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1331 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1332 // Check for substr count > string count
1333 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1334 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1335 Node* if_gt = generate_slow_guard(bol, nullptr);
1336 if (if_gt != nullptr) {
1337 phi->init_req(1, intcon(-1));
1338 region->init_req(1, if_gt);
1339 }
1340 if (!stopped()) {
1341 // Check for substr count == 0
1342 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1343 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1344 Node* if_zero = generate_slow_guard(bol, nullptr);
1345 if (if_zero != nullptr) {
1346 phi->init_req(2, intcon(0));
1347 region->init_req(2, if_zero);
1348 }
1349 }
1350 if (!stopped()) {
1351 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1352 }
1353 return nullptr;
1354 }
1355
1356 //-----------------------------inline_string_indexOfChar-----------------------
1357 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1358 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1359 return false;
1360 }
1361 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1362 return false;
1363 }
1364 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1365 Node* src = argument(0); // byte[]
1366 Node* int_ch = argument(1);
1367 Node* from_index = argument(2);
1368 Node* max = argument(3);
1369
1370 src = must_be_not_null(src, true);
1371
1372 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1373 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1374 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1375
1376 // Range checks
1377 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1378
1379 // Check for int_ch >= 0
1380 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1381 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1382 {
1383 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1384 uncommon_trap(Deoptimization::Reason_intrinsic,
1385 Deoptimization::Action_maybe_recompile);
1386 }
1387 if (stopped()) {
1388 return true;
1389 }
1390
1391 RegionNode* region = new RegionNode(3);
1392 Node* phi = new PhiNode(region, TypeInt::INT);
1393
1394 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1395 C->set_has_split_ifs(true); // Has chance for split-if optimization
1396 _gvn.transform(result);
1397
1398 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1399 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1400
1401 Node* if_lt = generate_slow_guard(bol, nullptr);
1402 if (if_lt != nullptr) {
1403 // result == -1
1404 phi->init_req(2, result);
1405 region->init_req(2, if_lt);
1406 }
1407 if (!stopped()) {
1408 result = _gvn.transform(new AddINode(result, from_index));
1409 phi->init_req(1, result);
1410 region->init_req(1, control());
1411 }
1412 set_control(_gvn.transform(region));
1413 record_for_igvn(region);
1414 set_result(_gvn.transform(phi));
1415 clear_upper_avx();
1416
1417 return true;
1418 }
1419 //---------------------------inline_string_copy---------------------
1420 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1421 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1422 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1423 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1424 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1425 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1426 bool LibraryCallKit::inline_string_copy(bool compress) {
1427 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1428 return false;
1429 }
1430 int nargs = 5; // 2 oops, 3 ints
1431 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1432
1433 Node* src = argument(0);
1434 Node* src_offset = argument(1);
1435 Node* dst = argument(2);
1436 Node* dst_offset = argument(3);
1437 Node* length = argument(4);
1438
1439 // Check for allocation before we add nodes that would confuse
1440 // tightly_coupled_allocation()
1441 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1442
1443 // Figure out the size and type of the elements we will be copying.
1444 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1445 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1446 if (src_type == nullptr || dst_type == nullptr) {
1447 return false;
1448 }
1449 BasicType src_elem = src_type->elem()->array_element_basic_type();
1450 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1451 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1452 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1453 "Unsupported array types for inline_string_copy");
1454
1455 src = must_be_not_null(src, true);
1456 dst = must_be_not_null(dst, true);
1457
1458 // Convert char[] offsets to byte[] offsets
1459 bool convert_src = (compress && src_elem == T_BYTE);
1460 bool convert_dst = (!compress && dst_elem == T_BYTE);
1461 if (convert_src) {
1462 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1463 } else if (convert_dst) {
1464 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1465 }
1466
1467 // Range checks
1468 generate_string_range_check(src, src_offset, length, convert_src);
1469 generate_string_range_check(dst, dst_offset, length, convert_dst);
1470 if (stopped()) {
1471 return true;
1472 }
1473
1474 Node* src_start = array_element_address(src, src_offset, src_elem);
1475 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1476 // 'src_start' points to src array + scaled offset
1477 // 'dst_start' points to dst array + scaled offset
1478 Node* count = nullptr;
1479 if (compress) {
1480 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1481 } else {
1482 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1483 }
1484
1485 if (alloc != nullptr) {
1486 if (alloc->maybe_set_complete(&_gvn)) {
1487 // "You break it, you buy it."
1488 InitializeNode* init = alloc->initialization();
1489 assert(init->is_complete(), "we just did this");
1490 init->set_complete_with_arraycopy();
1491 assert(dst->is_CheckCastPP(), "sanity");
1492 assert(dst->in(0)->in(0) == init, "dest pinned");
1493 }
1494 // Do not let stores that initialize this object be reordered with
1495 // a subsequent store that would make this object accessible by
1496 // other threads.
1497 // Record what AllocateNode this StoreStore protects so that
1498 // escape analysis can go from the MemBarStoreStoreNode to the
1499 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1500 // based on the escape status of the AllocateNode.
1501 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1502 }
1503 if (compress) {
1504 set_result(_gvn.transform(count));
1505 }
1506 clear_upper_avx();
1507
1508 return true;
1509 }
1510
1511 #ifdef _LP64
1512 #define XTOP ,top() /*additional argument*/
1513 #else //_LP64
1514 #define XTOP /*no additional argument*/
1515 #endif //_LP64
1516
1517 //------------------------inline_string_toBytesU--------------------------
1518 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1519 bool LibraryCallKit::inline_string_toBytesU() {
1520 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1521 return false;
1522 }
1523 // Get the arguments.
1524 Node* value = argument(0);
1525 Node* offset = argument(1);
1526 Node* length = argument(2);
1527
1528 Node* newcopy = nullptr;
1529
1530 // Set the original stack and the reexecute bit for the interpreter to reexecute
1531 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1532 { PreserveReexecuteState preexecs(this);
1533 jvms()->set_should_reexecute(true);
1534
1535 // Check if a null path was taken unconditionally.
1536 value = null_check(value);
1537
1538 RegionNode* bailout = new RegionNode(1);
1539 record_for_igvn(bailout);
1540
1541 // Range checks
1542 generate_negative_guard(offset, bailout);
1543 generate_negative_guard(length, bailout);
1544 generate_limit_guard(offset, length, load_array_length(value), bailout);
1545 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1546 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1547
1548 if (bailout->req() > 1) {
1549 PreserveJVMState pjvms(this);
1550 set_control(_gvn.transform(bailout));
1551 uncommon_trap(Deoptimization::Reason_intrinsic,
1552 Deoptimization::Action_maybe_recompile);
1553 }
1554 if (stopped()) {
1555 return true;
1556 }
1557
1558 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1559 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1560 newcopy = new_array(klass_node, size, 0); // no arguments to push
1561 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1562 guarantee(alloc != nullptr, "created above");
1563
1564 // Calculate starting addresses.
1565 Node* src_start = array_element_address(value, offset, T_CHAR);
1566 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1567
1568 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1569 const TypeInt* toffset = gvn().type(offset)->is_int();
1570 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1571
1572 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1573 const char* copyfunc_name = "arraycopy";
1574 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1575 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1576 OptoRuntime::fast_arraycopy_Type(),
1577 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1578 src_start, dst_start, ConvI2X(length) XTOP);
1579 // Do not let reads from the cloned object float above the arraycopy.
1580 if (alloc->maybe_set_complete(&_gvn)) {
1581 // "You break it, you buy it."
1582 InitializeNode* init = alloc->initialization();
1583 assert(init->is_complete(), "we just did this");
1584 init->set_complete_with_arraycopy();
1585 assert(newcopy->is_CheckCastPP(), "sanity");
1586 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1587 }
1588 // Do not let stores that initialize this object be reordered with
1589 // a subsequent store that would make this object accessible by
1590 // other threads.
1591 // Record what AllocateNode this StoreStore protects so that
1592 // escape analysis can go from the MemBarStoreStoreNode to the
1593 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1594 // based on the escape status of the AllocateNode.
1595 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1596 } // original reexecute is set back here
1597
1598 C->set_has_split_ifs(true); // Has chance for split-if optimization
1599 if (!stopped()) {
1600 set_result(newcopy);
1601 }
1602 clear_upper_avx();
1603
1604 return true;
1605 }
1606
1607 //------------------------inline_string_getCharsU--------------------------
1608 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1609 bool LibraryCallKit::inline_string_getCharsU() {
1610 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1611 return false;
1612 }
1613
1614 // Get the arguments.
1615 Node* src = argument(0);
1616 Node* src_begin = argument(1);
1617 Node* src_end = argument(2); // exclusive offset (i < src_end)
1618 Node* dst = argument(3);
1619 Node* dst_begin = argument(4);
1620
1621 // Check for allocation before we add nodes that would confuse
1622 // tightly_coupled_allocation()
1623 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1624
1625 // Check if a null path was taken unconditionally.
1626 src = null_check(src);
1627 dst = null_check(dst);
1628 if (stopped()) {
1629 return true;
1630 }
1631
1632 // Get length and convert char[] offset to byte[] offset
1633 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1634 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1635
1636 // Range checks
1637 generate_string_range_check(src, src_begin, length, true);
1638 generate_string_range_check(dst, dst_begin, length, false);
1639 if (stopped()) {
1640 return true;
1641 }
1642
1643 if (!stopped()) {
1644 // Calculate starting addresses.
1645 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1646 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1647
1648 // Check if array addresses are aligned to HeapWordSize
1649 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1650 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1651 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1652 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1653
1654 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1655 const char* copyfunc_name = "arraycopy";
1656 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1657 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1658 OptoRuntime::fast_arraycopy_Type(),
1659 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1660 src_start, dst_start, ConvI2X(length) XTOP);
1661 // Do not let reads from the cloned object float above the arraycopy.
1662 if (alloc != nullptr) {
1663 if (alloc->maybe_set_complete(&_gvn)) {
1664 // "You break it, you buy it."
1665 InitializeNode* init = alloc->initialization();
1666 assert(init->is_complete(), "we just did this");
1667 init->set_complete_with_arraycopy();
1668 assert(dst->is_CheckCastPP(), "sanity");
1669 assert(dst->in(0)->in(0) == init, "dest pinned");
1670 }
1671 // Do not let stores that initialize this object be reordered with
1672 // a subsequent store that would make this object accessible by
1673 // other threads.
1674 // Record what AllocateNode this StoreStore protects so that
1675 // escape analysis can go from the MemBarStoreStoreNode to the
1676 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1677 // based on the escape status of the AllocateNode.
1678 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1679 } else {
1680 insert_mem_bar(Op_MemBarCPUOrder);
1681 }
1682 }
1683
1684 C->set_has_split_ifs(true); // Has chance for split-if optimization
1685 return true;
1686 }
1687
1688 //----------------------inline_string_char_access----------------------------
1689 // Store/Load char to/from byte[] array.
1690 // static void StringUTF16.putChar(byte[] val, int index, int c)
1691 // static char StringUTF16.getChar(byte[] val, int index)
1692 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1693 Node* value = argument(0);
1694 Node* index = argument(1);
1695 Node* ch = is_store ? argument(2) : nullptr;
1696
1697 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1698 // correctly requires matched array shapes.
1699 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1700 "sanity: byte[] and char[] bases agree");
1701 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1702 "sanity: byte[] and char[] scales agree");
1703
1704 // Bail when getChar over constants is requested: constant folding would
1705 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1706 // Java method would constant fold nicely instead.
1707 if (!is_store && value->is_Con() && index->is_Con()) {
1708 return false;
1709 }
1710
1711 // Save state and restore on bailout
1712 uint old_sp = sp();
1713 SafePointNode* old_map = clone_map();
1714
1715 value = must_be_not_null(value, true);
1716
1717 Node* adr = array_element_address(value, index, T_CHAR);
1718 if (adr->is_top()) {
1719 set_map(old_map);
1720 set_sp(old_sp);
1721 return false;
1722 }
1723 destruct_map_clone(old_map);
1724 if (is_store) {
1725 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1726 } else {
1727 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);
1728 set_result(ch);
1729 }
1730 return true;
1731 }
1732
1733
1734 //------------------------------inline_math-----------------------------------
1735 // public static double Math.abs(double)
1736 // public static double Math.sqrt(double)
1737 // public static double Math.log(double)
1738 // public static double Math.log10(double)
1739 // public static double Math.round(double)
1740 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1741 Node* arg = argument(0);
1742 Node* n = nullptr;
1743 switch (id) {
1744 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1745 case vmIntrinsics::_dsqrt:
1746 case vmIntrinsics::_dsqrt_strict:
1747 n = new SqrtDNode(C, control(), arg); break;
1748 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1749 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1750 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1751 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1752 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1753 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1754 default: fatal_unexpected_iid(id); break;
1755 }
1756 set_result(_gvn.transform(n));
1757 return true;
1758 }
1759
1760 //------------------------------inline_math-----------------------------------
1761 // public static float Math.abs(float)
1762 // public static int Math.abs(int)
1763 // public static long Math.abs(long)
1764 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1765 Node* arg = argument(0);
1766 Node* n = nullptr;
1767 switch (id) {
1768 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1769 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1770 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1771 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1772 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1773 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1774 default: fatal_unexpected_iid(id); break;
1775 }
1776 set_result(_gvn.transform(n));
1777 return true;
1778 }
1779
1780 //------------------------------runtime_math-----------------------------
1781 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1782 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1783 "must be (DD)D or (D)D type");
1784
1785 // Inputs
1786 Node* a = argument(0);
1787 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1788
1789 const TypePtr* no_memory_effects = nullptr;
1790 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1791 no_memory_effects,
1792 a, top(), b, b ? top() : nullptr);
1793 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1794 #ifdef ASSERT
1795 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1796 assert(value_top == top(), "second value must be top");
1797 #endif
1798
1799 set_result(value);
1800 return true;
1801 }
1802
1803 //------------------------------inline_math_pow-----------------------------
1804 bool LibraryCallKit::inline_math_pow() {
1805 Node* exp = argument(2);
1806 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1807 if (d != nullptr) {
1808 if (d->getd() == 2.0) {
1809 // Special case: pow(x, 2.0) => x * x
1810 Node* base = argument(0);
1811 set_result(_gvn.transform(new MulDNode(base, base)));
1812 return true;
1813 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1814 // Special case: pow(x, 0.5) => sqrt(x)
1815 Node* base = argument(0);
1816 Node* zero = _gvn.zerocon(T_DOUBLE);
1817
1818 RegionNode* region = new RegionNode(3);
1819 Node* phi = new PhiNode(region, Type::DOUBLE);
1820
1821 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1822 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1823 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1824 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1825 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1826
1827 Node* if_pow = generate_slow_guard(test, nullptr);
1828 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1829 phi->init_req(1, value_sqrt);
1830 region->init_req(1, control());
1831
1832 if (if_pow != nullptr) {
1833 set_control(if_pow);
1834 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1835 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1836 const TypePtr* no_memory_effects = nullptr;
1837 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1838 no_memory_effects, base, top(), exp, top());
1839 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1840 #ifdef ASSERT
1841 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1842 assert(value_top == top(), "second value must be top");
1843 #endif
1844 phi->init_req(2, value_pow);
1845 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1846 }
1847
1848 C->set_has_split_ifs(true); // Has chance for split-if optimization
1849 set_control(_gvn.transform(region));
1850 record_for_igvn(region);
1851 set_result(_gvn.transform(phi));
1852
1853 return true;
1854 }
1855 }
1856
1857 return StubRoutines::dpow() != nullptr ?
1858 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1859 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1860 }
1861
1862 //------------------------------inline_math_native-----------------------------
1863 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1864 switch (id) {
1865 case vmIntrinsics::_dsin:
1866 return StubRoutines::dsin() != nullptr ?
1867 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1868 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1869 case vmIntrinsics::_dcos:
1870 return StubRoutines::dcos() != nullptr ?
1871 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1872 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1873 case vmIntrinsics::_dtan:
1874 return StubRoutines::dtan() != nullptr ?
1875 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1876 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1877 case vmIntrinsics::_dtanh:
1878 return StubRoutines::dtanh() != nullptr ?
1879 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1880 case vmIntrinsics::_dexp:
1881 return StubRoutines::dexp() != nullptr ?
1882 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1883 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1884 case vmIntrinsics::_dlog:
1885 return StubRoutines::dlog() != nullptr ?
1886 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1887 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1888 case vmIntrinsics::_dlog10:
1889 return StubRoutines::dlog10() != nullptr ?
1890 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1891 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1892
1893 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1894 case vmIntrinsics::_ceil:
1895 case vmIntrinsics::_floor:
1896 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1897
1898 case vmIntrinsics::_dsqrt:
1899 case vmIntrinsics::_dsqrt_strict:
1900 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1901 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1902 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1903 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1904 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1905
1906 case vmIntrinsics::_dpow: return inline_math_pow();
1907 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1908 case vmIntrinsics::_fcopySign: return inline_math(id);
1909 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1910 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1911 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1912
1913 // These intrinsics are not yet correctly implemented
1914 case vmIntrinsics::_datan2:
1915 return false;
1916
1917 default:
1918 fatal_unexpected_iid(id);
1919 return false;
1920 }
1921 }
1922
1923 //----------------------------inline_notify-----------------------------------*
1924 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1925 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1926 address func;
1927 if (id == vmIntrinsics::_notify) {
1928 func = OptoRuntime::monitor_notify_Java();
1929 } else {
1930 func = OptoRuntime::monitor_notifyAll_Java();
1931 }
1932 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1933 make_slow_call_ex(call, env()->Throwable_klass(), false);
1934 return true;
1935 }
1936
1937
1938 //----------------------------inline_min_max-----------------------------------
1939 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1940 Node* a = nullptr;
1941 Node* b = nullptr;
1942 Node* n = nullptr;
1943 switch (id) {
1944 case vmIntrinsics::_min:
1945 case vmIntrinsics::_max:
1946 case vmIntrinsics::_minF:
1947 case vmIntrinsics::_maxF:
1948 case vmIntrinsics::_minF_strict:
1949 case vmIntrinsics::_maxF_strict:
1950 case vmIntrinsics::_min_strict:
1951 case vmIntrinsics::_max_strict:
1952 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
1953 a = argument(0);
1954 b = argument(1);
1955 break;
1956 case vmIntrinsics::_minD:
1957 case vmIntrinsics::_maxD:
1958 case vmIntrinsics::_minD_strict:
1959 case vmIntrinsics::_maxD_strict:
1960 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
1961 a = argument(0);
1962 b = argument(2);
1963 break;
1964 case vmIntrinsics::_minL:
1965 case vmIntrinsics::_maxL:
1966 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
1967 a = argument(0);
1968 b = argument(2);
1969 break;
1970 default:
1971 fatal_unexpected_iid(id);
1972 break;
1973 }
1974
1975 switch (id) {
1976 case vmIntrinsics::_min:
1977 case vmIntrinsics::_min_strict:
1978 n = new MinINode(a, b);
1979 break;
1980 case vmIntrinsics::_max:
1981 case vmIntrinsics::_max_strict:
1982 n = new MaxINode(a, b);
1983 break;
1984 case vmIntrinsics::_minF:
1985 case vmIntrinsics::_minF_strict:
1986 n = new MinFNode(a, b);
1987 break;
1988 case vmIntrinsics::_maxF:
1989 case vmIntrinsics::_maxF_strict:
1990 n = new MaxFNode(a, b);
1991 break;
1992 case vmIntrinsics::_minD:
1993 case vmIntrinsics::_minD_strict:
1994 n = new MinDNode(a, b);
1995 break;
1996 case vmIntrinsics::_maxD:
1997 case vmIntrinsics::_maxD_strict:
1998 n = new MaxDNode(a, b);
1999 break;
2000 case vmIntrinsics::_minL:
2001 n = new MinLNode(_gvn.C, a, b);
2002 break;
2003 case vmIntrinsics::_maxL:
2004 n = new MaxLNode(_gvn.C, a, b);
2005 break;
2006 default:
2007 fatal_unexpected_iid(id);
2008 break;
2009 }
2010
2011 set_result(_gvn.transform(n));
2012 return true;
2013 }
2014
2015 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
2016 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2017 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2018 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2019 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2020
2021 {
2022 PreserveJVMState pjvms(this);
2023 PreserveReexecuteState preexecs(this);
2024 jvms()->set_should_reexecute(true);
2025
2026 set_control(slow_path);
2027 set_i_o(i_o());
2028
2029 uncommon_trap(Deoptimization::Reason_intrinsic,
2030 Deoptimization::Action_none);
2031 }
2032
2033 set_control(fast_path);
2034 set_result(math);
2035 }
2036
2037 template <typename OverflowOp>
2038 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2039 typedef typename OverflowOp::MathOp MathOp;
2040
2041 MathOp* mathOp = new MathOp(arg1, arg2);
2042 Node* operation = _gvn.transform( mathOp );
2043 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2044 inline_math_mathExact(operation, ofcheck);
2045 return true;
2046 }
2047
2048 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2049 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2050 }
2051
2052 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2053 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2054 }
2055
2056 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2057 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2058 }
2059
2060 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2061 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2062 }
2063
2064 bool LibraryCallKit::inline_math_negateExactI() {
2065 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2066 }
2067
2068 bool LibraryCallKit::inline_math_negateExactL() {
2069 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2070 }
2071
2072 bool LibraryCallKit::inline_math_multiplyExactI() {
2073 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2074 }
2075
2076 bool LibraryCallKit::inline_math_multiplyExactL() {
2077 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2078 }
2079
2080 bool LibraryCallKit::inline_math_multiplyHigh() {
2081 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2082 return true;
2083 }
2084
2085 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2086 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2087 return true;
2088 }
2089
2090 inline int
2091 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2092 const TypePtr* base_type = TypePtr::NULL_PTR;
2093 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2094 if (base_type == nullptr) {
2095 // Unknown type.
2096 return Type::AnyPtr;
2097 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2098 // Since this is a null+long form, we have to switch to a rawptr.
2099 base = _gvn.transform(new CastX2PNode(offset));
2100 offset = MakeConX(0);
2101 return Type::RawPtr;
2102 } else if (base_type->base() == Type::RawPtr) {
2103 return Type::RawPtr;
2104 } else if (base_type->isa_oopptr()) {
2105 // Base is never null => always a heap address.
2106 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2107 return Type::OopPtr;
2108 }
2109 // Offset is small => always a heap address.
2110 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2111 if (offset_type != nullptr &&
2112 base_type->offset() == 0 && // (should always be?)
2113 offset_type->_lo >= 0 &&
2114 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2115 return Type::OopPtr;
2116 } else if (type == T_OBJECT) {
2117 // off heap access to an oop doesn't make any sense. Has to be on
2118 // heap.
2119 return Type::OopPtr;
2120 }
2121 // Otherwise, it might either be oop+off or null+addr.
2122 return Type::AnyPtr;
2123 } else {
2124 // No information:
2125 return Type::AnyPtr;
2126 }
2127 }
2128
2129 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2130 Node* uncasted_base = base;
2131 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2132 if (kind == Type::RawPtr) {
2133 return basic_plus_adr(top(), uncasted_base, offset);
2134 } else if (kind == Type::AnyPtr) {
2135 assert(base == uncasted_base, "unexpected base change");
2136 if (can_cast) {
2137 if (!_gvn.type(base)->speculative_maybe_null() &&
2138 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2139 // According to profiling, this access is always on
2140 // heap. Casting the base to not null and thus avoiding membars
2141 // around the access should allow better optimizations
2142 Node* null_ctl = top();
2143 base = null_check_oop(base, &null_ctl, true, true, true);
2144 assert(null_ctl->is_top(), "no null control here");
2145 return basic_plus_adr(base, offset);
2146 } else if (_gvn.type(base)->speculative_always_null() &&
2147 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2148 // According to profiling, this access is always off
2149 // heap.
2150 base = null_assert(base);
2151 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2152 offset = MakeConX(0);
2153 return basic_plus_adr(top(), raw_base, offset);
2154 }
2155 }
2156 // We don't know if it's an on heap or off heap access. Fall back
2157 // to raw memory access.
2158 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2159 return basic_plus_adr(top(), raw, offset);
2160 } else {
2161 assert(base == uncasted_base, "unexpected base change");
2162 // We know it's an on heap access so base can't be null
2163 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2164 base = must_be_not_null(base, true);
2165 }
2166 return basic_plus_adr(base, offset);
2167 }
2168 }
2169
2170 //--------------------------inline_number_methods-----------------------------
2171 // inline int Integer.numberOfLeadingZeros(int)
2172 // inline int Long.numberOfLeadingZeros(long)
2173 //
2174 // inline int Integer.numberOfTrailingZeros(int)
2175 // inline int Long.numberOfTrailingZeros(long)
2176 //
2177 // inline int Integer.bitCount(int)
2178 // inline int Long.bitCount(long)
2179 //
2180 // inline char Character.reverseBytes(char)
2181 // inline short Short.reverseBytes(short)
2182 // inline int Integer.reverseBytes(int)
2183 // inline long Long.reverseBytes(long)
2184 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2185 Node* arg = argument(0);
2186 Node* n = nullptr;
2187 switch (id) {
2188 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2189 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2190 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2191 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2192 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2193 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2194 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2195 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2196 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2197 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2198 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2199 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2200 default: fatal_unexpected_iid(id); break;
2201 }
2202 set_result(_gvn.transform(n));
2203 return true;
2204 }
2205
2206 //--------------------------inline_bitshuffle_methods-----------------------------
2207 // inline int Integer.compress(int, int)
2208 // inline int Integer.expand(int, int)
2209 // inline long Long.compress(long, long)
2210 // inline long Long.expand(long, long)
2211 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2212 Node* n = nullptr;
2213 switch (id) {
2214 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2215 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2216 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2217 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2218 default: fatal_unexpected_iid(id); break;
2219 }
2220 set_result(_gvn.transform(n));
2221 return true;
2222 }
2223
2224 //--------------------------inline_number_methods-----------------------------
2225 // inline int Integer.compareUnsigned(int, int)
2226 // inline int Long.compareUnsigned(long, long)
2227 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2228 Node* arg1 = argument(0);
2229 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2230 Node* n = nullptr;
2231 switch (id) {
2232 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2233 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2234 default: fatal_unexpected_iid(id); break;
2235 }
2236 set_result(_gvn.transform(n));
2237 return true;
2238 }
2239
2240 //--------------------------inline_unsigned_divmod_methods-----------------------------
2241 // inline int Integer.divideUnsigned(int, int)
2242 // inline int Integer.remainderUnsigned(int, int)
2243 // inline long Long.divideUnsigned(long, long)
2244 // inline long Long.remainderUnsigned(long, long)
2245 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2246 Node* n = nullptr;
2247 switch (id) {
2248 case vmIntrinsics::_divideUnsigned_i: {
2249 zero_check_int(argument(1));
2250 // Compile-time detect of null-exception
2251 if (stopped()) {
2252 return true; // keep the graph constructed so far
2253 }
2254 n = new UDivINode(control(), argument(0), argument(1));
2255 break;
2256 }
2257 case vmIntrinsics::_divideUnsigned_l: {
2258 zero_check_long(argument(2));
2259 // Compile-time detect of null-exception
2260 if (stopped()) {
2261 return true; // keep the graph constructed so far
2262 }
2263 n = new UDivLNode(control(), argument(0), argument(2));
2264 break;
2265 }
2266 case vmIntrinsics::_remainderUnsigned_i: {
2267 zero_check_int(argument(1));
2268 // Compile-time detect of null-exception
2269 if (stopped()) {
2270 return true; // keep the graph constructed so far
2271 }
2272 n = new UModINode(control(), argument(0), argument(1));
2273 break;
2274 }
2275 case vmIntrinsics::_remainderUnsigned_l: {
2276 zero_check_long(argument(2));
2277 // Compile-time detect of null-exception
2278 if (stopped()) {
2279 return true; // keep the graph constructed so far
2280 }
2281 n = new UModLNode(control(), argument(0), argument(2));
2282 break;
2283 }
2284 default: fatal_unexpected_iid(id); break;
2285 }
2286 set_result(_gvn.transform(n));
2287 return true;
2288 }
2289
2290 //----------------------------inline_unsafe_access----------------------------
2291
2292 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2293 // Attempt to infer a sharper value type from the offset and base type.
2294 ciKlass* sharpened_klass = nullptr;
2295 bool null_free = false;
2296
2297 // See if it is an instance field, with an object type.
2298 if (alias_type->field() != nullptr) {
2299 if (alias_type->field()->type()->is_klass()) {
2300 sharpened_klass = alias_type->field()->type()->as_klass();
2301 null_free = alias_type->field()->is_null_free();
2302 }
2303 }
2304
2305 const TypeOopPtr* result = nullptr;
2306 // See if it is a narrow oop array.
2307 if (adr_type->isa_aryptr()) {
2308 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2309 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2310 null_free = adr_type->is_aryptr()->is_null_free();
2311 if (elem_type != nullptr && elem_type->is_loaded()) {
2312 // Sharpen the value type.
2313 result = elem_type;
2314 }
2315 }
2316 }
2317
2318 // The sharpened class might be unloaded if there is no class loader
2319 // contraint in place.
2320 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2321 // Sharpen the value type.
2322 result = TypeOopPtr::make_from_klass(sharpened_klass);
2323 if (null_free) {
2324 result = result->join_speculative(TypePtr::NOTNULL)->is_oopptr();
2325 }
2326 }
2327 if (result != nullptr) {
2328 #ifndef PRODUCT
2329 if (C->print_intrinsics() || C->print_inlining()) {
2330 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2331 tty->print(" sharpened value: "); result->dump(); tty->cr();
2332 }
2333 #endif
2334 }
2335 return result;
2336 }
2337
2338 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2339 switch (kind) {
2340 case Relaxed:
2341 return MO_UNORDERED;
2342 case Opaque:
2343 return MO_RELAXED;
2344 case Acquire:
2345 return MO_ACQUIRE;
2346 case Release:
2347 return MO_RELEASE;
2348 case Volatile:
2349 return MO_SEQ_CST;
2350 default:
2351 ShouldNotReachHere();
2352 return 0;
2353 }
2354 }
2355
2356 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned, const bool is_flat) {
2357 if (callee()->is_static()) return false; // caller must have the capability!
2358 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2359 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2360 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2361 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2362
2363 if (is_reference_type(type)) {
2364 decorators |= ON_UNKNOWN_OOP_REF;
2365 }
2366
2367 if (unaligned) {
2368 decorators |= C2_UNALIGNED;
2369 }
2370
2371 #ifndef PRODUCT
2372 {
2373 ResourceMark rm;
2374 // Check the signatures.
2375 ciSignature* sig = callee()->signature();
2376 #ifdef ASSERT
2377 if (!is_store) {
2378 // Object getReference(Object base, int/long offset), etc.
