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 case vmIntrinsics::_tryUpdateEpochField: return inline_native_try_update_epoch();
500 #endif
501 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
502 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
503 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
504 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
505 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
506 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
507 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
508 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
509 case vmIntrinsics::_getLength: return inline_native_getLength();
510 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
511 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
512 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
513 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
514 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
515 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
516 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
517
518 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
519 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
520
521 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
522
523 case vmIntrinsics::_isInstance:
524 case vmIntrinsics::_isHidden:
525 case vmIntrinsics::_getSuperclass: return inline_native_Class_query(intrinsic_id());
526
527 case vmIntrinsics::_floatToRawIntBits:
528 case vmIntrinsics::_floatToIntBits:
529 case vmIntrinsics::_intBitsToFloat:
530 case vmIntrinsics::_doubleToRawLongBits:
531 case vmIntrinsics::_doubleToLongBits:
532 case vmIntrinsics::_longBitsToDouble:
533 case vmIntrinsics::_floatToFloat16:
534 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
535 case vmIntrinsics::_sqrt_float16: return inline_fp16_operations(intrinsic_id(), 1);
536 case vmIntrinsics::_fma_float16: return inline_fp16_operations(intrinsic_id(), 3);
537 case vmIntrinsics::_floatIsFinite:
538 case vmIntrinsics::_floatIsInfinite:
539 case vmIntrinsics::_doubleIsFinite:
540 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
541
542 case vmIntrinsics::_numberOfLeadingZeros_i:
543 case vmIntrinsics::_numberOfLeadingZeros_l:
544 case vmIntrinsics::_numberOfTrailingZeros_i:
545 case vmIntrinsics::_numberOfTrailingZeros_l:
546 case vmIntrinsics::_bitCount_i:
547 case vmIntrinsics::_bitCount_l:
548 case vmIntrinsics::_reverse_i:
549 case vmIntrinsics::_reverse_l:
550 case vmIntrinsics::_reverseBytes_i:
551 case vmIntrinsics::_reverseBytes_l:
552 case vmIntrinsics::_reverseBytes_s:
553 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
554
555 case vmIntrinsics::_compress_i:
556 case vmIntrinsics::_compress_l:
557 case vmIntrinsics::_expand_i:
558 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
559
560 case vmIntrinsics::_compareUnsigned_i:
561 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
562
563 case vmIntrinsics::_divideUnsigned_i:
564 case vmIntrinsics::_divideUnsigned_l:
565 case vmIntrinsics::_remainderUnsigned_i:
566 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
567
568 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
569
570 case vmIntrinsics::_Reference_get0: return inline_reference_get0();
571 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
572 case vmIntrinsics::_Reference_reachabilityFence: return inline_reference_reachabilityFence();
573 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
574 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
575 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
576
577 case vmIntrinsics::_Class_cast: return inline_Class_cast();
578
579 case vmIntrinsics::_aescrypt_encryptBlock:
580 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
581
582 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
583 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
584 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
585
586 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
587 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
588 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
589
590 case vmIntrinsics::_counterMode_AESCrypt:
591 return inline_counterMode_AESCrypt(intrinsic_id());
592
593 case vmIntrinsics::_galoisCounterMode_AESCrypt:
594 return inline_galoisCounterMode_AESCrypt();
595
596 case vmIntrinsics::_md5_implCompress:
597 case vmIntrinsics::_sha_implCompress:
598 case vmIntrinsics::_sha2_implCompress:
599 case vmIntrinsics::_sha5_implCompress:
600 case vmIntrinsics::_sha3_implCompress:
601 return inline_digestBase_implCompress(intrinsic_id());
602 case vmIntrinsics::_double_keccak:
603 case vmIntrinsics::_quad_keccak:
604 return inline_keccak(intrinsic_id());
605
606 case vmIntrinsics::_digestBase_implCompressMB:
607 return inline_digestBase_implCompressMB(predicate);
608
609 case vmIntrinsics::_multiplyToLen:
610 return inline_multiplyToLen();
611
612 case vmIntrinsics::_squareToLen:
613 return inline_squareToLen();
614
615 case vmIntrinsics::_mulAdd:
616 return inline_mulAdd();
617
618 case vmIntrinsics::_montgomeryMultiply:
619 return inline_montgomeryMultiply();
620 case vmIntrinsics::_montgomerySquare:
621 return inline_montgomerySquare();
622
623 case vmIntrinsics::_bigIntegerRightShiftWorker:
624 return inline_bigIntegerShift(true);
625 case vmIntrinsics::_bigIntegerLeftShiftWorker:
626 return inline_bigIntegerShift(false);
627
628 case vmIntrinsics::_vectorizedMismatch:
629 return inline_vectorizedMismatch();
630
631 case vmIntrinsics::_ghash_processBlocks:
632 return inline_ghash_processBlocks();
633 case vmIntrinsics::_chacha20Block:
634 return inline_chacha20Block();
635 case vmIntrinsics::_kyberNtt:
636 return inline_kyberNtt();
637 case vmIntrinsics::_kyberInverseNtt:
638 return inline_kyberInverseNtt();
639 case vmIntrinsics::_kyberNttMult:
640 return inline_kyberNttMult();
641 case vmIntrinsics::_kyberAddPoly_2:
642 return inline_kyberAddPoly_2();
643 case vmIntrinsics::_kyberAddPoly_3:
644 return inline_kyberAddPoly_3();
645 case vmIntrinsics::_kyber12To16:
646 return inline_kyber12To16();
647 case vmIntrinsics::_kyberBarrettReduce:
648 return inline_kyberBarrettReduce();
649 case vmIntrinsics::_dilithiumAlmostNtt:
650 return inline_dilithiumAlmostNtt();
651 case vmIntrinsics::_dilithiumAlmostInverseNtt:
652 return inline_dilithiumAlmostInverseNtt();
653 case vmIntrinsics::_dilithiumNttMult:
654 return inline_dilithiumNttMult();
655 case vmIntrinsics::_dilithiumMontMulByConstant:
656 return inline_dilithiumMontMulByConstant();
657 case vmIntrinsics::_dilithiumDecomposePoly:
658 return inline_dilithiumDecomposePoly();
659 case vmIntrinsics::_base64_encodeBlock:
660 return inline_base64_encodeBlock();
661 case vmIntrinsics::_base64_decodeBlock:
662 return inline_base64_decodeBlock();
663 case vmIntrinsics::_poly1305_processBlocks:
664 return inline_poly1305_processBlocks();
665 case vmIntrinsics::_intpoly_montgomeryMult_P256:
666 return inline_intpoly_montgomeryMult_P256();
667 case vmIntrinsics::_intpoly_assign:
668 return inline_intpoly_assign();
669 case vmIntrinsics::_encodeISOArray:
670 case vmIntrinsics::_encodeByteISOArray:
671 return inline_encodeISOArray(false);
672 case vmIntrinsics::_encodeAsciiArray:
673 return inline_encodeISOArray(true);
674
675 case vmIntrinsics::_updateCRC32:
676 return inline_updateCRC32();
677 case vmIntrinsics::_updateBytesCRC32:
678 return inline_updateBytesCRC32();
679 case vmIntrinsics::_updateByteBufferCRC32:
680 return inline_updateByteBufferCRC32();
681
682 case vmIntrinsics::_updateBytesCRC32C:
683 return inline_updateBytesCRC32C();
684 case vmIntrinsics::_updateDirectByteBufferCRC32C:
685 return inline_updateDirectByteBufferCRC32C();
686
687 case vmIntrinsics::_updateBytesAdler32:
688 return inline_updateBytesAdler32();
689 case vmIntrinsics::_updateByteBufferAdler32:
690 return inline_updateByteBufferAdler32();
691
692 case vmIntrinsics::_profileBoolean:
693 return inline_profileBoolean();
694 case vmIntrinsics::_isCompileConstant:
695 return inline_isCompileConstant();
696
697 case vmIntrinsics::_countPositives:
698 return inline_countPositives();
699
700 case vmIntrinsics::_fmaD:
701 case vmIntrinsics::_fmaF:
702 return inline_fma(intrinsic_id());
703
704 case vmIntrinsics::_isDigit:
705 case vmIntrinsics::_isLowerCase:
706 case vmIntrinsics::_isUpperCase:
707 case vmIntrinsics::_isWhitespace:
708 return inline_character_compare(intrinsic_id());
709
710 case vmIntrinsics::_min:
711 case vmIntrinsics::_max:
712 case vmIntrinsics::_min_strict:
713 case vmIntrinsics::_max_strict:
714 case vmIntrinsics::_minL:
715 case vmIntrinsics::_maxL:
716 case vmIntrinsics::_minF:
717 case vmIntrinsics::_maxF:
718 case vmIntrinsics::_minD:
719 case vmIntrinsics::_maxD:
720 case vmIntrinsics::_minF_strict:
721 case vmIntrinsics::_maxF_strict:
722 case vmIntrinsics::_minD_strict:
723 case vmIntrinsics::_maxD_strict:
724 return inline_min_max(intrinsic_id());
725
726 case vmIntrinsics::_VectorUnaryOp:
727 return inline_vector_nary_operation(1);
728 case vmIntrinsics::_VectorBinaryOp:
729 return inline_vector_nary_operation(2);
730 case vmIntrinsics::_VectorUnaryLibOp:
731 return inline_vector_call(1);
732 case vmIntrinsics::_VectorBinaryLibOp:
733 return inline_vector_call(2);
734 case vmIntrinsics::_VectorTernaryOp:
735 return inline_vector_nary_operation(3);
736 case vmIntrinsics::_VectorFromBitsCoerced:
737 return inline_vector_frombits_coerced();
738 case vmIntrinsics::_VectorMaskOp:
739 return inline_vector_mask_operation();
740 case vmIntrinsics::_VectorLoadOp:
741 return inline_vector_mem_operation(/*is_store=*/false);
742 case vmIntrinsics::_VectorLoadMaskedOp:
743 return inline_vector_mem_masked_operation(/*is_store*/false);
744 case vmIntrinsics::_VectorStoreOp:
745 return inline_vector_mem_operation(/*is_store=*/true);
746 case vmIntrinsics::_VectorStoreMaskedOp:
747 return inline_vector_mem_masked_operation(/*is_store=*/true);
748 case vmIntrinsics::_VectorGatherOp:
749 return inline_vector_gather_scatter(/*is_scatter*/ false);
750 case vmIntrinsics::_VectorScatterOp:
751 return inline_vector_gather_scatter(/*is_scatter*/ true);
752 case vmIntrinsics::_VectorReductionCoerced:
753 return inline_vector_reduction();
754 case vmIntrinsics::_VectorTest:
755 return inline_vector_test();
756 case vmIntrinsics::_VectorBlend:
757 return inline_vector_blend();
758 case vmIntrinsics::_VectorRearrange:
759 return inline_vector_rearrange();
760 case vmIntrinsics::_VectorSelectFrom:
761 return inline_vector_select_from();
762 case vmIntrinsics::_VectorCompare:
763 return inline_vector_compare();
764 case vmIntrinsics::_VectorBroadcastInt:
765 return inline_vector_broadcast_int();
766 case vmIntrinsics::_VectorConvert:
767 return inline_vector_convert();
768 case vmIntrinsics::_VectorInsert:
769 return inline_vector_insert();
770 case vmIntrinsics::_VectorExtract:
771 return inline_vector_extract();
772 case vmIntrinsics::_VectorCompressExpand:
773 return inline_vector_compress_expand();
774 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
775 return inline_vector_select_from_two_vectors();
776 case vmIntrinsics::_IndexVector:
777 return inline_index_vector();
778 case vmIntrinsics::_IndexPartiallyInUpperRange:
779 return inline_index_partially_in_upper_range();
780
781 case vmIntrinsics::_getObjectSize:
782 return inline_getObjectSize();
783
784 case vmIntrinsics::_blackhole:
785 return inline_blackhole();
786
787 default:
788 // If you get here, it may be that someone has added a new intrinsic
789 // to the list in vmIntrinsics.hpp without implementing it here.
790 #ifndef PRODUCT
791 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
792 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
793 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
794 }
795 #endif
796 return false;
797 }
798 }
799
800 Node* LibraryCallKit::try_to_predicate(int predicate) {
801 if (!jvms()->has_method()) {
802 // Root JVMState has a null method.
803 assert(map()->memory()->Opcode() == Op_Parm, "");
804 // Insert the memory aliasing node
805 set_all_memory(reset_memory());
806 }
807 assert(merged_memory(), "");
808
809 switch (intrinsic_id()) {
810 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
811 return inline_cipherBlockChaining_AESCrypt_predicate(false);
812 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
813 return inline_cipherBlockChaining_AESCrypt_predicate(true);
814 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
815 return inline_electronicCodeBook_AESCrypt_predicate(false);
816 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
817 return inline_electronicCodeBook_AESCrypt_predicate(true);
818 case vmIntrinsics::_counterMode_AESCrypt:
819 return inline_counterMode_AESCrypt_predicate();
820 case vmIntrinsics::_digestBase_implCompressMB:
821 return inline_digestBase_implCompressMB_predicate(predicate);
822 case vmIntrinsics::_galoisCounterMode_AESCrypt:
823 return inline_galoisCounterMode_AESCrypt_predicate();
824
825 default:
826 // If you get here, it may be that someone has added a new intrinsic
827 // to the list in vmIntrinsics.hpp without implementing it here.
828 #ifndef PRODUCT
829 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
830 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
831 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
832 }
833 #endif
834 Node* slow_ctl = control();
835 set_control(top()); // No fast path intrinsic
836 return slow_ctl;
837 }
838 }
839
840 //------------------------------set_result-------------------------------
841 // Helper function for finishing intrinsics.
842 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
843 record_for_igvn(region);
844 set_control(_gvn.transform(region));
845 set_result( _gvn.transform(value));
846 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
847 }
848
849 RegionNode* LibraryCallKit::create_bailout() {
850 RegionNode* bailout = new RegionNode(1);
851 record_for_igvn(bailout);
852 return bailout;
853 }
854
855 bool LibraryCallKit::check_bailout(RegionNode* bailout) {
856 if (bailout->req() > 1) {
857 bailout = _gvn.transform(bailout)->as_Region();
858 Node* frame = _gvn.transform(new ParmNode(C->start(), TypeFunc::FramePtr));
859 Node* halt = _gvn.transform(new HaltNode(bailout, frame, "unexpected guard failure in intrinsic"));
860 C->root()->add_req(halt);
861 }
862 return stopped();
863 }
864
865 //------------------------------generate_guard---------------------------
866 // Helper function for generating guarded fast-slow graph structures.
867 // The given 'test', if true, guards a slow path. If the test fails
868 // then a fast path can be taken. (We generally hope it fails.)
869 // In all cases, GraphKit::control() is updated to the fast path.
870 // The returned value represents the control for the slow path.
871 // The return value is never 'top'; it is either a valid control
872 // or null if it is obvious that the slow path can never be taken.
873 // Also, if region and the slow control are not null, the slow edge
874 // is appended to the region.
875 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
876 if (stopped()) {
877 // Already short circuited.
878 return nullptr;
879 }
880
881 // Build an if node and its projections.
882 // If test is true we take the slow path, which we assume is uncommon.
883 if (_gvn.type(test) == TypeInt::ZERO) {
884 // The slow branch is never taken. No need to build this guard.
885 return nullptr;
886 }
887
888 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
889
890 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
891 if (if_slow == top()) {
892 // The slow branch is never taken. No need to build this guard.
893 return nullptr;
894 }
895
896 if (region != nullptr)
897 region->add_req(if_slow);
898
899 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
900 set_control(if_fast);
901
902 return if_slow;
903 }
904
905 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
906 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
907 }
908 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
909 return generate_guard(test, region, PROB_FAIR);
910 }
911
912 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
913 Node** pos_index, bool with_opaque) {
914 if (stopped())
915 return nullptr; // already stopped
916 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
917 return nullptr; // index is already adequately typed
918 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
919 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
920 if (with_opaque) {
921 bol_lt = _gvn.transform(new OpaqueConstantBoolNode(C, bol_lt, false));
922 }
923 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
924 if (is_neg != nullptr && pos_index != nullptr) {
925 // Emulate effect of Parse::adjust_map_after_if.
926 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
927 (*pos_index) = _gvn.transform(ccast);
928 }
929 return is_neg;
930 }
931
932 // Make sure that 'position' is a valid limit index, in [0..length].
933 // There are two equivalent plans for checking this:
934 // A. (offset + copyLength) unsigned<= arrayLength
935 // B. offset <= (arrayLength - copyLength)
936 // We require that all of the values above, except for the sum and
937 // difference, are already known to be non-negative.
938 // Plan A is robust in the face of overflow, if offset and copyLength
939 // are both hugely positive.
940 //
941 // Plan B is less direct and intuitive, but it does not overflow at
942 // all, since the difference of two non-negatives is always
943 // representable. Whenever Java methods must perform the equivalent
944 // check they generally use Plan B instead of Plan A.
945 // For the moment we use Plan A.
946 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
947 Node* subseq_length,
948 Node* array_length,
949 RegionNode* region,
950 bool with_opaque) {
951 if (stopped())
952 return nullptr; // already stopped
953 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
954 if (zero_offset && subseq_length->eqv_uncast(array_length))
955 return nullptr; // common case of whole-array copy
956 Node* last = subseq_length;
957 if (!zero_offset) // last += offset
958 last = _gvn.transform(new AddINode(last, offset));
959 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
960 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
961 if (with_opaque) {
962 bol_lt = _gvn.transform(new OpaqueConstantBoolNode(C, bol_lt, false));
963 }
964 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
965 return is_over;
966 }
967
968 // Emit range checks for the given String.value byte array
969 void LibraryCallKit::generate_string_range_check(Node* array,
970 Node* offset,
971 Node* count,
972 bool char_count,
973 RegionNode* region) {
974 if (stopped()) {
975 return; // already stopped
976 }
977 if (char_count) {
978 // Convert char count to byte count
979 count = _gvn.transform(new LShiftINode(count, intcon(1)));
980 }
981 // Offset and count must not be negative
982 generate_negative_guard(offset, region, nullptr, true);
983 generate_negative_guard(count, region, nullptr, true);
984 // Offset + count must not exceed length of array
985 generate_limit_guard(offset, count, load_array_length(array), region, true);
986 }
987
988 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
989 bool is_immutable) {
990 ciKlass* thread_klass = env()->Thread_klass();
991 const Type* thread_type
992 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
993
994 Node* thread = _gvn.transform(new ThreadLocalNode());
995 Node* p = off_heap_plus_addr(thread, in_bytes(handle_offset));
996 tls_output = thread;
997
998 Node* thread_obj_handle
999 = (is_immutable
1000 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
1001 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
1002 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
1003 thread_obj_handle = _gvn.transform(thread_obj_handle);
1004
1005 DecoratorSet decorators = IN_NATIVE;
1006 if (is_immutable) {
1007 decorators |= C2_IMMUTABLE_MEMORY;
1008 }
1009 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
1010 }
1011
1012 //--------------------------generate_current_thread--------------------
1013 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1014 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
1015 /*is_immutable*/false);
1016 }
1017
1018 //--------------------------generate_virtual_thread--------------------
1019 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
1020 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
1021 !C->method()->changes_current_thread());
1022 }
1023
1024 //------------------------------make_string_method_node------------------------
1025 // Helper method for String intrinsic functions. This version is called with
1026 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1027 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1028 // containing the lengths of str1 and str2.
1029 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1030 Node* result = nullptr;
1031 switch (opcode) {
1032 case Op_StrIndexOf:
1033 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1034 str1_start, cnt1, str2_start, cnt2, ae);
1035 break;
1036 case Op_StrComp:
1037 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1038 str1_start, cnt1, str2_start, cnt2, ae);
1039 break;
1040 case Op_StrEquals:
1041 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1042 // Use the constant length if there is one because optimized match rule may exist.
1043 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1044 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1045 break;
1046 default:
1047 ShouldNotReachHere();
1048 return nullptr;
1049 }
1050
1051 // All these intrinsics have checks.
1052 C->set_has_split_ifs(true); // Has chance for split-if optimization
1053 clear_upper_avx();
1054
1055 return _gvn.transform(result);
1056 }
1057
1058 //------------------------------inline_string_compareTo------------------------
1059 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1060 Node* arg1 = argument(0);
1061 Node* arg2 = argument(1);
1062
1063 arg1 = must_be_not_null(arg1, true);
1064 arg2 = must_be_not_null(arg2, true);
1065
1066 // Get start addr and length of first argument
1067 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1068 Node* arg1_cnt = load_array_length(arg1);
1069
1070 // Get start addr and length of second argument
1071 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1072 Node* arg2_cnt = load_array_length(arg2);
1073
1074 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1075 set_result(result);
1076 return true;
1077 }
1078
1079 //------------------------------inline_string_equals------------------------
1080 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1081 Node* arg1 = argument(0);
1082 Node* arg2 = argument(1);
1083
1084 // paths (plus control) merge
1085 RegionNode* region = new RegionNode(3);
1086 Node* phi = new PhiNode(region, TypeInt::BOOL);
1087
1088 if (!stopped()) {
1089
1090 arg1 = must_be_not_null(arg1, true);
1091 arg2 = must_be_not_null(arg2, true);
1092
1093 // Get start addr and length of first argument
1094 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1095 Node* arg1_cnt = load_array_length(arg1);
1096
1097 // Get start addr and length of second argument
1098 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1099 Node* arg2_cnt = load_array_length(arg2);
1100
1101 // Check for arg1_cnt != arg2_cnt
1102 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1103 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1104 Node* if_ne = generate_slow_guard(bol, nullptr);
1105 if (if_ne != nullptr) {
1106 phi->init_req(2, intcon(0));
1107 region->init_req(2, if_ne);
1108 }
1109
1110 // Check for count == 0 is done by assembler code for StrEquals.
1111
1112 if (!stopped()) {
1113 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1114 phi->init_req(1, equals);
1115 region->init_req(1, control());
1116 }
1117 }
1118
1119 // post merge
1120 set_control(_gvn.transform(region));
1121 record_for_igvn(region);
1122
1123 set_result(_gvn.transform(phi));
1124 return true;
1125 }
1126
1127 //------------------------------inline_array_equals----------------------------
1128 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1129 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1130 Node* arg1 = argument(0);
1131 Node* arg2 = argument(1);
1132
1133 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1134 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), mtype, arg1, arg2, ae)));
1135 clear_upper_avx();
1136
1137 return true;
1138 }
1139
1140
1141 //------------------------------inline_countPositives------------------------------
1142 // int java.lang.StringCoding#countPositives0(byte[] ba, int off, int len)
1143 bool LibraryCallKit::inline_countPositives() {
1144 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1145 // no receiver since it is static method
1146 Node* ba = argument(0);
1147 Node* offset = argument(1);
1148 Node* len = argument(2);
1149
1150 ba = must_be_not_null(ba, true);
1151 RegionNode* bailout = create_bailout();
1152 generate_string_range_check(ba, offset, len, false, bailout);
1153 if (check_bailout(bailout)) {
1154 return true;
1155 }
1156
1157 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1158 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1159 set_result(_gvn.transform(result));
1160 clear_upper_avx();
1161 return true;
1162 }
1163
1164 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1165 Node* index = argument(0);
1166 Node* length = bt == T_INT ? argument(1) : argument(2);
1167 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1168 return false;
1169 }
1170
1171 // check that length is positive
1172 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1173 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1174
1175 {
1176 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1177 uncommon_trap(Deoptimization::Reason_intrinsic,
1178 Deoptimization::Action_make_not_entrant);
1179 }
1180
1181 if (stopped()) {
1182 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1183 return true;
1184 }
1185
1186 // length is now known positive, add a cast node to make this explicit
1187 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1188 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1189 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1190 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1191 casted_length = _gvn.transform(casted_length);
1192 replace_in_map(length, casted_length);
1193 length = casted_length;
1194
1195 // Use an unsigned comparison for the range check itself
1196 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1197 BoolTest::mask btest = BoolTest::lt;
1198 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1199 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1200 _gvn.set_type(rc, rc->Value(&_gvn));
1201 if (!rc_bool->is_Con()) {
1202 record_for_igvn(rc);
1203 }
1204 set_control(_gvn.transform(new IfTrueNode(rc)));
1205 {
1206 PreserveJVMState pjvms(this);
1207 set_control(_gvn.transform(new IfFalseNode(rc)));
1208 uncommon_trap(Deoptimization::Reason_range_check,
1209 Deoptimization::Action_make_not_entrant);
1210 }
1211
1212 if (stopped()) {
1213 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1214 return true;
1215 }
1216
1217 // index is now known to be >= 0 and < length, cast it
1218 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1219 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1220 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1221 result = _gvn.transform(result);
1222 set_result(result);
1223 replace_in_map(index, result);
1224 return true;
1225 }
1226
1227 //------------------------------inline_string_indexOf------------------------
1228 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1229 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1230 return false;
1231 }
1232 Node* src = argument(0);
1233 Node* tgt = argument(1);
1234
1235 // Make the merge point
1236 RegionNode* result_rgn = new RegionNode(4);
1237 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1238
1239 src = must_be_not_null(src, true);
1240 tgt = must_be_not_null(tgt, true);
1241
1242 // Get start addr and length of source string
1243 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1244 Node* src_count = load_array_length(src);
1245
1246 // Get start addr and length of substring
1247 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1248 Node* tgt_count = load_array_length(tgt);
1249
1250 Node* result = nullptr;
1251 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1252
1253 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1254 // Divide src size by 2 if String is UTF16 encoded
1255 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1256 }
1257 if (ae == StrIntrinsicNode::UU) {
1258 // Divide substring size by 2 if String is UTF16 encoded
1259 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1260 }
1261
1262 if (call_opt_stub) {
1263 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1264 StubRoutines::_string_indexof_array[ae],
1265 "stringIndexOf", TypePtr::BOTTOM, src_start,
1266 src_count, tgt_start, tgt_count);
1267 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1268 } else {
1269 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1270 result_rgn, result_phi, ae);
1271 }
1272 if (result != nullptr) {
1273 result_phi->init_req(3, result);
1274 result_rgn->init_req(3, control());
1275 }
1276 set_control(_gvn.transform(result_rgn));
1277 record_for_igvn(result_rgn);
1278 set_result(_gvn.transform(result_phi));
1279
1280 return true;
1281 }
1282
1283 //-----------------------------inline_string_indexOfI-----------------------
1284 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1285 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1286 return false;
1287 }
1288
1289 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1290 Node* src = argument(0); // byte[]
1291 Node* src_count = argument(1); // char count
1292 Node* tgt = argument(2); // byte[]
1293 Node* tgt_count = argument(3); // char count
1294 Node* from_index = argument(4); // char index
1295
1296 src = must_be_not_null(src, true);
1297 tgt = must_be_not_null(tgt, true);
1298
1299 // Multiply byte array index by 2 if String is UTF16 encoded
1300 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1301 src_count = _gvn.transform(new SubINode(src_count, from_index));
1302 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1303 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1304
1305 // Range checks
1306 RegionNode* bailout = create_bailout();
1307 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL, bailout);
1308 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU, bailout);
1309 if (check_bailout(bailout)) {
1310 return true;
1311 }
1312
1313 RegionNode* region = new RegionNode(5);
1314 Node* phi = new PhiNode(region, TypeInt::INT);
1315 Node* result = nullptr;
1316
1317 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1318
1319 if (call_opt_stub) {
1320 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1321 StubRoutines::_string_indexof_array[ae],
1322 "stringIndexOf", TypePtr::BOTTOM, src_start,
1323 src_count, tgt_start, tgt_count);
1324 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1325 } else {
1326 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1327 region, phi, ae);
1328 }
1329 if (result != nullptr) {
1330 // The result is index relative to from_index if substring was found, -1 otherwise.
1331 // Generate code which will fold into cmove.
1332 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1333 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1334
1335 Node* if_lt = generate_slow_guard(bol, nullptr);
1336 if (if_lt != nullptr) {
1337 // result == -1
1338 phi->init_req(3, result);
1339 region->init_req(3, if_lt);
1340 }
1341 if (!stopped()) {
1342 result = _gvn.transform(new AddINode(result, from_index));
1343 phi->init_req(4, result);
1344 region->init_req(4, control());
1345 }
1346 }
1347
1348 set_control(_gvn.transform(region));
1349 record_for_igvn(region);
1350 set_result(_gvn.transform(phi));
1351 clear_upper_avx();
1352
1353 return true;
1354 }
1355
1356 // Create StrIndexOfNode with fast path checks
1357 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1358 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1359 // Check for substr count > string count
1360 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1361 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1362 Node* if_gt = generate_slow_guard(bol, nullptr);
1363 if (if_gt != nullptr) {
1364 phi->init_req(1, intcon(-1));
1365 region->init_req(1, if_gt);
1366 }
1367 if (!stopped()) {
1368 // Check for substr count == 0
1369 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1370 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1371 Node* if_zero = generate_slow_guard(bol, nullptr);
1372 if (if_zero != nullptr) {
1373 phi->init_req(2, intcon(0));
1374 region->init_req(2, if_zero);
1375 }
1376 }
1377 if (!stopped()) {
1378 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1379 }
1380 return nullptr;
1381 }
1382
1383 //-----------------------------inline_string_indexOfChar-----------------------
1384 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1385 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1386 return false;
1387 }
1388 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1389 return false;
1390 }
1391 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1392 Node* src = argument(0); // byte[]
1393 Node* int_ch = argument(1);
1394 Node* from_index = argument(2);
1395 Node* max = argument(3);
1396
1397 src = must_be_not_null(src, true);
1398
1399 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1400 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1401 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1402
1403 // Range checks
1404 RegionNode* bailout = create_bailout();
1405 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U, bailout);
1406 if (check_bailout(bailout)) {
1407 return true;
1408 }
1409
1410 // Check for int_ch >= 0
1411 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1412 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1413 {
1414 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1415 uncommon_trap(Deoptimization::Reason_intrinsic,
1416 Deoptimization::Action_maybe_recompile);
1417 }
1418 if (stopped()) {
1419 return true;
1420 }
1421
1422 RegionNode* region = new RegionNode(3);
1423 Node* phi = new PhiNode(region, TypeInt::INT);
1424
1425 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1426 C->set_has_split_ifs(true); // Has chance for split-if optimization
1427 _gvn.transform(result);
1428
1429 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1430 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1431
1432 Node* if_lt = generate_slow_guard(bol, nullptr);
1433 if (if_lt != nullptr) {
1434 // result == -1
1435 phi->init_req(2, result);
1436 region->init_req(2, if_lt);
1437 }
1438 if (!stopped()) {
1439 result = _gvn.transform(new AddINode(result, from_index));
1440 phi->init_req(1, result);
1441 region->init_req(1, control());
1442 }
1443 set_control(_gvn.transform(region));
1444 record_for_igvn(region);
1445 set_result(_gvn.transform(phi));
1446 clear_upper_avx();
1447
1448 return true;
1449 }
1450 //---------------------------inline_string_copy---------------------
1451 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1452 // int StringUTF16.compress0(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1453 // int StringUTF16.compress0(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1454 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1455 // void StringLatin1.inflate0(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1456 // void StringLatin1.inflate0(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1457 bool LibraryCallKit::inline_string_copy(bool compress) {
1458 int nargs = 5; // 2 oops, 3 ints
1459 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1460
1461 Node* src = argument(0);
1462 Node* src_offset = argument(1);
1463 Node* dst = argument(2);
1464 Node* dst_offset = argument(3);
1465 Node* length = argument(4);
1466
1467 // Check for allocation before we add nodes that would confuse
1468 // tightly_coupled_allocation()
1469 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1470
1471 // Figure out the size and type of the elements we will be copying.
1472 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1473 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1474 if (src_type == nullptr || dst_type == nullptr) {
1475 return false;
1476 }
1477 BasicType src_elem = src_type->elem()->array_element_basic_type();
1478 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1479 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1480 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1481 "Unsupported array types for inline_string_copy");
1482
1483 src = must_be_not_null(src, true);
1484 dst = must_be_not_null(dst, true);
1485
1486 // Convert char[] offsets to byte[] offsets
1487 bool convert_src = (compress && src_elem == T_BYTE);
1488 bool convert_dst = (!compress && dst_elem == T_BYTE);
1489 if (convert_src) {
1490 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1491 } else if (convert_dst) {
1492 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1493 }
1494
1495 // Range checks
1496 RegionNode* bailout = create_bailout();
1497 generate_string_range_check(src, src_offset, length, convert_src, bailout);
1498 generate_string_range_check(dst, dst_offset, length, convert_dst, bailout);
1499 if (check_bailout(bailout)) {
1500 return true;
1501 }
1502
1503 Node* src_start = array_element_address(src, src_offset, src_elem);
1504 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1505 // 'src_start' points to src array + scaled offset
1506 // 'dst_start' points to dst array + scaled offset
1507 Node* count = nullptr;
1508 if (compress) {
1509 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1510 } else {
1511 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1512 }
1513
1514 if (alloc != nullptr) {
1515 if (alloc->maybe_set_complete(&_gvn)) {
1516 // "You break it, you buy it."
1517 InitializeNode* init = alloc->initialization();
1518 assert(init->is_complete(), "we just did this");
1519 init->set_complete_with_arraycopy();
1520 assert(dst->is_CheckCastPP(), "sanity");
1521 assert(dst->in(0)->in(0) == init, "dest pinned");
1522 }
1523 // Do not let stores that initialize this object be reordered with
1524 // a subsequent store that would make this object accessible by
1525 // other threads.
1526 // Record what AllocateNode this StoreStore protects so that
1527 // escape analysis can go from the MemBarStoreStoreNode to the
1528 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1529 // based on the escape status of the AllocateNode.
1530 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1531 }
1532 if (compress) {
1533 set_result(_gvn.transform(count));
1534 }
1535 clear_upper_avx();
1536
1537 return true;
1538 }
1539
1540 #ifdef _LP64
1541 #define XTOP ,top() /*additional argument*/
1542 #else //_LP64
1543 #define XTOP /*no additional argument*/
1544 #endif //_LP64
1545
1546 //------------------------inline_string_toBytesU--------------------------
1547 // public static byte[] StringUTF16.toBytes0(char[] value, int off, int len)
1548 bool LibraryCallKit::inline_string_toBytesU() {
1549 // Get the arguments.
