1 /*
2 * Copyright (c) 1999, 2026, 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/ciSymbols.hpp"
27 #include "ci/ciUtilities.inline.hpp"
28 #include "classfile/vmIntrinsics.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "gc/shared/barrierSet.hpp"
32 #include "jfr/support/jfrIntrinsics.hpp"
33 #include "memory/resourceArea.hpp"
34 #include "oops/klass.inline.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "opto/addnode.hpp"
37 #include "opto/arraycopynode.hpp"
38 #include "opto/c2compiler.hpp"
39 #include "opto/castnode.hpp"
40 #include "opto/cfgnode.hpp"
41 #include "opto/convertnode.hpp"
42 #include "opto/countbitsnode.hpp"
43 #include "opto/idealKit.hpp"
44 #include "opto/library_call.hpp"
45 #include "opto/mathexactnode.hpp"
46 #include "opto/mulnode.hpp"
47 #include "opto/narrowptrnode.hpp"
48 #include "opto/opaquenode.hpp"
49 #include "opto/parse.hpp"
50 #include "opto/rootnode.hpp"
51 #include "opto/runtime.hpp"
52 #include "opto/subnode.hpp"
53 #include "opto/vectornode.hpp"
54 #include "prims/jvmtiExport.hpp"
55 #include "prims/jvmtiThreadState.hpp"
56 #include "prims/unsafe.hpp"
57 #include "runtime/jniHandles.inline.hpp"
58 #include "runtime/mountUnmountDisabler.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 assert(jvms->map() == kit.map(), "Out of sync JVM state");
140 if (jvms->has_method()) {
141 // Not a root compile.
142 const char* msg;
143 if (callee->intrinsic_candidate()) {
144 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
145 } else {
146 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
147 : "failed to inline (intrinsic), method not annotated";
148 }
149 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
150 C->inline_printer()->record(callee, jvms, InliningResult::FAILURE, msg);
151 } else {
152 // Root compile
153 ResourceMark rm;
154 stringStream msg_stream;
155 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
156 vmIntrinsics::name_at(intrinsic_id()),
157 is_virtual() ? " (virtual)" : "", bci);
158 const char *msg = msg_stream.freeze();
159 log_debug(jit, inlining)("%s", msg);
160 if (C->print_intrinsics() || C->print_inlining()) {
161 tty->print("%s", msg);
162 }
163 }
164 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
165
166 return nullptr;
167 }
168
169 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
170 LibraryCallKit kit(jvms, this);
171 Compile* C = kit.C;
172 int nodes = C->unique();
173 _last_predicate = predicate;
174 #ifndef PRODUCT
175 assert(is_predicated() && predicate < predicates_count(), "sanity");
176 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
177 char buf[1000];
178 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
179 tty->print_cr("Predicate for intrinsic %s", str);
180 }
181 #endif
182 ciMethod* callee = kit.callee();
183 const int bci = kit.bci();
184
185 Node* slow_ctl = kit.try_to_predicate(predicate);
186 if (!kit.failing()) {
187 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
188 : "(intrinsic, predicate)";
189 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
190 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
191
192 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
193 if (C->log()) {
194 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
195 vmIntrinsics::name_at(intrinsic_id()),
196 (is_virtual() ? " virtual='1'" : ""),
197 C->unique() - nodes);
198 }
199 return slow_ctl; // Could be null if the check folds.
200 }
201
202 // The intrinsic bailed out
203 if (jvms->has_method()) {
204 // Not a root compile.
205 const char* msg = "failed to generate predicate for intrinsic";
206 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
207 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
208 } else {
209 // Root compile
210 ResourceMark rm;
211 stringStream msg_stream;
212 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
213 vmIntrinsics::name_at(intrinsic_id()),
214 is_virtual() ? " (virtual)" : "", bci);
215 const char *msg = msg_stream.freeze();
216 log_debug(jit, inlining)("%s", msg);
217 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
218 }
219 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
220 return nullptr;
221 }
222
223 bool LibraryCallKit::try_to_inline(int predicate) {
224 // Handle symbolic names for otherwise undistinguished boolean switches:
225 const bool is_store = true;
226 const bool is_compress = true;
227 const bool is_static = true;
228 const bool is_volatile = true;
229
230 if (!jvms()->has_method()) {
231 // Root JVMState has a null method.
232 assert(map()->memory()->Opcode() == Op_Parm, "");
233 // Insert the memory aliasing node
234 set_all_memory(reset_memory());
235 }
236 assert(merged_memory(), "");
237
238 switch (intrinsic_id()) {
239 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
240 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
241 case vmIntrinsics::_getClass: return inline_native_getClass();
242
243 case vmIntrinsics::_ceil:
244 case vmIntrinsics::_floor:
245 case vmIntrinsics::_rint:
246 case vmIntrinsics::_dsin:
247 case vmIntrinsics::_dcos:
248 case vmIntrinsics::_dtan:
249 case vmIntrinsics::_dsinh:
250 case vmIntrinsics::_dtanh:
251 case vmIntrinsics::_dcbrt:
252 case vmIntrinsics::_dabs:
253 case vmIntrinsics::_fabs:
254 case vmIntrinsics::_iabs:
255 case vmIntrinsics::_labs:
256 case vmIntrinsics::_datan2:
257 case vmIntrinsics::_dsqrt:
258 case vmIntrinsics::_dsqrt_strict:
259 case vmIntrinsics::_dexp:
260 case vmIntrinsics::_dlog:
261 case vmIntrinsics::_dlog10:
262 case vmIntrinsics::_dpow:
263 case vmIntrinsics::_dcopySign:
264 case vmIntrinsics::_fcopySign:
265 case vmIntrinsics::_dsignum:
266 case vmIntrinsics::_roundF:
267 case vmIntrinsics::_roundD:
268 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
269
270 case vmIntrinsics::_notify:
271 case vmIntrinsics::_notifyAll:
272 return inline_notify(intrinsic_id());
273
274 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
275 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
276 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
277 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
278 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
279 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
280 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
281 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
282 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
283 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
284 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
285 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
286 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
287 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
288
289 case vmIntrinsics::_arraycopy: return inline_arraycopy();
290
291 case vmIntrinsics::_arraySort: return inline_array_sort();
292 case vmIntrinsics::_arrayPartition: return inline_array_partition();
293
294 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
295 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
296 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
297 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
298
299 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
300 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
301 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
302 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
303 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
304 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
305 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
306 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
307
308 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
309
310 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
311
312 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
313 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
314 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
315 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
316
317 case vmIntrinsics::_compressStringC:
318 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
319 case vmIntrinsics::_inflateStringC:
320 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
321
322 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
323 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
324 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
325 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
326 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
327 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
328 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
329 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
330 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
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
342 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
343 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
344 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
345 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
346 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
347 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
348 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
349 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
350 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
351
352 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
353 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
354 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
355 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
356 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
357 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
358 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
359 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
360 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
361
362 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
363 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
364 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
365 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
366
367 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
368 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
369 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
370 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
371
372 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
373 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
374 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
375 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
376 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
377 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
378 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
379 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
380 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
381
382 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
383 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
384 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
385 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
386 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
387 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
388 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
389 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
390 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
391
392 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
393 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
394 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
395 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
396 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
397 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
398 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
399 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
400 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
401
402 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
403 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
404 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
405 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
406 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
407 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
408 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
409 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
410 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
411
412 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
413 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
414 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
415 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
416 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
417
418 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
419 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
420 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
421 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
422 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
423 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
424 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
425 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
426 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
427 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
428 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
429 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
430 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
431 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
432 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
433 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
434 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
435 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
436 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
437 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
438
439 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
440 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
441 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
442 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
443 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
444 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
445 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
446 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
447 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
448 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
449 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
450 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
451 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
452 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
453 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
454
455 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
456 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
457 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
458 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
459
460 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
461 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
462 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
463 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
464 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
465
466 case vmIntrinsics::_loadFence:
467 case vmIntrinsics::_storeFence:
468 case vmIntrinsics::_storeStoreFence:
469 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
470
471 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
472
473 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
474 case vmIntrinsics::_currentThread: return inline_native_currentThread();
475 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
476
477 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
478 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
479
480 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
481 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
482
483 case vmIntrinsics::_vthreadEndFirstTransition: return inline_native_vthread_end_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_end_first_transition_Java()),
484 "endFirstTransition", true);
485 case vmIntrinsics::_vthreadStartFinalTransition: return inline_native_vthread_start_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_start_final_transition_Java()),
486 "startFinalTransition", true);
487 case vmIntrinsics::_vthreadStartTransition: return inline_native_vthread_start_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_start_transition_Java()),
488 "startTransition", false);
489 case vmIntrinsics::_vthreadEndTransition: return inline_native_vthread_end_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_end_transition_Java()),
490 "endTransition", false);
491 #if INCLUDE_JVMTI
492 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
493 #endif
494
495 #ifdef JFR_HAVE_INTRINSICS
496 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
497 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
498 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
499 #endif
500 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
501 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
502 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
503 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
504 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
505 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
506 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
507 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
508 case vmIntrinsics::_getLength: return inline_native_getLength();
509 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
510 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
511 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
512 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
513 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
514 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
515 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
516
517 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
518 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
519
520 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
521
522 case vmIntrinsics::_isInstance:
523 case vmIntrinsics::_isHidden:
524 case vmIntrinsics::_getSuperclass: return inline_native_Class_query(intrinsic_id());
525
526 case vmIntrinsics::_floatToRawIntBits:
527 case vmIntrinsics::_floatToIntBits:
528 case vmIntrinsics::_intBitsToFloat:
529 case vmIntrinsics::_doubleToRawLongBits:
530 case vmIntrinsics::_doubleToLongBits:
531 case vmIntrinsics::_longBitsToDouble:
532 case vmIntrinsics::_floatToFloat16:
533 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
534 case vmIntrinsics::_sqrt_float16: return inline_fp16_operations(intrinsic_id(), 1);
535 case vmIntrinsics::_fma_float16: return inline_fp16_operations(intrinsic_id(), 3);
536 case vmIntrinsics::_floatIsFinite:
537 case vmIntrinsics::_floatIsInfinite:
538 case vmIntrinsics::_doubleIsFinite:
539 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
540
541 case vmIntrinsics::_numberOfLeadingZeros_i:
542 case vmIntrinsics::_numberOfLeadingZeros_l:
543 case vmIntrinsics::_numberOfTrailingZeros_i:
544 case vmIntrinsics::_numberOfTrailingZeros_l:
545 case vmIntrinsics::_bitCount_i:
546 case vmIntrinsics::_bitCount_l:
547 case vmIntrinsics::_reverse_i:
548 case vmIntrinsics::_reverse_l:
549 case vmIntrinsics::_reverseBytes_i:
550 case vmIntrinsics::_reverseBytes_l:
551 case vmIntrinsics::_reverseBytes_s:
552 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
553
554 case vmIntrinsics::_compress_i:
555 case vmIntrinsics::_compress_l:
556 case vmIntrinsics::_expand_i:
557 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
558
559 case vmIntrinsics::_compareUnsigned_i:
560 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
561
562 case vmIntrinsics::_divideUnsigned_i:
563 case vmIntrinsics::_divideUnsigned_l:
564 case vmIntrinsics::_remainderUnsigned_i:
565 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
566
567 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
568
569 case vmIntrinsics::_Reference_get0: return inline_reference_get0();
570 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
571 case vmIntrinsics::_Reference_reachabilityFence: return inline_reference_reachabilityFence();
572 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
573 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
574 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
575
576 case vmIntrinsics::_Class_cast: return inline_Class_cast();
577
578 case vmIntrinsics::_aescrypt_encryptBlock:
579 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
580
581 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
582 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
583 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
584
585 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
586 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
587 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
588
589 case vmIntrinsics::_counterMode_AESCrypt:
590 return inline_counterMode_AESCrypt(intrinsic_id());
591
592 case vmIntrinsics::_galoisCounterMode_AESCrypt:
593 return inline_galoisCounterMode_AESCrypt();
594
595 case vmIntrinsics::_md5_implCompress:
596 case vmIntrinsics::_sha_implCompress:
597 case vmIntrinsics::_sha2_implCompress:
598 case vmIntrinsics::_sha5_implCompress:
599 case vmIntrinsics::_sha3_implCompress:
600 return inline_digestBase_implCompress(intrinsic_id());
601 case vmIntrinsics::_double_keccak:
602 return inline_double_keccak();
603
604 case vmIntrinsics::_digestBase_implCompressMB:
605 return inline_digestBase_implCompressMB(predicate);
606
607 case vmIntrinsics::_multiplyToLen:
608 return inline_multiplyToLen();
609
610 case vmIntrinsics::_squareToLen:
611 return inline_squareToLen();
612
613 case vmIntrinsics::_mulAdd:
614 return inline_mulAdd();
615
616 case vmIntrinsics::_montgomeryMultiply:
617 return inline_montgomeryMultiply();
618 case vmIntrinsics::_montgomerySquare:
619 return inline_montgomerySquare();
620
621 case vmIntrinsics::_bigIntegerRightShiftWorker:
622 return inline_bigIntegerShift(true);
623 case vmIntrinsics::_bigIntegerLeftShiftWorker:
624 return inline_bigIntegerShift(false);
625
626 case vmIntrinsics::_vectorizedMismatch:
627 return inline_vectorizedMismatch();
628
629 case vmIntrinsics::_ghash_processBlocks:
630 return inline_ghash_processBlocks();
631 case vmIntrinsics::_chacha20Block:
632 return inline_chacha20Block();
633 case vmIntrinsics::_kyberNtt:
634 return inline_kyberNtt();
635 case vmIntrinsics::_kyberInverseNtt:
636 return inline_kyberInverseNtt();
637 case vmIntrinsics::_kyberNttMult:
638 return inline_kyberNttMult();
639 case vmIntrinsics::_kyberAddPoly_2:
640 return inline_kyberAddPoly_2();
641 case vmIntrinsics::_kyberAddPoly_3:
642 return inline_kyberAddPoly_3();
643 case vmIntrinsics::_kyber12To16:
644 return inline_kyber12To16();
645 case vmIntrinsics::_kyberBarrettReduce:
646 return inline_kyberBarrettReduce();
647 case vmIntrinsics::_dilithiumAlmostNtt:
648 return inline_dilithiumAlmostNtt();
649 case vmIntrinsics::_dilithiumAlmostInverseNtt:
650 return inline_dilithiumAlmostInverseNtt();
651 case vmIntrinsics::_dilithiumNttMult:
652 return inline_dilithiumNttMult();
653 case vmIntrinsics::_dilithiumMontMulByConstant:
654 return inline_dilithiumMontMulByConstant();
655 case vmIntrinsics::_dilithiumDecomposePoly:
656 return inline_dilithiumDecomposePoly();
657 case vmIntrinsics::_base64_encodeBlock:
658 return inline_base64_encodeBlock();
659 case vmIntrinsics::_base64_decodeBlock:
660 return inline_base64_decodeBlock();
661 case vmIntrinsics::_poly1305_processBlocks:
662 return inline_poly1305_processBlocks();
663 case vmIntrinsics::_intpoly_montgomeryMult_P256:
664 return inline_intpoly_montgomeryMult_P256();
665 case vmIntrinsics::_intpoly_assign:
666 return inline_intpoly_assign();
667 case vmIntrinsics::_encodeISOArray:
668 case vmIntrinsics::_encodeByteISOArray:
669 return inline_encodeISOArray(false);
670 case vmIntrinsics::_encodeAsciiArray:
671 return inline_encodeISOArray(true);
672
673 case vmIntrinsics::_updateCRC32:
674 return inline_updateCRC32();
675 case vmIntrinsics::_updateBytesCRC32:
676 return inline_updateBytesCRC32();
677 case vmIntrinsics::_updateByteBufferCRC32:
678 return inline_updateByteBufferCRC32();
679
680 case vmIntrinsics::_updateBytesCRC32C:
681 return inline_updateBytesCRC32C();
682 case vmIntrinsics::_updateDirectByteBufferCRC32C:
683 return inline_updateDirectByteBufferCRC32C();
684
685 case vmIntrinsics::_updateBytesAdler32:
686 return inline_updateBytesAdler32();
687 case vmIntrinsics::_updateByteBufferAdler32:
688 return inline_updateByteBufferAdler32();
689
690 case vmIntrinsics::_profileBoolean:
691 return inline_profileBoolean();
692 case vmIntrinsics::_isCompileConstant:
693 return inline_isCompileConstant();
694
695 case vmIntrinsics::_countPositives:
696 return inline_countPositives();
697
698 case vmIntrinsics::_fmaD:
699 case vmIntrinsics::_fmaF:
700 return inline_fma(intrinsic_id());
701
702 case vmIntrinsics::_isDigit:
703 case vmIntrinsics::_isLowerCase:
704 case vmIntrinsics::_isUpperCase:
705 case vmIntrinsics::_isWhitespace:
706 return inline_character_compare(intrinsic_id());
707
708 case vmIntrinsics::_min:
709 case vmIntrinsics::_max:
710 case vmIntrinsics::_min_strict:
711 case vmIntrinsics::_max_strict:
712 case vmIntrinsics::_minL:
713 case vmIntrinsics::_maxL:
714 case vmIntrinsics::_minF:
715 case vmIntrinsics::_maxF:
716 case vmIntrinsics::_minD:
717 case vmIntrinsics::_maxD:
718 case vmIntrinsics::_minF_strict:
719 case vmIntrinsics::_maxF_strict:
720 case vmIntrinsics::_minD_strict:
721 case vmIntrinsics::_maxD_strict:
722 return inline_min_max(intrinsic_id());
723
724 case vmIntrinsics::_VectorUnaryOp:
725 return inline_vector_nary_operation(1);
726 case vmIntrinsics::_VectorBinaryOp:
727 return inline_vector_nary_operation(2);
728 case vmIntrinsics::_VectorUnaryLibOp:
729 return inline_vector_call(1);
730 case vmIntrinsics::_VectorBinaryLibOp:
731 return inline_vector_call(2);
732 case vmIntrinsics::_VectorTernaryOp:
733 return inline_vector_nary_operation(3);
734 case vmIntrinsics::_VectorFromBitsCoerced:
735 return inline_vector_frombits_coerced();
736 case vmIntrinsics::_VectorMaskOp:
737 return inline_vector_mask_operation();
738 case vmIntrinsics::_VectorLoadOp:
739 return inline_vector_mem_operation(/*is_store=*/false);
740 case vmIntrinsics::_VectorLoadMaskedOp:
741 return inline_vector_mem_masked_operation(/*is_store*/false);
742 case vmIntrinsics::_VectorStoreOp:
743 return inline_vector_mem_operation(/*is_store=*/true);
744 case vmIntrinsics::_VectorStoreMaskedOp:
745 return inline_vector_mem_masked_operation(/*is_store=*/true);
746 case vmIntrinsics::_VectorGatherOp:
747 return inline_vector_gather_scatter(/*is_scatter*/ false);
748 case vmIntrinsics::_VectorScatterOp:
749 return inline_vector_gather_scatter(/*is_scatter*/ true);
750 case vmIntrinsics::_VectorReductionCoerced:
751 return inline_vector_reduction();
752 case vmIntrinsics::_VectorTest:
753 return inline_vector_test();
754 case vmIntrinsics::_VectorBlend:
755 return inline_vector_blend();
756 case vmIntrinsics::_VectorRearrange:
757 return inline_vector_rearrange();
758 case vmIntrinsics::_VectorSelectFrom:
759 return inline_vector_select_from();
760 case vmIntrinsics::_VectorCompare:
761 return inline_vector_compare();
762 case vmIntrinsics::_VectorBroadcastInt:
763 return inline_vector_broadcast_int();
764 case vmIntrinsics::_VectorConvert:
765 return inline_vector_convert();
766 case vmIntrinsics::_VectorInsert:
767 return inline_vector_insert();
768 case vmIntrinsics::_VectorExtract:
769 return inline_vector_extract();
770 case vmIntrinsics::_VectorCompressExpand:
771 return inline_vector_compress_expand();
772 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
773 return inline_vector_select_from_two_vectors();
774 case vmIntrinsics::_IndexVector:
775 return inline_index_vector();
776 case vmIntrinsics::_IndexPartiallyInUpperRange:
777 return inline_index_partially_in_upper_range();
778
779 case vmIntrinsics::_getObjectSize:
780 return inline_getObjectSize();
781
782 case vmIntrinsics::_blackhole:
783 return inline_blackhole();
784
785 default:
786 // If you get here, it may be that someone has added a new intrinsic
787 // to the list in vmIntrinsics.hpp without implementing it here.
788 #ifndef PRODUCT
789 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
790 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
791 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
792 }
793 #endif
794 return false;
795 }
796 }
797
798 Node* LibraryCallKit::try_to_predicate(int predicate) {
799 if (!jvms()->has_method()) {
800 // Root JVMState has a null method.
801 assert(map()->memory()->Opcode() == Op_Parm, "");
802 // Insert the memory aliasing node
803 set_all_memory(reset_memory());
804 }
805 assert(merged_memory(), "");
806
807 switch (intrinsic_id()) {
808 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
809 return inline_cipherBlockChaining_AESCrypt_predicate(false);
810 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
811 return inline_cipherBlockChaining_AESCrypt_predicate(true);
812 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
813 return inline_electronicCodeBook_AESCrypt_predicate(false);
814 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
815 return inline_electronicCodeBook_AESCrypt_predicate(true);
816 case vmIntrinsics::_counterMode_AESCrypt:
817 return inline_counterMode_AESCrypt_predicate();
818 case vmIntrinsics::_digestBase_implCompressMB:
819 return inline_digestBase_implCompressMB_predicate(predicate);
820 case vmIntrinsics::_galoisCounterMode_AESCrypt:
821 return inline_galoisCounterMode_AESCrypt_predicate();
822
823 default:
824 // If you get here, it may be that someone has added a new intrinsic
825 // to the list in vmIntrinsics.hpp without implementing it here.
826 #ifndef PRODUCT
827 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
828 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
829 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
830 }
831 #endif
832 Node* slow_ctl = control();
833 set_control(top()); // No fast path intrinsic
834 return slow_ctl;
835 }
836 }
837
838 //------------------------------set_result-------------------------------
839 // Helper function for finishing intrinsics.
840 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
841 record_for_igvn(region);
842 set_control(_gvn.transform(region));
843 set_result( _gvn.transform(value));
844 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
845 }
846
847 RegionNode* LibraryCallKit::create_bailout() {
848 RegionNode* bailout = new RegionNode(1);
849 record_for_igvn(bailout);
850 return bailout;
851 }
852
853 bool LibraryCallKit::check_bailout(RegionNode* bailout) {
854 if (bailout->req() > 1) {
855 bailout = _gvn.transform(bailout)->as_Region();
856 Node* frame = _gvn.transform(new ParmNode(C->start(), TypeFunc::FramePtr));
857 Node* halt = _gvn.transform(new HaltNode(bailout, frame, "unexpected guard failure in intrinsic"));
858 C->root()->add_req(halt);
859 }
860 return stopped();
861 }
862
863 //------------------------------generate_guard---------------------------
864 // Helper function for generating guarded fast-slow graph structures.
865 // The given 'test', if true, guards a slow path. If the test fails
866 // then a fast path can be taken. (We generally hope it fails.)
867 // In all cases, GraphKit::control() is updated to the fast path.
868 // The returned value represents the control for the slow path.
869 // The return value is never 'top'; it is either a valid control
870 // or null if it is obvious that the slow path can never be taken.
871 // Also, if region and the slow control are not null, the slow edge
872 // is appended to the region.
873 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
874 if (stopped()) {
875 // Already short circuited.
876 return nullptr;
877 }
878
879 // Build an if node and its projections.
880 // If test is true we take the slow path, which we assume is uncommon.
881 if (_gvn.type(test) == TypeInt::ZERO) {
882 // The slow branch is never taken. No need to build this guard.
883 return nullptr;
884 }
885
886 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
887
888 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
889 if (if_slow == top()) {
890 // The slow branch is never taken. No need to build this guard.
891 return nullptr;
892 }
893
894 if (region != nullptr)
895 region->add_req(if_slow);
896
897 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
898 set_control(if_fast);
899
900 return if_slow;
901 }
902
903 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
904 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
905 }
906 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
907 return generate_guard(test, region, PROB_FAIR);
908 }
909
910 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
911 Node** pos_index, bool with_opaque) {
912 if (stopped())
913 return nullptr; // already stopped
914 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
915 return nullptr; // index is already adequately typed
916 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
917 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
918 if (with_opaque) {
919 bol_lt = _gvn.transform(new OpaqueConstantBoolNode(C, bol_lt, false));
920 }
921 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
922 if (is_neg != nullptr && pos_index != nullptr) {
923 // Emulate effect of Parse::adjust_map_after_if.
924 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
925 (*pos_index) = _gvn.transform(ccast);
926 }
927 return is_neg;
928 }
929
930 // Make sure that 'position' is a valid limit index, in [0..length].
931 // There are two equivalent plans for checking this:
932 // A. (offset + copyLength) unsigned<= arrayLength
933 // B. offset <= (arrayLength - copyLength)
934 // We require that all of the values above, except for the sum and
935 // difference, are already known to be non-negative.
936 // Plan A is robust in the face of overflow, if offset and copyLength
937 // are both hugely positive.
938 //
939 // Plan B is less direct and intuitive, but it does not overflow at
940 // all, since the difference of two non-negatives is always
941 // representable. Whenever Java methods must perform the equivalent
942 // check they generally use Plan B instead of Plan A.
943 // For the moment we use Plan A.
944 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
945 Node* subseq_length,
946 Node* array_length,
947 RegionNode* region,
948 bool with_opaque) {
949 if (stopped())
950 return nullptr; // already stopped
951 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
952 if (zero_offset && subseq_length->eqv_uncast(array_length))
953 return nullptr; // common case of whole-array copy
954 Node* last = subseq_length;
955 if (!zero_offset) // last += offset
956 last = _gvn.transform(new AddINode(last, offset));
957 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
958 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
959 if (with_opaque) {
960 bol_lt = _gvn.transform(new OpaqueConstantBoolNode(C, bol_lt, false));
961 }
962 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
963 return is_over;
964 }
965
966 // Emit range checks for the given String.value byte array
967 void LibraryCallKit::generate_string_range_check(Node* array,
968 Node* offset,
969 Node* count,
970 bool char_count,
971 RegionNode* region) {
972 if (stopped()) {
973 return; // already stopped
974 }
975 if (char_count) {
976 // Convert char count to byte count
977 count = _gvn.transform(new LShiftINode(count, intcon(1)));
978 }
979 // Offset and count must not be negative
980 generate_negative_guard(offset, region, nullptr, true);
981 generate_negative_guard(count, region, nullptr, true);
982 // Offset + count must not exceed length of array
983 generate_limit_guard(offset, count, load_array_length(array), region, true);
984 }
985
986 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
987 bool is_immutable) {
988 ciKlass* thread_klass = env()->Thread_klass();
989 const Type* thread_type
990 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
991
992 Node* thread = _gvn.transform(new ThreadLocalNode());
993 Node* p = off_heap_plus_addr(thread, in_bytes(handle_offset));
994 tls_output = thread;
995
996 Node* thread_obj_handle
997 = (is_immutable
998 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
999 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
1000 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
1001 thread_obj_handle = _gvn.transform(thread_obj_handle);
1002
1003 DecoratorSet decorators = IN_NATIVE;
1004 if (is_immutable) {
1005 decorators |= C2_IMMUTABLE_MEMORY;
1006 }
1007 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
1008 }
1009
1010 //--------------------------generate_current_thread--------------------
1011 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1012 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
1013 /*is_immutable*/false);
1014 }
1015
1016 //--------------------------generate_virtual_thread--------------------
1017 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
1018 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
1019 !C->method()->changes_current_thread());
1020 }
1021
1022 //------------------------------make_string_method_node------------------------
1023 // Helper method for String intrinsic functions. This version is called with
1024 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1025 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1026 // containing the lengths of str1 and str2.
1027 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1028 Node* result = nullptr;
1029 switch (opcode) {
1030 case Op_StrIndexOf:
1031 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1032 str1_start, cnt1, str2_start, cnt2, ae);
1033 break;
1034 case Op_StrComp:
1035 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1036 str1_start, cnt1, str2_start, cnt2, ae);
1037 break;
1038 case Op_StrEquals:
1039 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1040 // Use the constant length if there is one because optimized match rule may exist.
1041 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1042 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1043 break;
1044 default:
1045 ShouldNotReachHere();
1046 return nullptr;
1047 }
1048
1049 // All these intrinsics have checks.
1050 C->set_has_split_ifs(true); // Has chance for split-if optimization
1051 clear_upper_avx();
1052
1053 return _gvn.transform(result);
1054 }
1055
1056 //------------------------------inline_string_compareTo------------------------
1057 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1058 Node* arg1 = argument(0);
1059 Node* arg2 = argument(1);
1060
1061 arg1 = must_be_not_null(arg1, true);
1062 arg2 = must_be_not_null(arg2, true);
1063
1064 // Get start addr and length of first argument
1065 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1066 Node* arg1_cnt = load_array_length(arg1);
1067
1068 // Get start addr and length of second argument
1069 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1070 Node* arg2_cnt = load_array_length(arg2);
1071
1072 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1073 set_result(result);
1074 return true;
1075 }
1076
1077 //------------------------------inline_string_equals------------------------
1078 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1079 Node* arg1 = argument(0);
1080 Node* arg2 = argument(1);
1081
1082 // paths (plus control) merge
1083 RegionNode* region = new RegionNode(3);
1084 Node* phi = new PhiNode(region, TypeInt::BOOL);
1085
1086 if (!stopped()) {
1087
1088 arg1 = must_be_not_null(arg1, true);
1089 arg2 = must_be_not_null(arg2, true);
1090
1091 // Get start addr and length of first argument
1092 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1093 Node* arg1_cnt = load_array_length(arg1);
1094
1095 // Get start addr and length of second argument
1096 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1097 Node* arg2_cnt = load_array_length(arg2);
1098
1099 // Check for arg1_cnt != arg2_cnt
1100 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1101 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1102 Node* if_ne = generate_slow_guard(bol, nullptr);
1103 if (if_ne != nullptr) {
1104 phi->init_req(2, intcon(0));
1105 region->init_req(2, if_ne);
1106 }
1107
1108 // Check for count == 0 is done by assembler code for StrEquals.
1109
1110 if (!stopped()) {
1111 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1112 phi->init_req(1, equals);
1113 region->init_req(1, control());
1114 }
1115 }
1116
1117 // post merge
1118 set_control(_gvn.transform(region));
1119 record_for_igvn(region);
1120
1121 set_result(_gvn.transform(phi));
1122 return true;
1123 }
1124
1125 //------------------------------inline_array_equals----------------------------
1126 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1127 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1128 Node* arg1 = argument(0);
1129 Node* arg2 = argument(1);
1130
1131 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1132 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1133 clear_upper_avx();
1134
1135 return true;
1136 }
1137
1138
1139 //------------------------------inline_countPositives------------------------------
1140 // int java.lang.StringCoding#countPositives0(byte[] ba, int off, int len)
1141 bool LibraryCallKit::inline_countPositives() {
1142 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1143 // no receiver since it is static method
1144 Node* ba = argument(0);
1145 Node* offset = argument(1);
1146 Node* len = argument(2);
1147
1148 ba = must_be_not_null(ba, true);
1149 RegionNode* bailout = create_bailout();
1150 generate_string_range_check(ba, offset, len, false, bailout);
1151 if (check_bailout(bailout)) {
1152 return true;
1153 }
1154
1155 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1156 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1157 set_result(_gvn.transform(result));
1158 clear_upper_avx();
1159 return true;
1160 }
1161
1162 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1163 Node* index = argument(0);
1164 Node* length = bt == T_INT ? argument(1) : argument(2);
1165 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1166 return false;
1167 }
1168
1169 // check that length is positive
1170 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1171 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1172
1173 {
1174 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1175 uncommon_trap(Deoptimization::Reason_intrinsic,
1176 Deoptimization::Action_make_not_entrant);
1177 }
1178
1179 if (stopped()) {
1180 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1181 return true;
1182 }
1183
1184 // length is now known positive, add a cast node to make this explicit
1185 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1186 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1187 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1188 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1189 casted_length = _gvn.transform(casted_length);
1190 replace_in_map(length, casted_length);
1191 length = casted_length;
1192
1193 // Use an unsigned comparison for the range check itself
1194 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1195 BoolTest::mask btest = BoolTest::lt;
1196 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1197 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1198 _gvn.set_type(rc, rc->Value(&_gvn));
1199 if (!rc_bool->is_Con()) {
1200 record_for_igvn(rc);
1201 }
1202 set_control(_gvn.transform(new IfTrueNode(rc)));
1203 {
1204 PreserveJVMState pjvms(this);
1205 set_control(_gvn.transform(new IfFalseNode(rc)));
1206 uncommon_trap(Deoptimization::Reason_range_check,
1207 Deoptimization::Action_make_not_entrant);
1208 }
1209
1210 if (stopped()) {
1211 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1212 return true;
1213 }
1214
1215 // index is now known to be >= 0 and < length, cast it
1216 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1217 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1218 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1219 result = _gvn.transform(result);
1220 set_result(result);
1221 replace_in_map(index, result);
1222 return true;
1223 }
1224
1225 //------------------------------inline_string_indexOf------------------------
1226 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1227 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1228 return false;
1229 }
1230 Node* src = argument(0);
1231 Node* tgt = argument(1);
1232
1233 // Make the merge point
1234 RegionNode* result_rgn = new RegionNode(4);
1235 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1236
1237 src = must_be_not_null(src, true);
1238 tgt = must_be_not_null(tgt, true);
1239
1240 // Get start addr and length of source string
1241 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1242 Node* src_count = load_array_length(src);
1243
1244 // Get start addr and length of substring
1245 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1246 Node* tgt_count = load_array_length(tgt);
1247
1248 Node* result = nullptr;
1249 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1250
1251 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1252 // Divide src size by 2 if String is UTF16 encoded
1253 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1254 }
1255 if (ae == StrIntrinsicNode::UU) {
1256 // Divide substring size by 2 if String is UTF16 encoded
1257 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1258 }
1259
1260 if (call_opt_stub) {
1261 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1262 StubRoutines::_string_indexof_array[ae],
1263 "stringIndexOf", TypePtr::BOTTOM, src_start,
1264 src_count, tgt_start, tgt_count);
1265 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1266 } else {
1267 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1268 result_rgn, result_phi, ae);
1269 }
1270 if (result != nullptr) {
1271 result_phi->init_req(3, result);
1272 result_rgn->init_req(3, control());
1273 }
1274 set_control(_gvn.transform(result_rgn));
1275 record_for_igvn(result_rgn);
1276 set_result(_gvn.transform(result_phi));
1277
1278 return true;
1279 }
1280
1281 //-----------------------------inline_string_indexOfI-----------------------
1282 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1283 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1284 return false;
1285 }
1286
1287 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1288 Node* src = argument(0); // byte[]
1289 Node* src_count = argument(1); // char count
1290 Node* tgt = argument(2); // byte[]
1291 Node* tgt_count = argument(3); // char count
1292 Node* from_index = argument(4); // char index
1293
1294 src = must_be_not_null(src, true);
1295 tgt = must_be_not_null(tgt, true);
1296
1297 // Multiply byte array index by 2 if String is UTF16 encoded
1298 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1299 src_count = _gvn.transform(new SubINode(src_count, from_index));
1300 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1301 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1302
1303 // Range checks
1304 RegionNode* bailout = create_bailout();
1305 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL, bailout);
1306 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU, bailout);
1307 if (check_bailout(bailout)) {
1308 return true;
1309 }
1310
1311 RegionNode* region = new RegionNode(5);
1312 Node* phi = new PhiNode(region, TypeInt::INT);
1313 Node* result = nullptr;
1314
1315 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1316
1317 if (call_opt_stub) {
1318 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1319 StubRoutines::_string_indexof_array[ae],
1320 "stringIndexOf", TypePtr::BOTTOM, src_start,
1321 src_count, tgt_start, tgt_count);
1322 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1323 } else {
1324 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1325 region, phi, ae);
1326 }
1327 if (result != nullptr) {
1328 // The result is index relative to from_index if substring was found, -1 otherwise.