2379 BasicType rtype = sig->return_type()->basic_type();
2380 assert(rtype == type, "getter must return the expected value");
2381 assert(sig->count() == 2 || (is_flat && sig->count() == 3), "oop getter has 2 or 3 arguments");
2382 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2383 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2384 } else {
2385 // void putReference(Object base, int/long offset, Object x), etc.
2386 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2387 assert(sig->count() == 3 || (is_flat && sig->count() == 4), "oop putter has 3 arguments");
2388 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2389 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2390 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2391 assert(vtype == type, "putter must accept the expected value");
2392 }
2393 #endif // ASSERT
2394 }
2395 #endif //PRODUCT
2396
2397 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2398
2399 Node* receiver = argument(0); // type: oop
2400
2401 // Build address expression.
2402 Node* heap_base_oop = top();
2403
2404 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2405 Node* base = argument(1); // type: oop
2406 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2407 Node* offset = argument(2); // type: long
2408 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2409 // to be plain byte offsets, which are also the same as those accepted
2410 // by oopDesc::field_addr.
2411 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2412 "fieldOffset must be byte-scaled");
2413
2414 ciInlineKlass* inline_klass = nullptr;
2415 if (is_flat) {
2416 const TypeInstPtr* cls = _gvn.type(argument(4))->isa_instptr();
2417 if (cls == nullptr || cls->const_oop() == nullptr) {
2418 return false;
2419 }
2420 ciType* mirror_type = cls->const_oop()->as_instance()->java_mirror_type();
2421 if (!mirror_type->is_inlinetype()) {
2422 return false;
2423 }
2424 inline_klass = mirror_type->as_inline_klass();
2425 }
2426
2427 if (base->is_InlineType()) {
2428 InlineTypeNode* vt = base->as_InlineType();
2429 if (is_store) {
2430 if (!vt->is_allocated(&_gvn)) {
2431 return false;
2432 }
2433 base = vt->get_oop();
2434 } else {
2435 if (offset->is_Con()) {
2436 long off = find_long_con(offset, 0);
2437 ciInlineKlass* vk = vt->type()->inline_klass();
2438 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2439 return false;
2440 }
2441
2442 ciField* field = vk->get_non_flat_field_by_offset(off);
2443 if (field != nullptr) {
2444 BasicType bt = type2field[field->type()->basic_type()];
2445 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2446 bt = T_OBJECT;
2447 }
2448 if (bt == type && (!field->is_flat() || field->type() == inline_klass)) {
2449 Node* value = vt->field_value_by_offset(off, false);
2450 if (value->is_InlineType()) {
2451 value = value->as_InlineType()->adjust_scalarization_depth(this);
2452 }
2453 set_result(value);
2454 return true;
2455 }
2456 }
2457 }
2458 {
2459 // Re-execute the unsafe access if allocation triggers deoptimization.
2460 PreserveReexecuteState preexecs(this);
2461 jvms()->set_should_reexecute(true);
2462 vt = vt->buffer(this);
2463 }
2464 base = vt->get_oop();
2465 }
2466 }
2467
2468 // 32-bit machines ignore the high half!
2469 offset = ConvL2X(offset);
2470
2471 // Save state and restore on bailout
2472 uint old_sp = sp();
2473 SafePointNode* old_map = clone_map();
2474
2475 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2476 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2477
2478 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2479 if (type != T_OBJECT && (inline_klass == nullptr || !inline_klass->has_object_fields())) {
2480 decorators |= IN_NATIVE; // off-heap primitive access
2481 } else {
2482 set_map(old_map);
2483 set_sp(old_sp);
2484 return false; // off-heap oop accesses are not supported
2485 }
2486 } else {
2487 heap_base_oop = base; // on-heap or mixed access
2488 }
2489
2490 // Can base be null? Otherwise, always on-heap access.
2491 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2492
2493 if (!can_access_non_heap) {
2494 decorators |= IN_HEAP;
2495 }
2496
2497 Node* val = is_store ? argument(4 + (is_flat ? 1 : 0)) : nullptr;
2498
2499 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2500 if (adr_type == TypePtr::NULL_PTR) {
2501 set_map(old_map);
2502 set_sp(old_sp);
2503 return false; // off-heap access with zero address
2504 }
2505
2506 // Try to categorize the address.
2507 Compile::AliasType* alias_type = C->alias_type(adr_type);
2508 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2509
2510 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2511 alias_type->adr_type() == TypeAryPtr::RANGE) {
2512 set_map(old_map);
2513 set_sp(old_sp);
2514 return false; // not supported
2515 }
2516
2517 bool mismatched = false;
2518 BasicType bt = T_ILLEGAL;
2519 ciField* field = nullptr;
2520 if (adr_type->isa_instptr()) {
2521 const TypeInstPtr* instptr = adr_type->is_instptr();
2522 ciInstanceKlass* k = instptr->instance_klass();
2523 int off = instptr->offset();
2524 if (instptr->const_oop() != nullptr &&
2525 k == ciEnv::current()->Class_klass() &&
2526 instptr->offset() >= (k->size_helper() * wordSize)) {
2527 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2528 field = k->get_field_by_offset(off, true);
2529 } else {
2530 field = k->get_non_flat_field_by_offset(off);
2531 }
2532 if (field != nullptr) {
2533 bt = type2field[field->type()->basic_type()];
2534 }
2535 if (bt != alias_type->basic_type()) {
2536 // Type mismatch. Is it an access to a nested flat field?
2537 field = k->get_field_by_offset(off, false);
2538 if (field != nullptr) {
2539 bt = type2field[field->type()->basic_type()];
2540 }
2541 }
2542 assert(bt == alias_type->basic_type() || is_flat, "should match");
2543 } else {
2544 bt = alias_type->basic_type();
2545 }
2546
2547 if (bt != T_ILLEGAL) {
2548 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2549 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2550 // Alias type doesn't differentiate between byte[] and boolean[]).
2551 // Use address type to get the element type.
2552 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2553 }
2554 if (is_reference_type(bt, true)) {
2555 // accessing an array field with getReference is not a mismatch
2556 bt = T_OBJECT;
2557 }
2558 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2559 // Don't intrinsify mismatched object accesses
2560 set_map(old_map);
2561 set_sp(old_sp);
2562 return false;
2563 }
2564 mismatched = (bt != type);
2565 } else if (alias_type->adr_type()->isa_oopptr()) {
2566 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2567 }
2568
2569 if (is_flat) {
2570 if (adr_type->isa_instptr()) {
2571 if (field == nullptr || field->type() != inline_klass) {
2572 mismatched = true;
2573 }
2574 } else if (adr_type->isa_aryptr()) {
2575 const Type* elem = adr_type->is_aryptr()->elem();
2576 if (!adr_type->is_flat() || elem->inline_klass() != inline_klass) {
2577 mismatched = true;
2578 }
2579 } else {
2580 mismatched = true;
2581 }
2582 if (is_store) {
2583 const Type* val_t = _gvn.type(val);
2584 if (!val_t->is_inlinetypeptr() || val_t->inline_klass() != inline_klass) {
2585 set_map(old_map);
2586 set_sp(old_sp);
2587 return false;
2588 }
2589 }
2590 }
2591
2592 destruct_map_clone(old_map);
2593 assert(!mismatched || is_flat || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2594
2595 if (mismatched) {
2596 decorators |= C2_MISMATCHED;
2597 }
2598
2599 // First guess at the value type.
2600 const Type *value_type = Type::get_const_basic_type(type);
2601
2602 // Figure out the memory ordering.
2603 decorators |= mo_decorator_for_access_kind(kind);
2604
2605 if (!is_store) {
2606 if (type == T_OBJECT && !is_flat) {
2607 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2608 if (tjp != nullptr) {
2609 value_type = tjp;
2610 }
2611 }
2612 }
2613
2614 receiver = null_check(receiver);
2615 if (stopped()) {
2616 return true;
2617 }
2618 // Heap pointers get a null-check from the interpreter,
2619 // as a courtesy. However, this is not guaranteed by Unsafe,
2620 // and it is not possible to fully distinguish unintended nulls
2621 // from intended ones in this API.
2622
2623 if (!is_store) {
2624 Node* p = nullptr;
2625 // Try to constant fold a load from a constant field
2626
2627 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !field->is_flat() && !mismatched) {
2628 // final or stable field
2629 p = make_constant_from_field(field, heap_base_oop);
2630 }
2631
2632 if (p == nullptr) { // Could not constant fold the load
2633 if (is_flat) {
2634 if (adr_type->isa_instptr() && !mismatched) {
2635 ciInstanceKlass* holder = adr_type->is_instptr()->instance_klass();
2636 int offset = adr_type->is_instptr()->offset();
2637 p = InlineTypeNode::make_from_flat(this, inline_klass, base, base, nullptr, holder, offset, false, -1, decorators);
2638 } else {
2639 p = InlineTypeNode::make_from_flat(this, inline_klass, base, adr, nullptr, nullptr, 0, false, -1, decorators);
2640 }
2641 } else {
2642 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2643 const TypeOopPtr* ptr = value_type->make_oopptr();
2644 if (ptr != nullptr && ptr->is_inlinetypeptr()) {
2645 // Load a non-flattened inline type from memory
2646 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass());
2647 }
2648 }
2649 // Normalize the value returned by getBoolean in the following cases
2650 if (type == T_BOOLEAN &&
2651 (mismatched ||
2652 heap_base_oop == top() || // - heap_base_oop is null or
2653 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2654 // and the unsafe access is made to large offset
2655 // (i.e., larger than the maximum offset necessary for any
2656 // field access)
2657 ) {
2658 IdealKit ideal = IdealKit(this);
2659 #define __ ideal.
2660 IdealVariable normalized_result(ideal);
2661 __ declarations_done();
2662 __ set(normalized_result, p);
2663 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2664 __ set(normalized_result, ideal.ConI(1));
2665 ideal.end_if();
2666 final_sync(ideal);
2667 p = __ value(normalized_result);
2668 #undef __
2669 }
2670 }
2671 if (type == T_ADDRESS) {
2672 p = gvn().transform(new CastP2XNode(nullptr, p));
2673 p = ConvX2UL(p);
2674 }
2675 // The load node has the control of the preceding MemBarCPUOrder. All
2676 // following nodes will have the control of the MemBarCPUOrder inserted at
2677 // the end of this method. So, pushing the load onto the stack at a later
2678 // point is fine.
2679 set_result(p);
2680 } else {
2681 if (bt == T_ADDRESS) {
2682 // Repackage the long as a pointer.
2683 val = ConvL2X(val);
2684 val = gvn().transform(new CastX2PNode(val));
2685 }
2686 if (is_flat) {
2687 if (adr_type->isa_instptr() && !mismatched) {
2688 ciInstanceKlass* holder = adr_type->is_instptr()->instance_klass();
2689 int offset = adr_type->is_instptr()->offset();
2690 val->as_InlineType()->store_flat(this, base, base, nullptr, holder, offset, false, -1, decorators);
2691 } else {
2692 val->as_InlineType()->store_flat(this, base, adr, nullptr, val->bottom_type()->inline_klass(), 0, false, -1, decorators);
2693 }
2694 } else {
2695 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2696 }
2697 }
2698
2699 if (argument(1)->is_InlineType() && is_store) {
2700 InlineTypeNode* value = InlineTypeNode::make_from_oop(this, base, _gvn.type(argument(1))->inline_klass());
2701 value = value->make_larval(this, false);
2702 replace_in_map(argument(1), value);
2703 }
2704
2705 return true;
2706 }
2707
2708 bool LibraryCallKit::inline_unsafe_make_private_buffer() {
2709 Node* receiver = argument(0);
2710 Node* value = argument(1);
2711
2712 const Type* type = gvn().type(value);
2713 if (!type->is_inlinetypeptr()) {
2714 C->record_method_not_compilable("value passed to Unsafe::makePrivateBuffer is not of a constant value type");
2715 return false;
2716 }
2717
2718 null_check(receiver);
2719 if (stopped()) {
2720 return true;
2721 }
2722
2723 value = null_check(value);
2724 if (stopped()) {
2725 return true;
2726 }
2727
2728 ciInlineKlass* vk = type->inline_klass();
2729 Node* klass = makecon(TypeKlassPtr::make(vk));
2730 Node* obj = new_instance(klass);
2731 AllocateNode::Ideal_allocation(obj)->_larval = true;
2732
2733 assert(value->is_InlineType(), "must be an InlineTypeNode");
2734 value->as_InlineType()->store(this, obj, obj, vk);
2735
2736 set_result(obj);
2737 return true;
2738 }
2739
2740 bool LibraryCallKit::inline_unsafe_finish_private_buffer() {
2741 Node* receiver = argument(0);
2742 Node* buffer = argument(1);
2743
2744 const Type* type = gvn().type(buffer);
2745 if (!type->is_inlinetypeptr()) {
2746 C->record_method_not_compilable("value passed to Unsafe::finishPrivateBuffer is not of a constant value type");
2747 return false;
2748 }
2749
2750 AllocateNode* alloc = AllocateNode::Ideal_allocation(buffer);
2751 if (alloc == nullptr) {
2752 C->record_method_not_compilable("value passed to Unsafe::finishPrivateBuffer must be allocated by Unsafe::makePrivateBuffer");
2753 return false;
2754 }
2755
2756 null_check(receiver);
2757 if (stopped()) {
2758 return true;
2759 }
2760
2761 // Unset the larval bit in the object header
2762 Node* old_header = make_load(control(), buffer, TypeX_X, TypeX_X->basic_type(), MemNode::unordered, LoadNode::Pinned);
2763 Node* new_header = gvn().transform(new AndXNode(old_header, MakeConX(~markWord::larval_bit_in_place)));
2764 access_store_at(buffer, buffer, type->is_ptr(), new_header, TypeX_X, TypeX_X->basic_type(), MO_UNORDERED | IN_HEAP);
2765
2766 // We must ensure that the buffer is properly published
2767 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
2768 assert(!type->maybe_null(), "result of an allocation should not be null");
2769 set_result(InlineTypeNode::make_from_oop(this, buffer, type->inline_klass(), false));
2770 return true;
2771 }
2772
2773 //----------------------------inline_unsafe_load_store----------------------------
2774 // This method serves a couple of different customers (depending on LoadStoreKind):
2775 //
2776 // LS_cmp_swap:
2777 //
2778 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2779 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2780 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2781 //
2782 // LS_cmp_swap_weak:
2783 //
2784 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2785 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2786 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2787 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2788 //
2789 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2790 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2791 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2792 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2793 //
2794 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2795 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2796 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2797 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2798 //
2799 // LS_cmp_exchange:
2800 //
2801 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2802 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2803 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2804 //
2805 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2806 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2807 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2808 //
2809 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2810 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2811 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2812 //
2813 // LS_get_add:
2814 //
2815 // int getAndAddInt( Object o, long offset, int delta)
2816 // long getAndAddLong(Object o, long offset, long delta)
2817 //
2818 // LS_get_set:
2819 //
2820 // int getAndSet(Object o, long offset, int newValue)
2821 // long getAndSet(Object o, long offset, long newValue)
2822 // Object getAndSet(Object o, long offset, Object newValue)
2823 //
2824 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2825 // This basic scheme here is the same as inline_unsafe_access, but
2826 // differs in enough details that combining them would make the code
2827 // overly confusing. (This is a true fact! I originally combined
2828 // them, but even I was confused by it!) As much code/comments as
2829 // possible are retained from inline_unsafe_access though to make
2830 // the correspondences clearer. - dl
2831
2832 if (callee()->is_static()) return false; // caller must have the capability!
2833
2834 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2835 decorators |= mo_decorator_for_access_kind(access_kind);
2836
2837 #ifndef PRODUCT
2838 BasicType rtype;
2839 {
2840 ResourceMark rm;
2841 // Check the signatures.
2842 ciSignature* sig = callee()->signature();
2843 rtype = sig->return_type()->basic_type();
2844 switch(kind) {
2845 case LS_get_add:
2846 case LS_get_set: {
2847 // Check the signatures.
2848 #ifdef ASSERT
2849 assert(rtype == type, "get and set must return the expected type");
2850 assert(sig->count() == 3, "get and set has 3 arguments");
2851 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2852 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2853 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2854 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2855 #endif // ASSERT
2856 break;
2857 }
2858 case LS_cmp_swap:
2859 case LS_cmp_swap_weak: {
2860 // Check the signatures.
2861 #ifdef ASSERT
2862 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2863 assert(sig->count() == 4, "CAS has 4 arguments");
2864 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2865 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2866 #endif // ASSERT
2867 break;
2868 }
2869 case LS_cmp_exchange: {
2870 // Check the signatures.
2871 #ifdef ASSERT
2872 assert(rtype == type, "CAS must return the expected type");
2873 assert(sig->count() == 4, "CAS has 4 arguments");
2874 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2875 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2876 #endif // ASSERT
2877 break;
2878 }
2879 default:
2880 ShouldNotReachHere();
2881 }
2882 }
2883 #endif //PRODUCT
2884
2885 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2886
2887 // Get arguments:
2888 Node* receiver = nullptr;
2889 Node* base = nullptr;
2890 Node* offset = nullptr;
2891 Node* oldval = nullptr;
2892 Node* newval = nullptr;
2893 switch(kind) {
2894 case LS_cmp_swap:
2895 case LS_cmp_swap_weak:
2896 case LS_cmp_exchange: {
2897 const bool two_slot_type = type2size[type] == 2;
2898 receiver = argument(0); // type: oop
2899 base = argument(1); // type: oop
2900 offset = argument(2); // type: long
2901 oldval = argument(4); // type: oop, int, or long
2902 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2903 break;
2904 }
2905 case LS_get_add:
2906 case LS_get_set: {
2907 receiver = argument(0); // type: oop
2908 base = argument(1); // type: oop
2909 offset = argument(2); // type: long
2910 oldval = nullptr;
2911 newval = argument(4); // type: oop, int, or long
2912 break;
2913 }
2914 default:
2915 ShouldNotReachHere();
2916 }
2917
2918 // Build field offset expression.
2919 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2920 // to be plain byte offsets, which are also the same as those accepted
2921 // by oopDesc::field_addr.
2922 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2923 // 32-bit machines ignore the high half of long offsets
2924 offset = ConvL2X(offset);
2925 // Save state and restore on bailout
2926 uint old_sp = sp();
2927 SafePointNode* old_map = clone_map();
2928 Node* adr = make_unsafe_address(base, offset,type, false);
2929 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2930
2931 Compile::AliasType* alias_type = C->alias_type(adr_type);
2932 BasicType bt = alias_type->basic_type();
2933 if (bt != T_ILLEGAL &&
2934 (is_reference_type(bt) != (type == T_OBJECT))) {
2935 // Don't intrinsify mismatched object accesses.
2936 set_map(old_map);
2937 set_sp(old_sp);
2938 return false;
2939 }
2940
2941 destruct_map_clone(old_map);
2942
2943 // For CAS, unlike inline_unsafe_access, there seems no point in
2944 // trying to refine types. Just use the coarse types here.
2945 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2946 const Type *value_type = Type::get_const_basic_type(type);
2947
2948 switch (kind) {
2949 case LS_get_set:
2950 case LS_cmp_exchange: {
2951 if (type == T_OBJECT) {
2952 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2953 if (tjp != nullptr) {
2954 value_type = tjp;
2955 }
2956 }
2957 break;
2958 }
2959 case LS_cmp_swap:
2960 case LS_cmp_swap_weak:
2961 case LS_get_add:
2962 break;
2963 default:
2964 ShouldNotReachHere();
2965 }
2966
2967 // Null check receiver.
2968 receiver = null_check(receiver);
2969 if (stopped()) {
2970 return true;
2971 }
2972
2973 int alias_idx = C->get_alias_index(adr_type);
2974
2975 if (is_reference_type(type)) {
2976 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2977
2978 if (oldval != nullptr && oldval->is_InlineType()) {
2979 // Re-execute the unsafe access if allocation triggers deoptimization.
2980 PreserveReexecuteState preexecs(this);
2981 jvms()->set_should_reexecute(true);
2982 oldval = oldval->as_InlineType()->buffer(this)->get_oop();
2983 }
2984 if (newval != nullptr && newval->is_InlineType()) {
2985 // Re-execute the unsafe access if allocation triggers deoptimization.
2986 PreserveReexecuteState preexecs(this);
2987 jvms()->set_should_reexecute(true);
2988 newval = newval->as_InlineType()->buffer(this)->get_oop();
2989 }
2990
2991 // Transformation of a value which could be null pointer (CastPP #null)
2992 // could be delayed during Parse (for example, in adjust_map_after_if()).
2993 // Execute transformation here to avoid barrier generation in such case.
2994 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2995 newval = _gvn.makecon(TypePtr::NULL_PTR);
2996
2997 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2998 // Refine the value to a null constant, when it is known to be null
2999 oldval = _gvn.makecon(TypePtr::NULL_PTR);
3000 }
3001 }
3002
3003 Node* result = nullptr;
3004 switch (kind) {
3005 case LS_cmp_exchange: {
3006 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
3007 oldval, newval, value_type, type, decorators);
3008 break;
3009 }
3010 case LS_cmp_swap_weak:
3011 decorators |= C2_WEAK_CMPXCHG;
3012 case LS_cmp_swap: {
3013 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
3014 oldval, newval, value_type, type, decorators);
3015 break;
3016 }
3017 case LS_get_set: {
3018 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
3019 newval, value_type, type, decorators);
3020 break;
3021 }
3022 case LS_get_add: {
3023 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
3024 newval, value_type, type, decorators);
3025 break;
3026 }
3027 default:
3028 ShouldNotReachHere();
3029 }
3030
3031 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3032 set_result(result);
3033 return true;
3034 }
3035
3036 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3037 // Regardless of form, don't allow previous ld/st to move down,
3038 // then issue acquire, release, or volatile mem_bar.
3039 insert_mem_bar(Op_MemBarCPUOrder);
3040 switch(id) {
3041 case vmIntrinsics::_loadFence:
3042 insert_mem_bar(Op_LoadFence);
3043 return true;
3044 case vmIntrinsics::_storeFence:
3045 insert_mem_bar(Op_StoreFence);
3046 return true;
3047 case vmIntrinsics::_storeStoreFence:
3048 insert_mem_bar(Op_StoreStoreFence);
3049 return true;
3050 case vmIntrinsics::_fullFence:
3051 insert_mem_bar(Op_MemBarVolatile);
3052 return true;
3053 default:
3054 fatal_unexpected_iid(id);
3055 return false;
3056 }
3057 }
3058
3059 bool LibraryCallKit::inline_onspinwait() {
3060 insert_mem_bar(Op_OnSpinWait);
3061 return true;
3062 }
3063
3064 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3065 if (!kls->is_Con()) {
3066 return true;
3067 }
3068 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
3069 if (klsptr == nullptr) {
3070 return true;
3071 }
3072 ciInstanceKlass* ik = klsptr->instance_klass();
3073 // don't need a guard for a klass that is already initialized
3074 return !ik->is_initialized();
3075 }
3076
3077 //----------------------------inline_unsafe_writeback0-------------------------
3078 // public native void Unsafe.writeback0(long address)
3079 bool LibraryCallKit::inline_unsafe_writeback0() {
3080 if (!Matcher::has_match_rule(Op_CacheWB)) {
3081 return false;
3082 }
3083 #ifndef PRODUCT
3084 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
3085 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
3086 ciSignature* sig = callee()->signature();
3087 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
3088 #endif
3089 null_check_receiver(); // null-check, then ignore
3090 Node *addr = argument(1);
3091 addr = new CastX2PNode(addr);
3092 addr = _gvn.transform(addr);
3093 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
3094 flush = _gvn.transform(flush);
3095 set_memory(flush, TypeRawPtr::BOTTOM);
3096 return true;
3097 }
3098
3099 //----------------------------inline_unsafe_writeback0-------------------------
3100 // public native void Unsafe.writeback0(long address)
3101 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
3102 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
3103 return false;
3104 }
3105 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
3106 return false;
3107 }
3108 #ifndef PRODUCT
3109 assert(Matcher::has_match_rule(Op_CacheWB),
3110 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
3111 : "found match rule for CacheWBPostSync but not CacheWB"));
3112
3113 #endif
3114 null_check_receiver(); // null-check, then ignore
3115 Node *sync;
3116 if (is_pre) {
3117 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3118 } else {
3119 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3120 }
3121 sync = _gvn.transform(sync);
3122 set_memory(sync, TypeRawPtr::BOTTOM);
3123 return true;
3124 }
3125
3126 //----------------------------inline_unsafe_allocate---------------------------
3127 // public native Object Unsafe.allocateInstance(Class<?> cls);
3128 bool LibraryCallKit::inline_unsafe_allocate() {
3129
3130 #if INCLUDE_JVMTI
3131 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3132 return false;
3133 }
3134 #endif //INCLUDE_JVMTI
3135
3136 if (callee()->is_static()) return false; // caller must have the capability!
3137
3138 null_check_receiver(); // null-check, then ignore
3139 Node* cls = null_check(argument(1));
3140 if (stopped()) return true;
3141
3142 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
3143 kls = null_check(kls);
3144 if (stopped()) return true; // argument was like int.class
3145
3146 #if INCLUDE_JVMTI
3147 // Don't try to access new allocated obj in the intrinsic.
3148 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
3149 // Deoptimize and allocate in interpreter instead.
3150 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
3151 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
3152 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
3153 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
3154 {
3155 BuildCutout unless(this, tst, PROB_MAX);
3156 uncommon_trap(Deoptimization::Reason_intrinsic,
3157 Deoptimization::Action_make_not_entrant);
3158 }
3159 if (stopped()) {
3160 return true;
3161 }
3162 #endif //INCLUDE_JVMTI
3163
3164 Node* test = nullptr;
3165 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3166 // Note: The argument might still be an illegal value like
3167 // Serializable.class or Object[].class. The runtime will handle it.
3168 // But we must make an explicit check for initialization.
3169 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3170 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3171 // can generate code to load it as unsigned byte.
3172 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
3173 Node* bits = intcon(InstanceKlass::fully_initialized);
3174 test = _gvn.transform(new SubINode(inst, bits));
3175 // The 'test' is non-zero if we need to take a slow path.
3176 }
3177 Node* obj = nullptr;
3178 const TypeInstKlassPtr* tkls = _gvn.type(kls)->isa_instklassptr();
3179 if (tkls != nullptr && tkls->instance_klass()->is_inlinetype()) {
3180 obj = InlineTypeNode::make_all_zero(_gvn, tkls->instance_klass()->as_inline_klass())->buffer(this);
3181 } else {
3182 obj = new_instance(kls, test);
3183 }
3184 set_result(obj);
3185 return true;
3186 }
3187
3188 //------------------------inline_native_time_funcs--------------
3189 // inline code for System.currentTimeMillis() and System.nanoTime()
3190 // these have the same type and signature
3191 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3192 const TypeFunc* tf = OptoRuntime::void_long_Type();
3193 const TypePtr* no_memory_effects = nullptr;
3194 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3195 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3196 #ifdef ASSERT
3197 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3198 assert(value_top == top(), "second value must be top");
3199 #endif
3200 set_result(value);
3201 return true;
3202 }
3203
3204
3205 #if INCLUDE_JVMTI
3206
3207 // When notifications are disabled then just update the VTMS transition bit and return.
3208 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
3209 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
3210 if (!DoJVMTIVirtualThreadTransitions) {
3211 return true;
3212 }
3213 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3214 IdealKit ideal(this);
3215
3216 Node* ONE = ideal.ConI(1);
3217 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
3218 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
3219 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3220
3221 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
3222 sync_kit(ideal);
3223 // if notifyJvmti enabled then make a call to the given SharedRuntime function
3224 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
3225 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
3226 ideal.sync_kit(this);
3227 } ideal.else_(); {
3228 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
3229 Node* thread = ideal.thread();
3230 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
3231 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
3232
3233 sync_kit(ideal);
3234 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3235 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3236
3237 ideal.sync_kit(this);
3238 } ideal.end_if();
3239 final_sync(ideal);
3240
3241 return true;
3242 }
3243
3244 // Always update the is_disable_suspend bit.
3245 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3246 if (!DoJVMTIVirtualThreadTransitions) {
3247 return true;
3248 }
3249 IdealKit ideal(this);
3250
3251 {
3252 // unconditionally update the is_disable_suspend bit in current JavaThread
3253 Node* thread = ideal.thread();
3254 Node* arg = _gvn.transform(argument(0)); // argument for notification
3255 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3256 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3257
3258 sync_kit(ideal);
3259 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3260 ideal.sync_kit(this);
3261 }
3262 final_sync(ideal);
3263
3264 return true;
3265 }
3266
3267 #endif // INCLUDE_JVMTI
3268
3269 #ifdef JFR_HAVE_INTRINSICS
3270
3271 /**
3272 * if oop->klass != null
3273 * // normal class
3274 * epoch = _epoch_state ? 2 : 1
3275 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3276 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3277 * }
3278 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3279 * else
3280 * // primitive class
3281 * if oop->array_klass != null
3282 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3283 * else
3284 * id = LAST_TYPE_ID + 1 // void class path
3285 * if (!signaled)
3286 * signaled = true
3287 */
3288 bool LibraryCallKit::inline_native_classID() {
3289 Node* cls = argument(0);
3290
3291 IdealKit ideal(this);
3292 #define __ ideal.
3293 IdealVariable result(ideal); __ declarations_done();
3294 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3295 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3296 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3297
3298
3299 __ if_then(kls, BoolTest::ne, null()); {
3300 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3301 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3302
3303 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3304 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3305 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3306 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3307 mask = _gvn.transform(new OrLNode(mask, epoch));
3308 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3309
3310 float unlikely = PROB_UNLIKELY(0.999);
3311 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3312 sync_kit(ideal);
3313 make_runtime_call(RC_LEAF,
3314 OptoRuntime::class_id_load_barrier_Type(),
3315 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3316 "class id load barrier",
3317 TypePtr::BOTTOM,
3318 kls);
3319 ideal.sync_kit(this);
3320 } __ end_if();
3321
3322 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3323 } __ else_(); {
3324 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3325 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3326 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3327 __ if_then(array_kls, BoolTest::ne, null()); {
3328 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3329 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3330 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3331 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3332 } __ else_(); {
3333 // void class case
3334 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3335 } __ end_if();
3336
3337 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3338 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3339 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3340 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3341 } __ end_if();
3342 } __ end_if();
3343
3344 final_sync(ideal);
3345 set_result(ideal.value(result));
3346 #undef __
3347 return true;
3348 }
3349
3350 //------------------------inline_native_jvm_commit------------------
3351 bool LibraryCallKit::inline_native_jvm_commit() {
3352 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3353
3354 // Save input memory and i_o state.