1550 assert(callee()->signature()->size() == 3, "character array encoder requires 3 arguments");
1551 Node* value = argument(0);
1552 Node* offset = argument(1);
1553 Node* length = argument(2);
1554
1555 Node* newcopy = nullptr;
1556
1557 // Set the original stack and the reexecute bit for the interpreter to reexecute
1558 // the bytecode that invokes StringUTF16.toBytes0() if deoptimization happens.
1559 { PreserveReexecuteState preexecs(this);
1560 jvms()->set_should_reexecute(true);
1561
1562 value = must_be_not_null(value, true);
1563 RegionNode* bailout = create_bailout();
1564 generate_negative_guard(offset, bailout, nullptr, true);
1565 generate_negative_guard(length, bailout, nullptr, true);
1566 generate_limit_guard(offset, length, load_array_length(value), bailout, true);
1567 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1568 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout, true);
1569 if (check_bailout(bailout)) {
1570 return true;
1571 }
1572
1573 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1574 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1575 newcopy = new_array(klass_node, size, 0); // no arguments to push
1576 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1577 guarantee(alloc != nullptr, "created above");
1578
1579 // Calculate starting addresses.
1580 Node* src_start = array_element_address(value, offset, T_CHAR);
1581 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1582
1583 // Check if dst array address is aligned to HeapWordSize
1584 bool aligned = (arrayOopDesc::base_offset_in_bytes(T_BYTE) % HeapWordSize == 0);
1585 // If true, then check if src array address is aligned to HeapWordSize
1586 if (aligned) {
1587 const TypeInt* toffset = gvn().type(offset)->is_int();
1588 aligned = toffset->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) +
1589 toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1590 }
1591
1592 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1593 const char* copyfunc_name = "arraycopy";
1594 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1595 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1596 OptoRuntime::fast_arraycopy_Type(),
1597 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1598 src_start, dst_start, ConvI2X(length) XTOP);
1599 // Do not let reads from the cloned object float above the arraycopy.
1600 if (alloc->maybe_set_complete(&_gvn)) {
1601 // "You break it, you buy it."
1602 InitializeNode* init = alloc->initialization();
1603 assert(init->is_complete(), "we just did this");
1604 init->set_complete_with_arraycopy();
1605 assert(newcopy->is_CheckCastPP(), "sanity");
1606 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1607 }
1608 // Do not let stores that initialize this object be reordered with
1609 // a subsequent store that would make this object accessible by
1610 // other threads.
1611 // Record what AllocateNode this StoreStore protects so that
1612 // escape analysis can go from the MemBarStoreStoreNode to the
1613 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1614 // based on the escape status of the AllocateNode.
1615 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1616 } // original reexecute is set back here
1617
1618 C->set_has_split_ifs(true); // Has chance for split-if optimization
1619 if (!stopped()) {
1620 set_result(newcopy);
1621 }
1622 clear_upper_avx();
1623
1624 return true;
1625 }
1626
1627 //------------------------inline_string_getCharsU--------------------------
1628 // public void StringUTF16.getChars0(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1629 bool LibraryCallKit::inline_string_getCharsU() {
1630 assert(callee()->signature()->size() == 5, "StringUTF16.getChars0() has 5 arguments");
1631 // Get the arguments.
1632 Node* src = argument(0);
1633 Node* src_begin = argument(1);
1634 Node* src_end = argument(2); // exclusive offset (i < src_end)
1635 Node* dst = argument(3);
1636 Node* dst_begin = argument(4);
1637
1638 // Check for allocation before we add nodes that would confuse
1639 // tightly_coupled_allocation()
1640 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1641
1642 // Check if a null path was taken unconditionally.
1643 src = must_be_not_null(src, true);
1644 dst = must_be_not_null(dst, true);
1645 if (stopped()) {
1646 return true;
1647 }
1648
1649 // Get length and convert char[] offset to byte[] offset
1650 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1651 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1652
1653 // Range checks
1654 RegionNode* bailout = create_bailout();
1655 generate_string_range_check(src, src_begin, length, true, bailout);
1656 generate_string_range_check(dst, dst_begin, length, false, bailout);
1657 if (check_bailout(bailout)) {
1658 return true;
1659 }
1660
1661 // Calculate starting addresses.
1662 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1663 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1664
1665 // Check if array addresses are aligned to HeapWordSize
1666 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1667 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1668 bool aligned = tsrc->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_BYTE) + tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1669 tdst->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) + tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1670
1671 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1672 const char* copyfunc_name = "arraycopy";
1673 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1674 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1675 OptoRuntime::fast_arraycopy_Type(),
1676 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1677 src_start, dst_start, ConvI2X(length) XTOP);
1678 // Do not let reads from the cloned object float above the arraycopy.
1679 if (alloc != nullptr) {
1680 if (alloc->maybe_set_complete(&_gvn)) {
1681 // "You break it, you buy it."
1682 InitializeNode* init = alloc->initialization();
1683 assert(init->is_complete(), "we just did this");
1684 init->set_complete_with_arraycopy();
1685 assert(dst->is_CheckCastPP(), "sanity");
1686 assert(dst->in(0)->in(0) == init, "dest pinned");
1687 }
1688 // Do not let stores that initialize this object be reordered with
1689 // a subsequent store that would make this object accessible by
1690 // other threads.
1691 // Record what AllocateNode this StoreStore protects so that
1692 // escape analysis can go from the MemBarStoreStoreNode to the
1693 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1694 // based on the escape status of the AllocateNode.
1695 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1696 } else {
1697 insert_mem_bar(Op_MemBarCPUOrder);
1698 }
1699
1700 C->set_has_split_ifs(true); // Has chance for split-if optimization
1701 return true;
1702 }
1703
1704 //----------------------inline_string_char_access----------------------------
1705 // Store/Load char to/from byte[] array.
1706 // static void StringUTF16.putChar(byte[] val, int index, int c)
1707 // static char StringUTF16.getChar(byte[] val, int index)
1708 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1709 Node* ch;
1710 if (is_store) {
1711 assert(callee()->signature()->size() == 3, "StringUTF16.putChar() has 3 arguments");
1712 ch = argument(2);
1713 } else {
1714 assert(callee()->signature()->size() == 2, "StringUTF16.getChar() has 2 arguments");
1715 ch = nullptr;
1716 }
1717 Node* value = argument(0);
1718 Node* index = argument(1);
1719
1720 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1721 // correctly requires matched array shapes.
1722 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1723 "sanity: byte[] and char[] bases agree");
1724 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1725 "sanity: byte[] and char[] scales agree");
1726
1727 // Bail when getChar over constants is requested: constant folding would
1728 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1729 // Java method would constant fold nicely instead.
1730 if (!is_store && value->is_Con() && index->is_Con()) {
1731 return false;
1732 }
1733
1734 // Save state and restore on bailout
1735 SavedState old_state(this);
1736
1737 value = must_be_not_null(value, true);
1738
1739 Node* adr = array_element_address(value, index, T_CHAR);
1740 if (adr->is_top()) {
1741 return false;
1742 }
1743 old_state.discard();
1744 if (is_store) {
1745 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1746 } else {
1747 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);
1748 set_result(ch);
1749 }
1750 return true;
1751 }
1752
1753
1754 //------------------------------inline_math-----------------------------------
1755 // public static double Math.abs(double)
1756 // public static double Math.sqrt(double)
1757 // public static double Math.log(double)
1758 // public static double Math.log10(double)
1759 // public static double Math.round(double)
1760 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1761 Node* arg = argument(0);
1762 Node* n = nullptr;
1763 switch (id) {
1764 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1765 case vmIntrinsics::_dsqrt:
1766 case vmIntrinsics::_dsqrt_strict:
1767 n = new SqrtDNode(C, control(), arg); break;
1768 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1769 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1770 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1771 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1772 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1773 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1774 default: fatal_unexpected_iid(id); break;
1775 }
1776 set_result(_gvn.transform(n));
1777 return true;
1778 }
1779
1780 //------------------------------inline_math-----------------------------------
1781 // public static float Math.abs(float)
1782 // public static int Math.abs(int)
1783 // public static long Math.abs(long)
1784 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1785 Node* arg = argument(0);
1786 Node* n = nullptr;
1787 switch (id) {
1788 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1789 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1790 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1791 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1792 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1793 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1794 default: fatal_unexpected_iid(id); break;
1795 }
1796 set_result(_gvn.transform(n));
1797 return true;
1798 }
1799
1800 //------------------------------runtime_math-----------------------------
1801 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1802 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1803 "must be (DD)D or (D)D type");
1804
1805 // Inputs
1806 Node* a = argument(0);
1807 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1808
1809 const TypePtr* no_memory_effects = nullptr;
1810 Node* trig = make_runtime_call(RC_LEAF | RC_PURE, call_type, funcAddr, funcName,
1811 no_memory_effects,
1812 a, top(), b, b ? top() : nullptr);
1813 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1814 #ifdef ASSERT
1815 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1816 assert(value_top == top(), "second value must be top");
1817 #endif
1818
1819 set_result(value);
1820 return true;
1821 }
1822
1823 //------------------------------inline_math_pow-----------------------------
1824 bool LibraryCallKit::inline_math_pow() {
1825 Node* base = argument(0);
1826 Node* exp = argument(2);
1827
1828 CallNode* pow = new PowDNode(C, base, exp);
1829 set_predefined_input_for_runtime_call(pow);
1830 pow = _gvn.transform(pow)->as_CallLeafPure();
1831 set_predefined_output_for_runtime_call(pow);
1832 Node* result = _gvn.transform(new ProjNode(pow, TypeFunc::Parms + 0));
1833 record_for_igvn(pow);
1834 set_result(result);
1835 return true;
1836 }
1837
1838 //------------------------------inline_math_native-----------------------------
1839 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1840 switch (id) {
1841 case vmIntrinsics::_dsin:
1842 return StubRoutines::dsin() != nullptr ?
1843 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1844 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1845 case vmIntrinsics::_dcos:
1846 return StubRoutines::dcos() != nullptr ?
1847 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1848 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1849 case vmIntrinsics::_dtan:
1850 return StubRoutines::dtan() != nullptr ?
1851 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1852 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1853 case vmIntrinsics::_dsinh:
1854 return StubRoutines::dsinh() != nullptr ?
1855 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsinh(), "dsinh") : false;
1856 case vmIntrinsics::_dtanh:
1857 return StubRoutines::dtanh() != nullptr ?
1858 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1859 case vmIntrinsics::_dcbrt:
1860 return StubRoutines::dcbrt() != nullptr ?
1861 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcbrt(), "dcbrt") : false;
1862 case vmIntrinsics::_dexp:
1863 return StubRoutines::dexp() != nullptr ?
1864 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1865 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1866 case vmIntrinsics::_dlog:
1867 return StubRoutines::dlog() != nullptr ?
1868 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1869 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1870 case vmIntrinsics::_dlog10:
1871 return StubRoutines::dlog10() != nullptr ?
1872 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1873 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1874
1875 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1876 case vmIntrinsics::_ceil:
1877 case vmIntrinsics::_floor:
1878 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1879
1880 case vmIntrinsics::_dsqrt:
1881 case vmIntrinsics::_dsqrt_strict:
1882 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1883 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1884 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1885 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1886 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1887
1888 case vmIntrinsics::_dpow: return inline_math_pow();
1889 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1890 case vmIntrinsics::_fcopySign: return inline_math(id);
1891 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1892 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1893 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1894
1895 // These intrinsics are not yet correctly implemented
1896 case vmIntrinsics::_datan2:
1897 return false;
1898
1899 default:
1900 fatal_unexpected_iid(id);
1901 return false;
1902 }
1903 }
1904
1905 //----------------------------inline_notify-----------------------------------*
1906 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1907 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1908 address func;
1909 if (id == vmIntrinsics::_notify) {
1910 func = OptoRuntime::monitor_notify_Java();
1911 } else {
1912 func = OptoRuntime::monitor_notifyAll_Java();
1913 }
1914 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1915 make_slow_call_ex(call, env()->Throwable_klass(), false);
1916 return true;
1917 }
1918
1919
1920 //----------------------------inline_min_max-----------------------------------
1921 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1922 Node* a = nullptr;
1923 Node* b = nullptr;
1924 Node* n = nullptr;
1925 switch (id) {
1926 case vmIntrinsics::_min:
1927 case vmIntrinsics::_max:
1928 case vmIntrinsics::_minF:
1929 case vmIntrinsics::_maxF:
1930 case vmIntrinsics::_minF_strict:
1931 case vmIntrinsics::_maxF_strict:
1932 case vmIntrinsics::_min_strict:
1933 case vmIntrinsics::_max_strict:
1934 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
1935 a = argument(0);
1936 b = argument(1);
1937 break;
1938 case vmIntrinsics::_minD:
1939 case vmIntrinsics::_maxD:
1940 case vmIntrinsics::_minD_strict:
1941 case vmIntrinsics::_maxD_strict:
1942 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
1943 a = argument(0);
1944 b = argument(2);
1945 break;
1946 case vmIntrinsics::_minL:
1947 case vmIntrinsics::_maxL:
1948 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
1949 a = argument(0);
1950 b = argument(2);
1951 break;
1952 default:
1953 fatal_unexpected_iid(id);
1954 break;
1955 }
1956
1957 switch (id) {
1958 case vmIntrinsics::_min:
1959 case vmIntrinsics::_min_strict:
1960 n = new MinINode(a, b);
1961 break;
1962 case vmIntrinsics::_max:
1963 case vmIntrinsics::_max_strict:
1964 n = new MaxINode(a, b);
1965 break;
1966 case vmIntrinsics::_minF:
1967 case vmIntrinsics::_minF_strict:
1968 n = new MinFNode(a, b);
1969 break;
1970 case vmIntrinsics::_maxF:
1971 case vmIntrinsics::_maxF_strict:
1972 n = new MaxFNode(a, b);
1973 break;
1974 case vmIntrinsics::_minD:
1975 case vmIntrinsics::_minD_strict:
1976 n = new MinDNode(a, b);
1977 break;
1978 case vmIntrinsics::_maxD:
1979 case vmIntrinsics::_maxD_strict:
1980 n = new MaxDNode(a, b);
1981 break;
1982 case vmIntrinsics::_minL:
1983 n = new MinLNode(_gvn.C, a, b);
1984 break;
1985 case vmIntrinsics::_maxL:
1986 n = new MaxLNode(_gvn.C, a, b);
1987 break;
1988 default:
1989 fatal_unexpected_iid(id);
1990 break;
1991 }
1992
1993 set_result(_gvn.transform(n));
1994 return true;
1995 }
1996
1997 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
1998 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
1999 env()->ArithmeticException_instance())) {
2000 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
2001 // so let's bail out intrinsic rather than risking deopting again.
2002 return false;
2003 }
2004
2005 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2006 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2007 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2008 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2009
2010 {
2011 PreserveJVMState pjvms(this);
2012 PreserveReexecuteState preexecs(this);
2013 jvms()->set_should_reexecute(true);
2014
2015 set_control(slow_path);
2016 set_i_o(i_o());
2017
2018 builtin_throw(Deoptimization::Reason_intrinsic,
2019 env()->ArithmeticException_instance(),
2020 /*allow_too_many_traps*/ false);
2021 }
2022
2023 set_control(fast_path);
2024 set_result(math);
2025 return true;
2026 }
2027
2028 template <typename OverflowOp>
2029 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2030 typedef typename OverflowOp::MathOp MathOp;
2031
2032 MathOp* mathOp = new MathOp(arg1, arg2);
2033 Node* operation = _gvn.transform( mathOp );
2034 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2035 return inline_math_mathExact(operation, ofcheck);
2036 }
2037
2038 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2039 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2040 }
2041
2042 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2043 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2044 }
2045
2046 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2047 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2048 }
2049
2050 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2051 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2052 }
2053
2054 bool LibraryCallKit::inline_math_negateExactI() {
2055 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2056 }
2057
2058 bool LibraryCallKit::inline_math_negateExactL() {
2059 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2060 }
2061
2062 bool LibraryCallKit::inline_math_multiplyExactI() {
2063 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2064 }
2065
2066 bool LibraryCallKit::inline_math_multiplyExactL() {
2067 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2068 }
2069
2070 bool LibraryCallKit::inline_math_multiplyHigh() {
2071 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2072 return true;
2073 }
2074
2075 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2076 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2077 return true;
2078 }
2079
2080 inline int
2081 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2082 const TypePtr* base_type = TypePtr::NULL_PTR;
2083 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2084 if (base_type == nullptr) {
2085 // Unknown type.
2086 return Type::AnyPtr;
2087 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2088 // Since this is a null+long form, we have to switch to a rawptr.
2089 base = _gvn.transform(new CastX2PNode(offset));
2090 offset = MakeConX(0);
2091 return Type::RawPtr;
2092 } else if (base_type->base() == Type::RawPtr) {
2093 return Type::RawPtr;
2094 } else if (base_type->isa_oopptr()) {
2095 // Base is never null => always a heap address.
2096 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2097 return Type::OopPtr;
2098 }
2099 // Offset is small => always a heap address.
2100 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2101 if (offset_type != nullptr &&
2102 base_type->offset() == 0 && // (should always be?)
2103 offset_type->_lo >= 0 &&
2104 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2105 return Type::OopPtr;
2106 } else if (type == T_OBJECT) {
2107 // off heap access to an oop doesn't make any sense. Has to be on
2108 // heap.
2109 return Type::OopPtr;
2110 }
2111 // Otherwise, it might either be oop+off or null+addr.
2112 return Type::AnyPtr;
2113 } else {
2114 // No information:
2115 return Type::AnyPtr;
2116 }
2117 }
2118
2119 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2120 Node* uncasted_base = base;
2121 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2122 if (kind == Type::RawPtr) {
2123 return off_heap_plus_addr(uncasted_base, offset);
2124 } else if (kind == Type::AnyPtr) {
2125 assert(base == uncasted_base, "unexpected base change");
2126 if (can_cast) {
2127 if (!_gvn.type(base)->speculative_maybe_null() &&
2128 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2129 // According to profiling, this access is always on
2130 // heap. Casting the base to not null and thus avoiding membars
2131 // around the access should allow better optimizations
2132 Node* null_ctl = top();
2133 base = null_check_oop(base, &null_ctl, true, true, true);
2134 assert(null_ctl->is_top(), "no null control here");
2135 return basic_plus_adr(base, offset);
2136 } else if (_gvn.type(base)->speculative_always_null() &&
2137 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2138 // According to profiling, this access is always off
2139 // heap.
2140 base = null_assert(base);
2141 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2142 offset = MakeConX(0);
2143 return off_heap_plus_addr(raw_base, offset);
2144 }
2145 }
2146 // We don't know if it's an on heap or off heap access. Fall back
2147 // to raw memory access.
2148 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2149 return off_heap_plus_addr(raw, offset);
2150 } else {
2151 assert(base == uncasted_base, "unexpected base change");
2152 // We know it's an on heap access so base can't be null
2153 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2154 base = must_be_not_null(base, true);
2155 }
2156 return basic_plus_adr(base, offset);
2157 }
2158 }
2159
2160 //--------------------------inline_number_methods-----------------------------
2161 // inline int Integer.numberOfLeadingZeros(int)
2162 // inline int Long.numberOfLeadingZeros(long)
2163 //
2164 // inline int Integer.numberOfTrailingZeros(int)
2165 // inline int Long.numberOfTrailingZeros(long)
2166 //
2167 // inline int Integer.bitCount(int)
2168 // inline int Long.bitCount(long)
2169 //
2170 // inline char Character.reverseBytes(char)
2171 // inline short Short.reverseBytes(short)
2172 // inline int Integer.reverseBytes(int)
2173 // inline long Long.reverseBytes(long)
2174 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2175 Node* arg = argument(0);
2176 Node* n = nullptr;
2177 switch (id) {
2178 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2179 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2180 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2181 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2182 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2183 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2184 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2185 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2186 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2187 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2188 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2189 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2190 default: fatal_unexpected_iid(id); break;
2191 }
2192 set_result(_gvn.transform(n));
2193 return true;
2194 }
2195
2196 //--------------------------inline_bitshuffle_methods-----------------------------
2197 // inline int Integer.compress(int, int)
2198 // inline int Integer.expand(int, int)
2199 // inline long Long.compress(long, long)
2200 // inline long Long.expand(long, long)
2201 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2202 Node* n = nullptr;
2203 switch (id) {
2204 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2205 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2206 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2207 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2208 default: fatal_unexpected_iid(id); break;
2209 }
2210 set_result(_gvn.transform(n));
2211 return true;
2212 }
2213
2214 //--------------------------inline_number_methods-----------------------------
2215 // inline int Integer.compareUnsigned(int, int)
2216 // inline int Long.compareUnsigned(long, long)
2217 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2218 Node* arg1 = argument(0);
2219 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2220 Node* n = nullptr;
2221 switch (id) {
2222 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2223 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2224 default: fatal_unexpected_iid(id); break;
2225 }
2226 set_result(_gvn.transform(n));
2227 return true;
2228 }
2229
2230 //--------------------------inline_unsigned_divmod_methods-----------------------------
2231 // inline int Integer.divideUnsigned(int, int)
2232 // inline int Integer.remainderUnsigned(int, int)
2233 // inline long Long.divideUnsigned(long, long)
2234 // inline long Long.remainderUnsigned(long, long)
2235 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2236 Node* n = nullptr;
2237 switch (id) {
2238 case vmIntrinsics::_divideUnsigned_i: {
2239 zero_check_int(argument(1));
2240 // Compile-time detect of null-exception
2241 if (stopped()) {
2242 return true; // keep the graph constructed so far
2243 }
2244 n = new UDivINode(control(), argument(0), argument(1));
2245 break;
2246 }
2247 case vmIntrinsics::_divideUnsigned_l: {
2248 zero_check_long(argument(2));
2249 // Compile-time detect of null-exception
2250 if (stopped()) {
2251 return true; // keep the graph constructed so far
2252 }
2253 n = new UDivLNode(control(), argument(0), argument(2));
2254 break;
2255 }
2256 case vmIntrinsics::_remainderUnsigned_i: {
2257 zero_check_int(argument(1));
2258 // Compile-time detect of null-exception
2259 if (stopped()) {
2260 return true; // keep the graph constructed so far
2261 }
2262 n = new UModINode(control(), argument(0), argument(1));
2263 break;
2264 }
2265 case vmIntrinsics::_remainderUnsigned_l: {
2266 zero_check_long(argument(2));
2267 // Compile-time detect of null-exception
2268 if (stopped()) {
2269 return true; // keep the graph constructed so far
2270 }
2271 n = new UModLNode(control(), argument(0), argument(2));
2272 break;
2273 }
2274 default: fatal_unexpected_iid(id); break;
2275 }
2276 set_result(_gvn.transform(n));
2277 return true;
2278 }
2279
2280 //----------------------------inline_unsafe_access----------------------------
2281
2282 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2283 // Attempt to infer a sharper value type from the offset and base type.
2284 ciKlass* sharpened_klass = nullptr;
2285
2286 // See if it is an instance field, with an object type.
2287 if (alias_type->field() != nullptr) {
2288 if (alias_type->field()->type()->is_klass()) {
2289 sharpened_klass = alias_type->field()->type()->as_klass();
2290 }
2291 }
2292
2293 const TypeOopPtr* result = nullptr;
2294 // See if it is a narrow oop array.
2295 if (adr_type->isa_aryptr()) {
2296 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2297 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2298 if (elem_type != nullptr && elem_type->is_loaded()) {
2299 // Sharpen the value type.
2300 result = elem_type;
2301 }
2302 }
2303 }
2304
2305 // The sharpened class might be unloaded if there is no class loader
2306 // contraint in place.
2307 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2308 // Sharpen the value type.
2309 result = TypeOopPtr::make_from_klass(sharpened_klass);
2310 }
2311 if (result != nullptr) {
2312 #ifndef PRODUCT
2313 if (C->print_intrinsics() || C->print_inlining()) {
2314 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2315 tty->print(" sharpened value: "); result->dump(); tty->cr();
2316 }
2317 #endif
2318 }
2319 return result;
2320 }
2321
2322 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2323 switch (kind) {
2324 case Relaxed:
2325 return MO_UNORDERED;
2326 case Opaque:
2327 return MO_RELAXED;
2328 case Acquire:
2329 return MO_ACQUIRE;
2330 case Release:
2331 return MO_RELEASE;
2332 case Volatile:
2333 return MO_SEQ_CST;
2334 default:
2335 ShouldNotReachHere();
2336 return 0;
2337 }
2338 }
2339
2340 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2341 if (callee()->is_static()) return false; // caller must have the capability!
2342 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2343 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2344 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2345 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2346
2347 if (is_reference_type(type)) {
2348 decorators |= ON_UNKNOWN_OOP_REF;
2349 }
2350
2351 if (unaligned) {
2352 decorators |= C2_UNALIGNED;
2353 }
2354
2355 #ifndef PRODUCT
2356 {
2357 ResourceMark rm;
2358 // Check the signatures.
2359 ciSignature* sig = callee()->signature();
2360 #ifdef ASSERT
2361 if (!is_store) {
2362 // Object getReference(Object base, int/long offset), etc.
2363 BasicType rtype = sig->return_type()->basic_type();
2364 assert(rtype == type, "getter must return the expected value");
2365 assert(sig->count() == 2, "oop getter has 2 arguments");
2366 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2367 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2368 } else {
2369 // void putReference(Object base, int/long offset, Object x), etc.
2370 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2371 assert(sig->count() == 3, "oop putter has 3 arguments");
2372 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2373 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2374 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2375 assert(vtype == type, "putter must accept the expected value");
2376 }
2377 #endif // ASSERT
2378 }
2379 #endif //PRODUCT
2380
2381 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2382
2383 Node* receiver = argument(0); // type: oop
2384
2385 // Build address expression.
2386 Node* heap_base_oop = top();
2387
2388 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2389 Node* base = argument(1); // type: oop
2390 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2391 Node* offset = argument(2); // type: long
2392 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2393 // to be plain byte offsets, which are also the same as those accepted
2394 // by oopDesc::field_addr.
2395 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2396 "fieldOffset must be byte-scaled");
2397 // 32-bit machines ignore the high half!
2398 offset = ConvL2X(offset);
2399
2400 // Save state and restore on bailout
2401 SavedState old_state(this);
2402
2403 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2404 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2405
2406 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2407 if (type != T_OBJECT) {
2408 decorators |= IN_NATIVE; // off-heap primitive access
2409 } else {
2410 return false; // off-heap oop accesses are not supported
2411 }
2412 } else {
2413 heap_base_oop = base; // on-heap or mixed access
2414 }
2415
2416 // Can base be null? Otherwise, always on-heap access.
2417 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2418
2419 if (!can_access_non_heap) {
2420 decorators |= IN_HEAP;
2421 }
2422
2423 Node* val = is_store ? argument(4) : nullptr;
2424
2425 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2426 if (adr_type == TypePtr::NULL_PTR) {
2427 return false; // off-heap access with zero address
2428 }
2429
2430 // Try to categorize the address.
2431 Compile::AliasType* alias_type = C->alias_type(adr_type);
2432 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2433
2434 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2435 alias_type->adr_type() == TypeAryPtr::RANGE) {
2436 return false; // not supported
2437 }
2438
2439 bool mismatched = false;
2440 BasicType bt = alias_type->basic_type();
2441 if (bt != T_ILLEGAL) {
2442 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2443 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2444 // Alias type doesn't differentiate between byte[] and boolean[]).
2445 // Use address type to get the element type.
2446 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2447 }
2448 if (is_reference_type(bt, true)) {
2449 // accessing an array field with getReference is not a mismatch
2450 bt = T_OBJECT;
2451 }
2452 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2453 // Don't intrinsify mismatched object accesses
2454 return false;
2455 }
2456 mismatched = (bt != type);
2457 } else if (alias_type->adr_type()->isa_oopptr()) {
2458 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2459 }
2460
2461 old_state.discard();
2462 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2463
2464 if (mismatched) {
2465 decorators |= C2_MISMATCHED;
2466 }
2467
2468 // First guess at the value type.
2469 const Type *value_type = Type::get_const_basic_type(type);
2470
2471 // Figure out the memory ordering.
2472 decorators |= mo_decorator_for_access_kind(kind);
2473
2474 if (!is_store && type == T_OBJECT) {
2475 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2476 if (tjp != nullptr) {
2477 value_type = tjp;
2478 }
2479 }
2480
2481 receiver = null_check(receiver);
2482 if (stopped()) {
2483 return true;
2484 }
2485 // Heap pointers get a null-check from the interpreter,
2486 // as a courtesy. However, this is not guaranteed by Unsafe,
2487 // and it is not possible to fully distinguish unintended nulls
2488 // from intended ones in this API.
2489
2490 if (!is_store) {
2491 Node* p = nullptr;
2492 // Try to constant fold a load from a constant field
2493 ciField* field = alias_type->field();
2494 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) {
2495 // final or stable field
2496 p = make_constant_from_field(field, heap_base_oop);
2497 }
2498
2499 if (p == nullptr) { // Could not constant fold the load
2500 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2501 // Normalize the value returned by getBoolean in the following cases
2502 if (type == T_BOOLEAN &&
2503 (mismatched ||
2504 heap_base_oop == top() || // - heap_base_oop is null or
2505 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2506 // and the unsafe access is made to large offset
2507 // (i.e., larger than the maximum offset necessary for any
2508 // field access)
2509 ) {
2510 IdealKit ideal = IdealKit(this);
2511 #define __ ideal.
2512 IdealVariable normalized_result(ideal);
2513 __ declarations_done();
2514 __ set(normalized_result, p);
2515 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2516 __ set(normalized_result, ideal.ConI(1));
2517 ideal.end_if();
2518 final_sync(ideal);
2519 p = __ value(normalized_result);
2520 #undef __
2521 }
2522 }
2523 if (type == T_ADDRESS) {
2524 p = gvn().transform(new CastP2XNode(nullptr, p));
2525 p = ConvX2UL(p);
2526 }
2527 // The load node has the control of the preceding MemBarCPUOrder. All
2528 // following nodes will have the control of the MemBarCPUOrder inserted at
2529 // the end of this method. So, pushing the load onto the stack at a later
2530 // point is fine.
2531 set_result(p);
2532 } else {
2533 if (bt == T_ADDRESS) {
2534 // Repackage the long as a pointer.
2535 val = ConvL2X(val);
2536 val = gvn().transform(new CastX2PNode(val));
2537 }
2538 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2539 }
2540
2541 return true;
2542 }
2543
2544 //----------------------------inline_unsafe_load_store----------------------------
2545 // This method serves a couple of different customers (depending on LoadStoreKind):
2546 //
2547 // LS_cmp_swap:
2548 //
2549 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2550 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2551 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2552 //
2553 // LS_cmp_swap_weak:
2554 //
2555 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2556 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2557 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2558 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2559 //
2560 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2561 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2562 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2563 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2564 //
2565 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2566 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2567 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2568 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2569 //
2570 // LS_cmp_exchange:
2571 //
2572 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2573 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2574 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2575 //
2576 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2577 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2578 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2579 //
2580 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2581 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2582 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2583 //
2584 // LS_get_add:
2585 //
2586 // int getAndAddInt( Object o, long offset, int delta)
2587 // long getAndAddLong(Object o, long offset, long delta)
2588 //
2589 // LS_get_set:
2590 //
2591 // int getAndSet(Object o, long offset, int newValue)
2592 // long getAndSet(Object o, long offset, long newValue)
2593 // Object getAndSet(Object o, long offset, Object newValue)
2594 //
2595 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2596 // This basic scheme here is the same as inline_unsafe_access, but
2597 // differs in enough details that combining them would make the code
2598 // overly confusing. (This is a true fact! I originally combined
2599 // them, but even I was confused by it!) As much code/comments as
2600 // possible are retained from inline_unsafe_access though to make
2601 // the correspondences clearer. - dl
2602
2603 if (callee()->is_static()) return false; // caller must have the capability!
2604
2605 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2606 decorators |= mo_decorator_for_access_kind(access_kind);
2607
2608 #ifndef PRODUCT
2609 BasicType rtype;
2610 {
2611 ResourceMark rm;
2612 // Check the signatures.
2613 ciSignature* sig = callee()->signature();
2614 rtype = sig->return_type()->basic_type();
2615 switch(kind) {
2616 case LS_get_add:
2617 case LS_get_set: {
2618 // Check the signatures.
2619 #ifdef ASSERT
2620 assert(rtype == type, "get and set must return the expected type");
2621 assert(sig->count() == 3, "get and set has 3 arguments");
2622 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2623 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2624 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2625 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2626 #endif // ASSERT
2627 break;
2628 }
2629 case LS_cmp_swap:
2630 case LS_cmp_swap_weak: {
2631 // Check the signatures.
2632 #ifdef ASSERT
2633 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2634 assert(sig->count() == 4, "CAS has 4 arguments");
2635 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2636 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2637 #endif // ASSERT
2638 break;
2639 }
2640 case LS_cmp_exchange: {
2641 // Check the signatures.
2642 #ifdef ASSERT
2643 assert(rtype == type, "CAS must return the expected type");
2644 assert(sig->count() == 4, "CAS has 4 arguments");
2645 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2646 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2647 #endif // ASSERT
2648 break;
2649 }
2650 default:
2651 ShouldNotReachHere();
2652 }
2653 }
2654 #endif //PRODUCT
2655
2656 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2657
2658 // Get arguments:
2659 Node* receiver = nullptr;
2660 Node* base = nullptr;
2661 Node* offset = nullptr;
2662 Node* oldval = nullptr;
2663 Node* newval = nullptr;
2664 switch(kind) {
2665 case LS_cmp_swap:
2666 case LS_cmp_swap_weak:
2667 case LS_cmp_exchange: {
2668 const bool two_slot_type = type2size[type] == 2;
2669 receiver = argument(0); // type: oop
2670 base = argument(1); // type: oop
2671 offset = argument(2); // type: long
2672 oldval = argument(4); // type: oop, int, or long
2673 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2674 break;
2675 }
2676 case LS_get_add:
2677 case LS_get_set: {
2678 receiver = argument(0); // type: oop
2679 base = argument(1); // type: oop
2680 offset = argument(2); // type: long
2681 oldval = nullptr;
2682 newval = argument(4); // type: oop, int, or long
2683 break;
2684 }
2685 default:
2686 ShouldNotReachHere();
2687 }
2688
2689 // Build field offset expression.