1329 // Generate code which will fold into cmove.
1330 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1331 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1332
1333 Node* if_lt = generate_slow_guard(bol, nullptr);
1334 if (if_lt != nullptr) {
1335 // result == -1
1336 phi->init_req(3, result);
1337 region->init_req(3, if_lt);
1338 }
1339 if (!stopped()) {
1340 result = _gvn.transform(new AddINode(result, from_index));
1341 phi->init_req(4, result);
1342 region->init_req(4, control());
1343 }
1344 }
1345
1346 set_control(_gvn.transform(region));
1347 record_for_igvn(region);
1348 set_result(_gvn.transform(phi));
1349 clear_upper_avx();
1350
1351 return true;
1352 }
1353
1354 // Create StrIndexOfNode with fast path checks
1355 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1356 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1357 // Check for substr count > string count
1358 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1359 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1360 Node* if_gt = generate_slow_guard(bol, nullptr);
1361 if (if_gt != nullptr) {
1362 phi->init_req(1, intcon(-1));
1363 region->init_req(1, if_gt);
1364 }
1365 if (!stopped()) {
1366 // Check for substr count == 0
1367 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1368 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1369 Node* if_zero = generate_slow_guard(bol, nullptr);
1370 if (if_zero != nullptr) {
1371 phi->init_req(2, intcon(0));
1372 region->init_req(2, if_zero);
1373 }
1374 }
1375 if (!stopped()) {
1376 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1377 }
1378 return nullptr;
1379 }
1380
1381 //-----------------------------inline_string_indexOfChar-----------------------
1382 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1383 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1384 return false;
1385 }
1386 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1387 return false;
1388 }
1389 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1390 Node* src = argument(0); // byte[]
1391 Node* int_ch = argument(1);
1392 Node* from_index = argument(2);
1393 Node* max = argument(3);
1394
1395 src = must_be_not_null(src, true);
1396
1397 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1398 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1399 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1400
1401 // Range checks
1402 RegionNode* bailout = create_bailout();
1403 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U, bailout);
1404 if (check_bailout(bailout)) {
1405 return true;
1406 }
1407
1408 // Check for int_ch >= 0
1409 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1410 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1411 {
1412 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1413 uncommon_trap(Deoptimization::Reason_intrinsic,
1414 Deoptimization::Action_maybe_recompile);
1415 }
1416 if (stopped()) {
1417 return true;
1418 }
1419
1420 RegionNode* region = new RegionNode(3);
1421 Node* phi = new PhiNode(region, TypeInt::INT);
1422
1423 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1424 C->set_has_split_ifs(true); // Has chance for split-if optimization
1425 _gvn.transform(result);
1426
1427 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1428 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1429
1430 Node* if_lt = generate_slow_guard(bol, nullptr);
1431 if (if_lt != nullptr) {
1432 // result == -1
1433 phi->init_req(2, result);
1434 region->init_req(2, if_lt);
1435 }
1436 if (!stopped()) {
1437 result = _gvn.transform(new AddINode(result, from_index));
1438 phi->init_req(1, result);
1439 region->init_req(1, control());
1440 }
1441 set_control(_gvn.transform(region));
1442 record_for_igvn(region);
1443 set_result(_gvn.transform(phi));
1444 clear_upper_avx();
1445
1446 return true;
1447 }
1448 //---------------------------inline_string_copy---------------------
1449 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1450 // int StringUTF16.compress0(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1451 // int StringUTF16.compress0(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1452 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1453 // void StringLatin1.inflate0(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1454 // void StringLatin1.inflate0(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1455 bool LibraryCallKit::inline_string_copy(bool compress) {
1456 int nargs = 5; // 2 oops, 3 ints
1457 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1458
1459 Node* src = argument(0);
1460 Node* src_offset = argument(1);
1461 Node* dst = argument(2);
1462 Node* dst_offset = argument(3);
1463 Node* length = argument(4);
1464
1465 // Check for allocation before we add nodes that would confuse
1466 // tightly_coupled_allocation()
1467 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1468
1469 // Figure out the size and type of the elements we will be copying.
1470 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1471 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1472 if (src_type == nullptr || dst_type == nullptr) {
1473 return false;
1474 }
1475 BasicType src_elem = src_type->elem()->array_element_basic_type();
1476 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1477 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1478 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1479 "Unsupported array types for inline_string_copy");
1480
1481 src = must_be_not_null(src, true);
1482 dst = must_be_not_null(dst, true);
1483
1484 // Convert char[] offsets to byte[] offsets
1485 bool convert_src = (compress && src_elem == T_BYTE);
1486 bool convert_dst = (!compress && dst_elem == T_BYTE);
1487 if (convert_src) {
1488 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1489 } else if (convert_dst) {
1490 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1491 }
1492
1493 // Range checks
1494 RegionNode* bailout = create_bailout();
1495 generate_string_range_check(src, src_offset, length, convert_src, bailout);
1496 generate_string_range_check(dst, dst_offset, length, convert_dst, bailout);
1497 if (check_bailout(bailout)) {
1498 return true;
1499 }
1500
1501 Node* src_start = array_element_address(src, src_offset, src_elem);
1502 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1503 // 'src_start' points to src array + scaled offset
1504 // 'dst_start' points to dst array + scaled offset
1505 Node* count = nullptr;
1506 if (compress) {
1507 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1508 } else {
1509 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1510 }
1511
1512 if (alloc != nullptr) {
1513 if (alloc->maybe_set_complete(&_gvn)) {
1514 // "You break it, you buy it."
1515 InitializeNode* init = alloc->initialization();
1516 assert(init->is_complete(), "we just did this");
1517 init->set_complete_with_arraycopy();
1518 assert(dst->is_CheckCastPP(), "sanity");
1519 assert(dst->in(0)->in(0) == init, "dest pinned");
1520 }
1521 // Do not let stores that initialize this object be reordered with
1522 // a subsequent store that would make this object accessible by
1523 // other threads.
1524 // Record what AllocateNode this StoreStore protects so that
1525 // escape analysis can go from the MemBarStoreStoreNode to the
1526 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1527 // based on the escape status of the AllocateNode.
1528 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1529 }
1530 if (compress) {
1531 set_result(_gvn.transform(count));
1532 }
1533 clear_upper_avx();
1534
1535 return true;
1536 }
1537
1538 #ifdef _LP64
1539 #define XTOP ,top() /*additional argument*/
1540 #else //_LP64
1541 #define XTOP /*no additional argument*/
1542 #endif //_LP64
1543
1544 //------------------------inline_string_toBytesU--------------------------
1545 // public static byte[] StringUTF16.toBytes0(char[] value, int off, int len)
1546 bool LibraryCallKit::inline_string_toBytesU() {
1547 // Get the arguments.
1548 assert(callee()->signature()->size() == 3, "character array encoder requires 3 arguments");
1549 Node* value = argument(0);
1550 Node* offset = argument(1);
1551 Node* length = argument(2);
1552
1553 Node* newcopy = nullptr;
1554
1555 // Set the original stack and the reexecute bit for the interpreter to reexecute
1556 // the bytecode that invokes StringUTF16.toBytes0() if deoptimization happens.
1557 { PreserveReexecuteState preexecs(this);
1558 jvms()->set_should_reexecute(true);
1559
1560 value = must_be_not_null(value, true);
1561 RegionNode* bailout = create_bailout();
1562 generate_negative_guard(offset, bailout, nullptr, true);
1563 generate_negative_guard(length, bailout, nullptr, true);
1564 generate_limit_guard(offset, length, load_array_length(value), bailout, true);
1565 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1566 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout, true);
1567 if (check_bailout(bailout)) {
1568 return true;
1569 }
1570
1571 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1572 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1573 newcopy = new_array(klass_node, size, 0); // no arguments to push
1574 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1575 guarantee(alloc != nullptr, "created above");
1576
1577 // Calculate starting addresses.
1578 Node* src_start = array_element_address(value, offset, T_CHAR);
1579 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1580
1581 // Check if dst array address is aligned to HeapWordSize
1582 bool aligned = (arrayOopDesc::base_offset_in_bytes(T_BYTE) % HeapWordSize == 0);
1583 // If true, then check if src array address is aligned to HeapWordSize
1584 if (aligned) {
1585 const TypeInt* toffset = gvn().type(offset)->is_int();
1586 aligned = toffset->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) +
1587 toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1588 }
1589
1590 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1591 const char* copyfunc_name = "arraycopy";
1592 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1593 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1594 OptoRuntime::fast_arraycopy_Type(),
1595 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1596 src_start, dst_start, ConvI2X(length) XTOP);
1597 // Do not let reads from the cloned object float above the arraycopy.
1598 if (alloc->maybe_set_complete(&_gvn)) {
1599 // "You break it, you buy it."
1600 InitializeNode* init = alloc->initialization();
1601 assert(init->is_complete(), "we just did this");
1602 init->set_complete_with_arraycopy();
1603 assert(newcopy->is_CheckCastPP(), "sanity");
1604 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1605 }
1606 // Do not let stores that initialize this object be reordered with
1607 // a subsequent store that would make this object accessible by
1608 // other threads.
1609 // Record what AllocateNode this StoreStore protects so that
1610 // escape analysis can go from the MemBarStoreStoreNode to the
1611 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1612 // based on the escape status of the AllocateNode.
1613 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1614 } // original reexecute is set back here
1615
1616 C->set_has_split_ifs(true); // Has chance for split-if optimization
1617 if (!stopped()) {
1618 set_result(newcopy);
1619 }
1620 clear_upper_avx();
1621
1622 return true;
1623 }
1624
1625 //------------------------inline_string_getCharsU--------------------------
1626 // public void StringUTF16.getChars0(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1627 bool LibraryCallKit::inline_string_getCharsU() {
1628 assert(callee()->signature()->size() == 5, "StringUTF16.getChars0() has 5 arguments");
1629 // Get the arguments.
1630 Node* src = argument(0);
1631 Node* src_begin = argument(1);
1632 Node* src_end = argument(2); // exclusive offset (i < src_end)
1633 Node* dst = argument(3);
1634 Node* dst_begin = argument(4);
1635
1636 // Check for allocation before we add nodes that would confuse
1637 // tightly_coupled_allocation()
1638 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1639
1640 // Check if a null path was taken unconditionally.
1641 src = must_be_not_null(src, true);
1642 dst = must_be_not_null(dst, true);
1643 if (stopped()) {
1644 return true;
1645 }
1646
1647 // Get length and convert char[] offset to byte[] offset
1648 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1649 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1650
1651 // Range checks
1652 RegionNode* bailout = create_bailout();
1653 generate_string_range_check(src, src_begin, length, true, bailout);
1654 generate_string_range_check(dst, dst_begin, length, false, bailout);
1655 if (check_bailout(bailout)) {
1656 return true;
1657 }
1658
1659 // Calculate starting addresses.
1660 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1661 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1662
1663 // Check if array addresses are aligned to HeapWordSize
1664 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1665 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1666 bool aligned = tsrc->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_BYTE) + tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1667 tdst->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) + tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1668
1669 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1670 const char* copyfunc_name = "arraycopy";
1671 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1672 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1673 OptoRuntime::fast_arraycopy_Type(),
1674 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1675 src_start, dst_start, ConvI2X(length) XTOP);
1676 // Do not let reads from the cloned object float above the arraycopy.
1677 if (alloc != nullptr) {
1678 if (alloc->maybe_set_complete(&_gvn)) {
1679 // "You break it, you buy it."
1680 InitializeNode* init = alloc->initialization();
1681 assert(init->is_complete(), "we just did this");
1682 init->set_complete_with_arraycopy();
1683 assert(dst->is_CheckCastPP(), "sanity");
1684 assert(dst->in(0)->in(0) == init, "dest pinned");
1685 }
1686 // Do not let stores that initialize this object be reordered with
1687 // a subsequent store that would make this object accessible by
1688 // other threads.
1689 // Record what AllocateNode this StoreStore protects so that
1690 // escape analysis can go from the MemBarStoreStoreNode to the
1691 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1692 // based on the escape status of the AllocateNode.
1693 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1694 } else {
1695 insert_mem_bar(Op_MemBarCPUOrder);
1696 }
1697
1698 C->set_has_split_ifs(true); // Has chance for split-if optimization
1699 return true;
1700 }
1701
1702 //----------------------inline_string_char_access----------------------------
1703 // Store/Load char to/from byte[] array.
1704 // static void StringUTF16.putChar(byte[] val, int index, int c)
1705 // static char StringUTF16.getChar(byte[] val, int index)
1706 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1707 Node* ch;
1708 if (is_store) {
1709 assert(callee()->signature()->size() == 3, "StringUTF16.putChar() has 3 arguments");
1710 ch = argument(2);
1711 } else {
1712 assert(callee()->signature()->size() == 2, "StringUTF16.getChar() has 2 arguments");
1713 ch = nullptr;
1714 }
1715 Node* value = argument(0);
1716 Node* index = argument(1);
1717
1718 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1719 // correctly requires matched array shapes.
1720 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1721 "sanity: byte[] and char[] bases agree");
1722 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1723 "sanity: byte[] and char[] scales agree");
1724
1725 // Bail when getChar over constants is requested: constant folding would
1726 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1727 // Java method would constant fold nicely instead.
1728 if (!is_store && value->is_Con() && index->is_Con()) {
1729 return false;
1730 }
1731
1732 // Save state and restore on bailout
1733 SavedState old_state(this);
1734
1735 value = must_be_not_null(value, true);
1736
1737 Node* adr = array_element_address(value, index, T_CHAR);
1738 if (adr->is_top()) {
1739 return false;
1740 }
1741 old_state.discard();
1742 if (is_store) {
1743 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1744 } else {
1745 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);
1746 set_result(ch);
1747 }
1748 return true;
1749 }
1750
1751
1752 //------------------------------inline_math-----------------------------------
1753 // public static double Math.abs(double)
1754 // public static double Math.sqrt(double)
1755 // public static double Math.log(double)
1756 // public static double Math.log10(double)
1757 // public static double Math.round(double)
1758 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1759 Node* arg = argument(0);
1760 Node* n = nullptr;
1761 switch (id) {
1762 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1763 case vmIntrinsics::_dsqrt:
1764 case vmIntrinsics::_dsqrt_strict:
1765 n = new SqrtDNode(C, control(), arg); break;
1766 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1767 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1768 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1769 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1770 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1771 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1772 default: fatal_unexpected_iid(id); break;
1773 }
1774 set_result(_gvn.transform(n));
1775 return true;
1776 }
1777
1778 //------------------------------inline_math-----------------------------------
1779 // public static float Math.abs(float)
1780 // public static int Math.abs(int)
1781 // public static long Math.abs(long)
1782 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1783 Node* arg = argument(0);
1784 Node* n = nullptr;
1785 switch (id) {
1786 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1787 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1788 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1789 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1790 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1791 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1792 default: fatal_unexpected_iid(id); break;
1793 }
1794 set_result(_gvn.transform(n));
1795 return true;
1796 }
1797
1798 //------------------------------runtime_math-----------------------------
1799 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1800 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1801 "must be (DD)D or (D)D type");
1802
1803 // Inputs
1804 Node* a = argument(0);
1805 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1806
1807 const TypePtr* no_memory_effects = nullptr;
1808 Node* trig = make_runtime_call(RC_LEAF | RC_PURE, call_type, funcAddr, funcName,
1809 no_memory_effects,
1810 a, top(), b, b ? top() : nullptr);
1811 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1812 #ifdef ASSERT
1813 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1814 assert(value_top == top(), "second value must be top");
1815 #endif
1816
1817 set_result(value);
1818 return true;
1819 }
1820
1821 //------------------------------inline_math_pow-----------------------------
1822 bool LibraryCallKit::inline_math_pow() {
1823 Node* base = argument(0);
1824 Node* exp = argument(2);
1825
1826 CallNode* pow = new PowDNode(C, base, exp);
1827 set_predefined_input_for_runtime_call(pow);
1828 pow = _gvn.transform(pow)->as_CallLeafPure();
1829 set_predefined_output_for_runtime_call(pow);
1830 Node* result = _gvn.transform(new ProjNode(pow, TypeFunc::Parms + 0));
1831 record_for_igvn(pow);
1832 set_result(result);
1833 return true;
1834 }
1835
1836 //------------------------------inline_math_native-----------------------------
1837 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1838 switch (id) {
1839 case vmIntrinsics::_dsin:
1840 return StubRoutines::dsin() != nullptr ?
1841 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1842 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1843 case vmIntrinsics::_dcos:
1844 return StubRoutines::dcos() != nullptr ?
1845 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1846 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1847 case vmIntrinsics::_dtan:
1848 return StubRoutines::dtan() != nullptr ?
1849 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1850 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1851 case vmIntrinsics::_dsinh:
1852 return StubRoutines::dsinh() != nullptr ?
1853 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsinh(), "dsinh") : false;
1854 case vmIntrinsics::_dtanh:
1855 return StubRoutines::dtanh() != nullptr ?
1856 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1857 case vmIntrinsics::_dcbrt:
1858 return StubRoutines::dcbrt() != nullptr ?
1859 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcbrt(), "dcbrt") : false;
1860 case vmIntrinsics::_dexp:
1861 return StubRoutines::dexp() != nullptr ?
1862 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1863 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1864 case vmIntrinsics::_dlog:
1865 return StubRoutines::dlog() != nullptr ?
1866 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1867 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1868 case vmIntrinsics::_dlog10:
1869 return StubRoutines::dlog10() != nullptr ?
1870 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1871 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1872
1873 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1874 case vmIntrinsics::_ceil:
1875 case vmIntrinsics::_floor:
1876 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1877
1878 case vmIntrinsics::_dsqrt:
1879 case vmIntrinsics::_dsqrt_strict:
1880 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1881 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1882 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1883 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1884 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1885
1886 case vmIntrinsics::_dpow: return inline_math_pow();
1887 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1888 case vmIntrinsics::_fcopySign: return inline_math(id);
1889 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1890 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1891 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1892
1893 // These intrinsics are not yet correctly implemented
1894 case vmIntrinsics::_datan2:
1895 return false;
1896
1897 default:
1898 fatal_unexpected_iid(id);
1899 return false;
1900 }
1901 }
1902
1903 //----------------------------inline_notify-----------------------------------*
1904 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1905 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1906 address func;
1907 if (id == vmIntrinsics::_notify) {
1908 func = OptoRuntime::monitor_notify_Java();
1909 } else {
1910 func = OptoRuntime::monitor_notifyAll_Java();
1911 }
1912 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1913 make_slow_call_ex(call, env()->Throwable_klass(), false);
1914 return true;
1915 }
1916
1917
1918 //----------------------------inline_min_max-----------------------------------
1919 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1920 Node* a = nullptr;
1921 Node* b = nullptr;
1922 Node* n = nullptr;
1923 switch (id) {
1924 case vmIntrinsics::_min:
1925 case vmIntrinsics::_max:
1926 case vmIntrinsics::_minF:
1927 case vmIntrinsics::_maxF:
1928 case vmIntrinsics::_minF_strict:
1929 case vmIntrinsics::_maxF_strict:
1930 case vmIntrinsics::_min_strict:
1931 case vmIntrinsics::_max_strict:
1932 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
1933 a = argument(0);
1934 b = argument(1);
1935 break;
1936 case vmIntrinsics::_minD:
1937 case vmIntrinsics::_maxD:
1938 case vmIntrinsics::_minD_strict:
1939 case vmIntrinsics::_maxD_strict:
1940 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
1941 a = argument(0);
1942 b = argument(2);
1943 break;
1944 case vmIntrinsics::_minL:
1945 case vmIntrinsics::_maxL:
1946 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
1947 a = argument(0);
1948 b = argument(2);
1949 break;
1950 default:
1951 fatal_unexpected_iid(id);
1952 break;
1953 }
1954
1955 switch (id) {
1956 case vmIntrinsics::_min:
1957 case vmIntrinsics::_min_strict:
1958 n = new MinINode(a, b);
1959 break;
1960 case vmIntrinsics::_max:
1961 case vmIntrinsics::_max_strict:
1962 n = new MaxINode(a, b);
1963 break;
1964 case vmIntrinsics::_minF:
1965 case vmIntrinsics::_minF_strict:
1966 n = new MinFNode(a, b);
1967 break;
1968 case vmIntrinsics::_maxF:
1969 case vmIntrinsics::_maxF_strict:
1970 n = new MaxFNode(a, b);
1971 break;
1972 case vmIntrinsics::_minD:
1973 case vmIntrinsics::_minD_strict:
1974 n = new MinDNode(a, b);
1975 break;
1976 case vmIntrinsics::_maxD:
1977 case vmIntrinsics::_maxD_strict:
1978 n = new MaxDNode(a, b);
1979 break;
1980 case vmIntrinsics::_minL:
1981 n = new MinLNode(_gvn.C, a, b);
1982 break;
1983 case vmIntrinsics::_maxL:
1984 n = new MaxLNode(_gvn.C, a, b);
1985 break;
1986 default:
1987 fatal_unexpected_iid(id);
1988 break;
1989 }
1990
1991 set_result(_gvn.transform(n));
1992 return true;
1993 }
1994
1995 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
1996 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
1997 env()->ArithmeticException_instance())) {
1998 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
1999 // so let's bail out intrinsic rather than risking deopting again.
2000 return false;
2001 }
2002
2003 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2004 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2005 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2006 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2007
2008 {
2009 PreserveJVMState pjvms(this);
2010 PreserveReexecuteState preexecs(this);
2011 jvms()->set_should_reexecute(true);
2012
2013 set_control(slow_path);
2014 set_i_o(i_o());
2015
2016 builtin_throw(Deoptimization::Reason_intrinsic,
2017 env()->ArithmeticException_instance(),
2018 /*allow_too_many_traps*/ false);
2019 }
2020
2021 set_control(fast_path);
2022 set_result(math);
2023 return true;
2024 }
2025
2026 template <typename OverflowOp>
2027 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2028 typedef typename OverflowOp::MathOp MathOp;
2029
2030 MathOp* mathOp = new MathOp(arg1, arg2);
2031 Node* operation = _gvn.transform( mathOp );
2032 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2033 return inline_math_mathExact(operation, ofcheck);
2034 }
2035
2036 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2037 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2038 }
2039
2040 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2041 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2042 }
2043
2044 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2045 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2046 }
2047
2048 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2049 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2050 }
2051
2052 bool LibraryCallKit::inline_math_negateExactI() {
2053 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2054 }
2055
2056 bool LibraryCallKit::inline_math_negateExactL() {
2057 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2058 }
2059
2060 bool LibraryCallKit::inline_math_multiplyExactI() {
2061 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2062 }
2063
2064 bool LibraryCallKit::inline_math_multiplyExactL() {
2065 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2066 }
2067
2068 bool LibraryCallKit::inline_math_multiplyHigh() {
2069 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2070 return true;
2071 }
2072
2073 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2074 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2075 return true;
2076 }
2077
2078 inline int
2079 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2080 const TypePtr* base_type = TypePtr::NULL_PTR;
2081 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2082 if (base_type == nullptr) {
2083 // Unknown type.
2084 return Type::AnyPtr;
2085 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2086 // Since this is a null+long form, we have to switch to a rawptr.
2087 base = _gvn.transform(new CastX2PNode(offset));
2088 offset = MakeConX(0);
2089 return Type::RawPtr;
2090 } else if (base_type->base() == Type::RawPtr) {
2091 return Type::RawPtr;
2092 } else if (base_type->isa_oopptr()) {
2093 // Base is never null => always a heap address.
2094 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2095 return Type::OopPtr;
2096 }
2097 // Offset is small => always a heap address.
2098 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2099 if (offset_type != nullptr &&
2100 base_type->offset() == 0 && // (should always be?)
2101 offset_type->_lo >= 0 &&
2102 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2103 return Type::OopPtr;
2104 } else if (type == T_OBJECT) {
2105 // off heap access to an oop doesn't make any sense. Has to be on
2106 // heap.
2107 return Type::OopPtr;
2108 }
2109 // Otherwise, it might either be oop+off or null+addr.
2110 return Type::AnyPtr;
2111 } else {
2112 // No information:
2113 return Type::AnyPtr;
2114 }
2115 }
2116
2117 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2118 Node* uncasted_base = base;
2119 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2120 if (kind == Type::RawPtr) {
2121 return off_heap_plus_addr(uncasted_base, offset);
2122 } else if (kind == Type::AnyPtr) {
2123 assert(base == uncasted_base, "unexpected base change");
2124 if (can_cast) {
2125 if (!_gvn.type(base)->speculative_maybe_null() &&
2126 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2127 // According to profiling, this access is always on
2128 // heap. Casting the base to not null and thus avoiding membars
2129 // around the access should allow better optimizations
2130 Node* null_ctl = top();
2131 base = null_check_oop(base, &null_ctl, true, true, true);
2132 assert(null_ctl->is_top(), "no null control here");
2133 return basic_plus_adr(base, offset);
2134 } else if (_gvn.type(base)->speculative_always_null() &&
2135 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2136 // According to profiling, this access is always off
2137 // heap.
2138 base = null_assert(base);
2139 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2140 offset = MakeConX(0);
2141 return off_heap_plus_addr(raw_base, offset);
2142 }
2143 }
2144 // We don't know if it's an on heap or off heap access. Fall back
2145 // to raw memory access.
2146 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2147 return off_heap_plus_addr(raw, offset);
2148 } else {
2149 assert(base == uncasted_base, "unexpected base change");
2150 // We know it's an on heap access so base can't be null
2151 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2152 base = must_be_not_null(base, true);
2153 }
2154 return basic_plus_adr(base, offset);
2155 }
2156 }
2157
2158 //--------------------------inline_number_methods-----------------------------
2159 // inline int Integer.numberOfLeadingZeros(int)
2160 // inline int Long.numberOfLeadingZeros(long)
2161 //
2162 // inline int Integer.numberOfTrailingZeros(int)
2163 // inline int Long.numberOfTrailingZeros(long)
2164 //
2165 // inline int Integer.bitCount(int)
2166 // inline int Long.bitCount(long)
2167 //
2168 // inline char Character.reverseBytes(char)
2169 // inline short Short.reverseBytes(short)
2170 // inline int Integer.reverseBytes(int)
2171 // inline long Long.reverseBytes(long)
2172 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2173 Node* arg = argument(0);
2174 Node* n = nullptr;
2175 switch (id) {
2176 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2177 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2178 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2179 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2180 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2181 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2182 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2183 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2184 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2185 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2186 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2187 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2188 default: fatal_unexpected_iid(id); break;
2189 }
2190 set_result(_gvn.transform(n));
2191 return true;
2192 }
2193
2194 //--------------------------inline_bitshuffle_methods-----------------------------
2195 // inline int Integer.compress(int, int)
2196 // inline int Integer.expand(int, int)
2197 // inline long Long.compress(long, long)
2198 // inline long Long.expand(long, long)
2199 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2200 Node* n = nullptr;
2201 switch (id) {
2202 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2203 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2204 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2205 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2206 default: fatal_unexpected_iid(id); break;
2207 }
2208 set_result(_gvn.transform(n));
2209 return true;
2210 }
2211
2212 //--------------------------inline_number_methods-----------------------------
2213 // inline int Integer.compareUnsigned(int, int)
2214 // inline int Long.compareUnsigned(long, long)
2215 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2216 Node* arg1 = argument(0);
2217 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2218 Node* n = nullptr;
2219 switch (id) {
2220 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2221 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2222 default: fatal_unexpected_iid(id); break;
2223 }
2224 set_result(_gvn.transform(n));
2225 return true;
2226 }
2227
2228 //--------------------------inline_unsigned_divmod_methods-----------------------------
2229 // inline int Integer.divideUnsigned(int, int)
2230 // inline int Integer.remainderUnsigned(int, int)
2231 // inline long Long.divideUnsigned(long, long)
2232 // inline long Long.remainderUnsigned(long, long)
2233 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2234 Node* n = nullptr;
2235 switch (id) {
2236 case vmIntrinsics::_divideUnsigned_i: {
2237 zero_check_int(argument(1));
2238 // Compile-time detect of null-exception
2239 if (stopped()) {
2240 return true; // keep the graph constructed so far
2241 }
2242 n = new UDivINode(control(), argument(0), argument(1));
2243 break;
2244 }
2245 case vmIntrinsics::_divideUnsigned_l: {
2246 zero_check_long(argument(2));
2247 // Compile-time detect of null-exception
2248 if (stopped()) {
2249 return true; // keep the graph constructed so far
2250 }
2251 n = new UDivLNode(control(), argument(0), argument(2));
2252 break;
2253 }
2254 case vmIntrinsics::_remainderUnsigned_i: {
2255 zero_check_int(argument(1));
2256 // Compile-time detect of null-exception
2257 if (stopped()) {
2258 return true; // keep the graph constructed so far
2259 }
2260 n = new UModINode(control(), argument(0), argument(1));
2261 break;
2262 }
2263 case vmIntrinsics::_remainderUnsigned_l: {
2264 zero_check_long(argument(2));
2265 // Compile-time detect of null-exception
2266 if (stopped()) {
2267 return true; // keep the graph constructed so far
2268 }
2269 n = new UModLNode(control(), argument(0), argument(2));
2270 break;
2271 }
2272 default: fatal_unexpected_iid(id); break;
2273 }
2274 set_result(_gvn.transform(n));
2275 return true;
2276 }
2277
2278 //----------------------------inline_unsafe_access----------------------------
2279
2280 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2281 // Attempt to infer a sharper value type from the offset and base type.
2282 ciKlass* sharpened_klass = nullptr;
2283
2284 // See if it is an instance field, with an object type.
2285 if (alias_type->field() != nullptr) {
2286 if (alias_type->field()->type()->is_klass()) {
2287 sharpened_klass = alias_type->field()->type()->as_klass();
2288 }
2289 }
2290
2291 const TypeOopPtr* result = nullptr;
2292 // See if it is a narrow oop array.
2293 if (adr_type->isa_aryptr()) {
2294 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2295 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2296 if (elem_type != nullptr && elem_type->is_loaded()) {
2297 // Sharpen the value type.
2298 result = elem_type;
2299 }
2300 }
2301 }
2302
2303 // The sharpened class might be unloaded if there is no class loader
2304 // contraint in place.
2305 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2306 // Sharpen the value type.
2307 result = TypeOopPtr::make_from_klass(sharpened_klass);
2308 }
2309 if (result != nullptr) {
2310 #ifndef PRODUCT
2311 if (C->print_intrinsics() || C->print_inlining()) {
2312 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2313 tty->print(" sharpened value: "); result->dump(); tty->cr();
2314 }
2315 #endif
2316 }
2317 return result;
2318 }
2319
2320 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2321 switch (kind) {
2322 case Relaxed:
2323 return MO_UNORDERED;
2324 case Opaque:
2325 return MO_RELAXED;
2326 case Acquire:
2327 return MO_ACQUIRE;
2328 case Release:
2329 return MO_RELEASE;
2330 case Volatile:
2331 return MO_SEQ_CST;
2332 default:
2333 ShouldNotReachHere();
2334 return 0;
2335 }
2336 }
2337
2338 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2339 if (callee()->is_static()) return false; // caller must have the capability!
2340 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2341 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2342 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2343 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2344
2345 if (is_reference_type(type)) {
2346 decorators |= ON_UNKNOWN_OOP_REF;
2347 }
2348
2349 if (unaligned) {
2350 decorators |= C2_UNALIGNED;
2351 }
2352
2353 #ifndef PRODUCT
2354 {
2355 ResourceMark rm;
2356 // Check the signatures.
2357 ciSignature* sig = callee()->signature();
2358 #ifdef ASSERT
2359 if (!is_store) {
2360 // Object getReference(Object base, int/long offset), etc.
2361 BasicType rtype = sig->return_type()->basic_type();
2362 assert(rtype == type, "getter must return the expected value");
2363 assert(sig->count() == 2, "oop getter has 2 arguments");
2364 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2365 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2366 } else {
2367 // void putReference(Object base, int/long offset, Object x), etc.
2368 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2369 assert(sig->count() == 3, "oop putter has 3 arguments");
2370 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2371 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2372 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2373 assert(vtype == type, "putter must accept the expected value");
2374 }
2375 #endif // ASSERT
2376 }
2377 #endif //PRODUCT
2378
2379 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2380
2381 Node* receiver = argument(0); // type: oop
2382
2383 // Build address expression.
2384 Node* heap_base_oop = top();
2385
2386 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2387 Node* base = argument(1); // type: oop
2388 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2389 Node* offset = argument(2); // type: long
2390 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2391 // to be plain byte offsets, which are also the same as those accepted
2392 // by oopDesc::field_addr.
2393 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2394 "fieldOffset must be byte-scaled");
2395 // 32-bit machines ignore the high half!
2396 offset = ConvL2X(offset);
2397
2398 // Save state and restore on bailout
2399 SavedState old_state(this);
2400
2401 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2402 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2403
2404 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2405 if (type != T_OBJECT) {
2406 decorators |= IN_NATIVE; // off-heap primitive access
2407 } else {
2408 return false; // off-heap oop accesses are not supported
2409 }
2410 } else {
2411 heap_base_oop = base; // on-heap or mixed access
2412 }
2413
2414 // Can base be null? Otherwise, always on-heap access.
2415 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2416
2417 if (!can_access_non_heap) {
2418 decorators |= IN_HEAP;
2419 }
2420
2421 Node* val = is_store ? argument(4) : nullptr;
2422
2423 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2424 if (adr_type == TypePtr::NULL_PTR) {
2425 return false; // off-heap access with zero address
2426 }
2427
2428 // Try to categorize the address.
2429 Compile::AliasType* alias_type = C->alias_type(adr_type);
2430 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2431
2432 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2433 alias_type->adr_type() == TypeAryPtr::RANGE) {
2434 return false; // not supported
2435 }
2436
2437 bool mismatched = false;
2438 BasicType bt = alias_type->basic_type();
2439 if (bt != T_ILLEGAL) {
2440 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2441 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2442 // Alias type doesn't differentiate between byte[] and boolean[]).
2443 // Use address type to get the element type.
2444 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2445 }
2446 if (is_reference_type(bt, true)) {
2447 // accessing an array field with getReference is not a mismatch
2448 bt = T_OBJECT;
2449 }
2450 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2451 // Don't intrinsify mismatched object accesses
2452 return false;
2453 }
2454 mismatched = (bt != type);
2455 } else if (alias_type->adr_type()->isa_oopptr()) {
2456 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2457 }
2458
2459 old_state.discard();
2460 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2461
2462 if (mismatched) {
2463 decorators |= C2_MISMATCHED;
2464 }
2465
2466 // First guess at the value type.
2467 const Type *value_type = Type::get_const_basic_type(type);
2468
2469 // Figure out the memory ordering.