3355 Node* input_memory_state = reset_memory();
3356 set_all_memory(input_memory_state);
3357 Node* input_io_state = i_o();
3358
3359 // TLS.
3360 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3361 // Jfr java buffer.
3362 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3363 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3364 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3365
3366 // Load the current value of the notified field in the JfrThreadLocal.
3367 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3368 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3369
3370 // Test for notification.
3371 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3372 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3373 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3374
3375 // True branch, is notified.
3376 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3377 set_control(is_notified);
3378
3379 // Reset notified state.
3380 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3381 Node* notified_reset_memory = reset_memory();
3382
3383 // 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.
3384 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3385 // Convert the machine-word to a long.
3386 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3387
3388 // False branch, not notified.
3389 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3390 set_control(not_notified);
3391 set_all_memory(input_memory_state);
3392
3393 // Arg is the next position as a long.
3394 Node* arg = argument(0);
3395 // Convert long to machine-word.
3396 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3397
3398 // Store the next_position to the underlying jfr java buffer.
3399 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3400
3401 Node* commit_memory = reset_memory();
3402 set_all_memory(commit_memory);
3403
3404 // 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.
3405 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3406 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3407 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3408
3409 // And flags with lease constant.
3410 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3411
3412 // Branch on lease to conditionalize returning the leased java buffer.
3413 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3414 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3415 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3416
3417 // False branch, not a lease.
3418 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3419
3420 // True branch, is lease.
3421 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3422 set_control(is_lease);
3423
3424 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3425 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3426 OptoRuntime::void_void_Type(),
3427 SharedRuntime::jfr_return_lease(),
3428 "return_lease", TypePtr::BOTTOM);
3429 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3430
3431 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3432 record_for_igvn(lease_compare_rgn);
3433 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3434 record_for_igvn(lease_compare_mem);
3435 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3436 record_for_igvn(lease_compare_io);
3437 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3438 record_for_igvn(lease_result_value);
3439
3440 // Update control and phi nodes.
3441 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3442 lease_compare_rgn->init_req(_false_path, not_lease);
3443
3444 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3445 lease_compare_mem->init_req(_false_path, commit_memory);
3446
3447 lease_compare_io->init_req(_true_path, i_o());
3448 lease_compare_io->init_req(_false_path, input_io_state);
3449
3450 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3451 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3452
3453 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3454 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3455 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3456 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3457
3458 // Update control and phi nodes.
3459 result_rgn->init_req(_true_path, is_notified);
3460 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3461
3462 result_mem->init_req(_true_path, notified_reset_memory);
3463 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3464
3465 result_io->init_req(_true_path, input_io_state);
3466 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3467
3468 result_value->init_req(_true_path, current_pos);
3469 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3470
3471 // Set output state.
3472 set_control(_gvn.transform(result_rgn));
3473 set_all_memory(_gvn.transform(result_mem));
3474 set_i_o(_gvn.transform(result_io));
3475 set_result(result_rgn, result_value);
3476 return true;
3477 }
3478
3479 /*
3480 * The intrinsic is a model of this pseudo-code:
3481 *
3482 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3483 * jobject h_event_writer = tl->java_event_writer();
3484 * if (h_event_writer == nullptr) {
3485 * return nullptr;
3486 * }
3487 * oop threadObj = Thread::threadObj();
3488 * oop vthread = java_lang_Thread::vthread(threadObj);
3489 * traceid tid;
3490 * bool pinVirtualThread;
3491 * bool excluded;
3492 * if (vthread != threadObj) { // i.e. current thread is virtual
3493 * tid = java_lang_Thread::tid(vthread);
3494 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3495 * pinVirtualThread = VMContinuations;
3496 * excluded = vthread_epoch_raw & excluded_mask;
3497 * if (!excluded) {
3498 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3499 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3500 * if (vthread_epoch != current_epoch) {
3501 * write_checkpoint();
3502 * }
3503 * }
3504 * } else {
3505 * tid = java_lang_Thread::tid(threadObj);
3506 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3507 * pinVirtualThread = false;
3508 * excluded = thread_epoch_raw & excluded_mask;
3509 * }
3510 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3511 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3512 * if (tid_in_event_writer != tid) {
3513 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3514 * setField(event_writer, "excluded", excluded);
3515 * setField(event_writer, "threadID", tid);
3516 * }
3517 * return event_writer
3518 */
3519 bool LibraryCallKit::inline_native_getEventWriter() {
3520 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3521
3522 // Save input memory and i_o state.
3523 Node* input_memory_state = reset_memory();
3524 set_all_memory(input_memory_state);
3525 Node* input_io_state = i_o();
3526
3527 // The most significant bit of the u2 is used to denote thread exclusion
3528 Node* excluded_shift = _gvn.intcon(15);
3529 Node* excluded_mask = _gvn.intcon(1 << 15);
3530 // The epoch generation is the range [1-32767]
3531 Node* epoch_mask = _gvn.intcon(32767);
3532
3533 // TLS
3534 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3535
3536 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3537 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3538
3539 // Load the eventwriter jobject handle.
3540 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3541
3542 // Null check the jobject handle.
3543 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3544 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3545 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3546
3547 // False path, jobj is null.
3548 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3549
3550 // True path, jobj is not null.
3551 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3552
3553 set_control(jobj_is_not_null);
3554
3555 // Load the threadObj for the CarrierThread.
3556 Node* threadObj = generate_current_thread(tls_ptr);
3557
3558 // Load the vthread.
3559 Node* vthread = generate_virtual_thread(tls_ptr);
3560
3561 // If vthread != threadObj, this is a virtual thread.
3562 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3563 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3564 IfNode* iff_vthread_not_equal_threadObj =
3565 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3566
3567 // False branch, fallback to threadObj.
3568 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3569 set_control(vthread_equal_threadObj);
3570
3571 // Load the tid field from the vthread object.
3572 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3573
3574 // Load the raw epoch value from the threadObj.
3575 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3576 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3577 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3578 TypeInt::CHAR, T_CHAR,
3579 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3580
3581 // Mask off the excluded information from the epoch.
3582 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3583
3584 // True branch, this is a virtual thread.
3585 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3586 set_control(vthread_not_equal_threadObj);
3587
3588 // Load the tid field from the vthread object.
3589 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3590
3591 // Continuation support determines if a virtual thread should be pinned.
3592 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3593 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3594
3595 // Load the raw epoch value from the vthread.
3596 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3597 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3598 TypeInt::CHAR, T_CHAR,
3599 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3600
3601 // Mask off the excluded information from the epoch.
3602 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3603
3604 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3605 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3606 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3607 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3608
3609 // False branch, vthread is excluded, no need to write epoch info.
3610 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3611
3612 // True branch, vthread is included, update epoch info.
3613 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3614 set_control(included);
3615
3616 // Get epoch value.
3617 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3618
3619 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3620 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3621 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3622
3623 // Compare the epoch in the vthread to the current epoch generation.
3624 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3625 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3626 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3627
3628 // False path, epoch is equal, checkpoint information is valid.
3629 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3630
3631 // True path, epoch is not equal, write a checkpoint for the vthread.
3632 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3633
3634 set_control(epoch_is_not_equal);
3635
3636 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3637 // The call also updates the native thread local thread id and the vthread with the current epoch.
3638 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3639 OptoRuntime::jfr_write_checkpoint_Type(),
3640 SharedRuntime::jfr_write_checkpoint(),
3641 "write_checkpoint", TypePtr::BOTTOM);
3642 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3643
3644 // vthread epoch != current epoch
3645 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3646 record_for_igvn(epoch_compare_rgn);
3647 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3648 record_for_igvn(epoch_compare_mem);
3649 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3650 record_for_igvn(epoch_compare_io);
3651
3652 // Update control and phi nodes.
3653 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3654 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3655 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3656 epoch_compare_mem->init_req(_false_path, input_memory_state);
3657 epoch_compare_io->init_req(_true_path, i_o());
3658 epoch_compare_io->init_req(_false_path, input_io_state);
3659
3660 // excluded != true
3661 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3662 record_for_igvn(exclude_compare_rgn);
3663 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3664 record_for_igvn(exclude_compare_mem);
3665 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3666 record_for_igvn(exclude_compare_io);
3667
3668 // Update control and phi nodes.
3669 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3670 exclude_compare_rgn->init_req(_false_path, excluded);
3671 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3672 exclude_compare_mem->init_req(_false_path, input_memory_state);
3673 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3674 exclude_compare_io->init_req(_false_path, input_io_state);
3675
3676 // vthread != threadObj
3677 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3678 record_for_igvn(vthread_compare_rgn);
3679 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3680 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3681 record_for_igvn(vthread_compare_io);
3682 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3683 record_for_igvn(tid);
3684 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3685 record_for_igvn(exclusion);
3686 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3687 record_for_igvn(pinVirtualThread);
3688
3689 // Update control and phi nodes.
3690 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3691 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3692 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3693 vthread_compare_mem->init_req(_false_path, input_memory_state);
3694 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3695 vthread_compare_io->init_req(_false_path, input_io_state);
3696 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3697 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3698 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3699 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3700 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3701 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3702
3703 // Update branch state.
3704 set_control(_gvn.transform(vthread_compare_rgn));
3705 set_all_memory(_gvn.transform(vthread_compare_mem));
3706 set_i_o(_gvn.transform(vthread_compare_io));
3707
3708 // Load the event writer oop by dereferencing the jobject handle.
3709 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3710 assert(klass_EventWriter->is_loaded(), "invariant");
3711 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3712 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3713 const TypeOopPtr* const xtype = aklass->as_instance_type();
3714 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3715 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3716
3717 // Load the current thread id from the event writer object.
3718 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3719 // Get the field offset to, conditionally, store an updated tid value later.
3720 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3721 // Get the field offset to, conditionally, store an updated exclusion value later.
3722 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3723 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3724 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3725
3726 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3727 record_for_igvn(event_writer_tid_compare_rgn);
3728 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3729 record_for_igvn(event_writer_tid_compare_mem);
3730 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3731 record_for_igvn(event_writer_tid_compare_io);
3732
3733 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3734 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3735 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3736 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3737
3738 // False path, tids are the same.
3739 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3740
3741 // True path, tid is not equal, need to update the tid in the event writer.
3742 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3743 record_for_igvn(tid_is_not_equal);
3744
3745 // Store the pin state to the event writer.
3746 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3747
3748 // Store the exclusion state to the event writer.
3749 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3750 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3751
3752 // Store the tid to the event writer.
3753 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3754
3755 // Update control and phi nodes.
3756 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3757 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3758 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3759 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3760 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3761 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3762
3763 // Result of top level CFG, Memory, IO and Value.
3764 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3765 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3766 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3767 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3768
3769 // Result control.
3770 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3771 result_rgn->init_req(_false_path, jobj_is_null);
3772
3773 // Result memory.
3774 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3775 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3776
3777 // Result IO.
3778 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3779 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3780
3781 // Result value.
3782 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3783 result_value->init_req(_false_path, null()); // return null
3784
3785 // Set output state.
3786 set_control(_gvn.transform(result_rgn));
3787 set_all_memory(_gvn.transform(result_mem));
3788 set_i_o(_gvn.transform(result_io));
3789 set_result(result_rgn, result_value);
3790 return true;
3791 }
3792
3793 /*
3794 * The intrinsic is a model of this pseudo-code:
3795 *
3796 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3797 * if (carrierThread != thread) { // is virtual thread
3798 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3799 * bool excluded = vthread_epoch_raw & excluded_mask;
3800 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3801 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3802 * if (!excluded) {
3803 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3804 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3805 * }
3806 * Atomic::release_store(&tl->_vthread, true);
3807 * return;
3808 * }
3809 * Atomic::release_store(&tl->_vthread, false);
3810 */
3811 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3812 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3813
3814 Node* input_memory_state = reset_memory();
3815 set_all_memory(input_memory_state);
3816
3817 // The most significant bit of the u2 is used to denote thread exclusion
3818 Node* excluded_mask = _gvn.intcon(1 << 15);
3819 // The epoch generation is the range [1-32767]
3820 Node* epoch_mask = _gvn.intcon(32767);
3821
3822 Node* const carrierThread = generate_current_thread(jt);
3823 // If thread != carrierThread, this is a virtual thread.
3824 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3825 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3826 IfNode* iff_thread_not_equal_carrierThread =
3827 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3828
3829 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3830
3831 // False branch, is carrierThread.
3832 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3833 // Store release
3834 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3835
3836 set_all_memory(input_memory_state);
3837
3838 // True branch, is virtual thread.
3839 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3840 set_control(thread_not_equal_carrierThread);
3841
3842 // Load the raw epoch value from the vthread.
3843 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3844 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
3845 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3846
3847 // Mask off the excluded information from the epoch.
3848 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3849
3850 // Load the tid field from the thread.
3851 Node* tid = load_field_from_object(thread, "tid", "J");
3852
3853 // Store the vthread tid to the jfr thread local.
3854 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3855 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
3856
3857 // Branch is_excluded to conditionalize updating the epoch .
3858 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3859 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3860 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3861
3862 // True branch, vthread is excluded, no need to write epoch info.
3863 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3864 set_control(excluded);
3865 Node* vthread_is_excluded = _gvn.intcon(1);
3866
3867 // False branch, vthread is included, update epoch info.
3868 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3869 set_control(included);
3870 Node* vthread_is_included = _gvn.intcon(0);
3871
3872 // Get epoch value.
3873 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3874
3875 // Store the vthread epoch to the jfr thread local.
3876 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3877 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
3878
3879 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3880 record_for_igvn(excluded_rgn);
3881 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3882 record_for_igvn(excluded_mem);
3883 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3884 record_for_igvn(exclusion);
3885
3886 // Merge the excluded control and memory.
3887 excluded_rgn->init_req(_true_path, excluded);
3888 excluded_rgn->init_req(_false_path, included);
3889 excluded_mem->init_req(_true_path, tid_memory);
3890 excluded_mem->init_req(_false_path, included_memory);
3891 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3892 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3893
3894 // Set intermediate state.
3895 set_control(_gvn.transform(excluded_rgn));
3896 set_all_memory(excluded_mem);
3897
3898 // Store the vthread exclusion state to the jfr thread local.
3899 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3900 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
3901
3902 // Store release
3903 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
3904
3905 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3906 record_for_igvn(thread_compare_rgn);
3907 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3908 record_for_igvn(thread_compare_mem);
3909 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3910 record_for_igvn(vthread);
3911
3912 // Merge the thread_compare control and memory.
3913 thread_compare_rgn->init_req(_true_path, control());
3914 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3915 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3916 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3917
3918 // Set output state.
3919 set_control(_gvn.transform(thread_compare_rgn));
3920 set_all_memory(_gvn.transform(thread_compare_mem));
3921 }
3922
3923 #endif // JFR_HAVE_INTRINSICS
3924
3925 //------------------------inline_native_currentCarrierThread------------------
3926 bool LibraryCallKit::inline_native_currentCarrierThread() {
3927 Node* junk = nullptr;
3928 set_result(generate_current_thread(junk));
3929 return true;
3930 }
3931
3932 //------------------------inline_native_currentThread------------------
3933 bool LibraryCallKit::inline_native_currentThread() {
3934 Node* junk = nullptr;
3935 set_result(generate_virtual_thread(junk));
3936 return true;
3937 }
3938
3939 //------------------------inline_native_setVthread------------------
3940 bool LibraryCallKit::inline_native_setCurrentThread() {
3941 assert(C->method()->changes_current_thread(),
3942 "method changes current Thread but is not annotated ChangesCurrentThread");
3943 Node* arr = argument(1);
3944 Node* thread = _gvn.transform(new ThreadLocalNode());
3945 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3946 Node* thread_obj_handle
3947 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3948 thread_obj_handle = _gvn.transform(thread_obj_handle);
3949 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3950 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3951
3952 // Change the _monitor_owner_id of the JavaThread
3953 Node* tid = load_field_from_object(arr, "tid", "J");
3954 Node* monitor_owner_id_offset = basic_plus_adr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
3955 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
3956
3957 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3958 return true;
3959 }
3960
3961 const Type* LibraryCallKit::scopedValueCache_type() {
3962 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3963 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3964 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS, /* stable= */ false, /* flat= */ false, /* not_flat= */ true, /* not_null_free= */ true);
3965
3966 // Because we create the scopedValue cache lazily we have to make the
3967 // type of the result BotPTR.
3968 bool xk = etype->klass_is_exact();
3969 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, TypeAryPtr::Offset(0));
3970 return objects_type;
3971 }
3972
3973 Node* LibraryCallKit::scopedValueCache_helper() {
3974 Node* thread = _gvn.transform(new ThreadLocalNode());
3975 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3976 // We cannot use immutable_memory() because we might flip onto a
3977 // different carrier thread, at which point we'll need to use that
3978 // carrier thread's cache.
3979 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3980 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3981 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3982 }
3983
3984 //------------------------inline_native_scopedValueCache------------------
3985 bool LibraryCallKit::inline_native_scopedValueCache() {
3986 Node* cache_obj_handle = scopedValueCache_helper();
3987 const Type* objects_type = scopedValueCache_type();
3988 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3989
3990 return true;
3991 }
3992
3993 //------------------------inline_native_setScopedValueCache------------------
3994 bool LibraryCallKit::inline_native_setScopedValueCache() {
3995 Node* arr = argument(0);
3996 Node* cache_obj_handle = scopedValueCache_helper();
3997 const Type* objects_type = scopedValueCache_type();
3998
3999 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
4000 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
4001
4002 return true;
4003 }
4004
4005 //------------------------inline_native_Continuation_pin and unpin-----------
4006
4007 // Shared implementation routine for both pin and unpin.
4008 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
4009 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
4010
4011 // Save input memory.
4012 Node* input_memory_state = reset_memory();
4013 set_all_memory(input_memory_state);
4014
4015 // TLS
4016 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
4017 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
4018 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
4019
4020 // Null check the last continuation object.
4021 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
4022 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
4023 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
4024
4025 // False path, last continuation is null.
4026 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
4027
4028 // True path, last continuation is not null.
4029 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
4030
4031 set_control(continuation_is_not_null);
4032
4033 // Load the pin count from the last continuation.
4034 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
4035 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
4036
4037 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
4038 Node* pin_count_rhs;
4039 if (unpin) {
4040 pin_count_rhs = _gvn.intcon(0);
4041 } else {
4042 pin_count_rhs = _gvn.intcon(UINT32_MAX);
4043 }
4044 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
4045 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
4046 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
4047
4048 // True branch, pin count over/underflow.
4049 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
4050 {
4051 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
4052 // which will throw IllegalStateException for pin count over/underflow.
4053 // No memory changed so far - we can use memory create by reset_memory()
4054 // at the beginning of this intrinsic. No need to call reset_memory() again.
4055 PreserveJVMState pjvms(this);
4056 set_control(pin_count_over_underflow);
4057 uncommon_trap(Deoptimization::Reason_intrinsic,
4058 Deoptimization::Action_none);
4059 assert(stopped(), "invariant");
4060 }
4061
4062 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
4063 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
4064 set_control(valid_pin_count);
4065
4066 Node* next_pin_count;
4067 if (unpin) {
4068 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
4069 } else {
4070 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
4071 }
4072
4073 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
4074
4075 // Result of top level CFG and Memory.
4076 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
4077 record_for_igvn(result_rgn);
4078 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
4079 record_for_igvn(result_mem);
4080
4081 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
4082 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
4083 result_mem->init_req(_true_path, _gvn.transform(reset_memory()));
4084 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
4085
4086 // Set output state.
4087 set_control(_gvn.transform(result_rgn));
4088 set_all_memory(_gvn.transform(result_mem));
4089
4090 return true;
4091 }
4092
4093 //-----------------------load_klass_from_mirror_common-------------------------
4094 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
4095 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
4096 // and branch to the given path on the region.
4097 // If never_see_null, take an uncommon trap on null, so we can optimistically
4098 // compile for the non-null case.
4099 // If the region is null, force never_see_null = true.
4100 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
4101 bool never_see_null,
4102 RegionNode* region,
4103 int null_path,
4104 int offset) {
4105 if (region == nullptr) never_see_null = true;
4106 Node* p = basic_plus_adr(mirror, offset);
4107 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4108 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
4109 Node* null_ctl = top();
4110 kls = null_check_oop(kls, &null_ctl, never_see_null);
4111 if (region != nullptr) {
4112 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
4113 region->init_req(null_path, null_ctl);
4114 } else {
4115 assert(null_ctl == top(), "no loose ends");
4116 }
4117 return kls;
4118 }
4119
4120 //--------------------(inline_native_Class_query helpers)---------------------
4121 // Use this for JVM_ACC_INTERFACE.
4122 // Fall through if (mods & mask) == bits, take the guard otherwise.
4123 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
4124 ByteSize offset, const Type* type, BasicType bt) {
4125 // Branch around if the given klass has the given modifier bit set.
4126 // Like generate_guard, adds a new path onto the region.
4127 Node* modp = basic_plus_adr(kls, in_bytes(offset));
4128 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
4129 Node* mask = intcon(modifier_mask);
4130 Node* bits = intcon(modifier_bits);
4131 Node* mbit = _gvn.transform(new AndINode(mods, mask));
4132 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
4133 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
4134 return generate_fair_guard(bol, region);
4135 }
4136
4137 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
4138 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
4139 Klass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
4140 }
4141
4142 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
4143 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
4144 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
4145 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
4146 }
4147
4148 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
4149 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
4150 }
4151
4152 //-------------------------inline_native_Class_query-------------------
4153 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
4154 const Type* return_type = TypeInt::BOOL;
4155 Node* prim_return_value = top(); // what happens if it's a primitive class?
4156 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4157 bool expect_prim = false; // most of these guys expect to work on refs
4158
4159 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
4160
4161 Node* mirror = argument(0);
4162 Node* obj = top();
4163
4164 switch (id) {
4165 case vmIntrinsics::_isInstance:
4166 // nothing is an instance of a primitive type
4167 prim_return_value = intcon(0);
4168 obj = argument(1);
4169 break;
4170 case vmIntrinsics::_isHidden:
4171 prim_return_value = intcon(0);
4172 break;
4173 case vmIntrinsics::_getSuperclass:
4174 prim_return_value = null();
4175 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
4176 break;
4177 case vmIntrinsics::_getClassAccessFlags:
4178 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
4179 return_type = TypeInt::CHAR;
4180 break;
4181 default:
4182 fatal_unexpected_iid(id);
4183 break;
4184 }
4185
4186 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4187 if (mirror_con == nullptr) return false; // cannot happen?
4188
4189 #ifndef PRODUCT
4190 if (C->print_intrinsics() || C->print_inlining()) {
4191 ciType* k = mirror_con->java_mirror_type();
4192 if (k) {
4193 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4194 k->print_name();
4195 tty->cr();
4196 }
4197 }
4198 #endif
4199
4200 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4201 RegionNode* region = new RegionNode(PATH_LIMIT);
4202 record_for_igvn(region);
4203 PhiNode* phi = new PhiNode(region, return_type);
4204
4205 // The mirror will never be null of Reflection.getClassAccessFlags, however
4206 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4207 // if it is. See bug 4774291.
4208
4209 // For Reflection.getClassAccessFlags(), the null check occurs in
4210 // the wrong place; see inline_unsafe_access(), above, for a similar
4211 // situation.
4212 mirror = null_check(mirror);
4213 // If mirror or obj is dead, only null-path is taken.
4214 if (stopped()) return true;
4215
4216 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4217
4218 // Now load the mirror's klass metaobject, and null-check it.
4219 // Side-effects region with the control path if the klass is null.
4220 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4221 // If kls is null, we have a primitive mirror.
4222 phi->init_req(_prim_path, prim_return_value);
4223 if (stopped()) { set_result(region, phi); return true; }
4224 bool safe_for_replace = (region->in(_prim_path) == top());
4225
4226 Node* p; // handy temp
4227 Node* null_ctl;
4228
4229 // Now that we have the non-null klass, we can perform the real query.
4230 // For constant classes, the query will constant-fold in LoadNode::Value.
4231 Node* query_value = top();
4232 switch (id) {
4233 case vmIntrinsics::_isInstance:
4234 // nothing is an instance of a primitive type
4235 query_value = gen_instanceof(obj, kls, safe_for_replace);
4236 break;
4237
4238 case vmIntrinsics::_isHidden:
4239 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4240 if (generate_hidden_class_guard(kls, region) != nullptr)
4241 // A guard was added. If the guard is taken, it was an hidden class.
4242 phi->add_req(intcon(1));
4243 // If we fall through, it's a plain class.
4244 query_value = intcon(0);
4245 break;
4246
4247
4248 case vmIntrinsics::_getSuperclass:
4249 // The rules here are somewhat unfortunate, but we can still do better
4250 // with random logic than with a JNI call.
4251 // Interfaces store null or Object as _super, but must report null.
4252 // Arrays store an intermediate super as _super, but must report Object.
4253 // Other types can report the actual _super.
4254 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4255 if (generate_interface_guard(kls, region) != nullptr)
4256 // A guard was added. If the guard is taken, it was an interface.
4257 phi->add_req(null());
4258 if (generate_array_guard(kls, region) != nullptr)
4259 // A guard was added. If the guard is taken, it was an array.
4260 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4261 // If we fall through, it's a plain class. Get its _super.
4262 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4263 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4264 null_ctl = top();
4265 kls = null_check_oop(kls, &null_ctl);
4266 if (null_ctl != top()) {
4267 // If the guard is taken, Object.superClass is null (both klass and mirror).
4268 region->add_req(null_ctl);
4269 phi ->add_req(null());
4270 }
4271 if (!stopped()) {
4272 query_value = load_mirror_from_klass(kls);
4273 }
4274 break;
4275
4276 case vmIntrinsics::_getClassAccessFlags:
4277 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
4278 query_value = make_load(nullptr, p, TypeInt::CHAR, T_CHAR, MemNode::unordered);
4279 break;
4280
4281 default:
4282 fatal_unexpected_iid(id);
4283 break;
4284 }
4285
4286 // Fall-through is the normal case of a query to a real class.
4287 phi->init_req(1, query_value);
4288 region->init_req(1, control());
4289
4290 C->set_has_split_ifs(true); // Has chance for split-if optimization
4291 set_result(region, phi);
4292 return true;
4293 }
4294
4295
4296 //-------------------------inline_Class_cast-------------------
4297 bool LibraryCallKit::inline_Class_cast() {
4298 Node* mirror = argument(0); // Class
4299 Node* obj = argument(1);
4300 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4301 if (mirror_con == nullptr) {
4302 return false; // dead path (mirror->is_top()).
4303 }
4304 if (obj == nullptr || obj->is_top()) {
4305 return false; // dead path
4306 }
4307 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4308
4309 // First, see if Class.cast() can be folded statically.
4310 // java_mirror_type() returns non-null for compile-time Class constants.
4311 bool is_null_free_array = false;
4312 ciType* tm = mirror_con->java_mirror_type(&is_null_free_array);
4313 if (tm != nullptr && tm->is_klass() &&
4314 tp != nullptr) {
4315 if (!tp->is_loaded()) {
4316 // Don't use intrinsic when class is not loaded.
4317 return false;
4318 } else {
4319 const TypeKlassPtr* tklass = TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces);
4320 if (is_null_free_array) {
4321 tklass = tklass->is_aryklassptr()->cast_to_null_free();
4322 }
4323 int static_res = C->static_subtype_check(tklass, tp->as_klass_type());
4324 if (static_res == Compile::SSC_always_true) {
4325 // isInstance() is true - fold the code.
4326 set_result(obj);
4327 return true;
4328 } else if (static_res == Compile::SSC_always_false) {
4329 // Don't use intrinsic, have to throw ClassCastException.
4330 // If the reference is null, the non-intrinsic bytecode will
4331 // be optimized appropriately.
4332 return false;
4333 }
4334 }
4335 }
4336
4337 // Bailout intrinsic and do normal inlining if exception path is frequent.
4338 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4339 return false;
4340 }
4341
4342 // Generate dynamic checks.
4343 // Class.cast() is java implementation of _checkcast bytecode.
4344 // Do checkcast (Parse::do_checkcast()) optimizations here.
4345
4346 mirror = null_check(mirror);
4347 // If mirror is dead, only null-path is taken.
4348 if (stopped()) {
4349 return true;
4350 }
4351
4352 // Not-subtype or the mirror's klass ptr is nullptr (in case it is a primitive).
4353 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
4354 RegionNode* region = new RegionNode(PATH_LIMIT);
4355 record_for_igvn(region);
4356
4357 // Now load the mirror's klass metaobject, and null-check it.
4358 // If kls is null, we have a primitive mirror and
4359 // nothing is an instance of a primitive type.
4360 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4361
4362 Node* res = top();
4363 Node* io = i_o();
4364 Node* mem = merged_memory();
4365 if (!stopped()) {
4366
4367 Node* bad_type_ctrl = top();
4368 // Do checkcast optimizations.
4369 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4370 region->init_req(_bad_type_path, bad_type_ctrl);
4371 }
4372 if (region->in(_prim_path) != top() ||
4373 region->in(_bad_type_path) != top() ||
4374 region->in(_npe_path) != top()) {
4375 // Let Interpreter throw ClassCastException.
4376 PreserveJVMState pjvms(this);
4377 set_control(_gvn.transform(region));
4378 // Set IO and memory because gen_checkcast may override them when buffering inline types
4379 set_i_o(io);
4380 set_all_memory(mem);
4381 uncommon_trap(Deoptimization::Reason_intrinsic,
4382 Deoptimization::Action_maybe_recompile);
4383 }
4384 if (!stopped()) {
4385 set_result(res);
4386 }
4387 return true;
4388 }
4389
4390
4391 //--------------------------inline_native_subtype_check------------------------
4392 // This intrinsic takes the JNI calls out of the heart of
4393 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4394 bool LibraryCallKit::inline_native_subtype_check() {
4395 // Pull both arguments off the stack.
4396 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4397 args[0] = argument(0);
4398 args[1] = argument(1);
4399 Node* klasses[2]; // corresponding Klasses: superk, subk
4400 klasses[0] = klasses[1] = top();
4401
4402 enum {
4403 // A full decision tree on {superc is prim, subc is prim}:
4404 _prim_0_path = 1, // {P,N} => false
4405 // {P,P} & superc!=subc => false
4406 _prim_same_path, // {P,P} & superc==subc => true
4407 _prim_1_path, // {N,P} => false
4408 _ref_subtype_path, // {N,N} & subtype check wins => true
4409 _both_ref_path, // {N,N} & subtype check loses => false
4410 PATH_LIMIT
4411 };
4412
4413 RegionNode* region = new RegionNode(PATH_LIMIT);
4414 RegionNode* prim_region = new RegionNode(2);
4415 Node* phi = new PhiNode(region, TypeInt::BOOL);
4416 record_for_igvn(region);
4417 record_for_igvn(prim_region);
4418
4419 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4420 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4421 int class_klass_offset = java_lang_Class::klass_offset();
4422
4423 // First null-check both mirrors and load each mirror's klass metaobject.