2690 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2691 // to be plain byte offsets, which are also the same as those accepted
2692 // by oopDesc::field_addr.
2693 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2694 // 32-bit machines ignore the high half of long offsets
2695 offset = ConvL2X(offset);
2696 // Save state and restore on bailout
2697 SavedState old_state(this);
2698 Node* adr = make_unsafe_address(base, offset,type, false);
2699 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2700
2701 Compile::AliasType* alias_type = C->alias_type(adr_type);
2702 BasicType bt = alias_type->basic_type();
2703 if (bt != T_ILLEGAL &&
2704 (is_reference_type(bt) != (type == T_OBJECT))) {
2705 // Don't intrinsify mismatched object accesses.
2706 return false;
2707 }
2708
2709 old_state.discard();
2710
2711 // For CAS, unlike inline_unsafe_access, there seems no point in
2712 // trying to refine types. Just use the coarse types here.
2713 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2714 const Type *value_type = Type::get_const_basic_type(type);
2715
2716 switch (kind) {
2717 case LS_get_set:
2718 case LS_cmp_exchange: {
2719 if (type == T_OBJECT) {
2720 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2721 if (tjp != nullptr) {
2722 value_type = tjp;
2723 }
2724 }
2725 break;
2726 }
2727 case LS_cmp_swap:
2728 case LS_cmp_swap_weak:
2729 case LS_get_add:
2730 break;
2731 default:
2732 ShouldNotReachHere();
2733 }
2734
2735 // Null check receiver.
2736 receiver = null_check(receiver);
2737 if (stopped()) {
2738 return true;
2739 }
2740
2741 int alias_idx = C->get_alias_index(adr_type);
2742
2743 if (is_reference_type(type)) {
2744 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2745
2746 // Transformation of a value which could be null pointer (CastPP #null)
2747 // could be delayed during Parse (for example, in adjust_map_after_if()).
2748 // Execute transformation here to avoid barrier generation in such case.
2749 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2750 newval = _gvn.makecon(TypePtr::NULL_PTR);
2751
2752 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2753 // Refine the value to a null constant, when it is known to be null
2754 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2755 }
2756 }
2757
2758 Node* result = nullptr;
2759 switch (kind) {
2760 case LS_cmp_exchange: {
2761 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2762 oldval, newval, value_type, type, decorators);
2763 break;
2764 }
2765 case LS_cmp_swap_weak:
2766 decorators |= C2_WEAK_CMPXCHG;
2767 case LS_cmp_swap: {
2768 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2769 oldval, newval, value_type, type, decorators);
2770 break;
2771 }
2772 case LS_get_set: {
2773 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2774 newval, value_type, type, decorators);
2775 break;
2776 }
2777 case LS_get_add: {
2778 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2779 newval, value_type, type, decorators);
2780 break;
2781 }
2782 default:
2783 ShouldNotReachHere();
2784 }
2785
2786 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2787 set_result(result);
2788 return true;
2789 }
2790
2791 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2792 // Regardless of form, don't allow previous ld/st to move down,
2793 // then issue acquire, release, or volatile mem_bar.
2794 insert_mem_bar(Op_MemBarCPUOrder);
2795 switch(id) {
2796 case vmIntrinsics::_loadFence:
2797 insert_mem_bar(Op_LoadFence);
2798 return true;
2799 case vmIntrinsics::_storeFence:
2800 insert_mem_bar(Op_StoreFence);
2801 return true;
2802 case vmIntrinsics::_storeStoreFence:
2803 insert_mem_bar(Op_StoreStoreFence);
2804 return true;
2805 case vmIntrinsics::_fullFence:
2806 insert_mem_bar(Op_MemBarFull);
2807 return true;
2808 default:
2809 fatal_unexpected_iid(id);
2810 return false;
2811 }
2812 }
2813
2814 bool LibraryCallKit::inline_onspinwait() {
2815 insert_mem_bar(Op_OnSpinWait);
2816 return true;
2817 }
2818
2819 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2820 if (!kls->is_Con()) {
2821 return true;
2822 }
2823 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2824 if (klsptr == nullptr) {
2825 return true;
2826 }
2827 ciInstanceKlass* ik = klsptr->instance_klass();
2828 // don't need a guard for a klass that is already initialized
2829 return !ik->is_initialized();
2830 }
2831
2832 //----------------------------inline_unsafe_writeback0-------------------------
2833 // public native void Unsafe.writeback0(long address)
2834 bool LibraryCallKit::inline_unsafe_writeback0() {
2835 if (!Matcher::has_match_rule(Op_CacheWB)) {
2836 return false;
2837 }
2838 #ifndef PRODUCT
2839 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2840 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2841 ciSignature* sig = callee()->signature();
2842 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2843 #endif
2844 null_check_receiver(); // null-check, then ignore
2845 Node *addr = argument(1);
2846 addr = new CastX2PNode(addr);
2847 addr = _gvn.transform(addr);
2848 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2849 flush = _gvn.transform(flush);
2850 set_memory(flush, TypeRawPtr::BOTTOM);
2851 return true;
2852 }
2853
2854 //----------------------------inline_unsafe_writeback0-------------------------
2855 // public native void Unsafe.writeback0(long address)
2856 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2857 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2858 return false;
2859 }
2860 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2861 return false;
2862 }
2863 #ifndef PRODUCT
2864 assert(Matcher::has_match_rule(Op_CacheWB),
2865 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2866 : "found match rule for CacheWBPostSync but not CacheWB"));
2867
2868 #endif
2869 null_check_receiver(); // null-check, then ignore
2870 Node *sync;
2871 if (is_pre) {
2872 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2873 } else {
2874 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2875 }
2876 sync = _gvn.transform(sync);
2877 set_memory(sync, TypeRawPtr::BOTTOM);
2878 return true;
2879 }
2880
2881 //----------------------------inline_unsafe_allocate---------------------------
2882 // public native Object Unsafe.allocateInstance(Class<?> cls);
2883 bool LibraryCallKit::inline_unsafe_allocate() {
2884
2885 #if INCLUDE_JVMTI
2886 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
2887 return false;
2888 }
2889 #endif //INCLUDE_JVMTI
2890
2891 if (callee()->is_static()) return false; // caller must have the capability!
2892
2893 null_check_receiver(); // null-check, then ignore
2894 Node* cls = null_check(argument(1));
2895 if (stopped()) return true;
2896
2897 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
2898 kls = null_check(kls);
2899 if (stopped()) return true; // argument was like int.class
2900
2901 #if INCLUDE_JVMTI
2902 // Don't try to access new allocated obj in the intrinsic.
2903 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
2904 // Deoptimize and allocate in interpreter instead.
2905 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
2906 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
2907 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
2908 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
2909 {
2910 BuildCutout unless(this, tst, PROB_MAX);
2911 uncommon_trap(Deoptimization::Reason_intrinsic,
2912 Deoptimization::Action_make_not_entrant);
2913 }
2914 if (stopped()) {
2915 return true;
2916 }
2917 #endif //INCLUDE_JVMTI
2918
2919 Node* test = nullptr;
2920 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2921 // Note: The argument might still be an illegal value like
2922 // Serializable.class or Object[].class. The runtime will handle it.
2923 // But we must make an explicit check for initialization.
2924 Node* insp = off_heap_plus_addr(kls, in_bytes(InstanceKlass::init_state_offset()));
2925 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2926 // can generate code to load it as unsigned byte.
2927 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
2928 Node* bits = intcon(InstanceKlass::fully_initialized);
2929 test = _gvn.transform(new SubINode(inst, bits));
2930 // The 'test' is non-zero if we need to take a slow path.
2931 }
2932
2933 Node* obj = new_instance(kls, test);
2934 set_result(obj);
2935 return true;
2936 }
2937
2938 //------------------------inline_native_time_funcs--------------
2939 // inline code for System.currentTimeMillis() and System.nanoTime()
2940 // these have the same type and signature
2941 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2942 const TypeFunc* tf = OptoRuntime::void_long_Type();
2943 const TypePtr* no_memory_effects = nullptr;
2944 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2945 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2946 #ifdef ASSERT
2947 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2948 assert(value_top == top(), "second value must be top");
2949 #endif
2950 set_result(value);
2951 return true;
2952 }
2953
2954 //--------------------inline_native_vthread_start_transition--------------------
2955 // inline void startTransition(boolean is_mount);
2956 // inline void startFinalTransition();
2957 // Pseudocode of implementation:
2958 //
2959 // java_lang_Thread::set_is_in_vthread_transition(vthread, true);
2960 // carrier->set_is_in_vthread_transition(true);
2961 // OrderAccess::storeload();
2962 // int disable_requests = java_lang_Thread::vthread_transition_disable_count(vthread)
2963 // + global_vthread_transition_disable_count();
2964 // if (disable_requests > 0) {
2965 // slow path: runtime call
2966 // }
2967 bool LibraryCallKit::inline_native_vthread_start_transition(address funcAddr, const char* funcName, bool is_final_transition) {
2968 Node* vt_oop = must_be_not_null(argument(0), true); // VirtualThread this argument
2969 IdealKit ideal(this);
2970
2971 Node* thread = ideal.thread();
2972 Node* jt_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
2973 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
2974 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2975 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2976 insert_mem_bar(Op_MemBarStoreLoad);
2977 ideal.sync_kit(this);
2978
2979 Node* global_disable_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::global_vthread_transition_disable_count_address()));
2980 Node* global_disable = ideal.load(ideal.ctrl(), global_disable_addr, TypeInt::INT, T_INT, Compile::AliasIdxRaw, true /*require_atomic_access*/);
2981 Node* vt_disable_addr = basic_plus_adr(vt_oop, java_lang_Thread::vthread_transition_disable_count_offset());
2982 const TypePtr* vt_disable_addr_t = _gvn.type(vt_disable_addr)->is_ptr();
2983 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*/);
2984 Node* disabled = _gvn.transform(new AddINode(global_disable, vt_disable));
2985
2986 ideal.if_then(disabled, BoolTest::ne, ideal.ConI(0)); {
2987 sync_kit(ideal);
2988 Node* is_mount = is_final_transition ? ideal.ConI(0) : argument(1);
2989 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
2990 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
2991 ideal.sync_kit(this);
2992 }
2993 ideal.end_if();
2994
2995 final_sync(ideal);
2996 return true;
2997 }
2998
2999 bool LibraryCallKit::inline_native_vthread_end_transition(address funcAddr, const char* funcName, bool is_first_transition) {
3000 Node* vt_oop = must_be_not_null(argument(0), true); // VirtualThread this argument
3001 IdealKit ideal(this);
3002
3003 Node* _notify_jvmti_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::notify_jvmti_events_address()));
3004 Node* _notify_jvmti = ideal.load(ideal.ctrl(), _notify_jvmti_addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3005
3006 ideal.if_then(_notify_jvmti, BoolTest::eq, ideal.ConI(1)); {
3007 sync_kit(ideal);
3008 Node* is_mount = is_first_transition ? ideal.ConI(1) : argument(1);
3009 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3010 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3011 ideal.sync_kit(this);
3012 } ideal.else_(); {
3013 Node* thread = ideal.thread();
3014 Node* jt_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3015 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3016
3017 sync_kit(ideal);
3018 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3019 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3020 ideal.sync_kit(this);
3021 } ideal.end_if();
3022
3023 final_sync(ideal);
3024 return true;
3025 }
3026
3027 #if INCLUDE_JVMTI
3028
3029 // Always update the is_disable_suspend bit.
3030 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3031 if (!DoJVMTIVirtualThreadTransitions) {
3032 return true;
3033 }
3034 IdealKit ideal(this);
3035
3036 {
3037 // unconditionally update the is_disable_suspend bit in current JavaThread
3038 Node* thread = ideal.thread();
3039 Node* arg = argument(0); // argument for notification
3040 Node* addr = off_heap_plus_addr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3041 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3042
3043 sync_kit(ideal);
3044 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3045 ideal.sync_kit(this);
3046 }
3047 final_sync(ideal);
3048
3049 return true;
3050 }
3051
3052 #endif // INCLUDE_JVMTI
3053
3054 #ifdef JFR_HAVE_INTRINSICS
3055
3056 /**
3057 * if oop->klass != null
3058 * // normal class
3059 * epoch = _epoch_state ? 2 : 1
3060 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3061 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3062 * }
3063 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3064 * else
3065 * // primitive class
3066 * if oop->array_klass != null
3067 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3068 * else
3069 * id = LAST_TYPE_ID + 1 // void class path
3070 * if (!signaled)
3071 * signaled = true
3072 */
3073 bool LibraryCallKit::inline_native_classID() {
3074 Node* cls = argument(0);
3075
3076 IdealKit ideal(this);
3077 #define __ ideal.
3078 IdealVariable result(ideal); __ declarations_done();
3079 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3080 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3081 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3082
3083
3084 __ if_then(kls, BoolTest::ne, null()); {
3085 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3086 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3087
3088 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3089 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3090 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3091 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3092 mask = _gvn.transform(new OrLNode(mask, epoch));
3093 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3094
3095 float unlikely = PROB_UNLIKELY(0.999);
3096 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3097 sync_kit(ideal);
3098 make_runtime_call(RC_LEAF,
3099 OptoRuntime::class_id_load_barrier_Type(),
3100 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3101 "class id load barrier",
3102 TypePtr::BOTTOM,
3103 kls);
3104 ideal.sync_kit(this);
3105 } __ end_if();
3106
3107 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3108 } __ else_(); {
3109 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3110 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3111 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3112 __ if_then(array_kls, BoolTest::ne, null()); {
3113 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3114 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3115 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3116 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3117 } __ else_(); {
3118 // void class case
3119 ideal.set(result, longcon(LAST_TYPE_ID + 1));
3120 } __ end_if();
3121
3122 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3123 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3124 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3125 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3126 } __ end_if();
3127 } __ end_if();
3128
3129 final_sync(ideal);
3130 set_result(ideal.value(result));
3131 #undef __
3132 return true;
3133 }
3134
3135 //------------------------inline_native_jvm_commit------------------
3136 bool LibraryCallKit::inline_native_jvm_commit() {
3137 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3138
3139 // Save input memory and i_o state.
3140 Node* input_memory_state = reset_memory();
3141 set_all_memory(input_memory_state);
3142 Node* input_io_state = i_o();
3143
3144 // TLS.
3145 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3146 // Jfr java buffer.
3147 Node* java_buffer_offset = _gvn.transform(AddPNode::make_off_heap(tls_ptr, MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR))));
3148 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3149 Node* java_buffer_pos_offset = _gvn.transform(AddPNode::make_off_heap(java_buffer, MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET))));
3150
3151 // Load the current value of the notified field in the JfrThreadLocal.
3152 Node* notified_offset = off_heap_plus_addr(tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3153 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3154
3155 // Test for notification.
3156 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3157 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3158 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3159
3160 // True branch, is notified.
3161 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3162 set_control(is_notified);
3163
3164 // Reset notified state.
3165 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3166 Node* notified_reset_memory = reset_memory();
3167
3168 // 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.
3169 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3170 // Convert the machine-word to a long.
3171 Node* current_pos = ConvX2L(current_pos_X);
3172
3173 // False branch, not notified.
3174 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3175 set_control(not_notified);
3176 set_all_memory(input_memory_state);
3177
3178 // Arg is the next position as a long.
3179 Node* arg = argument(0);
3180 // Convert long to machine-word.
3181 Node* next_pos_X = ConvL2X(arg);
3182
3183 // Store the next_position to the underlying jfr java buffer.
3184 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3185
3186 Node* commit_memory = reset_memory();
3187 set_all_memory(commit_memory);
3188
3189 // 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.
3190 Node* java_buffer_flags_offset = _gvn.transform(AddPNode::make_off_heap(java_buffer, MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET))));
3191 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3192 Node* lease_constant = _gvn.intcon(4);
3193
3194 // And flags with lease constant.
3195 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3196
3197 // Branch on lease to conditionalize returning the leased java buffer.
3198 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3199 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3200 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3201
3202 // False branch, not a lease.
3203 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3204
3205 // True branch, is lease.
3206 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3207 set_control(is_lease);
3208
3209 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3210 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3211 OptoRuntime::void_void_Type(),
3212 SharedRuntime::jfr_return_lease(),
3213 "return_lease", TypePtr::BOTTOM);
3214 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3215
3216 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3217 record_for_igvn(lease_compare_rgn);
3218 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3219 record_for_igvn(lease_compare_mem);
3220 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3221 record_for_igvn(lease_compare_io);
3222 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3223 record_for_igvn(lease_result_value);
3224
3225 // Update control and phi nodes.
3226 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3227 lease_compare_rgn->init_req(_false_path, not_lease);
3228
3229 lease_compare_mem->init_req(_true_path, reset_memory());
3230 lease_compare_mem->init_req(_false_path, commit_memory);
3231
3232 lease_compare_io->init_req(_true_path, i_o());
3233 lease_compare_io->init_req(_false_path, input_io_state);
3234
3235 lease_result_value->init_req(_true_path, _gvn.longcon(0)); // if the lease was returned, return 0L.
3236 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3237
3238 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3239 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3240 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3241 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3242
3243 // Update control and phi nodes.
3244 result_rgn->init_req(_true_path, is_notified);
3245 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3246
3247 result_mem->init_req(_true_path, notified_reset_memory);
3248 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3249
3250 result_io->init_req(_true_path, input_io_state);
3251 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3252
3253 result_value->init_req(_true_path, current_pos);
3254 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3255
3256 // Set output state.
3257 set_control(_gvn.transform(result_rgn));
3258 set_all_memory(_gvn.transform(result_mem));
3259 set_i_o(_gvn.transform(result_io));
3260 set_result(result_rgn, result_value);
3261 return true;
3262 }
3263
3264 /*
3265 * The intrinsic is a model of this pseudo-code:
3266 *
3267 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3268 * jobject h_event_writer = tl->java_event_writer();
3269 * if (h_event_writer == nullptr) {
3270 * return nullptr;
3271 * }
3272 * oop threadObj = Thread::threadObj();
3273 * oop vthread = java_lang_Thread::vthread(threadObj);
3274 * traceid tid;
3275 * bool pinVirtualThread;
3276 * bool excluded;
3277 * if (vthread != threadObj) { // i.e. current thread is virtual
3278 * tid = java_lang_Thread::tid(vthread);
3279 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3280 * pinVirtualThread = VMContinuations;
3281 * excluded = vthread_epoch_raw & excluded_mask;
3282 * if (!excluded) {
3283 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3284 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3285 * if (vthread_epoch != current_epoch) {
3286 * write_checkpoint();
3287 * }
3288 * }
3289 * } else {
3290 * tid = java_lang_Thread::tid(threadObj);
3291 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3292 * pinVirtualThread = false;
3293 * excluded = thread_epoch_raw & excluded_mask;
3294 * }
3295 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3296 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3297 * if (tid_in_event_writer != tid) {
3298 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3299 * setField(event_writer, "excluded", excluded);
3300 * setField(event_writer, "threadID", tid);
3301 * }
3302 * return event_writer
3303 */
3304 bool LibraryCallKit::inline_native_getEventWriter() {
3305 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3306
3307 // Save input memory and i_o state.
3308 Node* input_memory_state = reset_memory();
3309 set_all_memory(input_memory_state);
3310 Node* input_io_state = i_o();
3311
3312 // The most significant bit of the u2 is used to denote thread exclusion
3313 Node* excluded_shift = _gvn.intcon(15);
3314 Node* excluded_mask = _gvn.intcon(1 << 15);
3315 // The epoch generation is the range [1-32767]
3316 Node* epoch_mask = _gvn.intcon(32767);
3317
3318 // TLS
3319 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3320
3321 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3322 Node* jobj_ptr = off_heap_plus_addr(tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3323
3324 // Load the eventwriter jobject handle.
3325 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3326
3327 // Null check the jobject handle.
3328 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3329 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3330 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3331
3332 // False path, jobj is null.
3333 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3334
3335 // True path, jobj is not null.
3336 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3337
3338 set_control(jobj_is_not_null);
3339
3340 // Load the threadObj for the CarrierThread.
3341 Node* threadObj = generate_current_thread(tls_ptr);
3342
3343 // Load the vthread.
3344 Node* vthread = generate_virtual_thread(tls_ptr);
3345
3346 // If vthread != threadObj, this is a virtual thread.
3347 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3348 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3349 IfNode* iff_vthread_not_equal_threadObj =
3350 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3351
3352 // False branch, fallback to threadObj.
3353 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3354 set_control(vthread_equal_threadObj);
3355
3356 // Load the tid field from the vthread object.
3357 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3358
3359 // Load the raw epoch value from the threadObj.
3360 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3361 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3362 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3363 TypeInt::CHAR, T_CHAR,
3364 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3365
3366 // Mask off the excluded information from the epoch.
3367 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3368
3369 // True branch, this is a virtual thread.
3370 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3371 set_control(vthread_not_equal_threadObj);
3372
3373 // Load the tid field from the vthread object.
3374 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3375
3376 // Continuation support determines if a virtual thread should be pinned.
3377 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3378 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3379
3380 // Load the raw epoch value from the vthread.
3381 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3382 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3383 TypeInt::CHAR, T_CHAR,
3384 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3385
3386 // Mask off the excluded information from the epoch.
3387 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, excluded_mask));
3388
3389 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3390 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, excluded_mask));
3391 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3392 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3393
3394 // False branch, vthread is excluded, no need to write epoch info.
3395 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3396
3397 // True branch, vthread is included, update epoch info.
3398 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3399 set_control(included);
3400
3401 // Get epoch value.
3402 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, epoch_mask));
3403
3404 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3405 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3406 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3407
3408 // Compare the epoch in the vthread to the current epoch generation.
3409 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3410 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3411 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3412
3413 // False path, epoch is equal, checkpoint information is valid.
3414 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3415
3416 // True path, epoch is not equal, write a checkpoint for the vthread.
3417 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3418
3419 set_control(epoch_is_not_equal);
3420
3421 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3422 // The call also updates the native thread local thread id and the vthread with the current epoch.
3423 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3424 OptoRuntime::jfr_write_checkpoint_Type(),
3425 SharedRuntime::jfr_write_checkpoint(),
3426 "write_checkpoint", TypePtr::BOTTOM);
3427 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3428
3429 // vthread epoch != current epoch
3430 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3431 record_for_igvn(epoch_compare_rgn);
3432 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3433 record_for_igvn(epoch_compare_mem);
3434 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3435 record_for_igvn(epoch_compare_io);
3436
3437 // Update control and phi nodes.
3438 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3439 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3440 epoch_compare_mem->init_req(_true_path, reset_memory());
3441 epoch_compare_mem->init_req(_false_path, input_memory_state);
3442 epoch_compare_io->init_req(_true_path, i_o());
3443 epoch_compare_io->init_req(_false_path, input_io_state);
3444
3445 // excluded != true
3446 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3447 record_for_igvn(exclude_compare_rgn);
3448 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3449 record_for_igvn(exclude_compare_mem);
3450 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3451 record_for_igvn(exclude_compare_io);
3452
3453 // Update control and phi nodes.
3454 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3455 exclude_compare_rgn->init_req(_false_path, excluded);
3456 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3457 exclude_compare_mem->init_req(_false_path, input_memory_state);
3458 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3459 exclude_compare_io->init_req(_false_path, input_io_state);
3460
3461 // vthread != threadObj
3462 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3463 record_for_igvn(vthread_compare_rgn);
3464 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3465 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3466 record_for_igvn(vthread_compare_io);
3467 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3468 record_for_igvn(tid);
3469 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3470 record_for_igvn(exclusion);
3471 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3472 record_for_igvn(pinVirtualThread);
3473
3474 // Update control and phi nodes.
3475 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3476 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3477 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3478 vthread_compare_mem->init_req(_false_path, input_memory_state);
3479 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3480 vthread_compare_io->init_req(_false_path, input_io_state);
3481 tid->init_req(_true_path, vthread_tid);
3482 tid->init_req(_false_path, thread_obj_tid);
3483 exclusion->init_req(_true_path, vthread_is_excluded);
3484 exclusion->init_req(_false_path, threadObj_is_excluded);
3485 pinVirtualThread->init_req(_true_path, continuation_support);
3486 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3487
3488 // Update branch state.
3489 set_control(_gvn.transform(vthread_compare_rgn));
3490 set_all_memory(_gvn.transform(vthread_compare_mem));
3491 set_i_o(_gvn.transform(vthread_compare_io));
3492
3493 // Load the event writer oop by dereferencing the jobject handle.
3494 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3495 assert(klass_EventWriter->is_loaded(), "invariant");
3496 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3497 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3498 const TypeOopPtr* const xtype = aklass->as_instance_type();
3499 Node* jobj_untagged = _gvn.transform(AddPNode::make_off_heap(jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3500 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3501
3502 // Load the current thread id from the event writer object.
3503 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3504 // Get the field offset to, conditionally, store an updated tid value later.
3505 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3506 // Get the field offset to, conditionally, store an updated exclusion value later.
3507 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3508 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3509 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3510
3511 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3512 record_for_igvn(event_writer_tid_compare_rgn);
3513 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3514 record_for_igvn(event_writer_tid_compare_mem);
3515 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3516 record_for_igvn(event_writer_tid_compare_io);
3517
3518 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3519 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3520 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3521 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3522
3523 // False path, tids are the same.
3524 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3525
3526 // True path, tid is not equal, need to update the tid in the event writer.
3527 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3528 record_for_igvn(tid_is_not_equal);
3529
3530 // Store the pin state to the event writer.
3531 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3532
3533 // Store the exclusion state to the event writer.
3534 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3535 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3536
3537 // Store the tid to the event writer.
3538 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3539
3540 // Update control and phi nodes.
3541 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3542 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3543 event_writer_tid_compare_mem->init_req(_true_path, reset_memory());
3544 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3545 event_writer_tid_compare_io->init_req(_true_path, i_o());
3546 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3547
3548 // Result of top level CFG, Memory, IO and Value.
3549 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3550 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3551 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3552 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3553
3554 // Result control.
3555 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3556 result_rgn->init_req(_false_path, jobj_is_null);
3557
3558 // Result memory.
3559 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3560 result_mem->init_req(_false_path, input_memory_state);
3561
3562 // Result IO.
3563 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3564 result_io->init_req(_false_path, input_io_state);
3565
3566 // Result value.
3567 result_value->init_req(_true_path, event_writer); // return event writer oop
3568 result_value->init_req(_false_path, null()); // return null
3569
3570 // Set output state.
3571 set_control(_gvn.transform(result_rgn));
3572 set_all_memory(_gvn.transform(result_mem));
3573 set_i_o(_gvn.transform(result_io));
3574 set_result(result_rgn, result_value);
3575 return true;
3576 }
3577
3578 /*
3579 * The intrinsic is a model of this pseudo-code:
3580 *
3581 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3582 * if (carrierThread != thread) { // is virtual thread
3583 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3584 * bool excluded = vthread_epoch_raw & excluded_mask;
3585 * AtomicAccess::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3586 * AtomicAccess::store(&tl->_contextual_thread_excluded, is_excluded);
3587 * if (!excluded) {
3588 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3589 * AtomicAccess::store(&tl->_vthread_epoch, vthread_epoch);
3590 * }
3591 * AtomicAccess::release_store(&tl->_vthread, true);
3592 * return;
3593 * }
3594 * AtomicAccess::release_store(&tl->_vthread, false);
3595 */
3596 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3597 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3598
3599 Node* input_memory_state = reset_memory();
3600 set_all_memory(input_memory_state);
3601
3602 // The most significant bit of the u2 is used to denote thread exclusion
3603 Node* excluded_mask = _gvn.intcon(1 << 15);
3604 // The epoch generation is the range [1-32767]
3605 Node* epoch_mask = _gvn.intcon(32767);
3606
3607 Node* const carrierThread = generate_current_thread(jt);
3608 // If thread != carrierThread, this is a virtual thread.
3609 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3610 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3611 IfNode* iff_thread_not_equal_carrierThread =
3612 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3613
3614 Node* vthread_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3615
3616 // False branch, is carrierThread.
3617 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3618 // Store release
3619 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3620
3621 set_all_memory(input_memory_state);
3622
3623 // True branch, is virtual thread.
3624 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3625 set_control(thread_not_equal_carrierThread);
3626
3627 // Load the raw epoch value from the vthread.
3628 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3629 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
3630 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3631
3632 // Mask off the excluded information from the epoch.
3633 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, excluded_mask));
3634
3635 // Load the tid field from the thread.
3636 Node* tid = load_field_from_object(thread, "tid", "J");
3637
3638 // Store the vthread tid to the jfr thread local.
3639 Node* thread_id_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3640 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
3641
3642 // Branch is_excluded to conditionalize updating the epoch .
3643 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, excluded_mask));
3644 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3645 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3646
3647 // True branch, vthread is excluded, no need to write epoch info.
3648 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3649 set_control(excluded);
3650 Node* vthread_is_excluded = _gvn.intcon(1);
3651
3652 // False branch, vthread is included, update epoch info.
3653 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3654 set_control(included);
3655 Node* vthread_is_included = _gvn.intcon(0);
3656
3657 // Get epoch value.
3658 Node* epoch = _gvn.transform(new AndINode(epoch_raw, epoch_mask));
3659
3660 // Store the vthread epoch to the jfr thread local.
3661 Node* vthread_epoch_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3662 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
3663
3664 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3665 record_for_igvn(excluded_rgn);
3666 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3667 record_for_igvn(excluded_mem);
3668 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3669 record_for_igvn(exclusion);
3670
3671 // Merge the excluded control and memory.
3672 excluded_rgn->init_req(_true_path, excluded);
3673 excluded_rgn->init_req(_false_path, included);
3674 excluded_mem->init_req(_true_path, tid_memory);
3675 excluded_mem->init_req(_false_path, included_memory);
3676 exclusion->init_req(_true_path, vthread_is_excluded);
3677 exclusion->init_req(_false_path, vthread_is_included);
3678
3679 // Set intermediate state.
3680 set_control(_gvn.transform(excluded_rgn));
3681 set_all_memory(excluded_mem);
3682
3683 // Store the vthread exclusion state to the jfr thread local.
3684 Node* thread_local_excluded_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3685 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
3686
3687 // Store release
3688 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
3689
3690 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3691 record_for_igvn(thread_compare_rgn);
3692 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3693 record_for_igvn(thread_compare_mem);
3694 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3695 record_for_igvn(vthread);
3696
3697 // Merge the thread_compare control and memory.
3698 thread_compare_rgn->init_req(_true_path, control());
3699 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3700 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3701 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3702
3703 // Set output state.
3704 set_control(_gvn.transform(thread_compare_rgn));
3705 set_all_memory(_gvn.transform(thread_compare_mem));
3706 }
3707
3708 //------------------------inline_native_try_update_epoch------------------
3709 //
3710 // The generated code is a function of the argument type.
3711 //
3712 bool LibraryCallKit::inline_native_try_update_epoch() {
3713 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3714
3715 // Save input memory.
3716 Node* input_memory_state = reset_memory();
3717 set_all_memory(input_memory_state);
3718
3719 // Argument is an oop whose class has an injected instance field,
3720 // called 'jfr_epoch' of type T_INT, used for holding a jfr epoch value.
3721 Node* oop = argument(0);
3722 const TypeInstPtr* tinst = _gvn.type(oop)->isa_instptr();
3723 assert(tinst != nullptr, "oop is null");
3724 assert(tinst->is_loaded(), "klass is not loaded");
3725 ciInstanceKlass* const ik = tinst->instance_klass();
3726
3727 ciField* const field = ik->get_injected_instance_field_by_name(ciSymbol::make("jfr_epoch"),
3728 ciSymbol::make("I"));
3729
3730 assert(field != nullptr, "field 'jfr_epoch' of type I not injected in klass %s", ik->name()->as_utf8());
3731
3732 const int jfr_epoch_field_offset = field->offset_in_bytes();
3733 Node* oop_epoch_field_offset = basic_plus_adr(oop, jfr_epoch_field_offset);
3734 const TypePtr* adr_type = _gvn.type(oop_epoch_field_offset)->isa_ptr();
3735 const int alias_idx = C->get_alias_index(adr_type);
3736 BasicType bt = field->layout_type();
3737 const Type * oop_epoch_field_type = Type::get_const_basic_type(bt);
3738
3739 // Load the epoch value from the oop.
3740 Node* oop_epoch = access_load_at(oop,
3741 oop_epoch_field_offset,
3742 adr_type, oop_epoch_field_type,
3743 bt, IN_HEAP | MO_UNORDERED);
3744
3745 // Load the current JFR epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3746 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3747 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3748
3749 // Compare the epoch in the oop against the current JFR epoch generation.
3750 Node* const epochs_cmp = _gvn.transform(new CmpINode(current_epoch_generation, oop_epoch));
3751 Node* epochs_equal_test = _gvn.transform(new BoolNode(epochs_cmp, BoolTest::eq));
3752 IfNode* iff_epochs_equal = create_and_map_if(control(), epochs_equal_test, PROB_LIKELY(0.999), COUNT_UNKNOWN);
3753
3754 // True path.
3755 Node* epochs_are_equal = _gvn.transform(new IfTrueNode(iff_epochs_equal));
3756
3757 // False path.
3758 Node* epochs_are_not_equal = _gvn.transform(new IfFalseNode(iff_epochs_equal));
3759
3760 set_control(_gvn.transform(epochs_are_not_equal));
3761
3762 // Attempt to cas the current JFR epoch generation into the oop epoch field.
3763 DecoratorSet decorators = IN_HEAP;
3764 decorators |= mo_decorator_for_access_kind(Volatile);
3765
3766 Node* result = access_atomic_cmpxchg_val_at(oop,
3767 oop_epoch_field_offset,
3768 adr_type, alias_idx,
3769 oop_epoch, // expected value
3770 current_epoch_generation, // new value
3771 oop_epoch_field_type,
3772 bt,
3773 decorators);
3774
3775 // Compare the result of the cas operation to the expected value.