2470 decorators |= mo_decorator_for_access_kind(kind);
2471
2472 if (!is_store && type == T_OBJECT) {
2473 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2474 if (tjp != nullptr) {
2475 value_type = tjp;
2476 }
2477 }
2478
2479 receiver = null_check(receiver);
2480 if (stopped()) {
2481 return true;
2482 }
2483 // Heap pointers get a null-check from the interpreter,
2484 // as a courtesy. However, this is not guaranteed by Unsafe,
2485 // and it is not possible to fully distinguish unintended nulls
2486 // from intended ones in this API.
2487
2488 if (!is_store) {
2489 Node* p = nullptr;
2490 // Try to constant fold a load from a constant field
2491 ciField* field = alias_type->field();
2492 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) {
2493 // final or stable field
2494 p = make_constant_from_field(field, heap_base_oop);
2495 }
2496
2497 if (p == nullptr) { // Could not constant fold the load
2498 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2499 // Normalize the value returned by getBoolean in the following cases
2500 if (type == T_BOOLEAN &&
2501 (mismatched ||
2502 heap_base_oop == top() || // - heap_base_oop is null or
2503 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2504 // and the unsafe access is made to large offset
2505 // (i.e., larger than the maximum offset necessary for any
2506 // field access)
2507 ) {
2508 IdealKit ideal = IdealKit(this);
2509 #define __ ideal.
2510 IdealVariable normalized_result(ideal);
2511 __ declarations_done();
2512 __ set(normalized_result, p);
2513 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2514 __ set(normalized_result, ideal.ConI(1));
2515 ideal.end_if();
2516 final_sync(ideal);
2517 p = __ value(normalized_result);
2518 #undef __
2519 }
2520 }
2521 if (type == T_ADDRESS) {
2522 p = gvn().transform(new CastP2XNode(nullptr, p));
2523 p = ConvX2UL(p);
2524 }
2525 // The load node has the control of the preceding MemBarCPUOrder. All
2526 // following nodes will have the control of the MemBarCPUOrder inserted at
2527 // the end of this method. So, pushing the load onto the stack at a later
2528 // point is fine.
2529 set_result(p);
2530 } else {
2531 if (bt == T_ADDRESS) {
2532 // Repackage the long as a pointer.
2533 val = ConvL2X(val);
2534 val = gvn().transform(new CastX2PNode(val));
2535 }
2536 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2537 }
2538
2539 return true;
2540 }
2541
2542 //----------------------------inline_unsafe_load_store----------------------------
2543 // This method serves a couple of different customers (depending on LoadStoreKind):
2544 //
2545 // LS_cmp_swap:
2546 //
2547 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2548 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2549 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2550 //
2551 // LS_cmp_swap_weak:
2552 //
2553 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2554 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2555 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2556 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2557 //
2558 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2559 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2560 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2561 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2562 //
2563 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2564 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2565 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2566 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2567 //
2568 // LS_cmp_exchange:
2569 //
2570 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2571 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2572 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2573 //
2574 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2575 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2576 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2577 //
2578 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2579 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2580 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2581 //
2582 // LS_get_add:
2583 //
2584 // int getAndAddInt( Object o, long offset, int delta)
2585 // long getAndAddLong(Object o, long offset, long delta)
2586 //
2587 // LS_get_set:
2588 //
2589 // int getAndSet(Object o, long offset, int newValue)
2590 // long getAndSet(Object o, long offset, long newValue)
2591 // Object getAndSet(Object o, long offset, Object newValue)
2592 //
2593 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2594 // This basic scheme here is the same as inline_unsafe_access, but
2595 // differs in enough details that combining them would make the code
2596 // overly confusing. (This is a true fact! I originally combined
2597 // them, but even I was confused by it!) As much code/comments as
2598 // possible are retained from inline_unsafe_access though to make
2599 // the correspondences clearer. - dl
2600
2601 if (callee()->is_static()) return false; // caller must have the capability!
2602
2603 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2604 decorators |= mo_decorator_for_access_kind(access_kind);
2605
2606 #ifndef PRODUCT
2607 BasicType rtype;
2608 {
2609 ResourceMark rm;
2610 // Check the signatures.
2611 ciSignature* sig = callee()->signature();
2612 rtype = sig->return_type()->basic_type();
2613 switch(kind) {
2614 case LS_get_add:
2615 case LS_get_set: {
2616 // Check the signatures.
2617 #ifdef ASSERT
2618 assert(rtype == type, "get and set must return the expected type");
2619 assert(sig->count() == 3, "get and set has 3 arguments");
2620 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2621 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2622 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2623 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2624 #endif // ASSERT
2625 break;
2626 }
2627 case LS_cmp_swap:
2628 case LS_cmp_swap_weak: {
2629 // Check the signatures.
2630 #ifdef ASSERT
2631 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2632 assert(sig->count() == 4, "CAS has 4 arguments");
2633 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2634 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2635 #endif // ASSERT
2636 break;
2637 }
2638 case LS_cmp_exchange: {
2639 // Check the signatures.
2640 #ifdef ASSERT
2641 assert(rtype == type, "CAS must return the expected type");
2642 assert(sig->count() == 4, "CAS has 4 arguments");
2643 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2644 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2645 #endif // ASSERT
2646 break;
2647 }
2648 default:
2649 ShouldNotReachHere();
2650 }
2651 }
2652 #endif //PRODUCT
2653
2654 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2655
2656 // Get arguments:
2657 Node* receiver = nullptr;
2658 Node* base = nullptr;
2659 Node* offset = nullptr;
2660 Node* oldval = nullptr;
2661 Node* newval = nullptr;
2662 switch(kind) {
2663 case LS_cmp_swap:
2664 case LS_cmp_swap_weak:
2665 case LS_cmp_exchange: {
2666 const bool two_slot_type = type2size[type] == 2;
2667 receiver = argument(0); // type: oop
2668 base = argument(1); // type: oop
2669 offset = argument(2); // type: long
2670 oldval = argument(4); // type: oop, int, or long
2671 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2672 break;
2673 }
2674 case LS_get_add:
2675 case LS_get_set: {
2676 receiver = argument(0); // type: oop
2677 base = argument(1); // type: oop
2678 offset = argument(2); // type: long
2679 oldval = nullptr;
2680 newval = argument(4); // type: oop, int, or long
2681 break;
2682 }
2683 default:
2684 ShouldNotReachHere();
2685 }
2686
2687 // Build field offset expression.
2688 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2689 // to be plain byte offsets, which are also the same as those accepted
2690 // by oopDesc::field_addr.
2691 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2692 // 32-bit machines ignore the high half of long offsets
2693 offset = ConvL2X(offset);
2694 // Save state and restore on bailout
2695 SavedState old_state(this);
2696 Node* adr = make_unsafe_address(base, offset,type, false);
2697 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2698
2699 Compile::AliasType* alias_type = C->alias_type(adr_type);
2700 BasicType bt = alias_type->basic_type();
2701 if (bt != T_ILLEGAL &&
2702 (is_reference_type(bt) != (type == T_OBJECT))) {
2703 // Don't intrinsify mismatched object accesses.
2704 return false;
2705 }
2706
2707 old_state.discard();
2708
2709 // For CAS, unlike inline_unsafe_access, there seems no point in
2710 // trying to refine types. Just use the coarse types here.
2711 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2712 const Type *value_type = Type::get_const_basic_type(type);
2713
2714 switch (kind) {
2715 case LS_get_set:
2716 case LS_cmp_exchange: {
2717 if (type == T_OBJECT) {
2718 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2719 if (tjp != nullptr) {
2720 value_type = tjp;
2721 }
2722 }
2723 break;
2724 }
2725 case LS_cmp_swap:
2726 case LS_cmp_swap_weak:
2727 case LS_get_add:
2728 break;
2729 default:
2730 ShouldNotReachHere();
2731 }
2732
2733 // Null check receiver.
2734 receiver = null_check(receiver);
2735 if (stopped()) {
2736 return true;
2737 }
2738
2739 int alias_idx = C->get_alias_index(adr_type);
2740
2741 if (is_reference_type(type)) {
2742 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2743
2744 // Transformation of a value which could be null pointer (CastPP #null)
2745 // could be delayed during Parse (for example, in adjust_map_after_if()).
2746 // Execute transformation here to avoid barrier generation in such case.
2747 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2748 newval = _gvn.makecon(TypePtr::NULL_PTR);
2749
2750 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2751 // Refine the value to a null constant, when it is known to be null
2752 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2753 }
2754 }
2755
2756 Node* result = nullptr;
2757 switch (kind) {
2758 case LS_cmp_exchange: {
2759 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2760 oldval, newval, value_type, type, decorators);
2761 break;
2762 }
2763 case LS_cmp_swap_weak:
2764 decorators |= C2_WEAK_CMPXCHG;
2765 case LS_cmp_swap: {
2766 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2767 oldval, newval, value_type, type, decorators);
2768 break;
2769 }
2770 case LS_get_set: {
2771 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2772 newval, value_type, type, decorators);
2773 break;
2774 }
2775 case LS_get_add: {
2776 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2777 newval, value_type, type, decorators);
2778 break;
2779 }
2780 default:
2781 ShouldNotReachHere();
2782 }
2783
2784 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2785 set_result(result);
2786 return true;
2787 }
2788
2789 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2790 // Regardless of form, don't allow previous ld/st to move down,
2791 // then issue acquire, release, or volatile mem_bar.
2792 insert_mem_bar(Op_MemBarCPUOrder);
2793 switch(id) {
2794 case vmIntrinsics::_loadFence:
2795 insert_mem_bar(Op_LoadFence);
2796 return true;
2797 case vmIntrinsics::_storeFence:
2798 insert_mem_bar(Op_StoreFence);
2799 return true;
2800 case vmIntrinsics::_storeStoreFence:
2801 insert_mem_bar(Op_StoreStoreFence);
2802 return true;
2803 case vmIntrinsics::_fullFence:
2804 insert_mem_bar(Op_MemBarFull);
2805 return true;
2806 default:
2807 fatal_unexpected_iid(id);
2808 return false;
2809 }
2810 }
2811
2812 bool LibraryCallKit::inline_onspinwait() {
2813 insert_mem_bar(Op_OnSpinWait);
2814 return true;
2815 }
2816
2817 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2818 if (!kls->is_Con()) {
2819 return true;
2820 }
2821 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2822 if (klsptr == nullptr) {
2823 return true;
2824 }
2825 ciInstanceKlass* ik = klsptr->instance_klass();
2826 // don't need a guard for a klass that is already initialized
2827 return !ik->is_initialized();
2828 }
2829
2830 //----------------------------inline_unsafe_writeback0-------------------------
2831 // public native void Unsafe.writeback0(long address)
2832 bool LibraryCallKit::inline_unsafe_writeback0() {
2833 if (!Matcher::has_match_rule(Op_CacheWB)) {
2834 return false;
2835 }
2836 #ifndef PRODUCT
2837 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2838 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2839 ciSignature* sig = callee()->signature();
2840 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2841 #endif
2842 null_check_receiver(); // null-check, then ignore
2843 Node *addr = argument(1);
2844 addr = new CastX2PNode(addr);
2845 addr = _gvn.transform(addr);
2846 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2847 flush = _gvn.transform(flush);
2848 set_memory(flush, TypeRawPtr::BOTTOM);
2849 return true;
2850 }
2851
2852 //----------------------------inline_unsafe_writeback0-------------------------
2853 // public native void Unsafe.writeback0(long address)
2854 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2855 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2856 return false;
2857 }
2858 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2859 return false;
2860 }
2861 #ifndef PRODUCT
2862 assert(Matcher::has_match_rule(Op_CacheWB),
2863 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2864 : "found match rule for CacheWBPostSync but not CacheWB"));
2865
2866 #endif
2867 null_check_receiver(); // null-check, then ignore
2868 Node *sync;
2869 if (is_pre) {
2870 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2871 } else {
2872 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2873 }
2874 sync = _gvn.transform(sync);
2875 set_memory(sync, TypeRawPtr::BOTTOM);
2876 return true;
2877 }
2878
2879 //----------------------------inline_unsafe_allocate---------------------------
2880 // public native Object Unsafe.allocateInstance(Class<?> cls);
2881 bool LibraryCallKit::inline_unsafe_allocate() {
2882
2883 #if INCLUDE_JVMTI
2884 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
2885 return false;
2886 }
2887 #endif //INCLUDE_JVMTI
2888
2889 if (callee()->is_static()) return false; // caller must have the capability!
2890
2891 null_check_receiver(); // null-check, then ignore
2892 Node* cls = null_check(argument(1));
2893 if (stopped()) return true;
2894
2895 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
2896 kls = null_check(kls);
2897 if (stopped()) return true; // argument was like int.class
2898
2899 #if INCLUDE_JVMTI
2900 // Don't try to access new allocated obj in the intrinsic.
2901 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
2902 // Deoptimize and allocate in interpreter instead.
2903 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
2904 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
2905 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
2906 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
2907 {
2908 BuildCutout unless(this, tst, PROB_MAX);
2909 uncommon_trap(Deoptimization::Reason_intrinsic,
2910 Deoptimization::Action_make_not_entrant);
2911 }
2912 if (stopped()) {
2913 return true;
2914 }
2915 #endif //INCLUDE_JVMTI
2916
2917 Node* test = nullptr;
2918 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2919 // Note: The argument might still be an illegal value like
2920 // Serializable.class or Object[].class. The runtime will handle it.
2921 // But we must make an explicit check for initialization.
2922 Node* insp = off_heap_plus_addr(kls, in_bytes(InstanceKlass::init_state_offset()));
2923 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2924 // can generate code to load it as unsigned byte.
2925 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
2926 Node* bits = intcon(InstanceKlass::fully_initialized);
2927 test = _gvn.transform(new SubINode(inst, bits));
2928 // The 'test' is non-zero if we need to take a slow path.
2929 }
2930
2931 Node* obj = new_instance(kls, test);
2932 set_result(obj);
2933 return true;
2934 }
2935
2936 //------------------------inline_native_time_funcs--------------
2937 // inline code for System.currentTimeMillis() and System.nanoTime()
2938 // these have the same type and signature
2939 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2940 const TypeFunc* tf = OptoRuntime::void_long_Type();
2941 const TypePtr* no_memory_effects = nullptr;
2942 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2943 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2944 #ifdef ASSERT
2945 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2946 assert(value_top == top(), "second value must be top");
2947 #endif
2948 set_result(value);
2949 return true;
2950 }
2951
2952 //--------------------inline_native_vthread_start_transition--------------------
2953 // inline void startTransition(boolean is_mount);
2954 // inline void startFinalTransition();
2955 // Pseudocode of implementation:
2956 //
2957 // java_lang_Thread::set_is_in_vthread_transition(vthread, true);
2958 // carrier->set_is_in_vthread_transition(true);
2959 // OrderAccess::storeload();
2960 // int disable_requests = java_lang_Thread::vthread_transition_disable_count(vthread)
2961 // + global_vthread_transition_disable_count();
2962 // if (disable_requests > 0) {
2963 // slow path: runtime call
2964 // }
2965 bool LibraryCallKit::inline_native_vthread_start_transition(address funcAddr, const char* funcName, bool is_final_transition) {
2966 Node* vt_oop = must_be_not_null(argument(0), true); // VirtualThread this argument
2967 IdealKit ideal(this);
2968
2969 Node* thread = ideal.thread();
2970 Node* jt_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
2971 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
2972 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2973 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2974 insert_mem_bar(Op_MemBarStoreLoad);
2975 ideal.sync_kit(this);
2976
2977 Node* global_disable_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::global_vthread_transition_disable_count_address()));
2978 Node* global_disable = ideal.load(ideal.ctrl(), global_disable_addr, TypeInt::INT, T_INT, Compile::AliasIdxRaw, true /*require_atomic_access*/);
2979 Node* vt_disable_addr = basic_plus_adr(vt_oop, java_lang_Thread::vthread_transition_disable_count_offset());
2980 const TypePtr* vt_disable_addr_t = _gvn.type(vt_disable_addr)->is_ptr();
2981 Node* vt_disable = ideal.load(ideal.ctrl(), vt_disable_addr, TypeInt::INT, T_INT, C->get_alias_index(vt_disable_addr_t), true /*require_atomic_access*/);
2982 Node* disabled = _gvn.transform(new AddINode(global_disable, vt_disable));
2983
2984 ideal.if_then(disabled, BoolTest::ne, ideal.ConI(0)); {
2985 sync_kit(ideal);
2986 Node* is_mount = is_final_transition ? ideal.ConI(0) : argument(1);
2987 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
2988 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
2989 ideal.sync_kit(this);
2990 }
2991 ideal.end_if();
2992
2993 final_sync(ideal);
2994 return true;
2995 }
2996
2997 bool LibraryCallKit::inline_native_vthread_end_transition(address funcAddr, const char* funcName, bool is_first_transition) {
2998 Node* vt_oop = must_be_not_null(argument(0), true); // VirtualThread this argument
2999 IdealKit ideal(this);
3000
3001 Node* _notify_jvmti_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::notify_jvmti_events_address()));
3002 Node* _notify_jvmti = ideal.load(ideal.ctrl(), _notify_jvmti_addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3003
3004 ideal.if_then(_notify_jvmti, BoolTest::eq, ideal.ConI(1)); {
3005 sync_kit(ideal);
3006 Node* is_mount = is_first_transition ? ideal.ConI(1) : argument(1);
3007 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3008 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3009 ideal.sync_kit(this);
3010 } ideal.else_(); {
3011 Node* thread = ideal.thread();
3012 Node* jt_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3013 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3014
3015 sync_kit(ideal);
3016 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3017 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3018 ideal.sync_kit(this);
3019 } ideal.end_if();
3020
3021 final_sync(ideal);
3022 return true;
3023 }
3024
3025 #if INCLUDE_JVMTI
3026
3027 // Always update the is_disable_suspend bit.
3028 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3029 if (!DoJVMTIVirtualThreadTransitions) {
3030 return true;
3031 }
3032 IdealKit ideal(this);
3033
3034 {
3035 // unconditionally update the is_disable_suspend bit in current JavaThread
3036 Node* thread = ideal.thread();
3037 Node* arg = argument(0); // argument for notification
3038 Node* addr = off_heap_plus_addr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3039 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3040
3041 sync_kit(ideal);
3042 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3043 ideal.sync_kit(this);
3044 }
3045 final_sync(ideal);
3046
3047 return true;
3048 }
3049
3050 #endif // INCLUDE_JVMTI
3051
3052 #ifdef JFR_HAVE_INTRINSICS
3053
3054 /**
3055 * if oop->klass != null
3056 * // normal class
3057 * epoch = _epoch_state ? 2 : 1
3058 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3059 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3060 * }
3061 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3062 * else
3063 * // primitive class
3064 * if oop->array_klass != null
3065 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3066 * else
3067 * id = LAST_TYPE_ID + 1 // void class path
3068 * if (!signaled)
3069 * signaled = true
3070 */
3071 bool LibraryCallKit::inline_native_classID() {
3072 Node* cls = argument(0);
3073
3074 IdealKit ideal(this);
3075 #define __ ideal.
3076 IdealVariable result(ideal); __ declarations_done();
3077 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3078 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3079 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3080
3081
3082 __ if_then(kls, BoolTest::ne, null()); {
3083 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3084 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3085
3086 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3087 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3088 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3089 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3090 mask = _gvn.transform(new OrLNode(mask, epoch));
3091 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3092
3093 float unlikely = PROB_UNLIKELY(0.999);
3094 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3095 sync_kit(ideal);
3096 make_runtime_call(RC_LEAF,
3097 OptoRuntime::class_id_load_barrier_Type(),
3098 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3099 "class id load barrier",
3100 TypePtr::BOTTOM,
3101 kls);
3102 ideal.sync_kit(this);
3103 } __ end_if();
3104
3105 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3106 } __ else_(); {
3107 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3108 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3109 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3110 __ if_then(array_kls, BoolTest::ne, null()); {
3111 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3112 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3113 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3114 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3115 } __ else_(); {
3116 // void class case
3117 ideal.set(result, longcon(LAST_TYPE_ID + 1));
3118 } __ end_if();
3119
3120 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3121 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3122 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3123 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3124 } __ end_if();
3125 } __ end_if();
3126
3127 final_sync(ideal);
3128 set_result(ideal.value(result));
3129 #undef __
3130 return true;
3131 }
3132
3133 //------------------------inline_native_jvm_commit------------------
3134 bool LibraryCallKit::inline_native_jvm_commit() {
3135 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3136
3137 // Save input memory and i_o state.
3138 Node* input_memory_state = reset_memory();
3139 set_all_memory(input_memory_state);
3140 Node* input_io_state = i_o();
3141
3142 // TLS.
3143 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3144 // Jfr java buffer.
3145 Node* java_buffer_offset = _gvn.transform(AddPNode::make_off_heap(tls_ptr, MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR))));
3146 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3147 Node* java_buffer_pos_offset = _gvn.transform(AddPNode::make_off_heap(java_buffer, MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET))));
3148
3149 // Load the current value of the notified field in the JfrThreadLocal.
3150 Node* notified_offset = off_heap_plus_addr(tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3151 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3152
3153 // Test for notification.
3154 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3155 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3156 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3157
3158 // True branch, is notified.
3159 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3160 set_control(is_notified);
3161
3162 // Reset notified state.
3163 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3164 Node* notified_reset_memory = reset_memory();
3165
3166 // 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.
3167 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3168 // Convert the machine-word to a long.
3169 Node* current_pos = ConvX2L(current_pos_X);
3170
3171 // False branch, not notified.
3172 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3173 set_control(not_notified);
3174 set_all_memory(input_memory_state);
3175
3176 // Arg is the next position as a long.
3177 Node* arg = argument(0);
3178 // Convert long to machine-word.
3179 Node* next_pos_X = ConvL2X(arg);
3180
3181 // Store the next_position to the underlying jfr java buffer.
3182 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3183
3184 Node* commit_memory = reset_memory();
3185 set_all_memory(commit_memory);
3186
3187 // 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.
3188 Node* java_buffer_flags_offset = _gvn.transform(AddPNode::make_off_heap(java_buffer, MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET))));
3189 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3190 Node* lease_constant = _gvn.intcon(4);
3191
3192 // And flags with lease constant.
3193 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3194
3195 // Branch on lease to conditionalize returning the leased java buffer.
3196 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3197 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3198 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3199
3200 // False branch, not a lease.
3201 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3202
3203 // True branch, is lease.
3204 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3205 set_control(is_lease);
3206
3207 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3208 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3209 OptoRuntime::void_void_Type(),
3210 SharedRuntime::jfr_return_lease(),
3211 "return_lease", TypePtr::BOTTOM);
3212 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3213
3214 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3215 record_for_igvn(lease_compare_rgn);
3216 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3217 record_for_igvn(lease_compare_mem);
3218 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3219 record_for_igvn(lease_compare_io);
3220 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3221 record_for_igvn(lease_result_value);
3222
3223 // Update control and phi nodes.
3224 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3225 lease_compare_rgn->init_req(_false_path, not_lease);
3226
3227 lease_compare_mem->init_req(_true_path, reset_memory());
3228 lease_compare_mem->init_req(_false_path, commit_memory);
3229
3230 lease_compare_io->init_req(_true_path, i_o());
3231 lease_compare_io->init_req(_false_path, input_io_state);
3232
3233 lease_result_value->init_req(_true_path, _gvn.longcon(0)); // if the lease was returned, return 0L.
3234 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3235
3236 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3237 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3238 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3239 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3240
3241 // Update control and phi nodes.
3242 result_rgn->init_req(_true_path, is_notified);
3243 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3244
3245 result_mem->init_req(_true_path, notified_reset_memory);
3246 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3247
3248 result_io->init_req(_true_path, input_io_state);
3249 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3250
3251 result_value->init_req(_true_path, current_pos);
3252 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3253
3254 // Set output state.
3255 set_control(_gvn.transform(result_rgn));
3256 set_all_memory(_gvn.transform(result_mem));
3257 set_i_o(_gvn.transform(result_io));
3258 set_result(result_rgn, result_value);
3259 return true;
3260 }
3261
3262 /*
3263 * The intrinsic is a model of this pseudo-code:
3264 *
3265 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3266 * jobject h_event_writer = tl->java_event_writer();
3267 * if (h_event_writer == nullptr) {
3268 * return nullptr;
3269 * }
3270 * oop threadObj = Thread::threadObj();
3271 * oop vthread = java_lang_Thread::vthread(threadObj);
3272 * traceid tid;
3273 * bool pinVirtualThread;
3274 * bool excluded;
3275 * if (vthread != threadObj) { // i.e. current thread is virtual
3276 * tid = java_lang_Thread::tid(vthread);
3277 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3278 * pinVirtualThread = VMContinuations;
3279 * excluded = vthread_epoch_raw & excluded_mask;
3280 * if (!excluded) {
3281 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3282 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3283 * if (vthread_epoch != current_epoch) {
3284 * write_checkpoint();
3285 * }
3286 * }
3287 * } else {
3288 * tid = java_lang_Thread::tid(threadObj);
3289 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3290 * pinVirtualThread = false;
3291 * excluded = thread_epoch_raw & excluded_mask;
3292 * }
3293 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3294 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3295 * if (tid_in_event_writer != tid) {
3296 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3297 * setField(event_writer, "excluded", excluded);
3298 * setField(event_writer, "threadID", tid);
3299 * }
3300 * return event_writer
3301 */
3302 bool LibraryCallKit::inline_native_getEventWriter() {
3303 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3304
3305 // Save input memory and i_o state.
3306 Node* input_memory_state = reset_memory();
3307 set_all_memory(input_memory_state);
3308 Node* input_io_state = i_o();
3309
3310 // The most significant bit of the u2 is used to denote thread exclusion
3311 Node* excluded_shift = _gvn.intcon(15);
3312 Node* excluded_mask = _gvn.intcon(1 << 15);
3313 // The epoch generation is the range [1-32767]
3314 Node* epoch_mask = _gvn.intcon(32767);
3315
3316 // TLS
3317 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3318
3319 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3320 Node* jobj_ptr = off_heap_plus_addr(tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3321
3322 // Load the eventwriter jobject handle.
3323 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3324
3325 // Null check the jobject handle.
3326 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3327 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3328 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3329
3330 // False path, jobj is null.
3331 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3332
3333 // True path, jobj is not null.
3334 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3335
3336 set_control(jobj_is_not_null);
3337
3338 // Load the threadObj for the CarrierThread.
3339 Node* threadObj = generate_current_thread(tls_ptr);
3340
3341 // Load the vthread.
3342 Node* vthread = generate_virtual_thread(tls_ptr);
3343
3344 // If vthread != threadObj, this is a virtual thread.
3345 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3346 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3347 IfNode* iff_vthread_not_equal_threadObj =
3348 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3349
3350 // False branch, fallback to threadObj.
3351 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3352 set_control(vthread_equal_threadObj);
3353
3354 // Load the tid field from the vthread object.
3355 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3356
3357 // Load the raw epoch value from the threadObj.
3358 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3359 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3360 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3361 TypeInt::CHAR, T_CHAR,
3362 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3363
3364 // Mask off the excluded information from the epoch.
3365 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3366
3367 // True branch, this is a virtual thread.
3368 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3369 set_control(vthread_not_equal_threadObj);
3370
3371 // Load the tid field from the vthread object.
3372 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3373
3374 // Continuation support determines if a virtual thread should be pinned.
3375 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3376 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3377
3378 // Load the raw epoch value from the vthread.
3379 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3380 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3381 TypeInt::CHAR, T_CHAR,
3382 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3383
3384 // Mask off the excluded information from the epoch.
3385 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, excluded_mask));
3386
3387 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3388 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, excluded_mask));
3389 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3390 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3391
3392 // False branch, vthread is excluded, no need to write epoch info.
3393 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3394
3395 // True branch, vthread is included, update epoch info.
3396 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3397 set_control(included);
3398
3399 // Get epoch value.
3400 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, epoch_mask));
3401
3402 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3403 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3404 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3405
3406 // Compare the epoch in the vthread to the current epoch generation.
3407 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3408 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3409 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3410
3411 // False path, epoch is equal, checkpoint information is valid.
3412 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3413
3414 // True path, epoch is not equal, write a checkpoint for the vthread.
3415 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3416
3417 set_control(epoch_is_not_equal);
3418
3419 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3420 // The call also updates the native thread local thread id and the vthread with the current epoch.
3421 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3422 OptoRuntime::jfr_write_checkpoint_Type(),
3423 SharedRuntime::jfr_write_checkpoint(),
3424 "write_checkpoint", TypePtr::BOTTOM);
3425 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3426
3427 // vthread epoch != current epoch
3428 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3429 record_for_igvn(epoch_compare_rgn);
3430 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3431 record_for_igvn(epoch_compare_mem);
3432 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3433 record_for_igvn(epoch_compare_io);
3434
3435 // Update control and phi nodes.
3436 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3437 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3438 epoch_compare_mem->init_req(_true_path, reset_memory());
3439 epoch_compare_mem->init_req(_false_path, input_memory_state);
3440 epoch_compare_io->init_req(_true_path, i_o());
3441 epoch_compare_io->init_req(_false_path, input_io_state);
3442
3443 // excluded != true
3444 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3445 record_for_igvn(exclude_compare_rgn);
3446 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3447 record_for_igvn(exclude_compare_mem);
3448 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3449 record_for_igvn(exclude_compare_io);
3450
3451 // Update control and phi nodes.
3452 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3453 exclude_compare_rgn->init_req(_false_path, excluded);
3454 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3455 exclude_compare_mem->init_req(_false_path, input_memory_state);
3456 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3457 exclude_compare_io->init_req(_false_path, input_io_state);
3458
3459 // vthread != threadObj
3460 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3461 record_for_igvn(vthread_compare_rgn);
3462 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3463 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3464 record_for_igvn(vthread_compare_io);
3465 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3466 record_for_igvn(tid);
3467 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3468 record_for_igvn(exclusion);
3469 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3470 record_for_igvn(pinVirtualThread);
3471
3472 // Update control and phi nodes.
3473 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3474 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3475 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3476 vthread_compare_mem->init_req(_false_path, input_memory_state);
3477 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3478 vthread_compare_io->init_req(_false_path, input_io_state);
3479 tid->init_req(_true_path, vthread_tid);
3480 tid->init_req(_false_path, thread_obj_tid);
3481 exclusion->init_req(_true_path, vthread_is_excluded);
3482 exclusion->init_req(_false_path, threadObj_is_excluded);
3483 pinVirtualThread->init_req(_true_path, continuation_support);
3484 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3485
3486 // Update branch state.
3487 set_control(_gvn.transform(vthread_compare_rgn));
3488 set_all_memory(_gvn.transform(vthread_compare_mem));
3489 set_i_o(_gvn.transform(vthread_compare_io));
3490
3491 // Load the event writer oop by dereferencing the jobject handle.
3492 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3493 assert(klass_EventWriter->is_loaded(), "invariant");
3494 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3495 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3496 const TypeOopPtr* const xtype = aklass->as_instance_type();
3497 Node* jobj_untagged = _gvn.transform(AddPNode::make_off_heap(jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3498 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3499
3500 // Load the current thread id from the event writer object.
3501 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3502 // Get the field offset to, conditionally, store an updated tid value later.
3503 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3504 // Get the field offset to, conditionally, store an updated exclusion value later.
3505 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3506 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3507 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3508
3509 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3510 record_for_igvn(event_writer_tid_compare_rgn);
3511 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3512 record_for_igvn(event_writer_tid_compare_mem);
3513 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3514 record_for_igvn(event_writer_tid_compare_io);
3515
3516 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3517 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3518 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3519 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3520
3521 // False path, tids are the same.
3522 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3523
3524 // True path, tid is not equal, need to update the tid in the event writer.
3525 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3526 record_for_igvn(tid_is_not_equal);
3527
3528 // Store the pin state to the event writer.
3529 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3530
3531 // Store the exclusion state to the event writer.
3532 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3533 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3534
3535 // Store the tid to the event writer.
3536 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3537
3538 // Update control and phi nodes.
3539 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3540 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3541 event_writer_tid_compare_mem->init_req(_true_path, reset_memory());
3542 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3543 event_writer_tid_compare_io->init_req(_true_path, i_o());
3544 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3545
3546 // Result of top level CFG, Memory, IO and Value.
3547 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3548 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3549 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3550 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3551
3552 // Result control.
3553 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3554 result_rgn->init_req(_false_path, jobj_is_null);
3555
3556 // Result memory.
3557 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3558 result_mem->init_req(_false_path, input_memory_state);
3559
3560 // Result IO.
3561 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3562 result_io->init_req(_false_path, input_io_state);
3563
3564 // Result value.
3565 result_value->init_req(_true_path, event_writer); // return event writer oop
3566 result_value->init_req(_false_path, null()); // return null
3567
3568 // Set output state.
3569 set_control(_gvn.transform(result_rgn));
3570 set_all_memory(_gvn.transform(result_mem));
3571 set_i_o(_gvn.transform(result_io));
3572 set_result(result_rgn, result_value);
3573 return true;
3574 }
3575
3576 /*
3577 * The intrinsic is a model of this pseudo-code:
3578 *
3579 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3580 * if (carrierThread != thread) { // is virtual thread
3581 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3582 * bool excluded = vthread_epoch_raw & excluded_mask;
3583 * AtomicAccess::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3584 * AtomicAccess::store(&tl->_contextual_thread_excluded, is_excluded);
3585 * if (!excluded) {
3586 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3587 * AtomicAccess::store(&tl->_vthread_epoch, vthread_epoch);
3588 * }
3589 * AtomicAccess::release_store(&tl->_vthread, true);
3590 * return;
3591 * }
3592 * AtomicAccess::release_store(&tl->_vthread, false);
3593 */
3594 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3595 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3596
3597 Node* input_memory_state = reset_memory();
3598 set_all_memory(input_memory_state);
3599
3600 // The most significant bit of the u2 is used to denote thread exclusion
3601 Node* excluded_mask = _gvn.intcon(1 << 15);
3602 // The epoch generation is the range [1-32767]
3603 Node* epoch_mask = _gvn.intcon(32767);
3604
3605 Node* const carrierThread = generate_current_thread(jt);
3606 // If thread != carrierThread, this is a virtual thread.
3607 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3608 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3609 IfNode* iff_thread_not_equal_carrierThread =
3610 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3611
3612 Node* vthread_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3613
3614 // False branch, is carrierThread.
3615 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3616 // Store release
3617 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3618
3619 set_all_memory(input_memory_state);
3620
3621 // True branch, is virtual thread.
3622 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3623 set_control(thread_not_equal_carrierThread);
3624
3625 // Load the raw epoch value from the vthread.
3626 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3627 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
3628 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3629
3630 // Mask off the excluded information from the epoch.
3631 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, excluded_mask));
3632
3633 // Load the tid field from the thread.
3634 Node* tid = load_field_from_object(thread, "tid", "J");
3635
3636 // Store the vthread tid to the jfr thread local.
3637 Node* thread_id_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3638 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
3639
3640 // Branch is_excluded to conditionalize updating the epoch .
3641 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, excluded_mask));
3642 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3643 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3644
3645 // True branch, vthread is excluded, no need to write epoch info.
3646 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3647 set_control(excluded);
3648 Node* vthread_is_excluded = _gvn.intcon(1);
3649
3650 // False branch, vthread is included, update epoch info.
3651 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3652 set_control(included);
3653 Node* vthread_is_included = _gvn.intcon(0);
3654
3655 // Get epoch value.
3656 Node* epoch = _gvn.transform(new AndINode(epoch_raw, epoch_mask));
3657
3658 // Store the vthread epoch to the jfr thread local.