4424 int which_arg;
4425 for (which_arg = 0; which_arg <= 1; which_arg++) {
4426 Node* arg = args[which_arg];
4427 arg = null_check(arg);
4428 if (stopped()) break;
4429 args[which_arg] = arg;
4430
4431 Node* p = basic_plus_adr(arg, class_klass_offset);
4432 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4433 klasses[which_arg] = _gvn.transform(kls);
4434 }
4435
4436 // Having loaded both klasses, test each for null.
4437 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4438 for (which_arg = 0; which_arg <= 1; which_arg++) {
4439 Node* kls = klasses[which_arg];
4440 Node* null_ctl = top();
4441 kls = null_check_oop(kls, &null_ctl, never_see_null);
4442 if (which_arg == 0) {
4443 prim_region->init_req(1, null_ctl);
4444 } else {
4445 region->init_req(_prim_1_path, null_ctl);
4446 }
4447 if (stopped()) break;
4448 klasses[which_arg] = kls;
4449 }
4450
4451 if (!stopped()) {
4452 // now we have two reference types, in klasses[0..1]
4453 Node* subk = klasses[1]; // the argument to isAssignableFrom
4454 Node* superk = klasses[0]; // the receiver
4455 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4456 region->set_req(_ref_subtype_path, control());
4457 }
4458
4459 // If both operands are primitive (both klasses null), then
4460 // we must return true when they are identical primitives.
4461 // It is convenient to test this after the first null klass check.
4462 // This path is also used if superc is a value mirror.
4463 set_control(_gvn.transform(prim_region));
4464 if (!stopped()) {
4465 // Since superc is primitive, make a guard for the superc==subc case.
4466 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4467 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4468 generate_fair_guard(bol_eq, region);
4469 if (region->req() == PATH_LIMIT+1) {
4470 // A guard was added. If the added guard is taken, superc==subc.
4471 region->swap_edges(PATH_LIMIT, _prim_same_path);
4472 region->del_req(PATH_LIMIT);
4473 }
4474 region->set_req(_prim_0_path, control()); // Not equal after all.
4475 }
4476
4477 // these are the only paths that produce 'true':
4478 phi->set_req(_prim_same_path, intcon(1));
4479 phi->set_req(_ref_subtype_path, intcon(1));
4480
4481 // pull together the cases:
4482 assert(region->req() == PATH_LIMIT, "sane region");
4483 for (uint i = 1; i < region->req(); i++) {
4484 Node* ctl = region->in(i);
4485 if (ctl == nullptr || ctl == top()) {
4486 region->set_req(i, top());
4487 phi ->set_req(i, top());
4488 } else if (phi->in(i) == nullptr) {
4489 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4490 }
4491 }
4492
4493 set_control(_gvn.transform(region));
4494 set_result(_gvn.transform(phi));
4495 return true;
4496 }
4497
4498 //---------------------generate_array_guard_common------------------------
4499 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind, Node** obj) {
4500
4501 if (stopped()) {
4502 return nullptr;
4503 }
4504
4505 // Like generate_guard, adds a new path onto the region.
4506 jint layout_con = 0;
4507 Node* layout_val = get_layout_helper(kls, layout_con);
4508 if (layout_val == nullptr) {
4509 bool query = 0;
4510 switch(kind) {
4511 case ObjectArray: query = Klass::layout_helper_is_objArray(layout_con); break;
4512 case NonObjectArray: query = !Klass::layout_helper_is_objArray(layout_con); break;
4513 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
4514 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
4515 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
4516 default:
4517 ShouldNotReachHere();
4518 }
4519 if (!query) {
4520 return nullptr; // never a branch
4521 } else { // always a branch
4522 Node* always_branch = control();
4523 if (region != nullptr)
4524 region->add_req(always_branch);
4525 set_control(top());
4526 return always_branch;
4527 }
4528 }
4529 unsigned int value = 0;
4530 BoolTest::mask btest = BoolTest::illegal;
4531 switch(kind) {
4532 case ObjectArray:
4533 case NonObjectArray: {
4534 value = Klass::_lh_array_tag_obj_value;
4535 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4536 btest = (kind == ObjectArray) ? BoolTest::eq : BoolTest::ne;
4537 break;
4538 }
4539 case TypeArray: {
4540 value = Klass::_lh_array_tag_type_value;
4541 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4542 btest = BoolTest::eq;
4543 break;
4544 }
4545 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
4546 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
4547 default:
4548 ShouldNotReachHere();
4549 }
4550 // Now test the correct condition.
4551 jint nval = (jint)value;
4552 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4553 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4554 Node* ctrl = generate_fair_guard(bol, region);
4555 Node* is_array_ctrl = kind == NonArray ? control() : ctrl;
4556 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4557 // Keep track of the fact that 'obj' is an array to prevent
4558 // array specific accesses from floating above the guard.
4559 Node* cast = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4560 // Check for top because in rare cases, the type system can determine that
4561 // the object can't be an array but the layout helper check is not folded.
4562 if (!cast->is_top()) {
4563 *obj = cast;
4564 }
4565 }
4566 return ctrl;
4567 }
4568
4569 // public static native Object[] newNullRestrictedAtomicArray(Class<?> componentType, int length, Object initVal);
4570 // public static native Object[] newNullRestrictedNonAtomicArray(Class<?> componentType, int length, Object initVal);
4571 // public static native Object[] newNullableAtomicArray(Class<?> componentType, int length);
4572 bool LibraryCallKit::inline_newArray(bool null_free, bool atomic) {
4573 assert(null_free || atomic, "nullable implies atomic");
4574 Node* componentType = argument(0);
4575 Node* length = argument(1);
4576 Node* init_val = null_free ? argument(2) : nullptr;
4577
4578 const TypeInstPtr* tp = _gvn.type(componentType)->isa_instptr();
4579 if (tp != nullptr) {
4580 ciInstanceKlass* ik = tp->instance_klass();
4581 if (ik == C->env()->Class_klass()) {
4582 ciType* t = tp->java_mirror_type();
4583 if (t != nullptr && t->is_inlinetype()) {
4584 ciInlineKlass* vk = t->as_inline_klass();
4585 bool flat = vk->maybe_flat_in_array();
4586 if (flat && atomic) {
4587 // Only flat if we have a corresponding atomic layout
4588 flat = null_free ? vk->has_atomic_layout() : vk->has_nullable_atomic_layout();
4589 }
4590 // TODO 8350865 refactor
4591 if (flat && !atomic) {
4592 flat = vk->has_non_atomic_layout();
4593 }
4594
4595 // TOOD 8350865 ZGC needs card marks on initializing oop stores
4596 if (UseZGC && null_free && !flat) {
4597 return false;
4598 }
4599
4600 ciArrayKlass* array_klass = ciArrayKlass::make(t, flat, null_free, atomic);
4601 if (array_klass->is_loaded() && array_klass->element_klass()->as_inline_klass()->is_initialized()) {
4602 const TypeAryKlassPtr* array_klass_type = TypeKlassPtr::make(array_klass, Type::trust_interfaces)->is_aryklassptr();
4603 if (null_free) {
4604 if (init_val->is_InlineType()) {
4605 if (array_klass_type->is_flat() && init_val->as_InlineType()->is_all_zero(&gvn(), /* flat */ true)) {
4606 // Zeroing is enough because the init value is the all-zero value
4607 init_val = nullptr;
4608 } else {
4609 init_val = init_val->as_InlineType()->buffer(this);
4610 }
4611 }
4612 // TODO 8350865 Should we add a check of the init_val type (maybe in debug only + halt)?
4613 }
4614 Node* obj = new_array(makecon(array_klass_type), length, 0, nullptr, false, init_val);
4615 const TypeAryPtr* arytype = gvn().type(obj)->is_aryptr();
4616 assert(arytype->is_null_free() == null_free, "inconsistency");
4617 assert(arytype->is_not_null_free() == !null_free, "inconsistency");
4618 assert(arytype->is_flat() == flat, "inconsistency");
4619 assert(arytype->is_aryptr()->is_not_flat() == !flat, "inconsistency");
4620 set_result(obj);
4621 return true;
4622 }
4623 }
4624 }
4625 }
4626 return false;
4627 }
4628
4629 //-----------------------inline_native_newArray--------------------------
4630 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
4631 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4632 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4633 Node* mirror;
4634 Node* count_val;
4635 if (uninitialized) {
4636 null_check_receiver();
4637 mirror = argument(1);
4638 count_val = argument(2);
4639 } else {
4640 mirror = argument(0);
4641 count_val = argument(1);
4642 }
4643
4644 mirror = null_check(mirror);
4645 // If mirror or obj is dead, only null-path is taken.
4646 if (stopped()) return true;
4647
4648 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4649 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4650 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4651 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4652 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4653
4654 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4655 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4656 result_reg, _slow_path);
4657 Node* normal_ctl = control();
4658 Node* no_array_ctl = result_reg->in(_slow_path);
4659
4660 // Generate code for the slow case. We make a call to newArray().
4661 set_control(no_array_ctl);
4662 if (!stopped()) {
4663 // Either the input type is void.class, or else the
4664 // array klass has not yet been cached. Either the
4665 // ensuing call will throw an exception, or else it
4666 // will cache the array klass for next time.
4667 PreserveJVMState pjvms(this);
4668 CallJavaNode* slow_call = nullptr;
4669 if (uninitialized) {
4670 // Generate optimized virtual call (holder class 'Unsafe' is final)
4671 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4672 } else {
4673 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4674 }
4675 Node* slow_result = set_results_for_java_call(slow_call);
4676 // this->control() comes from set_results_for_java_call
4677 result_reg->set_req(_slow_path, control());
4678 result_val->set_req(_slow_path, slow_result);
4679 result_io ->set_req(_slow_path, i_o());
4680 result_mem->set_req(_slow_path, reset_memory());
4681 }
4682
4683 set_control(normal_ctl);
4684 if (!stopped()) {
4685 // Normal case: The array type has been cached in the java.lang.Class.
4686 // The following call works fine even if the array type is polymorphic.
4687 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4688 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4689 result_reg->init_req(_normal_path, control());
4690 result_val->init_req(_normal_path, obj);
4691 result_io ->init_req(_normal_path, i_o());
4692 result_mem->init_req(_normal_path, reset_memory());
4693
4694 if (uninitialized) {
4695 // Mark the allocation so that zeroing is skipped
4696 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4697 alloc->maybe_set_complete(&_gvn);
4698 }
4699 }
4700
4701 // Return the combined state.
4702 set_i_o( _gvn.transform(result_io) );
4703 set_all_memory( _gvn.transform(result_mem));
4704
4705 C->set_has_split_ifs(true); // Has chance for split-if optimization
4706 set_result(result_reg, result_val);
4707 return true;
4708 }
4709
4710 //----------------------inline_native_getLength--------------------------
4711 // public static native int java.lang.reflect.Array.getLength(Object array);
4712 bool LibraryCallKit::inline_native_getLength() {
4713 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4714
4715 Node* array = null_check(argument(0));
4716 // If array is dead, only null-path is taken.
4717 if (stopped()) return true;
4718
4719 // Deoptimize if it is a non-array.
4720 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
4721
4722 if (non_array != nullptr) {
4723 PreserveJVMState pjvms(this);
4724 set_control(non_array);
4725 uncommon_trap(Deoptimization::Reason_intrinsic,
4726 Deoptimization::Action_maybe_recompile);
4727 }
4728
4729 // If control is dead, only non-array-path is taken.
4730 if (stopped()) return true;
4731
4732 // The works fine even if the array type is polymorphic.
4733 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4734 Node* result = load_array_length(array);
4735
4736 C->set_has_split_ifs(true); // Has chance for split-if optimization
4737 set_result(result);
4738 return true;
4739 }
4740
4741 //------------------------inline_array_copyOf----------------------------
4742 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4743 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4744 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4745 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4746
4747 // Get the arguments.
4748 Node* original = argument(0);
4749 Node* start = is_copyOfRange? argument(1): intcon(0);
4750 Node* end = is_copyOfRange? argument(2): argument(1);
4751 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4752
4753 Node* newcopy = nullptr;
4754
4755 // Set the original stack and the reexecute bit for the interpreter to reexecute
4756 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4757 { PreserveReexecuteState preexecs(this);
4758 jvms()->set_should_reexecute(true);
4759
4760 array_type_mirror = null_check(array_type_mirror);
4761 original = null_check(original);
4762
4763 // Check if a null path was taken unconditionally.
4764 if (stopped()) return true;
4765
4766 Node* orig_length = load_array_length(original);
4767
4768 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4769 klass_node = null_check(klass_node);
4770
4771 RegionNode* bailout = new RegionNode(1);
4772 record_for_igvn(bailout);
4773
4774 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4775 // Bail out if that is so.
4776 // Inline type array may have object field that would require a
4777 // write barrier. Conservatively, go to slow path.
4778 // TODO 8251971: Optimize for the case when flat src/dst are later found
4779 // to not contain oops (i.e., move this check to the macro expansion phase).
4780 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4781 const TypeAryPtr* orig_t = _gvn.type(original)->isa_aryptr();
4782 const TypeKlassPtr* tklass = _gvn.type(klass_node)->is_klassptr();
4783 bool exclude_flat = UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing) &&
4784 // Can src array be flat and contain oops?
4785 (orig_t == nullptr || (!orig_t->is_not_flat() && (!orig_t->is_flat() || orig_t->elem()->inline_klass()->contains_oops()))) &&
4786 // Can dest array be flat and contain oops?
4787 tklass->can_be_inline_array() && (!tklass->is_flat() || tklass->is_aryklassptr()->elem()->is_instklassptr()->instance_klass()->as_inline_klass()->contains_oops());
4788 Node* not_objArray = exclude_flat ? generate_non_objArray_guard(klass_node, bailout) : generate_typeArray_guard(klass_node, bailout);
4789 if (not_objArray != nullptr) {
4790 // Improve the klass node's type from the new optimistic assumption:
4791 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4792 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0));
4793 Node* cast = new CastPPNode(control(), klass_node, akls);
4794 klass_node = _gvn.transform(cast);
4795 }
4796
4797 // Bail out if either start or end is negative.
4798 generate_negative_guard(start, bailout, &start);
4799 generate_negative_guard(end, bailout, &end);
4800
4801 Node* length = end;
4802 if (_gvn.type(start) != TypeInt::ZERO) {
4803 length = _gvn.transform(new SubINode(end, start));
4804 }
4805
4806 // Bail out if length is negative (i.e., if start > end).
4807 // Without this the new_array would throw
4808 // NegativeArraySizeException but IllegalArgumentException is what
4809 // should be thrown
4810 generate_negative_guard(length, bailout, &length);
4811
4812 // Handle inline type arrays
4813 bool can_validate = !too_many_traps(Deoptimization::Reason_class_check);
4814 if (!stopped()) {
4815 // TODO JDK-8329224
4816 if (!orig_t->is_null_free()) {
4817 // Not statically known to be null free, add a check
4818 generate_fair_guard(null_free_array_test(original), bailout);
4819 }
4820 orig_t = _gvn.type(original)->isa_aryptr();
4821 if (orig_t != nullptr && orig_t->is_flat()) {
4822 // Src is flat, check that dest is flat as well
4823 if (exclude_flat) {
4824 // Dest can't be flat, bail out
4825 bailout->add_req(control());
4826 set_control(top());
4827 } else {
4828 generate_fair_guard(flat_array_test(klass_node, /* flat = */ false), bailout);
4829 }
4830 // TODO 8350865 This is not correct anymore. Write tests and fix logic similar to arraycopy.
4831 } else if (UseArrayFlattening && (orig_t == nullptr || !orig_t->is_not_flat()) &&
4832 // If dest is flat, src must be flat as well (guaranteed by src <: dest check if validated).
4833 ((!tklass->is_flat() && tklass->can_be_inline_array()) || !can_validate)) {
4834 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
4835 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
4836 generate_fair_guard(flat_array_test(load_object_klass(original)), bailout);
4837 if (orig_t != nullptr) {
4838 orig_t = orig_t->cast_to_not_flat();
4839 original = _gvn.transform(new CheckCastPPNode(control(), original, orig_t));
4840 }
4841 }
4842 if (!can_validate) {
4843 // No validation. The subtype check emitted at macro expansion time will not go to the slow
4844 // path but call checkcast_arraycopy which can not handle flat/null-free inline type arrays.
4845 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat/null-free.
4846 generate_fair_guard(flat_array_test(klass_node), bailout);
4847 generate_fair_guard(null_free_array_test(original), bailout);
4848 }
4849 }
4850
4851 // Bail out if start is larger than the original length
4852 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4853 generate_negative_guard(orig_tail, bailout, &orig_tail);
4854
4855 if (bailout->req() > 1) {
4856 PreserveJVMState pjvms(this);
4857 set_control(_gvn.transform(bailout));
4858 uncommon_trap(Deoptimization::Reason_intrinsic,
4859 Deoptimization::Action_maybe_recompile);
4860 }
4861
4862 if (!stopped()) {
4863 // How many elements will we copy from the original?
4864 // The answer is MinI(orig_tail, length).
4865 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
4866
4867 // Generate a direct call to the right arraycopy function(s).
4868 // We know the copy is disjoint but we might not know if the
4869 // oop stores need checking.
4870 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4871 // This will fail a store-check if x contains any non-nulls.
4872
4873 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4874 // loads/stores but it is legal only if we're sure the
4875 // Arrays.copyOf would succeed. So we need all input arguments
4876 // to the copyOf to be validated, including that the copy to the
4877 // new array won't trigger an ArrayStoreException. That subtype
4878 // check can be optimized if we know something on the type of
4879 // the input array from type speculation.
4880 if (_gvn.type(klass_node)->singleton()) {
4881 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4882 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4883
4884 int test = C->static_subtype_check(superk, subk);
4885 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4886 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4887 if (t_original->speculative_type() != nullptr) {
4888 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4889 }
4890 }
4891 }
4892
4893 bool validated = false;
4894 // Reason_class_check rather than Reason_intrinsic because we
4895 // want to intrinsify even if this traps.
4896 if (can_validate) {
4897 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4898
4899 if (not_subtype_ctrl != top()) {
4900 PreserveJVMState pjvms(this);
4901 set_control(not_subtype_ctrl);
4902 uncommon_trap(Deoptimization::Reason_class_check,
4903 Deoptimization::Action_make_not_entrant);
4904 assert(stopped(), "Should be stopped");
4905 }
4906 validated = true;
4907 }
4908
4909 if (!stopped()) {
4910 newcopy = new_array(klass_node, length, 0); // no arguments to push
4911
4912 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4913 load_object_klass(original), klass_node);
4914 if (!is_copyOfRange) {
4915 ac->set_copyof(validated);
4916 } else {
4917 ac->set_copyofrange(validated);
4918 }
4919 Node* n = _gvn.transform(ac);
4920 if (n == ac) {
4921 ac->connect_outputs(this);
4922 } else {
4923 assert(validated, "shouldn't transform if all arguments not validated");
4924 set_all_memory(n);
4925 }
4926 }
4927 }
4928 } // original reexecute is set back here
4929
4930 C->set_has_split_ifs(true); // Has chance for split-if optimization
4931 if (!stopped()) {
4932 set_result(newcopy);
4933 }
4934 return true;
4935 }
4936
4937
4938 //----------------------generate_virtual_guard---------------------------
4939 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4940 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4941 RegionNode* slow_region) {
4942 ciMethod* method = callee();
4943 int vtable_index = method->vtable_index();
4944 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4945 "bad index %d", vtable_index);
4946 // Get the Method* out of the appropriate vtable entry.
4947 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4948 vtable_index*vtableEntry::size_in_bytes() +
4949 in_bytes(vtableEntry::method_offset());
4950 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4951 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4952
4953 // Compare the target method with the expected method (e.g., Object.hashCode).
4954 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4955
4956 Node* native_call = makecon(native_call_addr);
4957 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4958 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4959
4960 return generate_slow_guard(test_native, slow_region);
4961 }
4962
4963 //-----------------------generate_method_call----------------------------
4964 // Use generate_method_call to make a slow-call to the real
4965 // method if the fast path fails. An alternative would be to
4966 // use a stub like OptoRuntime::slow_arraycopy_Java.
4967 // This only works for expanding the current library call,
4968 // not another intrinsic. (E.g., don't use this for making an
4969 // arraycopy call inside of the copyOf intrinsic.)
4970 CallJavaNode*
4971 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4972 // When compiling the intrinsic method itself, do not use this technique.
4973 guarantee(callee() != C->method(), "cannot make slow-call to self");
4974
4975 ciMethod* method = callee();
4976 // ensure the JVMS we have will be correct for this call
4977 guarantee(method_id == method->intrinsic_id(), "must match");
4978
4979 const TypeFunc* tf = TypeFunc::make(method);
4980 if (res_not_null) {
4981 assert(tf->return_type() == T_OBJECT, "");
4982 const TypeTuple* range = tf->range_cc();
4983 const Type** fields = TypeTuple::fields(range->cnt());
4984 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4985 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4986 tf = TypeFunc::make(tf->domain_cc(), new_range);
4987 }
4988 CallJavaNode* slow_call;
4989 if (is_static) {
4990 assert(!is_virtual, "");
4991 slow_call = new CallStaticJavaNode(C, tf,
4992 SharedRuntime::get_resolve_static_call_stub(), method);
4993 } else if (is_virtual) {
4994 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4995 int vtable_index = Method::invalid_vtable_index;
4996 if (UseInlineCaches) {
4997 // Suppress the vtable call
4998 } else {
4999 // hashCode and clone are not a miranda methods,
5000 // so the vtable index is fixed.
5001 // No need to use the linkResolver to get it.
5002 vtable_index = method->vtable_index();
5003 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5004 "bad index %d", vtable_index);
5005 }
5006 slow_call = new CallDynamicJavaNode(tf,
5007 SharedRuntime::get_resolve_virtual_call_stub(),
5008 method, vtable_index);
5009 } else { // neither virtual nor static: opt_virtual
5010 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5011 slow_call = new CallStaticJavaNode(C, tf,
5012 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
5013 slow_call->set_optimized_virtual(true);
5014 }
5015 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
5016 // To be able to issue a direct call (optimized virtual or virtual)
5017 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
5018 // about the method being invoked should be attached to the call site to
5019 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
5020 slow_call->set_override_symbolic_info(true);
5021 }
5022 set_arguments_for_java_call(slow_call);
5023 set_edges_for_java_call(slow_call);
5024 return slow_call;
5025 }
5026
5027
5028 /**
5029 * Build special case code for calls to hashCode on an object. This call may
5030 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
5031 * slightly different code.
5032 */
5033 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
5034 assert(is_static == callee()->is_static(), "correct intrinsic selection");
5035 assert(!(is_virtual && is_static), "either virtual, special, or static");
5036
5037 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
5038
5039 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5040 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
5041 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
5042 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5043 Node* obj = argument(0);
5044
5045 // Don't intrinsify hashcode on inline types for now.
5046 // The "is locked" runtime check below also serves as inline type check and goes to the slow path.
5047 if (gvn().type(obj)->is_inlinetypeptr()) {
5048 return false;
5049 }
5050
5051 if (!is_static) {
5052 // Check for hashing null object
5053 obj = null_check_receiver();
5054 if (stopped()) return true; // unconditionally null
5055 result_reg->init_req(_null_path, top());
5056 result_val->init_req(_null_path, top());
5057 } else {
5058 // Do a null check, and return zero if null.
5059 // System.identityHashCode(null) == 0
5060 Node* null_ctl = top();
5061 obj = null_check_oop(obj, &null_ctl);
5062 result_reg->init_req(_null_path, null_ctl);
5063 result_val->init_req(_null_path, _gvn.intcon(0));
5064 }
5065
5066 // Unconditionally null? Then return right away.
5067 if (stopped()) {
5068 set_control( result_reg->in(_null_path));
5069 if (!stopped())
5070 set_result(result_val->in(_null_path));
5071 return true;
5072 }
5073
5074 // We only go to the fast case code if we pass a number of guards. The
5075 // paths which do not pass are accumulated in the slow_region.
5076 RegionNode* slow_region = new RegionNode(1);
5077 record_for_igvn(slow_region);
5078
5079 // If this is a virtual call, we generate a funny guard. We pull out
5080 // the vtable entry corresponding to hashCode() from the target object.
5081 // If the target method which we are calling happens to be the native
5082 // Object hashCode() method, we pass the guard. We do not need this
5083 // guard for non-virtual calls -- the caller is known to be the native
5084 // Object hashCode().
5085 if (is_virtual) {
5086 // After null check, get the object's klass.
5087 Node* obj_klass = load_object_klass(obj);
5088 generate_virtual_guard(obj_klass, slow_region);
5089 }
5090
5091 // Get the header out of the object, use LoadMarkNode when available
5092 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
5093 // The control of the load must be null. Otherwise, the load can move before
5094 // the null check after castPP removal.
5095 Node* no_ctrl = nullptr;
5096 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
5097
5098 if (!UseObjectMonitorTable) {
5099 // Test the header to see if it is safe to read w.r.t. locking.
5100 // This also serves as guard against inline types
5101 Node *lock_mask = _gvn.MakeConX(markWord::inline_type_mask_in_place);
5102 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
5103 if (LockingMode == LM_LIGHTWEIGHT) {
5104 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
5105 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
5106 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
5107
5108 generate_slow_guard(test_monitor, slow_region);
5109 } else {
5110 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
5111 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val));
5112 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne));
5113
5114 generate_slow_guard(test_not_unlocked, slow_region);
5115 }
5116 }
5117
5118 // Get the hash value and check to see that it has been properly assigned.
5119 // We depend on hash_mask being at most 32 bits and avoid the use of
5120 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
5121 // vm: see markWord.hpp.
5122 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
5123 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
5124 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
5125 // This hack lets the hash bits live anywhere in the mark object now, as long
5126 // as the shift drops the relevant bits into the low 32 bits. Note that
5127 // Java spec says that HashCode is an int so there's no point in capturing
5128 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
5129 hshifted_header = ConvX2I(hshifted_header);
5130 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
5131
5132 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
5133 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
5134 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
5135
5136 generate_slow_guard(test_assigned, slow_region);
5137
5138 Node* init_mem = reset_memory();
5139 // fill in the rest of the null path:
5140 result_io ->init_req(_null_path, i_o());
5141 result_mem->init_req(_null_path, init_mem);
5142
5143 result_val->init_req(_fast_path, hash_val);
5144 result_reg->init_req(_fast_path, control());
5145 result_io ->init_req(_fast_path, i_o());
5146 result_mem->init_req(_fast_path, init_mem);
5147
5148 // Generate code for the slow case. We make a call to hashCode().
5149 set_control(_gvn.transform(slow_region));
5150 if (!stopped()) {
5151 // No need for PreserveJVMState, because we're using up the present state.
5152 set_all_memory(init_mem);
5153 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
5154 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
5155 Node* slow_result = set_results_for_java_call(slow_call);
5156 // this->control() comes from set_results_for_java_call
5157 result_reg->init_req(_slow_path, control());
5158 result_val->init_req(_slow_path, slow_result);
5159 result_io ->set_req(_slow_path, i_o());
5160 result_mem ->set_req(_slow_path, reset_memory());
5161 }
5162
5163 // Return the combined state.
5164 set_i_o( _gvn.transform(result_io) );
5165 set_all_memory( _gvn.transform(result_mem));
5166
5167 set_result(result_reg, result_val);
5168 return true;
5169 }
5170
5171 //---------------------------inline_native_getClass----------------------------
5172 // public final native Class<?> java.lang.Object.getClass();
5173 //
5174 // Build special case code for calls to getClass on an object.
5175 bool LibraryCallKit::inline_native_getClass() {
5176 Node* obj = argument(0);
5177 if (obj->is_InlineType()) {
5178 const Type* t = _gvn.type(obj);
5179 if (t->maybe_null()) {
5180 null_check(obj);
5181 }
5182 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror())));
5183 return true;
5184 }
5185 obj = null_check_receiver();
5186 if (stopped()) return true;
5187 set_result(load_mirror_from_klass(load_object_klass(obj)));
5188 return true;
5189 }
5190
5191 //-----------------inline_native_Reflection_getCallerClass---------------------
5192 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
5193 //
5194 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
5195 //
5196 // NOTE: This code must perform the same logic as JVM_GetCallerClass
5197 // in that it must skip particular security frames and checks for
5198 // caller sensitive methods.
5199 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
5200 #ifndef PRODUCT
5201 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5202 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
5203 }
5204 #endif
5205
5206 if (!jvms()->has_method()) {
5207 #ifndef PRODUCT
5208 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5209 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
5210 }
5211 #endif
5212 return false;
5213 }
5214
5215 // Walk back up the JVM state to find the caller at the required
5216 // depth.
5217 JVMState* caller_jvms = jvms();
5218
5219 // Cf. JVM_GetCallerClass
5220 // NOTE: Start the loop at depth 1 because the current JVM state does
5221 // not include the Reflection.getCallerClass() frame.
5222 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
5223 ciMethod* m = caller_jvms->method();
5224 switch (n) {
5225 case 0:
5226 fatal("current JVM state does not include the Reflection.getCallerClass frame");
5227 break;
5228 case 1:
5229 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
5230 if (!m->caller_sensitive()) {
5231 #ifndef PRODUCT
5232 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5233 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
5234 }
5235 #endif
5236 return false; // bail-out; let JVM_GetCallerClass do the work
5237 }
5238 break;
5239 default:
5240 if (!m->is_ignored_by_security_stack_walk()) {
5241 // We have reached the desired frame; return the holder class.
5242 // Acquire method holder as java.lang.Class and push as constant.
5243 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
5244 ciInstance* caller_mirror = caller_klass->java_mirror();
5245 set_result(makecon(TypeInstPtr::make(caller_mirror)));
5246
5247 #ifndef PRODUCT
5248 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5249 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());
5250 tty->print_cr(" JVM state at this point:");
5251 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5252 ciMethod* m = jvms()->of_depth(i)->method();
5253 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5254 }
5255 }
5256 #endif
5257 return true;
5258 }
5259 break;
5260 }
5261 }
5262
5263 #ifndef PRODUCT
5264 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5265 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
5266 tty->print_cr(" JVM state at this point:");
5267 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5268 ciMethod* m = jvms()->of_depth(i)->method();
5269 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5270 }
5271 }
5272 #endif
5273
5274 return false; // bail-out; let JVM_GetCallerClass do the work
5275 }
5276
5277 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
5278 Node* arg = argument(0);
5279 Node* result = nullptr;
5280
5281 switch (id) {
5282 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
5283 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
5284 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5285 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5286 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5287 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5288
5289 case vmIntrinsics::_doubleToLongBits: {
5290 // two paths (plus control) merge in a wood
5291 RegionNode *r = new RegionNode(3);
5292 Node *phi = new PhiNode(r, TypeLong::LONG);
5293
5294 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5295 // Build the boolean node
5296 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5297
5298 // Branch either way.