3776 Node* const cas_cmp_to_expected_value = _gvn.transform(new CmpINode(result, oop_epoch));
3777 Node* cas_operation_test = _gvn.transform(new BoolNode(cas_cmp_to_expected_value, BoolTest::eq));
3778 IfNode* iff_cas_success = create_and_map_if(control(), cas_operation_test, PROB_LIKELY(0.999), COUNT_UNKNOWN);
3779
3780 // True path.
3781 Node* cas_success = _gvn.transform(new IfTrueNode(iff_cas_success));
3782
3783 // False path.
3784 Node* cas_failure = _gvn.transform(new IfFalseNode(iff_cas_success));
3785
3786 // Cas result region and phi nodes.
3787 RegionNode* cas_operation_rgn = new RegionNode(PATH_LIMIT);
3788 record_for_igvn(cas_operation_rgn);
3789 PhiNode* cas_operation_mem = new PhiNode(cas_operation_rgn, Type::MEMORY, TypePtr::BOTTOM);
3790 record_for_igvn(cas_operation_mem);
3791 PhiNode* cas_result = new PhiNode(cas_operation_rgn, TypeInt::BOOL);
3792 record_for_igvn(cas_result);
3793
3794 cas_operation_rgn->init_req(_true_path, _gvn.transform(cas_success));
3795 cas_operation_rgn->init_req(_false_path, _gvn.transform(cas_failure));
3796 cas_operation_mem->init_req(_true_path, reset_memory());
3797 cas_operation_mem->init_req(_false_path, input_memory_state);
3798 cas_result->init_req(_true_path, _gvn.intcon(1));
3799 cas_result->init_req(_false_path, _gvn.intcon(0));
3800
3801 // Epoch compare region and phi nodes.
3802 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3803 record_for_igvn(epoch_compare_rgn);
3804 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3805 record_for_igvn(epoch_compare_mem);
3806 PhiNode* result_value = new PhiNode(epoch_compare_rgn, TypeInt::BOOL);
3807 record_for_igvn(result_value);
3808
3809 epoch_compare_rgn->init_req(_true_path, _gvn.transform(epochs_are_equal));
3810 epoch_compare_rgn->init_req(_false_path, _gvn.transform(cas_operation_rgn));
3811 epoch_compare_mem->init_req(_true_path, _gvn.transform(input_memory_state));
3812 epoch_compare_mem->init_req(_false_path, _gvn.transform(cas_operation_mem));
3813 result_value->init_req(_true_path, _gvn.intcon(0));
3814 result_value->init_req(_false_path, _gvn.transform(cas_result));
3815
3816 // Set output state.
3817 set_result(epoch_compare_rgn, result_value);
3818 set_all_memory(_gvn.transform(epoch_compare_mem));
3819
3820 return true;
3821 }
3822
3823 #endif // JFR_HAVE_INTRINSICS
3824
3825 //------------------------inline_native_currentCarrierThread------------------
3826 bool LibraryCallKit::inline_native_currentCarrierThread() {
3827 Node* junk = nullptr;
3828 set_result(generate_current_thread(junk));
3829 return true;
3830 }
3831
3832 //------------------------inline_native_currentThread------------------
3833 bool LibraryCallKit::inline_native_currentThread() {
3834 Node* junk = nullptr;
3835 set_result(generate_virtual_thread(junk));
3836 return true;
3837 }
3838
3839 //------------------------inline_native_setVthread------------------
3840 bool LibraryCallKit::inline_native_setCurrentThread() {
3841 assert(C->method()->changes_current_thread(),
3842 "method changes current Thread but is not annotated ChangesCurrentThread");
3843 Node* arr = argument(1);
3844 Node* thread = _gvn.transform(new ThreadLocalNode());
3845 Node* p = off_heap_plus_addr(thread, in_bytes(JavaThread::vthread_offset()));
3846 Node* thread_obj_handle
3847 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3848 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3849 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3850
3851 // Change the _monitor_owner_id of the JavaThread
3852 Node* tid = load_field_from_object(arr, "tid", "J");
3853 Node* monitor_owner_id_offset = off_heap_plus_addr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
3854 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
3855
3856 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3857 return true;
3858 }
3859
3860 const Type* LibraryCallKit::scopedValueCache_type() {
3861 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3862 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3863 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3864
3865 // Because we create the scopedValue cache lazily we have to make the
3866 // type of the result BotPTR.
3867 bool xk = etype->klass_is_exact();
3868 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3869 return objects_type;
3870 }
3871
3872 Node* LibraryCallKit::scopedValueCache_helper() {
3873 Node* thread = _gvn.transform(new ThreadLocalNode());
3874 Node* p = off_heap_plus_addr(thread, in_bytes(JavaThread::scopedValueCache_offset()));
3875 // We cannot use immutable_memory() because we might flip onto a
3876 // different carrier thread, at which point we'll need to use that
3877 // carrier thread's cache.
3878 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3879 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3880 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3881 }
3882
3883 //------------------------inline_native_scopedValueCache------------------
3884 bool LibraryCallKit::inline_native_scopedValueCache() {
3885 Node* cache_obj_handle = scopedValueCache_helper();
3886 const Type* objects_type = scopedValueCache_type();
3887 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3888
3889 return true;
3890 }
3891
3892 //------------------------inline_native_setScopedValueCache------------------
3893 bool LibraryCallKit::inline_native_setScopedValueCache() {
3894 Node* arr = argument(0);
3895 Node* cache_obj_handle = scopedValueCache_helper();
3896 const Type* objects_type = scopedValueCache_type();
3897
3898 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3899 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3900
3901 return true;
3902 }
3903
3904 //------------------------inline_native_Continuation_pin and unpin-----------
3905
3906 // Shared implementation routine for both pin and unpin.
3907 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
3908 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3909
3910 // Save input memory.
3911 Node* input_memory_state = reset_memory();
3912 set_all_memory(input_memory_state);
3913
3914 // TLS
3915 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3916 Node* last_continuation_offset = off_heap_plus_addr(tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
3917 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
3918
3919 // Null check the last continuation object.
3920 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
3921 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
3922 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3923
3924 // False path, last continuation is null.
3925 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
3926
3927 // True path, last continuation is not null.
3928 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
3929
3930 set_control(continuation_is_not_null);
3931
3932 // Load the pin count from the last continuation.
3933 Node* pin_count_offset = off_heap_plus_addr(last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
3934 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
3935
3936 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
3937 Node* pin_count_rhs;
3938 if (unpin) {
3939 pin_count_rhs = _gvn.intcon(0);
3940 } else {
3941 pin_count_rhs = _gvn.intcon(UINT32_MAX);
3942 }
3943 Node* pin_count_cmp = _gvn.transform(new CmpUNode(pin_count, pin_count_rhs));
3944 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
3945 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
3946
3947 // True branch, pin count over/underflow.
3948 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
3949 {
3950 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
3951 // which will throw IllegalStateException for pin count over/underflow.
3952 // No memory changed so far - we can use memory create by reset_memory()
3953 // at the beginning of this intrinsic. No need to call reset_memory() again.
3954 PreserveJVMState pjvms(this);
3955 set_control(pin_count_over_underflow);
3956 uncommon_trap(Deoptimization::Reason_intrinsic,
3957 Deoptimization::Action_none);
3958 assert(stopped(), "invariant");
3959 }
3960
3961 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
3962 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
3963 set_control(valid_pin_count);
3964
3965 Node* next_pin_count;
3966 if (unpin) {
3967 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
3968 } else {
3969 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
3970 }
3971
3972 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
3973
3974 // Result of top level CFG and Memory.
3975 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3976 record_for_igvn(result_rgn);
3977 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3978 record_for_igvn(result_mem);
3979
3980 result_rgn->init_req(_true_path, valid_pin_count);
3981 result_rgn->init_req(_false_path, continuation_is_null);
3982 result_mem->init_req(_true_path, reset_memory());
3983 result_mem->init_req(_false_path, input_memory_state);
3984
3985 // Set output state.
3986 set_control(_gvn.transform(result_rgn));
3987 set_all_memory(_gvn.transform(result_mem));
3988
3989 return true;
3990 }
3991
3992 //---------------------------load_mirror_from_klass----------------------------
3993 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3994 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3995 Node* p = off_heap_plus_addr(klass, in_bytes(Klass::java_mirror_offset()));
3996 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3997 // mirror = ((OopHandle)mirror)->resolve();
3998 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3999 }
4000
4001 //-----------------------load_klass_from_mirror_common-------------------------
4002 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
4003 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
4004 // and branch to the given path on the region.
4005 // If never_see_null, take an uncommon trap on null, so we can optimistically
4006 // compile for the non-null case.
4007 // If the region is null, force never_see_null = true.
4008 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
4009 bool never_see_null,
4010 RegionNode* region,
4011 int null_path,
4012 int offset) {
4013 if (region == nullptr) never_see_null = true;
4014 Node* p = basic_plus_adr(mirror, offset);
4015 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4016 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
4017 Node* null_ctl = top();
4018 kls = null_check_oop(kls, &null_ctl, never_see_null);
4019 if (region != nullptr) {
4020 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
4021 region->init_req(null_path, null_ctl);
4022 } else {
4023 assert(null_ctl == top(), "no loose ends");
4024 }
4025 return kls;
4026 }
4027
4028 //--------------------(inline_native_Class_query helpers)---------------------
4029 // Use this for JVM_ACC_INTERFACE.
4030 // Fall through if (mods & mask) == bits, take the guard otherwise.
4031 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
4032 ByteSize offset, const Type* type, BasicType bt) {
4033 // Branch around if the given klass has the given modifier bit set.
4034 // Like generate_guard, adds a new path onto the region.
4035 Node* modp = off_heap_plus_addr(kls, in_bytes(offset));
4036 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
4037 Node* mask = intcon(modifier_mask);
4038 Node* bits = intcon(modifier_bits);
4039 Node* mbit = _gvn.transform(new AndINode(mods, mask));
4040 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
4041 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
4042 return generate_fair_guard(bol, region);
4043 }
4044 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
4045 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
4046 InstanceKlass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
4047 }
4048
4049 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
4050 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
4051 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
4052 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
4053 }
4054
4055 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
4056 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
4057 }
4058
4059 //-------------------------inline_native_Class_query-------------------
4060 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
4061 const Type* return_type = TypeInt::BOOL;
4062 Node* prim_return_value = top(); // what happens if it's a primitive class?
4063 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4064 bool expect_prim = false; // most of these guys expect to work on refs
4065
4066 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
4067
4068 Node* mirror = argument(0);
4069 Node* obj = top();
4070
4071 switch (id) {
4072 case vmIntrinsics::_isInstance:
4073 // nothing is an instance of a primitive type
4074 prim_return_value = intcon(0);
4075 obj = argument(1);
4076 break;
4077 case vmIntrinsics::_isHidden:
4078 prim_return_value = intcon(0);
4079 break;
4080 case vmIntrinsics::_getSuperclass:
4081 prim_return_value = null();
4082 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
4083 break;
4084 default:
4085 fatal_unexpected_iid(id);
4086 break;
4087 }
4088
4089 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4090 if (mirror_con == nullptr) return false; // cannot happen?
4091
4092 #ifndef PRODUCT
4093 if (C->print_intrinsics() || C->print_inlining()) {
4094 ciType* k = mirror_con->java_mirror_type();
4095 if (k) {
4096 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4097 k->print_name();
4098 tty->cr();
4099 }
4100 }
4101 #endif
4102
4103 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4104 RegionNode* region = new RegionNode(PATH_LIMIT);
4105 record_for_igvn(region);
4106 PhiNode* phi = new PhiNode(region, return_type);
4107
4108 // The mirror will never be null of Reflection.getClassAccessFlags, however
4109 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4110 // if it is. See bug 4774291.
4111
4112 // For Reflection.getClassAccessFlags(), the null check occurs in
4113 // the wrong place; see inline_unsafe_access(), above, for a similar
4114 // situation.
4115 mirror = null_check(mirror);
4116 // If mirror or obj is dead, only null-path is taken.
4117 if (stopped()) return true;
4118
4119 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4120
4121 // Now load the mirror's klass metaobject, and null-check it.
4122 // Side-effects region with the control path if the klass is null.
4123 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4124 // If kls is null, we have a primitive mirror.
4125 phi->init_req(_prim_path, prim_return_value);
4126 if (stopped()) { set_result(region, phi); return true; }
4127 bool safe_for_replace = (region->in(_prim_path) == top());
4128
4129 Node* p; // handy temp
4130 Node* null_ctl;
4131
4132 // Now that we have the non-null klass, we can perform the real query.
4133 // For constant classes, the query will constant-fold in LoadNode::Value.
4134 Node* query_value = top();
4135 switch (id) {
4136 case vmIntrinsics::_isInstance:
4137 // nothing is an instance of a primitive type
4138 query_value = gen_instanceof(obj, kls, safe_for_replace);
4139 break;
4140
4141 case vmIntrinsics::_isHidden:
4142 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4143 if (generate_hidden_class_guard(kls, region) != nullptr)
4144 // A guard was added. If the guard is taken, it was an hidden class.
4145 phi->add_req(intcon(1));
4146 // If we fall through, it's a plain class.
4147 query_value = intcon(0);
4148 break;
4149
4150
4151 case vmIntrinsics::_getSuperclass:
4152 // The rules here are somewhat unfortunate, but we can still do better
4153 // with random logic than with a JNI call.
4154 // Interfaces store null or Object as _super, but must report null.
4155 // Arrays store an intermediate super as _super, but must report Object.
4156 // Other types can report the actual _super.
4157 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4158 if (generate_array_guard(kls, region) != nullptr) {
4159 // A guard was added. If the guard is taken, it was an array.
4160 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4161 }
4162 // Check for interface after array since this checks AccessFlags offset into InstanceKlass.
4163 // In other words, we are accessing subtype-specific information, so we need to determine the subtype first.
4164 if (generate_interface_guard(kls, region) != nullptr) {
4165 // A guard was added. If the guard is taken, it was an interface.
4166 phi->add_req(null());
4167 }
4168 // If we fall through, it's a plain class. Get its _super.
4169 p = off_heap_plus_addr(kls, in_bytes(Klass::super_offset()));
4170 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4171 null_ctl = top();
4172 kls = null_check_oop(kls, &null_ctl);
4173 if (null_ctl != top()) {
4174 // If the guard is taken, Object.superClass is null (both klass and mirror).
4175 region->add_req(null_ctl);
4176 phi ->add_req(null());
4177 }
4178 if (!stopped()) {
4179 query_value = load_mirror_from_klass(kls);
4180 }
4181 break;
4182
4183 default:
4184 fatal_unexpected_iid(id);
4185 break;
4186 }
4187
4188 // Fall-through is the normal case of a query to a real class.
4189 phi->init_req(1, query_value);
4190 region->init_req(1, control());
4191
4192 C->set_has_split_ifs(true); // Has chance for split-if optimization
4193 set_result(region, phi);
4194 return true;
4195 }
4196
4197 //-------------------------inline_Class_cast-------------------
4198 bool LibraryCallKit::inline_Class_cast() {
4199 Node* mirror = argument(0); // Class
4200 Node* obj = argument(1);
4201 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4202 if (mirror_con == nullptr) {
4203 return false; // dead path (mirror->is_top()).
4204 }
4205 if (obj == nullptr || obj->is_top()) {
4206 return false; // dead path
4207 }
4208 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4209
4210 // First, see if Class.cast() can be folded statically.
4211 // java_mirror_type() returns non-null for compile-time Class constants.
4212 ciType* tm = mirror_con->java_mirror_type();
4213 if (tm != nullptr && tm->is_klass() &&
4214 tp != nullptr) {
4215 if (!tp->is_loaded()) {
4216 // Don't use intrinsic when class is not loaded.
4217 return false;
4218 } else {
4219 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
4220 if (static_res == Compile::SSC_always_true) {
4221 // isInstance() is true - fold the code.
4222 set_result(obj);
4223 return true;
4224 } else if (static_res == Compile::SSC_always_false) {
4225 // Don't use intrinsic, have to throw ClassCastException.
4226 // If the reference is null, the non-intrinsic bytecode will
4227 // be optimized appropriately.
4228 return false;
4229 }
4230 }
4231 }
4232
4233 // Bailout intrinsic and do normal inlining if exception path is frequent.
4234 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4235 return false;
4236 }
4237
4238 // Generate dynamic checks.
4239 // Class.cast() is java implementation of _checkcast bytecode.
4240 // Do checkcast (Parse::do_checkcast()) optimizations here.
4241
4242 mirror = null_check(mirror);
4243 // If mirror is dead, only null-path is taken.
4244 if (stopped()) {
4245 return true;
4246 }
4247
4248 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive).
4249 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
4250 RegionNode* region = new RegionNode(PATH_LIMIT);
4251 record_for_igvn(region);
4252
4253 // Now load the mirror's klass metaobject, and null-check it.
4254 // If kls is null, we have a primitive mirror and
4255 // nothing is an instance of a primitive type.
4256 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4257
4258 Node* res = top();
4259 if (!stopped()) {
4260 Node* bad_type_ctrl = top();
4261 // Do checkcast optimizations.
4262 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4263 region->init_req(_bad_type_path, bad_type_ctrl);
4264 }
4265 if (region->in(_prim_path) != top() ||
4266 region->in(_bad_type_path) != top()) {
4267 // Let Interpreter throw ClassCastException.
4268 PreserveJVMState pjvms(this);
4269 set_control(_gvn.transform(region));
4270 uncommon_trap(Deoptimization::Reason_intrinsic,
4271 Deoptimization::Action_maybe_recompile);
4272 }
4273 if (!stopped()) {
4274 set_result(res);
4275 }
4276 return true;
4277 }
4278
4279
4280 //--------------------------inline_native_subtype_check------------------------
4281 // This intrinsic takes the JNI calls out of the heart of
4282 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4283 bool LibraryCallKit::inline_native_subtype_check() {
4284 // Pull both arguments off the stack.
4285 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4286 args[0] = argument(0);
4287 args[1] = argument(1);
4288 Node* klasses[2]; // corresponding Klasses: superk, subk
4289 klasses[0] = klasses[1] = top();
4290
4291 enum {
4292 // A full decision tree on {superc is prim, subc is prim}:
4293 _prim_0_path = 1, // {P,N} => false
4294 // {P,P} & superc!=subc => false
4295 _prim_same_path, // {P,P} & superc==subc => true
4296 _prim_1_path, // {N,P} => false
4297 _ref_subtype_path, // {N,N} & subtype check wins => true
4298 _both_ref_path, // {N,N} & subtype check loses => false
4299 PATH_LIMIT
4300 };
4301
4302 RegionNode* region = new RegionNode(PATH_LIMIT);
4303 Node* phi = new PhiNode(region, TypeInt::BOOL);
4304 record_for_igvn(region);
4305
4306 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4307 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4308 int class_klass_offset = java_lang_Class::klass_offset();
4309
4310 // First null-check both mirrors and load each mirror's klass metaobject.
4311 int which_arg;
4312 for (which_arg = 0; which_arg <= 1; which_arg++) {
4313 Node* arg = args[which_arg];
4314 arg = null_check(arg);
4315 if (stopped()) break;
4316 args[which_arg] = arg;
4317
4318 Node* p = basic_plus_adr(arg, class_klass_offset);
4319 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4320 klasses[which_arg] = _gvn.transform(kls);
4321 }
4322
4323 // Having loaded both klasses, test each for null.
4324 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4325 for (which_arg = 0; which_arg <= 1; which_arg++) {
4326 Node* kls = klasses[which_arg];
4327 Node* null_ctl = top();
4328 kls = null_check_oop(kls, &null_ctl, never_see_null);
4329 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
4330 region->init_req(prim_path, null_ctl);
4331 if (stopped()) break;
4332 klasses[which_arg] = kls;
4333 }
4334
4335 if (!stopped()) {
4336 // now we have two reference types, in klasses[0..1]
4337 Node* subk = klasses[1]; // the argument to isAssignableFrom
4338 Node* superk = klasses[0]; // the receiver
4339 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4340 // now we have a successful reference subtype check
4341 region->set_req(_ref_subtype_path, control());
4342 }
4343
4344 // If both operands are primitive (both klasses null), then
4345 // we must return true when they are identical primitives.
4346 // It is convenient to test this after the first null klass check.
4347 set_control(region->in(_prim_0_path)); // go back to first null check
4348 if (!stopped()) {
4349 // Since superc is primitive, make a guard for the superc==subc case.
4350 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4351 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4352 generate_guard(bol_eq, region, PROB_FAIR);
4353 if (region->req() == PATH_LIMIT+1) {
4354 // A guard was added. If the added guard is taken, superc==subc.
4355 region->swap_edges(PATH_LIMIT, _prim_same_path);
4356 region->del_req(PATH_LIMIT);
4357 }
4358 region->set_req(_prim_0_path, control()); // Not equal after all.
4359 }
4360
4361 // these are the only paths that produce 'true':
4362 phi->set_req(_prim_same_path, intcon(1));
4363 phi->set_req(_ref_subtype_path, intcon(1));
4364
4365 // pull together the cases:
4366 assert(region->req() == PATH_LIMIT, "sane region");
4367 for (uint i = 1; i < region->req(); i++) {
4368 Node* ctl = region->in(i);
4369 if (ctl == nullptr || ctl == top()) {
4370 region->set_req(i, top());
4371 phi ->set_req(i, top());
4372 } else if (phi->in(i) == nullptr) {
4373 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4374 }
4375 }
4376
4377 set_control(_gvn.transform(region));
4378 set_result(_gvn.transform(phi));
4379 return true;
4380 }
4381
4382 //---------------------generate_array_guard_common------------------------
4383 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
4384 bool obj_array, bool not_array, Node** obj) {
4385
4386 if (stopped()) {
4387 return nullptr;
4388 }
4389
4390 // If obj_array/non_array==false/false:
4391 // Branch around if the given klass is in fact an array (either obj or prim).
4392 // If obj_array/non_array==false/true:
4393 // Branch around if the given klass is not an array klass of any kind.
4394 // If obj_array/non_array==true/true:
4395 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
4396 // If obj_array/non_array==true/false:
4397 // Branch around if the kls is an oop array (Object[] or subtype)
4398 //
4399 // Like generate_guard, adds a new path onto the region.
4400 jint layout_con = 0;
4401 Node* layout_val = get_layout_helper(kls, layout_con);
4402 if (layout_val == nullptr) {
4403 bool query = (obj_array
4404 ? Klass::layout_helper_is_objArray(layout_con)
4405 : Klass::layout_helper_is_array(layout_con));
4406 if (query == not_array) {
4407 return nullptr; // never a branch
4408 } else { // always a branch
4409 Node* always_branch = control();
4410 if (region != nullptr)
4411 region->add_req(always_branch);
4412 set_control(top());
4413 return always_branch;
4414 }
4415 }
4416 // Now test the correct condition.
4417 jint nval = (obj_array
4418 ? (jint)(Klass::_lh_array_tag_type_value
4419 << Klass::_lh_array_tag_shift)
4420 : Klass::_lh_neutral_value);
4421 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4422 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
4423 // invert the test if we are looking for a non-array
4424 if (not_array) btest = BoolTest(btest).negate();
4425 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4426 Node* ctrl = generate_fair_guard(bol, region);
4427 Node* is_array_ctrl = not_array ? control() : ctrl;
4428 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4429 // Keep track of the fact that 'obj' is an array to prevent
4430 // array specific accesses from floating above the guard.
4431 *obj = _gvn.transform(new CheckCastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4432 }
4433 return ctrl;
4434 }
4435
4436
4437 //-----------------------inline_native_newArray--------------------------
4438 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4439 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4440 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4441 Node* mirror;
4442 Node* count_val;
4443 if (uninitialized) {
4444 null_check_receiver();
4445 mirror = argument(1);
4446 count_val = argument(2);
4447 } else {
4448 mirror = argument(0);
4449 count_val = argument(1);
4450 }
4451
4452 mirror = null_check(mirror);
4453 // If mirror or obj is dead, only null-path is taken.
4454 if (stopped()) return true;
4455
4456 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4457 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4458 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4459 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4460 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4461
4462 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4463 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4464 result_reg, _slow_path);
4465 Node* normal_ctl = control();
4466 Node* no_array_ctl = result_reg->in(_slow_path);
4467
4468 // Generate code for the slow case. We make a call to newArray().
4469 set_control(no_array_ctl);
4470 if (!stopped()) {
4471 // Either the input type is void.class, or else the
4472 // array klass has not yet been cached. Either the
4473 // ensuing call will throw an exception, or else it
4474 // will cache the array klass for next time.
4475 PreserveJVMState pjvms(this);
4476 CallJavaNode* slow_call = nullptr;
4477 if (uninitialized) {
4478 // Generate optimized virtual call (holder class 'Unsafe' is final)
4479 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4480 } else {
4481 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4482 }
4483 Node* slow_result = set_results_for_java_call(slow_call);
4484 // this->control() comes from set_results_for_java_call
4485 result_reg->set_req(_slow_path, control());
4486 result_val->set_req(_slow_path, slow_result);
4487 result_io ->set_req(_slow_path, i_o());
4488 result_mem->set_req(_slow_path, reset_memory());
4489 }
4490
4491 set_control(normal_ctl);
4492 if (!stopped()) {
4493 // Normal case: The array type has been cached in the java.lang.Class.
4494 // The following call works fine even if the array type is polymorphic.
4495 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4496 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4497 result_reg->init_req(_normal_path, control());
4498 result_val->init_req(_normal_path, obj);
4499 result_io ->init_req(_normal_path, i_o());
4500 result_mem->init_req(_normal_path, reset_memory());
4501
4502 if (uninitialized) {
4503 // Mark the allocation so that zeroing is skipped
4504 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4505 alloc->maybe_set_complete(&_gvn);
4506 }
4507 }
4508
4509 // Return the combined state.
4510 set_i_o( _gvn.transform(result_io) );
4511 set_all_memory( _gvn.transform(result_mem));
4512
4513 C->set_has_split_ifs(true); // Has chance for split-if optimization
4514 set_result(result_reg, result_val);
4515 return true;
4516 }
4517
4518 //----------------------inline_native_getLength--------------------------
4519 // public static native int java.lang.reflect.Array.getLength(Object array);
4520 bool LibraryCallKit::inline_native_getLength() {
4521 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4522
4523 Node* array = null_check(argument(0));
4524 // If array is dead, only null-path is taken.
4525 if (stopped()) return true;
4526
4527 // Deoptimize if it is a non-array.
4528 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
4529
4530 if (non_array != nullptr) {
4531 PreserveJVMState pjvms(this);
4532 set_control(non_array);
4533 uncommon_trap(Deoptimization::Reason_intrinsic,
4534 Deoptimization::Action_maybe_recompile);
4535 }
4536
4537 // If control is dead, only non-array-path is taken.
4538 if (stopped()) return true;
4539
4540 // The works fine even if the array type is polymorphic.
4541 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4542 Node* result = load_array_length(array);
4543
4544 C->set_has_split_ifs(true); // Has chance for split-if optimization
4545 set_result(result);
4546 return true;
4547 }
4548
4549 //------------------------inline_array_copyOf----------------------------
4550 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4551 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4552 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4553 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4554
4555 // Get the arguments.
4556 Node* original = argument(0);
4557 Node* start = is_copyOfRange? argument(1): intcon(0);
4558 Node* end = is_copyOfRange? argument(2): argument(1);
4559 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4560
4561 Node* newcopy = nullptr;
4562
4563 // Set the original stack and the reexecute bit for the interpreter to reexecute
4564 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4565 { PreserveReexecuteState preexecs(this);
4566 jvms()->set_should_reexecute(true);
4567
4568 array_type_mirror = null_check(array_type_mirror);
4569 original = null_check(original);
4570
4571 // Check if a null path was taken unconditionally.
4572 if (stopped()) return true;
4573
4574 Node* orig_length = load_array_length(original);
4575
4576 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4577 klass_node = null_check(klass_node);
4578
4579 RegionNode* bailout = new RegionNode(1);
4580 record_for_igvn(bailout);
4581
4582 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4583 // Bail out if that is so.
4584 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4585 if (not_objArray != nullptr) {
4586 // Improve the klass node's type from the new optimistic assumption:
4587 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4588 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4589 Node* cast = new CastPPNode(control(), klass_node, akls);
4590 klass_node = _gvn.transform(cast);
4591 }
4592
4593 // Bail out if either start or end is negative.
4594 generate_negative_guard(start, bailout, &start);
4595 generate_negative_guard(end, bailout, &end);
4596
4597 Node* length = end;
4598 if (_gvn.type(start) != TypeInt::ZERO) {
4599 length = _gvn.transform(new SubINode(end, start));
4600 }
4601
4602 // Bail out if length is negative (i.e., if start > end).
4603 // Without this the new_array would throw
4604 // NegativeArraySizeException but IllegalArgumentException is what
4605 // should be thrown
4606 generate_negative_guard(length, bailout, &length);
4607
4608 // Bail out if start is larger than the original length
4609 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4610 generate_negative_guard(orig_tail, bailout, &orig_tail);
4611
4612 if (bailout->req() > 1) {
4613 PreserveJVMState pjvms(this);
4614 set_control(_gvn.transform(bailout));
4615 uncommon_trap(Deoptimization::Reason_intrinsic,
4616 Deoptimization::Action_maybe_recompile);
4617 }
4618
4619 if (!stopped()) {
4620 // How many elements will we copy from the original?
4621 // The answer is MinI(orig_tail, length).
4622 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
4623
4624 // Generate a direct call to the right arraycopy function(s).
4625 // We know the copy is disjoint but we might not know if the
4626 // oop stores need checking.
4627 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4628 // This will fail a store-check if x contains any non-nulls.
4629
4630 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4631 // loads/stores but it is legal only if we're sure the
4632 // Arrays.copyOf would succeed. So we need all input arguments
4633 // to the copyOf to be validated, including that the copy to the
4634 // new array won't trigger an ArrayStoreException. That subtype
4635 // check can be optimized if we know something on the type of
4636 // the input array from type speculation.
4637 if (_gvn.type(klass_node)->singleton()) {
4638 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4639 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4640
4641 int test = C->static_subtype_check(superk, subk);
4642 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4643 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4644 if (t_original->speculative_type() != nullptr) {
4645 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4646 }
4647 }
4648 }
4649
4650 bool validated = false;
4651 // Reason_class_check rather than Reason_intrinsic because we
4652 // want to intrinsify even if this traps.
4653 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4654 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4655
4656 if (not_subtype_ctrl != top()) {
4657 PreserveJVMState pjvms(this);
4658 set_control(not_subtype_ctrl);
4659 uncommon_trap(Deoptimization::Reason_class_check,
4660 Deoptimization::Action_make_not_entrant);
4661 assert(stopped(), "Should be stopped");
4662 }
4663 validated = true;
4664 }
4665
4666 if (!stopped()) {
4667 newcopy = new_array(klass_node, length, 0); // no arguments to push
4668
4669 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4670 load_object_klass(original), klass_node);
4671 if (!is_copyOfRange) {
4672 ac->set_copyof(validated);
4673 } else {
4674 ac->set_copyofrange(validated);
4675 }
4676 Node* n = _gvn.transform(ac);
4677 if (n == ac) {
4678 ac->connect_outputs(this);
4679 } else {
4680 assert(validated, "shouldn't transform if all arguments not validated");
4681 set_all_memory(n);
4682 }
4683 }
4684 }
4685 } // original reexecute is set back here
4686
4687 C->set_has_split_ifs(true); // Has chance for split-if optimization
4688 if (!stopped()) {
4689 set_result(newcopy);
4690 }
4691 return true;
4692 }
4693
4694
4695 //----------------------generate_virtual_guard---------------------------
4696 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4697 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4698 RegionNode* slow_region) {
4699 ciMethod* method = callee();
4700 int vtable_index = method->vtable_index();
4701 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4702 "bad index %d", vtable_index);
4703 // Get the Method* out of the appropriate vtable entry.
4704 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4705 vtable_index*vtableEntry::size_in_bytes() +
4706 in_bytes(vtableEntry::method_offset());
4707 Node* entry_addr = off_heap_plus_addr(obj_klass, entry_offset);
4708 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4709
4710 // Compare the target method with the expected method (e.g., Object.hashCode).
4711 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4712
4713 Node* native_call = makecon(native_call_addr);
4714 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4715 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4716
4717 return generate_slow_guard(test_native, slow_region);
4718 }
4719
4720 //-----------------------generate_method_call----------------------------
4721 // Use generate_method_call to make a slow-call to the real
4722 // method if the fast path fails. An alternative would be to
4723 // use a stub like OptoRuntime::slow_arraycopy_Java.
4724 // This only works for expanding the current library call,
4725 // not another intrinsic. (E.g., don't use this for making an
4726 // arraycopy call inside of the copyOf intrinsic.)
4727 CallJavaNode*
4728 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4729 // When compiling the intrinsic method itself, do not use this technique.
4730 guarantee(callee() != C->method(), "cannot make slow-call to self");
4731
4732 ciMethod* method = callee();
4733 // ensure the JVMS we have will be correct for this call
4734 guarantee(method_id == method->intrinsic_id(), "must match");
4735
4736 const TypeFunc* tf = TypeFunc::make(method);
4737 if (res_not_null) {
4738 assert(tf->return_type() == T_OBJECT, "");
4739 const TypeTuple* range = tf->range();
4740 const Type** fields = TypeTuple::fields(range->cnt());
4741 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4742 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4743 tf = TypeFunc::make(tf->domain(), new_range);
4744 }
4745 CallJavaNode* slow_call;
4746 if (is_static) {
4747 assert(!is_virtual, "");
4748 slow_call = new CallStaticJavaNode(C, tf,
4749 SharedRuntime::get_resolve_static_call_stub(), method);
4750 } else if (is_virtual) {
4751 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4752 int vtable_index = Method::invalid_vtable_index;
4753 if (UseInlineCaches) {
4754 // Suppress the vtable call
4755 } else {
4756 // hashCode and clone are not a miranda methods,
4757 // so the vtable index is fixed.
4758 // No need to use the linkResolver to get it.