3659 Node* vthread_epoch_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3660 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
3661
3662 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3663 record_for_igvn(excluded_rgn);
3664 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3665 record_for_igvn(excluded_mem);
3666 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3667 record_for_igvn(exclusion);
3668
3669 // Merge the excluded control and memory.
3670 excluded_rgn->init_req(_true_path, excluded);
3671 excluded_rgn->init_req(_false_path, included);
3672 excluded_mem->init_req(_true_path, tid_memory);
3673 excluded_mem->init_req(_false_path, included_memory);
3674 exclusion->init_req(_true_path, vthread_is_excluded);
3675 exclusion->init_req(_false_path, vthread_is_included);
3676
3677 // Set intermediate state.
3678 set_control(_gvn.transform(excluded_rgn));
3679 set_all_memory(excluded_mem);
3680
3681 // Store the vthread exclusion state to the jfr thread local.
3682 Node* thread_local_excluded_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3683 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
3684
3685 // Store release
3686 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
3687
3688 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3689 record_for_igvn(thread_compare_rgn);
3690 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3691 record_for_igvn(thread_compare_mem);
3692 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3693 record_for_igvn(vthread);
3694
3695 // Merge the thread_compare control and memory.
3696 thread_compare_rgn->init_req(_true_path, control());
3697 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3698 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3699 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3700
3701 // Set output state.
3702 set_control(_gvn.transform(thread_compare_rgn));
3703 set_all_memory(_gvn.transform(thread_compare_mem));
3704 }
3705
3706 #endif // JFR_HAVE_INTRINSICS
3707
3708 //------------------------inline_native_currentCarrierThread------------------
3709 bool LibraryCallKit::inline_native_currentCarrierThread() {
3710 Node* junk = nullptr;
3711 set_result(generate_current_thread(junk));
3712 return true;
3713 }
3714
3715 //------------------------inline_native_currentThread------------------
3716 bool LibraryCallKit::inline_native_currentThread() {
3717 Node* junk = nullptr;
3718 set_result(generate_virtual_thread(junk));
3719 return true;
3720 }
3721
3722 //------------------------inline_native_setVthread------------------
3723 bool LibraryCallKit::inline_native_setCurrentThread() {
3724 assert(C->method()->changes_current_thread(),
3725 "method changes current Thread but is not annotated ChangesCurrentThread");
3726 Node* arr = argument(1);
3727 Node* thread = _gvn.transform(new ThreadLocalNode());
3728 Node* p = off_heap_plus_addr(thread, in_bytes(JavaThread::vthread_offset()));
3729 Node* thread_obj_handle
3730 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3731 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3732 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3733
3734 // Change the _monitor_owner_id of the JavaThread
3735 Node* tid = load_field_from_object(arr, "tid", "J");
3736 Node* monitor_owner_id_offset = off_heap_plus_addr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
3737 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
3738
3739 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3740 return true;
3741 }
3742
3743 const Type* LibraryCallKit::scopedValueCache_type() {
3744 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3745 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3746 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3747
3748 // Because we create the scopedValue cache lazily we have to make the
3749 // type of the result BotPTR.
3750 bool xk = etype->klass_is_exact();
3751 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3752 return objects_type;
3753 }
3754
3755 Node* LibraryCallKit::scopedValueCache_helper() {
3756 Node* thread = _gvn.transform(new ThreadLocalNode());
3757 Node* p = off_heap_plus_addr(thread, in_bytes(JavaThread::scopedValueCache_offset()));
3758 // We cannot use immutable_memory() because we might flip onto a
3759 // different carrier thread, at which point we'll need to use that
3760 // carrier thread's cache.
3761 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3762 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3763 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3764 }
3765
3766 //------------------------inline_native_scopedValueCache------------------
3767 bool LibraryCallKit::inline_native_scopedValueCache() {
3768 Node* cache_obj_handle = scopedValueCache_helper();
3769 const Type* objects_type = scopedValueCache_type();
3770 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3771
3772 return true;
3773 }
3774
3775 //------------------------inline_native_setScopedValueCache------------------
3776 bool LibraryCallKit::inline_native_setScopedValueCache() {
3777 Node* arr = argument(0);
3778 Node* cache_obj_handle = scopedValueCache_helper();
3779 const Type* objects_type = scopedValueCache_type();
3780
3781 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3782 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3783
3784 return true;
3785 }
3786
3787 //------------------------inline_native_Continuation_pin and unpin-----------
3788
3789 // Shared implementation routine for both pin and unpin.
3790 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
3791 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3792
3793 // Save input memory.
3794 Node* input_memory_state = reset_memory();
3795 set_all_memory(input_memory_state);
3796
3797 // TLS
3798 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3799 Node* last_continuation_offset = off_heap_plus_addr(tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
3800 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
3801
3802 // Null check the last continuation object.
3803 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
3804 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
3805 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3806
3807 // False path, last continuation is null.
3808 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
3809
3810 // True path, last continuation is not null.
3811 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
3812
3813 set_control(continuation_is_not_null);
3814
3815 // Load the pin count from the last continuation.
3816 Node* pin_count_offset = off_heap_plus_addr(last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
3817 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
3818
3819 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
3820 Node* pin_count_rhs;
3821 if (unpin) {
3822 pin_count_rhs = _gvn.intcon(0);
3823 } else {
3824 pin_count_rhs = _gvn.intcon(UINT32_MAX);
3825 }
3826 Node* pin_count_cmp = _gvn.transform(new CmpUNode(pin_count, pin_count_rhs));
3827 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
3828 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
3829
3830 // True branch, pin count over/underflow.
3831 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
3832 {
3833 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
3834 // which will throw IllegalStateException for pin count over/underflow.
3835 // No memory changed so far - we can use memory create by reset_memory()
3836 // at the beginning of this intrinsic. No need to call reset_memory() again.
3837 PreserveJVMState pjvms(this);
3838 set_control(pin_count_over_underflow);
3839 uncommon_trap(Deoptimization::Reason_intrinsic,
3840 Deoptimization::Action_none);
3841 assert(stopped(), "invariant");
3842 }
3843
3844 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
3845 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
3846 set_control(valid_pin_count);
3847
3848 Node* next_pin_count;
3849 if (unpin) {
3850 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
3851 } else {
3852 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
3853 }
3854
3855 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
3856
3857 // Result of top level CFG and Memory.
3858 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3859 record_for_igvn(result_rgn);
3860 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3861 record_for_igvn(result_mem);
3862
3863 result_rgn->init_req(_true_path, valid_pin_count);
3864 result_rgn->init_req(_false_path, continuation_is_null);
3865 result_mem->init_req(_true_path, reset_memory());
3866 result_mem->init_req(_false_path, input_memory_state);
3867
3868 // Set output state.
3869 set_control(_gvn.transform(result_rgn));
3870 set_all_memory(_gvn.transform(result_mem));
3871
3872 return true;
3873 }
3874
3875 //---------------------------load_mirror_from_klass----------------------------
3876 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3877 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3878 Node* p = off_heap_plus_addr(klass, in_bytes(Klass::java_mirror_offset()));
3879 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3880 // mirror = ((OopHandle)mirror)->resolve();
3881 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3882 }
3883
3884 //-----------------------load_klass_from_mirror_common-------------------------
3885 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3886 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3887 // and branch to the given path on the region.
3888 // If never_see_null, take an uncommon trap on null, so we can optimistically
3889 // compile for the non-null case.
3890 // If the region is null, force never_see_null = true.
3891 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3892 bool never_see_null,
3893 RegionNode* region,
3894 int null_path,
3895 int offset) {
3896 if (region == nullptr) never_see_null = true;
3897 Node* p = basic_plus_adr(mirror, offset);
3898 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3899 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3900 Node* null_ctl = top();
3901 kls = null_check_oop(kls, &null_ctl, never_see_null);
3902 if (region != nullptr) {
3903 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3904 region->init_req(null_path, null_ctl);
3905 } else {
3906 assert(null_ctl == top(), "no loose ends");
3907 }
3908 return kls;
3909 }
3910
3911 //--------------------(inline_native_Class_query helpers)---------------------
3912 // Use this for JVM_ACC_INTERFACE.
3913 // Fall through if (mods & mask) == bits, take the guard otherwise.
3914 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
3915 ByteSize offset, const Type* type, BasicType bt) {
3916 // Branch around if the given klass has the given modifier bit set.
3917 // Like generate_guard, adds a new path onto the region.
3918 Node* modp = off_heap_plus_addr(kls, in_bytes(offset));
3919 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
3920 Node* mask = intcon(modifier_mask);
3921 Node* bits = intcon(modifier_bits);
3922 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3923 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3924 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3925 return generate_fair_guard(bol, region);
3926 }
3927 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3928 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
3929 InstanceKlass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
3930 }
3931
3932 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
3933 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3934 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
3935 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
3936 }
3937
3938 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3939 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
3940 }
3941
3942 //-------------------------inline_native_Class_query-------------------
3943 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3944 const Type* return_type = TypeInt::BOOL;
3945 Node* prim_return_value = top(); // what happens if it's a primitive class?
3946 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3947 bool expect_prim = false; // most of these guys expect to work on refs
3948
3949 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3950
3951 Node* mirror = argument(0);
3952 Node* obj = top();
3953
3954 switch (id) {
3955 case vmIntrinsics::_isInstance:
3956 // nothing is an instance of a primitive type
3957 prim_return_value = intcon(0);
3958 obj = argument(1);
3959 break;
3960 case vmIntrinsics::_isHidden:
3961 prim_return_value = intcon(0);
3962 break;
3963 case vmIntrinsics::_getSuperclass:
3964 prim_return_value = null();
3965 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3966 break;
3967 default:
3968 fatal_unexpected_iid(id);
3969 break;
3970 }
3971
3972 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3973 if (mirror_con == nullptr) return false; // cannot happen?
3974
3975 #ifndef PRODUCT
3976 if (C->print_intrinsics() || C->print_inlining()) {
3977 ciType* k = mirror_con->java_mirror_type();
3978 if (k) {
3979 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3980 k->print_name();
3981 tty->cr();
3982 }
3983 }
3984 #endif
3985
3986 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3987 RegionNode* region = new RegionNode(PATH_LIMIT);
3988 record_for_igvn(region);
3989 PhiNode* phi = new PhiNode(region, return_type);
3990
3991 // The mirror will never be null of Reflection.getClassAccessFlags, however
3992 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3993 // if it is. See bug 4774291.
3994
3995 // For Reflection.getClassAccessFlags(), the null check occurs in
3996 // the wrong place; see inline_unsafe_access(), above, for a similar
3997 // situation.
3998 mirror = null_check(mirror);
3999 // If mirror or obj is dead, only null-path is taken.
4000 if (stopped()) return true;
4001
4002 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4003
4004 // Now load the mirror's klass metaobject, and null-check it.
4005 // Side-effects region with the control path if the klass is null.
4006 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4007 // If kls is null, we have a primitive mirror.
4008 phi->init_req(_prim_path, prim_return_value);
4009 if (stopped()) { set_result(region, phi); return true; }
4010 bool safe_for_replace = (region->in(_prim_path) == top());
4011
4012 Node* p; // handy temp
4013 Node* null_ctl;
4014
4015 // Now that we have the non-null klass, we can perform the real query.
4016 // For constant classes, the query will constant-fold in LoadNode::Value.
4017 Node* query_value = top();
4018 switch (id) {
4019 case vmIntrinsics::_isInstance:
4020 // nothing is an instance of a primitive type
4021 query_value = gen_instanceof(obj, kls, safe_for_replace);
4022 break;
4023
4024 case vmIntrinsics::_isHidden:
4025 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4026 if (generate_hidden_class_guard(kls, region) != nullptr)
4027 // A guard was added. If the guard is taken, it was an hidden class.
4028 phi->add_req(intcon(1));
4029 // If we fall through, it's a plain class.
4030 query_value = intcon(0);
4031 break;
4032
4033
4034 case vmIntrinsics::_getSuperclass:
4035 // The rules here are somewhat unfortunate, but we can still do better
4036 // with random logic than with a JNI call.
4037 // Interfaces store null or Object as _super, but must report null.
4038 // Arrays store an intermediate super as _super, but must report Object.
4039 // Other types can report the actual _super.
4040 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4041 if (generate_array_guard(kls, region) != nullptr) {
4042 // A guard was added. If the guard is taken, it was an array.
4043 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4044 }
4045 // Check for interface after array since this checks AccessFlags offset into InstanceKlass.
4046 // In other words, we are accessing subtype-specific information, so we need to determine the subtype first.
4047 if (generate_interface_guard(kls, region) != nullptr) {
4048 // A guard was added. If the guard is taken, it was an interface.
4049 phi->add_req(null());
4050 }
4051 // If we fall through, it's a plain class. Get its _super.
4052 p = off_heap_plus_addr(kls, in_bytes(Klass::super_offset()));
4053 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4054 null_ctl = top();
4055 kls = null_check_oop(kls, &null_ctl);
4056 if (null_ctl != top()) {
4057 // If the guard is taken, Object.superClass is null (both klass and mirror).
4058 region->add_req(null_ctl);
4059 phi ->add_req(null());
4060 }
4061 if (!stopped()) {
4062 query_value = load_mirror_from_klass(kls);
4063 }
4064 break;
4065
4066 default:
4067 fatal_unexpected_iid(id);
4068 break;
4069 }
4070
4071 // Fall-through is the normal case of a query to a real class.
4072 phi->init_req(1, query_value);
4073 region->init_req(1, control());
4074
4075 C->set_has_split_ifs(true); // Has chance for split-if optimization
4076 set_result(region, phi);
4077 return true;
4078 }
4079
4080 //-------------------------inline_Class_cast-------------------
4081 bool LibraryCallKit::inline_Class_cast() {
4082 Node* mirror = argument(0); // Class
4083 Node* obj = argument(1);
4084 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4085 if (mirror_con == nullptr) {
4086 return false; // dead path (mirror->is_top()).
4087 }
4088 if (obj == nullptr || obj->is_top()) {
4089 return false; // dead path
4090 }
4091 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4092
4093 // First, see if Class.cast() can be folded statically.
4094 // java_mirror_type() returns non-null for compile-time Class constants.
4095 ciType* tm = mirror_con->java_mirror_type();
4096 if (tm != nullptr && tm->is_klass() &&
4097 tp != nullptr) {
4098 if (!tp->is_loaded()) {
4099 // Don't use intrinsic when class is not loaded.
4100 return false;
4101 } else {
4102 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
4103 if (static_res == Compile::SSC_always_true) {
4104 // isInstance() is true - fold the code.
4105 set_result(obj);
4106 return true;
4107 } else if (static_res == Compile::SSC_always_false) {
4108 // Don't use intrinsic, have to throw ClassCastException.
4109 // If the reference is null, the non-intrinsic bytecode will
4110 // be optimized appropriately.
4111 return false;
4112 }
4113 }
4114 }
4115
4116 // Bailout intrinsic and do normal inlining if exception path is frequent.
4117 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4118 return false;
4119 }
4120
4121 // Generate dynamic checks.
4122 // Class.cast() is java implementation of _checkcast bytecode.
4123 // Do checkcast (Parse::do_checkcast()) optimizations here.
4124
4125 mirror = null_check(mirror);
4126 // If mirror is dead, only null-path is taken.
4127 if (stopped()) {
4128 return true;
4129 }
4130
4131 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive).
4132 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
4133 RegionNode* region = new RegionNode(PATH_LIMIT);
4134 record_for_igvn(region);
4135
4136 // Now load the mirror's klass metaobject, and null-check it.
4137 // If kls is null, we have a primitive mirror and
4138 // nothing is an instance of a primitive type.
4139 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4140
4141 Node* res = top();
4142 if (!stopped()) {
4143 Node* bad_type_ctrl = top();
4144 // Do checkcast optimizations.
4145 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4146 region->init_req(_bad_type_path, bad_type_ctrl);
4147 }
4148 if (region->in(_prim_path) != top() ||
4149 region->in(_bad_type_path) != top()) {
4150 // Let Interpreter throw ClassCastException.
4151 PreserveJVMState pjvms(this);
4152 set_control(_gvn.transform(region));
4153 uncommon_trap(Deoptimization::Reason_intrinsic,
4154 Deoptimization::Action_maybe_recompile);
4155 }
4156 if (!stopped()) {
4157 set_result(res);
4158 }
4159 return true;
4160 }
4161
4162
4163 //--------------------------inline_native_subtype_check------------------------
4164 // This intrinsic takes the JNI calls out of the heart of
4165 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4166 bool LibraryCallKit::inline_native_subtype_check() {
4167 // Pull both arguments off the stack.
4168 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4169 args[0] = argument(0);
4170 args[1] = argument(1);
4171 Node* klasses[2]; // corresponding Klasses: superk, subk
4172 klasses[0] = klasses[1] = top();
4173
4174 enum {
4175 // A full decision tree on {superc is prim, subc is prim}:
4176 _prim_0_path = 1, // {P,N} => false
4177 // {P,P} & superc!=subc => false
4178 _prim_same_path, // {P,P} & superc==subc => true
4179 _prim_1_path, // {N,P} => false
4180 _ref_subtype_path, // {N,N} & subtype check wins => true
4181 _both_ref_path, // {N,N} & subtype check loses => false
4182 PATH_LIMIT
4183 };
4184
4185 RegionNode* region = new RegionNode(PATH_LIMIT);
4186 Node* phi = new PhiNode(region, TypeInt::BOOL);
4187 record_for_igvn(region);
4188
4189 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4190 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4191 int class_klass_offset = java_lang_Class::klass_offset();
4192
4193 // First null-check both mirrors and load each mirror's klass metaobject.
4194 int which_arg;
4195 for (which_arg = 0; which_arg <= 1; which_arg++) {
4196 Node* arg = args[which_arg];
4197 arg = null_check(arg);
4198 if (stopped()) break;
4199 args[which_arg] = arg;
4200
4201 Node* p = basic_plus_adr(arg, class_klass_offset);
4202 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4203 klasses[which_arg] = _gvn.transform(kls);
4204 }
4205
4206 // Having loaded both klasses, test each for null.
4207 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4208 for (which_arg = 0; which_arg <= 1; which_arg++) {
4209 Node* kls = klasses[which_arg];
4210 Node* null_ctl = top();
4211 kls = null_check_oop(kls, &null_ctl, never_see_null);
4212 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
4213 region->init_req(prim_path, null_ctl);
4214 if (stopped()) break;
4215 klasses[which_arg] = kls;
4216 }
4217
4218 if (!stopped()) {
4219 // now we have two reference types, in klasses[0..1]
4220 Node* subk = klasses[1]; // the argument to isAssignableFrom
4221 Node* superk = klasses[0]; // the receiver
4222 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4223 // now we have a successful reference subtype check
4224 region->set_req(_ref_subtype_path, control());
4225 }
4226
4227 // If both operands are primitive (both klasses null), then
4228 // we must return true when they are identical primitives.
4229 // It is convenient to test this after the first null klass check.
4230 set_control(region->in(_prim_0_path)); // go back to first null check
4231 if (!stopped()) {
4232 // Since superc is primitive, make a guard for the superc==subc case.
4233 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4234 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4235 generate_guard(bol_eq, region, PROB_FAIR);
4236 if (region->req() == PATH_LIMIT+1) {
4237 // A guard was added. If the added guard is taken, superc==subc.
4238 region->swap_edges(PATH_LIMIT, _prim_same_path);
4239 region->del_req(PATH_LIMIT);
4240 }
4241 region->set_req(_prim_0_path, control()); // Not equal after all.
4242 }
4243
4244 // these are the only paths that produce 'true':
4245 phi->set_req(_prim_same_path, intcon(1));
4246 phi->set_req(_ref_subtype_path, intcon(1));
4247
4248 // pull together the cases:
4249 assert(region->req() == PATH_LIMIT, "sane region");
4250 for (uint i = 1; i < region->req(); i++) {
4251 Node* ctl = region->in(i);
4252 if (ctl == nullptr || ctl == top()) {
4253 region->set_req(i, top());
4254 phi ->set_req(i, top());
4255 } else if (phi->in(i) == nullptr) {
4256 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4257 }
4258 }
4259
4260 set_control(_gvn.transform(region));
4261 set_result(_gvn.transform(phi));
4262 return true;
4263 }
4264
4265 //---------------------generate_array_guard_common------------------------
4266 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
4267 bool obj_array, bool not_array, Node** obj) {
4268
4269 if (stopped()) {
4270 return nullptr;
4271 }
4272
4273 // If obj_array/non_array==false/false:
4274 // Branch around if the given klass is in fact an array (either obj or prim).
4275 // If obj_array/non_array==false/true:
4276 // Branch around if the given klass is not an array klass of any kind.
4277 // If obj_array/non_array==true/true:
4278 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
4279 // If obj_array/non_array==true/false:
4280 // Branch around if the kls is an oop array (Object[] or subtype)
4281 //
4282 // Like generate_guard, adds a new path onto the region.
4283 jint layout_con = 0;
4284 Node* layout_val = get_layout_helper(kls, layout_con);
4285 if (layout_val == nullptr) {
4286 bool query = (obj_array
4287 ? Klass::layout_helper_is_objArray(layout_con)
4288 : Klass::layout_helper_is_array(layout_con));
4289 if (query == not_array) {
4290 return nullptr; // never a branch
4291 } else { // always a branch
4292 Node* always_branch = control();
4293 if (region != nullptr)
4294 region->add_req(always_branch);
4295 set_control(top());
4296 return always_branch;
4297 }
4298 }
4299 // Now test the correct condition.
4300 jint nval = (obj_array
4301 ? (jint)(Klass::_lh_array_tag_type_value
4302 << Klass::_lh_array_tag_shift)
4303 : Klass::_lh_neutral_value);
4304 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4305 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
4306 // invert the test if we are looking for a non-array
4307 if (not_array) btest = BoolTest(btest).negate();
4308 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4309 Node* ctrl = generate_fair_guard(bol, region);
4310 Node* is_array_ctrl = not_array ? control() : ctrl;
4311 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4312 // Keep track of the fact that 'obj' is an array to prevent
4313 // array specific accesses from floating above the guard.
4314 *obj = _gvn.transform(new CheckCastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4315 }
4316 return ctrl;
4317 }
4318
4319
4320 //-----------------------inline_native_newArray--------------------------
4321 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4322 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4323 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4324 Node* mirror;
4325 Node* count_val;
4326 if (uninitialized) {
4327 null_check_receiver();
4328 mirror = argument(1);
4329 count_val = argument(2);
4330 } else {
4331 mirror = argument(0);
4332 count_val = argument(1);
4333 }
4334
4335 mirror = null_check(mirror);
4336 // If mirror or obj is dead, only null-path is taken.
4337 if (stopped()) return true;
4338
4339 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4340 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4341 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4342 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4343 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4344
4345 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4346 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4347 result_reg, _slow_path);
4348 Node* normal_ctl = control();
4349 Node* no_array_ctl = result_reg->in(_slow_path);
4350
4351 // Generate code for the slow case. We make a call to newArray().
4352 set_control(no_array_ctl);
4353 if (!stopped()) {
4354 // Either the input type is void.class, or else the
4355 // array klass has not yet been cached. Either the
4356 // ensuing call will throw an exception, or else it
4357 // will cache the array klass for next time.
4358 PreserveJVMState pjvms(this);
4359 CallJavaNode* slow_call = nullptr;
4360 if (uninitialized) {
4361 // Generate optimized virtual call (holder class 'Unsafe' is final)
4362 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4363 } else {
4364 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4365 }
4366 Node* slow_result = set_results_for_java_call(slow_call);
4367 // this->control() comes from set_results_for_java_call
4368 result_reg->set_req(_slow_path, control());
4369 result_val->set_req(_slow_path, slow_result);
4370 result_io ->set_req(_slow_path, i_o());
4371 result_mem->set_req(_slow_path, reset_memory());
4372 }
4373
4374 set_control(normal_ctl);
4375 if (!stopped()) {
4376 // Normal case: The array type has been cached in the java.lang.Class.
4377 // The following call works fine even if the array type is polymorphic.
4378 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4379 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4380 result_reg->init_req(_normal_path, control());
4381 result_val->init_req(_normal_path, obj);
4382 result_io ->init_req(_normal_path, i_o());
4383 result_mem->init_req(_normal_path, reset_memory());
4384
4385 if (uninitialized) {
4386 // Mark the allocation so that zeroing is skipped
4387 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4388 alloc->maybe_set_complete(&_gvn);
4389 }
4390 }
4391
4392 // Return the combined state.
4393 set_i_o( _gvn.transform(result_io) );
4394 set_all_memory( _gvn.transform(result_mem));
4395
4396 C->set_has_split_ifs(true); // Has chance for split-if optimization
4397 set_result(result_reg, result_val);
4398 return true;
4399 }
4400
4401 //----------------------inline_native_getLength--------------------------
4402 // public static native int java.lang.reflect.Array.getLength(Object array);
4403 bool LibraryCallKit::inline_native_getLength() {
4404 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4405
4406 Node* array = null_check(argument(0));
4407 // If array is dead, only null-path is taken.
4408 if (stopped()) return true;
4409
4410 // Deoptimize if it is a non-array.
4411 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
4412
4413 if (non_array != nullptr) {
4414 PreserveJVMState pjvms(this);
4415 set_control(non_array);
4416 uncommon_trap(Deoptimization::Reason_intrinsic,
4417 Deoptimization::Action_maybe_recompile);
4418 }
4419
4420 // If control is dead, only non-array-path is taken.
4421 if (stopped()) return true;
4422
4423 // The works fine even if the array type is polymorphic.
4424 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4425 Node* result = load_array_length(array);
4426
4427 C->set_has_split_ifs(true); // Has chance for split-if optimization
4428 set_result(result);
4429 return true;
4430 }
4431
4432 //------------------------inline_array_copyOf----------------------------
4433 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4434 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4435 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4436 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4437
4438 // Get the arguments.
4439 Node* original = argument(0);
4440 Node* start = is_copyOfRange? argument(1): intcon(0);
4441 Node* end = is_copyOfRange? argument(2): argument(1);
4442 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4443
4444 Node* newcopy = nullptr;
4445
4446 // Set the original stack and the reexecute bit for the interpreter to reexecute
4447 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4448 { PreserveReexecuteState preexecs(this);
4449 jvms()->set_should_reexecute(true);
4450
4451 array_type_mirror = null_check(array_type_mirror);
4452 original = null_check(original);
4453
4454 // Check if a null path was taken unconditionally.
4455 if (stopped()) return true;
4456
4457 Node* orig_length = load_array_length(original);
4458
4459 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4460 klass_node = null_check(klass_node);
4461
4462 RegionNode* bailout = new RegionNode(1);
4463 record_for_igvn(bailout);
4464
4465 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4466 // Bail out if that is so.
4467 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4468 if (not_objArray != nullptr) {
4469 // Improve the klass node's type from the new optimistic assumption:
4470 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4471 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4472 Node* cast = new CastPPNode(control(), klass_node, akls);
4473 klass_node = _gvn.transform(cast);
4474 }
4475
4476 // Bail out if either start or end is negative.
4477 generate_negative_guard(start, bailout, &start);
4478 generate_negative_guard(end, bailout, &end);
4479
4480 Node* length = end;
4481 if (_gvn.type(start) != TypeInt::ZERO) {
4482 length = _gvn.transform(new SubINode(end, start));
4483 }
4484
4485 // Bail out if length is negative (i.e., if start > end).
4486 // Without this the new_array would throw
4487 // NegativeArraySizeException but IllegalArgumentException is what
4488 // should be thrown
4489 generate_negative_guard(length, bailout, &length);
4490
4491 // Bail out if start is larger than the original length
4492 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4493 generate_negative_guard(orig_tail, bailout, &orig_tail);
4494
4495 if (bailout->req() > 1) {
4496 PreserveJVMState pjvms(this);
4497 set_control(_gvn.transform(bailout));
4498 uncommon_trap(Deoptimization::Reason_intrinsic,
4499 Deoptimization::Action_maybe_recompile);
4500 }
4501
4502 if (!stopped()) {
4503 // How many elements will we copy from the original?
4504 // The answer is MinI(orig_tail, length).
4505 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
4506
4507 // Generate a direct call to the right arraycopy function(s).
4508 // We know the copy is disjoint but we might not know if the
4509 // oop stores need checking.
4510 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4511 // This will fail a store-check if x contains any non-nulls.
4512
4513 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4514 // loads/stores but it is legal only if we're sure the
4515 // Arrays.copyOf would succeed. So we need all input arguments
4516 // to the copyOf to be validated, including that the copy to the
4517 // new array won't trigger an ArrayStoreException. That subtype
4518 // check can be optimized if we know something on the type of
4519 // the input array from type speculation.
4520 if (_gvn.type(klass_node)->singleton()) {
4521 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4522 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4523
4524 int test = C->static_subtype_check(superk, subk);
4525 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4526 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4527 if (t_original->speculative_type() != nullptr) {
4528 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4529 }
4530 }
4531 }
4532
4533 bool validated = false;
4534 // Reason_class_check rather than Reason_intrinsic because we
4535 // want to intrinsify even if this traps.
4536 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4537 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4538
4539 if (not_subtype_ctrl != top()) {
4540 PreserveJVMState pjvms(this);
4541 set_control(not_subtype_ctrl);
4542 uncommon_trap(Deoptimization::Reason_class_check,
4543 Deoptimization::Action_make_not_entrant);
4544 assert(stopped(), "Should be stopped");
4545 }
4546 validated = true;
4547 }
4548
4549 if (!stopped()) {
4550 newcopy = new_array(klass_node, length, 0); // no arguments to push
4551
4552 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4553 load_object_klass(original), klass_node);
4554 if (!is_copyOfRange) {
4555 ac->set_copyof(validated);
4556 } else {
4557 ac->set_copyofrange(validated);
4558 }
4559 Node* n = _gvn.transform(ac);
4560 if (n == ac) {
4561 ac->connect_outputs(this);
4562 } else {
4563 assert(validated, "shouldn't transform if all arguments not validated");
4564 set_all_memory(n);
4565 }
4566 }
4567 }
4568 } // original reexecute is set back here
4569
4570 C->set_has_split_ifs(true); // Has chance for split-if optimization
4571 if (!stopped()) {
4572 set_result(newcopy);
4573 }
4574 return true;
4575 }
4576
4577
4578 //----------------------generate_virtual_guard---------------------------
4579 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4580 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4581 RegionNode* slow_region) {
4582 ciMethod* method = callee();
4583 int vtable_index = method->vtable_index();
4584 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4585 "bad index %d", vtable_index);
4586 // Get the Method* out of the appropriate vtable entry.
4587 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4588 vtable_index*vtableEntry::size_in_bytes() +
4589 in_bytes(vtableEntry::method_offset());
4590 Node* entry_addr = off_heap_plus_addr(obj_klass, entry_offset);
4591 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4592
4593 // Compare the target method with the expected method (e.g., Object.hashCode).
4594 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4595
4596 Node* native_call = makecon(native_call_addr);
4597 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4598 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4599
4600 return generate_slow_guard(test_native, slow_region);
4601 }
4602
4603 //-----------------------generate_method_call----------------------------
4604 // Use generate_method_call to make a slow-call to the real
4605 // method if the fast path fails. An alternative would be to
4606 // use a stub like OptoRuntime::slow_arraycopy_Java.
4607 // This only works for expanding the current library call,
4608 // not another intrinsic. (E.g., don't use this for making an
4609 // arraycopy call inside of the copyOf intrinsic.)
4610 CallJavaNode*
4611 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4612 // When compiling the intrinsic method itself, do not use this technique.
4613 guarantee(callee() != C->method(), "cannot make slow-call to self");
4614
4615 ciMethod* method = callee();
4616 // ensure the JVMS we have will be correct for this call
4617 guarantee(method_id == method->intrinsic_id(), "must match");
4618
4619 const TypeFunc* tf = TypeFunc::make(method);
4620 if (res_not_null) {
4621 assert(tf->return_type() == T_OBJECT, "");
4622 const TypeTuple* range = tf->range();
4623 const Type** fields = TypeTuple::fields(range->cnt());
4624 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4625 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4626 tf = TypeFunc::make(tf->domain(), new_range);
4627 }
4628 CallJavaNode* slow_call;
4629 if (is_static) {
4630 assert(!is_virtual, "");
4631 slow_call = new CallStaticJavaNode(C, tf,
4632 SharedRuntime::get_resolve_static_call_stub(), method);
4633 } else if (is_virtual) {
4634 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4635 int vtable_index = Method::invalid_vtable_index;
4636 if (UseInlineCaches) {
4637 // Suppress the vtable call
4638 } else {
4639 // hashCode and clone are not a miranda methods,
4640 // so the vtable index is fixed.
4641 // No need to use the linkResolver to get it.
4642 vtable_index = method->vtable_index();
4643 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4644 "bad index %d", vtable_index);
4645 }
4646 slow_call = new CallDynamicJavaNode(tf,
4647 SharedRuntime::get_resolve_virtual_call_stub(),
4648 method, vtable_index);
4649 } else { // neither virtual nor static: opt_virtual
4650 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4651 slow_call = new CallStaticJavaNode(C, tf,
4652 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4653 slow_call->set_optimized_virtual(true);
4654 }
4655 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4656 // To be able to issue a direct call (optimized virtual or virtual)
4657 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4658 // about the method being invoked should be attached to the call site to
4659 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4660 slow_call->set_override_symbolic_info(true);
4661 }
4662 set_arguments_for_java_call(slow_call);
4663 set_edges_for_java_call(slow_call);
4664 return slow_call;
4665 }
4666
4667
4668 /**
4669 * Build special case code for calls to hashCode on an object. This call may
4670 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4671 * slightly different code.
4672 */
4673 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4674 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4675 assert(!(is_virtual && is_static), "either virtual, special, or static");
4676
4677 enum { _slow_path = 1, _null_path, _fast_path, _fast_path2, PATH_LIMIT };
4678
4679 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4680 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4681 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4682 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4683 Node* obj = nullptr;
4684 if (!is_static) {
4685 // Check for hashing null object
4686 obj = null_check_receiver();
4687 if (stopped()) return true; // unconditionally null
4688 result_reg->init_req(_null_path, top());
4689 result_val->init_req(_null_path, top());
4690 } else {
4691 // Do a null check, and return zero if null.
4692 // System.identityHashCode(null) == 0
4693 obj = argument(0);
4694 Node* null_ctl = top();
4695 obj = null_check_oop(obj, &null_ctl);
4696 result_reg->init_req(_null_path, null_ctl);
4697 result_val->init_req(_null_path, _gvn.intcon(0));
4698 }
4699
4700 // Unconditionally null? Then return right away.
4701 if (stopped()) {
4702 set_control( result_reg->in(_null_path));
4703 if (!stopped())
4704 set_result(result_val->in(_null_path));
4705 return true;
4706 }
4707
4708 // We only go to the fast case code if we pass a number of guards. The
4709 // paths which do not pass are accumulated in the slow_region.
4710 RegionNode* slow_region = new RegionNode(1);
4711 record_for_igvn(slow_region);
4712
4713 // If this is a virtual call, we generate a funny guard. We pull out
4714 // the vtable entry corresponding to hashCode() from the target object.
4715 // If the target method which we are calling happens to be the native
4716 // Object hashCode() method, we pass the guard. We do not need this
4717 // guard for non-virtual calls -- the caller is known to be the native
4718 // Object hashCode().
4719 if (is_virtual) {
4720 // After null check, get the object's klass.