5299 // NaN case is less traveled, which makes all the difference.
5300 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5301 Node *opt_isnan = _gvn.transform(ifisnan);
5302 assert( opt_isnan->is_If(), "Expect an IfNode");
5303 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5304 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5305
5306 set_control(iftrue);
5307
5308 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5309 Node *slow_result = longcon(nan_bits); // return NaN
5310 phi->init_req(1, _gvn.transform( slow_result ));
5311 r->init_req(1, iftrue);
5312
5313 // Else fall through
5314 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5315 set_control(iffalse);
5316
5317 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5318 r->init_req(2, iffalse);
5319
5320 // Post merge
5321 set_control(_gvn.transform(r));
5322 record_for_igvn(r);
5323
5324 C->set_has_split_ifs(true); // Has chance for split-if optimization
5325 result = phi;
5326 assert(result->bottom_type()->isa_long(), "must be");
5327 break;
5328 }
5329
5330 case vmIntrinsics::_floatToIntBits: {
5331 // two paths (plus control) merge in a wood
5332 RegionNode *r = new RegionNode(3);
5333 Node *phi = new PhiNode(r, TypeInt::INT);
5334
5335 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5336 // Build the boolean node
5337 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5338
5339 // Branch either way.
5340 // NaN case is less traveled, which makes all the difference.
5341 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5342 Node *opt_isnan = _gvn.transform(ifisnan);
5343 assert( opt_isnan->is_If(), "Expect an IfNode");
5344 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5345 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5346
5347 set_control(iftrue);
5348
5349 static const jint nan_bits = 0x7fc00000;
5350 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5351 phi->init_req(1, _gvn.transform( slow_result ));
5352 r->init_req(1, iftrue);
5353
5354 // Else fall through
5355 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5356 set_control(iffalse);
5357
5358 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5359 r->init_req(2, iffalse);
5360
5361 // Post merge
5362 set_control(_gvn.transform(r));
5363 record_for_igvn(r);
5364
5365 C->set_has_split_ifs(true); // Has chance for split-if optimization
5366 result = phi;
5367 assert(result->bottom_type()->isa_int(), "must be");
5368 break;
5369 }
5370
5371 default:
5372 fatal_unexpected_iid(id);
5373 break;
5374 }
5375 set_result(_gvn.transform(result));
5376 return true;
5377 }
5378
5379 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5380 Node* arg = argument(0);
5381 Node* result = nullptr;
5382
5383 switch (id) {
5384 case vmIntrinsics::_floatIsInfinite:
5385 result = new IsInfiniteFNode(arg);
5386 break;
5387 case vmIntrinsics::_floatIsFinite:
5388 result = new IsFiniteFNode(arg);
5389 break;
5390 case vmIntrinsics::_doubleIsInfinite:
5391 result = new IsInfiniteDNode(arg);
5392 break;
5393 case vmIntrinsics::_doubleIsFinite:
5394 result = new IsFiniteDNode(arg);
5395 break;
5396 default:
5397 fatal_unexpected_iid(id);
5398 break;
5399 }
5400 set_result(_gvn.transform(result));
5401 return true;
5402 }
5403
5404 //----------------------inline_unsafe_copyMemory-------------------------
5405 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5406
5407 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5408 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5409 const Type* base_t = gvn.type(base);
5410
5411 bool in_native = (base_t == TypePtr::NULL_PTR);
5412 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5413 bool is_mixed = !in_heap && !in_native;
5414
5415 if (is_mixed) {
5416 return true; // mixed accesses can touch both on-heap and off-heap memory
5417 }
5418 if (in_heap) {
5419 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5420 if (!is_prim_array) {
5421 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5422 // there's not enough type information available to determine proper memory slice for it.
5423 return true;
5424 }
5425 }
5426 return false;
5427 }
5428
5429 bool LibraryCallKit::inline_unsafe_copyMemory() {
5430 if (callee()->is_static()) return false; // caller must have the capability!
5431 null_check_receiver(); // null-check receiver
5432 if (stopped()) return true;
5433
5434 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5435
5436 Node* src_base = argument(1); // type: oop
5437 Node* src_off = ConvL2X(argument(2)); // type: long
5438 Node* dst_base = argument(4); // type: oop
5439 Node* dst_off = ConvL2X(argument(5)); // type: long
5440 Node* size = ConvL2X(argument(7)); // type: long
5441
5442 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5443 "fieldOffset must be byte-scaled");
5444
5445 Node* src_addr = make_unsafe_address(src_base, src_off);
5446 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5447
5448 Node* thread = _gvn.transform(new ThreadLocalNode());
5449 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5450 BasicType doing_unsafe_access_bt = T_BYTE;
5451 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5452
5453 // update volatile field
5454 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5455
5456 int flags = RC_LEAF | RC_NO_FP;
5457
5458 const TypePtr* dst_type = TypePtr::BOTTOM;
5459
5460 // Adjust memory effects of the runtime call based on input values.
5461 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5462 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5463 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5464
5465 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5466 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5467 flags |= RC_NARROW_MEM; // narrow in memory
5468 }
5469 }
5470
5471 // Call it. Note that the length argument is not scaled.
5472 make_runtime_call(flags,
5473 OptoRuntime::fast_arraycopy_Type(),
5474 StubRoutines::unsafe_arraycopy(),
5475 "unsafe_arraycopy",
5476 dst_type,
5477 src_addr, dst_addr, size XTOP);
5478
5479 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5480
5481 return true;
5482 }
5483
5484 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5485 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5486 bool LibraryCallKit::inline_unsafe_setMemory() {
5487 if (callee()->is_static()) return false; // caller must have the capability!
5488 null_check_receiver(); // null-check receiver
5489 if (stopped()) return true;
5490
5491 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5492
5493 Node* dst_base = argument(1); // type: oop
5494 Node* dst_off = ConvL2X(argument(2)); // type: long
5495 Node* size = ConvL2X(argument(4)); // type: long
5496 Node* byte = argument(6); // type: byte
5497
5498 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5499 "fieldOffset must be byte-scaled");
5500
5501 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5502
5503 Node* thread = _gvn.transform(new ThreadLocalNode());
5504 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5505 BasicType doing_unsafe_access_bt = T_BYTE;
5506 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5507
5508 // update volatile field
5509 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5510
5511 int flags = RC_LEAF | RC_NO_FP;
5512
5513 const TypePtr* dst_type = TypePtr::BOTTOM;
5514
5515 // Adjust memory effects of the runtime call based on input values.
5516 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5517 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5518
5519 flags |= RC_NARROW_MEM; // narrow in memory
5520 }
5521
5522 // Call it. Note that the length argument is not scaled.
5523 make_runtime_call(flags,
5524 OptoRuntime::unsafe_setmemory_Type(),
5525 StubRoutines::unsafe_setmemory(),
5526 "unsafe_setmemory",
5527 dst_type,
5528 dst_addr, size XTOP, byte);
5529
5530 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5531
5532 return true;
5533 }
5534
5535 #undef XTOP
5536
5537 //----------------------inline_unsafe_isFlatArray------------------------
5538 // public native boolean Unsafe.isFlatArray(Class<?> arrayClass);
5539 // This intrinsic exploits assumptions made by the native implementation
5540 // (arrayClass is neither null nor primitive) to avoid unnecessary null checks.
5541 bool LibraryCallKit::inline_unsafe_isFlatArray() {
5542 Node* cls = argument(1);
5543 Node* p = basic_plus_adr(cls, java_lang_Class::klass_offset());
5544 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p,
5545 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT));
5546 Node* result = flat_array_test(kls);
5547 set_result(result);
5548 return true;
5549 }
5550
5551 //------------------------clone_coping-----------------------------------
5552 // Helper function for inline_native_clone.
5553 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5554 assert(obj_size != nullptr, "");
5555 Node* raw_obj = alloc_obj->in(1);
5556 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5557
5558 AllocateNode* alloc = nullptr;
5559 if (ReduceBulkZeroing &&
5560 // If we are implementing an array clone without knowing its source type
5561 // (can happen when compiling the array-guarded branch of a reflective
5562 // Object.clone() invocation), initialize the array within the allocation.
5563 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5564 // to a runtime clone call that assumes fully initialized source arrays.
5565 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5566 // We will be completely responsible for initializing this object -
5567 // mark Initialize node as complete.
5568 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5569 // The object was just allocated - there should be no any stores!
5570 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5571 // Mark as complete_with_arraycopy so that on AllocateNode
5572 // expansion, we know this AllocateNode is initialized by an array
5573 // copy and a StoreStore barrier exists after the array copy.
5574 alloc->initialization()->set_complete_with_arraycopy();
5575 }
5576
5577 Node* size = _gvn.transform(obj_size);
5578 access_clone(obj, alloc_obj, size, is_array);
5579
5580 // Do not let reads from the cloned object float above the arraycopy.
5581 if (alloc != nullptr) {
5582 // Do not let stores that initialize this object be reordered with
5583 // a subsequent store that would make this object accessible by
5584 // other threads.
5585 // Record what AllocateNode this StoreStore protects so that
5586 // escape analysis can go from the MemBarStoreStoreNode to the
5587 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5588 // based on the escape status of the AllocateNode.
5589 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5590 } else {
5591 insert_mem_bar(Op_MemBarCPUOrder);
5592 }
5593 }
5594
5595 //------------------------inline_native_clone----------------------------
5596 // protected native Object java.lang.Object.clone();
5597 //
5598 // Here are the simple edge cases:
5599 // null receiver => normal trap
5600 // virtual and clone was overridden => slow path to out-of-line clone
5601 // not cloneable or finalizer => slow path to out-of-line Object.clone
5602 //
5603 // The general case has two steps, allocation and copying.
5604 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5605 //
5606 // Copying also has two cases, oop arrays and everything else.
5607 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5608 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5609 //
5610 // These steps fold up nicely if and when the cloned object's klass
5611 // can be sharply typed as an object array, a type array, or an instance.
5612 //
5613 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5614 PhiNode* result_val;
5615
5616 // Set the reexecute bit for the interpreter to reexecute
5617 // the bytecode that invokes Object.clone if deoptimization happens.
5618 { PreserveReexecuteState preexecs(this);
5619 jvms()->set_should_reexecute(true);
5620
5621 Node* obj = argument(0);
5622 obj = null_check_receiver();
5623 if (stopped()) return true;
5624
5625 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5626 if (obj_type->is_inlinetypeptr()) {
5627 // If the object to clone is an inline type, we can simply return it (i.e. a nop) since inline types have
5628 // no identity.
5629 set_result(obj);
5630 return true;
5631 }
5632
5633 // If we are going to clone an instance, we need its exact type to
5634 // know the number and types of fields to convert the clone to
5635 // loads/stores. Maybe a speculative type can help us.
5636 if (!obj_type->klass_is_exact() &&
5637 obj_type->speculative_type() != nullptr &&
5638 obj_type->speculative_type()->is_instance_klass() &&
5639 !obj_type->speculative_type()->is_inlinetype()) {
5640 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5641 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5642 !spec_ik->has_injected_fields()) {
5643 if (!obj_type->isa_instptr() ||
5644 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5645 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5646 }
5647 }
5648 }
5649
5650 // Conservatively insert a memory barrier on all memory slices.
5651 // Do not let writes into the original float below the clone.
5652 insert_mem_bar(Op_MemBarCPUOrder);
5653
5654 // paths into result_reg:
5655 enum {
5656 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5657 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5658 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5659 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5660 PATH_LIMIT
5661 };
5662 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5663 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5664 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5665 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5666 record_for_igvn(result_reg);
5667
5668 Node* obj_klass = load_object_klass(obj);
5669 // We only go to the fast case code if we pass a number of guards.
5670 // The paths which do not pass are accumulated in the slow_region.
5671 RegionNode* slow_region = new RegionNode(1);
5672 record_for_igvn(slow_region);
5673
5674 Node* array_obj = obj;
5675 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5676 if (array_ctl != nullptr) {
5677 // It's an array.
5678 PreserveJVMState pjvms(this);
5679 set_control(array_ctl);
5680
5681 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5682 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr();
5683 if (UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) &&
5684 obj_type->can_be_inline_array() &&
5685 (ary_ptr == nullptr || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) {
5686 // Flat inline type array may have object field that would require a
5687 // write barrier. Conservatively, go to slow path.
5688 generate_fair_guard(flat_array_test(obj_klass), slow_region);
5689 }
5690
5691 if (!stopped()) {
5692 Node* obj_length = load_array_length(array_obj);
5693 Node* array_size = nullptr; // Size of the array without object alignment padding.
5694 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5695
5696 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5697 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5698 // If it is an oop array, it requires very special treatment,
5699 // because gc barriers are required when accessing the array.
5700 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5701 if (is_obja != nullptr) {
5702 PreserveJVMState pjvms2(this);
5703 set_control(is_obja);
5704 // Generate a direct call to the right arraycopy function(s).
5705 // Clones are always tightly coupled.
5706 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5707 ac->set_clone_oop_array();
5708 Node* n = _gvn.transform(ac);
5709 assert(n == ac, "cannot disappear");
5710 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5711
5712 result_reg->init_req(_objArray_path, control());
5713 result_val->init_req(_objArray_path, alloc_obj);
5714 result_i_o ->set_req(_objArray_path, i_o());
5715 result_mem ->set_req(_objArray_path, reset_memory());
5716 }
5717 }
5718 // Otherwise, there are no barriers to worry about.
5719 // (We can dispense with card marks if we know the allocation
5720 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5721 // causes the non-eden paths to take compensating steps to
5722 // simulate a fresh allocation, so that no further
5723 // card marks are required in compiled code to initialize
5724 // the object.)
5725
5726 if (!stopped()) {
5727 copy_to_clone(obj, alloc_obj, array_size, true);
5728
5729 // Present the results of the copy.
5730 result_reg->init_req(_array_path, control());
5731 result_val->init_req(_array_path, alloc_obj);
5732 result_i_o ->set_req(_array_path, i_o());
5733 result_mem ->set_req(_array_path, reset_memory());
5734 }
5735 }
5736 }
5737
5738 if (!stopped()) {
5739 // It's an instance (we did array above). Make the slow-path tests.
5740 // If this is a virtual call, we generate a funny guard. We grab
5741 // the vtable entry corresponding to clone() from the target object.
5742 // If the target method which we are calling happens to be the
5743 // Object clone() method, we pass the guard. We do not need this
5744 // guard for non-virtual calls; the caller is known to be the native
5745 // Object clone().
5746 if (is_virtual) {
5747 generate_virtual_guard(obj_klass, slow_region);
5748 }
5749
5750 // The object must be easily cloneable and must not have a finalizer.
5751 // Both of these conditions may be checked in a single test.
5752 // We could optimize the test further, but we don't care.
5753 generate_misc_flags_guard(obj_klass,
5754 // Test both conditions:
5755 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5756 // Must be cloneable but not finalizer:
5757 KlassFlags::_misc_is_cloneable_fast,
5758 slow_region);
5759 }
5760
5761 if (!stopped()) {
5762 // It's an instance, and it passed the slow-path tests.
5763 PreserveJVMState pjvms(this);
5764 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5765 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5766 // is reexecuted if deoptimization occurs and there could be problems when merging
5767 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5768 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5769
5770 copy_to_clone(obj, alloc_obj, obj_size, false);
5771
5772 // Present the results of the slow call.
5773 result_reg->init_req(_instance_path, control());
5774 result_val->init_req(_instance_path, alloc_obj);
5775 result_i_o ->set_req(_instance_path, i_o());
5776 result_mem ->set_req(_instance_path, reset_memory());
5777 }
5778
5779 // Generate code for the slow case. We make a call to clone().
5780 set_control(_gvn.transform(slow_region));
5781 if (!stopped()) {
5782 PreserveJVMState pjvms(this);
5783 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5784 // We need to deoptimize on exception (see comment above)
5785 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5786 // this->control() comes from set_results_for_java_call
5787 result_reg->init_req(_slow_path, control());
5788 result_val->init_req(_slow_path, slow_result);
5789 result_i_o ->set_req(_slow_path, i_o());
5790 result_mem ->set_req(_slow_path, reset_memory());
5791 }
5792
5793 // Return the combined state.
5794 set_control( _gvn.transform(result_reg));
5795 set_i_o( _gvn.transform(result_i_o));
5796 set_all_memory( _gvn.transform(result_mem));
5797 } // original reexecute is set back here
5798
5799 set_result(_gvn.transform(result_val));
5800 return true;
5801 }
5802
5803 // If we have a tightly coupled allocation, the arraycopy may take care
5804 // of the array initialization. If one of the guards we insert between
5805 // the allocation and the arraycopy causes a deoptimization, an
5806 // uninitialized array will escape the compiled method. To prevent that
5807 // we set the JVM state for uncommon traps between the allocation and
5808 // the arraycopy to the state before the allocation so, in case of
5809 // deoptimization, we'll reexecute the allocation and the
5810 // initialization.
5811 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5812 if (alloc != nullptr) {
5813 ciMethod* trap_method = alloc->jvms()->method();
5814 int trap_bci = alloc->jvms()->bci();
5815
5816 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5817 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5818 // Make sure there's no store between the allocation and the
5819 // arraycopy otherwise visible side effects could be rexecuted
5820 // in case of deoptimization and cause incorrect execution.
5821 bool no_interfering_store = true;
5822 Node* mem = alloc->in(TypeFunc::Memory);
5823 if (mem->is_MergeMem()) {
5824 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5825 Node* n = mms.memory();
5826 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5827 assert(n->is_Store(), "what else?");
5828 no_interfering_store = false;
5829 break;
5830 }
5831 }
5832 } else {
5833 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5834 Node* n = mms.memory();
5835 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5836 assert(n->is_Store(), "what else?");
5837 no_interfering_store = false;
5838 break;
5839 }
5840 }
5841 }
5842
5843 if (no_interfering_store) {
5844 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5845
5846 JVMState* saved_jvms = jvms();
5847 saved_reexecute_sp = _reexecute_sp;
5848
5849 set_jvms(sfpt->jvms());
5850 _reexecute_sp = jvms()->sp();
5851
5852 return saved_jvms;
5853 }
5854 }
5855 }
5856 return nullptr;
5857 }
5858
5859 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5860 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5861 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5862 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5863 uint size = alloc->req();
5864 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5865 old_jvms->set_map(sfpt);
5866 for (uint i = 0; i < size; i++) {
5867 sfpt->init_req(i, alloc->in(i));
5868 }
5869 int adjustment = 1;
5870 const TypeAryKlassPtr* ary_klass_ptr = alloc->in(AllocateNode::KlassNode)->bottom_type()->is_aryklassptr();
5871 if (ary_klass_ptr->is_null_free()) {
5872 // A null-free, tightly coupled array allocation can only come from LibraryCallKit::inline_newArray which
5873 // also requires the componentType and initVal on stack for re-execution.
5874 // Re-create and push the componentType.
5875 ciArrayKlass* klass = ary_klass_ptr->exact_klass()->as_array_klass();
5876 ciInstance* instance = klass->component_mirror_instance();
5877 const TypeInstPtr* t_instance = TypeInstPtr::make(instance);
5878 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), makecon(t_instance));
5879 adjustment++;
5880 }
5881 // re-push array length for deoptimization
5882 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment - 1, alloc->in(AllocateNode::ALength));
5883 if (ary_klass_ptr->is_null_free()) {
5884 // Re-create and push the initVal.
5885 Node* init_val = alloc->in(AllocateNode::InitValue);
5886 if (init_val == nullptr) {
5887 init_val = InlineTypeNode::make_all_zero(_gvn, ary_klass_ptr->elem()->is_instklassptr()->instance_klass()->as_inline_klass());
5888 } else if (UseCompressedOops) {
5889 init_val = _gvn.transform(new DecodeNNode(init_val, init_val->bottom_type()->make_ptr()));
5890 }
5891 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment, init_val);
5892 adjustment++;
5893 }
5894 old_jvms->set_sp(old_jvms->sp() + adjustment);
5895 old_jvms->set_monoff(old_jvms->monoff() + adjustment);
5896 old_jvms->set_scloff(old_jvms->scloff() + adjustment);
5897 old_jvms->set_endoff(old_jvms->endoff() + adjustment);
5898 old_jvms->set_should_reexecute(true);
5899
5900 sfpt->set_i_o(map()->i_o());
5901 sfpt->set_memory(map()->memory());
5902 sfpt->set_control(map()->control());
5903 return sfpt;
5904 }
5905
5906 // In case of a deoptimization, we restart execution at the
5907 // allocation, allocating a new array. We would leave an uninitialized
5908 // array in the heap that GCs wouldn't expect. Move the allocation
5909 // after the traps so we don't allocate the array if we
5910 // deoptimize. This is possible because tightly_coupled_allocation()
5911 // guarantees there's no observer of the allocated array at this point
5912 // and the control flow is simple enough.
5913 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5914 int saved_reexecute_sp, uint new_idx) {
5915 if (saved_jvms_before_guards != nullptr && !stopped()) {
5916 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5917
5918 assert(alloc != nullptr, "only with a tightly coupled allocation");
5919 // restore JVM state to the state at the arraycopy
5920 saved_jvms_before_guards->map()->set_control(map()->control());
5921 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5922 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5923 // If we've improved the types of some nodes (null check) while
5924 // emitting the guards, propagate them to the current state
5925 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5926 set_jvms(saved_jvms_before_guards);
5927 _reexecute_sp = saved_reexecute_sp;
5928
5929 // Remove the allocation from above the guards
5930 CallProjections* callprojs = alloc->extract_projections(true);
5931 InitializeNode* init = alloc->initialization();
5932 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5933 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5934 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5935
5936 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5937 // the allocation (i.e. is only valid if the allocation succeeds):
5938 // 1) replace CastIINode with AllocateArrayNode's length here
5939 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5940 //
5941 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5942 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5943 Node* init_control = init->proj_out(TypeFunc::Control);
5944 Node* alloc_length = alloc->Ideal_length();
5945 #ifdef ASSERT
5946 Node* prev_cast = nullptr;
5947 #endif
5948 for (uint i = 0; i < init_control->outcnt(); i++) {
5949 Node* init_out = init_control->raw_out(i);
5950 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5951 #ifdef ASSERT
5952 if (prev_cast == nullptr) {
5953 prev_cast = init_out;
5954 } else {
5955 if (prev_cast->cmp(*init_out) == false) {
5956 prev_cast->dump();
5957 init_out->dump();
5958 assert(false, "not equal CastIINode");
5959 }
5960 }
5961 #endif
5962 C->gvn_replace_by(init_out, alloc_length);
5963 }
5964 }
5965 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5966
5967 // move the allocation here (after the guards)
5968 _gvn.hash_delete(alloc);
5969 alloc->set_req(TypeFunc::Control, control());
5970 alloc->set_req(TypeFunc::I_O, i_o());
5971 Node *mem = reset_memory();
5972 set_all_memory(mem);
5973 alloc->set_req(TypeFunc::Memory, mem);
5974 set_control(init->proj_out_or_null(TypeFunc::Control));
5975 set_i_o(callprojs->fallthrough_ioproj);
5976
5977 // Update memory as done in GraphKit::set_output_for_allocation()
5978 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5979 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5980 if (ary_type->isa_aryptr() && length_type != nullptr) {
5981 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5982 }
5983 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5984 int elemidx = C->get_alias_index(telemref);
5985 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5986 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5987
5988 Node* allocx = _gvn.transform(alloc);
5989 assert(allocx == alloc, "where has the allocation gone?");
5990 assert(dest->is_CheckCastPP(), "not an allocation result?");
5991
5992 _gvn.hash_delete(dest);
5993 dest->set_req(0, control());
5994 Node* destx = _gvn.transform(dest);
5995 assert(destx == dest, "where has the allocation result gone?");
5996
5997 array_ideal_length(alloc, ary_type, true);
5998 }
5999 }
6000
6001 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
6002 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
6003 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
6004 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
6005 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
6006 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
6007 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
6008 JVMState* saved_jvms_before_guards) {
6009 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
6010 // There is at least one unrelated uncommon trap which needs to be replaced.
6011 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6012
6013 JVMState* saved_jvms = jvms();
6014 const int saved_reexecute_sp = _reexecute_sp;
6015 set_jvms(sfpt->jvms());
6016 _reexecute_sp = jvms()->sp();
6017
6018 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
6019
6020 // Restore state
6021 set_jvms(saved_jvms);
6022 _reexecute_sp = saved_reexecute_sp;
6023 }
6024 }
6025
6026 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
6027 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
6028 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
6029 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
6030 while (if_proj->is_IfProj()) {
6031 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
6032 if (uncommon_trap != nullptr) {
6033 create_new_uncommon_trap(uncommon_trap);
6034 }
6035 assert(if_proj->in(0)->is_If(), "must be If");
6036 if_proj = if_proj->in(0)->in(0);
6037 }
6038 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
6039 "must have reached control projection of init node");
6040 }
6041
6042 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
6043 const int trap_request = uncommon_trap_call->uncommon_trap_request();
6044 assert(trap_request != 0, "no valid UCT trap request");
6045 PreserveJVMState pjvms(this);
6046 set_control(uncommon_trap_call->in(0));
6047 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
6048 Deoptimization::trap_request_action(trap_request));
6049 assert(stopped(), "Should be stopped");
6050 _gvn.hash_delete(uncommon_trap_call);
6051 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
6052 }
6053
6054 // Common checks for array sorting intrinsics arguments.
6055 // Returns `true` if checks passed.
6056 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
6057 // check address of the class
6058 if (elementType == nullptr || elementType->is_top()) {
6059 return false; // dead path
6060 }
6061 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
6062 if (elem_klass == nullptr) {
6063 return false; // dead path
6064 }
6065 // java_mirror_type() returns non-null for compile-time Class constants only
6066 ciType* elem_type = elem_klass->java_mirror_type();
6067 if (elem_type == nullptr) {
6068 return false;
6069 }
6070 bt = elem_type->basic_type();
6071 // Disable the intrinsic if the CPU does not support SIMD sort
6072 if (!Matcher::supports_simd_sort(bt)) {
6073 return false;
6074 }
6075 // check address of the array
6076 if (obj == nullptr || obj->is_top()) {
6077 return false; // dead path
6078 }
6079 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
6080 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
6081 return false; // failed input validation
6082 }
6083 return true;
6084 }
6085
6086 //------------------------------inline_array_partition-----------------------
6087 bool LibraryCallKit::inline_array_partition() {
6088 address stubAddr = StubRoutines::select_array_partition_function();
6089 if (stubAddr == nullptr) {
6090 return false; // Intrinsic's stub is not implemented on this platform
6091 }
6092 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
6093
6094 // no receiver because it is a static method
6095 Node* elementType = argument(0);
6096 Node* obj = argument(1);
6097 Node* offset = argument(2); // long
6098 Node* fromIndex = argument(4);
6099 Node* toIndex = argument(5);
6100 Node* indexPivot1 = argument(6);
6101 Node* indexPivot2 = argument(7);
6102 // PartitionOperation: argument(8) is ignored
6103
6104 Node* pivotIndices = nullptr;
6105 BasicType bt = T_ILLEGAL;
6106
6107 if (!check_array_sort_arguments(elementType, obj, bt)) {
6108 return false;
6109 }
6110 null_check(obj);
6111 // If obj is dead, only null-path is taken.
6112 if (stopped()) {
6113 return true;
6114 }
6115 // Set the original stack and the reexecute bit for the interpreter to reexecute
6116 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
6117 { PreserveReexecuteState preexecs(this);
6118 jvms()->set_should_reexecute(true);
6119
6120 Node* obj_adr = make_unsafe_address(obj, offset);
6121
6122 // create the pivotIndices array of type int and size = 2
6123 Node* size = intcon(2);
6124 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
6125 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
6126 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
6127 guarantee(alloc != nullptr, "created above");
6128 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
6129
6130 // pass the basic type enum to the stub
6131 Node* elemType = intcon(bt);
6132
6133 // Call the stub
6134 const char *stubName = "array_partition_stub";
6135 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
6136 stubAddr, stubName, TypePtr::BOTTOM,
6137 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
6138 indexPivot1, indexPivot2);
6139
6140 } // original reexecute is set back here
6141
6142 if (!stopped()) {
6143 set_result(pivotIndices);
6144 }
6145
6146 return true;
6147 }
6148
6149
6150 //------------------------------inline_array_sort-----------------------
6151 bool LibraryCallKit::inline_array_sort() {
6152 address stubAddr = StubRoutines::select_arraysort_function();
6153 if (stubAddr == nullptr) {
6154 return false; // Intrinsic's stub is not implemented on this platform
6155 }
6156 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
6157
6158 // no receiver because it is a static method
6159 Node* elementType = argument(0);
6160 Node* obj = argument(1);
6161 Node* offset = argument(2); // long
6162 Node* fromIndex = argument(4);
6163 Node* toIndex = argument(5);
6164 // SortOperation: argument(6) is ignored
6165
6166 BasicType bt = T_ILLEGAL;
6167
6168 if (!check_array_sort_arguments(elementType, obj, bt)) {
6169 return false;
6170 }
6171 null_check(obj);
6172 // If obj is dead, only null-path is taken.
6173 if (stopped()) {
6174 return true;
6175 }
6176 Node* obj_adr = make_unsafe_address(obj, offset);
6177
6178 // pass the basic type enum to the stub
6179 Node* elemType = intcon(bt);
6180
6181 // Call the stub.
6182 const char *stubName = "arraysort_stub";
6183 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
6184 stubAddr, stubName, TypePtr::BOTTOM,
6185 obj_adr, elemType, fromIndex, toIndex);
6186
6187 return true;
6188 }
6189
6190
6191 //------------------------------inline_arraycopy-----------------------
6192 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
6193 // Object dest, int destPos,
6194 // int length);
6195 bool LibraryCallKit::inline_arraycopy() {
6196 // Get the arguments.
6197 Node* src = argument(0); // type: oop
6198 Node* src_offset = argument(1); // type: int
6199 Node* dest = argument(2); // type: oop
6200 Node* dest_offset = argument(3); // type: int
6201 Node* length = argument(4); // type: int
6202
6203 uint new_idx = C->unique();
6204
6205 // Check for allocation before we add nodes that would confuse
6206 // tightly_coupled_allocation()
6207 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
6208
6209 int saved_reexecute_sp = -1;
6210 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
6211 // See arraycopy_restore_alloc_state() comment
6212 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
6213 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
6214 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
6215 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
6216
6217 // The following tests must be performed
6218 // (1) src and dest are arrays.