4759 vtable_index = method->vtable_index();
4760 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4761 "bad index %d", vtable_index);
4762 }
4763 slow_call = new CallDynamicJavaNode(tf,
4764 SharedRuntime::get_resolve_virtual_call_stub(),
4765 method, vtable_index);
4766 } else { // neither virtual nor static: opt_virtual
4767 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4768 slow_call = new CallStaticJavaNode(C, tf,
4769 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4770 slow_call->set_optimized_virtual(true);
4771 }
4772 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4773 // To be able to issue a direct call (optimized virtual or virtual)
4774 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4775 // about the method being invoked should be attached to the call site to
4776 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4777 slow_call->set_override_symbolic_info(true);
4778 }
4779 set_arguments_for_java_call(slow_call);
4780 set_edges_for_java_call(slow_call);
4781 return slow_call;
4782 }
4783
4784
4785 /**
4786 * Build special case code for calls to hashCode on an object. This call may
4787 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4788 * slightly different code.
4789 */
4790 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4791 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4792 assert(!(is_virtual && is_static), "either virtual, special, or static");
4793
4794 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4795
4796 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4797 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4798 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4799 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4800 Node* obj = nullptr;
4801 if (!is_static) {
4802 // Check for hashing null object
4803 obj = null_check_receiver();
4804 if (stopped()) return true; // unconditionally null
4805 result_reg->init_req(_null_path, top());
4806 result_val->init_req(_null_path, top());
4807 } else {
4808 // Do a null check, and return zero if null.
4809 // System.identityHashCode(null) == 0
4810 obj = argument(0);
4811 Node* null_ctl = top();
4812 obj = null_check_oop(obj, &null_ctl);
4813 result_reg->init_req(_null_path, null_ctl);
4814 result_val->init_req(_null_path, _gvn.intcon(0));
4815 }
4816
4817 // Unconditionally null? Then return right away.
4818 if (stopped()) {
4819 set_control( result_reg->in(_null_path));
4820 if (!stopped())
4821 set_result(result_val->in(_null_path));
4822 return true;
4823 }
4824
4825 // We only go to the fast case code if we pass a number of guards. The
4826 // paths which do not pass are accumulated in the slow_region.
4827 RegionNode* slow_region = new RegionNode(1);
4828 record_for_igvn(slow_region);
4829
4830 // If this is a virtual call, we generate a funny guard. We pull out
4831 // the vtable entry corresponding to hashCode() from the target object.
4832 // If the target method which we are calling happens to be the native
4833 // Object hashCode() method, we pass the guard. We do not need this
4834 // guard for non-virtual calls -- the caller is known to be the native
4835 // Object hashCode().
4836 if (is_virtual) {
4837 // After null check, get the object's klass.
4838 Node* obj_klass = load_object_klass(obj);
4839 generate_virtual_guard(obj_klass, slow_region);
4840 }
4841
4842 // Get the header out of the object, use LoadMarkNode when available
4843 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4844 // The control of the load must be null. Otherwise, the load can move before
4845 // the null check after castPP removal.
4846 Node* no_ctrl = nullptr;
4847 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4848
4849 if (!UseObjectMonitorTable) {
4850 // Test the header to see if it is safe to read w.r.t. locking.
4851 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4852 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4853 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
4854 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
4855 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
4856
4857 generate_slow_guard(test_monitor, slow_region);
4858 }
4859
4860 // Get the hash value and check to see that it has been properly assigned.
4861 // We depend on hash_mask being at most 32 bits and avoid the use of
4862 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4863 // vm: see markWord.hpp.
4864 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4865 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4866 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4867 // This hack lets the hash bits live anywhere in the mark object now, as long
4868 // as the shift drops the relevant bits into the low 32 bits. Note that
4869 // Java spec says that HashCode is an int so there's no point in capturing
4870 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4871 hshifted_header = ConvX2I(hshifted_header);
4872 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4873
4874 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4875 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4876 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4877
4878 generate_slow_guard(test_assigned, slow_region);
4879
4880 Node* init_mem = reset_memory();
4881 // fill in the rest of the null path:
4882 result_io ->init_req(_null_path, i_o());
4883 result_mem->init_req(_null_path, init_mem);
4884
4885 result_val->init_req(_fast_path, hash_val);
4886 result_reg->init_req(_fast_path, control());
4887 result_io ->init_req(_fast_path, i_o());
4888 result_mem->init_req(_fast_path, init_mem);
4889
4890 // Generate code for the slow case. We make a call to hashCode().
4891 set_control(_gvn.transform(slow_region));
4892 if (!stopped()) {
4893 // No need for PreserveJVMState, because we're using up the present state.
4894 set_all_memory(init_mem);
4895 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4896 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4897 Node* slow_result = set_results_for_java_call(slow_call);
4898 // this->control() comes from set_results_for_java_call
4899 result_reg->init_req(_slow_path, control());
4900 result_val->init_req(_slow_path, slow_result);
4901 result_io ->set_req(_slow_path, i_o());
4902 result_mem ->set_req(_slow_path, reset_memory());
4903 }
4904
4905 // Return the combined state.
4906 set_i_o( _gvn.transform(result_io) );
4907 set_all_memory( _gvn.transform(result_mem));
4908
4909 set_result(result_reg, result_val);
4910 return true;
4911 }
4912
4913 //---------------------------inline_native_getClass----------------------------
4914 // public final native Class<?> java.lang.Object.getClass();
4915 //
4916 // Build special case code for calls to getClass on an object.
4917 bool LibraryCallKit::inline_native_getClass() {
4918 Node* obj = null_check_receiver();
4919 if (stopped()) return true;
4920 set_result(load_mirror_from_klass(load_object_klass(obj)));
4921 return true;
4922 }
4923
4924 //-----------------inline_native_Reflection_getCallerClass---------------------
4925 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4926 //
4927 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4928 //
4929 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4930 // in that it must skip particular security frames and checks for
4931 // caller sensitive methods.
4932 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4933 #ifndef PRODUCT
4934 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4935 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4936 }
4937 #endif
4938
4939 if (!jvms()->has_method()) {
4940 #ifndef PRODUCT
4941 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4942 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4943 }
4944 #endif
4945 return false;
4946 }
4947
4948 // Walk back up the JVM state to find the caller at the required
4949 // depth.
4950 JVMState* caller_jvms = jvms();
4951
4952 // Cf. JVM_GetCallerClass
4953 // NOTE: Start the loop at depth 1 because the current JVM state does
4954 // not include the Reflection.getCallerClass() frame.
4955 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4956 ciMethod* m = caller_jvms->method();
4957 switch (n) {
4958 case 0:
4959 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4960 break;
4961 case 1:
4962 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4963 if (!m->caller_sensitive()) {
4964 #ifndef PRODUCT
4965 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4966 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4967 }
4968 #endif
4969 return false; // bail-out; let JVM_GetCallerClass do the work
4970 }
4971 break;
4972 default:
4973 if (!m->is_ignored_by_security_stack_walk()) {
4974 // We have reached the desired frame; return the holder class.
4975 // Acquire method holder as java.lang.Class and push as constant.
4976 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4977 ciInstance* caller_mirror = caller_klass->java_mirror();
4978 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4979
4980 #ifndef PRODUCT
4981 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4982 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());
4983 tty->print_cr(" JVM state at this point:");
4984 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4985 ciMethod* m = jvms()->of_depth(i)->method();
4986 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4987 }
4988 }
4989 #endif
4990 return true;
4991 }
4992 break;
4993 }
4994 }
4995
4996 #ifndef PRODUCT
4997 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4998 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4999 tty->print_cr(" JVM state at this point:");
5000 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5001 ciMethod* m = jvms()->of_depth(i)->method();
5002 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5003 }
5004 }
5005 #endif
5006
5007 return false; // bail-out; let JVM_GetCallerClass do the work
5008 }
5009
5010 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
5011 Node* arg = argument(0);
5012 Node* result = nullptr;
5013
5014 switch (id) {
5015 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
5016 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
5017 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5018 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5019 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5020 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5021
5022 case vmIntrinsics::_doubleToLongBits: {
5023 // two paths (plus control) merge in a wood
5024 RegionNode *r = new RegionNode(3);
5025 Node *phi = new PhiNode(r, TypeLong::LONG);
5026
5027 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5028 // Build the boolean node
5029 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5030
5031 // Branch either way.
5032 // NaN case is less traveled, which makes all the difference.
5033 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5034 Node *opt_isnan = _gvn.transform(ifisnan);
5035 assert( opt_isnan->is_If(), "Expect an IfNode");
5036 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5037 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5038
5039 set_control(iftrue);
5040
5041 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5042 Node *slow_result = longcon(nan_bits); // return NaN
5043 phi->init_req(1, _gvn.transform( slow_result ));
5044 r->init_req(1, iftrue);
5045
5046 // Else fall through
5047 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5048 set_control(iffalse);
5049
5050 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5051 r->init_req(2, iffalse);
5052
5053 // Post merge
5054 set_control(_gvn.transform(r));
5055 record_for_igvn(r);
5056
5057 C->set_has_split_ifs(true); // Has chance for split-if optimization
5058 result = phi;
5059 assert(result->bottom_type()->isa_long(), "must be");
5060 break;
5061 }
5062
5063 case vmIntrinsics::_floatToIntBits: {
5064 // two paths (plus control) merge in a wood
5065 RegionNode *r = new RegionNode(3);
5066 Node *phi = new PhiNode(r, TypeInt::INT);
5067
5068 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5069 // Build the boolean node
5070 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5071
5072 // Branch either way.
5073 // NaN case is less traveled, which makes all the difference.
5074 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5075 Node *opt_isnan = _gvn.transform(ifisnan);
5076 assert( opt_isnan->is_If(), "Expect an IfNode");
5077 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5078 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5079
5080 set_control(iftrue);
5081
5082 static const jint nan_bits = 0x7fc00000;
5083 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5084 phi->init_req(1, _gvn.transform( slow_result ));
5085 r->init_req(1, iftrue);
5086
5087 // Else fall through
5088 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5089 set_control(iffalse);
5090
5091 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5092 r->init_req(2, iffalse);
5093
5094 // Post merge
5095 set_control(_gvn.transform(r));
5096 record_for_igvn(r);
5097
5098 C->set_has_split_ifs(true); // Has chance for split-if optimization
5099 result = phi;
5100 assert(result->bottom_type()->isa_int(), "must be");
5101 break;
5102 }
5103
5104 default:
5105 fatal_unexpected_iid(id);
5106 break;
5107 }
5108 set_result(_gvn.transform(result));
5109 return true;
5110 }
5111
5112 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5113 Node* arg = argument(0);
5114 Node* result = nullptr;
5115
5116 switch (id) {
5117 case vmIntrinsics::_floatIsInfinite:
5118 result = new IsInfiniteFNode(arg);
5119 break;
5120 case vmIntrinsics::_floatIsFinite:
5121 result = new IsFiniteFNode(arg);
5122 break;
5123 case vmIntrinsics::_doubleIsInfinite:
5124 result = new IsInfiniteDNode(arg);
5125 break;
5126 case vmIntrinsics::_doubleIsFinite:
5127 result = new IsFiniteDNode(arg);
5128 break;
5129 default:
5130 fatal_unexpected_iid(id);
5131 break;
5132 }
5133 set_result(_gvn.transform(result));
5134 return true;
5135 }
5136
5137 //----------------------inline_unsafe_copyMemory-------------------------
5138 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5139
5140 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5141 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5142 const Type* base_t = gvn.type(base);
5143
5144 bool in_native = (base_t == TypePtr::NULL_PTR);
5145 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5146 bool is_mixed = !in_heap && !in_native;
5147
5148 if (is_mixed) {
5149 return true; // mixed accesses can touch both on-heap and off-heap memory
5150 }
5151 if (in_heap) {
5152 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5153 if (!is_prim_array) {
5154 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5155 // there's not enough type information available to determine proper memory slice for it.
5156 return true;
5157 }
5158 }
5159 return false;
5160 }
5161
5162 bool LibraryCallKit::inline_unsafe_copyMemory() {
5163 if (callee()->is_static()) return false; // caller must have the capability!
5164 null_check_receiver(); // null-check receiver
5165 if (stopped()) return true;
5166
5167 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5168
5169 Node* src_base = argument(1); // type: oop
5170 Node* src_off = ConvL2X(argument(2)); // type: long
5171 Node* dst_base = argument(4); // type: oop
5172 Node* dst_off = ConvL2X(argument(5)); // type: long
5173 Node* size = ConvL2X(argument(7)); // type: long
5174
5175 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5176 "fieldOffset must be byte-scaled");
5177
5178 Node* src_addr = make_unsafe_address(src_base, src_off);
5179 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5180
5181 Node* thread = _gvn.transform(new ThreadLocalNode());
5182 Node* doing_unsafe_access_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5183 BasicType doing_unsafe_access_bt = T_BYTE;
5184 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5185
5186 // update volatile field
5187 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5188
5189 int flags = RC_LEAF | RC_NO_FP;
5190
5191 const TypePtr* dst_type = TypePtr::BOTTOM;
5192
5193 // Adjust memory effects of the runtime call based on input values.
5194 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5195 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5196 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5197
5198 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5199 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5200 flags |= RC_NARROW_MEM; // narrow in memory
5201 }
5202 }
5203
5204 // Call it. Note that the length argument is not scaled.
5205 make_runtime_call(flags,
5206 OptoRuntime::fast_arraycopy_Type(),
5207 StubRoutines::unsafe_arraycopy(),
5208 "unsafe_arraycopy",
5209 dst_type,
5210 src_addr, dst_addr, size XTOP);
5211
5212 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5213
5214 return true;
5215 }
5216
5217 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5218 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5219 bool LibraryCallKit::inline_unsafe_setMemory() {
5220 if (callee()->is_static()) return false; // caller must have the capability!
5221 null_check_receiver(); // null-check receiver
5222 if (stopped()) return true;
5223
5224 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5225
5226 Node* dst_base = argument(1); // type: oop
5227 Node* dst_off = ConvL2X(argument(2)); // type: long
5228 Node* size = ConvL2X(argument(4)); // type: long
5229 Node* byte = argument(6); // type: byte
5230
5231 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5232 "fieldOffset must be byte-scaled");
5233
5234 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5235
5236 Node* thread = _gvn.transform(new ThreadLocalNode());
5237 Node* doing_unsafe_access_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5238 BasicType doing_unsafe_access_bt = T_BYTE;
5239 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5240
5241 // update volatile field
5242 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5243
5244 int flags = RC_LEAF | RC_NO_FP;
5245
5246 const TypePtr* dst_type = TypePtr::BOTTOM;
5247
5248 // Adjust memory effects of the runtime call based on input values.
5249 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5250 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5251
5252 flags |= RC_NARROW_MEM; // narrow in memory
5253 }
5254
5255 // Call it. Note that the length argument is not scaled.
5256 make_runtime_call(flags,
5257 OptoRuntime::unsafe_setmemory_Type(),
5258 StubRoutines::unsafe_setmemory(),
5259 "unsafe_setmemory",
5260 dst_type,
5261 dst_addr, size XTOP, byte);
5262
5263 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5264
5265 return true;
5266 }
5267
5268 #undef XTOP
5269
5270 //------------------------clone_coping-----------------------------------
5271 // Helper function for inline_native_clone.
5272 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5273 assert(obj_size != nullptr, "");
5274 Node* raw_obj = alloc_obj->in(1);
5275 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5276
5277 AllocateNode* alloc = nullptr;
5278 if (ReduceBulkZeroing &&
5279 // If we are implementing an array clone without knowing its source type
5280 // (can happen when compiling the array-guarded branch of a reflective
5281 // Object.clone() invocation), initialize the array within the allocation.
5282 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5283 // to a runtime clone call that assumes fully initialized source arrays.
5284 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5285 // We will be completely responsible for initializing this object -
5286 // mark Initialize node as complete.
5287 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5288 // The object was just allocated - there should be no any stores!
5289 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5290 // Mark as complete_with_arraycopy so that on AllocateNode
5291 // expansion, we know this AllocateNode is initialized by an array
5292 // copy and a StoreStore barrier exists after the array copy.
5293 alloc->initialization()->set_complete_with_arraycopy();
5294 }
5295
5296 Node* size = _gvn.transform(obj_size);
5297 access_clone(obj, alloc_obj, size, is_array);
5298
5299 // Do not let reads from the cloned object float above the arraycopy.
5300 if (alloc != nullptr) {
5301 // Do not let stores that initialize this object be reordered with
5302 // a subsequent store that would make this object accessible by
5303 // other threads.
5304 // Record what AllocateNode this StoreStore protects so that
5305 // escape analysis can go from the MemBarStoreStoreNode to the
5306 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5307 // based on the escape status of the AllocateNode.
5308 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5309 } else {
5310 insert_mem_bar(Op_MemBarCPUOrder);
5311 }
5312 }
5313
5314 //------------------------inline_native_clone----------------------------
5315 // protected native Object java.lang.Object.clone();
5316 //
5317 // Here are the simple edge cases:
5318 // null receiver => normal trap
5319 // virtual and clone was overridden => slow path to out-of-line clone
5320 // not cloneable or finalizer => slow path to out-of-line Object.clone
5321 //
5322 // The general case has two steps, allocation and copying.
5323 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5324 //
5325 // Copying also has two cases, oop arrays and everything else.
5326 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5327 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5328 //
5329 // These steps fold up nicely if and when the cloned object's klass
5330 // can be sharply typed as an object array, a type array, or an instance.
5331 //
5332 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5333 PhiNode* result_val;
5334
5335 // Set the reexecute bit for the interpreter to reexecute
5336 // the bytecode that invokes Object.clone if deoptimization happens.
5337 { PreserveReexecuteState preexecs(this);
5338 jvms()->set_should_reexecute(true);
5339
5340 Node* obj = null_check_receiver();
5341 if (stopped()) return true;
5342
5343 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5344
5345 // If we are going to clone an instance, we need its exact type to
5346 // know the number and types of fields to convert the clone to
5347 // loads/stores. Maybe a speculative type can help us.
5348 if (!obj_type->klass_is_exact() &&
5349 obj_type->speculative_type() != nullptr &&
5350 obj_type->speculative_type()->is_instance_klass()) {
5351 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5352 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5353 !spec_ik->has_injected_fields()) {
5354 if (!obj_type->isa_instptr() ||
5355 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5356 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5357 }
5358 }
5359 }
5360
5361 // Conservatively insert a memory barrier on all memory slices.
5362 // Do not let writes into the original float below the clone.
5363 insert_mem_bar(Op_MemBarCPUOrder);
5364
5365 // paths into result_reg:
5366 enum {
5367 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5368 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5369 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5370 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5371 PATH_LIMIT
5372 };
5373 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5374 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5375 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5376 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5377 record_for_igvn(result_reg);
5378
5379 Node* obj_klass = load_object_klass(obj);
5380 Node* array_obj = obj;
5381 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5382 if (array_ctl != nullptr) {
5383 // It's an array.
5384 PreserveJVMState pjvms(this);
5385 set_control(array_ctl);
5386 Node* obj_length = load_array_length(array_obj);
5387 Node* array_size = nullptr; // Size of the array without object alignment padding.
5388 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5389
5390 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5391 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5392 // If it is an oop array, it requires very special treatment,
5393 // because gc barriers are required when accessing the array.
5394 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5395 if (is_obja != nullptr) {
5396 PreserveJVMState pjvms2(this);
5397 set_control(is_obja);
5398 // Generate a direct call to the right arraycopy function(s).
5399 // Clones are always tightly coupled.
5400 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5401 ac->set_clone_oop_array();
5402 Node* n = _gvn.transform(ac);
5403 assert(n == ac, "cannot disappear");
5404 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5405
5406 result_reg->init_req(_objArray_path, control());
5407 result_val->init_req(_objArray_path, alloc_obj);
5408 result_i_o ->set_req(_objArray_path, i_o());
5409 result_mem ->set_req(_objArray_path, reset_memory());
5410 }
5411 }
5412 // Otherwise, there are no barriers to worry about.
5413 // (We can dispense with card marks if we know the allocation
5414 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5415 // causes the non-eden paths to take compensating steps to
5416 // simulate a fresh allocation, so that no further
5417 // card marks are required in compiled code to initialize
5418 // the object.)
5419
5420 if (!stopped()) {
5421 copy_to_clone(array_obj, alloc_obj, array_size, true);
5422
5423 // Present the results of the copy.
5424 result_reg->init_req(_array_path, control());
5425 result_val->init_req(_array_path, alloc_obj);
5426 result_i_o ->set_req(_array_path, i_o());
5427 result_mem ->set_req(_array_path, reset_memory());
5428 }
5429 }
5430
5431 // We only go to the instance fast case code if we pass a number of guards.
5432 // The paths which do not pass are accumulated in the slow_region.
5433 RegionNode* slow_region = new RegionNode(1);
5434 record_for_igvn(slow_region);
5435 if (!stopped()) {
5436 // It's an instance (we did array above). Make the slow-path tests.
5437 // If this is a virtual call, we generate a funny guard. We grab
5438 // the vtable entry corresponding to clone() from the target object.
5439 // If the target method which we are calling happens to be the
5440 // Object clone() method, we pass the guard. We do not need this
5441 // guard for non-virtual calls; the caller is known to be the native
5442 // Object clone().
5443 if (is_virtual) {
5444 generate_virtual_guard(obj_klass, slow_region);
5445 }
5446
5447 // The object must be easily cloneable and must not have a finalizer.
5448 // Both of these conditions may be checked in a single test.
5449 // We could optimize the test further, but we don't care.
5450 generate_misc_flags_guard(obj_klass,
5451 // Test both conditions:
5452 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5453 // Must be cloneable but not finalizer:
5454 KlassFlags::_misc_is_cloneable_fast,
5455 slow_region);
5456 }
5457
5458 if (!stopped()) {
5459 // It's an instance, and it passed the slow-path tests.
5460 PreserveJVMState pjvms(this);
5461 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5462 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5463 // is reexecuted if deoptimization occurs and there could be problems when merging
5464 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5465 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5466
5467 copy_to_clone(obj, alloc_obj, obj_size, false);
5468
5469 // Present the results of the slow call.
5470 result_reg->init_req(_instance_path, control());
5471 result_val->init_req(_instance_path, alloc_obj);
5472 result_i_o ->set_req(_instance_path, i_o());
5473 result_mem ->set_req(_instance_path, reset_memory());
5474 }
5475
5476 // Generate code for the slow case. We make a call to clone().
5477 set_control(_gvn.transform(slow_region));
5478 if (!stopped()) {
5479 PreserveJVMState pjvms(this);
5480 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5481 // We need to deoptimize on exception (see comment above)
5482 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5483 // this->control() comes from set_results_for_java_call
5484 result_reg->init_req(_slow_path, control());
5485 result_val->init_req(_slow_path, slow_result);
5486 result_i_o ->set_req(_slow_path, i_o());
5487 result_mem ->set_req(_slow_path, reset_memory());
5488 }
5489
5490 // Return the combined state.
5491 set_control( _gvn.transform(result_reg));
5492 set_i_o( _gvn.transform(result_i_o));
5493 set_all_memory( _gvn.transform(result_mem));
5494 } // original reexecute is set back here
5495
5496 set_result(_gvn.transform(result_val));
5497 return true;
5498 }
5499
5500 // If we have a tightly coupled allocation, the arraycopy may take care
5501 // of the array initialization. If one of the guards we insert between
5502 // the allocation and the arraycopy causes a deoptimization, an
5503 // uninitialized array will escape the compiled method. To prevent that
5504 // we set the JVM state for uncommon traps between the allocation and
5505 // the arraycopy to the state before the allocation so, in case of
5506 // deoptimization, we'll reexecute the allocation and the
5507 // initialization.
5508 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5509 if (alloc != nullptr) {
5510 ciMethod* trap_method = alloc->jvms()->method();
5511 int trap_bci = alloc->jvms()->bci();
5512
5513 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5514 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5515 // Make sure there's no store between the allocation and the
5516 // arraycopy otherwise visible side effects could be rexecuted
5517 // in case of deoptimization and cause incorrect execution.
5518 bool no_interfering_store = true;
5519 Node* mem = alloc->in(TypeFunc::Memory);
5520 if (mem->is_MergeMem()) {
5521 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5522 Node* n = mms.memory();
5523 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5524 assert(n->is_Store(), "what else?");
5525 no_interfering_store = false;
5526 break;
5527 }
5528 }
5529 } else {
5530 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5531 Node* n = mms.memory();
5532 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5533 assert(n->is_Store(), "what else?");
5534 no_interfering_store = false;
5535 break;
5536 }
5537 }
5538 }
5539
5540 if (no_interfering_store) {
5541 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5542
5543 JVMState* saved_jvms = jvms();
5544 saved_reexecute_sp = _reexecute_sp;
5545
5546 set_jvms(sfpt->jvms());
5547 _reexecute_sp = jvms()->sp();
5548
5549 return saved_jvms;
5550 }
5551 }
5552 }
5553 return nullptr;
5554 }
5555
5556 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5557 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5558 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5559 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5560 uint size = alloc->req();
5561 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5562 old_jvms->set_map(sfpt);
5563 for (uint i = 0; i < size; i++) {
5564 sfpt->init_req(i, alloc->in(i));
5565 }
5566 // re-push array length for deoptimization
5567 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5568 old_jvms->set_sp(old_jvms->sp()+1);
5569 old_jvms->set_monoff(old_jvms->monoff()+1);
5570 old_jvms->set_scloff(old_jvms->scloff()+1);
5571 old_jvms->set_endoff(old_jvms->endoff()+1);
5572 old_jvms->set_should_reexecute(true);
5573
5574 sfpt->set_i_o(map()->i_o());
5575 sfpt->set_memory(map()->memory());
5576 sfpt->set_control(map()->control());
5577 return sfpt;
5578 }
5579
5580 // In case of a deoptimization, we restart execution at the
5581 // allocation, allocating a new array. We would leave an uninitialized
5582 // array in the heap that GCs wouldn't expect. Move the allocation
5583 // after the traps so we don't allocate the array if we
5584 // deoptimize. This is possible because tightly_coupled_allocation()
5585 // guarantees there's no observer of the allocated array at this point
5586 // and the control flow is simple enough.
5587 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5588 int saved_reexecute_sp, uint new_idx) {
5589 if (saved_jvms_before_guards != nullptr && !stopped()) {
5590 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5591
5592 assert(alloc != nullptr, "only with a tightly coupled allocation");
5593 // restore JVM state to the state at the arraycopy
5594 saved_jvms_before_guards->map()->set_control(map()->control());
5595 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5596 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5597 // If we've improved the types of some nodes (null check) while
5598 // emitting the guards, propagate them to the current state
5599 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5600 set_jvms(saved_jvms_before_guards);
5601 _reexecute_sp = saved_reexecute_sp;
5602
5603 // Remove the allocation from above the guards
5604 CallProjections callprojs;
5605 alloc->extract_projections(&callprojs, true);
5606 InitializeNode* init = alloc->initialization();
5607 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5608 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5609 init->replace_mem_projs_by(alloc_mem, C);
5610
5611 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5612 // the allocation (i.e. is only valid if the allocation succeeds):
5613 // 1) replace CastIINode with AllocateArrayNode's length here
5614 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5615 //
5616 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5617 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5618 Node* init_control = init->proj_out(TypeFunc::Control);
5619 Node* alloc_length = alloc->Ideal_length();
5620 #ifdef ASSERT
5621 Node* prev_cast = nullptr;
5622 #endif
5623 for (uint i = 0; i < init_control->outcnt(); i++) {
5624 Node* init_out = init_control->raw_out(i);
5625 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5626 #ifdef ASSERT
5627 if (prev_cast == nullptr) {
5628 prev_cast = init_out;
5629 } else {
5630 if (prev_cast->cmp(*init_out) == false) {
5631 prev_cast->dump();
5632 init_out->dump();
5633 assert(false, "not equal CastIINode");
5634 }
5635 }
5636 #endif
5637 C->gvn_replace_by(init_out, alloc_length);
5638 }
5639 }
5640 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5641
5642 // move the allocation here (after the guards)
5643 _gvn.hash_delete(alloc);
5644 alloc->set_req(TypeFunc::Control, control());
5645 alloc->set_req(TypeFunc::I_O, i_o());
5646 Node *mem = reset_memory();
5647 set_all_memory(mem);
5648 alloc->set_req(TypeFunc::Memory, mem);
5649 set_control(init->proj_out_or_null(TypeFunc::Control));
5650 set_i_o(callprojs.fallthrough_ioproj);
5651
5652 // Update memory as done in GraphKit::set_output_for_allocation()
5653 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5654 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5655 if (ary_type->isa_aryptr() && length_type != nullptr) {
5656 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5657 }
5658 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5659 int elemidx = C->get_alias_index(telemref);
5660 // Need to properly move every memory projection for the Initialize
5661 #ifdef ASSERT
5662 int mark_idx = C->get_alias_index(ary_type->add_offset(oopDesc::mark_offset_in_bytes()));
5663 int klass_idx = C->get_alias_index(ary_type->add_offset(oopDesc::klass_offset_in_bytes()));
5664 #endif
5665 auto move_proj = [&](ProjNode* proj) {
5666 int alias_idx = C->get_alias_index(proj->adr_type());
5667 assert(alias_idx == Compile::AliasIdxRaw ||
5668 alias_idx == elemidx ||
5669 alias_idx == mark_idx ||
5670 alias_idx == klass_idx, "should be raw memory or array element type");
5671 set_memory(proj, alias_idx);
5672 };
5673 init->for_each_proj(move_proj, TypeFunc::Memory);
5674
5675 Node* allocx = _gvn.transform(alloc);
5676 assert(allocx == alloc, "where has the allocation gone?");
5677 assert(dest->is_CheckCastPP(), "not an allocation result?");
5678
5679 _gvn.hash_delete(dest);
5680 dest->set_req(0, control());
5681 Node* destx = _gvn.transform(dest);
5682 assert(destx == dest, "where has the allocation result gone?");
5683
5684 array_ideal_length(alloc, ary_type, true);
5685 }
5686 }
5687
5688 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5689 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5690 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5691 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5692 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5693 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5694 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5695 JVMState* saved_jvms_before_guards) {
5696 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5697 // There is at least one unrelated uncommon trap which needs to be replaced.
5698 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5699
5700 JVMState* saved_jvms = jvms();
5701 const int saved_reexecute_sp = _reexecute_sp;
5702 set_jvms(sfpt->jvms());
5703 _reexecute_sp = jvms()->sp();
5704
5705 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5706
5707 // Restore state
5708 set_jvms(saved_jvms);
5709 _reexecute_sp = saved_reexecute_sp;
5710 }
5711 }
5712
5713 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5714 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5715 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5716 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5717 while (if_proj->is_IfProj()) {
5718 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5719 if (uncommon_trap != nullptr) {
5720 create_new_uncommon_trap(uncommon_trap);
5721 }
5722 assert(if_proj->in(0)->is_If(), "must be If");
5723 if_proj = if_proj->in(0)->in(0);
5724 }
5725 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5726 "must have reached control projection of init node");
5727 }
5728
5729 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5730 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5731 assert(trap_request != 0, "no valid UCT trap request");
5732 PreserveJVMState pjvms(this);
5733 set_control(uncommon_trap_call->in(0));
5734 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5735 Deoptimization::trap_request_action(trap_request));
5736 assert(stopped(), "Should be stopped");
5737 _gvn.hash_delete(uncommon_trap_call);
5738 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5739 }
5740
5741 // Common checks for array sorting intrinsics arguments.
5742 // Returns `true` if checks passed.
5743 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
5744 // check address of the class
5745 if (elementType == nullptr || elementType->is_top()) {
5746 return false; // dead path
5747 }
5748 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5749 if (elem_klass == nullptr) {
5750 return false; // dead path
5751 }
5752 // java_mirror_type() returns non-null for compile-time Class constants only
5753 ciType* elem_type = elem_klass->java_mirror_type();
5754 if (elem_type == nullptr) {
5755 return false;
5756 }
5757 bt = elem_type->basic_type();
5758 // Disable the intrinsic if the CPU does not support SIMD sort
5759 if (!Matcher::supports_simd_sort(bt)) {
5760 return false;
5761 }
5762 // check address of the array
5763 if (obj == nullptr || obj->is_top()) {
5764 return false; // dead path
5765 }
5766 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5767 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
5768 return false; // failed input validation
5769 }
5770 return true;
5771 }
5772
5773 //------------------------------inline_array_partition-----------------------
5774 bool LibraryCallKit::inline_array_partition() {
5775 address stubAddr = StubRoutines::select_array_partition_function();
5776 if (stubAddr == nullptr) {
5777 return false; // Intrinsic's stub is not implemented on this platform
5778 }
5779 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
5780
5781 // no receiver because it is a static method
5782 Node* elementType = argument(0);
5783 Node* obj = argument(1);
5784 Node* offset = argument(2); // long
5785 Node* fromIndex = argument(4);
5786 Node* toIndex = argument(5);
5787 Node* indexPivot1 = argument(6);
5788 Node* indexPivot2 = argument(7);
5789 // PartitionOperation: argument(8) is ignored
5790
5791 Node* pivotIndices = nullptr;
5792 BasicType bt = T_ILLEGAL;
5793
5794 if (!check_array_sort_arguments(elementType, obj, bt)) {
5795 return false;
5796 }
5797 null_check(obj);
5798 // If obj is dead, only null-path is taken.
5799 if (stopped()) {
5800 return true;
5801 }
5802 // Set the original stack and the reexecute bit for the interpreter to reexecute
5803 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
5804 { PreserveReexecuteState preexecs(this);
5805 jvms()->set_should_reexecute(true);
5806
5807 Node* obj_adr = make_unsafe_address(obj, offset);
5808
5809 // create the pivotIndices array of type int and size = 2
5810 Node* size = intcon(2);
5811 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
5812 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
5813 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
5814 guarantee(alloc != nullptr, "created above");
5815 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
5816
5817 // pass the basic type enum to the stub
5818 Node* elemType = intcon(bt);
5819
5820 // Call the stub
5821 const char *stubName = "array_partition_stub";
5822 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
5823 stubAddr, stubName, TypePtr::BOTTOM,
5824 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
5825 indexPivot1, indexPivot2);
5826
5827 } // original reexecute is set back here
5828
5829 if (!stopped()) {
5830 set_result(pivotIndices);
5831 }
5832
5833 return true;
5834 }
5835
5836
5837 //------------------------------inline_array_sort-----------------------
5838 bool LibraryCallKit::inline_array_sort() {
5839 address stubAddr = StubRoutines::select_arraysort_function();
5840 if (stubAddr == nullptr) {
5841 return false; // Intrinsic's stub is not implemented on this platform
5842 }
5843 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
5844
5845 // no receiver because it is a static method
5846 Node* elementType = argument(0);
5847 Node* obj = argument(1);
5848 Node* offset = argument(2); // long
5849 Node* fromIndex = argument(4);
5850 Node* toIndex = argument(5);
5851 // SortOperation: argument(6) is ignored
5852
5853 BasicType bt = T_ILLEGAL;
5854
5855 if (!check_array_sort_arguments(elementType, obj, bt)) {
5856 return false;
5857 }
5858 null_check(obj);
5859 // If obj is dead, only null-path is taken.