4721 Node* obj_klass = load_object_klass(obj);
4722 generate_virtual_guard(obj_klass, slow_region);
4723 }
4724
4725 if (UseCompactObjectHeaders) {
4726 // Get the header out of the object.
4727 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4728 // The control of the load must be null. Otherwise, the load can move before
4729 // the null check after castPP removal.
4730 Node* no_ctrl = nullptr;
4731 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4732
4733 // Test the header to see if the object is in hashed or copied state.
4734 Node* hashctrl_mask = _gvn.MakeConX(markWord::hashctrl_mask_in_place);
4735 Node* masked_header = _gvn.transform(new AndXNode(header, hashctrl_mask));
4736
4737 // Take slow-path when the object has not been hashed.
4738 Node* not_hashed_val = _gvn.MakeConX(0);
4739 Node* chk_hashed = _gvn.transform(new CmpXNode(masked_header, not_hashed_val));
4740 Node* test_hashed = _gvn.transform(new BoolNode(chk_hashed, BoolTest::eq));
4741
4742 generate_slow_guard(test_hashed, slow_region);
4743
4744 // Test whether the object is hashed or hashed&copied.
4745 Node* hashed_copied = _gvn.MakeConX(markWord::hashctrl_expanded_mask_in_place | markWord::hashctrl_hashed_mask_in_place);
4746 Node* chk_copied = _gvn.transform(new CmpXNode(masked_header, hashed_copied));
4747 // If true, then object has been hashed&copied, otherwise it's only hashed.
4748 Node* test_copied = _gvn.transform(new BoolNode(chk_copied, BoolTest::eq));
4749 IfNode* if_copied = create_and_map_if(control(), test_copied, PROB_FAIR, COUNT_UNKNOWN);
4750 Node* if_true = _gvn.transform(new IfTrueNode(if_copied));
4751 Node* if_false = _gvn.transform(new IfFalseNode(if_copied));
4752
4753 // Hashed&Copied path: read hash-code out of the object.
4754 set_control(if_true);
4755 // result_val->del_req(_fast_path2);
4756 // result_reg->del_req(_fast_path2);
4757 // result_io->del_req(_fast_path2);
4758 // result_mem->del_req(_fast_path2);
4759
4760 Node* obj_klass = load_object_klass(obj);
4761 Node* hash_addr;
4762 const TypeKlassPtr* klass_t = _gvn.type(obj_klass)->isa_klassptr();
4763 bool load_offset_runtime = true;
4764
4765 if (klass_t != nullptr) {
4766 if (klass_t->klass_is_exact() && klass_t->isa_instklassptr()) {
4767 ciInstanceKlass* ciKlass = reinterpret_cast<ciInstanceKlass*>(klass_t->is_instklassptr()->exact_klass());
4768 if (!ciKlass->is_mirror_instance_klass() && !ciKlass->is_reference_instance_klass()) {
4769 // We know the InstanceKlass, load hash_offset from there at compile-time.
4770 int hash_offset = ciKlass->hash_offset_in_bytes();
4771 hash_addr = basic_plus_adr(obj, hash_offset);
4772 Node* loaded_hash = make_load(control(), hash_addr, TypeInt::INT, T_INT, MemNode::unordered);
4773 result_val->init_req(_fast_path2, loaded_hash);
4774 result_reg->init_req(_fast_path2, control());
4775 load_offset_runtime = false;
4776 }
4777 }
4778 }
4779
4780 //tty->print_cr("Load hash-offset at runtime: %s", BOOL_TO_STR(load_offset_runtime));
4781
4782 if (load_offset_runtime) {
4783 // We don't know if it is an array or an exact type, figure it out at run-time.
4784 // If not an ordinary instance, then we need to take slow-path.
4785 Node* kind_addr = basic_plus_adr(top(), obj_klass, Klass::kind_offset_in_bytes());
4786 Node* kind = make_load(control(), kind_addr, TypeInt::INT, T_INT, MemNode::unordered);
4787 Node* instance_val = _gvn.intcon(Klass::InstanceKlassKind);
4788 Node* chk_inst = _gvn.transform(new CmpINode(kind, instance_val));
4789 Node* test_inst = _gvn.transform(new BoolNode(chk_inst, BoolTest::ne));
4790 generate_slow_guard(test_inst, slow_region);
4791
4792 // Otherwise it's an instance and we can read the hash_offset from the InstanceKlass.
4793 Node* hash_offset_addr = basic_plus_adr(top(), obj_klass, InstanceKlass::hash_offset_offset_in_bytes());
4794 Node* hash_offset = make_load(control(), hash_offset_addr, TypeInt::INT, T_INT, MemNode::unordered);
4795 // hash_offset->dump();
4796 Node* hash_addr = basic_plus_adr(obj, ConvI2X(hash_offset));
4797 Compile::current()->set_has_unsafe_access(true);
4798 Node* loaded_hash = make_load(control(), hash_addr, TypeInt::INT, T_INT, MemNode::unordered);
4799 result_val->init_req(_fast_path2, loaded_hash);
4800 result_reg->init_req(_fast_path2, control());
4801 }
4802
4803 // Hashed-only path: recompute hash-code from object address.
4804 set_control(if_false);
4805 if (hashCode == 6) {
4806 // Our constants.
4807 Node* M = _gvn.intcon(0x337954D5);
4808 Node* A = _gvn.intcon(0xAAAAAAAA);
4809 // Split object address into lo and hi 32 bits.
4810 Node* obj_addr = _gvn.transform(new CastP2XNode(nullptr, obj));
4811 Node* x = _gvn.transform(new ConvL2INode(obj_addr));
4812 Node* upper_addr = _gvn.transform(new URShiftLNode(obj_addr, _gvn.intcon(32)));
4813 Node* y = _gvn.transform(new ConvL2INode(upper_addr));
4814
4815 Node* H0 = _gvn.transform(new XorINode(x, y));
4816 Node* L0 = _gvn.transform(new XorINode(x, A));
4817
4818 // Full multiplication of two 32 bit values L0 and M into a hi/lo result in two 32 bit values V0 and U0.
4819 Node* L0_64 = _gvn.transform(new ConvI2LNode(L0));
4820 L0_64 = _gvn.transform(new AndLNode(L0_64, _gvn.longcon(0xFFFFFFFF)));
4821 Node* M_64 = _gvn.transform(new ConvI2LNode(M));
4822 // M_64 = _gvn.transform(new AndLNode(M_64, _gvn.longcon(0xFFFFFFFF)));
4823 Node* prod64 = _gvn.transform(new MulLNode(L0_64, M_64));
4824 Node* V0 = _gvn.transform(new ConvL2INode(prod64));
4825 Node* prod_upper = _gvn.transform(new URShiftLNode(prod64, _gvn.intcon(32)));
4826 Node* U0 = _gvn.transform(new ConvL2INode(prod_upper));
4827
4828 Node* Q0 = _gvn.transform(new MulINode(H0, M));
4829 Node* L1 = _gvn.transform(new XorINode(Q0, U0));
4830
4831 // Full multiplication of two 32 bit values L1 and M into a hi/lo result in two 32 bit values V1 and U1.
4832 Node* L1_64 = _gvn.transform(new ConvI2LNode(L1));
4833 L1_64 = _gvn.transform(new AndLNode(L1_64, _gvn.longcon(0xFFFFFFFF)));
4834 prod64 = _gvn.transform(new MulLNode(L1_64, M_64));
4835 Node* V1 = _gvn.transform(new ConvL2INode(prod64));
4836 prod_upper = _gvn.transform(new URShiftLNode(prod64, _gvn.intcon(32)));
4837 Node* U1 = _gvn.transform(new ConvL2INode(prod_upper));
4838
4839 Node* P1 = _gvn.transform(new XorINode(V0, M));
4840
4841 // Right rotate P1 by distance L1.
4842 Node* distance = _gvn.transform(new AndINode(L1, _gvn.intcon(32 - 1)));
4843 Node* inverse_distance = _gvn.transform(new SubINode(_gvn.intcon(32), distance));
4844 Node* ror_part1 = _gvn.transform(new URShiftINode(P1, distance));
4845 Node* ror_part2 = _gvn.transform(new LShiftINode(P1, inverse_distance));
4846 Node* Q1 = _gvn.transform(new OrINode(ror_part1, ror_part2));
4847
4848 Node* L2 = _gvn.transform(new XorINode(Q1, U1));
4849 Node* hash = _gvn.transform(new XorINode(V1, L2));
4850 Node* hash_truncated = _gvn.transform(new AndINode(hash, _gvn.intcon(markWord::hash_mask)));
4851
4852 result_val->init_req(_fast_path, hash_truncated);
4853 } else if (hashCode == 2) {
4854 result_val->init_req(_fast_path, _gvn.intcon(1));
4855 }
4856 } else {
4857 // Get the header out of the object, use LoadMarkNode when available
4858 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4859 // The control of the load must be null. Otherwise, the load can move before
4860 // the null check after castPP removal.
4861 Node* no_ctrl = nullptr;
4862 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4863
4864 if (!UseObjectMonitorTable) {
4865 // Test the header to see if it is safe to read w.r.t. locking.
4866 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4867 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4868 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
4869 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
4870 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
4871
4872 generate_slow_guard(test_monitor, slow_region);
4873 }
4874
4875 // Get the hash value and check to see that it has been properly assigned.
4876 // We depend on hash_mask being at most 32 bits and avoid the use of
4877 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4878 // vm: see markWord.hpp.
4879 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4880 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4881 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4882 // This hack lets the hash bits live anywhere in the mark object now, as long
4883 // as the shift drops the relevant bits into the low 32 bits. Note that
4884 // Java spec says that HashCode is an int so there's no point in capturing
4885 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4886 hshifted_header = ConvX2I(hshifted_header);
4887 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4888
4889 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4890 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4891 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4892
4893 generate_slow_guard(test_assigned, slow_region);
4894
4895 result_val->init_req(_fast_path, hash_val);
4896
4897 // _fast_path2 is not used here.
4898 result_val->del_req(_fast_path2);
4899 result_reg->del_req(_fast_path2);
4900 result_io->del_req(_fast_path2);
4901 result_mem->del_req(_fast_path2);
4902 }
4903
4904 Node* init_mem = reset_memory();
4905 // fill in the rest of the null path:
4906 result_io ->init_req(_null_path, i_o());
4907 result_mem->init_req(_null_path, init_mem);
4908
4909 result_reg->init_req(_fast_path, control());
4910 result_io ->init_req(_fast_path, i_o());
4911 result_mem->init_req(_fast_path, init_mem);
4912
4913 if (UseCompactObjectHeaders) {
4914 result_io->init_req(_fast_path2, i_o());
4915 result_mem->init_req(_fast_path2, init_mem);
4916 }
4917
4918 // Generate code for the slow case. We make a call to hashCode().
4919 assert(slow_region != nullptr, "must have slow_region");
4920 set_control(_gvn.transform(slow_region));
4921 if (!stopped()) {
4922 // No need for PreserveJVMState, because we're using up the present state.
4923 set_all_memory(init_mem);
4924 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4925 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4926 Node* slow_result = set_results_for_java_call(slow_call);
4927 // this->control() comes from set_results_for_java_call
4928 result_reg->init_req(_slow_path, control());
4929 result_val->init_req(_slow_path, slow_result);
4930 result_io ->set_req(_slow_path, i_o());
4931 result_mem ->set_req(_slow_path, reset_memory());
4932 }
4933
4934 // Return the combined state.
4935 set_i_o( _gvn.transform(result_io) );
4936 set_all_memory( _gvn.transform(result_mem));
4937
4938 set_result(result_reg, result_val);
4939 return true;
4940 }
4941
4942 //---------------------------inline_native_getClass----------------------------
4943 // public final native Class<?> java.lang.Object.getClass();
4944 //
4945 // Build special case code for calls to getClass on an object.
4946 bool LibraryCallKit::inline_native_getClass() {
4947 Node* obj = null_check_receiver();
4948 if (stopped()) return true;
4949 set_result(load_mirror_from_klass(load_object_klass(obj)));
4950 return true;
4951 }
4952
4953 //-----------------inline_native_Reflection_getCallerClass---------------------
4954 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4955 //
4956 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4957 //
4958 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4959 // in that it must skip particular security frames and checks for
4960 // caller sensitive methods.
4961 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4962 #ifndef PRODUCT
4963 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4964 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4965 }
4966 #endif
4967
4968 if (!jvms()->has_method()) {
4969 #ifndef PRODUCT
4970 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4971 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4972 }
4973 #endif
4974 return false;
4975 }
4976
4977 // Walk back up the JVM state to find the caller at the required
4978 // depth.
4979 JVMState* caller_jvms = jvms();
4980
4981 // Cf. JVM_GetCallerClass
4982 // NOTE: Start the loop at depth 1 because the current JVM state does
4983 // not include the Reflection.getCallerClass() frame.
4984 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4985 ciMethod* m = caller_jvms->method();
4986 switch (n) {
4987 case 0:
4988 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4989 break;
4990 case 1:
4991 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4992 if (!m->caller_sensitive()) {
4993 #ifndef PRODUCT
4994 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4995 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4996 }
4997 #endif
4998 return false; // bail-out; let JVM_GetCallerClass do the work
4999 }
5000 break;
5001 default:
5002 if (!m->is_ignored_by_security_stack_walk()) {
5003 // We have reached the desired frame; return the holder class.
5004 // Acquire method holder as java.lang.Class and push as constant.
5005 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
5006 ciInstance* caller_mirror = caller_klass->java_mirror();
5007 set_result(makecon(TypeInstPtr::make(caller_mirror)));
5008
5009 #ifndef PRODUCT
5010 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5011 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());
5012 tty->print_cr(" JVM state at this point:");
5013 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5014 ciMethod* m = jvms()->of_depth(i)->method();
5015 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5016 }
5017 }
5018 #endif
5019 return true;
5020 }
5021 break;
5022 }
5023 }
5024
5025 #ifndef PRODUCT
5026 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5027 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
5028 tty->print_cr(" JVM state at this point:");
5029 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5030 ciMethod* m = jvms()->of_depth(i)->method();
5031 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5032 }
5033 }
5034 #endif
5035
5036 return false; // bail-out; let JVM_GetCallerClass do the work
5037 }
5038
5039 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
5040 Node* arg = argument(0);
5041 Node* result = nullptr;
5042
5043 switch (id) {
5044 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
5045 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
5046 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5047 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5048 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5049 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5050
5051 case vmIntrinsics::_doubleToLongBits: {
5052 // two paths (plus control) merge in a wood
5053 RegionNode *r = new RegionNode(3);
5054 Node *phi = new PhiNode(r, TypeLong::LONG);
5055
5056 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5057 // Build the boolean node
5058 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5059
5060 // Branch either way.
5061 // NaN case is less traveled, which makes all the difference.
5062 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5063 Node *opt_isnan = _gvn.transform(ifisnan);
5064 assert( opt_isnan->is_If(), "Expect an IfNode");
5065 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5066 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5067
5068 set_control(iftrue);
5069
5070 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5071 Node *slow_result = longcon(nan_bits); // return NaN
5072 phi->init_req(1, _gvn.transform( slow_result ));
5073 r->init_req(1, iftrue);
5074
5075 // Else fall through
5076 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5077 set_control(iffalse);
5078
5079 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5080 r->init_req(2, iffalse);
5081
5082 // Post merge
5083 set_control(_gvn.transform(r));
5084 record_for_igvn(r);
5085
5086 C->set_has_split_ifs(true); // Has chance for split-if optimization
5087 result = phi;
5088 assert(result->bottom_type()->isa_long(), "must be");
5089 break;
5090 }
5091
5092 case vmIntrinsics::_floatToIntBits: {
5093 // two paths (plus control) merge in a wood
5094 RegionNode *r = new RegionNode(3);
5095 Node *phi = new PhiNode(r, TypeInt::INT);
5096
5097 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5098 // Build the boolean node
5099 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5100
5101 // Branch either way.
5102 // NaN case is less traveled, which makes all the difference.
5103 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5104 Node *opt_isnan = _gvn.transform(ifisnan);
5105 assert( opt_isnan->is_If(), "Expect an IfNode");
5106 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5107 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5108
5109 set_control(iftrue);
5110
5111 static const jint nan_bits = 0x7fc00000;
5112 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5113 phi->init_req(1, _gvn.transform( slow_result ));
5114 r->init_req(1, iftrue);
5115
5116 // Else fall through
5117 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5118 set_control(iffalse);
5119
5120 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5121 r->init_req(2, iffalse);
5122
5123 // Post merge
5124 set_control(_gvn.transform(r));
5125 record_for_igvn(r);
5126
5127 C->set_has_split_ifs(true); // Has chance for split-if optimization
5128 result = phi;
5129 assert(result->bottom_type()->isa_int(), "must be");
5130 break;
5131 }
5132
5133 default:
5134 fatal_unexpected_iid(id);
5135 break;
5136 }
5137 set_result(_gvn.transform(result));
5138 return true;
5139 }
5140
5141 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5142 Node* arg = argument(0);
5143 Node* result = nullptr;
5144
5145 switch (id) {
5146 case vmIntrinsics::_floatIsInfinite:
5147 result = new IsInfiniteFNode(arg);
5148 break;
5149 case vmIntrinsics::_floatIsFinite:
5150 result = new IsFiniteFNode(arg);
5151 break;
5152 case vmIntrinsics::_doubleIsInfinite:
5153 result = new IsInfiniteDNode(arg);
5154 break;
5155 case vmIntrinsics::_doubleIsFinite:
5156 result = new IsFiniteDNode(arg);
5157 break;
5158 default:
5159 fatal_unexpected_iid(id);
5160 break;
5161 }
5162 set_result(_gvn.transform(result));
5163 return true;
5164 }
5165
5166 //----------------------inline_unsafe_copyMemory-------------------------
5167 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5168
5169 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5170 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5171 const Type* base_t = gvn.type(base);
5172
5173 bool in_native = (base_t == TypePtr::NULL_PTR);
5174 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5175 bool is_mixed = !in_heap && !in_native;
5176
5177 if (is_mixed) {
5178 return true; // mixed accesses can touch both on-heap and off-heap memory
5179 }
5180 if (in_heap) {
5181 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5182 if (!is_prim_array) {
5183 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5184 // there's not enough type information available to determine proper memory slice for it.
5185 return true;
5186 }
5187 }
5188 return false;
5189 }
5190
5191 bool LibraryCallKit::inline_unsafe_copyMemory() {
5192 if (callee()->is_static()) return false; // caller must have the capability!
5193 null_check_receiver(); // null-check receiver
5194 if (stopped()) return true;
5195
5196 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5197
5198 Node* src_base = argument(1); // type: oop
5199 Node* src_off = ConvL2X(argument(2)); // type: long
5200 Node* dst_base = argument(4); // type: oop
5201 Node* dst_off = ConvL2X(argument(5)); // type: long
5202 Node* size = ConvL2X(argument(7)); // type: long
5203
5204 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5205 "fieldOffset must be byte-scaled");
5206
5207 Node* src_addr = make_unsafe_address(src_base, src_off);
5208 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5209
5210 Node* thread = _gvn.transform(new ThreadLocalNode());
5211 Node* doing_unsafe_access_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5212 BasicType doing_unsafe_access_bt = T_BYTE;
5213 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5214
5215 // update volatile field
5216 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5217
5218 int flags = RC_LEAF | RC_NO_FP;
5219
5220 const TypePtr* dst_type = TypePtr::BOTTOM;
5221
5222 // Adjust memory effects of the runtime call based on input values.
5223 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5224 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5225 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5226
5227 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5228 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5229 flags |= RC_NARROW_MEM; // narrow in memory
5230 }
5231 }
5232
5233 // Call it. Note that the length argument is not scaled.
5234 make_runtime_call(flags,
5235 OptoRuntime::fast_arraycopy_Type(),
5236 StubRoutines::unsafe_arraycopy(),
5237 "unsafe_arraycopy",
5238 dst_type,
5239 src_addr, dst_addr, size XTOP);
5240
5241 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5242
5243 return true;
5244 }
5245
5246 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5247 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5248 bool LibraryCallKit::inline_unsafe_setMemory() {
5249 if (callee()->is_static()) return false; // caller must have the capability!
5250 null_check_receiver(); // null-check receiver
5251 if (stopped()) return true;
5252
5253 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5254
5255 Node* dst_base = argument(1); // type: oop
5256 Node* dst_off = ConvL2X(argument(2)); // type: long
5257 Node* size = ConvL2X(argument(4)); // type: long
5258 Node* byte = argument(6); // type: byte
5259
5260 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5261 "fieldOffset must be byte-scaled");
5262
5263 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5264
5265 Node* thread = _gvn.transform(new ThreadLocalNode());
5266 Node* doing_unsafe_access_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5267 BasicType doing_unsafe_access_bt = T_BYTE;
5268 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5269
5270 // update volatile field
5271 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5272
5273 int flags = RC_LEAF | RC_NO_FP;
5274
5275 const TypePtr* dst_type = TypePtr::BOTTOM;
5276
5277 // Adjust memory effects of the runtime call based on input values.
5278 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5279 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5280
5281 flags |= RC_NARROW_MEM; // narrow in memory
5282 }
5283
5284 // Call it. Note that the length argument is not scaled.
5285 make_runtime_call(flags,
5286 OptoRuntime::unsafe_setmemory_Type(),
5287 StubRoutines::unsafe_setmemory(),
5288 "unsafe_setmemory",
5289 dst_type,
5290 dst_addr, size XTOP, byte);
5291
5292 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5293
5294 return true;
5295 }
5296
5297 #undef XTOP
5298
5299 //------------------------clone_coping-----------------------------------
5300 // Helper function for inline_native_clone.
5301 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5302 assert(obj_size != nullptr, "");
5303 Node* raw_obj = alloc_obj->in(1);
5304 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5305
5306 AllocateNode* alloc = nullptr;
5307 if (ReduceBulkZeroing &&
5308 // If we are implementing an array clone without knowing its source type
5309 // (can happen when compiling the array-guarded branch of a reflective
5310 // Object.clone() invocation), initialize the array within the allocation.
5311 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5312 // to a runtime clone call that assumes fully initialized source arrays.
5313 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5314 // We will be completely responsible for initializing this object -
5315 // mark Initialize node as complete.
5316 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5317 // The object was just allocated - there should be no any stores!
5318 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5319 // Mark as complete_with_arraycopy so that on AllocateNode
5320 // expansion, we know this AllocateNode is initialized by an array
5321 // copy and a StoreStore barrier exists after the array copy.
5322 alloc->initialization()->set_complete_with_arraycopy();
5323 }
5324
5325 Node* size = _gvn.transform(obj_size);
5326 access_clone(obj, alloc_obj, size, is_array);
5327
5328 // Do not let reads from the cloned object float above the arraycopy.
5329 if (alloc != nullptr) {
5330 // Do not let stores that initialize this object be reordered with
5331 // a subsequent store that would make this object accessible by
5332 // other threads.
5333 // Record what AllocateNode this StoreStore protects so that
5334 // escape analysis can go from the MemBarStoreStoreNode to the
5335 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5336 // based on the escape status of the AllocateNode.
5337 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5338 } else {
5339 insert_mem_bar(Op_MemBarCPUOrder);
5340 }
5341 }
5342
5343 //------------------------inline_native_clone----------------------------
5344 // protected native Object java.lang.Object.clone();
5345 //
5346 // Here are the simple edge cases:
5347 // null receiver => normal trap
5348 // virtual and clone was overridden => slow path to out-of-line clone
5349 // not cloneable or finalizer => slow path to out-of-line Object.clone
5350 //
5351 // The general case has two steps, allocation and copying.
5352 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5353 //
5354 // Copying also has two cases, oop arrays and everything else.
5355 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5356 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5357 //
5358 // These steps fold up nicely if and when the cloned object's klass
5359 // can be sharply typed as an object array, a type array, or an instance.
5360 //
5361 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5362 PhiNode* result_val;
5363
5364 // Set the reexecute bit for the interpreter to reexecute
5365 // the bytecode that invokes Object.clone if deoptimization happens.
5366 { PreserveReexecuteState preexecs(this);
5367 jvms()->set_should_reexecute(true);
5368
5369 Node* obj = null_check_receiver();
5370 if (stopped()) return true;
5371
5372 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5373
5374 // If we are going to clone an instance, we need its exact type to
5375 // know the number and types of fields to convert the clone to
5376 // loads/stores. Maybe a speculative type can help us.
5377 if (!obj_type->klass_is_exact() &&
5378 obj_type->speculative_type() != nullptr &&
5379 obj_type->speculative_type()->is_instance_klass()) {
5380 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5381 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5382 !spec_ik->has_injected_fields()) {
5383 if (!obj_type->isa_instptr() ||
5384 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5385 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5386 }
5387 }
5388 }
5389
5390 // Conservatively insert a memory barrier on all memory slices.
5391 // Do not let writes into the original float below the clone.
5392 insert_mem_bar(Op_MemBarCPUOrder);
5393
5394 // paths into result_reg:
5395 enum {
5396 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5397 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5398 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5399 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5400 PATH_LIMIT
5401 };
5402 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5403 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5404 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5405 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5406 record_for_igvn(result_reg);
5407
5408 Node* obj_klass = load_object_klass(obj);
5409 Node* array_obj = obj;
5410 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5411 if (array_ctl != nullptr) {
5412 // It's an array.
5413 PreserveJVMState pjvms(this);
5414 set_control(array_ctl);
5415 Node* obj_length = load_array_length(array_obj);
5416 Node* array_size = nullptr; // Size of the array without object alignment padding.
5417 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5418
5419 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5420 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5421 // If it is an oop array, it requires very special treatment,
5422 // because gc barriers are required when accessing the array.
5423 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5424 if (is_obja != nullptr) {
5425 PreserveJVMState pjvms2(this);
5426 set_control(is_obja);
5427 // Generate a direct call to the right arraycopy function(s).
5428 // Clones are always tightly coupled.
5429 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5430 ac->set_clone_oop_array();
5431 Node* n = _gvn.transform(ac);
5432 assert(n == ac, "cannot disappear");
5433 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5434
5435 result_reg->init_req(_objArray_path, control());
5436 result_val->init_req(_objArray_path, alloc_obj);
5437 result_i_o ->set_req(_objArray_path, i_o());
5438 result_mem ->set_req(_objArray_path, reset_memory());
5439 }
5440 }
5441 // Otherwise, there are no barriers to worry about.
5442 // (We can dispense with card marks if we know the allocation
5443 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5444 // causes the non-eden paths to take compensating steps to
5445 // simulate a fresh allocation, so that no further
5446 // card marks are required in compiled code to initialize
5447 // the object.)
5448
5449 if (!stopped()) {
5450 copy_to_clone(array_obj, alloc_obj, array_size, true);
5451
5452 // Present the results of the copy.
5453 result_reg->init_req(_array_path, control());
5454 result_val->init_req(_array_path, alloc_obj);
5455 result_i_o ->set_req(_array_path, i_o());
5456 result_mem ->set_req(_array_path, reset_memory());
5457 }
5458 }
5459
5460 // We only go to the instance fast case code if we pass a number of guards.
5461 // The paths which do not pass are accumulated in the slow_region.
5462 RegionNode* slow_region = new RegionNode(1);
5463 record_for_igvn(slow_region);
5464 if (!stopped()) {
5465 // It's an instance (we did array above). Make the slow-path tests.
5466 // If this is a virtual call, we generate a funny guard. We grab
5467 // the vtable entry corresponding to clone() from the target object.
5468 // If the target method which we are calling happens to be the
5469 // Object clone() method, we pass the guard. We do not need this
5470 // guard for non-virtual calls; the caller is known to be the native
5471 // Object clone().
5472 if (is_virtual) {
5473 generate_virtual_guard(obj_klass, slow_region);
5474 }
5475
5476 // The object must be easily cloneable and must not have a finalizer.
5477 // Both of these conditions may be checked in a single test.
5478 // We could optimize the test further, but we don't care.
5479 generate_misc_flags_guard(obj_klass,
5480 // Test both conditions:
5481 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5482 // Must be cloneable but not finalizer:
5483 KlassFlags::_misc_is_cloneable_fast,
5484 slow_region);
5485 }
5486
5487 if (!stopped()) {
5488 // It's an instance, and it passed the slow-path tests.
5489 PreserveJVMState pjvms(this);
5490 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5491 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5492 // is reexecuted if deoptimization occurs and there could be problems when merging
5493 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5494 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5495
5496 copy_to_clone(obj, alloc_obj, obj_size, false);
5497
5498 // Present the results of the slow call.
5499 result_reg->init_req(_instance_path, control());
5500 result_val->init_req(_instance_path, alloc_obj);
5501 result_i_o ->set_req(_instance_path, i_o());
5502 result_mem ->set_req(_instance_path, reset_memory());
5503 }
5504
5505 // Generate code for the slow case. We make a call to clone().
5506 set_control(_gvn.transform(slow_region));
5507 if (!stopped()) {
5508 PreserveJVMState pjvms(this);
5509 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5510 // We need to deoptimize on exception (see comment above)
5511 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5512 // this->control() comes from set_results_for_java_call
5513 result_reg->init_req(_slow_path, control());
5514 result_val->init_req(_slow_path, slow_result);
5515 result_i_o ->set_req(_slow_path, i_o());
5516 result_mem ->set_req(_slow_path, reset_memory());
5517 }
5518
5519 // Return the combined state.
5520 set_control( _gvn.transform(result_reg));
5521 set_i_o( _gvn.transform(result_i_o));
5522 set_all_memory( _gvn.transform(result_mem));
5523 } // original reexecute is set back here
5524
5525 set_result(_gvn.transform(result_val));
5526 return true;
5527 }
5528
5529 // If we have a tightly coupled allocation, the arraycopy may take care
5530 // of the array initialization. If one of the guards we insert between
5531 // the allocation and the arraycopy causes a deoptimization, an
5532 // uninitialized array will escape the compiled method. To prevent that
5533 // we set the JVM state for uncommon traps between the allocation and
5534 // the arraycopy to the state before the allocation so, in case of
5535 // deoptimization, we'll reexecute the allocation and the
5536 // initialization.
5537 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5538 if (alloc != nullptr) {
5539 ciMethod* trap_method = alloc->jvms()->method();
5540 int trap_bci = alloc->jvms()->bci();
5541
5542 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5543 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5544 // Make sure there's no store between the allocation and the
5545 // arraycopy otherwise visible side effects could be rexecuted
5546 // in case of deoptimization and cause incorrect execution.
5547 bool no_interfering_store = true;
5548 Node* mem = alloc->in(TypeFunc::Memory);
5549 if (mem->is_MergeMem()) {
5550 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5551 Node* n = mms.memory();
5552 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5553 assert(n->is_Store(), "what else?");
5554 no_interfering_store = false;
5555 break;
5556 }
5557 }
5558 } else {
5559 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5560 Node* n = mms.memory();
5561 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5562 assert(n->is_Store(), "what else?");
5563 no_interfering_store = false;
5564 break;
5565 }
5566 }
5567 }
5568
5569 if (no_interfering_store) {
5570 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5571
5572 JVMState* saved_jvms = jvms();
5573 saved_reexecute_sp = _reexecute_sp;
5574
5575 set_jvms(sfpt->jvms());
5576 _reexecute_sp = jvms()->sp();
5577
5578 return saved_jvms;
5579 }
5580 }
5581 }
5582 return nullptr;
5583 }
5584
5585 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5586 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5587 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5588 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5589 uint size = alloc->req();
5590 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5591 old_jvms->set_map(sfpt);
5592 for (uint i = 0; i < size; i++) {
5593 sfpt->init_req(i, alloc->in(i));
5594 }
5595 // re-push array length for deoptimization
5596 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5597 old_jvms->set_sp(old_jvms->sp()+1);
5598 old_jvms->set_monoff(old_jvms->monoff()+1);
5599 old_jvms->set_scloff(old_jvms->scloff()+1);
5600 old_jvms->set_endoff(old_jvms->endoff()+1);
5601 old_jvms->set_should_reexecute(true);
5602
5603 sfpt->set_i_o(map()->i_o());
5604 sfpt->set_memory(map()->memory());
5605 sfpt->set_control(map()->control());
5606 return sfpt;
5607 }
5608
5609 // In case of a deoptimization, we restart execution at the
5610 // allocation, allocating a new array. We would leave an uninitialized
5611 // array in the heap that GCs wouldn't expect. Move the allocation
5612 // after the traps so we don't allocate the array if we
5613 // deoptimize. This is possible because tightly_coupled_allocation()
5614 // guarantees there's no observer of the allocated array at this point
5615 // and the control flow is simple enough.
5616 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5617 int saved_reexecute_sp, uint new_idx) {
5618 if (saved_jvms_before_guards != nullptr && !stopped()) {
5619 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5620
5621 assert(alloc != nullptr, "only with a tightly coupled allocation");
5622 // restore JVM state to the state at the arraycopy
5623 saved_jvms_before_guards->map()->set_control(map()->control());
5624 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5625 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5626 // If we've improved the types of some nodes (null check) while
5627 // emitting the guards, propagate them to the current state
5628 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5629 set_jvms(saved_jvms_before_guards);
5630 _reexecute_sp = saved_reexecute_sp;
5631
5632 // Remove the allocation from above the guards
5633 CallProjections callprojs;
5634 alloc->extract_projections(&callprojs, true);
5635 InitializeNode* init = alloc->initialization();
5636 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5637 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5638 init->replace_mem_projs_by(alloc_mem, C);
5639
5640 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5641 // the allocation (i.e. is only valid if the allocation succeeds):
5642 // 1) replace CastIINode with AllocateArrayNode's length here
5643 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5644 //
5645 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5646 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5647 Node* init_control = init->proj_out(TypeFunc::Control);
5648 Node* alloc_length = alloc->Ideal_length();
5649 #ifdef ASSERT
5650 Node* prev_cast = nullptr;
5651 #endif
5652 for (uint i = 0; i < init_control->outcnt(); i++) {
5653 Node* init_out = init_control->raw_out(i);
5654 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5655 #ifdef ASSERT
5656 if (prev_cast == nullptr) {
5657 prev_cast = init_out;
5658 } else {
5659 if (prev_cast->cmp(*init_out) == false) {
5660 prev_cast->dump();
5661 init_out->dump();
5662 assert(false, "not equal CastIINode");
5663 }
5664 }
5665 #endif
5666 C->gvn_replace_by(init_out, alloc_length);
5667 }
5668 }
5669 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5670
5671 // move the allocation here (after the guards)
5672 _gvn.hash_delete(alloc);
5673 alloc->set_req(TypeFunc::Control, control());
5674 alloc->set_req(TypeFunc::I_O, i_o());
5675 Node *mem = reset_memory();
5676 set_all_memory(mem);
5677 alloc->set_req(TypeFunc::Memory, mem);
5678 set_control(init->proj_out_or_null(TypeFunc::Control));
5679 set_i_o(callprojs.fallthrough_ioproj);
5680
5681 // Update memory as done in GraphKit::set_output_for_allocation()
5682 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5683 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5684 if (ary_type->isa_aryptr() && length_type != nullptr) {
5685 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5686 }
5687 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5688 int elemidx = C->get_alias_index(telemref);
5689 // Need to properly move every memory projection for the Initialize
5690 #ifdef ASSERT
5691 int mark_idx = C->get_alias_index(ary_type->add_offset(oopDesc::mark_offset_in_bytes()));
5692 int klass_idx = C->get_alias_index(ary_type->add_offset(Type::klass_offset()));
5693 #endif
5694 auto move_proj = [&](ProjNode* proj) {
5695 int alias_idx = C->get_alias_index(proj->adr_type());
5696 assert(alias_idx == Compile::AliasIdxRaw ||
5697 alias_idx == elemidx ||
5698 alias_idx == mark_idx ||
5699 alias_idx == klass_idx, "should be raw memory or array element type");
5700 set_memory(proj, alias_idx);
5701 };
5702 init->for_each_proj(move_proj, TypeFunc::Memory);
5703
5704 Node* allocx = _gvn.transform(alloc);
5705 assert(allocx == alloc, "where has the allocation gone?");
5706 assert(dest->is_CheckCastPP(), "not an allocation result?");
5707
5708 _gvn.hash_delete(dest);
5709 dest->set_req(0, control());
5710 Node* destx = _gvn.transform(dest);
5711 assert(destx == dest, "where has the allocation result gone?");
5712
5713 array_ideal_length(alloc, ary_type, true);
5714 }
5715 }
5716
5717 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5718 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5719 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5720 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5721 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5722 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5723 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5724 JVMState* saved_jvms_before_guards) {
5725 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5726 // There is at least one unrelated uncommon trap which needs to be replaced.