6219 // (2) src and dest arrays must have elements of the same BasicType
6220 // (3) src and dest must not be null.
6221 // (4) src_offset must not be negative.
6222 // (5) dest_offset must not be negative.
6223 // (6) length must not be negative.
6224 // (7) src_offset + length must not exceed length of src.
6225 // (8) dest_offset + length must not exceed length of dest.
6226 // (9) each element of an oop array must be assignable
6227
6228 // (3) src and dest must not be null.
6229 // always do this here because we need the JVM state for uncommon traps
6230 Node* null_ctl = top();
6231 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
6232 assert(null_ctl->is_top(), "no null control here");
6233 dest = null_check(dest, T_ARRAY);
6234
6235 if (!can_emit_guards) {
6236 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
6237 // guards but the arraycopy node could still take advantage of a
6238 // tightly allocated allocation. tightly_coupled_allocation() is
6239 // called again to make sure it takes the null check above into
6240 // account: the null check is mandatory and if it caused an
6241 // uncommon trap to be emitted then the allocation can't be
6242 // considered tightly coupled in this context.
6243 alloc = tightly_coupled_allocation(dest);
6244 }
6245
6246 bool validated = false;
6247
6248 const Type* src_type = _gvn.type(src);
6249 const Type* dest_type = _gvn.type(dest);
6250 const TypeAryPtr* top_src = src_type->isa_aryptr();
6251 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6252
6253 // Do we have the type of src?
6254 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6255 // Do we have the type of dest?
6256 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6257 // Is the type for src from speculation?
6258 bool src_spec = false;
6259 // Is the type for dest from speculation?
6260 bool dest_spec = false;
6261
6262 if ((!has_src || !has_dest) && can_emit_guards) {
6263 // We don't have sufficient type information, let's see if
6264 // speculative types can help. We need to have types for both src
6265 // and dest so that it pays off.
6266
6267 // Do we already have or could we have type information for src
6268 bool could_have_src = has_src;
6269 // Do we already have or could we have type information for dest
6270 bool could_have_dest = has_dest;
6271
6272 ciKlass* src_k = nullptr;
6273 if (!has_src) {
6274 src_k = src_type->speculative_type_not_null();
6275 if (src_k != nullptr && src_k->is_array_klass()) {
6276 could_have_src = true;
6277 }
6278 }
6279
6280 ciKlass* dest_k = nullptr;
6281 if (!has_dest) {
6282 dest_k = dest_type->speculative_type_not_null();
6283 if (dest_k != nullptr && dest_k->is_array_klass()) {
6284 could_have_dest = true;
6285 }
6286 }
6287
6288 if (could_have_src && could_have_dest) {
6289 // This is going to pay off so emit the required guards
6290 if (!has_src) {
6291 src = maybe_cast_profiled_obj(src, src_k, true);
6292 src_type = _gvn.type(src);
6293 top_src = src_type->isa_aryptr();
6294 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6295 src_spec = true;
6296 }
6297 if (!has_dest) {
6298 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6299 dest_type = _gvn.type(dest);
6300 top_dest = dest_type->isa_aryptr();
6301 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6302 dest_spec = true;
6303 }
6304 }
6305 }
6306
6307 if (has_src && has_dest && can_emit_guards) {
6308 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
6309 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
6310 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
6311 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
6312
6313 if (src_elem == dest_elem && top_src->is_flat() == top_dest->is_flat() && src_elem == T_OBJECT) {
6314 // If both arrays are object arrays then having the exact types
6315 // for both will remove the need for a subtype check at runtime
6316 // before the call and may make it possible to pick a faster copy
6317 // routine (without a subtype check on every element)
6318 // Do we have the exact type of src?
6319 bool could_have_src = src_spec;
6320 // Do we have the exact type of dest?
6321 bool could_have_dest = dest_spec;
6322 ciKlass* src_k = nullptr;
6323 ciKlass* dest_k = nullptr;
6324 if (!src_spec) {
6325 src_k = src_type->speculative_type_not_null();
6326 if (src_k != nullptr && src_k->is_array_klass()) {
6327 could_have_src = true;
6328 }
6329 }
6330 if (!dest_spec) {
6331 dest_k = dest_type->speculative_type_not_null();
6332 if (dest_k != nullptr && dest_k->is_array_klass()) {
6333 could_have_dest = true;
6334 }
6335 }
6336 if (could_have_src && could_have_dest) {
6337 // If we can have both exact types, emit the missing guards
6338 if (could_have_src && !src_spec) {
6339 src = maybe_cast_profiled_obj(src, src_k, true);
6340 src_type = _gvn.type(src);
6341 top_src = src_type->isa_aryptr();
6342 }
6343 if (could_have_dest && !dest_spec) {
6344 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6345 dest_type = _gvn.type(dest);
6346 top_dest = dest_type->isa_aryptr();
6347 }
6348 }
6349 }
6350 }
6351
6352 ciMethod* trap_method = method();
6353 int trap_bci = bci();
6354 if (saved_jvms_before_guards != nullptr) {
6355 trap_method = alloc->jvms()->method();
6356 trap_bci = alloc->jvms()->bci();
6357 }
6358
6359 bool negative_length_guard_generated = false;
6360
6361 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6362 can_emit_guards && !src->is_top() && !dest->is_top()) {
6363 // validate arguments: enables transformation the ArrayCopyNode
6364 validated = true;
6365
6366 RegionNode* slow_region = new RegionNode(1);
6367 record_for_igvn(slow_region);
6368
6369 // (1) src and dest are arrays.
6370 generate_non_array_guard(load_object_klass(src), slow_region, &src);
6371 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
6372
6373 // (2) src and dest arrays must have elements of the same BasicType
6374 // done at macro expansion or at Ideal transformation time
6375
6376 // (4) src_offset must not be negative.
6377 generate_negative_guard(src_offset, slow_region);
6378
6379 // (5) dest_offset must not be negative.
6380 generate_negative_guard(dest_offset, slow_region);
6381
6382 // (7) src_offset + length must not exceed length of src.
6383 generate_limit_guard(src_offset, length,
6384 load_array_length(src),
6385 slow_region);
6386
6387 // (8) dest_offset + length must not exceed length of dest.
6388 generate_limit_guard(dest_offset, length,
6389 load_array_length(dest),
6390 slow_region);
6391
6392 // (6) length must not be negative.
6393 // This is also checked in generate_arraycopy() during macro expansion, but
6394 // we also have to check it here for the case where the ArrayCopyNode will
6395 // be eliminated by Escape Analysis.
6396 if (EliminateAllocations) {
6397 generate_negative_guard(length, slow_region);
6398 negative_length_guard_generated = true;
6399 }
6400
6401 // (9) each element of an oop array must be assignable
6402 Node* dest_klass = load_object_klass(dest);
6403 if (src != dest) {
6404 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6405 slow_region->add_req(not_subtype_ctrl);
6406 }
6407
6408 // TODO 8350865 Fix below logic. Also handle atomicity.
6409 generate_fair_guard(flat_array_test(src), slow_region);
6410 generate_fair_guard(flat_array_test(dest), slow_region);
6411
6412 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
6413 const Type* toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6414 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6415 src_type = _gvn.type(src);
6416 top_src = src_type->isa_aryptr();
6417
6418 // Handle flat inline type arrays (null-free arrays are handled by the subtype check above)
6419 if (!stopped() && UseArrayFlattening) {
6420 // If dest is flat, src must be flat as well (guaranteed by src <: dest check). Handle flat src here.
6421 assert(top_dest == nullptr || !top_dest->is_flat() || top_src->is_flat(), "src array must be flat");
6422 if (top_src != nullptr && top_src->is_flat()) {
6423 // Src is flat, check that dest is flat as well
6424 if (top_dest != nullptr && !top_dest->is_flat()) {
6425 generate_fair_guard(flat_array_test(dest_klass, /* flat = */ false), slow_region);
6426 // Since dest is flat and src <: dest, dest must have the same type as src.
6427 top_dest = top_src->cast_to_exactness(false);
6428 assert(top_dest->is_flat(), "dest must be flat");
6429 dest = _gvn.transform(new CheckCastPPNode(control(), dest, top_dest));
6430 }
6431 } else if (top_src == nullptr || !top_src->is_not_flat()) {
6432 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
6433 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
6434 assert(top_dest == nullptr || !top_dest->is_flat(), "dest array must not be flat");
6435 generate_fair_guard(flat_array_test(src), slow_region);
6436 if (top_src != nullptr) {
6437 top_src = top_src->cast_to_not_flat();
6438 src = _gvn.transform(new CheckCastPPNode(control(), src, top_src));
6439 }
6440 }
6441 }
6442
6443 {
6444 PreserveJVMState pjvms(this);
6445 set_control(_gvn.transform(slow_region));
6446 uncommon_trap(Deoptimization::Reason_intrinsic,
6447 Deoptimization::Action_make_not_entrant);
6448 assert(stopped(), "Should be stopped");
6449 }
6450 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6451 }
6452
6453 if (stopped()) {
6454 return true;
6455 }
6456
6457 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6458 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6459 // so the compiler has a chance to eliminate them: during macro expansion,
6460 // we have to set their control (CastPP nodes are eliminated).
6461 load_object_klass(src), load_object_klass(dest),
6462 load_array_length(src), load_array_length(dest));
6463
6464 ac->set_arraycopy(validated);
6465
6466 Node* n = _gvn.transform(ac);
6467 if (n == ac) {
6468 ac->connect_outputs(this);
6469 } else {
6470 assert(validated, "shouldn't transform if all arguments not validated");
6471 set_all_memory(n);
6472 }
6473 clear_upper_avx();
6474
6475
6476 return true;
6477 }
6478
6479
6480 // Helper function which determines if an arraycopy immediately follows
6481 // an allocation, with no intervening tests or other escapes for the object.
6482 AllocateArrayNode*
6483 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6484 if (stopped()) return nullptr; // no fast path
6485 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6486
6487 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6488 if (alloc == nullptr) return nullptr;
6489
6490 Node* rawmem = memory(Compile::AliasIdxRaw);
6491 // Is the allocation's memory state untouched?
6492 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6493 // Bail out if there have been raw-memory effects since the allocation.
6494 // (Example: There might have been a call or safepoint.)
6495 return nullptr;
6496 }
6497 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6498 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6499 return nullptr;
6500 }
6501
6502 // There must be no unexpected observers of this allocation.
6503 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6504 Node* obs = ptr->fast_out(i);
6505 if (obs != this->map()) {
6506 return nullptr;
6507 }
6508 }
6509
6510 // This arraycopy must unconditionally follow the allocation of the ptr.
6511 Node* alloc_ctl = ptr->in(0);
6512 Node* ctl = control();
6513 while (ctl != alloc_ctl) {
6514 // There may be guards which feed into the slow_region.
6515 // Any other control flow means that we might not get a chance
6516 // to finish initializing the allocated object.
6517 // Various low-level checks bottom out in uncommon traps. These
6518 // are considered safe since we've already checked above that
6519 // there is no unexpected observer of this allocation.
6520 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6521 assert(ctl->in(0)->is_If(), "must be If");
6522 ctl = ctl->in(0)->in(0);
6523 } else {
6524 return nullptr;
6525 }
6526 }
6527
6528 // If we get this far, we have an allocation which immediately
6529 // precedes the arraycopy, and we can take over zeroing the new object.
6530 // The arraycopy will finish the initialization, and provide
6531 // a new control state to which we will anchor the destination pointer.
6532
6533 return alloc;
6534 }
6535
6536 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6537 if (node->is_IfProj()) {
6538 Node* other_proj = node->as_IfProj()->other_if_proj();
6539 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6540 Node* obs = other_proj->fast_out(j);
6541 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6542 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6543 return obs->as_CallStaticJava();
6544 }
6545 }
6546 }
6547 return nullptr;
6548 }
6549
6550 //-------------inline_encodeISOArray-----------------------------------
6551 // encode char[] to byte[] in ISO_8859_1 or ASCII
6552 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6553 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6554 // no receiver since it is static method
6555 Node *src = argument(0);
6556 Node *src_offset = argument(1);
6557 Node *dst = argument(2);
6558 Node *dst_offset = argument(3);
6559 Node *length = argument(4);
6560
6561 src = must_be_not_null(src, true);
6562 dst = must_be_not_null(dst, true);
6563
6564 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6565 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6566 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6567 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6568 // failed array check
6569 return false;
6570 }
6571
6572 // Figure out the size and type of the elements we will be copying.
6573 BasicType src_elem = src_type->elem()->array_element_basic_type();
6574 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6575 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6576 return false;
6577 }
6578
6579 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6580 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6581 // 'src_start' points to src array + scaled offset
6582 // 'dst_start' points to dst array + scaled offset
6583
6584 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6585 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6586 enc = _gvn.transform(enc);
6587 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6588 set_memory(res_mem, mtype);
6589 set_result(enc);
6590 clear_upper_avx();
6591
6592 return true;
6593 }
6594
6595 //-------------inline_multiplyToLen-----------------------------------
6596 bool LibraryCallKit::inline_multiplyToLen() {
6597 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6598
6599 address stubAddr = StubRoutines::multiplyToLen();
6600 if (stubAddr == nullptr) {
6601 return false; // Intrinsic's stub is not implemented on this platform
6602 }
6603 const char* stubName = "multiplyToLen";
6604
6605 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6606
6607 // no receiver because it is a static method
6608 Node* x = argument(0);
6609 Node* xlen = argument(1);
6610 Node* y = argument(2);
6611 Node* ylen = argument(3);
6612 Node* z = argument(4);
6613
6614 x = must_be_not_null(x, true);
6615 y = must_be_not_null(y, true);
6616
6617 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6618 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6619 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6620 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6621 // failed array check
6622 return false;
6623 }
6624
6625 BasicType x_elem = x_type->elem()->array_element_basic_type();
6626 BasicType y_elem = y_type->elem()->array_element_basic_type();
6627 if (x_elem != T_INT || y_elem != T_INT) {
6628 return false;
6629 }
6630
6631 Node* x_start = array_element_address(x, intcon(0), x_elem);
6632 Node* y_start = array_element_address(y, intcon(0), y_elem);
6633 // 'x_start' points to x array + scaled xlen
6634 // 'y_start' points to y array + scaled ylen
6635
6636 Node* z_start = array_element_address(z, intcon(0), T_INT);
6637
6638 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6639 OptoRuntime::multiplyToLen_Type(),
6640 stubAddr, stubName, TypePtr::BOTTOM,
6641 x_start, xlen, y_start, ylen, z_start);
6642
6643 C->set_has_split_ifs(true); // Has chance for split-if optimization
6644 set_result(z);
6645 return true;
6646 }
6647
6648 //-------------inline_squareToLen------------------------------------
6649 bool LibraryCallKit::inline_squareToLen() {
6650 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6651
6652 address stubAddr = StubRoutines::squareToLen();
6653 if (stubAddr == nullptr) {
6654 return false; // Intrinsic's stub is not implemented on this platform
6655 }
6656 const char* stubName = "squareToLen";
6657
6658 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6659
6660 Node* x = argument(0);
6661 Node* len = argument(1);
6662 Node* z = argument(2);
6663 Node* zlen = argument(3);
6664
6665 x = must_be_not_null(x, true);
6666 z = must_be_not_null(z, true);
6667
6668 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6669 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6670 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6671 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6672 // failed array check
6673 return false;
6674 }
6675
6676 BasicType x_elem = x_type->elem()->array_element_basic_type();
6677 BasicType z_elem = z_type->elem()->array_element_basic_type();
6678 if (x_elem != T_INT || z_elem != T_INT) {
6679 return false;
6680 }
6681
6682
6683 Node* x_start = array_element_address(x, intcon(0), x_elem);
6684 Node* z_start = array_element_address(z, intcon(0), z_elem);
6685
6686 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6687 OptoRuntime::squareToLen_Type(),
6688 stubAddr, stubName, TypePtr::BOTTOM,
6689 x_start, len, z_start, zlen);
6690
6691 set_result(z);
6692 return true;
6693 }
6694
6695 //-------------inline_mulAdd------------------------------------------
6696 bool LibraryCallKit::inline_mulAdd() {
6697 assert(UseMulAddIntrinsic, "not implemented on this platform");
6698
6699 address stubAddr = StubRoutines::mulAdd();
6700 if (stubAddr == nullptr) {
6701 return false; // Intrinsic's stub is not implemented on this platform
6702 }
6703 const char* stubName = "mulAdd";
6704
6705 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6706
6707 Node* out = argument(0);
6708 Node* in = argument(1);
6709 Node* offset = argument(2);
6710 Node* len = argument(3);
6711 Node* k = argument(4);
6712
6713 in = must_be_not_null(in, true);
6714 out = must_be_not_null(out, true);
6715
6716 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6717 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6718 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6719 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6720 // failed array check
6721 return false;
6722 }
6723
6724 BasicType out_elem = out_type->elem()->array_element_basic_type();
6725 BasicType in_elem = in_type->elem()->array_element_basic_type();
6726 if (out_elem != T_INT || in_elem != T_INT) {
6727 return false;
6728 }
6729
6730 Node* outlen = load_array_length(out);
6731 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6732 Node* out_start = array_element_address(out, intcon(0), out_elem);
6733 Node* in_start = array_element_address(in, intcon(0), in_elem);
6734
6735 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6736 OptoRuntime::mulAdd_Type(),
6737 stubAddr, stubName, TypePtr::BOTTOM,
6738 out_start,in_start, new_offset, len, k);
6739 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6740 set_result(result);
6741 return true;
6742 }
6743
6744 //-------------inline_montgomeryMultiply-----------------------------------
6745 bool LibraryCallKit::inline_montgomeryMultiply() {
6746 address stubAddr = StubRoutines::montgomeryMultiply();
6747 if (stubAddr == nullptr) {
6748 return false; // Intrinsic's stub is not implemented on this platform
6749 }
6750
6751 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6752 const char* stubName = "montgomery_multiply";
6753
6754 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6755
6756 Node* a = argument(0);
6757 Node* b = argument(1);
6758 Node* n = argument(2);
6759 Node* len = argument(3);
6760 Node* inv = argument(4);
6761 Node* m = argument(6);
6762
6763 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6764 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6765 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6766 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6767 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6768 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6769 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6770 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6771 // failed array check
6772 return false;
6773 }
6774
6775 BasicType a_elem = a_type->elem()->array_element_basic_type();
6776 BasicType b_elem = b_type->elem()->array_element_basic_type();
6777 BasicType n_elem = n_type->elem()->array_element_basic_type();
6778 BasicType m_elem = m_type->elem()->array_element_basic_type();
6779 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6780 return false;
6781 }
6782
6783 // Make the call
6784 {
6785 Node* a_start = array_element_address(a, intcon(0), a_elem);
6786 Node* b_start = array_element_address(b, intcon(0), b_elem);
6787 Node* n_start = array_element_address(n, intcon(0), n_elem);
6788 Node* m_start = array_element_address(m, intcon(0), m_elem);
6789
6790 Node* call = make_runtime_call(RC_LEAF,
6791 OptoRuntime::montgomeryMultiply_Type(),
6792 stubAddr, stubName, TypePtr::BOTTOM,
6793 a_start, b_start, n_start, len, inv, top(),
6794 m_start);
6795 set_result(m);
6796 }
6797
6798 return true;
6799 }
6800
6801 bool LibraryCallKit::inline_montgomerySquare() {
6802 address stubAddr = StubRoutines::montgomerySquare();
6803 if (stubAddr == nullptr) {
6804 return false; // Intrinsic's stub is not implemented on this platform
6805 }
6806
6807 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6808 const char* stubName = "montgomery_square";
6809
6810 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6811
6812 Node* a = argument(0);
6813 Node* n = argument(1);
6814 Node* len = argument(2);
6815 Node* inv = argument(3);
6816 Node* m = argument(5);
6817
6818 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6819 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6820 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6821 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6822 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6823 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6824 // failed array check
6825 return false;
6826 }
6827
6828 BasicType a_elem = a_type->elem()->array_element_basic_type();
6829 BasicType n_elem = n_type->elem()->array_element_basic_type();
6830 BasicType m_elem = m_type->elem()->array_element_basic_type();
6831 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6832 return false;
6833 }
6834
6835 // Make the call
6836 {
6837 Node* a_start = array_element_address(a, intcon(0), a_elem);
6838 Node* n_start = array_element_address(n, intcon(0), n_elem);
6839 Node* m_start = array_element_address(m, intcon(0), m_elem);
6840
6841 Node* call = make_runtime_call(RC_LEAF,
6842 OptoRuntime::montgomerySquare_Type(),
6843 stubAddr, stubName, TypePtr::BOTTOM,
6844 a_start, n_start, len, inv, top(),
6845 m_start);
6846 set_result(m);
6847 }
6848
6849 return true;
6850 }
6851
6852 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6853 address stubAddr = nullptr;
6854 const char* stubName = nullptr;
6855
6856 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6857 if (stubAddr == nullptr) {
6858 return false; // Intrinsic's stub is not implemented on this platform
6859 }
6860
6861 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6862
6863 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6864
6865 Node* newArr = argument(0);
6866 Node* oldArr = argument(1);
6867 Node* newIdx = argument(2);
6868 Node* shiftCount = argument(3);
6869 Node* numIter = argument(4);
6870
6871 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6872 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6873 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6874 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6875 return false;
6876 }
6877
6878 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6879 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6880 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6881 return false;
6882 }
6883
6884 // Make the call
6885 {
6886 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6887 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6888
6889 Node* call = make_runtime_call(RC_LEAF,
6890 OptoRuntime::bigIntegerShift_Type(),
6891 stubAddr,
6892 stubName,
6893 TypePtr::BOTTOM,
6894 newArr_start,
6895 oldArr_start,
6896 newIdx,
6897 shiftCount,
6898 numIter);
6899 }
6900
6901 return true;
6902 }
6903
6904 //-------------inline_vectorizedMismatch------------------------------
6905 bool LibraryCallKit::inline_vectorizedMismatch() {
6906 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6907
6908 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6909 Node* obja = argument(0); // Object
6910 Node* aoffset = argument(1); // long
6911 Node* objb = argument(3); // Object
6912 Node* boffset = argument(4); // long
6913 Node* length = argument(6); // int
6914 Node* scale = argument(7); // int
6915
6916 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6917 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6918 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6919 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6920 scale == top()) {
6921 return false; // failed input validation
6922 }
6923
6924 Node* obja_adr = make_unsafe_address(obja, aoffset);
6925 Node* objb_adr = make_unsafe_address(objb, boffset);
6926
6927 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6928 //
6929 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6930 // if (length <= inline_limit) {
6931 // inline_path:
6932 // vmask = VectorMaskGen length
6933 // vload1 = LoadVectorMasked obja, vmask
6934 // vload2 = LoadVectorMasked objb, vmask
6935 // result1 = VectorCmpMasked vload1, vload2, vmask
6936 // } else {
6937 // call_stub_path:
6938 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6939 // }
6940 // exit_block:
6941 // return Phi(result1, result2);
6942 //
6943 enum { inline_path = 1, // input is small enough to process it all at once
6944 stub_path = 2, // input is too large; call into the VM
6945 PATH_LIMIT = 3
6946 };
6947
6948 Node* exit_block = new RegionNode(PATH_LIMIT);
6949 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6950 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6951
6952 Node* call_stub_path = control();
6953
6954 BasicType elem_bt = T_ILLEGAL;
6955
6956 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6957 if (scale_t->is_con()) {
6958 switch (scale_t->get_con()) {
6959 case 0: elem_bt = T_BYTE; break;
6960 case 1: elem_bt = T_SHORT; break;
6961 case 2: elem_bt = T_INT; break;
6962 case 3: elem_bt = T_LONG; break;
6963
6964 default: elem_bt = T_ILLEGAL; break; // not supported
6965 }
6966 }
6967
6968 int inline_limit = 0;
6969 bool do_partial_inline = false;
6970
6971 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6972 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6973 do_partial_inline = inline_limit >= 16;
6974 }
6975
6976 if (do_partial_inline) {
6977 assert(elem_bt != T_ILLEGAL, "sanity");
6978
6979 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6980 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6981 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6982
6983 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6984 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6985 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6986
6987 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6988
6989 if (!stopped()) {
6990 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6991
6992 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6993 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6994 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6995 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6996
6997 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6998 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6999 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
7000 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
7001
7002 exit_block->init_req(inline_path, control());
7003 memory_phi->init_req(inline_path, map()->memory());
7004 result_phi->init_req(inline_path, result);
7005
7006 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
7007 clear_upper_avx();
7008 }
7009 }
7010 }
7011
7012 if (call_stub_path != nullptr) {
7013 set_control(call_stub_path);
7014
7015 Node* call = make_runtime_call(RC_LEAF,
7016 OptoRuntime::vectorizedMismatch_Type(),
7017 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
7018 obja_adr, objb_adr, length, scale);
7019
7020 exit_block->init_req(stub_path, control());
7021 memory_phi->init_req(stub_path, map()->memory());
7022 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
7023 }
7024
7025 exit_block = _gvn.transform(exit_block);
7026 memory_phi = _gvn.transform(memory_phi);
7027 result_phi = _gvn.transform(result_phi);
7028
7029 set_control(exit_block);
7030 set_all_memory(memory_phi);
7031 set_result(result_phi);
7032
7033 return true;
7034 }
7035
7036 //------------------------------inline_vectorizedHashcode----------------------------
7037 bool LibraryCallKit::inline_vectorizedHashCode() {
7038 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
7039
7040 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
7041 Node* array = argument(0);
7042 Node* offset = argument(1);
7043 Node* length = argument(2);
7044 Node* initialValue = argument(3);
7045 Node* basic_type = argument(4);
7046
7047 if (basic_type == top()) {
7048 return false; // failed input validation
7049 }
7050
7051 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
7052 if (!basic_type_t->is_con()) {
7053 return false; // Only intrinsify if mode argument is constant
7054 }
7055
7056 array = must_be_not_null(array, true);
7057
7058 BasicType bt = (BasicType)basic_type_t->get_con();
7059
7060 // Resolve address of first element
7061 Node* array_start = array_element_address(array, offset, bt);
7062
7063 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
7064 array_start, length, initialValue, basic_type)));
7065 clear_upper_avx();
7066
7067 return true;
7068 }
7069
7070 /**
7071 * Calculate CRC32 for byte.
7072 * int java.util.zip.CRC32.update(int crc, int b)
7073 */
7074 bool LibraryCallKit::inline_updateCRC32() {
7075 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7076 assert(callee()->signature()->size() == 2, "update has 2 parameters");
7077 // no receiver since it is static method
7078 Node* crc = argument(0); // type: int
7079 Node* b = argument(1); // type: int
7080
7081 /*
7082 * int c = ~ crc;
7083 * b = timesXtoThe32[(b ^ c) & 0xFF];
7084 * b = b ^ (c >>> 8);
7085 * crc = ~b;
7086 */
7087
7088 Node* M1 = intcon(-1);
7089 crc = _gvn.transform(new XorINode(crc, M1));
7090 Node* result = _gvn.transform(new XorINode(crc, b));
7091 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
7092
7093 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
7094 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
7095 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
7096 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
7097
7098 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
7099 result = _gvn.transform(new XorINode(crc, result));
7100 result = _gvn.transform(new XorINode(result, M1));
7101 set_result(result);
7102 return true;
7103 }
7104
7105 /**
7106 * Calculate CRC32 for byte[] array.
7107 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
7108 */
7109 bool LibraryCallKit::inline_updateBytesCRC32() {
7110 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7111 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7112 // no receiver since it is static method
7113 Node* crc = argument(0); // type: int
7114 Node* src = argument(1); // type: oop
7115 Node* offset = argument(2); // type: int
7116 Node* length = argument(3); // type: int
7117
7118 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7119 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7120 // failed array check
7121 return false;
7122 }
7123
7124 // Figure out the size and type of the elements we will be copying.
7125 BasicType src_elem = src_type->elem()->array_element_basic_type();
7126 if (src_elem != T_BYTE) {
7127 return false;
7128 }
7129
7130 // 'src_start' points to src array + scaled offset
7131 src = must_be_not_null(src, true);
7132 Node* src_start = array_element_address(src, offset, src_elem);
7133
7134 // We assume that range check is done by caller.
7135 // TODO: generate range check (offset+length < src.length) in debug VM.
7136
7137 // Call the stub.
7138 address stubAddr = StubRoutines::updateBytesCRC32();
7139 const char *stubName = "updateBytesCRC32";
7140
7141 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7142 stubAddr, stubName, TypePtr::BOTTOM,
7143 crc, src_start, length);
7144 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7145 set_result(result);
7146 return true;
7147 }
7148
7149 /**
7150 * Calculate CRC32 for ByteBuffer.
7151 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
7152 */
7153 bool LibraryCallKit::inline_updateByteBufferCRC32() {
7154 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7155 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7156 // no receiver since it is static method
7157 Node* crc = argument(0); // type: int
7158 Node* src = argument(1); // type: long
7159 Node* offset = argument(3); // type: int
7160 Node* length = argument(4); // type: int
7161
7162 src = ConvL2X(src); // adjust Java long to machine word
7163 Node* base = _gvn.transform(new CastX2PNode(src));
7164 offset = ConvI2X(offset);
7165
7166 // 'src_start' points to src array + scaled offset
7167 Node* src_start = basic_plus_adr(top(), base, offset);
7168
7169 // Call the stub.
7170 address stubAddr = StubRoutines::updateBytesCRC32();
7171 const char *stubName = "updateBytesCRC32";
7172
7173 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7174 stubAddr, stubName, TypePtr::BOTTOM,
7175 crc, src_start, length);
7176 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7177 set_result(result);
7178 return true;
7179 }
7180
7181 //------------------------------get_table_from_crc32c_class-----------------------
7182 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
7183 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
7184 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
7185
7186 return table;
7187 }
7188
7189 //------------------------------inline_updateBytesCRC32C-----------------------
7190 //
7191 // Calculate CRC32C for byte[] array.
7192 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
7193 //
7194 bool LibraryCallKit::inline_updateBytesCRC32C() {
7195 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7196 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7197 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7198 // no receiver since it is a static method
7199 Node* crc = argument(0); // type: int
7200 Node* src = argument(1); // type: oop
7201 Node* offset = argument(2); // type: int
7202 Node* end = argument(3); // type: int
7203
7204 Node* length = _gvn.transform(new SubINode(end, offset));
7205
7206 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7207 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7208 // failed array check
7209 return false;
7210 }
7211
7212 // Figure out the size and type of the elements we will be copying.