5860 if (stopped()) {
5861 return true;
5862 }
5863 Node* obj_adr = make_unsafe_address(obj, offset);
5864
5865 // pass the basic type enum to the stub
5866 Node* elemType = intcon(bt);
5867
5868 // Call the stub.
5869 const char *stubName = "arraysort_stub";
5870 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
5871 stubAddr, stubName, TypePtr::BOTTOM,
5872 obj_adr, elemType, fromIndex, toIndex);
5873
5874 return true;
5875 }
5876
5877
5878 //------------------------------inline_arraycopy-----------------------
5879 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5880 // Object dest, int destPos,
5881 // int length);
5882 bool LibraryCallKit::inline_arraycopy() {
5883 // Get the arguments.
5884 Node* src = argument(0); // type: oop
5885 Node* src_offset = argument(1); // type: int
5886 Node* dest = argument(2); // type: oop
5887 Node* dest_offset = argument(3); // type: int
5888 Node* length = argument(4); // type: int
5889
5890 uint new_idx = C->unique();
5891
5892 // Check for allocation before we add nodes that would confuse
5893 // tightly_coupled_allocation()
5894 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5895
5896 int saved_reexecute_sp = -1;
5897 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5898 // See arraycopy_restore_alloc_state() comment
5899 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5900 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5901 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5902 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5903
5904 // The following tests must be performed
5905 // (1) src and dest are arrays.
5906 // (2) src and dest arrays must have elements of the same BasicType
5907 // (3) src and dest must not be null.
5908 // (4) src_offset must not be negative.
5909 // (5) dest_offset must not be negative.
5910 // (6) length must not be negative.
5911 // (7) src_offset + length must not exceed length of src.
5912 // (8) dest_offset + length must not exceed length of dest.
5913 // (9) each element of an oop array must be assignable
5914
5915 // (3) src and dest must not be null.
5916 // always do this here because we need the JVM state for uncommon traps
5917 Node* null_ctl = top();
5918 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5919 assert(null_ctl->is_top(), "no null control here");
5920 dest = null_check(dest, T_ARRAY);
5921
5922 if (!can_emit_guards) {
5923 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5924 // guards but the arraycopy node could still take advantage of a
5925 // tightly allocated allocation. tightly_coupled_allocation() is
5926 // called again to make sure it takes the null check above into
5927 // account: the null check is mandatory and if it caused an
5928 // uncommon trap to be emitted then the allocation can't be
5929 // considered tightly coupled in this context.
5930 alloc = tightly_coupled_allocation(dest);
5931 }
5932
5933 bool validated = false;
5934
5935 const Type* src_type = _gvn.type(src);
5936 const Type* dest_type = _gvn.type(dest);
5937 const TypeAryPtr* top_src = src_type->isa_aryptr();
5938 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5939
5940 // Do we have the type of src?
5941 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5942 // Do we have the type of dest?
5943 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5944 // Is the type for src from speculation?
5945 bool src_spec = false;
5946 // Is the type for dest from speculation?
5947 bool dest_spec = false;
5948
5949 if ((!has_src || !has_dest) && can_emit_guards) {
5950 // We don't have sufficient type information, let's see if
5951 // speculative types can help. We need to have types for both src
5952 // and dest so that it pays off.
5953
5954 // Do we already have or could we have type information for src
5955 bool could_have_src = has_src;
5956 // Do we already have or could we have type information for dest
5957 bool could_have_dest = has_dest;
5958
5959 ciKlass* src_k = nullptr;
5960 if (!has_src) {
5961 src_k = src_type->speculative_type_not_null();
5962 if (src_k != nullptr && src_k->is_array_klass()) {
5963 could_have_src = true;
5964 }
5965 }
5966
5967 ciKlass* dest_k = nullptr;
5968 if (!has_dest) {
5969 dest_k = dest_type->speculative_type_not_null();
5970 if (dest_k != nullptr && dest_k->is_array_klass()) {
5971 could_have_dest = true;
5972 }
5973 }
5974
5975 if (could_have_src && could_have_dest) {
5976 // This is going to pay off so emit the required guards
5977 if (!has_src) {
5978 src = maybe_cast_profiled_obj(src, src_k, true);
5979 src_type = _gvn.type(src);
5980 top_src = src_type->isa_aryptr();
5981 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5982 src_spec = true;
5983 }
5984 if (!has_dest) {
5985 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5986 dest_type = _gvn.type(dest);
5987 top_dest = dest_type->isa_aryptr();
5988 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5989 dest_spec = true;
5990 }
5991 }
5992 }
5993
5994 if (has_src && has_dest && can_emit_guards) {
5995 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5996 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5997 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5998 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5999
6000 if (src_elem == dest_elem && src_elem == T_OBJECT) {
6001 // If both arrays are object arrays then having the exact types
6002 // for both will remove the need for a subtype check at runtime
6003 // before the call and may make it possible to pick a faster copy
6004 // routine (without a subtype check on every element)
6005 // Do we have the exact type of src?
6006 bool could_have_src = src_spec;
6007 // Do we have the exact type of dest?
6008 bool could_have_dest = dest_spec;
6009 ciKlass* src_k = nullptr;
6010 ciKlass* dest_k = nullptr;
6011 if (!src_spec) {
6012 src_k = src_type->speculative_type_not_null();
6013 if (src_k != nullptr && src_k->is_array_klass()) {
6014 could_have_src = true;
6015 }
6016 }
6017 if (!dest_spec) {
6018 dest_k = dest_type->speculative_type_not_null();
6019 if (dest_k != nullptr && dest_k->is_array_klass()) {
6020 could_have_dest = true;
6021 }
6022 }
6023 if (could_have_src && could_have_dest) {
6024 // If we can have both exact types, emit the missing guards
6025 if (could_have_src && !src_spec) {
6026 src = maybe_cast_profiled_obj(src, src_k, true);
6027 }
6028 if (could_have_dest && !dest_spec) {
6029 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6030 }
6031 }
6032 }
6033 }
6034
6035 ciMethod* trap_method = method();
6036 int trap_bci = bci();
6037 if (saved_jvms_before_guards != nullptr) {
6038 trap_method = alloc->jvms()->method();
6039 trap_bci = alloc->jvms()->bci();
6040 }
6041
6042 bool negative_length_guard_generated = false;
6043
6044 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6045 can_emit_guards &&
6046 !src->is_top() && !dest->is_top()) {
6047 // validate arguments: enables transformation the ArrayCopyNode
6048 validated = true;
6049
6050 RegionNode* slow_region = new RegionNode(1);
6051 record_for_igvn(slow_region);
6052
6053 // (1) src and dest are arrays.
6054 generate_non_array_guard(load_object_klass(src), slow_region, &src);
6055 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
6056
6057 // (2) src and dest arrays must have elements of the same BasicType
6058 // done at macro expansion or at Ideal transformation time
6059
6060 // (4) src_offset must not be negative.
6061 generate_negative_guard(src_offset, slow_region);
6062
6063 // (5) dest_offset must not be negative.
6064 generate_negative_guard(dest_offset, slow_region);
6065
6066 // (7) src_offset + length must not exceed length of src.
6067 generate_limit_guard(src_offset, length,
6068 load_array_length(src),
6069 slow_region);
6070
6071 // (8) dest_offset + length must not exceed length of dest.
6072 generate_limit_guard(dest_offset, length,
6073 load_array_length(dest),
6074 slow_region);
6075
6076 // (6) length must not be negative.
6077 // This is also checked in generate_arraycopy() during macro expansion, but
6078 // we also have to check it here for the case where the ArrayCopyNode will
6079 // be eliminated by Escape Analysis.
6080 if (EliminateAllocations) {
6081 generate_negative_guard(length, slow_region);
6082 negative_length_guard_generated = true;
6083 }
6084
6085 // (9) each element of an oop array must be assignable
6086 Node* dest_klass = load_object_klass(dest);
6087 if (src != dest) {
6088 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6089
6090 if (not_subtype_ctrl != top()) {
6091 PreserveJVMState pjvms(this);
6092 set_control(not_subtype_ctrl);
6093 uncommon_trap(Deoptimization::Reason_intrinsic,
6094 Deoptimization::Action_make_not_entrant);
6095 assert(stopped(), "Should be stopped");
6096 }
6097 }
6098 {
6099 PreserveJVMState pjvms(this);
6100 set_control(_gvn.transform(slow_region));
6101 uncommon_trap(Deoptimization::Reason_intrinsic,
6102 Deoptimization::Action_make_not_entrant);
6103 assert(stopped(), "Should be stopped");
6104 }
6105
6106 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
6107 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6108 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6109 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6110 }
6111
6112 if (stopped()) {
6113 return true;
6114 }
6115
6116 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6117 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6118 // so the compiler has a chance to eliminate them: during macro expansion,
6119 // we have to set their control (CastPP nodes are eliminated).
6120 load_object_klass(src), load_object_klass(dest),
6121 load_array_length(src), load_array_length(dest));
6122
6123 ac->set_arraycopy(validated);
6124
6125 Node* n = _gvn.transform(ac);
6126 if (n == ac) {
6127 ac->connect_outputs(this);
6128 } else {
6129 assert(validated, "shouldn't transform if all arguments not validated");
6130 set_all_memory(n);
6131 }
6132 clear_upper_avx();
6133
6134
6135 return true;
6136 }
6137
6138
6139 // Helper function which determines if an arraycopy immediately follows
6140 // an allocation, with no intervening tests or other escapes for the object.
6141 AllocateArrayNode*
6142 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6143 if (stopped()) return nullptr; // no fast path
6144 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6145
6146 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6147 if (alloc == nullptr) return nullptr;
6148
6149 Node* rawmem = memory(Compile::AliasIdxRaw);
6150 // Is the allocation's memory state untouched?
6151 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6152 // Bail out if there have been raw-memory effects since the allocation.
6153 // (Example: There might have been a call or safepoint.)
6154 return nullptr;
6155 }
6156 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6157 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6158 return nullptr;
6159 }
6160
6161 // There must be no unexpected observers of this allocation.
6162 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6163 Node* obs = ptr->fast_out(i);
6164 if (obs != this->map()) {
6165 return nullptr;
6166 }
6167 }
6168
6169 // This arraycopy must unconditionally follow the allocation of the ptr.
6170 Node* alloc_ctl = ptr->in(0);
6171 Node* ctl = control();
6172 while (ctl != alloc_ctl) {
6173 // There may be guards which feed into the slow_region.
6174 // Any other control flow means that we might not get a chance
6175 // to finish initializing the allocated object.
6176 // Various low-level checks bottom out in uncommon traps. These
6177 // are considered safe since we've already checked above that
6178 // there is no unexpected observer of this allocation.
6179 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6180 assert(ctl->in(0)->is_If(), "must be If");
6181 ctl = ctl->in(0)->in(0);
6182 } else {
6183 return nullptr;
6184 }
6185 }
6186
6187 // If we get this far, we have an allocation which immediately
6188 // precedes the arraycopy, and we can take over zeroing the new object.
6189 // The arraycopy will finish the initialization, and provide
6190 // a new control state to which we will anchor the destination pointer.
6191
6192 return alloc;
6193 }
6194
6195 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6196 if (node->is_IfProj()) {
6197 IfProjNode* other_proj = node->as_IfProj()->other_if_proj();
6198 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6199 Node* obs = other_proj->fast_out(j);
6200 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6201 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6202 return obs->as_CallStaticJava();
6203 }
6204 }
6205 }
6206 return nullptr;
6207 }
6208
6209 //-------------inline_encodeISOArray-----------------------------------
6210 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6211 // int java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6212 // int java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6213 // encode char[] to byte[] in ISO_8859_1 or ASCII
6214 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6215 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6216 // no receiver since it is static method
6217 Node *src = argument(0);
6218 Node *src_offset = argument(1);
6219 Node *dst = argument(2);
6220 Node *dst_offset = argument(3);
6221 Node *length = argument(4);
6222
6223 // Cast source & target arrays to not-null
6224 src = must_be_not_null(src, true);
6225 dst = must_be_not_null(dst, true);
6226 if (stopped()) {
6227 return true;
6228 }
6229
6230 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6231 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6232 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6233 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6234 // failed array check
6235 return false;
6236 }
6237
6238 // Figure out the size and type of the elements we will be copying.
6239 BasicType src_elem = src_type->elem()->array_element_basic_type();
6240 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6241 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6242 return false;
6243 }
6244
6245 // Check source & target bounds
6246 RegionNode* bailout = create_bailout();
6247 generate_string_range_check(src, src_offset, length, src_elem == T_BYTE, bailout);
6248 generate_string_range_check(dst, dst_offset, length, false, bailout);
6249 if (check_bailout(bailout)) {
6250 return true;
6251 }
6252
6253 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6254 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6255 // 'src_start' points to src array + scaled offset
6256 // 'dst_start' points to dst array + scaled offset
6257
6258 // See GraphKit::compress_string
6259 const TypePtr* adr_type;
6260 Node* mem = capture_memory(adr_type, src_type, dst_type);
6261 Node* enc = new EncodeISOArrayNode(control(), mem, adr_type, src_start, dst_start, length, ascii);
6262 enc = _gvn.transform(enc);
6263 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6264 memory_effect(res_mem, src_type, dst_type);
6265
6266 set_result(enc);
6267 clear_upper_avx();
6268
6269 return true;
6270 }
6271
6272 //-------------inline_multiplyToLen-----------------------------------
6273 bool LibraryCallKit::inline_multiplyToLen() {
6274 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6275
6276 address stubAddr = StubRoutines::multiplyToLen();
6277 if (stubAddr == nullptr) {
6278 return false; // Intrinsic's stub is not implemented on this platform
6279 }
6280 const char* stubName = "multiplyToLen";
6281
6282 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6283
6284 // no receiver because it is a static method
6285 Node* x = argument(0);
6286 Node* xlen = argument(1);
6287 Node* y = argument(2);
6288 Node* ylen = argument(3);
6289 Node* z = argument(4);
6290
6291 x = must_be_not_null(x, true);
6292 y = must_be_not_null(y, true);
6293
6294 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6295 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6296 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6297 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6298 // failed array check
6299 return false;
6300 }
6301
6302 BasicType x_elem = x_type->elem()->array_element_basic_type();
6303 BasicType y_elem = y_type->elem()->array_element_basic_type();
6304 if (x_elem != T_INT || y_elem != T_INT) {
6305 return false;
6306 }
6307
6308 Node* x_start = array_element_address(x, intcon(0), x_elem);
6309 Node* y_start = array_element_address(y, intcon(0), y_elem);
6310 // 'x_start' points to x array + scaled xlen
6311 // 'y_start' points to y array + scaled ylen
6312
6313 Node* z_start = array_element_address(z, intcon(0), T_INT);
6314
6315 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6316 OptoRuntime::multiplyToLen_Type(),
6317 stubAddr, stubName, TypePtr::BOTTOM,
6318 x_start, xlen, y_start, ylen, z_start);
6319
6320 C->set_has_split_ifs(true); // Has chance for split-if optimization
6321 set_result(z);
6322 return true;
6323 }
6324
6325 //-------------inline_squareToLen------------------------------------
6326 bool LibraryCallKit::inline_squareToLen() {
6327 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6328
6329 address stubAddr = StubRoutines::squareToLen();
6330 if (stubAddr == nullptr) {
6331 return false; // Intrinsic's stub is not implemented on this platform
6332 }
6333 const char* stubName = "squareToLen";
6334
6335 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6336
6337 Node* x = argument(0);
6338 Node* len = argument(1);
6339 Node* z = argument(2);
6340 Node* zlen = argument(3);
6341
6342 x = must_be_not_null(x, true);
6343 z = must_be_not_null(z, true);
6344
6345 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6346 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6347 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6348 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6349 // failed array check
6350 return false;
6351 }
6352
6353 BasicType x_elem = x_type->elem()->array_element_basic_type();
6354 BasicType z_elem = z_type->elem()->array_element_basic_type();
6355 if (x_elem != T_INT || z_elem != T_INT) {
6356 return false;
6357 }
6358
6359
6360 Node* x_start = array_element_address(x, intcon(0), x_elem);
6361 Node* z_start = array_element_address(z, intcon(0), z_elem);
6362
6363 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6364 OptoRuntime::squareToLen_Type(),
6365 stubAddr, stubName, TypePtr::BOTTOM,
6366 x_start, len, z_start, zlen);
6367
6368 set_result(z);
6369 return true;
6370 }
6371
6372 //-------------inline_mulAdd------------------------------------------
6373 bool LibraryCallKit::inline_mulAdd() {
6374 assert(UseMulAddIntrinsic, "not implemented on this platform");
6375
6376 address stubAddr = StubRoutines::mulAdd();
6377 if (stubAddr == nullptr) {
6378 return false; // Intrinsic's stub is not implemented on this platform
6379 }
6380 const char* stubName = "mulAdd";
6381
6382 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6383
6384 Node* out = argument(0);
6385 Node* in = argument(1);
6386 Node* offset = argument(2);
6387 Node* len = argument(3);
6388 Node* k = argument(4);
6389
6390 in = must_be_not_null(in, true);
6391 out = must_be_not_null(out, true);
6392
6393 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6394 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6395 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6396 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6397 // failed array check
6398 return false;
6399 }
6400
6401 BasicType out_elem = out_type->elem()->array_element_basic_type();
6402 BasicType in_elem = in_type->elem()->array_element_basic_type();
6403 if (out_elem != T_INT || in_elem != T_INT) {
6404 return false;
6405 }
6406
6407 Node* outlen = load_array_length(out);
6408 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6409 Node* out_start = array_element_address(out, intcon(0), out_elem);
6410 Node* in_start = array_element_address(in, intcon(0), in_elem);
6411
6412 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6413 OptoRuntime::mulAdd_Type(),
6414 stubAddr, stubName, TypePtr::BOTTOM,
6415 out_start,in_start, new_offset, len, k);
6416 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6417 set_result(result);
6418 return true;
6419 }
6420
6421 //-------------inline_montgomeryMultiply-----------------------------------
6422 bool LibraryCallKit::inline_montgomeryMultiply() {
6423 address stubAddr = StubRoutines::montgomeryMultiply();
6424 if (stubAddr == nullptr) {
6425 return false; // Intrinsic's stub is not implemented on this platform
6426 }
6427
6428 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6429 const char* stubName = "montgomery_multiply";
6430
6431 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6432
6433 Node* a = argument(0);
6434 Node* b = argument(1);
6435 Node* n = argument(2);
6436 Node* len = argument(3);
6437 Node* inv = argument(4);
6438 Node* m = argument(6);
6439
6440 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6441 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6442 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6443 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6444 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6445 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6446 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6447 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6448 // failed array check
6449 return false;
6450 }
6451
6452 BasicType a_elem = a_type->elem()->array_element_basic_type();
6453 BasicType b_elem = b_type->elem()->array_element_basic_type();
6454 BasicType n_elem = n_type->elem()->array_element_basic_type();
6455 BasicType m_elem = m_type->elem()->array_element_basic_type();
6456 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6457 return false;
6458 }
6459
6460 // Make the call
6461 {
6462 Node* a_start = array_element_address(a, intcon(0), a_elem);
6463 Node* b_start = array_element_address(b, intcon(0), b_elem);
6464 Node* n_start = array_element_address(n, intcon(0), n_elem);
6465 Node* m_start = array_element_address(m, intcon(0), m_elem);
6466
6467 Node* call = make_runtime_call(RC_LEAF,
6468 OptoRuntime::montgomeryMultiply_Type(),
6469 stubAddr, stubName, TypePtr::BOTTOM,
6470 a_start, b_start, n_start, len, inv, top(),
6471 m_start);
6472 set_result(m);
6473 }
6474
6475 return true;
6476 }
6477
6478 bool LibraryCallKit::inline_montgomerySquare() {
6479 address stubAddr = StubRoutines::montgomerySquare();
6480 if (stubAddr == nullptr) {
6481 return false; // Intrinsic's stub is not implemented on this platform
6482 }
6483
6484 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6485 const char* stubName = "montgomery_square";
6486
6487 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6488
6489 Node* a = argument(0);
6490 Node* n = argument(1);
6491 Node* len = argument(2);
6492 Node* inv = argument(3);
6493 Node* m = argument(5);
6494
6495 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6496 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6497 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6498 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6499 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6500 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6501 // failed array check
6502 return false;
6503 }
6504
6505 BasicType a_elem = a_type->elem()->array_element_basic_type();
6506 BasicType n_elem = n_type->elem()->array_element_basic_type();
6507 BasicType m_elem = m_type->elem()->array_element_basic_type();
6508 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6509 return false;
6510 }
6511
6512 // Make the call
6513 {
6514 Node* a_start = array_element_address(a, intcon(0), a_elem);
6515 Node* n_start = array_element_address(n, intcon(0), n_elem);
6516 Node* m_start = array_element_address(m, intcon(0), m_elem);
6517
6518 Node* call = make_runtime_call(RC_LEAF,
6519 OptoRuntime::montgomerySquare_Type(),
6520 stubAddr, stubName, TypePtr::BOTTOM,
6521 a_start, n_start, len, inv, top(),
6522 m_start);
6523 set_result(m);
6524 }
6525
6526 return true;
6527 }
6528
6529 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6530 address stubAddr = nullptr;
6531 const char* stubName = nullptr;
6532
6533 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6534 if (stubAddr == nullptr) {
6535 return false; // Intrinsic's stub is not implemented on this platform
6536 }
6537
6538 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6539
6540 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6541
6542 Node* newArr = argument(0);
6543 Node* oldArr = argument(1);
6544 Node* newIdx = argument(2);
6545 Node* shiftCount = argument(3);
6546 Node* numIter = argument(4);
6547
6548 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6549 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6550 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6551 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6552 return false;
6553 }
6554
6555 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6556 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6557 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6558 return false;
6559 }
6560
6561 // Make the call
6562 {
6563 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6564 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6565
6566 Node* call = make_runtime_call(RC_LEAF,
6567 OptoRuntime::bigIntegerShift_Type(),
6568 stubAddr,
6569 stubName,
6570 TypePtr::BOTTOM,
6571 newArr_start,
6572 oldArr_start,
6573 newIdx,
6574 shiftCount,
6575 numIter);
6576 }
6577
6578 return true;
6579 }
6580
6581 //-------------inline_vectorizedMismatch------------------------------
6582 bool LibraryCallKit::inline_vectorizedMismatch() {
6583 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6584
6585 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6586 Node* obja = argument(0); // Object
6587 Node* aoffset = argument(1); // long
6588 Node* objb = argument(3); // Object
6589 Node* boffset = argument(4); // long
6590 Node* length = argument(6); // int
6591 Node* scale = argument(7); // int
6592
6593 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6594 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6595 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6596 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6597 scale == top()) {
6598 return false; // failed input validation
6599 }
6600
6601 Node* obja_adr = make_unsafe_address(obja, aoffset);
6602 Node* objb_adr = make_unsafe_address(objb, boffset);
6603
6604 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6605 //
6606 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6607 // if (length <= inline_limit) {
6608 // inline_path:
6609 // vmask = VectorMaskGen length
6610 // vload1 = LoadVectorMasked obja, vmask
6611 // vload2 = LoadVectorMasked objb, vmask
6612 // result1 = VectorCmpMasked vload1, vload2, vmask
6613 // } else {
6614 // call_stub_path:
6615 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6616 // }
6617 // exit_block:
6618 // return Phi(result1, result2);
6619 //
6620 enum { inline_path = 1, // input is small enough to process it all at once
6621 stub_path = 2, // input is too large; call into the VM
6622 PATH_LIMIT = 3
6623 };
6624
6625 Node* exit_block = new RegionNode(PATH_LIMIT);
6626 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6627 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6628
6629 Node* call_stub_path = control();
6630
6631 BasicType elem_bt = T_ILLEGAL;
6632
6633 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6634 if (scale_t->is_con()) {
6635 switch (scale_t->get_con()) {
6636 case 0: elem_bt = T_BYTE; break;
6637 case 1: elem_bt = T_SHORT; break;
6638 case 2: elem_bt = T_INT; break;
6639 case 3: elem_bt = T_LONG; break;
6640
6641 default: elem_bt = T_ILLEGAL; break; // not supported
6642 }
6643 }
6644
6645 int inline_limit = 0;
6646 bool do_partial_inline = false;
6647
6648 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6649 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6650 do_partial_inline = inline_limit >= 16;
6651 }
6652
6653 if (do_partial_inline) {
6654 assert(elem_bt != T_ILLEGAL, "sanity");
6655
6656 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6657 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6658 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6659
6660 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6661 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6662 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6663
6664 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6665
6666 if (!stopped()) {
6667 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6668
6669 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6670 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6671 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6672 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6673
6674 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6675 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6676 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6677 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6678
6679 exit_block->init_req(inline_path, control());
6680 memory_phi->init_req(inline_path, map()->memory());
6681 result_phi->init_req(inline_path, result);
6682
6683 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6684 clear_upper_avx();
6685 }
6686 }
6687 }
6688
6689 if (call_stub_path != nullptr) {
6690 set_control(call_stub_path);
6691
6692 Node* call = make_runtime_call(RC_LEAF,
6693 OptoRuntime::vectorizedMismatch_Type(),
6694 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6695 obja_adr, objb_adr, length, scale);
6696
6697 exit_block->init_req(stub_path, control());
6698 memory_phi->init_req(stub_path, map()->memory());
6699 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6700 }
6701
6702 exit_block = _gvn.transform(exit_block);
6703 memory_phi = _gvn.transform(memory_phi);
6704 result_phi = _gvn.transform(result_phi);
6705
6706 record_for_igvn(exit_block);
6707 record_for_igvn(memory_phi);
6708 record_for_igvn(result_phi);
6709
6710 set_control(exit_block);
6711 set_all_memory(memory_phi);
6712 set_result(result_phi);
6713
6714 return true;
6715 }
6716
6717 //------------------------------inline_vectorizedHashcode----------------------------
6718 bool LibraryCallKit::inline_vectorizedHashCode() {
6719 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6720
6721 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6722 Node* array = argument(0);
6723 Node* offset = argument(1);
6724 Node* length = argument(2);
6725 Node* initialValue = argument(3);
6726 Node* basic_type = argument(4);
6727
6728 if (basic_type == top()) {
6729 return false; // failed input validation
6730 }
6731
6732 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6733 if (!basic_type_t->is_con()) {
6734 return false; // Only intrinsify if mode argument is constant
6735 }
6736
6737 array = must_be_not_null(array, true);
6738
6739 BasicType bt = (BasicType)basic_type_t->get_con();
6740
6741 // Resolve address of first element
6742 Node* array_start = array_element_address(array, offset, bt);
6743
6744 const TypeAryPtr* in_adr_type = TypeAryPtr::get_array_body_type(bt);
6745 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(in_adr_type), in_adr_type,
6746 array_start, length, initialValue, basic_type)));
6747 clear_upper_avx();
6748
6749 return true;
6750 }
6751
6752 /**
6753 * Calculate CRC32 for byte.
6754 * int java.util.zip.CRC32.update(int crc, int b)
6755 */
6756 bool LibraryCallKit::inline_updateCRC32() {
6757 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6758 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6759 // no receiver since it is static method
6760 Node* crc = argument(0); // type: int
6761 Node* b = argument(1); // type: int
6762
6763 /*
6764 * int c = ~ crc;
6765 * b = timesXtoThe32[(b ^ c) & 0xFF];
6766 * b = b ^ (c >>> 8);
6767 * crc = ~b;
6768 */
6769
6770 Node* M1 = intcon(-1);
6771 crc = _gvn.transform(new XorINode(crc, M1));
6772 Node* result = _gvn.transform(new XorINode(crc, b));
6773 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6774
6775 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6776 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6777 Node* adr = off_heap_plus_addr(base, ConvI2X(offset));
6778 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6779
6780 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6781 result = _gvn.transform(new XorINode(crc, result));
6782 result = _gvn.transform(new XorINode(result, M1));
6783 set_result(result);
6784 return true;
6785 }
6786
6787 /**
6788 * Calculate CRC32 for byte[] array.
6789 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6790 */
6791 bool LibraryCallKit::inline_updateBytesCRC32() {
6792 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6793 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6794 // no receiver since it is static method
6795 Node* crc = argument(0); // type: int
6796 Node* src = argument(1); // type: oop
6797 Node* offset = argument(2); // type: int
6798 Node* length = argument(3); // type: int
6799
6800 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6801 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6802 // failed array check
6803 return false;
6804 }
6805
6806 // Figure out the size and type of the elements we will be copying.
6807 BasicType src_elem = src_type->elem()->array_element_basic_type();
6808 if (src_elem != T_BYTE) {
6809 return false;
6810 }
6811
6812 // 'src_start' points to src array + scaled offset
6813 src = must_be_not_null(src, true);
6814 Node* src_start = array_element_address(src, offset, src_elem);
6815
6816 // We assume that range check is done by caller.
6817 // TODO: generate range check (offset+length < src.length) in debug VM.
6818
6819 // Call the stub.
6820 address stubAddr = StubRoutines::updateBytesCRC32();
6821 const char *stubName = "updateBytesCRC32";
6822
6823 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6824 stubAddr, stubName, TypePtr::BOTTOM,
6825 crc, src_start, length);
6826 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6827 set_result(result);
6828 return true;
6829 }
6830
6831 /**
6832 * Calculate CRC32 for ByteBuffer.
6833 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6834 */
6835 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6836 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6837 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6838 // no receiver since it is static method
6839 Node* crc = argument(0); // type: int
6840 Node* src = argument(1); // type: long
6841 Node* offset = argument(3); // type: int
6842 Node* length = argument(4); // type: int
6843
6844 src = ConvL2X(src); // adjust Java long to machine word
6845 Node* base = _gvn.transform(new CastX2PNode(src));
6846 offset = ConvI2X(offset);
6847
6848 // 'src_start' points to src array + scaled offset
6849 Node* src_start = off_heap_plus_addr(base, offset);
6850
6851 // Call the stub.
6852 address stubAddr = StubRoutines::updateBytesCRC32();
6853 const char *stubName = "updateBytesCRC32";
6854
6855 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6856 stubAddr, stubName, TypePtr::BOTTOM,
6857 crc, src_start, length);
6858 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6859 set_result(result);
6860 return true;
6861 }
6862
6863 //------------------------------get_table_from_crc32c_class-----------------------
6864 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6865 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6866 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6867
6868 return table;
6869 }
6870
6871 //------------------------------inline_updateBytesCRC32C-----------------------
6872 //
6873 // Calculate CRC32C for byte[] array.
6874 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6875 //
6876 bool LibraryCallKit::inline_updateBytesCRC32C() {
6877 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6878 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6879 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6880 // no receiver since it is a static method
6881 Node* crc = argument(0); // type: int
6882 Node* src = argument(1); // type: oop
6883 Node* offset = argument(2); // type: int
6884 Node* end = argument(3); // type: int
6885
6886 Node* length = _gvn.transform(new SubINode(end, offset));
6887
6888 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6889 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6890 // failed array check
6891 return false;
6892 }
6893
6894 // Figure out the size and type of the elements we will be copying.
6895 BasicType src_elem = src_type->elem()->array_element_basic_type();
6896 if (src_elem != T_BYTE) {
6897 return false;
6898 }
6899
6900 // 'src_start' points to src array + scaled offset
6901 src = must_be_not_null(src, true);
6902 Node* src_start = array_element_address(src, offset, src_elem);
6903
6904 // static final int[] byteTable in class CRC32C
6905 Node* table = get_table_from_crc32c_class(callee()->holder());
6906 table = must_be_not_null(table, true);
6907 Node* table_start = array_element_address(table, intcon(0), T_INT);
6908
6909 // We assume that range check is done by caller.
6910 // TODO: generate range check (offset+length < src.length) in debug VM.
6911
6912 // Call the stub.
6913 address stubAddr = StubRoutines::updateBytesCRC32C();
6914 const char *stubName = "updateBytesCRC32C";
6915
6916 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6917 stubAddr, stubName, TypePtr::BOTTOM,
6918 crc, src_start, length, table_start);
6919 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6920 set_result(result);
6921 return true;
6922 }
6923
6924 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6925 //
6926 // Calculate CRC32C for DirectByteBuffer.
6927 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6928 //
6929 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6930 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6931 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6932 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6933 // no receiver since it is a static method
6934 Node* crc = argument(0); // type: int
6935 Node* src = argument(1); // type: long
6936 Node* offset = argument(3); // type: int
6937 Node* end = argument(4); // type: int
6938
6939 Node* length = _gvn.transform(new SubINode(end, offset));
6940
6941 src = ConvL2X(src); // adjust Java long to machine word
6942 Node* base = _gvn.transform(new CastX2PNode(src));
6943 offset = ConvI2X(offset);
6944
6945 // 'src_start' points to src array + scaled offset
6946 Node* src_start = off_heap_plus_addr(base, offset);
6947
6948 // static final int[] byteTable in class CRC32C
6949 Node* table = get_table_from_crc32c_class(callee()->holder());
6950 table = must_be_not_null(table, true);
6951 Node* table_start = array_element_address(table, intcon(0), T_INT);
6952
6953 // Call the stub.
6954 address stubAddr = StubRoutines::updateBytesCRC32C();
6955 const char *stubName = "updateBytesCRC32C";
6956
6957 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6958 stubAddr, stubName, TypePtr::BOTTOM,
6959 crc, src_start, length, table_start);
6960 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6961 set_result(result);
6962 return true;
6963 }
6964
6965 //------------------------------inline_updateBytesAdler32----------------------
6966 //
6967 // Calculate Adler32 checksum for byte[] array.