5727 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5728
5729 JVMState* saved_jvms = jvms();
5730 const int saved_reexecute_sp = _reexecute_sp;
5731 set_jvms(sfpt->jvms());
5732 _reexecute_sp = jvms()->sp();
5733
5734 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5735
5736 // Restore state
5737 set_jvms(saved_jvms);
5738 _reexecute_sp = saved_reexecute_sp;
5739 }
5740 }
5741
5742 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5743 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5744 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5745 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5746 while (if_proj->is_IfProj()) {
5747 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5748 if (uncommon_trap != nullptr) {
5749 create_new_uncommon_trap(uncommon_trap);
5750 }
5751 assert(if_proj->in(0)->is_If(), "must be If");
5752 if_proj = if_proj->in(0)->in(0);
5753 }
5754 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5755 "must have reached control projection of init node");
5756 }
5757
5758 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5759 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5760 assert(trap_request != 0, "no valid UCT trap request");
5761 PreserveJVMState pjvms(this);
5762 set_control(uncommon_trap_call->in(0));
5763 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5764 Deoptimization::trap_request_action(trap_request));
5765 assert(stopped(), "Should be stopped");
5766 _gvn.hash_delete(uncommon_trap_call);
5767 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5768 }
5769
5770 // Common checks for array sorting intrinsics arguments.
5771 // Returns `true` if checks passed.
5772 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
5773 // check address of the class
5774 if (elementType == nullptr || elementType->is_top()) {
5775 return false; // dead path
5776 }
5777 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5778 if (elem_klass == nullptr) {
5779 return false; // dead path
5780 }
5781 // java_mirror_type() returns non-null for compile-time Class constants only
5782 ciType* elem_type = elem_klass->java_mirror_type();
5783 if (elem_type == nullptr) {
5784 return false;
5785 }
5786 bt = elem_type->basic_type();
5787 // Disable the intrinsic if the CPU does not support SIMD sort
5788 if (!Matcher::supports_simd_sort(bt)) {
5789 return false;
5790 }
5791 // check address of the array
5792 if (obj == nullptr || obj->is_top()) {
5793 return false; // dead path
5794 }
5795 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5796 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
5797 return false; // failed input validation
5798 }
5799 return true;
5800 }
5801
5802 //------------------------------inline_array_partition-----------------------
5803 bool LibraryCallKit::inline_array_partition() {
5804 address stubAddr = StubRoutines::select_array_partition_function();
5805 if (stubAddr == nullptr) {
5806 return false; // Intrinsic's stub is not implemented on this platform
5807 }
5808 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
5809
5810 // no receiver because it is a static method
5811 Node* elementType = argument(0);
5812 Node* obj = argument(1);
5813 Node* offset = argument(2); // long
5814 Node* fromIndex = argument(4);
5815 Node* toIndex = argument(5);
5816 Node* indexPivot1 = argument(6);
5817 Node* indexPivot2 = argument(7);
5818 // PartitionOperation: argument(8) is ignored
5819
5820 Node* pivotIndices = nullptr;
5821 BasicType bt = T_ILLEGAL;
5822
5823 if (!check_array_sort_arguments(elementType, obj, bt)) {
5824 return false;
5825 }
5826 null_check(obj);
5827 // If obj is dead, only null-path is taken.
5828 if (stopped()) {
5829 return true;
5830 }
5831 // Set the original stack and the reexecute bit for the interpreter to reexecute
5832 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
5833 { PreserveReexecuteState preexecs(this);
5834 jvms()->set_should_reexecute(true);
5835
5836 Node* obj_adr = make_unsafe_address(obj, offset);
5837
5838 // create the pivotIndices array of type int and size = 2
5839 Node* size = intcon(2);
5840 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
5841 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
5842 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
5843 guarantee(alloc != nullptr, "created above");
5844 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
5845
5846 // pass the basic type enum to the stub
5847 Node* elemType = intcon(bt);
5848
5849 // Call the stub
5850 const char *stubName = "array_partition_stub";
5851 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
5852 stubAddr, stubName, TypePtr::BOTTOM,
5853 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
5854 indexPivot1, indexPivot2);
5855
5856 } // original reexecute is set back here
5857
5858 if (!stopped()) {
5859 set_result(pivotIndices);
5860 }
5861
5862 return true;
5863 }
5864
5865
5866 //------------------------------inline_array_sort-----------------------
5867 bool LibraryCallKit::inline_array_sort() {
5868 address stubAddr = StubRoutines::select_arraysort_function();
5869 if (stubAddr == nullptr) {
5870 return false; // Intrinsic's stub is not implemented on this platform
5871 }
5872 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
5873
5874 // no receiver because it is a static method
5875 Node* elementType = argument(0);
5876 Node* obj = argument(1);
5877 Node* offset = argument(2); // long
5878 Node* fromIndex = argument(4);
5879 Node* toIndex = argument(5);
5880 // SortOperation: argument(6) is ignored
5881
5882 BasicType bt = T_ILLEGAL;
5883
5884 if (!check_array_sort_arguments(elementType, obj, bt)) {
5885 return false;
5886 }
5887 null_check(obj);
5888 // If obj is dead, only null-path is taken.
5889 if (stopped()) {
5890 return true;
5891 }
5892 Node* obj_adr = make_unsafe_address(obj, offset);
5893
5894 // pass the basic type enum to the stub
5895 Node* elemType = intcon(bt);
5896
5897 // Call the stub.
5898 const char *stubName = "arraysort_stub";
5899 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
5900 stubAddr, stubName, TypePtr::BOTTOM,
5901 obj_adr, elemType, fromIndex, toIndex);
5902
5903 return true;
5904 }
5905
5906
5907 //------------------------------inline_arraycopy-----------------------
5908 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5909 // Object dest, int destPos,
5910 // int length);
5911 bool LibraryCallKit::inline_arraycopy() {
5912 // Get the arguments.
5913 Node* src = argument(0); // type: oop
5914 Node* src_offset = argument(1); // type: int
5915 Node* dest = argument(2); // type: oop
5916 Node* dest_offset = argument(3); // type: int
5917 Node* length = argument(4); // type: int
5918
5919 uint new_idx = C->unique();
5920
5921 // Check for allocation before we add nodes that would confuse
5922 // tightly_coupled_allocation()
5923 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5924
5925 int saved_reexecute_sp = -1;
5926 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5927 // See arraycopy_restore_alloc_state() comment
5928 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5929 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5930 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5931 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5932
5933 // The following tests must be performed
5934 // (1) src and dest are arrays.
5935 // (2) src and dest arrays must have elements of the same BasicType
5936 // (3) src and dest must not be null.
5937 // (4) src_offset must not be negative.
5938 // (5) dest_offset must not be negative.
5939 // (6) length must not be negative.
5940 // (7) src_offset + length must not exceed length of src.
5941 // (8) dest_offset + length must not exceed length of dest.
5942 // (9) each element of an oop array must be assignable
5943
5944 // (3) src and dest must not be null.
5945 // always do this here because we need the JVM state for uncommon traps
5946 Node* null_ctl = top();
5947 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5948 assert(null_ctl->is_top(), "no null control here");
5949 dest = null_check(dest, T_ARRAY);
5950
5951 if (!can_emit_guards) {
5952 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5953 // guards but the arraycopy node could still take advantage of a
5954 // tightly allocated allocation. tightly_coupled_allocation() is
5955 // called again to make sure it takes the null check above into
5956 // account: the null check is mandatory and if it caused an
5957 // uncommon trap to be emitted then the allocation can't be
5958 // considered tightly coupled in this context.
5959 alloc = tightly_coupled_allocation(dest);
5960 }
5961
5962 bool validated = false;
5963
5964 const Type* src_type = _gvn.type(src);
5965 const Type* dest_type = _gvn.type(dest);
5966 const TypeAryPtr* top_src = src_type->isa_aryptr();
5967 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5968
5969 // Do we have the type of src?
5970 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5971 // Do we have the type of dest?
5972 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5973 // Is the type for src from speculation?
5974 bool src_spec = false;
5975 // Is the type for dest from speculation?
5976 bool dest_spec = false;
5977
5978 if ((!has_src || !has_dest) && can_emit_guards) {
5979 // We don't have sufficient type information, let's see if
5980 // speculative types can help. We need to have types for both src
5981 // and dest so that it pays off.
5982
5983 // Do we already have or could we have type information for src
5984 bool could_have_src = has_src;
5985 // Do we already have or could we have type information for dest
5986 bool could_have_dest = has_dest;
5987
5988 ciKlass* src_k = nullptr;
5989 if (!has_src) {
5990 src_k = src_type->speculative_type_not_null();
5991 if (src_k != nullptr && src_k->is_array_klass()) {
5992 could_have_src = true;
5993 }
5994 }
5995
5996 ciKlass* dest_k = nullptr;
5997 if (!has_dest) {
5998 dest_k = dest_type->speculative_type_not_null();
5999 if (dest_k != nullptr && dest_k->is_array_klass()) {
6000 could_have_dest = true;
6001 }
6002 }
6003
6004 if (could_have_src && could_have_dest) {
6005 // This is going to pay off so emit the required guards
6006 if (!has_src) {
6007 src = maybe_cast_profiled_obj(src, src_k, true);
6008 src_type = _gvn.type(src);
6009 top_src = src_type->isa_aryptr();
6010 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6011 src_spec = true;
6012 }
6013 if (!has_dest) {
6014 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6015 dest_type = _gvn.type(dest);
6016 top_dest = dest_type->isa_aryptr();
6017 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6018 dest_spec = true;
6019 }
6020 }
6021 }
6022
6023 if (has_src && has_dest && can_emit_guards) {
6024 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
6025 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
6026 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
6027 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
6028
6029 if (src_elem == dest_elem && src_elem == T_OBJECT) {
6030 // If both arrays are object arrays then having the exact types
6031 // for both will remove the need for a subtype check at runtime
6032 // before the call and may make it possible to pick a faster copy
6033 // routine (without a subtype check on every element)
6034 // Do we have the exact type of src?
6035 bool could_have_src = src_spec;
6036 // Do we have the exact type of dest?
6037 bool could_have_dest = dest_spec;
6038 ciKlass* src_k = nullptr;
6039 ciKlass* dest_k = nullptr;
6040 if (!src_spec) {
6041 src_k = src_type->speculative_type_not_null();
6042 if (src_k != nullptr && src_k->is_array_klass()) {
6043 could_have_src = true;
6044 }
6045 }
6046 if (!dest_spec) {
6047 dest_k = dest_type->speculative_type_not_null();
6048 if (dest_k != nullptr && dest_k->is_array_klass()) {
6049 could_have_dest = true;
6050 }
6051 }
6052 if (could_have_src && could_have_dest) {
6053 // If we can have both exact types, emit the missing guards
6054 if (could_have_src && !src_spec) {
6055 src = maybe_cast_profiled_obj(src, src_k, true);
6056 }
6057 if (could_have_dest && !dest_spec) {
6058 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6059 }
6060 }
6061 }
6062 }
6063
6064 ciMethod* trap_method = method();
6065 int trap_bci = bci();
6066 if (saved_jvms_before_guards != nullptr) {
6067 trap_method = alloc->jvms()->method();
6068 trap_bci = alloc->jvms()->bci();
6069 }
6070
6071 bool negative_length_guard_generated = false;
6072
6073 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6074 can_emit_guards &&
6075 !src->is_top() && !dest->is_top()) {
6076 // validate arguments: enables transformation the ArrayCopyNode
6077 validated = true;
6078
6079 RegionNode* slow_region = new RegionNode(1);
6080 record_for_igvn(slow_region);
6081
6082 // (1) src and dest are arrays.
6083 generate_non_array_guard(load_object_klass(src), slow_region, &src);
6084 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
6085
6086 // (2) src and dest arrays must have elements of the same BasicType
6087 // done at macro expansion or at Ideal transformation time
6088
6089 // (4) src_offset must not be negative.
6090 generate_negative_guard(src_offset, slow_region);
6091
6092 // (5) dest_offset must not be negative.
6093 generate_negative_guard(dest_offset, slow_region);
6094
6095 // (7) src_offset + length must not exceed length of src.
6096 generate_limit_guard(src_offset, length,
6097 load_array_length(src),
6098 slow_region);
6099
6100 // (8) dest_offset + length must not exceed length of dest.
6101 generate_limit_guard(dest_offset, length,
6102 load_array_length(dest),
6103 slow_region);
6104
6105 // (6) length must not be negative.
6106 // This is also checked in generate_arraycopy() during macro expansion, but
6107 // we also have to check it here for the case where the ArrayCopyNode will
6108 // be eliminated by Escape Analysis.
6109 if (EliminateAllocations) {
6110 generate_negative_guard(length, slow_region);
6111 negative_length_guard_generated = true;
6112 }
6113
6114 // (9) each element of an oop array must be assignable
6115 Node* dest_klass = load_object_klass(dest);
6116 if (src != dest) {
6117 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6118
6119 if (not_subtype_ctrl != top()) {
6120 PreserveJVMState pjvms(this);
6121 set_control(not_subtype_ctrl);
6122 uncommon_trap(Deoptimization::Reason_intrinsic,
6123 Deoptimization::Action_make_not_entrant);
6124 assert(stopped(), "Should be stopped");
6125 }
6126 }
6127 {
6128 PreserveJVMState pjvms(this);
6129 set_control(_gvn.transform(slow_region));
6130 uncommon_trap(Deoptimization::Reason_intrinsic,
6131 Deoptimization::Action_make_not_entrant);
6132 assert(stopped(), "Should be stopped");
6133 }
6134
6135 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
6136 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6137 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6138 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6139 }
6140
6141 if (stopped()) {
6142 return true;
6143 }
6144
6145 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6146 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6147 // so the compiler has a chance to eliminate them: during macro expansion,
6148 // we have to set their control (CastPP nodes are eliminated).
6149 load_object_klass(src), load_object_klass(dest),
6150 load_array_length(src), load_array_length(dest));
6151
6152 ac->set_arraycopy(validated);
6153
6154 Node* n = _gvn.transform(ac);
6155 if (n == ac) {
6156 ac->connect_outputs(this);
6157 } else {
6158 assert(validated, "shouldn't transform if all arguments not validated");
6159 set_all_memory(n);
6160 }
6161 clear_upper_avx();
6162
6163
6164 return true;
6165 }
6166
6167
6168 // Helper function which determines if an arraycopy immediately follows
6169 // an allocation, with no intervening tests or other escapes for the object.
6170 AllocateArrayNode*
6171 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6172 if (stopped()) return nullptr; // no fast path
6173 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6174
6175 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6176 if (alloc == nullptr) return nullptr;
6177
6178 Node* rawmem = memory(Compile::AliasIdxRaw);
6179 // Is the allocation's memory state untouched?
6180 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6181 // Bail out if there have been raw-memory effects since the allocation.
6182 // (Example: There might have been a call or safepoint.)
6183 return nullptr;
6184 }
6185 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6186 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6187 return nullptr;
6188 }
6189
6190 // There must be no unexpected observers of this allocation.
6191 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6192 Node* obs = ptr->fast_out(i);
6193 if (obs != this->map()) {
6194 return nullptr;
6195 }
6196 }
6197
6198 // This arraycopy must unconditionally follow the allocation of the ptr.
6199 Node* alloc_ctl = ptr->in(0);
6200 Node* ctl = control();
6201 while (ctl != alloc_ctl) {
6202 // There may be guards which feed into the slow_region.
6203 // Any other control flow means that we might not get a chance
6204 // to finish initializing the allocated object.
6205 // Various low-level checks bottom out in uncommon traps. These
6206 // are considered safe since we've already checked above that
6207 // there is no unexpected observer of this allocation.
6208 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6209 assert(ctl->in(0)->is_If(), "must be If");
6210 ctl = ctl->in(0)->in(0);
6211 } else {
6212 return nullptr;
6213 }
6214 }
6215
6216 // If we get this far, we have an allocation which immediately
6217 // precedes the arraycopy, and we can take over zeroing the new object.
6218 // The arraycopy will finish the initialization, and provide
6219 // a new control state to which we will anchor the destination pointer.
6220
6221 return alloc;
6222 }
6223
6224 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6225 if (node->is_IfProj()) {
6226 IfProjNode* other_proj = node->as_IfProj()->other_if_proj();
6227 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6228 Node* obs = other_proj->fast_out(j);
6229 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6230 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6231 return obs->as_CallStaticJava();
6232 }
6233 }
6234 }
6235 return nullptr;
6236 }
6237
6238 //-------------inline_encodeISOArray-----------------------------------
6239 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6240 // int java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6241 // int java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6242 // encode char[] to byte[] in ISO_8859_1 or ASCII
6243 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6244 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6245 // no receiver since it is static method
6246 Node *src = argument(0);
6247 Node *src_offset = argument(1);
6248 Node *dst = argument(2);
6249 Node *dst_offset = argument(3);
6250 Node *length = argument(4);
6251
6252 // Cast source & target arrays to not-null
6253 src = must_be_not_null(src, true);
6254 dst = must_be_not_null(dst, true);
6255 if (stopped()) {
6256 return true;
6257 }
6258
6259 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6260 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6261 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6262 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6263 // failed array check
6264 return false;
6265 }
6266
6267 // Figure out the size and type of the elements we will be copying.
6268 BasicType src_elem = src_type->elem()->array_element_basic_type();
6269 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6270 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6271 return false;
6272 }
6273
6274 // Check source & target bounds
6275 RegionNode* bailout = create_bailout();
6276 generate_string_range_check(src, src_offset, length, src_elem == T_BYTE, bailout);
6277 generate_string_range_check(dst, dst_offset, length, false, bailout);
6278 if (check_bailout(bailout)) {
6279 return true;
6280 }
6281
6282 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6283 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6284 // 'src_start' points to src array + scaled offset
6285 // 'dst_start' points to dst array + scaled offset
6286
6287 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6288 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6289 enc = _gvn.transform(enc);
6290 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6291 set_memory(res_mem, mtype);
6292 set_result(enc);
6293 clear_upper_avx();
6294
6295 return true;
6296 }
6297
6298 //-------------inline_multiplyToLen-----------------------------------
6299 bool LibraryCallKit::inline_multiplyToLen() {
6300 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6301
6302 address stubAddr = StubRoutines::multiplyToLen();
6303 if (stubAddr == nullptr) {
6304 return false; // Intrinsic's stub is not implemented on this platform
6305 }
6306 const char* stubName = "multiplyToLen";
6307
6308 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6309
6310 // no receiver because it is a static method
6311 Node* x = argument(0);
6312 Node* xlen = argument(1);
6313 Node* y = argument(2);
6314 Node* ylen = argument(3);
6315 Node* z = argument(4);
6316
6317 x = must_be_not_null(x, true);
6318 y = must_be_not_null(y, true);
6319
6320 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6321 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6322 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6323 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6324 // failed array check
6325 return false;
6326 }
6327
6328 BasicType x_elem = x_type->elem()->array_element_basic_type();
6329 BasicType y_elem = y_type->elem()->array_element_basic_type();
6330 if (x_elem != T_INT || y_elem != T_INT) {
6331 return false;
6332 }
6333
6334 Node* x_start = array_element_address(x, intcon(0), x_elem);
6335 Node* y_start = array_element_address(y, intcon(0), y_elem);
6336 // 'x_start' points to x array + scaled xlen
6337 // 'y_start' points to y array + scaled ylen
6338
6339 Node* z_start = array_element_address(z, intcon(0), T_INT);
6340
6341 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6342 OptoRuntime::multiplyToLen_Type(),
6343 stubAddr, stubName, TypePtr::BOTTOM,
6344 x_start, xlen, y_start, ylen, z_start);
6345
6346 C->set_has_split_ifs(true); // Has chance for split-if optimization
6347 set_result(z);
6348 return true;
6349 }
6350
6351 //-------------inline_squareToLen------------------------------------
6352 bool LibraryCallKit::inline_squareToLen() {
6353 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6354
6355 address stubAddr = StubRoutines::squareToLen();
6356 if (stubAddr == nullptr) {
6357 return false; // Intrinsic's stub is not implemented on this platform
6358 }
6359 const char* stubName = "squareToLen";
6360
6361 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6362
6363 Node* x = argument(0);
6364 Node* len = argument(1);
6365 Node* z = argument(2);
6366 Node* zlen = argument(3);
6367
6368 x = must_be_not_null(x, true);
6369 z = must_be_not_null(z, true);
6370
6371 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6372 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6373 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6374 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6375 // failed array check
6376 return false;
6377 }
6378
6379 BasicType x_elem = x_type->elem()->array_element_basic_type();
6380 BasicType z_elem = z_type->elem()->array_element_basic_type();
6381 if (x_elem != T_INT || z_elem != T_INT) {
6382 return false;
6383 }
6384
6385
6386 Node* x_start = array_element_address(x, intcon(0), x_elem);
6387 Node* z_start = array_element_address(z, intcon(0), z_elem);
6388
6389 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6390 OptoRuntime::squareToLen_Type(),
6391 stubAddr, stubName, TypePtr::BOTTOM,
6392 x_start, len, z_start, zlen);
6393
6394 set_result(z);
6395 return true;
6396 }
6397
6398 //-------------inline_mulAdd------------------------------------------
6399 bool LibraryCallKit::inline_mulAdd() {
6400 assert(UseMulAddIntrinsic, "not implemented on this platform");
6401
6402 address stubAddr = StubRoutines::mulAdd();
6403 if (stubAddr == nullptr) {
6404 return false; // Intrinsic's stub is not implemented on this platform
6405 }
6406 const char* stubName = "mulAdd";
6407
6408 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6409
6410 Node* out = argument(0);
6411 Node* in = argument(1);
6412 Node* offset = argument(2);
6413 Node* len = argument(3);
6414 Node* k = argument(4);
6415
6416 in = must_be_not_null(in, true);
6417 out = must_be_not_null(out, true);
6418
6419 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6420 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6421 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6422 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6423 // failed array check
6424 return false;
6425 }
6426
6427 BasicType out_elem = out_type->elem()->array_element_basic_type();
6428 BasicType in_elem = in_type->elem()->array_element_basic_type();
6429 if (out_elem != T_INT || in_elem != T_INT) {
6430 return false;
6431 }
6432
6433 Node* outlen = load_array_length(out);
6434 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6435 Node* out_start = array_element_address(out, intcon(0), out_elem);
6436 Node* in_start = array_element_address(in, intcon(0), in_elem);
6437
6438 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6439 OptoRuntime::mulAdd_Type(),
6440 stubAddr, stubName, TypePtr::BOTTOM,
6441 out_start,in_start, new_offset, len, k);
6442 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6443 set_result(result);
6444 return true;
6445 }
6446
6447 //-------------inline_montgomeryMultiply-----------------------------------
6448 bool LibraryCallKit::inline_montgomeryMultiply() {
6449 address stubAddr = StubRoutines::montgomeryMultiply();
6450 if (stubAddr == nullptr) {
6451 return false; // Intrinsic's stub is not implemented on this platform
6452 }
6453
6454 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6455 const char* stubName = "montgomery_multiply";
6456
6457 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6458
6459 Node* a = argument(0);
6460 Node* b = argument(1);
6461 Node* n = argument(2);
6462 Node* len = argument(3);
6463 Node* inv = argument(4);
6464 Node* m = argument(6);
6465
6466 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6467 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6468 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6469 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6470 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6471 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6472 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6473 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6474 // failed array check
6475 return false;
6476 }
6477
6478 BasicType a_elem = a_type->elem()->array_element_basic_type();
6479 BasicType b_elem = b_type->elem()->array_element_basic_type();
6480 BasicType n_elem = n_type->elem()->array_element_basic_type();
6481 BasicType m_elem = m_type->elem()->array_element_basic_type();
6482 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6483 return false;
6484 }
6485
6486 // Make the call
6487 {
6488 Node* a_start = array_element_address(a, intcon(0), a_elem);
6489 Node* b_start = array_element_address(b, intcon(0), b_elem);
6490 Node* n_start = array_element_address(n, intcon(0), n_elem);
6491 Node* m_start = array_element_address(m, intcon(0), m_elem);
6492
6493 Node* call = make_runtime_call(RC_LEAF,
6494 OptoRuntime::montgomeryMultiply_Type(),
6495 stubAddr, stubName, TypePtr::BOTTOM,
6496 a_start, b_start, n_start, len, inv, top(),
6497 m_start);
6498 set_result(m);
6499 }
6500
6501 return true;
6502 }
6503
6504 bool LibraryCallKit::inline_montgomerySquare() {
6505 address stubAddr = StubRoutines::montgomerySquare();
6506 if (stubAddr == nullptr) {
6507 return false; // Intrinsic's stub is not implemented on this platform
6508 }
6509
6510 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6511 const char* stubName = "montgomery_square";
6512
6513 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6514
6515 Node* a = argument(0);
6516 Node* n = argument(1);
6517 Node* len = argument(2);
6518 Node* inv = argument(3);
6519 Node* m = argument(5);
6520
6521 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6522 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6523 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6524 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6525 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6526 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6527 // failed array check
6528 return false;
6529 }
6530
6531 BasicType a_elem = a_type->elem()->array_element_basic_type();
6532 BasicType n_elem = n_type->elem()->array_element_basic_type();
6533 BasicType m_elem = m_type->elem()->array_element_basic_type();
6534 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6535 return false;
6536 }
6537
6538 // Make the call
6539 {
6540 Node* a_start = array_element_address(a, intcon(0), a_elem);
6541 Node* n_start = array_element_address(n, intcon(0), n_elem);
6542 Node* m_start = array_element_address(m, intcon(0), m_elem);
6543
6544 Node* call = make_runtime_call(RC_LEAF,
6545 OptoRuntime::montgomerySquare_Type(),
6546 stubAddr, stubName, TypePtr::BOTTOM,
6547 a_start, n_start, len, inv, top(),
6548 m_start);
6549 set_result(m);
6550 }
6551
6552 return true;
6553 }
6554
6555 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6556 address stubAddr = nullptr;
6557 const char* stubName = nullptr;
6558
6559 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6560 if (stubAddr == nullptr) {
6561 return false; // Intrinsic's stub is not implemented on this platform
6562 }
6563
6564 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6565
6566 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6567
6568 Node* newArr = argument(0);
6569 Node* oldArr = argument(1);
6570 Node* newIdx = argument(2);
6571 Node* shiftCount = argument(3);
6572 Node* numIter = argument(4);
6573
6574 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6575 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6576 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6577 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6578 return false;
6579 }
6580
6581 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6582 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6583 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6584 return false;
6585 }
6586
6587 // Make the call
6588 {
6589 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6590 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6591
6592 Node* call = make_runtime_call(RC_LEAF,
6593 OptoRuntime::bigIntegerShift_Type(),
6594 stubAddr,
6595 stubName,
6596 TypePtr::BOTTOM,
6597 newArr_start,
6598 oldArr_start,
6599 newIdx,
6600 shiftCount,
6601 numIter);
6602 }
6603
6604 return true;
6605 }
6606
6607 //-------------inline_vectorizedMismatch------------------------------
6608 bool LibraryCallKit::inline_vectorizedMismatch() {
6609 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6610
6611 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6612 Node* obja = argument(0); // Object
6613 Node* aoffset = argument(1); // long
6614 Node* objb = argument(3); // Object
6615 Node* boffset = argument(4); // long
6616 Node* length = argument(6); // int
6617 Node* scale = argument(7); // int
6618
6619 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6620 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6621 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6622 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6623 scale == top()) {
6624 return false; // failed input validation
6625 }
6626
6627 Node* obja_adr = make_unsafe_address(obja, aoffset);
6628 Node* objb_adr = make_unsafe_address(objb, boffset);
6629
6630 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6631 //
6632 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6633 // if (length <= inline_limit) {
6634 // inline_path:
6635 // vmask = VectorMaskGen length
6636 // vload1 = LoadVectorMasked obja, vmask
6637 // vload2 = LoadVectorMasked objb, vmask
6638 // result1 = VectorCmpMasked vload1, vload2, vmask
6639 // } else {
6640 // call_stub_path:
6641 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6642 // }
6643 // exit_block:
6644 // return Phi(result1, result2);
6645 //
6646 enum { inline_path = 1, // input is small enough to process it all at once
6647 stub_path = 2, // input is too large; call into the VM
6648 PATH_LIMIT = 3
6649 };
6650
6651 Node* exit_block = new RegionNode(PATH_LIMIT);
6652 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6653 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6654
6655 Node* call_stub_path = control();
6656
6657 BasicType elem_bt = T_ILLEGAL;
6658
6659 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6660 if (scale_t->is_con()) {
6661 switch (scale_t->get_con()) {
6662 case 0: elem_bt = T_BYTE; break;
6663 case 1: elem_bt = T_SHORT; break;
6664 case 2: elem_bt = T_INT; break;
6665 case 3: elem_bt = T_LONG; break;
6666
6667 default: elem_bt = T_ILLEGAL; break; // not supported
6668 }
6669 }
6670
6671 int inline_limit = 0;
6672 bool do_partial_inline = false;
6673
6674 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6675 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6676 do_partial_inline = inline_limit >= 16;
6677 }
6678
6679 if (do_partial_inline) {
6680 assert(elem_bt != T_ILLEGAL, "sanity");
6681
6682 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6683 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6684 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6685
6686 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6687 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6688 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6689
6690 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6691
6692 if (!stopped()) {
6693 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6694
6695 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6696 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6697 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6698 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6699
6700 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6701 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6702 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6703 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6704
6705 exit_block->init_req(inline_path, control());
6706 memory_phi->init_req(inline_path, map()->memory());
6707 result_phi->init_req(inline_path, result);
6708
6709 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6710 clear_upper_avx();
6711 }
6712 }
6713 }
6714
6715 if (call_stub_path != nullptr) {
6716 set_control(call_stub_path);
6717
6718 Node* call = make_runtime_call(RC_LEAF,
6719 OptoRuntime::vectorizedMismatch_Type(),
6720 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6721 obja_adr, objb_adr, length, scale);
6722
6723 exit_block->init_req(stub_path, control());
6724 memory_phi->init_req(stub_path, map()->memory());
6725 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6726 }
6727
6728 exit_block = _gvn.transform(exit_block);
6729 memory_phi = _gvn.transform(memory_phi);
6730 result_phi = _gvn.transform(result_phi);
6731
6732 record_for_igvn(exit_block);
6733 record_for_igvn(memory_phi);
6734 record_for_igvn(result_phi);
6735
6736 set_control(exit_block);
6737 set_all_memory(memory_phi);
6738 set_result(result_phi);
6739
6740 return true;
6741 }
6742
6743 //------------------------------inline_vectorizedHashcode----------------------------
6744 bool LibraryCallKit::inline_vectorizedHashCode() {
6745 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6746
6747 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6748 Node* array = argument(0);
6749 Node* offset = argument(1);
6750 Node* length = argument(2);
6751 Node* initialValue = argument(3);
6752 Node* basic_type = argument(4);
6753
6754 if (basic_type == top()) {
6755 return false; // failed input validation
6756 }
6757
6758 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6759 if (!basic_type_t->is_con()) {
6760 return false; // Only intrinsify if mode argument is constant
6761 }
6762
6763 array = must_be_not_null(array, true);
6764
6765 BasicType bt = (BasicType)basic_type_t->get_con();
6766
6767 // Resolve address of first element
6768 Node* array_start = array_element_address(array, offset, bt);
6769
6770 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6771 array_start, length, initialValue, basic_type)));
6772 clear_upper_avx();
6773
6774 return true;
6775 }
6776
6777 /**
6778 * Calculate CRC32 for byte.
6779 * int java.util.zip.CRC32.update(int crc, int b)
6780 */
6781 bool LibraryCallKit::inline_updateCRC32() {
6782 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6783 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6784 // no receiver since it is static method
6785 Node* crc = argument(0); // type: int
6786 Node* b = argument(1); // type: int
6787
6788 /*
6789 * int c = ~ crc;
6790 * b = timesXtoThe32[(b ^ c) & 0xFF];
6791 * b = b ^ (c >>> 8);
6792 * crc = ~b;
6793 */
6794
6795 Node* M1 = intcon(-1);
6796 crc = _gvn.transform(new XorINode(crc, M1));
6797 Node* result = _gvn.transform(new XorINode(crc, b));
6798 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6799
6800 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6801 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6802 Node* adr = off_heap_plus_addr(base, ConvI2X(offset));
6803 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6804
6805 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6806 result = _gvn.transform(new XorINode(crc, result));
6807 result = _gvn.transform(new XorINode(result, M1));
6808 set_result(result);
6809 return true;
6810 }
6811
6812 /**
6813 * Calculate CRC32 for byte[] array.
6814 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6815 */
6816 bool LibraryCallKit::inline_updateBytesCRC32() {
6817 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6818 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6819 // no receiver since it is static method
6820 Node* crc = argument(0); // type: int
6821 Node* src = argument(1); // type: oop
6822 Node* offset = argument(2); // type: int
6823 Node* length = argument(3); // type: int
6824
6825 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6826 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6827 // failed array check
6828 return false;
6829 }
6830
6831 // Figure out the size and type of the elements we will be copying.
6832 BasicType src_elem = src_type->elem()->array_element_basic_type();
6833 if (src_elem != T_BYTE) {
6834 return false;
6835 }
6836
6837 // 'src_start' points to src array + scaled offset
6838 src = must_be_not_null(src, true);
6839 Node* src_start = array_element_address(src, offset, src_elem);
6840
6841 // We assume that range check is done by caller.
6842 // TODO: generate range check (offset+length < src.length) in debug VM.
6843
6844 // Call the stub.
6845 address stubAddr = StubRoutines::updateBytesCRC32();
6846 const char *stubName = "updateBytesCRC32";
6847
6848 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6849 stubAddr, stubName, TypePtr::BOTTOM,
6850 crc, src_start, length);
6851 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6852 set_result(result);
6853 return true;
6854 }
6855
6856 /**
6857 * Calculate CRC32 for ByteBuffer.
6858 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6859 */
6860 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6861 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6862 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6863 // no receiver since it is static method
6864 Node* crc = argument(0); // type: int
6865 Node* src = argument(1); // type: long
6866 Node* offset = argument(3); // type: int
6867 Node* length = argument(4); // type: int
6868
6869 src = ConvL2X(src); // adjust Java long to machine word
6870 Node* base = _gvn.transform(new CastX2PNode(src));
6871 offset = ConvI2X(offset);
6872
6873 // 'src_start' points to src array + scaled offset
6874 Node* src_start = off_heap_plus_addr(base, offset);
6875
6876 // Call the stub.