7213 BasicType src_elem = src_type->elem()->array_element_basic_type();
7214 if (src_elem != T_BYTE) {
7215 return false;
7216 }
7217
7218 // 'src_start' points to src array + scaled offset
7219 src = must_be_not_null(src, true);
7220 Node* src_start = array_element_address(src, offset, src_elem);
7221
7222 // static final int[] byteTable in class CRC32C
7223 Node* table = get_table_from_crc32c_class(callee()->holder());
7224 table = must_be_not_null(table, true);
7225 Node* table_start = array_element_address(table, intcon(0), T_INT);
7226
7227 // We assume that range check is done by caller.
7228 // TODO: generate range check (offset+length < src.length) in debug VM.
7229
7230 // Call the stub.
7231 address stubAddr = StubRoutines::updateBytesCRC32C();
7232 const char *stubName = "updateBytesCRC32C";
7233
7234 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7235 stubAddr, stubName, TypePtr::BOTTOM,
7236 crc, src_start, length, table_start);
7237 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7238 set_result(result);
7239 return true;
7240 }
7241
7242 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
7243 //
7244 // Calculate CRC32C for DirectByteBuffer.
7245 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
7246 //
7247 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
7248 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7249 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
7250 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7251 // no receiver since it is a static method
7252 Node* crc = argument(0); // type: int
7253 Node* src = argument(1); // type: long
7254 Node* offset = argument(3); // type: int
7255 Node* end = argument(4); // type: int
7256
7257 Node* length = _gvn.transform(new SubINode(end, offset));
7258
7259 src = ConvL2X(src); // adjust Java long to machine word
7260 Node* base = _gvn.transform(new CastX2PNode(src));
7261 offset = ConvI2X(offset);
7262
7263 // 'src_start' points to src array + scaled offset
7264 Node* src_start = basic_plus_adr(top(), base, offset);
7265
7266 // static final int[] byteTable in class CRC32C
7267 Node* table = get_table_from_crc32c_class(callee()->holder());
7268 table = must_be_not_null(table, true);
7269 Node* table_start = array_element_address(table, intcon(0), T_INT);
7270
7271 // Call the stub.
7272 address stubAddr = StubRoutines::updateBytesCRC32C();
7273 const char *stubName = "updateBytesCRC32C";
7274
7275 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7276 stubAddr, stubName, TypePtr::BOTTOM,
7277 crc, src_start, length, table_start);
7278 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7279 set_result(result);
7280 return true;
7281 }
7282
7283 //------------------------------inline_updateBytesAdler32----------------------
7284 //
7285 // Calculate Adler32 checksum for byte[] array.
7286 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
7287 //
7288 bool LibraryCallKit::inline_updateBytesAdler32() {
7289 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7290 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7291 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7292 // no receiver since it is static method
7293 Node* crc = argument(0); // type: int
7294 Node* src = argument(1); // type: oop
7295 Node* offset = argument(2); // type: int
7296 Node* length = argument(3); // type: int
7297
7298 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7299 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7300 // failed array check
7301 return false;
7302 }
7303
7304 // Figure out the size and type of the elements we will be copying.
7305 BasicType src_elem = src_type->elem()->array_element_basic_type();
7306 if (src_elem != T_BYTE) {
7307 return false;
7308 }
7309
7310 // 'src_start' points to src array + scaled offset
7311 Node* src_start = array_element_address(src, offset, src_elem);
7312
7313 // We assume that range check is done by caller.
7314 // TODO: generate range check (offset+length < src.length) in debug VM.
7315
7316 // Call the stub.
7317 address stubAddr = StubRoutines::updateBytesAdler32();
7318 const char *stubName = "updateBytesAdler32";
7319
7320 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7321 stubAddr, stubName, TypePtr::BOTTOM,
7322 crc, src_start, length);
7323 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7324 set_result(result);
7325 return true;
7326 }
7327
7328 //------------------------------inline_updateByteBufferAdler32---------------
7329 //
7330 // Calculate Adler32 checksum for DirectByteBuffer.
7331 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
7332 //
7333 bool LibraryCallKit::inline_updateByteBufferAdler32() {
7334 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7335 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7336 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7337 // no receiver since it is static method
7338 Node* crc = argument(0); // type: int
7339 Node* src = argument(1); // type: long
7340 Node* offset = argument(3); // type: int
7341 Node* length = argument(4); // type: int
7342
7343 src = ConvL2X(src); // adjust Java long to machine word
7344 Node* base = _gvn.transform(new CastX2PNode(src));
7345 offset = ConvI2X(offset);
7346
7347 // 'src_start' points to src array + scaled offset
7348 Node* src_start = basic_plus_adr(top(), base, offset);
7349
7350 // Call the stub.
7351 address stubAddr = StubRoutines::updateBytesAdler32();
7352 const char *stubName = "updateBytesAdler32";
7353
7354 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7355 stubAddr, stubName, TypePtr::BOTTOM,
7356 crc, src_start, length);
7357
7358 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7359 set_result(result);
7360 return true;
7361 }
7362
7363 //----------------------------inline_reference_get----------------------------
7364 // public T java.lang.ref.Reference.get();
7365 bool LibraryCallKit::inline_reference_get() {
7366 const int referent_offset = java_lang_ref_Reference::referent_offset();
7367
7368 // Get the argument:
7369 Node* reference_obj = null_check_receiver();
7370 if (stopped()) return true;
7371
7372 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7373 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7374 decorators, /*is_static*/ false, nullptr);
7375 if (result == nullptr) return false;
7376
7377 // Add memory barrier to prevent commoning reads from this field
7378 // across safepoint since GC can change its value.
7379 insert_mem_bar(Op_MemBarCPUOrder);
7380
7381 set_result(result);
7382 return true;
7383 }
7384
7385 //----------------------------inline_reference_refersTo0----------------------------
7386 // bool java.lang.ref.Reference.refersTo0();
7387 // bool java.lang.ref.PhantomReference.refersTo0();
7388 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7389 // Get arguments:
7390 Node* reference_obj = null_check_receiver();
7391 Node* other_obj = argument(1);
7392 if (stopped()) return true;
7393
7394 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7395 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7396 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7397 decorators, /*is_static*/ false, nullptr);
7398 if (referent == nullptr) return false;
7399
7400 // Add memory barrier to prevent commoning reads from this field
7401 // across safepoint since GC can change its value.
7402 insert_mem_bar(Op_MemBarCPUOrder);
7403
7404 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7405 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7406 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7407
7408 RegionNode* region = new RegionNode(3);
7409 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7410
7411 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7412 region->init_req(1, if_true);
7413 phi->init_req(1, intcon(1));
7414
7415 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7416 region->init_req(2, if_false);
7417 phi->init_req(2, intcon(0));
7418
7419 set_control(_gvn.transform(region));
7420 record_for_igvn(region);
7421 set_result(_gvn.transform(phi));
7422 return true;
7423 }
7424
7425 //----------------------------inline_reference_clear0----------------------------
7426 // void java.lang.ref.Reference.clear0();
7427 // void java.lang.ref.PhantomReference.clear0();
7428 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7429 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7430
7431 // Get arguments
7432 Node* reference_obj = null_check_receiver();
7433 if (stopped()) return true;
7434
7435 // Common access parameters
7436 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7437 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7438 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7439 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7440 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7441
7442 Node* referent = access_load_at(reference_obj,
7443 referent_field_addr,
7444 referent_field_addr_type,
7445 val_type,
7446 T_OBJECT,
7447 decorators);
7448
7449 IdealKit ideal(this);
7450 #define __ ideal.
7451 __ if_then(referent, BoolTest::ne, null());
7452 sync_kit(ideal);
7453 access_store_at(reference_obj,
7454 referent_field_addr,
7455 referent_field_addr_type,
7456 null(),
7457 val_type,
7458 T_OBJECT,
7459 decorators);
7460 __ sync_kit(this);
7461 __ end_if();
7462 final_sync(ideal);
7463 #undef __
7464
7465 return true;
7466 }
7467
7468 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7469 DecoratorSet decorators, bool is_static,
7470 ciInstanceKlass* fromKls) {
7471 if (fromKls == nullptr) {
7472 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7473 assert(tinst != nullptr, "obj is null");
7474 assert(tinst->is_loaded(), "obj is not loaded");
7475 fromKls = tinst->instance_klass();
7476 } else {
7477 assert(is_static, "only for static field access");
7478 }
7479 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7480 ciSymbol::make(fieldTypeString),
7481 is_static);
7482
7483 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7484 if (field == nullptr) return (Node *) nullptr;
7485
7486 if (is_static) {
7487 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7488 fromObj = makecon(tip);
7489 }
7490
7491 // Next code copied from Parse::do_get_xxx():
7492
7493 // Compute address and memory type.
7494 int offset = field->offset_in_bytes();
7495 bool is_vol = field->is_volatile();
7496 ciType* field_klass = field->type();
7497 assert(field_klass->is_loaded(), "should be loaded");
7498 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7499 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7500 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7501 "slice of address and input slice don't match");
7502 BasicType bt = field->layout_type();
7503
7504 // Build the resultant type of the load
7505 const Type *type;
7506 if (bt == T_OBJECT) {
7507 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7508 } else {
7509 type = Type::get_const_basic_type(bt);
7510 }
7511
7512 if (is_vol) {
7513 decorators |= MO_SEQ_CST;
7514 }
7515
7516 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7517 }
7518
7519 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7520 bool is_exact /* true */, bool is_static /* false */,
7521 ciInstanceKlass * fromKls /* nullptr */) {
7522 if (fromKls == nullptr) {
7523 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7524 assert(tinst != nullptr, "obj is null");
7525 assert(tinst->is_loaded(), "obj is not loaded");
7526 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7527 fromKls = tinst->instance_klass();
7528 }
7529 else {
7530 assert(is_static, "only for static field access");
7531 }
7532 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7533 ciSymbol::make(fieldTypeString),
7534 is_static);
7535
7536 assert(field != nullptr, "undefined field");
7537 assert(!field->is_volatile(), "not defined for volatile fields");
7538
7539 if (is_static) {
7540 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7541 fromObj = makecon(tip);
7542 }
7543
7544 // Next code copied from Parse::do_get_xxx():
7545
7546 // Compute address and memory type.
7547 int offset = field->offset_in_bytes();
7548 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7549
7550 return adr;
7551 }
7552
7553 //------------------------------inline_aescrypt_Block-----------------------
7554 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7555 address stubAddr = nullptr;
7556 const char *stubName;
7557 assert(UseAES, "need AES instruction support");
7558
7559 switch(id) {
7560 case vmIntrinsics::_aescrypt_encryptBlock:
7561 stubAddr = StubRoutines::aescrypt_encryptBlock();
7562 stubName = "aescrypt_encryptBlock";
7563 break;
7564 case vmIntrinsics::_aescrypt_decryptBlock:
7565 stubAddr = StubRoutines::aescrypt_decryptBlock();
7566 stubName = "aescrypt_decryptBlock";
7567 break;
7568 default:
7569 break;
7570 }
7571 if (stubAddr == nullptr) return false;
7572
7573 Node* aescrypt_object = argument(0);
7574 Node* src = argument(1);
7575 Node* src_offset = argument(2);
7576 Node* dest = argument(3);
7577 Node* dest_offset = argument(4);
7578
7579 src = must_be_not_null(src, true);
7580 dest = must_be_not_null(dest, true);
7581
7582 // (1) src and dest are arrays.
7583 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7584 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7585 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7586 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7587
7588 // for the quick and dirty code we will skip all the checks.
7589 // we are just trying to get the call to be generated.
7590 Node* src_start = src;
7591 Node* dest_start = dest;
7592 if (src_offset != nullptr || dest_offset != nullptr) {
7593 assert(src_offset != nullptr && dest_offset != nullptr, "");
7594 src_start = array_element_address(src, src_offset, T_BYTE);
7595 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7596 }
7597
7598 // now need to get the start of its expanded key array
7599 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7600 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7601 if (k_start == nullptr) return false;
7602
7603 // Call the stub.
7604 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7605 stubAddr, stubName, TypePtr::BOTTOM,
7606 src_start, dest_start, k_start);
7607
7608 return true;
7609 }
7610
7611 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7612 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7613 address stubAddr = nullptr;
7614 const char *stubName = nullptr;
7615
7616 assert(UseAES, "need AES instruction support");
7617
7618 switch(id) {
7619 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7620 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7621 stubName = "cipherBlockChaining_encryptAESCrypt";
7622 break;
7623 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7624 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7625 stubName = "cipherBlockChaining_decryptAESCrypt";
7626 break;
7627 default:
7628 break;
7629 }
7630 if (stubAddr == nullptr) return false;
7631
7632 Node* cipherBlockChaining_object = argument(0);
7633 Node* src = argument(1);
7634 Node* src_offset = argument(2);
7635 Node* len = argument(3);
7636 Node* dest = argument(4);
7637 Node* dest_offset = argument(5);
7638
7639 src = must_be_not_null(src, false);
7640 dest = must_be_not_null(dest, false);
7641
7642 // (1) src and dest are arrays.
7643 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7644 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7645 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7646 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7647
7648 // checks are the responsibility of the caller
7649 Node* src_start = src;
7650 Node* dest_start = dest;
7651 if (src_offset != nullptr || dest_offset != nullptr) {
7652 assert(src_offset != nullptr && dest_offset != nullptr, "");
7653 src_start = array_element_address(src, src_offset, T_BYTE);
7654 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7655 }
7656
7657 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7658 // (because of the predicated logic executed earlier).
7659 // so we cast it here safely.
7660 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7661
7662 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7663 if (embeddedCipherObj == nullptr) return false;
7664
7665 // cast it to what we know it will be at runtime
7666 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7667 assert(tinst != nullptr, "CBC obj is null");
7668 assert(tinst->is_loaded(), "CBC obj is not loaded");
7669 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7670 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7671
7672 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7673 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7674 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7675 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7676 aescrypt_object = _gvn.transform(aescrypt_object);
7677
7678 // we need to get the start of the aescrypt_object's expanded key array
7679 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7680 if (k_start == nullptr) return false;
7681
7682 // similarly, get the start address of the r vector
7683 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7684 if (objRvec == nullptr) return false;
7685 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7686
7687 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7688 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7689 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7690 stubAddr, stubName, TypePtr::BOTTOM,
7691 src_start, dest_start, k_start, r_start, len);
7692
7693 // return cipher length (int)
7694 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7695 set_result(retvalue);
7696 return true;
7697 }
7698
7699 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7700 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7701 address stubAddr = nullptr;
7702 const char *stubName = nullptr;
7703
7704 assert(UseAES, "need AES instruction support");
7705
7706 switch (id) {
7707 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7708 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7709 stubName = "electronicCodeBook_encryptAESCrypt";
7710 break;
7711 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7712 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7713 stubName = "electronicCodeBook_decryptAESCrypt";
7714 break;
7715 default:
7716 break;
7717 }
7718
7719 if (stubAddr == nullptr) return false;
7720
7721 Node* electronicCodeBook_object = argument(0);
7722 Node* src = argument(1);
7723 Node* src_offset = argument(2);
7724 Node* len = argument(3);
7725 Node* dest = argument(4);
7726 Node* dest_offset = argument(5);
7727
7728 // (1) src and dest are arrays.
7729 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7730 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7731 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7732 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7733
7734 // checks are the responsibility of the caller
7735 Node* src_start = src;
7736 Node* dest_start = dest;
7737 if (src_offset != nullptr || dest_offset != nullptr) {
7738 assert(src_offset != nullptr && dest_offset != nullptr, "");
7739 src_start = array_element_address(src, src_offset, T_BYTE);
7740 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7741 }
7742
7743 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7744 // (because of the predicated logic executed earlier).
7745 // so we cast it here safely.
7746 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7747
7748 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7749 if (embeddedCipherObj == nullptr) return false;
7750
7751 // cast it to what we know it will be at runtime
7752 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7753 assert(tinst != nullptr, "ECB obj is null");
7754 assert(tinst->is_loaded(), "ECB obj is not loaded");
7755 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7756 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7757
7758 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7759 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7760 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7761 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7762 aescrypt_object = _gvn.transform(aescrypt_object);
7763
7764 // we need to get the start of the aescrypt_object's expanded key array
7765 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7766 if (k_start == nullptr) return false;
7767
7768 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7769 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7770 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7771 stubAddr, stubName, TypePtr::BOTTOM,
7772 src_start, dest_start, k_start, len);
7773
7774 // return cipher length (int)
7775 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7776 set_result(retvalue);
7777 return true;
7778 }
7779
7780 //------------------------------inline_counterMode_AESCrypt-----------------------
7781 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7782 assert(UseAES, "need AES instruction support");
7783 if (!UseAESCTRIntrinsics) return false;
7784
7785 address stubAddr = nullptr;
7786 const char *stubName = nullptr;
7787 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7788 stubAddr = StubRoutines::counterMode_AESCrypt();
7789 stubName = "counterMode_AESCrypt";
7790 }
7791 if (stubAddr == nullptr) return false;
7792
7793 Node* counterMode_object = argument(0);
7794 Node* src = argument(1);
7795 Node* src_offset = argument(2);
7796 Node* len = argument(3);
7797 Node* dest = argument(4);
7798 Node* dest_offset = argument(5);
7799
7800 // (1) src and dest are arrays.
7801 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7802 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7803 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7804 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7805
7806 // checks are the responsibility of the caller
7807 Node* src_start = src;
7808 Node* dest_start = dest;
7809 if (src_offset != nullptr || dest_offset != nullptr) {
7810 assert(src_offset != nullptr && dest_offset != nullptr, "");
7811 src_start = array_element_address(src, src_offset, T_BYTE);
7812 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7813 }
7814
7815 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7816 // (because of the predicated logic executed earlier).
7817 // so we cast it here safely.
7818 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7819 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7820 if (embeddedCipherObj == nullptr) return false;
7821 // cast it to what we know it will be at runtime
7822 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7823 assert(tinst != nullptr, "CTR obj is null");
7824 assert(tinst->is_loaded(), "CTR obj is not loaded");
7825 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7826 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7827 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7828 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7829 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7830 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7831 aescrypt_object = _gvn.transform(aescrypt_object);
7832 // we need to get the start of the aescrypt_object's expanded key array
7833 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7834 if (k_start == nullptr) return false;
7835 // similarly, get the start address of the r vector
7836 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7837 if (obj_counter == nullptr) return false;
7838 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7839
7840 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7841 if (saved_encCounter == nullptr) return false;
7842 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7843 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7844
7845 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7846 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7847 OptoRuntime::counterMode_aescrypt_Type(),
7848 stubAddr, stubName, TypePtr::BOTTOM,
7849 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7850
7851 // return cipher length (int)
7852 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7853 set_result(retvalue);
7854 return true;
7855 }
7856
7857 //------------------------------get_key_start_from_aescrypt_object-----------------------
7858 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7859 #if defined(PPC64) || defined(S390) || defined(RISCV64)
7860 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7861 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7862 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7863 // The ppc64 and riscv64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7864 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7865 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7866 if (objSessionK == nullptr) {
7867 return (Node *) nullptr;
7868 }
7869 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7870 #else
7871 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7872 #endif // PPC64
7873 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7874 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7875
7876 // now have the array, need to get the start address of the K array
7877 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7878 return k_start;
7879 }
7880
7881 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7882 // Return node representing slow path of predicate check.
7883 // the pseudo code we want to emulate with this predicate is:
7884 // for encryption:
7885 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7886 // for decryption:
7887 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7888 // note cipher==plain is more conservative than the original java code but that's OK
7889 //
7890 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7891 // The receiver was checked for null already.
7892 Node* objCBC = argument(0);
7893
7894 Node* src = argument(1);
7895 Node* dest = argument(4);
7896
7897 // Load embeddedCipher field of CipherBlockChaining object.
7898 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7899
7900 // get AESCrypt klass for instanceOf check
7901 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7902 // will have same classloader as CipherBlockChaining object
7903 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7904 assert(tinst != nullptr, "CBCobj is null");
7905 assert(tinst->is_loaded(), "CBCobj is not loaded");
7906
7907 // we want to do an instanceof comparison against the AESCrypt class
7908 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7909 if (!klass_AESCrypt->is_loaded()) {
7910 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7911 Node* ctrl = control();
7912 set_control(top()); // no regular fast path
7913 return ctrl;
7914 }
7915
7916 src = must_be_not_null(src, true);
7917 dest = must_be_not_null(dest, true);
7918
7919 // Resolve oops to stable for CmpP below.
7920 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7921
7922 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7923 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7924 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7925
7926 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7927
7928 // for encryption, we are done
7929 if (!decrypting)
7930 return instof_false; // even if it is null
7931
7932 // for decryption, we need to add a further check to avoid
7933 // taking the intrinsic path when cipher and plain are the same
7934 // see the original java code for why.
7935 RegionNode* region = new RegionNode(3);
7936 region->init_req(1, instof_false);
7937
7938 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7939 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7940 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7941 region->init_req(2, src_dest_conjoint);
7942
7943 record_for_igvn(region);
7944 return _gvn.transform(region);
7945 }
7946
7947 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7948 // Return node representing slow path of predicate check.
7949 // the pseudo code we want to emulate with this predicate is:
7950 // for encryption:
7951 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7952 // for decryption:
7953 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7954 // note cipher==plain is more conservative than the original java code but that's OK
7955 //
7956 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7957 // The receiver was checked for null already.
7958 Node* objECB = argument(0);
7959
7960 // Load embeddedCipher field of ElectronicCodeBook object.
7961 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7962
7963 // get AESCrypt klass for instanceOf check
7964 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7965 // will have same classloader as ElectronicCodeBook object
7966 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7967 assert(tinst != nullptr, "ECBobj is null");
7968 assert(tinst->is_loaded(), "ECBobj is not loaded");
7969
7970 // we want to do an instanceof comparison against the AESCrypt class
7971 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7972 if (!klass_AESCrypt->is_loaded()) {
7973 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7974 Node* ctrl = control();
7975 set_control(top()); // no regular fast path
7976 return ctrl;
7977 }
7978 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7979
7980 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7981 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7982 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7983
7984 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7985
7986 // for encryption, we are done
7987 if (!decrypting)
7988 return instof_false; // even if it is null
7989
7990 // for decryption, we need to add a further check to avoid
7991 // taking the intrinsic path when cipher and plain are the same
7992 // see the original java code for why.
7993 RegionNode* region = new RegionNode(3);
7994 region->init_req(1, instof_false);
7995 Node* src = argument(1);
7996 Node* dest = argument(4);
7997 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7998 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7999 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8000 region->init_req(2, src_dest_conjoint);
8001
8002 record_for_igvn(region);
8003 return _gvn.transform(region);
8004 }
8005
8006 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
8007 // Return node representing slow path of predicate check.
8008 // the pseudo code we want to emulate with this predicate is:
8009 // for encryption:
8010 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8011 // for decryption:
8012 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8013 // note cipher==plain is more conservative than the original java code but that's OK
8014 //
8015
8016 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
8017 // The receiver was checked for null already.
8018 Node* objCTR = argument(0);
8019
8020 // Load embeddedCipher field of CipherBlockChaining object.
8021 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8022
8023 // get AESCrypt klass for instanceOf check
8024 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8025 // will have same classloader as CipherBlockChaining object
8026 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
8027 assert(tinst != nullptr, "CTRobj is null");
8028 assert(tinst->is_loaded(), "CTRobj is not loaded");
8029
8030 // we want to do an instanceof comparison against the AESCrypt class
8031 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8032 if (!klass_AESCrypt->is_loaded()) {
8033 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8034 Node* ctrl = control();
8035 set_control(top()); // no regular fast path
8036 return ctrl;
8037 }
8038
8039 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8040 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8041 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8042 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8043 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8044
8045 return instof_false; // even if it is null
8046 }
8047
8048 //------------------------------inline_ghash_processBlocks
8049 bool LibraryCallKit::inline_ghash_processBlocks() {
8050 address stubAddr;
8051 const char *stubName;
8052 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
8053
8054 stubAddr = StubRoutines::ghash_processBlocks();
8055 stubName = "ghash_processBlocks";
8056
8057 Node* data = argument(0);
8058 Node* offset = argument(1);
8059 Node* len = argument(2);
8060 Node* state = argument(3);
8061 Node* subkeyH = argument(4);
8062
8063 state = must_be_not_null(state, true);
8064 subkeyH = must_be_not_null(subkeyH, true);
8065 data = must_be_not_null(data, true);
8066
8067 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8068 assert(state_start, "state is null");
8069 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
8070 assert(subkeyH_start, "subkeyH is null");
8071 Node* data_start = array_element_address(data, offset, T_BYTE);
8072 assert(data_start, "data is null");
8073
8074 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
8075 OptoRuntime::ghash_processBlocks_Type(),
8076 stubAddr, stubName, TypePtr::BOTTOM,
8077 state_start, subkeyH_start, data_start, len);
8078 return true;
8079 }
8080
8081 //------------------------------inline_chacha20Block
8082 bool LibraryCallKit::inline_chacha20Block() {
8083 address stubAddr;
8084 const char *stubName;
8085 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
8086
8087 stubAddr = StubRoutines::chacha20Block();
8088 stubName = "chacha20Block";
8089
8090 Node* state = argument(0);
8091 Node* result = argument(1);
8092
8093 state = must_be_not_null(state, true);
8094 result = must_be_not_null(result, true);
8095
8096 Node* state_start = array_element_address(state, intcon(0), T_INT);
8097 assert(state_start, "state is null");
8098 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
8099 assert(result_start, "result is null");
8100
8101 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
8102 OptoRuntime::chacha20Block_Type(),
8103 stubAddr, stubName, TypePtr::BOTTOM,
8104 state_start, result_start);
8105 // return key stream length (int)
8106 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
8107 set_result(retvalue);
8108 return true;
8109 }
8110
8111 //------------------------------inline_dilithiumAlmostNtt
8112 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
8113 address stubAddr;
8114 const char *stubName;
8115 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8116 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
8117
8118 stubAddr = StubRoutines::dilithiumAlmostNtt();
8119 stubName = "dilithiumAlmostNtt";
8120 if (!stubAddr) return false;
8121
8122 Node* coeffs = argument(0);
8123 Node* ntt_zetas = argument(1);
8124
8125 coeffs = must_be_not_null(coeffs, true);
8126 ntt_zetas = must_be_not_null(ntt_zetas, true);
8127
8128 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8129 assert(coeffs_start, "coeffs is null");
8130 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
8131 assert(ntt_zetas_start, "ntt_zetas is null");
8132 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8133 OptoRuntime::dilithiumAlmostNtt_Type(),
8134 stubAddr, stubName, TypePtr::BOTTOM,
8135 coeffs_start, ntt_zetas_start);
8136 // return an int
8137 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
8138 set_result(retvalue);
8139 return true;
8140 }
8141
8142 //------------------------------inline_dilithiumAlmostInverseNtt
8143 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
8144 address stubAddr;
8145 const char *stubName;
8146 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8147 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
8148
8149 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
8150 stubName = "dilithiumAlmostInverseNtt";
8151 if (!stubAddr) return false;
8152
8153 Node* coeffs = argument(0);
8154 Node* zetas = argument(1);
8155
8156 coeffs = must_be_not_null(coeffs, true);
8157 zetas = must_be_not_null(zetas, true);
8158
8159 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8160 assert(coeffs_start, "coeffs is null");
8161 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
8162 assert(zetas_start, "inverseNtt_zetas is null");
8163 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8164 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
8165 stubAddr, stubName, TypePtr::BOTTOM,
8166 coeffs_start, zetas_start);
8167
8168 // return an int
8169 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
8170 set_result(retvalue);
8171 return true;
8172 }
8173
8174 //------------------------------inline_dilithiumNttMult
8175 bool LibraryCallKit::inline_dilithiumNttMult() {
8176 address stubAddr;
8177 const char *stubName;
8178 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8179 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
8180
8181 stubAddr = StubRoutines::dilithiumNttMult();
8182 stubName = "dilithiumNttMult";
8183 if (!stubAddr) return false;
8184
8185 Node* result = argument(0);
8186 Node* ntta = argument(1);
8187 Node* nttb = argument(2);
8188
8189 result = must_be_not_null(result, true);
8190 ntta = must_be_not_null(ntta, true);
8191 nttb = must_be_not_null(nttb, true);
8192
8193 Node* result_start = array_element_address(result, intcon(0), T_INT);
8194 assert(result_start, "result is null");
8195 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
8196 assert(ntta_start, "ntta is null");
8197 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
8198 assert(nttb_start, "nttb is null");
8199 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8200 OptoRuntime::dilithiumNttMult_Type(),
8201 stubAddr, stubName, TypePtr::BOTTOM,
8202 result_start, ntta_start, nttb_start);
8203
8204 // return an int
8205 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8206 set_result(retvalue);
8207
8208 return true;
8209 }
8210
8211 //------------------------------inline_dilithiumMontMulByConstant
8212 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8213 address stubAddr;
8214 const char *stubName;
8215 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8216 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8217
8218 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8219 stubName = "dilithiumMontMulByConstant";
8220 if (!stubAddr) return false;
8221
8222 Node* coeffs = argument(0);
8223 Node* constant = argument(1);
8224
8225 coeffs = must_be_not_null(coeffs, true);
8226
8227 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8228 assert(coeffs_start, "coeffs is null");
8229 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8230 OptoRuntime::dilithiumMontMulByConstant_Type(),
8231 stubAddr, stubName, TypePtr::BOTTOM,
8232 coeffs_start, constant);
8233
8234 // return an int
8235 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8236 set_result(retvalue);
8237 return true;
8238 }
8239
8240
8241 //------------------------------inline_dilithiumDecomposePoly
8242 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8243 address stubAddr;
8244 const char *stubName;
8245 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8246 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8247
8248 stubAddr = StubRoutines::dilithiumDecomposePoly();
8249 stubName = "dilithiumDecomposePoly";
8250 if (!stubAddr) return false;
8251
8252 Node* input = argument(0);
8253 Node* lowPart = argument(1);
8254 Node* highPart = argument(2);
8255 Node* twoGamma2 = argument(3);
8256 Node* multiplier = argument(4);
8257
8258 input = must_be_not_null(input, true);
8259 lowPart = must_be_not_null(lowPart, true);
8260 highPart = must_be_not_null(highPart, true);
8261
8262 Node* input_start = array_element_address(input, intcon(0), T_INT);
8263 assert(input_start, "input is null");
8264 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8265 assert(lowPart_start, "lowPart is null");
8266 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8267 assert(highPart_start, "highPart is null");
8268
8269 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8270 OptoRuntime::dilithiumDecomposePoly_Type(),
8271 stubAddr, stubName, TypePtr::BOTTOM,
8272 input_start, lowPart_start, highPart_start,
8273 twoGamma2, multiplier);
8274
8275 // return an int
8276 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8277 set_result(retvalue);
8278 return true;
8279 }
8280
8281 bool LibraryCallKit::inline_base64_encodeBlock() {
8282 address stubAddr;
8283 const char *stubName;
8284 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8285 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8286 stubAddr = StubRoutines::base64_encodeBlock();
8287 stubName = "encodeBlock";
8288
8289 if (!stubAddr) return false;
8290 Node* base64obj = argument(0);
8291 Node* src = argument(1);
8292 Node* offset = argument(2);
8293 Node* len = argument(3);
8294 Node* dest = argument(4);
8295 Node* dp = argument(5);
8296 Node* isURL = argument(6);
8297
8298 src = must_be_not_null(src, true);
8299 dest = must_be_not_null(dest, true);
8300
8301 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8302 assert(src_start, "source array is null");
8303 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8304 assert(dest_start, "destination array is null");
8305
8306 Node* base64 = make_runtime_call(RC_LEAF,
8307 OptoRuntime::base64_encodeBlock_Type(),
8308 stubAddr, stubName, TypePtr::BOTTOM,
8309 src_start, offset, len, dest_start, dp, isURL);
8310 return true;
8311 }
8312
8313 bool LibraryCallKit::inline_base64_decodeBlock() {
8314 address stubAddr;
8315 const char *stubName;
8316 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8317 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8318 stubAddr = StubRoutines::base64_decodeBlock();
8319 stubName = "decodeBlock";
8320
8321 if (!stubAddr) return false;
8322 Node* base64obj = argument(0);
8323 Node* src = argument(1);
8324 Node* src_offset = argument(2);
8325 Node* len = argument(3);
8326 Node* dest = argument(4);
8327 Node* dest_offset = argument(5);
8328 Node* isURL = argument(6);
8329 Node* isMIME = argument(7);
8330
8331 src = must_be_not_null(src, true);
8332 dest = must_be_not_null(dest, true);
8333
8334 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8335 assert(src_start, "source array is null");
8336 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8337 assert(dest_start, "destination array is null");
8338
8339 Node* call = make_runtime_call(RC_LEAF,
8340 OptoRuntime::base64_decodeBlock_Type(),
8341 stubAddr, stubName, TypePtr::BOTTOM,
8342 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8343 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8344 set_result(result);
8345 return true;
8346 }
8347
8348 bool LibraryCallKit::inline_poly1305_processBlocks() {
8349 address stubAddr;
8350 const char *stubName;
8351 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8352 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8353 stubAddr = StubRoutines::poly1305_processBlocks();
8354 stubName = "poly1305_processBlocks";
8355
8356 if (!stubAddr) return false;
8357 null_check_receiver(); // null-check receiver
8358 if (stopped()) return true;
8359
8360 Node* input = argument(1);
8361 Node* input_offset = argument(2);
8362 Node* len = argument(3);
8363 Node* alimbs = argument(4);
8364 Node* rlimbs = argument(5);
8365
8366 input = must_be_not_null(input, true);
8367 alimbs = must_be_not_null(alimbs, true);
8368 rlimbs = must_be_not_null(rlimbs, true);
8369
8370 Node* input_start = array_element_address(input, input_offset, T_BYTE);
8371 assert(input_start, "input array is null");
8372 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
8373 assert(acc_start, "acc array is null");
8374 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
8375 assert(r_start, "r array is null");
8376
8377 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8378 OptoRuntime::poly1305_processBlocks_Type(),
8379 stubAddr, stubName, TypePtr::BOTTOM,
8380 input_start, len, acc_start, r_start);
8381 return true;
8382 }
8383
8384 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
8385 address stubAddr;
8386 const char *stubName;
8387 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8388 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
8389 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
8390 stubName = "intpoly_montgomeryMult_P256";
8391
8392 if (!stubAddr) return false;
8393 null_check_receiver(); // null-check receiver
8394 if (stopped()) return true;
8395
8396 Node* a = argument(1);
8397 Node* b = argument(2);
8398 Node* r = argument(3);
8399
8400 a = must_be_not_null(a, true);
8401 b = must_be_not_null(b, true);
8402 r = must_be_not_null(r, true);
8403
8404 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8405 assert(a_start, "a array is null");
8406 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8407 assert(b_start, "b array is null");
8408 Node* r_start = array_element_address(r, intcon(0), T_LONG);
8409 assert(r_start, "r array is null");
8410
8411 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8412 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
8413 stubAddr, stubName, TypePtr::BOTTOM,
8414 a_start, b_start, r_start);
8415 return true;
8416 }
8417
8418 bool LibraryCallKit::inline_intpoly_assign() {
8419 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8420 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
8421 const char *stubName = "intpoly_assign";
8422 address stubAddr = StubRoutines::intpoly_assign();
8423 if (!stubAddr) return false;
8424
8425 Node* set = argument(0);
8426 Node* a = argument(1);
8427 Node* b = argument(2);
8428 Node* arr_length = load_array_length(a);
8429
8430 a = must_be_not_null(a, true);
8431 b = must_be_not_null(b, true);
8432
8433 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8434 assert(a_start, "a array is null");
8435 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8436 assert(b_start, "b array is null");
8437
8438 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8439 OptoRuntime::intpoly_assign_Type(),
8440 stubAddr, stubName, TypePtr::BOTTOM,
8441 set, a_start, b_start, arr_length);
8442 return true;
8443 }
8444
8445 //------------------------------inline_digestBase_implCompress-----------------------
8446 //
8447 // Calculate MD5 for single-block byte[] array.