6968 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6969 //
6970 bool LibraryCallKit::inline_updateBytesAdler32() {
6971 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6972 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6973 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6974 // no receiver since it is static method
6975 Node* crc = argument(0); // type: int
6976 Node* src = argument(1); // type: oop
6977 Node* offset = argument(2); // type: int
6978 Node* length = argument(3); // type: int
6979
6980 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6981 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6982 // failed array check
6983 return false;
6984 }
6985
6986 // Figure out the size and type of the elements we will be copying.
6987 BasicType src_elem = src_type->elem()->array_element_basic_type();
6988 if (src_elem != T_BYTE) {
6989 return false;
6990 }
6991
6992 // 'src_start' points to src array + scaled offset
6993 Node* src_start = array_element_address(src, offset, src_elem);
6994
6995 // We assume that range check is done by caller.
6996 // TODO: generate range check (offset+length < src.length) in debug VM.
6997
6998 // Call the stub.
6999 address stubAddr = StubRoutines::updateBytesAdler32();
7000 const char *stubName = "updateBytesAdler32";
7001
7002 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7003 stubAddr, stubName, TypePtr::BOTTOM,
7004 crc, src_start, length);
7005 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7006 set_result(result);
7007 return true;
7008 }
7009
7010 //------------------------------inline_updateByteBufferAdler32---------------
7011 //
7012 // Calculate Adler32 checksum for DirectByteBuffer.
7013 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
7014 //
7015 bool LibraryCallKit::inline_updateByteBufferAdler32() {
7016 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7017 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7018 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7019 // no receiver since it is static method
7020 Node* crc = argument(0); // type: int
7021 Node* src = argument(1); // type: long
7022 Node* offset = argument(3); // type: int
7023 Node* length = argument(4); // type: int
7024
7025 src = ConvL2X(src); // adjust Java long to machine word
7026 Node* base = _gvn.transform(new CastX2PNode(src));
7027 offset = ConvI2X(offset);
7028
7029 // 'src_start' points to src array + scaled offset
7030 Node* src_start = off_heap_plus_addr(base, offset);
7031
7032 // Call the stub.
7033 address stubAddr = StubRoutines::updateBytesAdler32();
7034 const char *stubName = "updateBytesAdler32";
7035
7036 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7037 stubAddr, stubName, TypePtr::BOTTOM,
7038 crc, src_start, length);
7039
7040 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7041 set_result(result);
7042 return true;
7043 }
7044
7045 //----------------------------inline_reference_get0----------------------------
7046 // public T java.lang.ref.Reference.get();
7047 bool LibraryCallKit::inline_reference_get0() {
7048 const int referent_offset = java_lang_ref_Reference::referent_offset();
7049
7050 // Get the argument:
7051 Node* reference_obj = null_check_receiver();
7052 if (stopped()) return true;
7053
7054 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7055 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7056 decorators, /*is_static*/ false,
7057 env()->Reference_klass());
7058 if (result == nullptr) return false;
7059
7060 // Add memory barrier to prevent commoning reads from this field
7061 // across safepoint since GC can change its value.
7062 insert_mem_bar(Op_MemBarCPUOrder);
7063
7064 set_result(result);
7065 return true;
7066 }
7067
7068 //----------------------------inline_reference_refersTo0----------------------------
7069 // bool java.lang.ref.Reference.refersTo0();
7070 // bool java.lang.ref.PhantomReference.refersTo0();
7071 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7072 // Get arguments:
7073 Node* reference_obj = null_check_receiver();
7074 Node* other_obj = argument(1);
7075 if (stopped()) return true;
7076
7077 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7078 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7079 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7080 decorators, /*is_static*/ false,
7081 env()->Reference_klass());
7082 if (referent == nullptr) return false;
7083
7084 // Add memory barrier to prevent commoning reads from this field
7085 // across safepoint since GC can change its value.
7086 insert_mem_bar(Op_MemBarCPUOrder);
7087
7088 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7089 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7090 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7091
7092 RegionNode* region = new RegionNode(3);
7093 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7094
7095 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7096 region->init_req(1, if_true);
7097 phi->init_req(1, intcon(1));
7098
7099 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7100 region->init_req(2, if_false);
7101 phi->init_req(2, intcon(0));
7102
7103 set_control(_gvn.transform(region));
7104 record_for_igvn(region);
7105 set_result(_gvn.transform(phi));
7106 return true;
7107 }
7108
7109 //----------------------------inline_reference_clear0----------------------------
7110 // void java.lang.ref.Reference.clear0();
7111 // void java.lang.ref.PhantomReference.clear0();
7112 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7113 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7114
7115 // Get arguments
7116 Node* reference_obj = null_check_receiver();
7117 if (stopped()) return true;
7118
7119 // Common access parameters
7120 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7121 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7122 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7123 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7124 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7125
7126 Node* referent = access_load_at(reference_obj,
7127 referent_field_addr,
7128 referent_field_addr_type,
7129 val_type,
7130 T_OBJECT,
7131 decorators);
7132
7133 IdealKit ideal(this);
7134 #define __ ideal.
7135 __ if_then(referent, BoolTest::ne, null());
7136 sync_kit(ideal);
7137 access_store_at(reference_obj,
7138 referent_field_addr,
7139 referent_field_addr_type,
7140 null(),
7141 val_type,
7142 T_OBJECT,
7143 decorators);
7144 __ sync_kit(this);
7145 __ end_if();
7146 final_sync(ideal);
7147 #undef __
7148
7149 return true;
7150 }
7151
7152 //-----------------------inline_reference_reachabilityFence-----------------
7153 // bool java.lang.ref.Reference.reachabilityFence();
7154 bool LibraryCallKit::inline_reference_reachabilityFence() {
7155 Node* referent = argument(0);
7156 insert_reachability_fence(referent);
7157 return true;
7158 }
7159
7160 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7161 DecoratorSet decorators, bool is_static,
7162 ciInstanceKlass* fromKls) {
7163 if (fromKls == nullptr) {
7164 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7165 assert(tinst != nullptr, "obj is null");
7166 assert(tinst->is_loaded(), "obj is not loaded");
7167 fromKls = tinst->instance_klass();
7168 }
7169 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7170 ciSymbol::make(fieldTypeString),
7171 is_static);
7172
7173 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7174 if (field == nullptr) return (Node *) nullptr;
7175
7176 if (is_static) {
7177 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7178 fromObj = makecon(tip);
7179 }
7180
7181 // Next code copied from Parse::do_get_xxx():
7182
7183 // Compute address and memory type.
7184 int offset = field->offset_in_bytes();
7185 bool is_vol = field->is_volatile();
7186 ciType* field_klass = field->type();
7187 assert(field_klass->is_loaded(), "should be loaded");
7188 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7189 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7190 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7191 "slice of address and input slice don't match");
7192 BasicType bt = field->layout_type();
7193
7194 // Build the resultant type of the load
7195 const Type *type;
7196 if (bt == T_OBJECT) {
7197 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7198 } else {
7199 type = Type::get_const_basic_type(bt);
7200 }
7201
7202 if (is_vol) {
7203 decorators |= MO_SEQ_CST;
7204 }
7205
7206 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7207 }
7208
7209 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7210 bool is_exact /* true */, bool is_static /* false */,
7211 ciInstanceKlass * fromKls /* nullptr */) {
7212 if (fromKls == nullptr) {
7213 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7214 assert(tinst != nullptr, "obj is null");
7215 assert(tinst->is_loaded(), "obj is not loaded");
7216 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7217 fromKls = tinst->instance_klass();
7218 }
7219 else {
7220 assert(is_static, "only for static field access");
7221 }
7222 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7223 ciSymbol::make(fieldTypeString),
7224 is_static);
7225
7226 assert(field != nullptr, "undefined field");
7227 assert(!field->is_volatile(), "not defined for volatile fields");
7228
7229 if (is_static) {
7230 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7231 fromObj = makecon(tip);
7232 }
7233
7234 // Next code copied from Parse::do_get_xxx():
7235
7236 // Compute address and memory type.
7237 int offset = field->offset_in_bytes();
7238 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7239
7240 return adr;
7241 }
7242
7243 //------------------------------inline_aescrypt_Block-----------------------
7244 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7245 address stubAddr = nullptr;
7246 const char *stubName;
7247 bool is_decrypt = false;
7248 assert(UseAES, "need AES instruction support");
7249
7250 switch(id) {
7251 case vmIntrinsics::_aescrypt_encryptBlock:
7252 stubAddr = StubRoutines::aescrypt_encryptBlock();
7253 stubName = "aescrypt_encryptBlock";
7254 break;
7255 case vmIntrinsics::_aescrypt_decryptBlock:
7256 stubAddr = StubRoutines::aescrypt_decryptBlock();
7257 stubName = "aescrypt_decryptBlock";
7258 is_decrypt = true;
7259 break;
7260 default:
7261 break;
7262 }
7263 if (stubAddr == nullptr) return false;
7264
7265 Node* aescrypt_object = argument(0);
7266 Node* src = argument(1);
7267 Node* src_offset = argument(2);
7268 Node* dest = argument(3);
7269 Node* dest_offset = argument(4);
7270
7271 src = must_be_not_null(src, true);
7272 dest = must_be_not_null(dest, true);
7273
7274 // (1) src and dest are arrays.
7275 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7276 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7277 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7278 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7279
7280 // for the quick and dirty code we will skip all the checks.
7281 // we are just trying to get the call to be generated.
7282 Node* src_start = src;
7283 Node* dest_start = dest;
7284 if (src_offset != nullptr || dest_offset != nullptr) {
7285 assert(src_offset != nullptr && dest_offset != nullptr, "");
7286 src_start = array_element_address(src, src_offset, T_BYTE);
7287 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7288 }
7289
7290 // now need to get the start of its expanded key array
7291 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7292 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7293 if (k_start == nullptr) return false;
7294
7295 // Call the stub.
7296 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7297 stubAddr, stubName, TypePtr::BOTTOM,
7298 src_start, dest_start, k_start);
7299
7300 return true;
7301 }
7302
7303 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7304 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7305 address stubAddr = nullptr;
7306 const char *stubName = nullptr;
7307 bool is_decrypt = false;
7308 assert(UseAES, "need AES instruction support");
7309
7310 switch(id) {
7311 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7312 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7313 stubName = "cipherBlockChaining_encryptAESCrypt";
7314 break;
7315 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7316 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7317 stubName = "cipherBlockChaining_decryptAESCrypt";
7318 is_decrypt = true;
7319 break;
7320 default:
7321 break;
7322 }
7323 if (stubAddr == nullptr) return false;
7324
7325 Node* cipherBlockChaining_object = argument(0);
7326 Node* src = argument(1);
7327 Node* src_offset = argument(2);
7328 Node* len = argument(3);
7329 Node* dest = argument(4);
7330 Node* dest_offset = argument(5);
7331
7332 src = must_be_not_null(src, false);
7333 dest = must_be_not_null(dest, false);
7334
7335 // (1) src and dest are arrays.
7336 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7337 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7338 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7339 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7340
7341 // checks are the responsibility of the caller
7342 Node* src_start = src;
7343 Node* dest_start = dest;
7344 if (src_offset != nullptr || dest_offset != nullptr) {
7345 assert(src_offset != nullptr && dest_offset != nullptr, "");
7346 src_start = array_element_address(src, src_offset, T_BYTE);
7347 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7348 }
7349
7350 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7351 // (because of the predicated logic executed earlier).
7352 // so we cast it here safely.
7353 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7354
7355 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7356 if (embeddedCipherObj == nullptr) return false;
7357
7358 // cast it to what we know it will be at runtime
7359 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7360 assert(tinst != nullptr, "CBC obj is null");
7361 assert(tinst->is_loaded(), "CBC obj is not loaded");
7362 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7363 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7364
7365 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7366 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7367 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7368 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7369 aescrypt_object = _gvn.transform(aescrypt_object);
7370
7371 // we need to get the start of the aescrypt_object's expanded key array
7372 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7373 if (k_start == nullptr) return false;
7374
7375 // similarly, get the start address of the r vector
7376 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7377 if (objRvec == nullptr) return false;
7378 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7379
7380 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7381 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7382 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7383 stubAddr, stubName, TypePtr::BOTTOM,
7384 src_start, dest_start, k_start, r_start, len);
7385
7386 // return cipher length (int)
7387 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7388 set_result(retvalue);
7389 return true;
7390 }
7391
7392 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7393 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7394 address stubAddr = nullptr;
7395 const char *stubName = nullptr;
7396 bool is_decrypt = false;
7397 assert(UseAES, "need AES instruction support");
7398
7399 switch (id) {
7400 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7401 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7402 stubName = "electronicCodeBook_encryptAESCrypt";
7403 break;
7404 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7405 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7406 stubName = "electronicCodeBook_decryptAESCrypt";
7407 is_decrypt = true;
7408 break;
7409 default:
7410 break;
7411 }
7412
7413 if (stubAddr == nullptr) return false;
7414
7415 Node* electronicCodeBook_object = argument(0);
7416 Node* src = argument(1);
7417 Node* src_offset = argument(2);
7418 Node* len = argument(3);
7419 Node* dest = argument(4);
7420 Node* dest_offset = argument(5);
7421
7422 // (1) src and dest are arrays.
7423 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7424 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7425 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7426 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7427
7428 // checks are the responsibility of the caller
7429 Node* src_start = src;
7430 Node* dest_start = dest;
7431 if (src_offset != nullptr || dest_offset != nullptr) {
7432 assert(src_offset != nullptr && dest_offset != nullptr, "");
7433 src_start = array_element_address(src, src_offset, T_BYTE);
7434 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7435 }
7436
7437 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7438 // (because of the predicated logic executed earlier).
7439 // so we cast it here safely.
7440 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7441
7442 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7443 if (embeddedCipherObj == nullptr) return false;
7444
7445 // cast it to what we know it will be at runtime
7446 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7447 assert(tinst != nullptr, "ECB obj is null");
7448 assert(tinst->is_loaded(), "ECB obj is not loaded");
7449 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7450 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7451
7452 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7453 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7454 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7455 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7456 aescrypt_object = _gvn.transform(aescrypt_object);
7457
7458 // we need to get the start of the aescrypt_object's expanded key array
7459 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7460 if (k_start == nullptr) return false;
7461
7462 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7463 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7464 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7465 stubAddr, stubName, TypePtr::BOTTOM,
7466 src_start, dest_start, k_start, len);
7467
7468 // return cipher length (int)
7469 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7470 set_result(retvalue);
7471 return true;
7472 }
7473
7474 //------------------------------inline_counterMode_AESCrypt-----------------------
7475 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7476 assert(UseAES, "need AES instruction support");
7477 if (!UseAESCTRIntrinsics) return false;
7478
7479 address stubAddr = nullptr;
7480 const char *stubName = nullptr;
7481 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7482 stubAddr = StubRoutines::counterMode_AESCrypt();
7483 stubName = "counterMode_AESCrypt";
7484 }
7485 if (stubAddr == nullptr) return false;
7486
7487 Node* counterMode_object = argument(0);
7488 Node* src = argument(1);
7489 Node* src_offset = argument(2);
7490 Node* len = argument(3);
7491 Node* dest = argument(4);
7492 Node* dest_offset = argument(5);
7493
7494 // (1) src and dest are arrays.
7495 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7496 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7497 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7498 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7499
7500 // checks are the responsibility of the caller
7501 Node* src_start = src;
7502 Node* dest_start = dest;
7503 if (src_offset != nullptr || dest_offset != nullptr) {
7504 assert(src_offset != nullptr && dest_offset != nullptr, "");
7505 src_start = array_element_address(src, src_offset, T_BYTE);
7506 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7507 }
7508
7509 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7510 // (because of the predicated logic executed earlier).
7511 // so we cast it here safely.
7512 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7513 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7514 if (embeddedCipherObj == nullptr) return false;
7515 // cast it to what we know it will be at runtime
7516 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7517 assert(tinst != nullptr, "CTR obj is null");
7518 assert(tinst->is_loaded(), "CTR obj is not loaded");
7519 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7520 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7521 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7522 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7523 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7524 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7525 aescrypt_object = _gvn.transform(aescrypt_object);
7526 // we need to get the start of the aescrypt_object's expanded key array
7527 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
7528 if (k_start == nullptr) return false;
7529 // similarly, get the start address of the r vector
7530 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7531 if (obj_counter == nullptr) return false;
7532 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7533
7534 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7535 if (saved_encCounter == nullptr) return false;
7536 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7537 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7538
7539 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7540 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7541 OptoRuntime::counterMode_aescrypt_Type(),
7542 stubAddr, stubName, TypePtr::BOTTOM,
7543 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7544
7545 // return cipher length (int)
7546 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7547 set_result(retvalue);
7548 return true;
7549 }
7550
7551 //------------------------------get_key_start_from_aescrypt_object-----------------------
7552 Node* LibraryCallKit::get_key_start_from_aescrypt_object(Node* aescrypt_object, bool is_decrypt) {
7553 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7554 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7555 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7556 // The following platform specific stubs of encryption and decryption use the same round keys.
7557 #if defined(PPC64) || defined(S390) || defined(RISCV64)
7558 bool use_decryption_key = false;
7559 #else
7560 bool use_decryption_key = is_decrypt;
7561 #endif
7562 Node* objAESCryptKey = load_field_from_object(aescrypt_object, use_decryption_key ? "sessionKd" : "sessionKe", "[I");
7563 assert(objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AES_Crypt");
7564 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7565
7566 // now have the array, need to get the start address of the selected key array
7567 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7568 return k_start;
7569 }
7570
7571 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7572 // Return node representing slow path of predicate check.
7573 // the pseudo code we want to emulate with this predicate is:
7574 // for encryption:
7575 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7576 // for decryption:
7577 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7578 // note cipher==plain is more conservative than the original java code but that's OK
7579 //
7580 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7581 // The receiver was checked for null already.
7582 Node* objCBC = argument(0);
7583
7584 Node* src = argument(1);
7585 Node* dest = argument(4);
7586
7587 // Load embeddedCipher field of CipherBlockChaining object.
7588 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7589
7590 // get AESCrypt klass for instanceOf check
7591 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7592 // will have same classloader as CipherBlockChaining object
7593 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7594 assert(tinst != nullptr, "CBCobj is null");
7595 assert(tinst->is_loaded(), "CBCobj is not loaded");
7596
7597 // we want to do an instanceof comparison against the AESCrypt class
7598 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7599 if (!klass_AESCrypt->is_loaded()) {
7600 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7601 Node* ctrl = control();
7602 set_control(top()); // no regular fast path
7603 return ctrl;
7604 }
7605
7606 src = must_be_not_null(src, true);
7607 dest = must_be_not_null(dest, true);
7608
7609 // Resolve oops to stable for CmpP below.
7610 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7611
7612 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7613 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7614 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7615
7616 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7617
7618 // for encryption, we are done
7619 if (!decrypting)
7620 return instof_false; // even if it is null
7621
7622 // for decryption, we need to add a further check to avoid
7623 // taking the intrinsic path when cipher and plain are the same
7624 // see the original java code for why.
7625 RegionNode* region = new RegionNode(3);
7626 region->init_req(1, instof_false);
7627
7628 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7629 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7630 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7631 region->init_req(2, src_dest_conjoint);
7632
7633 record_for_igvn(region);
7634 return _gvn.transform(region);
7635 }
7636
7637 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7638 // Return node representing slow path of predicate check.
7639 // the pseudo code we want to emulate with this predicate is:
7640 // for encryption:
7641 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7642 // for decryption:
7643 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7644 // note cipher==plain is more conservative than the original java code but that's OK
7645 //
7646 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7647 // The receiver was checked for null already.
7648 Node* objECB = argument(0);
7649
7650 // Load embeddedCipher field of ElectronicCodeBook object.
7651 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7652
7653 // get AESCrypt klass for instanceOf check
7654 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7655 // will have same classloader as ElectronicCodeBook object
7656 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7657 assert(tinst != nullptr, "ECBobj is null");
7658 assert(tinst->is_loaded(), "ECBobj is not loaded");
7659
7660 // we want to do an instanceof comparison against the AESCrypt class
7661 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7662 if (!klass_AESCrypt->is_loaded()) {
7663 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7664 Node* ctrl = control();
7665 set_control(top()); // no regular fast path
7666 return ctrl;
7667 }
7668 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7669
7670 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7671 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7672 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7673
7674 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7675
7676 // for encryption, we are done
7677 if (!decrypting)
7678 return instof_false; // even if it is null
7679
7680 // for decryption, we need to add a further check to avoid
7681 // taking the intrinsic path when cipher and plain are the same
7682 // see the original java code for why.
7683 RegionNode* region = new RegionNode(3);
7684 region->init_req(1, instof_false);
7685 Node* src = argument(1);
7686 Node* dest = argument(4);
7687 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7688 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7689 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7690 region->init_req(2, src_dest_conjoint);
7691
7692 record_for_igvn(region);
7693 return _gvn.transform(region);
7694 }
7695
7696 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7697 // Return node representing slow path of predicate check.
7698 // the pseudo code we want to emulate with this predicate is:
7699 // for encryption:
7700 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7701 // for decryption:
7702 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7703 // note cipher==plain is more conservative than the original java code but that's OK
7704 //
7705
7706 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7707 // The receiver was checked for null already.
7708 Node* objCTR = argument(0);
7709
7710 // Load embeddedCipher field of CipherBlockChaining object.
7711 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7712
7713 // get AESCrypt klass for instanceOf check
7714 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7715 // will have same classloader as CipherBlockChaining object
7716 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7717 assert(tinst != nullptr, "CTRobj is null");
7718 assert(tinst->is_loaded(), "CTRobj is not loaded");
7719
7720 // we want to do an instanceof comparison against the AESCrypt class
7721 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7722 if (!klass_AESCrypt->is_loaded()) {
7723 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7724 Node* ctrl = control();
7725 set_control(top()); // no regular fast path
7726 return ctrl;
7727 }
7728
7729 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7730 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7731 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7732 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7733 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7734
7735 return instof_false; // even if it is null
7736 }
7737
7738 //------------------------------inline_ghash_processBlocks
7739 bool LibraryCallKit::inline_ghash_processBlocks() {
7740 address stubAddr;
7741 const char *stubName;
7742 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7743
7744 stubAddr = StubRoutines::ghash_processBlocks();
7745 stubName = "ghash_processBlocks";
7746
7747 Node* data = argument(0);
7748 Node* offset = argument(1);
7749 Node* len = argument(2);
7750 Node* state = argument(3);
7751 Node* subkeyH = argument(4);
7752
7753 state = must_be_not_null(state, true);
7754 subkeyH = must_be_not_null(subkeyH, true);
7755 data = must_be_not_null(data, true);
7756
7757 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7758 assert(state_start, "state is null");
7759 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7760 assert(subkeyH_start, "subkeyH is null");
7761 Node* data_start = array_element_address(data, offset, T_BYTE);
7762 assert(data_start, "data is null");
7763
7764 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7765 OptoRuntime::ghash_processBlocks_Type(),
7766 stubAddr, stubName, TypePtr::BOTTOM,
7767 state_start, subkeyH_start, data_start, len);
7768 return true;
7769 }
7770
7771 //------------------------------inline_chacha20Block
7772 bool LibraryCallKit::inline_chacha20Block() {
7773 address stubAddr;
7774 const char *stubName;
7775 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7776
7777 stubAddr = StubRoutines::chacha20Block();
7778 stubName = "chacha20Block";
7779
7780 Node* state = argument(0);
7781 Node* result = argument(1);
7782
7783 state = must_be_not_null(state, true);
7784 result = must_be_not_null(result, true);
7785
7786 Node* state_start = array_element_address(state, intcon(0), T_INT);
7787 assert(state_start, "state is null");
7788 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7789 assert(result_start, "result is null");
7790
7791 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7792 OptoRuntime::chacha20Block_Type(),
7793 stubAddr, stubName, TypePtr::BOTTOM,
7794 state_start, result_start);
7795 // return key stream length (int)
7796 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7797 set_result(retvalue);
7798 return true;
7799 }
7800
7801 //------------------------------inline_kyberNtt
7802 bool LibraryCallKit::inline_kyberNtt() {
7803 address stubAddr;
7804 const char *stubName;
7805 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7806 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
7807
7808 stubAddr = StubRoutines::kyberNtt();
7809 stubName = "kyberNtt";
7810 if (!stubAddr) return false;
7811
7812 Node* coeffs = argument(0);
7813 Node* ntt_zetas = argument(1);
7814
7815 coeffs = must_be_not_null(coeffs, true);
7816 ntt_zetas = must_be_not_null(ntt_zetas, true);
7817
7818 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7819 assert(coeffs_start, "coeffs is null");
7820 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
7821 assert(ntt_zetas_start, "ntt_zetas is null");
7822 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7823 OptoRuntime::kyberNtt_Type(),
7824 stubAddr, stubName, TypePtr::BOTTOM,
7825 coeffs_start, ntt_zetas_start);
7826 // return an int
7827 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
7828 set_result(retvalue);
7829 return true;
7830 }
7831
7832 //------------------------------inline_kyberInverseNtt
7833 bool LibraryCallKit::inline_kyberInverseNtt() {
7834 address stubAddr;
7835 const char *stubName;
7836 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7837 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
7838
7839 stubAddr = StubRoutines::kyberInverseNtt();
7840 stubName = "kyberInverseNtt";
7841 if (!stubAddr) return false;
7842
7843 Node* coeffs = argument(0);
7844 Node* zetas = argument(1);
7845
7846 coeffs = must_be_not_null(coeffs, true);
7847 zetas = must_be_not_null(zetas, true);
7848
7849 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7850 assert(coeffs_start, "coeffs is null");
7851 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
7852 assert(zetas_start, "inverseNtt_zetas is null");
7853 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7854 OptoRuntime::kyberInverseNtt_Type(),
7855 stubAddr, stubName, TypePtr::BOTTOM,
7856 coeffs_start, zetas_start);
7857
7858 // return an int
7859 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
7860 set_result(retvalue);
7861 return true;
7862 }
7863
7864 //------------------------------inline_kyberNttMult
7865 bool LibraryCallKit::inline_kyberNttMult() {
7866 address stubAddr;
7867 const char *stubName;
7868 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7869 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
7870
7871 stubAddr = StubRoutines::kyberNttMult();
7872 stubName = "kyberNttMult";
7873 if (!stubAddr) return false;
7874
7875 Node* result = argument(0);
7876 Node* ntta = argument(1);
7877 Node* nttb = argument(2);
7878 Node* zetas = argument(3);
7879
7880 result = must_be_not_null(result, true);
7881 ntta = must_be_not_null(ntta, true);
7882 nttb = must_be_not_null(nttb, true);
7883 zetas = must_be_not_null(zetas, true);
7884
7885 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7886 assert(result_start, "result is null");
7887 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
7888 assert(ntta_start, "ntta is null");
7889 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
7890 assert(nttb_start, "nttb is null");
7891 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
7892 assert(zetas_start, "nttMult_zetas is null");
7893 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
7894 OptoRuntime::kyberNttMult_Type(),
7895 stubAddr, stubName, TypePtr::BOTTOM,
7896 result_start, ntta_start, nttb_start,
7897 zetas_start);
7898
7899 // return an int
7900 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
7901 set_result(retvalue);
7902
7903 return true;
7904 }
7905
7906 //------------------------------inline_kyberAddPoly_2
7907 bool LibraryCallKit::inline_kyberAddPoly_2() {
7908 address stubAddr;
7909 const char *stubName;
7910 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7911 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
7912
7913 stubAddr = StubRoutines::kyberAddPoly_2();
7914 stubName = "kyberAddPoly_2";
7915 if (!stubAddr) return false;
7916
7917 Node* result = argument(0);
7918 Node* a = argument(1);
7919 Node* b = argument(2);
7920
7921 result = must_be_not_null(result, true);
7922 a = must_be_not_null(a, true);
7923 b = must_be_not_null(b, true);
7924
7925 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7926 assert(result_start, "result is null");
7927 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
7928 assert(a_start, "a is null");
7929 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
7930 assert(b_start, "b is null");
7931 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
7932 OptoRuntime::kyberAddPoly_2_Type(),
7933 stubAddr, stubName, TypePtr::BOTTOM,
7934 result_start, a_start, b_start);
7935 // return an int
7936 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
7937 set_result(retvalue);
7938 return true;
7939 }
7940
7941 //------------------------------inline_kyberAddPoly_3
7942 bool LibraryCallKit::inline_kyberAddPoly_3() {
7943 address stubAddr;
7944 const char *stubName;
7945 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7946 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
7947
7948 stubAddr = StubRoutines::kyberAddPoly_3();
7949 stubName = "kyberAddPoly_3";
7950 if (!stubAddr) return false;
7951
7952 Node* result = argument(0);
7953 Node* a = argument(1);
7954 Node* b = argument(2);
7955 Node* c = argument(3);
7956
7957 result = must_be_not_null(result, true);
7958 a = must_be_not_null(a, true);
7959 b = must_be_not_null(b, true);
7960 c = must_be_not_null(c, true);
7961
7962 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7963 assert(result_start, "result is null");
7964 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
7965 assert(a_start, "a is null");
7966 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
7967 assert(b_start, "b is null");
7968 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
7969 assert(c_start, "c is null");
7970 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
7971 OptoRuntime::kyberAddPoly_3_Type(),
7972 stubAddr, stubName, TypePtr::BOTTOM,
7973 result_start, a_start, b_start, c_start);
7974 // return an int
7975 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
7976 set_result(retvalue);
7977 return true;
7978 }
7979
7980 //------------------------------inline_kyber12To16
7981 bool LibraryCallKit::inline_kyber12To16() {
7982 address stubAddr;
7983 const char *stubName;
7984 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7985 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
7986
7987 stubAddr = StubRoutines::kyber12To16();
7988 stubName = "kyber12To16";
7989 if (!stubAddr) return false;
7990
7991 Node* condensed = argument(0);
7992 Node* condensedOffs = argument(1);
7993 Node* parsed = argument(2);
7994 Node* parsedLength = argument(3);
7995
7996 condensed = must_be_not_null(condensed, true);
7997 parsed = must_be_not_null(parsed, true);
7998
7999 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
8000 assert(condensed_start, "condensed is null");
8001 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
8002 assert(parsed_start, "parsed is null");
8003 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
8004 OptoRuntime::kyber12To16_Type(),
8005 stubAddr, stubName, TypePtr::BOTTOM,
8006 condensed_start, condensedOffs, parsed_start, parsedLength);
8007 // return an int
8008 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
8009 set_result(retvalue);
8010 return true;
8011
8012 }
8013
8014 //------------------------------inline_kyberBarrettReduce
8015 bool LibraryCallKit::inline_kyberBarrettReduce() {
8016 address stubAddr;
8017 const char *stubName;
8018 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8019 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
8020
8021 stubAddr = StubRoutines::kyberBarrettReduce();
8022 stubName = "kyberBarrettReduce";
8023 if (!stubAddr) return false;
8024
8025 Node* coeffs = argument(0);
8026
8027 coeffs = must_be_not_null(coeffs, true);
8028
8029 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8030 assert(coeffs_start, "coeffs is null");
8031 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
8032 OptoRuntime::kyberBarrettReduce_Type(),
8033 stubAddr, stubName, TypePtr::BOTTOM,
8034 coeffs_start);
8035 // return an int
8036 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
8037 set_result(retvalue);
8038 return true;
8039 }
8040
8041 //------------------------------inline_dilithiumAlmostNtt
8042 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
8043 address stubAddr;
8044 const char *stubName;
8045 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8046 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
8047
8048 stubAddr = StubRoutines::dilithiumAlmostNtt();
8049 stubName = "dilithiumAlmostNtt";
8050 if (!stubAddr) return false;
8051
8052 Node* coeffs = argument(0);
8053 Node* ntt_zetas = argument(1);
8054
8055 coeffs = must_be_not_null(coeffs, true);
8056 ntt_zetas = must_be_not_null(ntt_zetas, true);
8057
8058 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8059 assert(coeffs_start, "coeffs is null");
8060 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
8061 assert(ntt_zetas_start, "ntt_zetas is null");
8062 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8063 OptoRuntime::dilithiumAlmostNtt_Type(),
8064 stubAddr, stubName, TypePtr::BOTTOM,
8065 coeffs_start, ntt_zetas_start);
8066 // return an int
8067 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
8068 set_result(retvalue);
8069 return true;
8070 }
8071
8072 //------------------------------inline_dilithiumAlmostInverseNtt
8073 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
8074 address stubAddr;
8075 const char *stubName;
8076 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8077 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
8078
8079 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
8080 stubName = "dilithiumAlmostInverseNtt";
8081 if (!stubAddr) return false;
8082
8083 Node* coeffs = argument(0);
8084 Node* zetas = argument(1);
8085
8086 coeffs = must_be_not_null(coeffs, true);
8087 zetas = must_be_not_null(zetas, true);
8088
8089 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8090 assert(coeffs_start, "coeffs is null");
8091 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
8092 assert(zetas_start, "inverseNtt_zetas is null");
8093 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8094 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
8095 stubAddr, stubName, TypePtr::BOTTOM,
8096 coeffs_start, zetas_start);
8097 // return an int
8098 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
8099 set_result(retvalue);
8100 return true;
8101 }
8102
8103 //------------------------------inline_dilithiumNttMult
8104 bool LibraryCallKit::inline_dilithiumNttMult() {
8105 address stubAddr;
8106 const char *stubName;
8107 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8108 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
8109
8110 stubAddr = StubRoutines::dilithiumNttMult();
8111 stubName = "dilithiumNttMult";
8112 if (!stubAddr) return false;
8113
8114 Node* result = argument(0);
8115 Node* ntta = argument(1);
8116 Node* nttb = argument(2);
8117 Node* zetas = argument(3);
8118
8119 result = must_be_not_null(result, true);
8120 ntta = must_be_not_null(ntta, true);
8121 nttb = must_be_not_null(nttb, true);
8122 zetas = must_be_not_null(zetas, true);
8123
8124 Node* result_start = array_element_address(result, intcon(0), T_INT);
8125 assert(result_start, "result is null");
8126 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
8127 assert(ntta_start, "ntta is null");
8128 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
8129 assert(nttb_start, "nttb is null");
8130 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8131 OptoRuntime::dilithiumNttMult_Type(),
8132 stubAddr, stubName, TypePtr::BOTTOM,
8133 result_start, ntta_start, nttb_start);
8134
8135 // return an int
8136 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8137 set_result(retvalue);
8138
8139 return true;
8140 }
8141
8142 //------------------------------inline_dilithiumMontMulByConstant
8143 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8144 address stubAddr;
8145 const char *stubName;
8146 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8147 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8148
8149 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8150 stubName = "dilithiumMontMulByConstant";
8151 if (!stubAddr) return false;
8152
8153 Node* coeffs = argument(0);
8154 Node* constant = argument(1);
8155
8156 coeffs = must_be_not_null(coeffs, true);
8157
8158 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8159 assert(coeffs_start, "coeffs is null");
8160 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8161 OptoRuntime::dilithiumMontMulByConstant_Type(),
8162 stubAddr, stubName, TypePtr::BOTTOM,
8163 coeffs_start, constant);
8164
8165 // return an int
8166 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8167 set_result(retvalue);
8168 return true;
8169 }
8170
8171
8172 //------------------------------inline_dilithiumDecomposePoly
8173 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8174 address stubAddr;
8175 const char *stubName;
8176 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8177 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8178
8179 stubAddr = StubRoutines::dilithiumDecomposePoly();
8180 stubName = "dilithiumDecomposePoly";
8181 if (!stubAddr) return false;
8182
8183 Node* input = argument(0);
8184 Node* lowPart = argument(1);
8185 Node* highPart = argument(2);
8186 Node* twoGamma2 = argument(3);
8187 Node* multiplier = argument(4);
8188
8189 input = must_be_not_null(input, true);
8190 lowPart = must_be_not_null(lowPart, true);
8191 highPart = must_be_not_null(highPart, true);
8192
8193 Node* input_start = array_element_address(input, intcon(0), T_INT);
8194 assert(input_start, "input is null");
8195 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8196 assert(lowPart_start, "lowPart is null");
8197 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8198 assert(highPart_start, "highPart is null");
8199
8200 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8201 OptoRuntime::dilithiumDecomposePoly_Type(),
8202 stubAddr, stubName, TypePtr::BOTTOM,
8203 input_start, lowPart_start, highPart_start,
8204 twoGamma2, multiplier);
8205
8206 // return an int
8207 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8208 set_result(retvalue);
8209 return true;
8210 }
8211
8212 bool LibraryCallKit::inline_base64_encodeBlock() {
8213 address stubAddr;
8214 const char *stubName;
8215 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8216 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8217 stubAddr = StubRoutines::base64_encodeBlock();
8218 stubName = "encodeBlock";
8219
8220 if (!stubAddr) return false;
8221 Node* base64obj = argument(0);
8222 Node* src = argument(1);
8223 Node* offset = argument(2);
8224 Node* len = argument(3);
8225 Node* dest = argument(4);
8226 Node* dp = argument(5);
8227 Node* isURL = argument(6);
8228
8229 src = must_be_not_null(src, true);
8230 dest = must_be_not_null(dest, true);
8231
8232 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8233 assert(src_start, "source array is null");
8234 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8235 assert(dest_start, "destination array is null");
8236
8237 Node* base64 = make_runtime_call(RC_LEAF,
8238 OptoRuntime::base64_encodeBlock_Type(),
8239 stubAddr, stubName, TypePtr::BOTTOM,
8240 src_start, offset, len, dest_start, dp, isURL);
8241 return true;
8242 }
8243
8244 bool LibraryCallKit::inline_base64_decodeBlock() {
8245 address stubAddr;
8246 const char *stubName;
8247 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8248 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8249 stubAddr = StubRoutines::base64_decodeBlock();
8250 stubName = "decodeBlock";
8251
8252 if (!stubAddr) return false;
8253 Node* base64obj = argument(0);
8254 Node* src = argument(1);
8255 Node* src_offset = argument(2);
8256 Node* len = argument(3);
8257 Node* dest = argument(4);
8258 Node* dest_offset = argument(5);
8259 Node* isURL = argument(6);
8260 Node* isMIME = argument(7);
8261
8262 src = must_be_not_null(src, true);
8263 dest = must_be_not_null(dest, true);
8264
8265 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8266 assert(src_start, "source array is null");
8267 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8268 assert(dest_start, "destination array is null");
8269
8270 Node* call = make_runtime_call(RC_LEAF,
8271 OptoRuntime::base64_decodeBlock_Type(),
8272 stubAddr, stubName, TypePtr::BOTTOM,
8273 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8274 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8275 set_result(result);
8276 return true;
8277 }
8278
8279 bool LibraryCallKit::inline_poly1305_processBlocks() {
8280 address stubAddr;
8281 const char *stubName;
8282 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8283 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8284 stubAddr = StubRoutines::poly1305_processBlocks();
8285 stubName = "poly1305_processBlocks";
8286
8287 if (!stubAddr) return false;
8288 null_check_receiver(); // null-check receiver
8289 if (stopped()) return true;
8290
8291 Node* input = argument(1);
8292 Node* input_offset = argument(2);
8293 Node* len = argument(3);
8294 Node* alimbs = argument(4);
8295 Node* rlimbs = argument(5);
8296
8297 input = must_be_not_null(input, true);
8298 alimbs = must_be_not_null(alimbs, true);
8299 rlimbs = must_be_not_null(rlimbs, true);
8300
8301 Node* input_start = array_element_address(input, input_offset, T_BYTE);
8302 assert(input_start, "input array is null");
8303 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
8304 assert(acc_start, "acc array is null");
8305 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
8306 assert(r_start, "r array is null");
8307
8308 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8309 OptoRuntime::poly1305_processBlocks_Type(),
8310 stubAddr, stubName, TypePtr::BOTTOM,
8311 input_start, len, acc_start, r_start);
8312 return true;
8313 }
8314
8315 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
8316 address stubAddr;
8317 const char *stubName;
8318 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8319 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
8320 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
8321 stubName = "intpoly_montgomeryMult_P256";
8322
8323 if (!stubAddr) return false;
8324 null_check_receiver(); // null-check receiver
8325 if (stopped()) return true;
8326
8327 Node* a = argument(1);
8328 Node* b = argument(2);
8329 Node* r = argument(3);
8330
8331 a = must_be_not_null(a, true);
8332 b = must_be_not_null(b, true);
8333 r = must_be_not_null(r, true);
8334
8335 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8336 assert(a_start, "a array is null");
8337 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8338 assert(b_start, "b array is null");
8339 Node* r_start = array_element_address(r, intcon(0), T_LONG);
8340 assert(r_start, "r array is null");
8341
8342 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8343 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
8344 stubAddr, stubName, TypePtr::BOTTOM,
8345 a_start, b_start, r_start);
8346 return true;
8347 }
8348
8349 bool LibraryCallKit::inline_intpoly_assign() {
8350 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8351 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
8352 const char *stubName = "intpoly_assign";
8353 address stubAddr = StubRoutines::intpoly_assign();
8354 if (!stubAddr) return false;
8355
8356 Node* set = argument(0);
8357 Node* a = argument(1);
8358 Node* b = argument(2);
8359 Node* arr_length = load_array_length(a);
8360
8361 a = must_be_not_null(a, true);
8362 b = must_be_not_null(b, true);
8363
8364 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8365 assert(a_start, "a array is null");
8366 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8367 assert(b_start, "b array is null");
8368
8369 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8370 OptoRuntime::intpoly_assign_Type(),
8371 stubAddr, stubName, TypePtr::BOTTOM,
8372 set, a_start, b_start, arr_length);
8373 return true;
8374 }
8375
8376 //------------------------------inline_digestBase_implCompress-----------------------
8377 //
8378 // Calculate MD5 for single-block byte[] array.