6877 address stubAddr = StubRoutines::updateBytesCRC32();
6878 const char *stubName = "updateBytesCRC32";
6879
6880 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6881 stubAddr, stubName, TypePtr::BOTTOM,
6882 crc, src_start, length);
6883 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6884 set_result(result);
6885 return true;
6886 }
6887
6888 //------------------------------get_table_from_crc32c_class-----------------------
6889 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6890 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6891 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6892
6893 return table;
6894 }
6895
6896 //------------------------------inline_updateBytesCRC32C-----------------------
6897 //
6898 // Calculate CRC32C for byte[] array.
6899 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6900 //
6901 bool LibraryCallKit::inline_updateBytesCRC32C() {
6902 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6903 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6904 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6905 // no receiver since it is a static method
6906 Node* crc = argument(0); // type: int
6907 Node* src = argument(1); // type: oop
6908 Node* offset = argument(2); // type: int
6909 Node* end = argument(3); // type: int
6910
6911 Node* length = _gvn.transform(new SubINode(end, offset));
6912
6913 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6914 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6915 // failed array check
6916 return false;
6917 }
6918
6919 // Figure out the size and type of the elements we will be copying.
6920 BasicType src_elem = src_type->elem()->array_element_basic_type();
6921 if (src_elem != T_BYTE) {
6922 return false;
6923 }
6924
6925 // 'src_start' points to src array + scaled offset
6926 src = must_be_not_null(src, true);
6927 Node* src_start = array_element_address(src, offset, src_elem);
6928
6929 // static final int[] byteTable in class CRC32C
6930 Node* table = get_table_from_crc32c_class(callee()->holder());
6931 table = must_be_not_null(table, true);
6932 Node* table_start = array_element_address(table, intcon(0), T_INT);
6933
6934 // We assume that range check is done by caller.
6935 // TODO: generate range check (offset+length < src.length) in debug VM.
6936
6937 // Call the stub.
6938 address stubAddr = StubRoutines::updateBytesCRC32C();
6939 const char *stubName = "updateBytesCRC32C";
6940
6941 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6942 stubAddr, stubName, TypePtr::BOTTOM,
6943 crc, src_start, length, table_start);
6944 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6945 set_result(result);
6946 return true;
6947 }
6948
6949 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6950 //
6951 // Calculate CRC32C for DirectByteBuffer.
6952 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6953 //
6954 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6955 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6956 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6957 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6958 // no receiver since it is a static method
6959 Node* crc = argument(0); // type: int
6960 Node* src = argument(1); // type: long
6961 Node* offset = argument(3); // type: int
6962 Node* end = argument(4); // type: int
6963
6964 Node* length = _gvn.transform(new SubINode(end, offset));
6965
6966 src = ConvL2X(src); // adjust Java long to machine word
6967 Node* base = _gvn.transform(new CastX2PNode(src));
6968 offset = ConvI2X(offset);
6969
6970 // 'src_start' points to src array + scaled offset
6971 Node* src_start = off_heap_plus_addr(base, offset);
6972
6973 // static final int[] byteTable in class CRC32C
6974 Node* table = get_table_from_crc32c_class(callee()->holder());
6975 table = must_be_not_null(table, true);
6976 Node* table_start = array_element_address(table, intcon(0), T_INT);
6977
6978 // Call the stub.
6979 address stubAddr = StubRoutines::updateBytesCRC32C();
6980 const char *stubName = "updateBytesCRC32C";
6981
6982 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6983 stubAddr, stubName, TypePtr::BOTTOM,
6984 crc, src_start, length, table_start);
6985 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6986 set_result(result);
6987 return true;
6988 }
6989
6990 //------------------------------inline_updateBytesAdler32----------------------
6991 //
6992 // Calculate Adler32 checksum for byte[] array.
6993 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6994 //
6995 bool LibraryCallKit::inline_updateBytesAdler32() {
6996 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6997 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6998 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6999 // no receiver since it is static method
7000 Node* crc = argument(0); // type: int
7001 Node* src = argument(1); // type: oop
7002 Node* offset = argument(2); // type: int
7003 Node* length = argument(3); // type: int
7004
7005 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7006 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7007 // failed array check
7008 return false;
7009 }
7010
7011 // Figure out the size and type of the elements we will be copying.
7012 BasicType src_elem = src_type->elem()->array_element_basic_type();
7013 if (src_elem != T_BYTE) {
7014 return false;
7015 }
7016
7017 // 'src_start' points to src array + scaled offset
7018 Node* src_start = array_element_address(src, offset, src_elem);
7019
7020 // We assume that range check is done by caller.
7021 // TODO: generate range check (offset+length < src.length) in debug VM.
7022
7023 // Call the stub.
7024 address stubAddr = StubRoutines::updateBytesAdler32();
7025 const char *stubName = "updateBytesAdler32";
7026
7027 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7028 stubAddr, stubName, TypePtr::BOTTOM,
7029 crc, src_start, length);
7030 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7031 set_result(result);
7032 return true;
7033 }
7034
7035 //------------------------------inline_updateByteBufferAdler32---------------
7036 //
7037 // Calculate Adler32 checksum for DirectByteBuffer.
7038 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
7039 //
7040 bool LibraryCallKit::inline_updateByteBufferAdler32() {
7041 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7042 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7043 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7044 // no receiver since it is static method
7045 Node* crc = argument(0); // type: int
7046 Node* src = argument(1); // type: long
7047 Node* offset = argument(3); // type: int
7048 Node* length = argument(4); // type: int
7049
7050 src = ConvL2X(src); // adjust Java long to machine word
7051 Node* base = _gvn.transform(new CastX2PNode(src));
7052 offset = ConvI2X(offset);
7053
7054 // 'src_start' points to src array + scaled offset
7055 Node* src_start = off_heap_plus_addr(base, offset);
7056
7057 // Call the stub.
7058 address stubAddr = StubRoutines::updateBytesAdler32();
7059 const char *stubName = "updateBytesAdler32";
7060
7061 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7062 stubAddr, stubName, TypePtr::BOTTOM,
7063 crc, src_start, length);
7064
7065 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7066 set_result(result);
7067 return true;
7068 }
7069
7070 //----------------------------inline_reference_get0----------------------------
7071 // public T java.lang.ref.Reference.get();
7072 bool LibraryCallKit::inline_reference_get0() {
7073 const int referent_offset = java_lang_ref_Reference::referent_offset();
7074
7075 // Get the argument:
7076 Node* reference_obj = null_check_receiver();
7077 if (stopped()) return true;
7078
7079 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7080 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7081 decorators, /*is_static*/ false,
7082 env()->Reference_klass());
7083 if (result == nullptr) return false;
7084
7085 // Add memory barrier to prevent commoning reads from this field
7086 // across safepoint since GC can change its value.
7087 insert_mem_bar(Op_MemBarCPUOrder);
7088
7089 set_result(result);
7090 return true;
7091 }
7092
7093 //----------------------------inline_reference_refersTo0----------------------------
7094 // bool java.lang.ref.Reference.refersTo0();
7095 // bool java.lang.ref.PhantomReference.refersTo0();
7096 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7097 // Get arguments:
7098 Node* reference_obj = null_check_receiver();
7099 Node* other_obj = argument(1);
7100 if (stopped()) return true;
7101
7102 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7103 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7104 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7105 decorators, /*is_static*/ false,
7106 env()->Reference_klass());
7107 if (referent == nullptr) return false;
7108
7109 // Add memory barrier to prevent commoning reads from this field
7110 // across safepoint since GC can change its value.
7111 insert_mem_bar(Op_MemBarCPUOrder);
7112
7113 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7114 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7115 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7116
7117 RegionNode* region = new RegionNode(3);
7118 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7119
7120 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7121 region->init_req(1, if_true);
7122 phi->init_req(1, intcon(1));
7123
7124 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7125 region->init_req(2, if_false);
7126 phi->init_req(2, intcon(0));
7127
7128 set_control(_gvn.transform(region));
7129 record_for_igvn(region);
7130 set_result(_gvn.transform(phi));
7131 return true;
7132 }
7133
7134 //----------------------------inline_reference_clear0----------------------------
7135 // void java.lang.ref.Reference.clear0();
7136 // void java.lang.ref.PhantomReference.clear0();
7137 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7138 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7139
7140 // Get arguments
7141 Node* reference_obj = null_check_receiver();
7142 if (stopped()) return true;
7143
7144 // Common access parameters
7145 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7146 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7147 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7148 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7149 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7150
7151 Node* referent = access_load_at(reference_obj,
7152 referent_field_addr,
7153 referent_field_addr_type,
7154 val_type,
7155 T_OBJECT,
7156 decorators);
7157
7158 IdealKit ideal(this);
7159 #define __ ideal.
7160 __ if_then(referent, BoolTest::ne, null());
7161 sync_kit(ideal);
7162 access_store_at(reference_obj,
7163 referent_field_addr,
7164 referent_field_addr_type,
7165 null(),
7166 val_type,
7167 T_OBJECT,
7168 decorators);
7169 __ sync_kit(this);
7170 __ end_if();
7171 final_sync(ideal);
7172 #undef __
7173
7174 return true;
7175 }
7176
7177 //-----------------------inline_reference_reachabilityFence-----------------
7178 // bool java.lang.ref.Reference.reachabilityFence();
7179 bool LibraryCallKit::inline_reference_reachabilityFence() {
7180 Node* referent = argument(0);
7181 insert_reachability_fence(referent);
7182 return true;
7183 }
7184
7185 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7186 DecoratorSet decorators, bool is_static,
7187 ciInstanceKlass* fromKls) {
7188 if (fromKls == nullptr) {
7189 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7190 assert(tinst != nullptr, "obj is null");
7191 assert(tinst->is_loaded(), "obj is not loaded");
7192 fromKls = tinst->instance_klass();
7193 }
7194 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7195 ciSymbol::make(fieldTypeString),
7196 is_static);
7197
7198 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7199 if (field == nullptr) return (Node *) nullptr;
7200
7201 if (is_static) {
7202 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7203 fromObj = makecon(tip);
7204 }
7205
7206 // Next code copied from Parse::do_get_xxx():
7207
7208 // Compute address and memory type.
7209 int offset = field->offset_in_bytes();
7210 bool is_vol = field->is_volatile();
7211 ciType* field_klass = field->type();
7212 assert(field_klass->is_loaded(), "should be loaded");
7213 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7214 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7215 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7216 "slice of address and input slice don't match");
7217 BasicType bt = field->layout_type();
7218
7219 // Build the resultant type of the load
7220 const Type *type;
7221 if (bt == T_OBJECT) {
7222 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7223 } else {
7224 type = Type::get_const_basic_type(bt);
7225 }
7226
7227 if (is_vol) {
7228 decorators |= MO_SEQ_CST;
7229 }
7230
7231 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7232 }
7233
7234 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7235 bool is_exact /* true */, bool is_static /* false */,
7236 ciInstanceKlass * fromKls /* nullptr */) {
7237 if (fromKls == nullptr) {
7238 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7239 assert(tinst != nullptr, "obj is null");
7240 assert(tinst->is_loaded(), "obj is not loaded");
7241 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7242 fromKls = tinst->instance_klass();
7243 }
7244 else {
7245 assert(is_static, "only for static field access");
7246 }
7247 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7248 ciSymbol::make(fieldTypeString),
7249 is_static);
7250
7251 assert(field != nullptr, "undefined field");
7252 assert(!field->is_volatile(), "not defined for volatile fields");
7253
7254 if (is_static) {
7255 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7256 fromObj = makecon(tip);
7257 }
7258
7259 // Next code copied from Parse::do_get_xxx():
7260
7261 // Compute address and memory type.
7262 int offset = field->offset_in_bytes();
7263 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7264
7265 return adr;
7266 }
7267
7268 //------------------------------inline_aescrypt_Block-----------------------
7269 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7270 address stubAddr = nullptr;
7271 const char *stubName;
7272 bool is_decrypt = false;
7273 assert(UseAES, "need AES instruction support");
7274
7275 switch(id) {
7276 case vmIntrinsics::_aescrypt_encryptBlock:
7277 stubAddr = StubRoutines::aescrypt_encryptBlock();
7278 stubName = "aescrypt_encryptBlock";
7279 break;
7280 case vmIntrinsics::_aescrypt_decryptBlock:
7281 stubAddr = StubRoutines::aescrypt_decryptBlock();
7282 stubName = "aescrypt_decryptBlock";
7283 is_decrypt = true;
7284 break;
7285 default:
7286 break;
7287 }
7288 if (stubAddr == nullptr) return false;
7289
7290 Node* aescrypt_object = argument(0);
7291 Node* src = argument(1);
7292 Node* src_offset = argument(2);
7293 Node* dest = argument(3);
7294 Node* dest_offset = argument(4);
7295
7296 src = must_be_not_null(src, true);
7297 dest = must_be_not_null(dest, true);
7298
7299 // (1) src and dest are arrays.
7300 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7301 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7302 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7303 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7304
7305 // for the quick and dirty code we will skip all the checks.
7306 // we are just trying to get the call to be generated.
7307 Node* src_start = src;
7308 Node* dest_start = dest;
7309 if (src_offset != nullptr || dest_offset != nullptr) {
7310 assert(src_offset != nullptr && dest_offset != nullptr, "");
7311 src_start = array_element_address(src, src_offset, T_BYTE);
7312 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7313 }
7314
7315 // now need to get the start of its expanded key array
7316 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7317 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7318 if (k_start == nullptr) return false;
7319
7320 // Call the stub.
7321 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7322 stubAddr, stubName, TypePtr::BOTTOM,
7323 src_start, dest_start, k_start);
7324
7325 return true;
7326 }
7327
7328 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7329 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7330 address stubAddr = nullptr;
7331 const char *stubName = nullptr;
7332 bool is_decrypt = false;
7333 assert(UseAES, "need AES instruction support");
7334
7335 switch(id) {
7336 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7337 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7338 stubName = "cipherBlockChaining_encryptAESCrypt";
7339 break;
7340 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7341 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7342 stubName = "cipherBlockChaining_decryptAESCrypt";
7343 is_decrypt = true;
7344 break;
7345 default:
7346 break;
7347 }
7348 if (stubAddr == nullptr) return false;
7349
7350 Node* cipherBlockChaining_object = argument(0);
7351 Node* src = argument(1);
7352 Node* src_offset = argument(2);
7353 Node* len = argument(3);
7354 Node* dest = argument(4);
7355 Node* dest_offset = argument(5);
7356
7357 src = must_be_not_null(src, false);
7358 dest = must_be_not_null(dest, false);
7359
7360 // (1) src and dest are arrays.
7361 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7362 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7363 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7364 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7365
7366 // checks are the responsibility of the caller
7367 Node* src_start = src;
7368 Node* dest_start = dest;
7369 if (src_offset != nullptr || dest_offset != nullptr) {
7370 assert(src_offset != nullptr && dest_offset != nullptr, "");
7371 src_start = array_element_address(src, src_offset, T_BYTE);
7372 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7373 }
7374
7375 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7376 // (because of the predicated logic executed earlier).
7377 // so we cast it here safely.
7378 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7379
7380 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7381 if (embeddedCipherObj == nullptr) return false;
7382
7383 // cast it to what we know it will be at runtime
7384 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7385 assert(tinst != nullptr, "CBC obj is null");
7386 assert(tinst->is_loaded(), "CBC obj is not loaded");
7387 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7388 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7389
7390 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7391 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7392 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7393 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7394 aescrypt_object = _gvn.transform(aescrypt_object);
7395
7396 // we need to get the start of the aescrypt_object's expanded key array
7397 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7398 if (k_start == nullptr) return false;
7399
7400 // similarly, get the start address of the r vector
7401 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7402 if (objRvec == nullptr) return false;
7403 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7404
7405 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7406 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7407 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7408 stubAddr, stubName, TypePtr::BOTTOM,
7409 src_start, dest_start, k_start, r_start, len);
7410
7411 // return cipher length (int)
7412 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7413 set_result(retvalue);
7414 return true;
7415 }
7416
7417 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7418 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7419 address stubAddr = nullptr;
7420 const char *stubName = nullptr;
7421 bool is_decrypt = false;
7422 assert(UseAES, "need AES instruction support");
7423
7424 switch (id) {
7425 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7426 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7427 stubName = "electronicCodeBook_encryptAESCrypt";
7428 break;
7429 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7430 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7431 stubName = "electronicCodeBook_decryptAESCrypt";
7432 is_decrypt = true;
7433 break;
7434 default:
7435 break;
7436 }
7437
7438 if (stubAddr == nullptr) return false;
7439
7440 Node* electronicCodeBook_object = argument(0);
7441 Node* src = argument(1);
7442 Node* src_offset = argument(2);
7443 Node* len = argument(3);
7444 Node* dest = argument(4);
7445 Node* dest_offset = argument(5);
7446
7447 // (1) src and dest are arrays.
7448 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7449 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7450 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7451 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7452
7453 // checks are the responsibility of the caller
7454 Node* src_start = src;
7455 Node* dest_start = dest;
7456 if (src_offset != nullptr || dest_offset != nullptr) {
7457 assert(src_offset != nullptr && dest_offset != nullptr, "");
7458 src_start = array_element_address(src, src_offset, T_BYTE);
7459 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7460 }
7461
7462 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7463 // (because of the predicated logic executed earlier).
7464 // so we cast it here safely.
7465 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7466
7467 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7468 if (embeddedCipherObj == nullptr) return false;
7469
7470 // cast it to what we know it will be at runtime
7471 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7472 assert(tinst != nullptr, "ECB obj is null");
7473 assert(tinst->is_loaded(), "ECB obj is not loaded");
7474 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7475 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7476
7477 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7478 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7479 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7480 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7481 aescrypt_object = _gvn.transform(aescrypt_object);
7482
7483 // we need to get the start of the aescrypt_object's expanded key array
7484 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7485 if (k_start == nullptr) return false;
7486
7487 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7488 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7489 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7490 stubAddr, stubName, TypePtr::BOTTOM,
7491 src_start, dest_start, k_start, len);
7492
7493 // return cipher length (int)
7494 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7495 set_result(retvalue);
7496 return true;
7497 }
7498
7499 //------------------------------inline_counterMode_AESCrypt-----------------------
7500 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7501 assert(UseAES, "need AES instruction support");
7502 if (!UseAESCTRIntrinsics) return false;
7503
7504 address stubAddr = nullptr;
7505 const char *stubName = nullptr;
7506 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7507 stubAddr = StubRoutines::counterMode_AESCrypt();
7508 stubName = "counterMode_AESCrypt";
7509 }
7510 if (stubAddr == nullptr) return false;
7511
7512 Node* counterMode_object = argument(0);
7513 Node* src = argument(1);
7514 Node* src_offset = argument(2);
7515 Node* len = argument(3);
7516 Node* dest = argument(4);
7517 Node* dest_offset = argument(5);
7518
7519 // (1) src and dest are arrays.
7520 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7521 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7522 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7523 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7524
7525 // checks are the responsibility of the caller
7526 Node* src_start = src;
7527 Node* dest_start = dest;
7528 if (src_offset != nullptr || dest_offset != nullptr) {
7529 assert(src_offset != nullptr && dest_offset != nullptr, "");
7530 src_start = array_element_address(src, src_offset, T_BYTE);
7531 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7532 }
7533
7534 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7535 // (because of the predicated logic executed earlier).
7536 // so we cast it here safely.
7537 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7538 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7539 if (embeddedCipherObj == nullptr) return false;
7540 // cast it to what we know it will be at runtime
7541 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7542 assert(tinst != nullptr, "CTR obj is null");
7543 assert(tinst->is_loaded(), "CTR obj is not loaded");
7544 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7545 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7546 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7547 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7548 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7549 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7550 aescrypt_object = _gvn.transform(aescrypt_object);
7551 // we need to get the start of the aescrypt_object's expanded key array
7552 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
7553 if (k_start == nullptr) return false;
7554 // similarly, get the start address of the r vector
7555 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7556 if (obj_counter == nullptr) return false;
7557 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7558
7559 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7560 if (saved_encCounter == nullptr) return false;
7561 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7562 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7563
7564 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7565 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7566 OptoRuntime::counterMode_aescrypt_Type(),
7567 stubAddr, stubName, TypePtr::BOTTOM,
7568 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7569
7570 // return cipher length (int)
7571 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7572 set_result(retvalue);
7573 return true;
7574 }
7575
7576 //------------------------------get_key_start_from_aescrypt_object-----------------------
7577 Node* LibraryCallKit::get_key_start_from_aescrypt_object(Node* aescrypt_object, bool is_decrypt) {
7578 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7579 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7580 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7581 // The following platform specific stubs of encryption and decryption use the same round keys.
7582 #if defined(PPC64) || defined(S390) || defined(RISCV64)
7583 bool use_decryption_key = false;
7584 #else
7585 bool use_decryption_key = is_decrypt;
7586 #endif
7587 Node* objAESCryptKey = load_field_from_object(aescrypt_object, use_decryption_key ? "sessionKd" : "sessionKe", "[I");
7588 assert(objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AES_Crypt");
7589 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7590
7591 // now have the array, need to get the start address of the selected key array
7592 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7593 return k_start;
7594 }
7595
7596 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7597 // Return node representing slow path of predicate check.
7598 // the pseudo code we want to emulate with this predicate is:
7599 // for encryption:
7600 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7601 // for decryption:
7602 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7603 // note cipher==plain is more conservative than the original java code but that's OK
7604 //
7605 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7606 // The receiver was checked for null already.
7607 Node* objCBC = argument(0);
7608
7609 Node* src = argument(1);
7610 Node* dest = argument(4);
7611
7612 // Load embeddedCipher field of CipherBlockChaining object.
7613 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7614
7615 // get AESCrypt klass for instanceOf check
7616 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7617 // will have same classloader as CipherBlockChaining object
7618 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7619 assert(tinst != nullptr, "CBCobj is null");
7620 assert(tinst->is_loaded(), "CBCobj is not loaded");
7621
7622 // we want to do an instanceof comparison against the AESCrypt class
7623 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7624 if (!klass_AESCrypt->is_loaded()) {
7625 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7626 Node* ctrl = control();
7627 set_control(top()); // no regular fast path
7628 return ctrl;
7629 }
7630
7631 src = must_be_not_null(src, true);
7632 dest = must_be_not_null(dest, true);
7633
7634 // Resolve oops to stable for CmpP below.
7635 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7636
7637 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7638 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7639 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7640
7641 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7642
7643 // for encryption, we are done
7644 if (!decrypting)
7645 return instof_false; // even if it is null
7646
7647 // for decryption, we need to add a further check to avoid
7648 // taking the intrinsic path when cipher and plain are the same
7649 // see the original java code for why.
7650 RegionNode* region = new RegionNode(3);
7651 region->init_req(1, instof_false);
7652
7653 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7654 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7655 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7656 region->init_req(2, src_dest_conjoint);
7657
7658 record_for_igvn(region);
7659 return _gvn.transform(region);
7660 }
7661
7662 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7663 // Return node representing slow path of predicate check.
7664 // the pseudo code we want to emulate with this predicate is:
7665 // for encryption:
7666 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7667 // for decryption:
7668 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7669 // note cipher==plain is more conservative than the original java code but that's OK
7670 //
7671 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7672 // The receiver was checked for null already.
7673 Node* objECB = argument(0);
7674
7675 // Load embeddedCipher field of ElectronicCodeBook object.
7676 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7677
7678 // get AESCrypt klass for instanceOf check
7679 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7680 // will have same classloader as ElectronicCodeBook object
7681 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7682 assert(tinst != nullptr, "ECBobj is null");
7683 assert(tinst->is_loaded(), "ECBobj is not loaded");
7684
7685 // we want to do an instanceof comparison against the AESCrypt class
7686 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7687 if (!klass_AESCrypt->is_loaded()) {
7688 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7689 Node* ctrl = control();
7690 set_control(top()); // no regular fast path
7691 return ctrl;
7692 }
7693 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7694
7695 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7696 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7697 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7698
7699 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7700
7701 // for encryption, we are done
7702 if (!decrypting)
7703 return instof_false; // even if it is null
7704
7705 // for decryption, we need to add a further check to avoid
7706 // taking the intrinsic path when cipher and plain are the same
7707 // see the original java code for why.
7708 RegionNode* region = new RegionNode(3);
7709 region->init_req(1, instof_false);
7710 Node* src = argument(1);
7711 Node* dest = argument(4);
7712 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7713 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7714 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7715 region->init_req(2, src_dest_conjoint);
7716
7717 record_for_igvn(region);
7718 return _gvn.transform(region);
7719 }
7720
7721 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7722 // Return node representing slow path of predicate check.
7723 // the pseudo code we want to emulate with this predicate is:
7724 // for encryption:
7725 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7726 // for decryption:
7727 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7728 // note cipher==plain is more conservative than the original java code but that's OK
7729 //
7730
7731 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7732 // The receiver was checked for null already.
7733 Node* objCTR = argument(0);
7734
7735 // Load embeddedCipher field of CipherBlockChaining object.
7736 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7737
7738 // get AESCrypt klass for instanceOf check
7739 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7740 // will have same classloader as CipherBlockChaining object
7741 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7742 assert(tinst != nullptr, "CTRobj is null");
7743 assert(tinst->is_loaded(), "CTRobj is not loaded");
7744
7745 // we want to do an instanceof comparison against the AESCrypt class
7746 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7747 if (!klass_AESCrypt->is_loaded()) {
7748 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7749 Node* ctrl = control();
7750 set_control(top()); // no regular fast path
7751 return ctrl;
7752 }
7753
7754 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7755 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7756 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7757 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7758 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7759
7760 return instof_false; // even if it is null
7761 }
7762
7763 //------------------------------inline_ghash_processBlocks
7764 bool LibraryCallKit::inline_ghash_processBlocks() {
7765 address stubAddr;
7766 const char *stubName;
7767 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7768
7769 stubAddr = StubRoutines::ghash_processBlocks();
7770 stubName = "ghash_processBlocks";
7771
7772 Node* data = argument(0);
7773 Node* offset = argument(1);
7774 Node* len = argument(2);
7775 Node* state = argument(3);
7776 Node* subkeyH = argument(4);
7777
7778 state = must_be_not_null(state, true);
7779 subkeyH = must_be_not_null(subkeyH, true);
7780 data = must_be_not_null(data, true);
7781
7782 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7783 assert(state_start, "state is null");
7784 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7785 assert(subkeyH_start, "subkeyH is null");
7786 Node* data_start = array_element_address(data, offset, T_BYTE);
7787 assert(data_start, "data is null");
7788
7789 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7790 OptoRuntime::ghash_processBlocks_Type(),
7791 stubAddr, stubName, TypePtr::BOTTOM,
7792 state_start, subkeyH_start, data_start, len);
7793 return true;
7794 }
7795
7796 //------------------------------inline_chacha20Block
7797 bool LibraryCallKit::inline_chacha20Block() {
7798 address stubAddr;
7799 const char *stubName;
7800 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7801
7802 stubAddr = StubRoutines::chacha20Block();
7803 stubName = "chacha20Block";
7804
7805 Node* state = argument(0);
7806 Node* result = argument(1);
7807
7808 state = must_be_not_null(state, true);
7809 result = must_be_not_null(result, true);
7810
7811 Node* state_start = array_element_address(state, intcon(0), T_INT);
7812 assert(state_start, "state is null");
7813 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7814 assert(result_start, "result is null");
7815
7816 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7817 OptoRuntime::chacha20Block_Type(),
7818 stubAddr, stubName, TypePtr::BOTTOM,
7819 state_start, result_start);
7820 // return key stream length (int)
7821 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7822 set_result(retvalue);
7823 return true;
7824 }
7825
7826 //------------------------------inline_kyberNtt
7827 bool LibraryCallKit::inline_kyberNtt() {
7828 address stubAddr;
7829 const char *stubName;
7830 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7831 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
7832
7833 stubAddr = StubRoutines::kyberNtt();
7834 stubName = "kyberNtt";
7835 if (!stubAddr) return false;
7836
7837 Node* coeffs = argument(0);
7838 Node* ntt_zetas = argument(1);
7839
7840 coeffs = must_be_not_null(coeffs, true);
7841 ntt_zetas = must_be_not_null(ntt_zetas, true);
7842
7843 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7844 assert(coeffs_start, "coeffs is null");
7845 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
7846 assert(ntt_zetas_start, "ntt_zetas is null");
7847 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7848 OptoRuntime::kyberNtt_Type(),
7849 stubAddr, stubName, TypePtr::BOTTOM,
7850 coeffs_start, ntt_zetas_start);
7851 // return an int
7852 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
7853 set_result(retvalue);
7854 return true;
7855 }
7856
7857 //------------------------------inline_kyberInverseNtt
7858 bool LibraryCallKit::inline_kyberInverseNtt() {
7859 address stubAddr;
7860 const char *stubName;
7861 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7862 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
7863
7864 stubAddr = StubRoutines::kyberInverseNtt();
7865 stubName = "kyberInverseNtt";
7866 if (!stubAddr) return false;
7867
7868 Node* coeffs = argument(0);
7869 Node* zetas = argument(1);
7870
7871 coeffs = must_be_not_null(coeffs, true);
7872 zetas = must_be_not_null(zetas, true);
7873
7874 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7875 assert(coeffs_start, "coeffs is null");
7876 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
7877 assert(zetas_start, "inverseNtt_zetas is null");
7878 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7879 OptoRuntime::kyberInverseNtt_Type(),
7880 stubAddr, stubName, TypePtr::BOTTOM,
7881 coeffs_start, zetas_start);
7882
7883 // return an int
7884 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
7885 set_result(retvalue);
7886 return true;
7887 }
7888
7889 //------------------------------inline_kyberNttMult
7890 bool LibraryCallKit::inline_kyberNttMult() {
7891 address stubAddr;
7892 const char *stubName;
7893 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7894 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
7895
7896 stubAddr = StubRoutines::kyberNttMult();
7897 stubName = "kyberNttMult";
7898 if (!stubAddr) return false;
7899
7900 Node* result = argument(0);
7901 Node* ntta = argument(1);
7902 Node* nttb = argument(2);
7903 Node* zetas = argument(3);
7904
7905 result = must_be_not_null(result, true);
7906 ntta = must_be_not_null(ntta, true);
7907 nttb = must_be_not_null(nttb, true);
7908 zetas = must_be_not_null(zetas, true);
7909
7910 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7911 assert(result_start, "result is null");
7912 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
7913 assert(ntta_start, "ntta is null");
7914 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
7915 assert(nttb_start, "nttb is null");
7916 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
7917 assert(zetas_start, "nttMult_zetas is null");
7918 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
7919 OptoRuntime::kyberNttMult_Type(),
7920 stubAddr, stubName, TypePtr::BOTTOM,
7921 result_start, ntta_start, nttb_start,
7922 zetas_start);
7923
7924 // return an int
7925 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
7926 set_result(retvalue);
7927
7928 return true;
7929 }
7930
7931 //------------------------------inline_kyberAddPoly_2
7932 bool LibraryCallKit::inline_kyberAddPoly_2() {
7933 address stubAddr;
7934 const char *stubName;
7935 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7936 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
7937
7938 stubAddr = StubRoutines::kyberAddPoly_2();
7939 stubName = "kyberAddPoly_2";
7940 if (!stubAddr) return false;
7941
7942 Node* result = argument(0);
7943 Node* a = argument(1);
7944 Node* b = argument(2);
7945
7946 result = must_be_not_null(result, true);
7947 a = must_be_not_null(a, true);
7948 b = must_be_not_null(b, true);
7949
7950 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7951 assert(result_start, "result is null");
7952 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
7953 assert(a_start, "a is null");
7954 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
7955 assert(b_start, "b is null");
7956 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
7957 OptoRuntime::kyberAddPoly_2_Type(),
7958 stubAddr, stubName, TypePtr::BOTTOM,
7959 result_start, a_start, b_start);
7960 // return an int
7961 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
7962 set_result(retvalue);
7963 return true;
7964 }
7965
7966 //------------------------------inline_kyberAddPoly_3
7967 bool LibraryCallKit::inline_kyberAddPoly_3() {
7968 address stubAddr;
7969 const char *stubName;
7970 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7971 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
7972
7973 stubAddr = StubRoutines::kyberAddPoly_3();
7974 stubName = "kyberAddPoly_3";
7975 if (!stubAddr) return false;
7976
7977 Node* result = argument(0);
7978 Node* a = argument(1);
7979 Node* b = argument(2);
7980 Node* c = argument(3);
7981
7982 result = must_be_not_null(result, true);
7983 a = must_be_not_null(a, true);
7984 b = must_be_not_null(b, true);
7985 c = must_be_not_null(c, true);
7986
7987 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7988 assert(result_start, "result is null");
7989 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
7990 assert(a_start, "a is null");
7991 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
7992 assert(b_start, "b is null");
7993 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
7994 assert(c_start, "c is null");
7995 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
7996 OptoRuntime::kyberAddPoly_3_Type(),
7997 stubAddr, stubName, TypePtr::BOTTOM,
7998 result_start, a_start, b_start, c_start);
7999 // return an int
8000 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
8001 set_result(retvalue);
8002 return true;
8003 }
8004
8005 //------------------------------inline_kyber12To16
8006 bool LibraryCallKit::inline_kyber12To16() {
8007 address stubAddr;
8008 const char *stubName;
8009 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8010 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
8011
8012 stubAddr = StubRoutines::kyber12To16();
8013 stubName = "kyber12To16";
8014 if (!stubAddr) return false;
8015
8016 Node* condensed = argument(0);
8017 Node* condensedOffs = argument(1);
8018 Node* parsed = argument(2);
8019 Node* parsedLength = argument(3);
8020
8021 condensed = must_be_not_null(condensed, true);
8022 parsed = must_be_not_null(parsed, true);
8023
8024 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
8025 assert(condensed_start, "condensed is null");
8026 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
8027 assert(parsed_start, "parsed is null");
8028 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
8029 OptoRuntime::kyber12To16_Type(),
8030 stubAddr, stubName, TypePtr::BOTTOM,
8031 condensed_start, condensedOffs, parsed_start, parsedLength);
8032 // return an int
8033 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
8034 set_result(retvalue);
8035 return true;
8036
8037 }
8038
8039 //------------------------------inline_kyberBarrettReduce
8040 bool LibraryCallKit::inline_kyberBarrettReduce() {
8041 address stubAddr;
8042 const char *stubName;
8043 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8044 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
8045
8046 stubAddr = StubRoutines::kyberBarrettReduce();
8047 stubName = "kyberBarrettReduce";
8048 if (!stubAddr) return false;
8049
8050 Node* coeffs = argument(0);
8051
8052 coeffs = must_be_not_null(coeffs, true);
8053
8054 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8055 assert(coeffs_start, "coeffs is null");
8056 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
8057 OptoRuntime::kyberBarrettReduce_Type(),
8058 stubAddr, stubName, TypePtr::BOTTOM,
8059 coeffs_start);
8060 // return an int
8061 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
8062 set_result(retvalue);
8063 return true;
8064 }
8065
8066 //------------------------------inline_dilithiumAlmostNtt
8067 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
8068 address stubAddr;
8069 const char *stubName;
8070 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8071 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
8072
8073 stubAddr = StubRoutines::dilithiumAlmostNtt();
8074 stubName = "dilithiumAlmostNtt";
8075 if (!stubAddr) return false;
8076
8077 Node* coeffs = argument(0);
8078 Node* ntt_zetas = argument(1);
8079
8080 coeffs = must_be_not_null(coeffs, true);
8081 ntt_zetas = must_be_not_null(ntt_zetas, true);
8082
8083 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8084 assert(coeffs_start, "coeffs is null");
8085 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
8086 assert(ntt_zetas_start, "ntt_zetas is null");
8087 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8088 OptoRuntime::dilithiumAlmostNtt_Type(),
8089 stubAddr, stubName, TypePtr::BOTTOM,
8090 coeffs_start, ntt_zetas_start);
8091 // return an int
8092 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
8093 set_result(retvalue);
8094 return true;
8095 }
8096
8097 //------------------------------inline_dilithiumAlmostInverseNtt
8098 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
8099 address stubAddr;
8100 const char *stubName;
8101 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8102 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
8103
8104 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
8105 stubName = "dilithiumAlmostInverseNtt";
8106 if (!stubAddr) return false;
8107
8108 Node* coeffs = argument(0);
8109 Node* zetas = argument(1);
8110
8111 coeffs = must_be_not_null(coeffs, true);
8112 zetas = must_be_not_null(zetas, true);
8113
8114 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8115 assert(coeffs_start, "coeffs is null");
8116 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
8117 assert(zetas_start, "inverseNtt_zetas is null");
8118 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8119 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
8120 stubAddr, stubName, TypePtr::BOTTOM,
8121 coeffs_start, zetas_start);
8122 // return an int
8123 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
8124 set_result(retvalue);
8125 return true;
8126 }
8127
8128 //------------------------------inline_dilithiumNttMult
8129 bool LibraryCallKit::inline_dilithiumNttMult() {
8130 address stubAddr;
8131 const char *stubName;
8132 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8133 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
8134
8135 stubAddr = StubRoutines::dilithiumNttMult();
8136 stubName = "dilithiumNttMult";
8137 if (!stubAddr) return false;
8138
8139 Node* result = argument(0);
8140 Node* ntta = argument(1);
8141 Node* nttb = argument(2);
8142 Node* zetas = argument(3);
8143
8144 result = must_be_not_null(result, true);
8145 ntta = must_be_not_null(ntta, true);
8146 nttb = must_be_not_null(nttb, true);
8147 zetas = must_be_not_null(zetas, true);
8148
8149 Node* result_start = array_element_address(result, intcon(0), T_INT);
8150 assert(result_start, "result is null");
8151 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
8152 assert(ntta_start, "ntta is null");
8153 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
8154 assert(nttb_start, "nttb is null");
8155 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8156 OptoRuntime::dilithiumNttMult_Type(),
8157 stubAddr, stubName, TypePtr::BOTTOM,
8158 result_start, ntta_start, nttb_start);
8159
8160 // return an int
8161 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8162 set_result(retvalue);
8163
8164 return true;
8165 }
8166
8167 //------------------------------inline_dilithiumMontMulByConstant
8168 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8169 address stubAddr;
8170 const char *stubName;
8171 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8172 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8173
8174 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8175 stubName = "dilithiumMontMulByConstant";
8176 if (!stubAddr) return false;
8177
8178 Node* coeffs = argument(0);
8179 Node* constant = argument(1);
8180
8181 coeffs = must_be_not_null(coeffs, true);
8182
8183 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8184 assert(coeffs_start, "coeffs is null");
8185 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8186 OptoRuntime::dilithiumMontMulByConstant_Type(),
8187 stubAddr, stubName, TypePtr::BOTTOM,
8188 coeffs_start, constant);
8189
8190 // return an int
8191 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8192 set_result(retvalue);
8193 return true;
8194 }
8195
8196
8197 //------------------------------inline_dilithiumDecomposePoly
8198 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8199 address stubAddr;
8200 const char *stubName;
8201 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8202 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8203
8204 stubAddr = StubRoutines::dilithiumDecomposePoly();
8205 stubName = "dilithiumDecomposePoly";
8206 if (!stubAddr) return false;
8207
8208 Node* input = argument(0);
8209 Node* lowPart = argument(1);
8210 Node* highPart = argument(2);
8211 Node* twoGamma2 = argument(3);
8212 Node* multiplier = argument(4);
8213
8214 input = must_be_not_null(input, true);
8215 lowPart = must_be_not_null(lowPart, true);
8216 highPart = must_be_not_null(highPart, true);
8217
8218 Node* input_start = array_element_address(input, intcon(0), T_INT);
8219 assert(input_start, "input is null");
8220 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8221 assert(lowPart_start, "lowPart is null");
8222 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8223 assert(highPart_start, "highPart is null");
8224
8225 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8226 OptoRuntime::dilithiumDecomposePoly_Type(),
8227 stubAddr, stubName, TypePtr::BOTTOM,
8228 input_start, lowPart_start, highPart_start,
8229 twoGamma2, multiplier);
8230
8231 // return an int
8232 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8233 set_result(retvalue);
8234 return true;
8235 }
8236
8237 bool LibraryCallKit::inline_base64_encodeBlock() {
8238 address stubAddr;
8239 const char *stubName;
8240 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8241 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8242 stubAddr = StubRoutines::base64_encodeBlock();
8243 stubName = "encodeBlock";
8244
8245 if (!stubAddr) return false;
8246 Node* base64obj = argument(0);
8247 Node* src = argument(1);
8248 Node* offset = argument(2);
8249 Node* len = argument(3);
8250 Node* dest = argument(4);
8251 Node* dp = argument(5);
8252 Node* isURL = argument(6);
8253
8254 src = must_be_not_null(src, true);
8255 dest = must_be_not_null(dest, true);
8256
8257 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8258 assert(src_start, "source array is null");
8259 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8260 assert(dest_start, "destination array is null");
8261
8262 Node* base64 = make_runtime_call(RC_LEAF,
8263 OptoRuntime::base64_encodeBlock_Type(),
8264 stubAddr, stubName, TypePtr::BOTTOM,
8265 src_start, offset, len, dest_start, dp, isURL);
8266 return true;
8267 }
8268
8269 bool LibraryCallKit::inline_base64_decodeBlock() {
8270 address stubAddr;
8271 const char *stubName;
8272 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8273 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8274 stubAddr = StubRoutines::base64_decodeBlock();
8275 stubName = "decodeBlock";
8276
8277 if (!stubAddr) return false;
8278 Node* base64obj = argument(0);
8279 Node* src = argument(1);
8280 Node* src_offset = argument(2);
8281 Node* len = argument(3);
8282 Node* dest = argument(4);
8283 Node* dest_offset = argument(5);
8284 Node* isURL = argument(6);
8285 Node* isMIME = argument(7);
8286
8287 src = must_be_not_null(src, true);
8288 dest = must_be_not_null(dest, true);
8289
8290 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8291 assert(src_start, "source array is null");
8292 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8293 assert(dest_start, "destination array is null");
8294
8295 Node* call = make_runtime_call(RC_LEAF,
8296 OptoRuntime::base64_decodeBlock_Type(),
8297 stubAddr, stubName, TypePtr::BOTTOM,
8298 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8299 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8300 set_result(result);
8301 return true;
8302 }
8303
8304 bool LibraryCallKit::inline_poly1305_processBlocks() {
8305 address stubAddr;
8306 const char *stubName;
8307 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8308 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8309 stubAddr = StubRoutines::poly1305_processBlocks();
8310 stubName = "poly1305_processBlocks";
8311
8312 if (!stubAddr) return false;
8313 null_check_receiver(); // null-check receiver
8314 if (stopped()) return true;
8315
8316 Node* input = argument(1);
8317 Node* input_offset = argument(2);
8318 Node* len = argument(3);
8319 Node* alimbs = argument(4);
8320 Node* rlimbs = argument(5);
8321
8322 input = must_be_not_null(input, true);
8323 alimbs = must_be_not_null(alimbs, true);
8324 rlimbs = must_be_not_null(rlimbs, true);
8325
8326 Node* input_start = array_element_address(input, input_offset, T_BYTE);
8327 assert(input_start, "input array is null");
8328 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
8329 assert(acc_start, "acc array is null");
8330 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
8331 assert(r_start, "r array is null");
8332
8333 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8334 OptoRuntime::poly1305_processBlocks_Type(),
8335 stubAddr, stubName, TypePtr::BOTTOM,
8336 input_start, len, acc_start, r_start);
8337 return true;
8338 }
8339
8340 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
8341 address stubAddr;
8342 const char *stubName;
8343 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8344 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
8345 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
8346 stubName = "intpoly_montgomeryMult_P256";
8347
8348 if (!stubAddr) return false;
8349 null_check_receiver(); // null-check receiver
8350 if (stopped()) return true;
8351
8352 Node* a = argument(1);
8353 Node* b = argument(2);
8354 Node* r = argument(3);
8355
8356 a = must_be_not_null(a, true);
8357 b = must_be_not_null(b, true);
8358 r = must_be_not_null(r, true);
8359
8360 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8361 assert(a_start, "a array is null");
8362 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8363 assert(b_start, "b array is null");
8364 Node* r_start = array_element_address(r, intcon(0), T_LONG);
8365 assert(r_start, "r array is null");
8366
8367 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8368 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
8369 stubAddr, stubName, TypePtr::BOTTOM,
8370 a_start, b_start, r_start);
8371 return true;
8372 }
8373
8374 bool LibraryCallKit::inline_intpoly_assign() {
8375 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8376 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
8377 const char *stubName = "intpoly_assign";
8378 address stubAddr = StubRoutines::intpoly_assign();
8379 if (!stubAddr) return false;
8380
8381 Node* set = argument(0);
8382 Node* a = argument(1);
8383 Node* b = argument(2);
8384 Node* arr_length = load_array_length(a);
8385
8386 a = must_be_not_null(a, true);
8387 b = must_be_not_null(b, true);
8388
8389 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8390 assert(a_start, "a array is null");
8391 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8392 assert(b_start, "b array is null");
8393
8394 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8395 OptoRuntime::intpoly_assign_Type(),
8396 stubAddr, stubName, TypePtr::BOTTOM,
8397 set, a_start, b_start, arr_length);
8398 return true;
8399 }
8400
8401 //------------------------------inline_digestBase_implCompress-----------------------
8402 //
8403 // Calculate MD5 for single-block byte[] array.