8448 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
8449 //
8450 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
8451 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
8452 //
8453 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
8454 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
8455 //
8456 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
8457 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
8458 //
8459 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
8460 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
8461 //
8462 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
8463 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
8464
8465 Node* digestBase_obj = argument(0);
8466 Node* src = argument(1); // type oop
8467 Node* ofs = argument(2); // type int
8468
8469 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8470 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8471 // failed array check
8472 return false;
8473 }
8474 // Figure out the size and type of the elements we will be copying.
8475 BasicType src_elem = src_type->elem()->array_element_basic_type();
8476 if (src_elem != T_BYTE) {
8477 return false;
8478 }
8479 // 'src_start' points to src array + offset
8480 src = must_be_not_null(src, true);
8481 Node* src_start = array_element_address(src, ofs, src_elem);
8482 Node* state = nullptr;
8483 Node* block_size = nullptr;
8484 address stubAddr;
8485 const char *stubName;
8486
8487 switch(id) {
8488 case vmIntrinsics::_md5_implCompress:
8489 assert(UseMD5Intrinsics, "need MD5 instruction support");
8490 state = get_state_from_digest_object(digestBase_obj, T_INT);
8491 stubAddr = StubRoutines::md5_implCompress();
8492 stubName = "md5_implCompress";
8493 break;
8494 case vmIntrinsics::_sha_implCompress:
8495 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
8496 state = get_state_from_digest_object(digestBase_obj, T_INT);
8497 stubAddr = StubRoutines::sha1_implCompress();
8498 stubName = "sha1_implCompress";
8499 break;
8500 case vmIntrinsics::_sha2_implCompress:
8501 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
8502 state = get_state_from_digest_object(digestBase_obj, T_INT);
8503 stubAddr = StubRoutines::sha256_implCompress();
8504 stubName = "sha256_implCompress";
8505 break;
8506 case vmIntrinsics::_sha5_implCompress:
8507 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
8508 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8509 stubAddr = StubRoutines::sha512_implCompress();
8510 stubName = "sha512_implCompress";
8511 break;
8512 case vmIntrinsics::_sha3_implCompress:
8513 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
8514 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8515 stubAddr = StubRoutines::sha3_implCompress();
8516 stubName = "sha3_implCompress";
8517 block_size = get_block_size_from_digest_object(digestBase_obj);
8518 if (block_size == nullptr) return false;
8519 break;
8520 default:
8521 fatal_unexpected_iid(id);
8522 return false;
8523 }
8524 if (state == nullptr) return false;
8525
8526 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
8527 if (stubAddr == nullptr) return false;
8528
8529 // Call the stub.
8530 Node* call;
8531 if (block_size == nullptr) {
8532 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
8533 stubAddr, stubName, TypePtr::BOTTOM,
8534 src_start, state);
8535 } else {
8536 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
8537 stubAddr, stubName, TypePtr::BOTTOM,
8538 src_start, state, block_size);
8539 }
8540
8541 return true;
8542 }
8543
8544 //------------------------------inline_double_keccak
8545 bool LibraryCallKit::inline_double_keccak() {
8546 address stubAddr;
8547 const char *stubName;
8548 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
8549 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
8550
8551 stubAddr = StubRoutines::double_keccak();
8552 stubName = "double_keccak";
8553 if (!stubAddr) return false;
8554
8555 Node* status0 = argument(0);
8556 Node* status1 = argument(1);
8557
8558 status0 = must_be_not_null(status0, true);
8559 status1 = must_be_not_null(status1, true);
8560
8561 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
8562 assert(status0_start, "status0 is null");
8563 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
8564 assert(status1_start, "status1 is null");
8565 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
8566 OptoRuntime::double_keccak_Type(),
8567 stubAddr, stubName, TypePtr::BOTTOM,
8568 status0_start, status1_start);
8569 // return an int
8570 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
8571 set_result(retvalue);
8572 return true;
8573 }
8574
8575
8576 //------------------------------inline_digestBase_implCompressMB-----------------------
8577 //
8578 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
8579 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
8580 //
8581 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
8582 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8583 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8584 assert((uint)predicate < 5, "sanity");
8585 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
8586
8587 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
8588 Node* src = argument(1); // byte[] array
8589 Node* ofs = argument(2); // type int
8590 Node* limit = argument(3); // type int
8591
8592 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8593 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8594 // failed array check
8595 return false;
8596 }
8597 // Figure out the size and type of the elements we will be copying.
8598 BasicType src_elem = src_type->elem()->array_element_basic_type();
8599 if (src_elem != T_BYTE) {
8600 return false;
8601 }
8602 // 'src_start' points to src array + offset
8603 src = must_be_not_null(src, false);
8604 Node* src_start = array_element_address(src, ofs, src_elem);
8605
8606 const char* klass_digestBase_name = nullptr;
8607 const char* stub_name = nullptr;
8608 address stub_addr = nullptr;
8609 BasicType elem_type = T_INT;
8610
8611 switch (predicate) {
8612 case 0:
8613 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
8614 klass_digestBase_name = "sun/security/provider/MD5";
8615 stub_name = "md5_implCompressMB";
8616 stub_addr = StubRoutines::md5_implCompressMB();
8617 }
8618 break;
8619 case 1:
8620 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
8621 klass_digestBase_name = "sun/security/provider/SHA";
8622 stub_name = "sha1_implCompressMB";
8623 stub_addr = StubRoutines::sha1_implCompressMB();
8624 }
8625 break;
8626 case 2:
8627 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
8628 klass_digestBase_name = "sun/security/provider/SHA2";
8629 stub_name = "sha256_implCompressMB";
8630 stub_addr = StubRoutines::sha256_implCompressMB();
8631 }
8632 break;
8633 case 3:
8634 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
8635 klass_digestBase_name = "sun/security/provider/SHA5";
8636 stub_name = "sha512_implCompressMB";
8637 stub_addr = StubRoutines::sha512_implCompressMB();
8638 elem_type = T_LONG;
8639 }
8640 break;
8641 case 4:
8642 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
8643 klass_digestBase_name = "sun/security/provider/SHA3";
8644 stub_name = "sha3_implCompressMB";
8645 stub_addr = StubRoutines::sha3_implCompressMB();
8646 elem_type = T_LONG;
8647 }
8648 break;
8649 default:
8650 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
8651 }
8652 if (klass_digestBase_name != nullptr) {
8653 assert(stub_addr != nullptr, "Stub is generated");
8654 if (stub_addr == nullptr) return false;
8655
8656 // get DigestBase klass to lookup for SHA klass
8657 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
8658 assert(tinst != nullptr, "digestBase_obj is not instance???");
8659 assert(tinst->is_loaded(), "DigestBase is not loaded");
8660
8661 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
8662 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
8663 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
8664 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
8665 }
8666 return false;
8667 }
8668
8669 //------------------------------inline_digestBase_implCompressMB-----------------------
8670 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
8671 BasicType elem_type, address stubAddr, const char *stubName,
8672 Node* src_start, Node* ofs, Node* limit) {
8673 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
8674 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8675 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
8676 digest_obj = _gvn.transform(digest_obj);
8677
8678 Node* state = get_state_from_digest_object(digest_obj, elem_type);
8679 if (state == nullptr) return false;
8680
8681 Node* block_size = nullptr;
8682 if (strcmp("sha3_implCompressMB", stubName) == 0) {
8683 block_size = get_block_size_from_digest_object(digest_obj);
8684 if (block_size == nullptr) return false;
8685 }
8686
8687 // Call the stub.
8688 Node* call;
8689 if (block_size == nullptr) {
8690 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8691 OptoRuntime::digestBase_implCompressMB_Type(false),
8692 stubAddr, stubName, TypePtr::BOTTOM,
8693 src_start, state, ofs, limit);
8694 } else {
8695 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8696 OptoRuntime::digestBase_implCompressMB_Type(true),
8697 stubAddr, stubName, TypePtr::BOTTOM,
8698 src_start, state, block_size, ofs, limit);
8699 }
8700
8701 // return ofs (int)
8702 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8703 set_result(result);
8704
8705 return true;
8706 }
8707
8708 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8709 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8710 assert(UseAES, "need AES instruction support");
8711 address stubAddr = nullptr;
8712 const char *stubName = nullptr;
8713 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8714 stubName = "galoisCounterMode_AESCrypt";
8715
8716 if (stubAddr == nullptr) return false;
8717
8718 Node* in = argument(0);
8719 Node* inOfs = argument(1);
8720 Node* len = argument(2);
8721 Node* ct = argument(3);
8722 Node* ctOfs = argument(4);
8723 Node* out = argument(5);
8724 Node* outOfs = argument(6);
8725 Node* gctr_object = argument(7);
8726 Node* ghash_object = argument(8);
8727
8728 // (1) in, ct and out are arrays.
8729 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8730 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8731 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8732 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8733 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8734 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8735
8736 // checks are the responsibility of the caller
8737 Node* in_start = in;
8738 Node* ct_start = ct;
8739 Node* out_start = out;
8740 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8741 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8742 in_start = array_element_address(in, inOfs, T_BYTE);
8743 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8744 out_start = array_element_address(out, outOfs, T_BYTE);
8745 }
8746
8747 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8748 // (because of the predicated logic executed earlier).
8749 // so we cast it here safely.
8750 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8751 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8752 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8753 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8754 Node* state = load_field_from_object(ghash_object, "state", "[J");
8755
8756 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8757 return false;
8758 }
8759 // cast it to what we know it will be at runtime
8760 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8761 assert(tinst != nullptr, "GCTR obj is null");
8762 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8763 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8764 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8765 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8766 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8767 const TypeOopPtr* xtype = aklass->as_instance_type();
8768 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8769 aescrypt_object = _gvn.transform(aescrypt_object);
8770 // we need to get the start of the aescrypt_object's expanded key array
8771 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8772 if (k_start == nullptr) return false;
8773 // similarly, get the start address of the r vector
8774 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8775 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8776 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8777
8778
8779 // Call the stub, passing params
8780 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8781 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8782 stubAddr, stubName, TypePtr::BOTTOM,
8783 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8784
8785 // return cipher length (int)
8786 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8787 set_result(retvalue);
8788
8789 return true;
8790 }
8791
8792 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8793 // Return node representing slow path of predicate check.
8794 // the pseudo code we want to emulate with this predicate is:
8795 // for encryption:
8796 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8797 // for decryption:
8798 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8799 // note cipher==plain is more conservative than the original java code but that's OK
8800 //
8801
8802 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8803 // The receiver was checked for null already.
8804 Node* objGCTR = argument(7);
8805 // Load embeddedCipher field of GCTR object.
8806 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8807 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8808
8809 // get AESCrypt klass for instanceOf check
8810 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8811 // will have same classloader as CipherBlockChaining object
8812 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8813 assert(tinst != nullptr, "GCTR obj is null");
8814 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8815
8816 // we want to do an instanceof comparison against the AESCrypt class
8817 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8818 if (!klass_AESCrypt->is_loaded()) {
8819 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8820 Node* ctrl = control();
8821 set_control(top()); // no regular fast path
8822 return ctrl;
8823 }
8824
8825 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8826 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8827 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8828 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8829 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8830
8831 return instof_false; // even if it is null
8832 }
8833
8834 //------------------------------get_state_from_digest_object-----------------------
8835 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8836 const char* state_type;
8837 switch (elem_type) {
8838 case T_BYTE: state_type = "[B"; break;
8839 case T_INT: state_type = "[I"; break;
8840 case T_LONG: state_type = "[J"; break;
8841 default: ShouldNotReachHere();
8842 }
8843 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8844 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8845 if (digest_state == nullptr) return (Node *) nullptr;
8846
8847 // now have the array, need to get the start address of the state array
8848 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8849 return state;
8850 }
8851
8852 //------------------------------get_block_size_from_sha3_object----------------------------------
8853 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8854 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8855 assert (block_size != nullptr, "sanity");
8856 return block_size;
8857 }
8858
8859 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8860 // Return node representing slow path of predicate check.
8861 // the pseudo code we want to emulate with this predicate is:
8862 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8863 //
8864 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8865 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8866 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8867 assert((uint)predicate < 5, "sanity");
8868
8869 // The receiver was checked for null already.
8870 Node* digestBaseObj = argument(0);
8871
8872 // get DigestBase klass for instanceOf check
8873 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8874 assert(tinst != nullptr, "digestBaseObj is null");
8875 assert(tinst->is_loaded(), "DigestBase is not loaded");
8876
8877 const char* klass_name = nullptr;
8878 switch (predicate) {
8879 case 0:
8880 if (UseMD5Intrinsics) {
8881 // we want to do an instanceof comparison against the MD5 class
8882 klass_name = "sun/security/provider/MD5";
8883 }
8884 break;
8885 case 1:
8886 if (UseSHA1Intrinsics) {
8887 // we want to do an instanceof comparison against the SHA class
8888 klass_name = "sun/security/provider/SHA";
8889 }
8890 break;
8891 case 2:
8892 if (UseSHA256Intrinsics) {
8893 // we want to do an instanceof comparison against the SHA2 class
8894 klass_name = "sun/security/provider/SHA2";
8895 }
8896 break;
8897 case 3:
8898 if (UseSHA512Intrinsics) {
8899 // we want to do an instanceof comparison against the SHA5 class
8900 klass_name = "sun/security/provider/SHA5";
8901 }
8902 break;
8903 case 4:
8904 if (UseSHA3Intrinsics) {
8905 // we want to do an instanceof comparison against the SHA3 class
8906 klass_name = "sun/security/provider/SHA3";
8907 }
8908 break;
8909 default:
8910 fatal("unknown SHA intrinsic predicate: %d", predicate);
8911 }
8912
8913 ciKlass* klass = nullptr;
8914 if (klass_name != nullptr) {
8915 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8916 }
8917 if ((klass == nullptr) || !klass->is_loaded()) {
8918 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8919 Node* ctrl = control();
8920 set_control(top()); // no intrinsic path
8921 return ctrl;
8922 }
8923 ciInstanceKlass* instklass = klass->as_instance_klass();
8924
8925 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8926 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8927 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8928 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8929
8930 return instof_false; // even if it is null
8931 }
8932
8933 //-------------inline_fma-----------------------------------
8934 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8935 Node *a = nullptr;
8936 Node *b = nullptr;
8937 Node *c = nullptr;
8938 Node* result = nullptr;
8939 switch (id) {
8940 case vmIntrinsics::_fmaD:
8941 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8942 // no receiver since it is static method
8943 a = argument(0);
8944 b = argument(2);
8945 c = argument(4);
8946 result = _gvn.transform(new FmaDNode(a, b, c));
8947 break;
8948 case vmIntrinsics::_fmaF:
8949 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8950 a = argument(0);
8951 b = argument(1);
8952 c = argument(2);
8953 result = _gvn.transform(new FmaFNode(a, b, c));
8954 break;
8955 default:
8956 fatal_unexpected_iid(id); break;
8957 }
8958 set_result(result);
8959 return true;
8960 }
8961
8962 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8963 // argument(0) is receiver
8964 Node* codePoint = argument(1);
8965 Node* n = nullptr;
8966
8967 switch (id) {
8968 case vmIntrinsics::_isDigit :
8969 n = new DigitNode(control(), codePoint);
8970 break;
8971 case vmIntrinsics::_isLowerCase :
8972 n = new LowerCaseNode(control(), codePoint);
8973 break;
8974 case vmIntrinsics::_isUpperCase :
8975 n = new UpperCaseNode(control(), codePoint);
8976 break;
8977 case vmIntrinsics::_isWhitespace :
8978 n = new WhitespaceNode(control(), codePoint);
8979 break;
8980 default:
8981 fatal_unexpected_iid(id);
8982 }
8983
8984 set_result(_gvn.transform(n));
8985 return true;
8986 }
8987
8988 bool LibraryCallKit::inline_profileBoolean() {
8989 Node* counts = argument(1);
8990 const TypeAryPtr* ary = nullptr;
8991 ciArray* aobj = nullptr;
8992 if (counts->is_Con()
8993 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8994 && (aobj = ary->const_oop()->as_array()) != nullptr
8995 && (aobj->length() == 2)) {
8996 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8997 jint false_cnt = aobj->element_value(0).as_int();
8998 jint true_cnt = aobj->element_value(1).as_int();
8999
9000 if (C->log() != nullptr) {
9001 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
9002 false_cnt, true_cnt);
9003 }
9004
9005 if (false_cnt + true_cnt == 0) {
9006 // According to profile, never executed.
9007 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9008 Deoptimization::Action_reinterpret);
9009 return true;
9010 }
9011
9012 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
9013 // is a number of each value occurrences.
9014 Node* result = argument(0);
9015 if (false_cnt == 0 || true_cnt == 0) {
9016 // According to profile, one value has been never seen.
9017 int expected_val = (false_cnt == 0) ? 1 : 0;
9018
9019 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
9020 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
9021
9022 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
9023 Node* fast_path = _gvn.transform(new IfTrueNode(check));
9024 Node* slow_path = _gvn.transform(new IfFalseNode(check));
9025
9026 { // Slow path: uncommon trap for never seen value and then reexecute
9027 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
9028 // the value has been seen at least once.
9029 PreserveJVMState pjvms(this);
9030 PreserveReexecuteState preexecs(this);
9031 jvms()->set_should_reexecute(true);
9032
9033 set_control(slow_path);
9034 set_i_o(i_o());
9035
9036 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9037 Deoptimization::Action_reinterpret);
9038 }
9039 // The guard for never seen value enables sharpening of the result and
9040 // returning a constant. It allows to eliminate branches on the same value
9041 // later on.
9042 set_control(fast_path);
9043 result = intcon(expected_val);
9044 }
9045 // Stop profiling.
9046 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
9047 // By replacing method body with profile data (represented as ProfileBooleanNode
9048 // on IR level) we effectively disable profiling.
9049 // It enables full speed execution once optimized code is generated.
9050 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
9051 C->record_for_igvn(profile);
9052 set_result(profile);
9053 return true;
9054 } else {
9055 // Continue profiling.
9056 // Profile data isn't available at the moment. So, execute method's bytecode version.
9057 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
9058 // is compiled and counters aren't available since corresponding MethodHandle
9059 // isn't a compile-time constant.
9060 return false;
9061 }
9062 }
9063
9064 bool LibraryCallKit::inline_isCompileConstant() {
9065 Node* n = argument(0);
9066 set_result(n->is_Con() ? intcon(1) : intcon(0));
9067 return true;
9068 }
9069
9070 //------------------------------- inline_getObjectSize --------------------------------------
9071 //
9072 // Calculate the runtime size of the object/array.
9073 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
9074 //
9075 bool LibraryCallKit::inline_getObjectSize() {
9076 Node* obj = argument(3);
9077 Node* klass_node = load_object_klass(obj);
9078
9079 jint layout_con = Klass::_lh_neutral_value;
9080 Node* layout_val = get_layout_helper(klass_node, layout_con);
9081 int layout_is_con = (layout_val == nullptr);
9082
9083 if (layout_is_con) {
9084 // Layout helper is constant, can figure out things at compile time.
9085
9086 if (Klass::layout_helper_is_instance(layout_con)) {
9087 // Instance case: layout_con contains the size itself.
9088 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
9089 set_result(size);
9090 } else {
9091 // Array case: size is round(header + element_size*arraylength).
9092 // Since arraylength is different for every array instance, we have to
9093 // compute the whole thing at runtime.
9094
9095 Node* arr_length = load_array_length(obj);
9096
9097 int round_mask = MinObjAlignmentInBytes - 1;
9098 int hsize = Klass::layout_helper_header_size(layout_con);
9099 int eshift = Klass::layout_helper_log2_element_size(layout_con);
9100
9101 if ((round_mask & ~right_n_bits(eshift)) == 0) {
9102 round_mask = 0; // strength-reduce it if it goes away completely
9103 }
9104 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
9105 Node* header_size = intcon(hsize + round_mask);
9106
9107 Node* lengthx = ConvI2X(arr_length);
9108 Node* headerx = ConvI2X(header_size);
9109
9110 Node* abody = lengthx;
9111 if (eshift != 0) {
9112 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
9113 }
9114 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
9115 if (round_mask != 0) {
9116 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
9117 }
9118 size = ConvX2L(size);
9119 set_result(size);
9120 }
9121 } else {
9122 // Layout helper is not constant, need to test for array-ness at runtime.
9123
9124 enum { _instance_path = 1, _array_path, PATH_LIMIT };
9125 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
9126 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
9127 record_for_igvn(result_reg);
9128
9129 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
9130 if (array_ctl != nullptr) {
9131 // Array case: size is round(header + element_size*arraylength).
9132 // Since arraylength is different for every array instance, we have to
9133 // compute the whole thing at runtime.
9134
9135 PreserveJVMState pjvms(this);
9136 set_control(array_ctl);
9137 Node* arr_length = load_array_length(obj);
9138
9139 int round_mask = MinObjAlignmentInBytes - 1;
9140 Node* mask = intcon(round_mask);
9141
9142 Node* hss = intcon(Klass::_lh_header_size_shift);
9143 Node* hsm = intcon(Klass::_lh_header_size_mask);
9144 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
9145 header_size = _gvn.transform(new AndINode(header_size, hsm));
9146 header_size = _gvn.transform(new AddINode(header_size, mask));
9147
9148 // There is no need to mask or shift this value.
9149 // The semantics of LShiftINode include an implicit mask to 0x1F.
9150 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
9151 Node* elem_shift = layout_val;
9152
9153 Node* lengthx = ConvI2X(arr_length);
9154 Node* headerx = ConvI2X(header_size);
9155
9156 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
9157 Node* size = _gvn.transform(new AddXNode(headerx, abody));
9158 if (round_mask != 0) {
9159 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
9160 }
9161 size = ConvX2L(size);
9162
9163 result_reg->init_req(_array_path, control());
9164 result_val->init_req(_array_path, size);
9165 }
9166
9167 if (!stopped()) {
9168 // Instance case: the layout helper gives us instance size almost directly,
9169 // but we need to mask out the _lh_instance_slow_path_bit.
9170 Node* size = ConvI2X(layout_val);
9171 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
9172 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
9173 size = _gvn.transform(new AndXNode(size, mask));
9174 size = ConvX2L(size);
9175
9176 result_reg->init_req(_instance_path, control());
9177 result_val->init_req(_instance_path, size);
9178 }
9179
9180 set_result(result_reg, result_val);
9181 }
9182
9183 return true;
9184 }
9185
9186 //------------------------------- inline_blackhole --------------------------------------
9187 //
9188 // Make sure all arguments to this node are alive.
9189 // This matches methods that were requested to be blackholed through compile commands.
9190 //
9191 bool LibraryCallKit::inline_blackhole() {
9192 assert(callee()->is_static(), "Should have been checked before: only static methods here");
9193 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
9194 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
9195
9196 // Blackhole node pinches only the control, not memory. This allows
9197 // the blackhole to be pinned in the loop that computes blackholed
9198 // values, but have no other side effects, like breaking the optimizations
9199 // across the blackhole.
9200
9201 Node* bh = _gvn.transform(new BlackholeNode(control()));
9202 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
9203
9204 // Bind call arguments as blackhole arguments to keep them alive
9205 uint nargs = callee()->arg_size();
9206 for (uint i = 0; i < nargs; i++) {
9207 bh->add_req(argument(i));
9208 }
9209
9210 return true;
9211 }
9212
9213 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9214 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9215 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9216 return nullptr; // box klass is not Float16
9217 }
9218
9219 // Null check; get notnull casted pointer
9220 Node* null_ctl = top();
9221 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9222 // If not_null_box is dead, only null-path is taken
9223 if (stopped()) {
9224 set_control(null_ctl);
9225 return nullptr;
9226 }
9227 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9228 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9229 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9230 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9231 }
9232
9233 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9234 PreserveReexecuteState preexecs(this);
9235 jvms()->set_should_reexecute(true);
9236
9237 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9238 Node* klass_node = makecon(klass_type);
9239 Node* box = new_instance(klass_node);
9240
9241 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9242 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9243
9244 Node* field_store = _gvn.transform(access_store_at(box,
9245 value_field,
9246 value_adr_type,
9247 value,
9248 TypeInt::SHORT,
9249 T_SHORT,
9250 IN_HEAP));
9251 set_memory(field_store, value_adr_type);
9252 return box;
9253 }
9254
9255 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9256 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9257 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9258 return false;
9259 }
9260
9261 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9262 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9263 return false;
9264 }
9265
9266 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9267 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9268 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9269 ciSymbols::short_signature(),
9270 false);
9271 assert(field != nullptr, "");
9272
9273 // Transformed nodes
9274 Node* fld1 = nullptr;
9275 Node* fld2 = nullptr;
9276 Node* fld3 = nullptr;
9277 switch(num_args) {
9278 case 3:
9279 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9280 if (fld3 == nullptr) {
9281 return false;
9282 }
9283 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9284 // fall-through
9285 case 2:
9286 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9287 if (fld2 == nullptr) {
9288 return false;
9289 }
9290 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9291 // fall-through
9292 case 1:
9293 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9294 if (fld1 == nullptr) {
9295 return false;
9296 }
9297 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9298 break;
9299 default: fatal("Unsupported number of arguments %d", num_args);
9300 }
9301
9302 Node* result = nullptr;
9303 switch (id) {
9304 // Unary operations
9305 case vmIntrinsics::_sqrt_float16:
9306 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9307 break;
9308 // Ternary operations
9309 case vmIntrinsics::_fma_float16:
9310 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9311 break;
9312 default:
9313 fatal_unexpected_iid(id);
9314 break;
9315 }
9316 result = _gvn.transform(new ReinterpretHF2SNode(result));
9317 set_result(box_fp16_value(float16_box_type, field, result));
9318 return true;
9319 }
9320