8379 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
8380 //
8381 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
8382 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
8383 //
8384 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
8385 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
8386 //
8387 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
8388 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
8389 //
8390 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
8391 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
8392 //
8393 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
8394 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
8395
8396 Node* digestBase_obj = argument(0);
8397 Node* src = argument(1); // type oop
8398 Node* ofs = argument(2); // type int
8399
8400 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8401 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8402 // failed array check
8403 return false;
8404 }
8405 // Figure out the size and type of the elements we will be copying.
8406 BasicType src_elem = src_type->elem()->array_element_basic_type();
8407 if (src_elem != T_BYTE) {
8408 return false;
8409 }
8410 // 'src_start' points to src array + offset
8411 src = must_be_not_null(src, true);
8412 Node* src_start = array_element_address(src, ofs, src_elem);
8413 Node* state = nullptr;
8414 Node* block_size = nullptr;
8415 address stubAddr;
8416 const char *stubName;
8417
8418 switch(id) {
8419 case vmIntrinsics::_md5_implCompress:
8420 assert(UseMD5Intrinsics, "need MD5 instruction support");
8421 state = get_state_from_digest_object(digestBase_obj, T_INT);
8422 stubAddr = StubRoutines::md5_implCompress();
8423 stubName = "md5_implCompress";
8424 break;
8425 case vmIntrinsics::_sha_implCompress:
8426 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
8427 state = get_state_from_digest_object(digestBase_obj, T_INT);
8428 stubAddr = StubRoutines::sha1_implCompress();
8429 stubName = "sha1_implCompress";
8430 break;
8431 case vmIntrinsics::_sha2_implCompress:
8432 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
8433 state = get_state_from_digest_object(digestBase_obj, T_INT);
8434 stubAddr = StubRoutines::sha256_implCompress();
8435 stubName = "sha256_implCompress";
8436 break;
8437 case vmIntrinsics::_sha5_implCompress:
8438 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
8439 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8440 stubAddr = StubRoutines::sha512_implCompress();
8441 stubName = "sha512_implCompress";
8442 break;
8443 case vmIntrinsics::_sha3_implCompress:
8444 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
8445 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8446 stubAddr = StubRoutines::sha3_implCompress();
8447 stubName = "sha3_implCompress";
8448 block_size = get_block_size_from_digest_object(digestBase_obj);
8449 if (block_size == nullptr) return false;
8450 break;
8451 default:
8452 fatal_unexpected_iid(id);
8453 return false;
8454 }
8455 if (state == nullptr) return false;
8456
8457 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
8458 if (stubAddr == nullptr) return false;
8459
8460 // Call the stub.
8461 Node* call;
8462 if (block_size == nullptr) {
8463 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
8464 stubAddr, stubName, TypePtr::BOTTOM,
8465 src_start, state);
8466 } else {
8467 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
8468 stubAddr, stubName, TypePtr::BOTTOM,
8469 src_start, state, block_size);
8470 }
8471
8472 return true;
8473 }
8474
8475 //------------------------------inline_keccak
8476 bool LibraryCallKit::inline_keccak(vmIntrinsics::ID id) {
8477 address stubAddr = nullptr;
8478 const char *stubName;
8479 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
8480 assert((id == vmIntrinsics::_double_keccak && callee()->signature()->size() == 2) ||
8481 (id == vmIntrinsics::_quad_keccak && callee()->signature()->size() == 4),
8482 "double_keccak wrong number of parameters");
8483
8484 int parmCnt = 0;
8485 switch (id) {
8486 case vmIntrinsics::_double_keccak:
8487 stubAddr = StubRoutines::double_keccak();
8488 stubName = "double_keccak";
8489 parmCnt = 2;
8490 break;
8491 case vmIntrinsics::_quad_keccak:
8492 stubAddr = StubRoutines::quad_keccak();
8493 stubName = "quad_keccak";
8494 parmCnt = 4;
8495 break;
8496 default:
8497 ShouldNotReachHere();
8498 }
8499
8500 if (!stubAddr) return false;
8501
8502 Node* state[4];
8503 for (int i = 0; i<parmCnt; i++) {
8504 state[i] = must_be_not_null(argument(i), true);
8505 state[i] = array_element_address(state[i], intcon(0), T_LONG);
8506 assert(state[i], "state[%d] is null", i);
8507 }
8508
8509 Node* keccak;
8510 switch (id) {
8511 case vmIntrinsics::_double_keccak:
8512 keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
8513 OptoRuntime::double_keccak_Type(),
8514 stubAddr, stubName, TypePtr::BOTTOM,
8515 state[0], state[1]);
8516 break;
8517 case vmIntrinsics::_quad_keccak:
8518 keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
8519 OptoRuntime::quad_keccak_Type(),
8520 stubAddr, stubName, TypePtr::BOTTOM,
8521 state[0], state[1], state[2], state[3]);
8522 break;
8523 default:
8524 ShouldNotReachHere();
8525 }
8526
8527 // return an int
8528 Node* retvalue = _gvn.transform(new ProjNode(keccak, TypeFunc::Parms));
8529 set_result(retvalue);
8530 return true;
8531 }
8532
8533
8534 //------------------------------inline_digestBase_implCompressMB-----------------------
8535 //
8536 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
8537 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
8538 //
8539 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
8540 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8541 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8542 assert((uint)predicate < 5, "sanity");
8543 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
8544
8545 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
8546 Node* src = argument(1); // byte[] array
8547 Node* ofs = argument(2); // type int
8548 Node* limit = argument(3); // type int
8549
8550 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8551 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8552 // failed array check
8553 return false;
8554 }
8555 // Figure out the size and type of the elements we will be copying.
8556 BasicType src_elem = src_type->elem()->array_element_basic_type();
8557 if (src_elem != T_BYTE) {
8558 return false;
8559 }
8560 // 'src_start' points to src array + offset
8561 src = must_be_not_null(src, false);
8562 Node* src_start = array_element_address(src, ofs, src_elem);
8563
8564 const char* klass_digestBase_name = nullptr;
8565 const char* stub_name = nullptr;
8566 address stub_addr = nullptr;
8567 BasicType elem_type = T_INT;
8568
8569 switch (predicate) {
8570 case 0:
8571 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
8572 klass_digestBase_name = "sun/security/provider/MD5";
8573 stub_name = "md5_implCompressMB";
8574 stub_addr = StubRoutines::md5_implCompressMB();
8575 }
8576 break;
8577 case 1:
8578 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
8579 klass_digestBase_name = "sun/security/provider/SHA";
8580 stub_name = "sha1_implCompressMB";
8581 stub_addr = StubRoutines::sha1_implCompressMB();
8582 }
8583 break;
8584 case 2:
8585 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
8586 klass_digestBase_name = "sun/security/provider/SHA2";
8587 stub_name = "sha256_implCompressMB";
8588 stub_addr = StubRoutines::sha256_implCompressMB();
8589 }
8590 break;
8591 case 3:
8592 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
8593 klass_digestBase_name = "sun/security/provider/SHA5";
8594 stub_name = "sha512_implCompressMB";
8595 stub_addr = StubRoutines::sha512_implCompressMB();
8596 elem_type = T_LONG;
8597 }
8598 break;
8599 case 4:
8600 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
8601 klass_digestBase_name = "sun/security/provider/SHA3";
8602 stub_name = "sha3_implCompressMB";
8603 stub_addr = StubRoutines::sha3_implCompressMB();
8604 elem_type = T_LONG;
8605 }
8606 break;
8607 default:
8608 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
8609 }
8610 if (klass_digestBase_name != nullptr) {
8611 assert(stub_addr != nullptr, "Stub is generated");
8612 if (stub_addr == nullptr) return false;
8613
8614 // get DigestBase klass to lookup for SHA klass
8615 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
8616 assert(tinst != nullptr, "digestBase_obj is not instance???");
8617 assert(tinst->is_loaded(), "DigestBase is not loaded");
8618
8619 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
8620 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
8621 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
8622 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
8623 }
8624 return false;
8625 }
8626
8627 //------------------------------inline_digestBase_implCompressMB-----------------------
8628 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
8629 BasicType elem_type, address stubAddr, const char *stubName,
8630 Node* src_start, Node* ofs, Node* limit) {
8631 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
8632 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8633 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
8634 digest_obj = _gvn.transform(digest_obj);
8635
8636 Node* state = get_state_from_digest_object(digest_obj, elem_type);
8637 if (state == nullptr) return false;
8638
8639 Node* block_size = nullptr;
8640 if (strcmp("sha3_implCompressMB", stubName) == 0) {
8641 block_size = get_block_size_from_digest_object(digest_obj);
8642 if (block_size == nullptr) return false;
8643 }
8644
8645 // Call the stub.
8646 Node* call;
8647 if (block_size == nullptr) {
8648 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8649 OptoRuntime::digestBase_implCompressMB_Type(false),
8650 stubAddr, stubName, TypePtr::BOTTOM,
8651 src_start, state, ofs, limit);
8652 } else {
8653 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8654 OptoRuntime::digestBase_implCompressMB_Type(true),
8655 stubAddr, stubName, TypePtr::BOTTOM,
8656 src_start, state, block_size, ofs, limit);
8657 }
8658
8659 // return ofs (int)
8660 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8661 set_result(result);
8662
8663 return true;
8664 }
8665
8666 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8667 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8668 assert(UseAES, "need AES instruction support");
8669 address stubAddr = nullptr;
8670 const char *stubName = nullptr;
8671 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8672 stubName = "galoisCounterMode_AESCrypt";
8673
8674 if (stubAddr == nullptr) return false;
8675
8676 Node* in = argument(0);
8677 Node* inOfs = argument(1);
8678 Node* len = argument(2);
8679 Node* ct = argument(3);
8680 Node* ctOfs = argument(4);
8681 Node* out = argument(5);
8682 Node* outOfs = argument(6);
8683 Node* gctr_object = argument(7);
8684 Node* ghash_object = argument(8);
8685
8686 // (1) in, ct and out are arrays.
8687 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8688 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8689 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8690 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8691 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8692 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8693
8694 // checks are the responsibility of the caller
8695 Node* in_start = in;
8696 Node* ct_start = ct;
8697 Node* out_start = out;
8698 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8699 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8700 in_start = array_element_address(in, inOfs, T_BYTE);
8701 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8702 out_start = array_element_address(out, outOfs, T_BYTE);
8703 }
8704
8705 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8706 // (because of the predicated logic executed earlier).
8707 // so we cast it here safely.
8708 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8709 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8710 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8711 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8712 Node* state = load_field_from_object(ghash_object, "state", "[J");
8713
8714 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8715 return false;
8716 }
8717 // cast it to what we know it will be at runtime
8718 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8719 assert(tinst != nullptr, "GCTR obj is null");
8720 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8721 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8722 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8723 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8724 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8725 const TypeOopPtr* xtype = aklass->as_instance_type();
8726 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8727 aescrypt_object = _gvn.transform(aescrypt_object);
8728 // we need to get the start of the aescrypt_object's expanded key array
8729 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
8730 if (k_start == nullptr) return false;
8731 // similarly, get the start address of the r vector
8732 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8733 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8734 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8735
8736
8737 // Call the stub, passing params
8738 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8739 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8740 stubAddr, stubName, TypePtr::BOTTOM,
8741 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8742
8743 // return cipher length (int)
8744 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8745 set_result(retvalue);
8746
8747 return true;
8748 }
8749
8750 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8751 // Return node representing slow path of predicate check.
8752 // the pseudo code we want to emulate with this predicate is:
8753 // for encryption:
8754 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8755 // for decryption:
8756 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8757 // note cipher==plain is more conservative than the original java code but that's OK
8758 //
8759
8760 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8761 // The receiver was checked for null already.
8762 Node* objGCTR = argument(7);
8763 // Load embeddedCipher field of GCTR object.
8764 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8765 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8766
8767 // get AESCrypt klass for instanceOf check
8768 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8769 // will have same classloader as CipherBlockChaining object
8770 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8771 assert(tinst != nullptr, "GCTR obj is null");
8772 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8773
8774 // we want to do an instanceof comparison against the AESCrypt class
8775 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8776 if (!klass_AESCrypt->is_loaded()) {
8777 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8778 Node* ctrl = control();
8779 set_control(top()); // no regular fast path
8780 return ctrl;
8781 }
8782
8783 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8784 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8785 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8786 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8787 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8788
8789 return instof_false; // even if it is null
8790 }
8791
8792 //------------------------------get_state_from_digest_object-----------------------
8793 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8794 const char* state_type;
8795 switch (elem_type) {
8796 case T_BYTE: state_type = "[B"; break;
8797 case T_INT: state_type = "[I"; break;
8798 case T_LONG: state_type = "[J"; break;
8799 default: ShouldNotReachHere();
8800 }
8801 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8802 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8803 if (digest_state == nullptr) return (Node *) nullptr;
8804
8805 // now have the array, need to get the start address of the state array
8806 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8807 return state;
8808 }
8809
8810 //------------------------------get_block_size_from_sha3_object----------------------------------
8811 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8812 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8813 assert (block_size != nullptr, "sanity");
8814 return block_size;
8815 }
8816
8817 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8818 // Return node representing slow path of predicate check.
8819 // the pseudo code we want to emulate with this predicate is:
8820 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8821 //
8822 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8823 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8824 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8825 assert((uint)predicate < 5, "sanity");
8826
8827 // The receiver was checked for null already.
8828 Node* digestBaseObj = argument(0);
8829
8830 // get DigestBase klass for instanceOf check
8831 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8832 assert(tinst != nullptr, "digestBaseObj is null");
8833 assert(tinst->is_loaded(), "DigestBase is not loaded");
8834
8835 const char* klass_name = nullptr;
8836 switch (predicate) {
8837 case 0:
8838 if (UseMD5Intrinsics) {
8839 // we want to do an instanceof comparison against the MD5 class
8840 klass_name = "sun/security/provider/MD5";
8841 }
8842 break;
8843 case 1:
8844 if (UseSHA1Intrinsics) {
8845 // we want to do an instanceof comparison against the SHA class
8846 klass_name = "sun/security/provider/SHA";
8847 }
8848 break;
8849 case 2:
8850 if (UseSHA256Intrinsics) {
8851 // we want to do an instanceof comparison against the SHA2 class
8852 klass_name = "sun/security/provider/SHA2";
8853 }
8854 break;
8855 case 3:
8856 if (UseSHA512Intrinsics) {
8857 // we want to do an instanceof comparison against the SHA5 class
8858 klass_name = "sun/security/provider/SHA5";
8859 }
8860 break;
8861 case 4:
8862 if (UseSHA3Intrinsics) {
8863 // we want to do an instanceof comparison against the SHA3 class
8864 klass_name = "sun/security/provider/SHA3";
8865 }
8866 break;
8867 default:
8868 fatal("unknown SHA intrinsic predicate: %d", predicate);
8869 }
8870
8871 ciKlass* klass = nullptr;
8872 if (klass_name != nullptr) {
8873 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8874 }
8875 if ((klass == nullptr) || !klass->is_loaded()) {
8876 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8877 Node* ctrl = control();
8878 set_control(top()); // no intrinsic path
8879 return ctrl;
8880 }
8881 ciInstanceKlass* instklass = klass->as_instance_klass();
8882
8883 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8884 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8885 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8886 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8887
8888 return instof_false; // even if it is null
8889 }
8890
8891 //-------------inline_fma-----------------------------------
8892 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8893 Node *a = nullptr;
8894 Node *b = nullptr;
8895 Node *c = nullptr;
8896 Node* result = nullptr;
8897 switch (id) {
8898 case vmIntrinsics::_fmaD:
8899 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8900 // no receiver since it is static method
8901 a = argument(0);
8902 b = argument(2);
8903 c = argument(4);
8904 result = _gvn.transform(new FmaDNode(a, b, c));
8905 break;
8906 case vmIntrinsics::_fmaF:
8907 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8908 a = argument(0);
8909 b = argument(1);
8910 c = argument(2);
8911 result = _gvn.transform(new FmaFNode(a, b, c));
8912 break;
8913 default:
8914 fatal_unexpected_iid(id); break;
8915 }
8916 set_result(result);
8917 return true;
8918 }
8919
8920 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8921 // argument(0) is receiver
8922 Node* codePoint = argument(1);
8923 Node* n = nullptr;
8924
8925 switch (id) {
8926 case vmIntrinsics::_isDigit :
8927 n = new DigitNode(control(), codePoint);
8928 break;
8929 case vmIntrinsics::_isLowerCase :
8930 n = new LowerCaseNode(control(), codePoint);
8931 break;
8932 case vmIntrinsics::_isUpperCase :
8933 n = new UpperCaseNode(control(), codePoint);
8934 break;
8935 case vmIntrinsics::_isWhitespace :
8936 n = new WhitespaceNode(control(), codePoint);
8937 break;
8938 default:
8939 fatal_unexpected_iid(id);
8940 }
8941
8942 set_result(_gvn.transform(n));
8943 return true;
8944 }
8945
8946 bool LibraryCallKit::inline_profileBoolean() {
8947 Node* counts = argument(1);
8948 const TypeAryPtr* ary = nullptr;
8949 ciArray* aobj = nullptr;
8950 if (counts->is_Con()
8951 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8952 && (aobj = ary->const_oop()->as_array()) != nullptr
8953 && (aobj->length() == 2)) {
8954 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8955 jint false_cnt = aobj->element_value(0).as_int();
8956 jint true_cnt = aobj->element_value(1).as_int();
8957
8958 if (C->log() != nullptr) {
8959 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8960 false_cnt, true_cnt);
8961 }
8962
8963 if (false_cnt + true_cnt == 0) {
8964 // According to profile, never executed.
8965 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8966 Deoptimization::Action_reinterpret);
8967 return true;
8968 }
8969
8970 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8971 // is a number of each value occurrences.
8972 Node* result = argument(0);
8973 if (false_cnt == 0 || true_cnt == 0) {
8974 // According to profile, one value has been never seen.
8975 int expected_val = (false_cnt == 0) ? 1 : 0;
8976
8977 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8978 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8979
8980 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8981 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8982 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8983
8984 { // Slow path: uncommon trap for never seen value and then reexecute
8985 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8986 // the value has been seen at least once.
8987 PreserveJVMState pjvms(this);
8988 PreserveReexecuteState preexecs(this);
8989 jvms()->set_should_reexecute(true);
8990
8991 set_control(slow_path);
8992 set_i_o(i_o());
8993
8994 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8995 Deoptimization::Action_reinterpret);
8996 }
8997 // The guard for never seen value enables sharpening of the result and
8998 // returning a constant. It allows to eliminate branches on the same value
8999 // later on.
9000 set_control(fast_path);
9001 result = intcon(expected_val);
9002 }
9003 // Stop profiling.
9004 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
9005 // By replacing method body with profile data (represented as ProfileBooleanNode
9006 // on IR level) we effectively disable profiling.
9007 // It enables full speed execution once optimized code is generated.
9008 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
9009 C->record_for_igvn(profile);
9010 set_result(profile);
9011 return true;
9012 } else {
9013 // Continue profiling.
9014 // Profile data isn't available at the moment. So, execute method's bytecode version.
9015 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
9016 // is compiled and counters aren't available since corresponding MethodHandle
9017 // isn't a compile-time constant.
9018 return false;
9019 }
9020 }
9021
9022 bool LibraryCallKit::inline_isCompileConstant() {
9023 Node* n = argument(0);
9024 set_result(n->is_Con() ? intcon(1) : intcon(0));
9025 return true;
9026 }
9027
9028 //------------------------------- inline_getObjectSize --------------------------------------
9029 //
9030 // Calculate the runtime size of the object/array.
9031 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
9032 //
9033 bool LibraryCallKit::inline_getObjectSize() {
9034 Node* obj = argument(3);
9035 Node* klass_node = load_object_klass(obj);
9036
9037 jint layout_con = Klass::_lh_neutral_value;
9038 Node* layout_val = get_layout_helper(klass_node, layout_con);
9039 int layout_is_con = (layout_val == nullptr);
9040
9041 if (layout_is_con) {
9042 // Layout helper is constant, can figure out things at compile time.
9043
9044 if (Klass::layout_helper_is_instance(layout_con)) {
9045 // Instance case: layout_con contains the size itself.
9046 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
9047 set_result(size);
9048 } else {
9049 // Array case: size is round(header + element_size*arraylength).
9050 // Since arraylength is different for every array instance, we have to
9051 // compute the whole thing at runtime.
9052
9053 Node* arr_length = load_array_length(obj);
9054
9055 int round_mask = MinObjAlignmentInBytes - 1;
9056 int hsize = Klass::layout_helper_header_size(layout_con);
9057 int eshift = Klass::layout_helper_log2_element_size(layout_con);
9058
9059 if ((round_mask & ~right_n_bits(eshift)) == 0) {
9060 round_mask = 0; // strength-reduce it if it goes away completely
9061 }
9062 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
9063 Node* header_size = intcon(hsize + round_mask);
9064
9065 Node* lengthx = ConvI2X(arr_length);
9066 Node* headerx = ConvI2X(header_size);
9067
9068 Node* abody = lengthx;
9069 if (eshift != 0) {
9070 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
9071 }
9072 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
9073 if (round_mask != 0) {
9074 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
9075 }
9076 size = ConvX2L(size);
9077 set_result(size);
9078 }
9079 } else {
9080 // Layout helper is not constant, need to test for array-ness at runtime.
9081
9082 enum { _instance_path = 1, _array_path, PATH_LIMIT };
9083 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
9084 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
9085 record_for_igvn(result_reg);
9086
9087 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
9088 if (array_ctl != nullptr) {
9089 // Array case: size is round(header + element_size*arraylength).
9090 // Since arraylength is different for every array instance, we have to
9091 // compute the whole thing at runtime.
9092
9093 PreserveJVMState pjvms(this);
9094 set_control(array_ctl);
9095 Node* arr_length = load_array_length(obj);
9096
9097 int round_mask = MinObjAlignmentInBytes - 1;
9098 Node* mask = intcon(round_mask);
9099
9100 Node* hss = intcon(Klass::_lh_header_size_shift);
9101 Node* hsm = intcon(Klass::_lh_header_size_mask);
9102 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
9103 header_size = _gvn.transform(new AndINode(header_size, hsm));
9104 header_size = _gvn.transform(new AddINode(header_size, mask));
9105
9106 // There is no need to mask or shift this value.
9107 // The semantics of LShiftINode include an implicit mask to 0x1F.
9108 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
9109 Node* elem_shift = layout_val;
9110
9111 Node* lengthx = ConvI2X(arr_length);
9112 Node* headerx = ConvI2X(header_size);
9113
9114 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
9115 Node* size = _gvn.transform(new AddXNode(headerx, abody));
9116 if (round_mask != 0) {
9117 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
9118 }
9119 size = ConvX2L(size);
9120
9121 result_reg->init_req(_array_path, control());
9122 result_val->init_req(_array_path, size);
9123 }
9124
9125 if (!stopped()) {
9126 // Instance case: the layout helper gives us instance size almost directly,
9127 // but we need to mask out the _lh_instance_slow_path_bit.
9128 Node* size = ConvI2X(layout_val);
9129 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
9130 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
9131 size = _gvn.transform(new AndXNode(size, mask));
9132 size = ConvX2L(size);
9133
9134 result_reg->init_req(_instance_path, control());
9135 result_val->init_req(_instance_path, size);
9136 }
9137
9138 set_result(result_reg, result_val);
9139 }
9140
9141 return true;
9142 }
9143
9144 //------------------------------- inline_blackhole --------------------------------------
9145 //
9146 // Make sure all arguments to this node are alive.
9147 // This matches methods that were requested to be blackholed through compile commands.
9148 //
9149 bool LibraryCallKit::inline_blackhole() {
9150 assert(callee()->is_static(), "Should have been checked before: only static methods here");
9151 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
9152 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
9153
9154 // Blackhole node pinches only the control, not memory. This allows
9155 // the blackhole to be pinned in the loop that computes blackholed
9156 // values, but have no other side effects, like breaking the optimizations
9157 // across the blackhole.
9158
9159 Node* bh = _gvn.transform(new BlackholeNode(control()));
9160 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
9161
9162 // Bind call arguments as blackhole arguments to keep them alive
9163 uint nargs = callee()->arg_size();
9164 for (uint i = 0; i < nargs; i++) {
9165 bh->add_req(argument(i));
9166 }
9167
9168 return true;
9169 }
9170
9171 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9172 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9173 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9174 return nullptr; // box klass is not Float16
9175 }
9176
9177 // Null check; get notnull casted pointer
9178 Node* null_ctl = top();
9179 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9180 // If not_null_box is dead, only null-path is taken
9181 if (stopped()) {
9182 set_control(null_ctl);
9183 return nullptr;
9184 }
9185 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9186 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9187 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9188 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9189 }
9190
9191 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9192 PreserveReexecuteState preexecs(this);
9193 jvms()->set_should_reexecute(true);
9194
9195 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9196 Node* klass_node = makecon(klass_type);
9197 Node* box = new_instance(klass_node);
9198
9199 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9200 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9201
9202 Node* field_store = _gvn.transform(access_store_at(box,
9203 value_field,
9204 value_adr_type,
9205 value,
9206 TypeInt::SHORT,
9207 T_SHORT,
9208 IN_HEAP));
9209 set_memory(field_store, value_adr_type);
9210 return box;
9211 }
9212
9213 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9214 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9215 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9216 return false;
9217 }
9218
9219 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9220 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9221 return false;
9222 }
9223
9224 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9225 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9226 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9227 ciSymbols::short_signature(),
9228 false);
9229 assert(field != nullptr, "");
9230
9231 // Transformed nodes
9232 Node* fld1 = nullptr;
9233 Node* fld2 = nullptr;
9234 Node* fld3 = nullptr;
9235 switch(num_args) {
9236 case 3:
9237 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9238 if (fld3 == nullptr) {
9239 return false;
9240 }
9241 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9242 // fall-through
9243 case 2:
9244 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9245 if (fld2 == nullptr) {
9246 return false;
9247 }
9248 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9249 // fall-through
9250 case 1:
9251 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9252 if (fld1 == nullptr) {
9253 return false;
9254 }
9255 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9256 break;
9257 default: fatal("Unsupported number of arguments %d", num_args);
9258 }
9259
9260 Node* result = nullptr;
9261 switch (id) {
9262 // Unary operations
9263 case vmIntrinsics::_sqrt_float16:
9264 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9265 break;
9266 // Ternary operations
9267 case vmIntrinsics::_fma_float16:
9268 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9269 break;
9270 default:
9271 fatal_unexpected_iid(id);
9272 break;
9273 }
9274 result = _gvn.transform(new ReinterpretHF2SNode(result));
9275 set_result(box_fp16_value(float16_box_type, field, result));
9276 return true;
9277 }
9278