8404 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
8405 //
8406 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
8407 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
8408 //
8409 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
8410 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
8411 //
8412 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
8413 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
8414 //
8415 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
8416 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
8417 //
8418 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
8419 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
8420
8421 Node* digestBase_obj = argument(0);
8422 Node* src = argument(1); // type oop
8423 Node* ofs = argument(2); // type int
8424
8425 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8426 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8427 // failed array check
8428 return false;
8429 }
8430 // Figure out the size and type of the elements we will be copying.
8431 BasicType src_elem = src_type->elem()->array_element_basic_type();
8432 if (src_elem != T_BYTE) {
8433 return false;
8434 }
8435 // 'src_start' points to src array + offset
8436 src = must_be_not_null(src, true);
8437 Node* src_start = array_element_address(src, ofs, src_elem);
8438 Node* state = nullptr;
8439 Node* block_size = nullptr;
8440 address stubAddr;
8441 const char *stubName;
8442
8443 switch(id) {
8444 case vmIntrinsics::_md5_implCompress:
8445 assert(UseMD5Intrinsics, "need MD5 instruction support");
8446 state = get_state_from_digest_object(digestBase_obj, T_INT);
8447 stubAddr = StubRoutines::md5_implCompress();
8448 stubName = "md5_implCompress";
8449 break;
8450 case vmIntrinsics::_sha_implCompress:
8451 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
8452 state = get_state_from_digest_object(digestBase_obj, T_INT);
8453 stubAddr = StubRoutines::sha1_implCompress();
8454 stubName = "sha1_implCompress";
8455 break;
8456 case vmIntrinsics::_sha2_implCompress:
8457 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
8458 state = get_state_from_digest_object(digestBase_obj, T_INT);
8459 stubAddr = StubRoutines::sha256_implCompress();
8460 stubName = "sha256_implCompress";
8461 break;
8462 case vmIntrinsics::_sha5_implCompress:
8463 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
8464 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8465 stubAddr = StubRoutines::sha512_implCompress();
8466 stubName = "sha512_implCompress";
8467 break;
8468 case vmIntrinsics::_sha3_implCompress:
8469 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
8470 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8471 stubAddr = StubRoutines::sha3_implCompress();
8472 stubName = "sha3_implCompress";
8473 block_size = get_block_size_from_digest_object(digestBase_obj);
8474 if (block_size == nullptr) return false;
8475 break;
8476 default:
8477 fatal_unexpected_iid(id);
8478 return false;
8479 }
8480 if (state == nullptr) return false;
8481
8482 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
8483 if (stubAddr == nullptr) return false;
8484
8485 // Call the stub.
8486 Node* call;
8487 if (block_size == nullptr) {
8488 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
8489 stubAddr, stubName, TypePtr::BOTTOM,
8490 src_start, state);
8491 } else {
8492 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
8493 stubAddr, stubName, TypePtr::BOTTOM,
8494 src_start, state, block_size);
8495 }
8496
8497 return true;
8498 }
8499
8500 //------------------------------inline_double_keccak
8501 bool LibraryCallKit::inline_double_keccak() {
8502 address stubAddr;
8503 const char *stubName;
8504 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
8505 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
8506
8507 stubAddr = StubRoutines::double_keccak();
8508 stubName = "double_keccak";
8509 if (!stubAddr) return false;
8510
8511 Node* status0 = argument(0);
8512 Node* status1 = argument(1);
8513
8514 status0 = must_be_not_null(status0, true);
8515 status1 = must_be_not_null(status1, true);
8516
8517 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
8518 assert(status0_start, "status0 is null");
8519 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
8520 assert(status1_start, "status1 is null");
8521 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
8522 OptoRuntime::double_keccak_Type(),
8523 stubAddr, stubName, TypePtr::BOTTOM,
8524 status0_start, status1_start);
8525 // return an int
8526 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
8527 set_result(retvalue);
8528 return true;
8529 }
8530
8531
8532 //------------------------------inline_digestBase_implCompressMB-----------------------
8533 //
8534 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
8535 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
8536 //
8537 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
8538 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8539 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8540 assert((uint)predicate < 5, "sanity");
8541 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
8542
8543 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
8544 Node* src = argument(1); // byte[] array
8545 Node* ofs = argument(2); // type int
8546 Node* limit = argument(3); // type int
8547
8548 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8549 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8550 // failed array check
8551 return false;
8552 }
8553 // Figure out the size and type of the elements we will be copying.
8554 BasicType src_elem = src_type->elem()->array_element_basic_type();
8555 if (src_elem != T_BYTE) {
8556 return false;
8557 }
8558 // 'src_start' points to src array + offset
8559 src = must_be_not_null(src, false);
8560 Node* src_start = array_element_address(src, ofs, src_elem);
8561
8562 const char* klass_digestBase_name = nullptr;
8563 const char* stub_name = nullptr;
8564 address stub_addr = nullptr;
8565 BasicType elem_type = T_INT;
8566
8567 switch (predicate) {
8568 case 0:
8569 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
8570 klass_digestBase_name = "sun/security/provider/MD5";
8571 stub_name = "md5_implCompressMB";
8572 stub_addr = StubRoutines::md5_implCompressMB();
8573 }
8574 break;
8575 case 1:
8576 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
8577 klass_digestBase_name = "sun/security/provider/SHA";
8578 stub_name = "sha1_implCompressMB";
8579 stub_addr = StubRoutines::sha1_implCompressMB();
8580 }
8581 break;
8582 case 2:
8583 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
8584 klass_digestBase_name = "sun/security/provider/SHA2";
8585 stub_name = "sha256_implCompressMB";
8586 stub_addr = StubRoutines::sha256_implCompressMB();
8587 }
8588 break;
8589 case 3:
8590 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
8591 klass_digestBase_name = "sun/security/provider/SHA5";
8592 stub_name = "sha512_implCompressMB";
8593 stub_addr = StubRoutines::sha512_implCompressMB();
8594 elem_type = T_LONG;
8595 }
8596 break;
8597 case 4:
8598 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
8599 klass_digestBase_name = "sun/security/provider/SHA3";
8600 stub_name = "sha3_implCompressMB";
8601 stub_addr = StubRoutines::sha3_implCompressMB();
8602 elem_type = T_LONG;
8603 }
8604 break;
8605 default:
8606 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
8607 }
8608 if (klass_digestBase_name != nullptr) {
8609 assert(stub_addr != nullptr, "Stub is generated");
8610 if (stub_addr == nullptr) return false;
8611
8612 // get DigestBase klass to lookup for SHA klass
8613 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
8614 assert(tinst != nullptr, "digestBase_obj is not instance???");
8615 assert(tinst->is_loaded(), "DigestBase is not loaded");
8616
8617 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
8618 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
8619 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
8620 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
8621 }
8622 return false;
8623 }
8624
8625 //------------------------------inline_digestBase_implCompressMB-----------------------
8626 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
8627 BasicType elem_type, address stubAddr, const char *stubName,
8628 Node* src_start, Node* ofs, Node* limit) {
8629 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
8630 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8631 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
8632 digest_obj = _gvn.transform(digest_obj);
8633
8634 Node* state = get_state_from_digest_object(digest_obj, elem_type);
8635 if (state == nullptr) return false;
8636
8637 Node* block_size = nullptr;
8638 if (strcmp("sha3_implCompressMB", stubName) == 0) {
8639 block_size = get_block_size_from_digest_object(digest_obj);
8640 if (block_size == nullptr) return false;
8641 }
8642
8643 // Call the stub.
8644 Node* call;
8645 if (block_size == nullptr) {
8646 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8647 OptoRuntime::digestBase_implCompressMB_Type(false),
8648 stubAddr, stubName, TypePtr::BOTTOM,
8649 src_start, state, ofs, limit);
8650 } else {
8651 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8652 OptoRuntime::digestBase_implCompressMB_Type(true),
8653 stubAddr, stubName, TypePtr::BOTTOM,
8654 src_start, state, block_size, ofs, limit);
8655 }
8656
8657 // return ofs (int)
8658 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8659 set_result(result);
8660
8661 return true;
8662 }
8663
8664 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8665 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8666 assert(UseAES, "need AES instruction support");
8667 address stubAddr = nullptr;
8668 const char *stubName = nullptr;
8669 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8670 stubName = "galoisCounterMode_AESCrypt";
8671
8672 if (stubAddr == nullptr) return false;
8673
8674 Node* in = argument(0);
8675 Node* inOfs = argument(1);
8676 Node* len = argument(2);
8677 Node* ct = argument(3);
8678 Node* ctOfs = argument(4);
8679 Node* out = argument(5);
8680 Node* outOfs = argument(6);
8681 Node* gctr_object = argument(7);
8682 Node* ghash_object = argument(8);
8683
8684 // (1) in, ct and out are arrays.
8685 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8686 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8687 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8688 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8689 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8690 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8691
8692 // checks are the responsibility of the caller
8693 Node* in_start = in;
8694 Node* ct_start = ct;
8695 Node* out_start = out;
8696 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8697 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8698 in_start = array_element_address(in, inOfs, T_BYTE);
8699 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8700 out_start = array_element_address(out, outOfs, T_BYTE);
8701 }
8702
8703 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8704 // (because of the predicated logic executed earlier).
8705 // so we cast it here safely.
8706 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8707 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8708 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8709 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8710 Node* state = load_field_from_object(ghash_object, "state", "[J");
8711
8712 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8713 return false;
8714 }
8715 // cast it to what we know it will be at runtime
8716 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8717 assert(tinst != nullptr, "GCTR obj is null");
8718 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8719 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8720 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8721 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8722 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8723 const TypeOopPtr* xtype = aklass->as_instance_type();
8724 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8725 aescrypt_object = _gvn.transform(aescrypt_object);
8726 // we need to get the start of the aescrypt_object's expanded key array
8727 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
8728 if (k_start == nullptr) return false;
8729 // similarly, get the start address of the r vector
8730 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8731 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8732 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8733
8734
8735 // Call the stub, passing params
8736 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8737 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8738 stubAddr, stubName, TypePtr::BOTTOM,
8739 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8740
8741 // return cipher length (int)
8742 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8743 set_result(retvalue);
8744
8745 return true;
8746 }
8747
8748 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8749 // Return node representing slow path of predicate check.
8750 // the pseudo code we want to emulate with this predicate is:
8751 // for encryption:
8752 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8753 // for decryption:
8754 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8755 // note cipher==plain is more conservative than the original java code but that's OK
8756 //
8757
8758 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8759 // The receiver was checked for null already.
8760 Node* objGCTR = argument(7);
8761 // Load embeddedCipher field of GCTR object.
8762 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8763 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8764
8765 // get AESCrypt klass for instanceOf check
8766 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8767 // will have same classloader as CipherBlockChaining object
8768 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8769 assert(tinst != nullptr, "GCTR obj is null");
8770 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8771
8772 // we want to do an instanceof comparison against the AESCrypt class
8773 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8774 if (!klass_AESCrypt->is_loaded()) {
8775 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8776 Node* ctrl = control();
8777 set_control(top()); // no regular fast path
8778 return ctrl;
8779 }
8780
8781 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8782 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8783 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8784 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8785 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8786
8787 return instof_false; // even if it is null
8788 }
8789
8790 //------------------------------get_state_from_digest_object-----------------------
8791 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8792 const char* state_type;
8793 switch (elem_type) {
8794 case T_BYTE: state_type = "[B"; break;
8795 case T_INT: state_type = "[I"; break;
8796 case T_LONG: state_type = "[J"; break;
8797 default: ShouldNotReachHere();
8798 }
8799 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8800 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8801 if (digest_state == nullptr) return (Node *) nullptr;
8802
8803 // now have the array, need to get the start address of the state array
8804 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8805 return state;
8806 }
8807
8808 //------------------------------get_block_size_from_sha3_object----------------------------------
8809 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8810 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8811 assert (block_size != nullptr, "sanity");
8812 return block_size;
8813 }
8814
8815 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8816 // Return node representing slow path of predicate check.
8817 // the pseudo code we want to emulate with this predicate is:
8818 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8819 //
8820 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8821 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8822 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8823 assert((uint)predicate < 5, "sanity");
8824
8825 // The receiver was checked for null already.
8826 Node* digestBaseObj = argument(0);
8827
8828 // get DigestBase klass for instanceOf check
8829 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8830 assert(tinst != nullptr, "digestBaseObj is null");
8831 assert(tinst->is_loaded(), "DigestBase is not loaded");
8832
8833 const char* klass_name = nullptr;
8834 switch (predicate) {
8835 case 0:
8836 if (UseMD5Intrinsics) {
8837 // we want to do an instanceof comparison against the MD5 class
8838 klass_name = "sun/security/provider/MD5";
8839 }
8840 break;
8841 case 1:
8842 if (UseSHA1Intrinsics) {
8843 // we want to do an instanceof comparison against the SHA class
8844 klass_name = "sun/security/provider/SHA";
8845 }
8846 break;
8847 case 2:
8848 if (UseSHA256Intrinsics) {
8849 // we want to do an instanceof comparison against the SHA2 class
8850 klass_name = "sun/security/provider/SHA2";
8851 }
8852 break;
8853 case 3:
8854 if (UseSHA512Intrinsics) {
8855 // we want to do an instanceof comparison against the SHA5 class
8856 klass_name = "sun/security/provider/SHA5";
8857 }
8858 break;
8859 case 4:
8860 if (UseSHA3Intrinsics) {
8861 // we want to do an instanceof comparison against the SHA3 class
8862 klass_name = "sun/security/provider/SHA3";
8863 }
8864 break;
8865 default:
8866 fatal("unknown SHA intrinsic predicate: %d", predicate);
8867 }
8868
8869 ciKlass* klass = nullptr;
8870 if (klass_name != nullptr) {
8871 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8872 }
8873 if ((klass == nullptr) || !klass->is_loaded()) {
8874 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8875 Node* ctrl = control();
8876 set_control(top()); // no intrinsic path
8877 return ctrl;
8878 }
8879 ciInstanceKlass* instklass = klass->as_instance_klass();
8880
8881 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8882 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8883 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8884 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8885
8886 return instof_false; // even if it is null
8887 }
8888
8889 //-------------inline_fma-----------------------------------
8890 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8891 Node *a = nullptr;
8892 Node *b = nullptr;
8893 Node *c = nullptr;
8894 Node* result = nullptr;
8895 switch (id) {
8896 case vmIntrinsics::_fmaD:
8897 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8898 // no receiver since it is static method
8899 a = argument(0);
8900 b = argument(2);
8901 c = argument(4);
8902 result = _gvn.transform(new FmaDNode(a, b, c));
8903 break;
8904 case vmIntrinsics::_fmaF:
8905 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8906 a = argument(0);
8907 b = argument(1);
8908 c = argument(2);
8909 result = _gvn.transform(new FmaFNode(a, b, c));
8910 break;
8911 default:
8912 fatal_unexpected_iid(id); break;
8913 }
8914 set_result(result);
8915 return true;
8916 }
8917
8918 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8919 // argument(0) is receiver
8920 Node* codePoint = argument(1);
8921 Node* n = nullptr;
8922
8923 switch (id) {
8924 case vmIntrinsics::_isDigit :
8925 n = new DigitNode(control(), codePoint);
8926 break;
8927 case vmIntrinsics::_isLowerCase :
8928 n = new LowerCaseNode(control(), codePoint);
8929 break;
8930 case vmIntrinsics::_isUpperCase :
8931 n = new UpperCaseNode(control(), codePoint);
8932 break;
8933 case vmIntrinsics::_isWhitespace :
8934 n = new WhitespaceNode(control(), codePoint);
8935 break;
8936 default:
8937 fatal_unexpected_iid(id);
8938 }
8939
8940 set_result(_gvn.transform(n));
8941 return true;
8942 }
8943
8944 bool LibraryCallKit::inline_profileBoolean() {
8945 Node* counts = argument(1);
8946 const TypeAryPtr* ary = nullptr;
8947 ciArray* aobj = nullptr;
8948 if (counts->is_Con()
8949 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8950 && (aobj = ary->const_oop()->as_array()) != nullptr
8951 && (aobj->length() == 2)) {
8952 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8953 jint false_cnt = aobj->element_value(0).as_int();
8954 jint true_cnt = aobj->element_value(1).as_int();
8955
8956 if (C->log() != nullptr) {
8957 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8958 false_cnt, true_cnt);
8959 }
8960
8961 if (false_cnt + true_cnt == 0) {
8962 // According to profile, never executed.
8963 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8964 Deoptimization::Action_reinterpret);
8965 return true;
8966 }
8967
8968 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8969 // is a number of each value occurrences.
8970 Node* result = argument(0);
8971 if (false_cnt == 0 || true_cnt == 0) {
8972 // According to profile, one value has been never seen.
8973 int expected_val = (false_cnt == 0) ? 1 : 0;
8974
8975 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8976 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8977
8978 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8979 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8980 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8981
8982 { // Slow path: uncommon trap for never seen value and then reexecute
8983 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8984 // the value has been seen at least once.
8985 PreserveJVMState pjvms(this);
8986 PreserveReexecuteState preexecs(this);
8987 jvms()->set_should_reexecute(true);
8988
8989 set_control(slow_path);
8990 set_i_o(i_o());
8991
8992 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8993 Deoptimization::Action_reinterpret);
8994 }
8995 // The guard for never seen value enables sharpening of the result and
8996 // returning a constant. It allows to eliminate branches on the same value
8997 // later on.
8998 set_control(fast_path);
8999 result = intcon(expected_val);
9000 }
9001 // Stop profiling.
9002 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
9003 // By replacing method body with profile data (represented as ProfileBooleanNode
9004 // on IR level) we effectively disable profiling.
9005 // It enables full speed execution once optimized code is generated.
9006 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
9007 C->record_for_igvn(profile);
9008 set_result(profile);
9009 return true;
9010 } else {
9011 // Continue profiling.
9012 // Profile data isn't available at the moment. So, execute method's bytecode version.
9013 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
9014 // is compiled and counters aren't available since corresponding MethodHandle
9015 // isn't a compile-time constant.
9016 return false;
9017 }
9018 }
9019
9020 bool LibraryCallKit::inline_isCompileConstant() {
9021 Node* n = argument(0);
9022 set_result(n->is_Con() ? intcon(1) : intcon(0));
9023 return true;
9024 }
9025
9026 //------------------------------- inline_getObjectSize --------------------------------------
9027 //
9028 // Calculate the runtime size of the object/array.
9029 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
9030 //
9031 bool LibraryCallKit::inline_getObjectSize() {
9032 Node* obj = argument(3);
9033 Node* klass_node = load_object_klass(obj);
9034
9035 jint layout_con = Klass::_lh_neutral_value;
9036 Node* layout_val = get_layout_helper(klass_node, layout_con);
9037 int layout_is_con = (layout_val == nullptr);
9038
9039 if (layout_is_con) {
9040 // Layout helper is constant, can figure out things at compile time.
9041
9042 if (Klass::layout_helper_is_instance(layout_con)) {
9043 // Instance case: layout_con contains the size itself.
9044 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
9045 set_result(size);
9046 } else {
9047 // Array case: size is round(header + element_size*arraylength).
9048 // Since arraylength is different for every array instance, we have to
9049 // compute the whole thing at runtime.
9050
9051 Node* arr_length = load_array_length(obj);
9052
9053 int round_mask = MinObjAlignmentInBytes - 1;
9054 int hsize = Klass::layout_helper_header_size(layout_con);
9055 int eshift = Klass::layout_helper_log2_element_size(layout_con);
9056
9057 if ((round_mask & ~right_n_bits(eshift)) == 0) {
9058 round_mask = 0; // strength-reduce it if it goes away completely
9059 }
9060 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
9061 Node* header_size = intcon(hsize + round_mask);
9062
9063 Node* lengthx = ConvI2X(arr_length);
9064 Node* headerx = ConvI2X(header_size);
9065
9066 Node* abody = lengthx;
9067 if (eshift != 0) {
9068 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
9069 }
9070 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
9071 if (round_mask != 0) {
9072 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
9073 }
9074 size = ConvX2L(size);
9075 set_result(size);
9076 }
9077 } else {
9078 // Layout helper is not constant, need to test for array-ness at runtime.
9079
9080 enum { _instance_path = 1, _array_path, PATH_LIMIT };
9081 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
9082 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
9083 record_for_igvn(result_reg);
9084
9085 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
9086 if (array_ctl != nullptr) {
9087 // Array case: size is round(header + element_size*arraylength).
9088 // Since arraylength is different for every array instance, we have to
9089 // compute the whole thing at runtime.
9090
9091 PreserveJVMState pjvms(this);
9092 set_control(array_ctl);
9093 Node* arr_length = load_array_length(obj);
9094
9095 int round_mask = MinObjAlignmentInBytes - 1;
9096 Node* mask = intcon(round_mask);
9097
9098 Node* hss = intcon(Klass::_lh_header_size_shift);
9099 Node* hsm = intcon(Klass::_lh_header_size_mask);
9100 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
9101 header_size = _gvn.transform(new AndINode(header_size, hsm));
9102 header_size = _gvn.transform(new AddINode(header_size, mask));
9103
9104 // There is no need to mask or shift this value.
9105 // The semantics of LShiftINode include an implicit mask to 0x1F.
9106 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
9107 Node* elem_shift = layout_val;
9108
9109 Node* lengthx = ConvI2X(arr_length);
9110 Node* headerx = ConvI2X(header_size);
9111
9112 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
9113 Node* size = _gvn.transform(new AddXNode(headerx, abody));
9114 if (round_mask != 0) {
9115 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
9116 }
9117 size = ConvX2L(size);
9118
9119 result_reg->init_req(_array_path, control());
9120 result_val->init_req(_array_path, size);
9121 }
9122
9123 if (!stopped()) {
9124 // Instance case: the layout helper gives us instance size almost directly,
9125 // but we need to mask out the _lh_instance_slow_path_bit.
9126 Node* size = ConvI2X(layout_val);
9127 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
9128 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
9129 size = _gvn.transform(new AndXNode(size, mask));
9130 size = ConvX2L(size);
9131
9132 result_reg->init_req(_instance_path, control());
9133 result_val->init_req(_instance_path, size);
9134 }
9135
9136 set_result(result_reg, result_val);
9137 }
9138
9139 return true;
9140 }
9141
9142 //------------------------------- inline_blackhole --------------------------------------
9143 //
9144 // Make sure all arguments to this node are alive.
9145 // This matches methods that were requested to be blackholed through compile commands.
9146 //
9147 bool LibraryCallKit::inline_blackhole() {
9148 assert(callee()->is_static(), "Should have been checked before: only static methods here");
9149 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
9150 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
9151
9152 // Blackhole node pinches only the control, not memory. This allows
9153 // the blackhole to be pinned in the loop that computes blackholed
9154 // values, but have no other side effects, like breaking the optimizations
9155 // across the blackhole.
9156
9157 Node* bh = _gvn.transform(new BlackholeNode(control()));
9158 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
9159
9160 // Bind call arguments as blackhole arguments to keep them alive
9161 uint nargs = callee()->arg_size();
9162 for (uint i = 0; i < nargs; i++) {
9163 bh->add_req(argument(i));
9164 }
9165
9166 return true;
9167 }
9168
9169 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9170 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9171 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9172 return nullptr; // box klass is not Float16
9173 }
9174
9175 // Null check; get notnull casted pointer
9176 Node* null_ctl = top();
9177 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9178 // If not_null_box is dead, only null-path is taken
9179 if (stopped()) {
9180 set_control(null_ctl);
9181 return nullptr;
9182 }
9183 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9184 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9185 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9186 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9187 }
9188
9189 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9190 PreserveReexecuteState preexecs(this);
9191 jvms()->set_should_reexecute(true);
9192
9193 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9194 Node* klass_node = makecon(klass_type);
9195 Node* box = new_instance(klass_node);
9196
9197 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9198 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9199
9200 Node* field_store = _gvn.transform(access_store_at(box,
9201 value_field,
9202 value_adr_type,
9203 value,
9204 TypeInt::SHORT,
9205 T_SHORT,
9206 IN_HEAP));
9207 set_memory(field_store, value_adr_type);
9208 return box;
9209 }
9210
9211 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9212 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9213 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9214 return false;
9215 }
9216
9217 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9218 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9219 return false;
9220 }
9221
9222 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9223 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9224 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9225 ciSymbols::short_signature(),
9226 false);
9227 assert(field != nullptr, "");
9228
9229 // Transformed nodes
9230 Node* fld1 = nullptr;
9231 Node* fld2 = nullptr;
9232 Node* fld3 = nullptr;
9233 switch(num_args) {
9234 case 3:
9235 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9236 if (fld3 == nullptr) {
9237 return false;
9238 }
9239 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9240 // fall-through
9241 case 2:
9242 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9243 if (fld2 == nullptr) {
9244 return false;
9245 }
9246 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9247 // fall-through
9248 case 1:
9249 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9250 if (fld1 == nullptr) {
9251 return false;
9252 }
9253 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9254 break;
9255 default: fatal("Unsupported number of arguments %d", num_args);
9256 }
9257
9258 Node* result = nullptr;
9259 switch (id) {
9260 // Unary operations
9261 case vmIntrinsics::_sqrt_float16:
9262 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9263 break;
9264 // Ternary operations
9265 case vmIntrinsics::_fma_float16:
9266 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9267 break;
9268 default:
9269 fatal_unexpected_iid(id);
9270 break;
9271 }
9272 result = _gvn.transform(new ReinterpretHF2SNode(result));
9273 set_result(box_fp16_value(float16_box_type, field, result));
9274 return true;
9275 }
9276